xyz.txt

// -----// IR Dump After AutoInputConversionPipelinePass (iree-auto-input-conversion) //----- //
module {
  func.func @fully_dynamic_pack_simple() {
    %0 = flow.tensor.dynamic_constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32> -> tensor<?x?xi32>
    %c0 = arith.constant 0 : index
    %c1 = arith.constant 1 : index
    %1 = util.unfoldable_constant 2 : index
    %dim = tensor.dim %0, %c0 : tensor<?x?xi32>
    %dim_0 = tensor.dim %0, %c1 : tensor<?x?xi32>
    %2 = arith.ceildivui %dim, %1 : index
    %3 = arith.ceildivui %dim_0, %1 : index
    %4 = tensor.empty(%2, %3, %1, %1) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %0 inner_dims_pos = [0, 1] inner_tiles = [%1, %1] into %4 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq_const(%cast, dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>) : tensor<2x2x2x2xi32>
    return
  }
}


// -----// IR Dump After IREEImportPublicPass (iree-import-public) //----- //
module {
  util.func public @fully_dynamic_pack_simple() {
    %0 = flow.tensor.dynamic_constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32> -> tensor<?x?xi32>
    %c0 = arith.constant 0 : index
    %c1 = arith.constant 1 : index
    %1 = util.unfoldable_constant 2 : index
    %dim = tensor.dim %0, %c0 : tensor<?x?xi32>
    %dim_0 = tensor.dim %0, %c1 : tensor<?x?xi32>
    %2 = arith.ceildivui %dim, %1 : index
    %3 = arith.ceildivui %dim_0, %1 : index
    %4 = tensor.empty(%2, %3, %1, %1) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %0 inner_dims_pos = [0, 1] inner_tiles = [%1, %1] into %4 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq_const(%cast, dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ImportMLProgramPass (iree-import-ml-program) //----- //
module {
  util.func public @fully_dynamic_pack_simple() {
    %0 = flow.tensor.dynamic_constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32> -> tensor<?x?xi32>
    %c0 = arith.constant 0 : index
    %c1 = arith.constant 1 : index
    %1 = util.unfoldable_constant 2 : index
    %dim = tensor.dim %0, %c0 : tensor<?x?xi32>
    %dim_0 = tensor.dim %0, %c1 : tensor<?x?xi32>
    %2 = arith.ceildivui %dim, %1 : index
    %3 = arith.ceildivui %dim_0, %1 : index
    %4 = tensor.empty(%2, %3, %1, %1) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %0 inner_dims_pos = [0, 1] inner_tiles = [%1, %1] into %4 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq_const(%cast, dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After SanitizeModuleNamesPass (iree-sanitize-module-names) //----- //
module {
  util.func public @fully_dynamic_pack_simple() {
    %0 = flow.tensor.dynamic_constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32> -> tensor<?x?xi32>
    %c0 = arith.constant 0 : index
    %c1 = arith.constant 1 : index
    %1 = util.unfoldable_constant 2 : index
    %dim = tensor.dim %0, %c0 : tensor<?x?xi32>
    %dim_0 = tensor.dim %0, %c1 : tensor<?x?xi32>
    %2 = arith.ceildivui %dim, %1 : index
    %3 = arith.ceildivui %dim_0, %1 : index
    %4 = tensor.empty(%2, %3, %1, %1) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %0 inner_dims_pos = [0, 1] inner_tiles = [%1, %1] into %4 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq_const(%cast, dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ConvertMeshToFlowPass (iree-convert-mesh-to-flow) //----- //
module {
  util.func public @fully_dynamic_pack_simple() {
    %c1 = arith.constant 1 : index
    %c0 = arith.constant 0 : index
    %0 = flow.tensor.dynamic_constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32> -> tensor<?x?xi32>
    %1 = util.unfoldable_constant 2 : index
    %dim = tensor.dim %0, %c0 : tensor<?x?xi32>
    %dim_0 = tensor.dim %0, %c1 : tensor<?x?xi32>
    %2 = arith.ceildivui %dim, %1 : index
    %3 = arith.ceildivui %dim_0, %1 : index
    %4 = tensor.empty(%2, %3, %1, %1) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %0 inner_dims_pos = [0, 1] inner_tiles = [%1, %1] into %4 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq_const(%cast, dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After DemoteF64ToF32Pass (iree-input-conversion-demote-f64-to-f32) //----- //
module {
  util.func public @fully_dynamic_pack_simple() {
    %c1 = arith.constant 1 : index
    %c0 = arith.constant 0 : index
    %0 = flow.tensor.dynamic_constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32> -> tensor<?x?xi32>
    %1 = util.unfoldable_constant 2 : index
    %dim = tensor.dim %0, %c0 : tensor<?x?xi32>
    %dim_0 = tensor.dim %0, %c1 : tensor<?x?xi32>
    %2 = arith.ceildivui %dim, %1 : index
    %3 = arith.ceildivui %dim_0, %1 : index
    %4 = tensor.empty(%2, %3, %1, %1) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %0 inner_dims_pos = [0, 1] inner_tiles = [%1, %1] into %4 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq_const(%cast, dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After mlir::iree_compiler::IREE::ABI::ConvertStreamableOpsPass (iree-abi-convert-streamable-ops) //----- //
module {
  util.func public @fully_dynamic_pack_simple() {
    %c1 = arith.constant 1 : index
    %c0 = arith.constant 0 : index
    %0 = flow.tensor.dynamic_constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32> -> tensor<?x?xi32>
    %1 = util.unfoldable_constant 2 : index
    %dim = tensor.dim %0, %c0 : tensor<?x?xi32>
    %dim_0 = tensor.dim %0, %c1 : tensor<?x?xi32>
    %2 = arith.ceildivui %dim, %1 : index
    %3 = arith.ceildivui %dim_0, %1 : index
    %4 = tensor.empty(%2, %3, %1, %1) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %0 inner_dims_pos = [0, 1] inner_tiles = [%1, %1] into %4 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq_const(%cast, dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After mlir::iree_compiler::IREE::ABI::WrapEntryPointsPass (iree-abi-wrap-entry-points) //----- //
module {
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c1 = arith.constant 1 : index
    %c0 = arith.constant 0 : index
    %0 = flow.tensor.dynamic_constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32> -> tensor<?x?xi32>
    %1 = util.unfoldable_constant 2 : index
    %dim = tensor.dim %0, %c0 : tensor<?x?xi32>
    %dim_0 = tensor.dim %0, %c1 : tensor<?x?xi32>
    %2 = arith.ceildivui %dim, %1 : index
    %3 = arith.ceildivui %dim_0, %1 : index
    %4 = tensor.empty(%2, %3, %1, %1) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %0 inner_dims_pos = [0, 1] inner_tiles = [%1, %1] into %4 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq_const(%cast, dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
  check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Inliner (inline) //----- //
module {
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
  check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
  check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SymbolDCE (symbol-dce) //----- //
module {
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After AssignLegacyTargetDevicesPass (iree-hal-assign-legacy-target-devices) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {hal.device.targets = [#device_target_local]} {
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After MaterializeTargetDevicesPass (iree-hal-materialize-target-devices) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ResolveDevicePromisesPass (iree-hal-resolve-device-promises) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ResolveDeviceAliasesPass (iree-hal-resolve-device-aliases) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After VerifyDevicesPass (iree-hal-verify-devices) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
    check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After LinalgQuantizedConvToConvPass (iree-global-opt-quantized-conv-to-conv) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After LinalgQuantizedConvToConvPass (iree-global-opt-quantized-conv-to-conv) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
  check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After LinalgQuantizedMatmulToMatmulPass (iree-global-opt-quantized-matmul-to-matmul) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After LinalgQuantizedMatmulToMatmulPass (iree-global-opt-quantized-matmul-to-matmul) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %cast = tensor.cast %pack : tensor<?x?x?x?xi32> to tensor<2x2x2x2xi32>
  check.expect_eq(%cast, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After RemoveZeroExtentTensorsPass (iree-global-opt-remove-zero-extent-tensors) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After DetachElementwiseFromNamedOpsPass (iree-global-opt-detach-elementwise-from-named-ops) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After LinalgNamedOpConversionPass (linalg-named-op-conversion) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Convert1X1FilterConv2DToMatmulPass (iree-global-opt-convert-1x1-filter-conv2d-to-matmul) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After RemoveZeroExtentTensorsPass (iree-global-opt-remove-zero-extent-tensors) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After DetachElementwiseFromNamedOpsPass (iree-global-opt-detach-elementwise-from-named-ops) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After LinalgNamedOpConversionPass (linalg-named-op-conversion) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After Convert1X1FilterConv2DToMatmulPass (iree-global-opt-convert-1x1-filter-conv2d-to-matmul) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After EraseUnusedLinalgOperandsPass (iree-global-opt-erase-unused-linalg-operands) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ExpandTensorShapesPass (iree-global-opt-expand-tensor-shapes) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ConvertElementwiseToLinalgPass (convert-elementwise-to-linalg) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After ConvertElementwiseToLinalgPass (convert-elementwise-to-linalg) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After RaiseSpecialOpsPass (iree-global-opt-raise-special-ops) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After RaiseSpecialOpsPass (iree-global-opt-raise-special-ops) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After DecomposeConcatPass (iree-global-opt-decompose-concat) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After GeneralizeLinalgNamedOpsPass (iree-global-opt-generalize-linalg-named-ops) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After DecomposeConcatPass (iree-global-opt-decompose-concat) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After GeneralizeLinalgNamedOpsPass (iree-global-opt-generalize-linalg-named-ops) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After FoldUnitExtentDimsPass (iree-dispatch-creation-fold-unit-extent-dims) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After DemoteContractionInputsToBF16Pass (iree-global-opt-demote-contraction-inputs-to-bf16) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After DemoteContractionInputsToBF16Pass (iree-global-opt-demote-contraction-inputs-to-bf16) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SetEncodingPass (iree-dispatch-creation-set-encoding) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SetEncodingPass (iree-dispatch-creation-set-encoding) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CPUMaterializeHostEncodingPass (iree-codegen-cpu-materialize-host-encoding) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After MaterializeHomogeneousEncodingsPass (iree-global-opt-materialize-homogeneous-encodings) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After CSE (cse) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After SimplifyPackUnpackPass (iree-global-opt-simplify-pack-unpack) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After DataLayoutPropagationPass (iree-global-opt-data-layout-propagation) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After GeneralizeLinalgNamedOpsPass (iree-global-opt-generalize-linalg-named-ops) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After GlobalLoopInvariantCodeMotionPass (iree-global-opt-loop-invariant-code-motion) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After DataLayoutPropagationPass (iree-global-opt-data-layout-propagation) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After GeneralizeLinalgNamedOpsPass (iree-global-opt-generalize-linalg-named-ops) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After GlobalLoopInvariantCodeMotionPass (iree-global-opt-loop-invariant-code-motion) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After CSE (cse) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After HoistIntoGlobals (iree-util-hoist-into-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After JitGlobalsPass (iree-consteval-jit-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After RaiseSpecialOpsPass (iree-global-opt-raise-special-ops) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After InjectTensorTracingPass (iree-flow-inject-tensor-tracing) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After RaiseSpecialOpsPass (iree-global-opt-raise-special-ops) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After InjectTensorTracingPass (iree-flow-inject-tensor-tracing) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After TensorPadToTensorInsertSlicePass (iree-dispatch-creation-tensor-pad-to-tensor-insert-slice) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FixedPointIterator (iree-util-fixed-point-iterator) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FusionPreprocessingPass (iree-dispatch-creation-fusion-preprocessing) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After FusionPreprocessingPass (iree-dispatch-creation-fusion-preprocessing) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ElementwiseOpFusionPass (iree-dispatch-creation-elementwise-op-fusion) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After ElementwiseOpFusionPass (iree-dispatch-creation-elementwise-op-fusion) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After BubbleUpExpandShapesPass (iree-dispatch-creation-bubble-up-expand-shapes) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After BubbleUpExtractSlicesPass (iree-dispatch-creation-bubble-up-extract-slices) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After BubbleUpExpandShapesPass (iree-dispatch-creation-bubble-up-expand-shapes) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ElementwiseOpFusionPass (iree-dispatch-creation-elementwise-op-fusion) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After BubbleUpExtractSlicesPass (iree-dispatch-creation-bubble-up-extract-slices) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SinkReshapesPass (iree-dispatch-creation-sink-reshapes) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After FuseMultiUseElementwiseProducerPass (iree-dispatch-creation-fuse-multi-use-elementwise-producer) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After ElementwiseOpFusionPass (iree-dispatch-creation-elementwise-op-fusion) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After SplitReductionPass (iree-dispatch-creation-split-reduction-ops) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After TransposeGenericOpsPass (iree-dispatch-creation-transpose-generic-ops) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After FormScalarDispatchesPass (iree-dispatch-creation-form-scalar-dispatches) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After FormDispatchRegionsPass (iree-dispatch-creation-form-dispatch-regions) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CloneProducersIntoDispatchRegionsPass (iree-dispatch-creation-clone-producers-into-dispatch-regions) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CollapseDimensionsPass (iree-dispatch-creation-collapse-dimensions) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After ConvertDispatchRegionsToWorkgroupsPass (iree-dispatch-creation-convert-dispatch-regions-to-workgroups) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After SinkReshapesPass (iree-dispatch-creation-sink-reshapes) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ConvertTensorToFlowPass (iree-dispatch-creation-convert-tensor-to-flow) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After FuseMultiUseElementwiseProducerPass (iree-dispatch-creation-fuse-multi-use-elementwise-producer) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SplitReductionPass (iree-dispatch-creation-split-reduction-ops) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After TransposeGenericOpsPass (iree-dispatch-creation-transpose-generic-ops) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After FormScalarDispatchesPass (iree-dispatch-creation-form-scalar-dispatches) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  %8 = flow.tensor.reshape %pack : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After MaterializeDefaultWorkgroupCountRegionPass (iree-dispatch-creation-materialize-default-workgroup-count-region) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After FormDispatchRegionsPass (iree-dispatch-creation-form-dispatch-regions) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %c0 = arith.constant 0 : index
  %c1 = arith.constant 1 : index
  %c2_0 = arith.constant 2 : index
  %c3 = arith.constant 3 : index
  %8 = flow.dispatch.region -> (tensor<?x?x?x?xi32>{%5, %6, %4, %4}) {
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %7 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    flow.return %pack : tensor<?x?x?x?xi32>
  }
  %9 = flow.tensor.reshape %8 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%9, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CloneProducersIntoDispatchRegionsPass (iree-dispatch-creation-clone-producers-into-dispatch-regions) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %c0 = arith.constant 0 : index
  %c1 = arith.constant 1 : index
  %c2_0 = arith.constant 2 : index
  %c3 = arith.constant 3 : index
  %8 = flow.dispatch.region -> (tensor<?x?x?x?xi32>{%5, %6, %4, %4}) {
    %10 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %10 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    flow.return %pack : tensor<?x?x?x?xi32>
  }
  %9 = flow.tensor.reshape %8 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%9, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CollapseDimensionsPass (iree-dispatch-creation-collapse-dimensions) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %c0 = arith.constant 0 : index
  %c1 = arith.constant 1 : index
  %c2_0 = arith.constant 2 : index
  %c3 = arith.constant 3 : index
  %8 = flow.dispatch.region -> (tensor<?x?x?x?xi32>{%5, %6, %4, %4}) {
    %10 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %3 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %10 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    flow.return %pack : tensor<?x?x?x?xi32>
  }
  %9 = flow.tensor.reshape %8 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%9, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ConvertDispatchRegionsToWorkgroupsPass (iree-dispatch-creation-convert-dispatch-regions-to-workgroups) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = tensor.empty(%5, %6, %4, %4) : tensor<?x?x?x?xi32>
  %c0 = arith.constant 0 : index
  %c1 = arith.constant 1 : index
  %c2_0 = arith.constant 2 : index
  %c3 = arith.constant 3 : index
  %8 = flow.dispatch.workgroups(%5, %6, %4, %3, %1, %2, %5, %6, %4, %4) : (index, index, index, tensor<?x?xi32>{%1, %2}, index, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4} =
      (%arg0: index, %arg1: index, %arg2: index, %arg3: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg4: index, %arg5: index, %arg6: index, %arg7: index, %arg8: index, %arg9: index, %arg10: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
    %10 = flow.dispatch.tensor.load %arg3, offsets = [0, 0], sizes = [%arg4, %arg5], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%arg4, %arg5} -> tensor<?x?xi32>
    %11 = tensor.empty(%arg6, %arg7, %arg9, %arg9) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %10 inner_dims_pos = [0, 1] inner_tiles = [%arg9, %arg9] into %11 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    flow.dispatch.tensor.store %pack, %arg10, offsets = [0, 0, 0, 0], sizes = [%arg6, %arg7, %arg9, %arg9], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%arg6, %arg7, %arg9, %arg9}
    flow.return
  }
  %9 = flow.tensor.reshape %8 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%9, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ConvertTensorToFlowPass (iree-dispatch-creation-convert-tensor-to-flow) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = flow.dispatch.workgroups(%5, %6, %4, %3, %1, %2, %5, %6, %4, %4) : (index, index, index, tensor<?x?xi32>{%1, %2}, index, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4} =
      (%arg0: index, %arg1: index, %arg2: index, %arg3: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg4: index, %arg5: index, %arg6: index, %arg7: index, %arg8: index, %arg9: index, %arg10: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
    %9 = flow.dispatch.tensor.load %arg3, offsets = [0, 0], sizes = [%arg4, %arg5], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%arg4, %arg5} -> tensor<?x?xi32>
    %10 = tensor.empty(%arg6, %arg7, %arg9, %arg9) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %9 inner_dims_pos = [0, 1] inner_tiles = [%arg9, %arg9] into %10 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    flow.dispatch.tensor.store %pack, %arg10, offsets = [0, 0, 0, 0], sizes = [%arg6, %arg7, %arg9, %arg9], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%arg6, %arg7, %arg9, %arg9}
    flow.return
  }
  %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = flow.dispatch.workgroups(%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4} =
      (%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
    %9 = flow.dispatch.tensor.load %arg0, offsets = [0, 0], sizes = [%arg1, %arg2], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%arg1, %arg2} -> tensor<?x?xi32>
    %10 = tensor.empty(%arg3, %arg4, %arg5, %arg5) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %9 inner_dims_pos = [0, 1] inner_tiles = [%arg5, %arg5] into %10 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    flow.dispatch.tensor.store %pack, %arg6, offsets = [0, 0, 0, 0], sizes = [%arg3, %arg4, %arg5, %arg5], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%arg3, %arg4, %arg5, %arg5}
    flow.return
  }
  %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After MaterializeDefaultWorkgroupCountRegionPass (iree-dispatch-creation-materialize-default-workgroup-count-region) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = flow.dispatch.workgroups[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4} =
      (%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
    %9 = flow.dispatch.workload.ordinal %arg1, 0 : index
    %10 = flow.dispatch.workload.ordinal %arg2, 1 : index
    %11 = flow.dispatch.workload.ordinal %arg3, 2 : index
    %12 = flow.dispatch.workload.ordinal %arg4, 3 : index
    %13 = flow.dispatch.workload.ordinal %arg5, 4 : index
    %14 = flow.dispatch.tensor.load %arg0, offsets = [0, 0], sizes = [%9, %10], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%9, %10} -> tensor<?x?xi32>
    %15 = tensor.empty(%11, %12, %13, %13) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %14 inner_dims_pos = [0, 1] inner_tiles = [%13, %13] into %15 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    flow.dispatch.tensor.store %pack, %arg6, offsets = [0, 0, 0, 0], sizes = [%11, %12, %13, %13], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%11, %12, %13, %13}
    flow.return
  } count(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
    %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
    flow.return %x, %y, %z : index, index, index
  }
  %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After VerifyInputLegalityPass (iree-verify-input-legality) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = flow.dispatch.workgroups[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4} =
        (%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
      %9 = flow.dispatch.workload.ordinal %arg1, 0 : index
      %10 = flow.dispatch.workload.ordinal %arg2, 1 : index
      %11 = flow.dispatch.workload.ordinal %arg3, 2 : index
      %12 = flow.dispatch.workload.ordinal %arg4, 3 : index
      %13 = flow.dispatch.workload.ordinal %arg5, 4 : index
      %14 = flow.dispatch.tensor.load %arg0, offsets = [0, 0], sizes = [%9, %10], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%9, %10} -> tensor<?x?xi32>
      %15 = tensor.empty(%11, %12, %13, %13) : tensor<?x?x?x?xi32>
      %pack = tensor.pack %14 inner_dims_pos = [0, 1] inner_tiles = [%13, %13] into %15 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
      flow.dispatch.tensor.store %pack, %arg6, offsets = [0, 0, 0, 0], sizes = [%11, %12, %13, %13], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%11, %12, %13, %13}
      flow.return
    } count(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After CaptureDynamicDimsPass (iree-flow-capture-dynamic-dims) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CaptureDynamicDimsPass (iree-flow-capture-dynamic-dims) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = flow.dispatch.workgroups[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4} =
      (%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
    %9 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%arg1, %arg2}
    %10 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%arg3, %arg4, %arg5, %arg5}
    %11 = flow.dispatch.workload.ordinal %arg1, 0 : index
    %12 = flow.dispatch.workload.ordinal %arg2, 1 : index
    %13 = flow.dispatch.workload.ordinal %arg3, 2 : index
    %14 = flow.dispatch.workload.ordinal %arg4, 3 : index
    %15 = flow.dispatch.workload.ordinal %arg5, 4 : index
    %16 = flow.dispatch.tensor.load %9, offsets = [0, 0], sizes = [%11, %12], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%11, %12} -> tensor<?x?xi32>
    %17 = tensor.empty(%13, %14, %15, %15) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %16 inner_dims_pos = [0, 1] inner_tiles = [%15, %15] into %17 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    flow.dispatch.tensor.store %pack, %10, offsets = [0, 0, 0, 0], sizes = [%13, %14, %15, %15], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%13, %14, %15, %15}
    flow.return
  } count(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
    %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
    flow.return %x, %y, %z : index, index, index
  }
  %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After InitializeEmptyTensorsPass (iree-flow-initialize-empty-tensors) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = flow.dispatch.workgroups[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4} =
      (%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
    %9 = flow.dispatch.workload.ordinal %arg1, 0 : index
    %10 = flow.dispatch.workload.ordinal %arg2, 1 : index
    %11 = flow.dispatch.workload.ordinal %arg3, 2 : index
    %12 = flow.dispatch.workload.ordinal %arg4, 3 : index
    %13 = flow.dispatch.workload.ordinal %arg5, 4 : index
    %14 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%9, %10}
    %15 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%11, %12, %13, %13}
    %16 = flow.dispatch.tensor.load %14, offsets = [0, 0], sizes = [%9, %10], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%9, %10} -> tensor<?x?xi32>
    %17 = tensor.empty(%11, %12, %13, %13) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %16 inner_dims_pos = [0, 1] inner_tiles = [%13, %13] into %17 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    flow.dispatch.tensor.store %pack, %15, offsets = [0, 0, 0, 0], sizes = [%11, %12, %13, %13], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%11, %12, %13, %13}
    flow.return
  } count(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
    %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
    flow.return %x, %y, %z : index, index, index
  }
  %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = flow.dispatch.workgroups[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4} =
      (%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
    %9 = flow.dispatch.workload.ordinal %arg1, 0 : index
    %10 = flow.dispatch.workload.ordinal %arg2, 1 : index
    %11 = flow.dispatch.workload.ordinal %arg3, 2 : index
    %12 = flow.dispatch.workload.ordinal %arg4, 3 : index
    %13 = flow.dispatch.workload.ordinal %arg5, 4 : index
    %14 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%9, %10}
    %15 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%11, %12, %13, %13}
    %16 = flow.dispatch.tensor.load %14, offsets = [0, 0], sizes = [%9, %10], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%9, %10} -> tensor<?x?xi32>
    %17 = tensor.empty(%11, %12, %13, %13) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %16 inner_dims_pos = [0, 1] inner_tiles = [%13, %13] into %17 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    flow.dispatch.tensor.store %pack, %15, offsets = [0, 0, 0, 0], sizes = [%11, %12, %13, %13], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%11, %12, %13, %13}
    flow.return
  } count(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
    %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
    flow.return %x, %y, %z : index, index, index
  }
  %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After InitializeEmptyTensorsPass (iree-flow-initialize-empty-tensors) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = flow.dispatch.workgroups[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4} =
      (%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
    %9 = flow.dispatch.workload.ordinal %arg1, 0 : index
    %10 = flow.dispatch.workload.ordinal %arg2, 1 : index
    %11 = flow.dispatch.workload.ordinal %arg3, 2 : index
    %12 = flow.dispatch.workload.ordinal %arg4, 3 : index
    %13 = flow.dispatch.workload.ordinal %arg5, 4 : index
    %14 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%9, %10}
    %15 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%11, %12, %13, %13}
    %16 = flow.dispatch.tensor.load %14, offsets = [0, 0], sizes = [%9, %10], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%9, %10} -> tensor<?x?xi32>
    %17 = tensor.empty(%11, %12, %13, %13) : tensor<?x?x?x?xi32>
    %pack = tensor.pack %16 inner_dims_pos = [0, 1] inner_tiles = [%13, %13] into %17 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
    flow.dispatch.tensor.store %pack, %15, offsets = [0, 0, 0, 0], sizes = [%11, %12, %13, %13], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%11, %12, %13, %13}
    flow.return
  } count(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
    %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
    flow.return %x, %y, %z : index, index, index
  }
  %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After OutlineDispatchExternsPass (iree-flow-outline-dispatch-externs) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = flow.dispatch.workgroups[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4} =
        (%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
      %9 = flow.dispatch.workload.ordinal %arg1, 0 : index
      %10 = flow.dispatch.workload.ordinal %arg2, 1 : index
      %11 = flow.dispatch.workload.ordinal %arg3, 2 : index
      %12 = flow.dispatch.workload.ordinal %arg4, 3 : index
      %13 = flow.dispatch.workload.ordinal %arg5, 4 : index
      %14 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%9, %10}
      %15 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%11, %12, %13, %13}
      %16 = flow.dispatch.tensor.load %14, offsets = [0, 0], sizes = [%9, %10], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%9, %10} -> tensor<?x?xi32>
      %17 = tensor.empty(%11, %12, %13, %13) : tensor<?x?x?x?xi32>
      %pack = tensor.pack %16 inner_dims_pos = [0, 1] inner_tiles = [%13, %13] into %17 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
      flow.dispatch.tensor.store %pack, %15, offsets = [0, 0, 0, 0], sizes = [%11, %12, %13, %13], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%11, %12, %13, %13}
      flow.return
    } count(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After OutlineDispatchRegionsPass (iree-flow-outline-dispatch-regions) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4}
    %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After AnnotateDispatchesPass (iree-flow-annotate-dispatches) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4}
    %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After StripDebugOps (iree-util-strip-debug-ops) //----- //
flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
  flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
    %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
    flow.return %x, %y, %z : index, index, index
  }
  builtin.module {
    func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
      %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
      %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
      %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
      %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
      %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
      %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
      %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
      %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
      %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
      %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
      flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
      return
    }
  }
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4}
  %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After DeduplicateExecutablesPass (iree-flow-deduplicate-executables) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.func private @_fully_dynamic_pack_simple() {
    %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4}
    %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After InjectTensorTracingPass (iree-flow-inject-tensor-tracing) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CleanupTensorShapesPass (iree-flow-cleanup-tensor-shapes) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After InjectTensorTracingPass (iree-flow-inject-tensor-tracing) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4}
  %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CleanupTensorShapesPass (iree-flow-cleanup-tensor-shapes) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %cst = arith.constant dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %c2 = arith.constant 2 : index
  %0 = flow.tensor.constant dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = flow.tensor.reshape %0 : tensor<4x4xi32> -> tensor<?x?xi32>{%1, %2}
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%3, %1, %2, %5, %6, %4) : (tensor<?x?xi32>{%1, %2}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%5, %6, %4, %4}
  %8 = flow.tensor.reshape %7 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%8, %cst) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After OutlineConstantsPass (iree-flow-outline-constants) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %c2 = arith.constant 2 : index
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %c4 = arith.constant 4 : index
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CanonicalizerPass (iree-flow-canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
  %3 = util.optimization_barrier %c2 : index
  %4 = arith.ceildivui %0, %3 : index
  %5 = arith.ceildivui %1, %3 : index
  %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
  %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
  %3 = util.optimization_barrier %c2 : index
  %4 = arith.ceildivui %0, %3 : index
  %5 = arith.ceildivui %1, %3 : index
  %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
  %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
  %3 = util.optimization_barrier %c2 : index
  %4 = arith.ceildivui %0, %3 : index
  %5 = arith.ceildivui %1, %3 : index
  %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
  %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FixedPointIterator (iree-util-fixed-point-iterator) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After SymbolDCE (symbol-dce) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After VerifyInputPass (iree-stream-verify-input) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
  %3 = util.optimization_barrier %c2 : index
  %4 = arith.ceildivui %0, %3 : index
  %5 = arith.ceildivui %1, %3 : index
  %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
  %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
  %3 = util.optimization_barrier %c2 : index
  %4 = arith.ceildivui %0, %3 : index
  %5 = arith.ceildivui %1, %3 : index
  %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
  %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
  %3 = util.optimization_barrier %c2 : index
  %4 = arith.ceildivui %0, %3 : index
  %5 = arith.ceildivui %1, %3 : index
  %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
  %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
  check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  flow.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    flow.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      flow.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !flow.dispatch.tensor<readonly:tensor<?x?xi32>>, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>) {
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = flow.dispatch.tie_shape %arg0 : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = flow.dispatch.tie_shape %arg6 : !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global private @__constant_tensor_4x4xi32 {inlining_policy = #util.inline.never, stream.affinity.default = #hal.device.affinity<@__device_0>} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.func private @_fully_dynamic_pack_simple() {
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : tensor<2x2x2x2xi32>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : tensor<4x4xi32>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = flow.tensor.reshape %__constant_tensor_4x4xi32 : tensor<4x4xi32> -> tensor<?x?xi32>{%0, %1}
    %3 = util.optimization_barrier %c2 : index
    %4 = arith.ceildivui %0, %3 : index
    %5 = arith.ceildivui %1, %3 : index
    %6 = flow.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %4, %5, %3](%2, %0, %1, %4, %5, %3) : (tensor<?x?xi32>{%0, %1}, index, index, index, index, index) -> tensor<?x?x?x?xi32>{%4, %5, %3, %3}
    %7 = flow.tensor.reshape %6 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} -> tensor<2x2x2x2xi32>
    check.expect_eq(%7, %__constant_tensor_2x2x2x2xi32) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ConvertToStreamPass (iree-stream-conversion) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_2x2x2x2xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_4x4xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_4x4xi32__size : index
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32__size = util.global.load @__constant_tensor_2x2x2x2xi32__size : index
    %0 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_2x2x2x2xi32__size}
    %__constant_tensor_4x4xi32 = util.global.load @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32__size = util.global.load @__constant_tensor_4x4xi32__size : index
    %1 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
    %2 = util.optimization_barrier %c4 : index
    %3 = util.optimization_barrier %c4 : index
    %4 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%2, %3} : index
    %5 = stream.tensor.clone on(#hal.device.affinity<@__device_0>) %1 : tensor<4x4xi32> in !stream.resource<*>{%__constant_tensor_4x4xi32__size} -> tensor<?x?xi32>{%2, %3} in !stream.resource<*>{%4}
    %6 = util.optimization_barrier %c2 : index
    %7 = arith.ceildivui %2, %6 : index
    %8 = arith.ceildivui %3, %6 : index
    %c0 = arith.constant 0 : index
    %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%7, %8, %6, %6} : index
    %10 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%2, %3, %7, %8, %6](%5[%c0 to %4 for %4], %2, %3, %7, %8, %6) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%9}
    %11 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
    %12 = stream.tensor.clone on(#hal.device.affinity<@__device_0>) %10 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} in !stream.resource<*>{%9} -> tensor<2x2x2x2xi32> in !stream.resource<*>{%11}
    %13 = stream.async.transfer %12 : !stream.resource<*>{%11} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%11}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%11} -> tensor<2x2x2x2xi32>
    %15 = stream.async.transfer %0 : !stream.resource<*>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
    %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After VerifyLoweringToTensorsPass (iree-stream-verify-lowering-to-tensors) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_2x2x2x2xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_4x4xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_4x4xi32__size : index
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32__size = util.global.load @__constant_tensor_2x2x2x2xi32__size : index
    %0 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_2x2x2x2xi32__size}
    %__constant_tensor_4x4xi32 = util.global.load @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32__size = util.global.load @__constant_tensor_4x4xi32__size : index
    %1 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
    %2 = util.optimization_barrier %c4 : index
    %3 = util.optimization_barrier %c4 : index
    %4 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%2, %3} : index
    %5 = stream.tensor.clone on(#hal.device.affinity<@__device_0>) %1 : tensor<4x4xi32> in !stream.resource<*>{%__constant_tensor_4x4xi32__size} -> tensor<?x?xi32>{%2, %3} in !stream.resource<*>{%4}
    %6 = util.optimization_barrier %c2 : index
    %7 = arith.ceildivui %2, %6 : index
    %8 = arith.ceildivui %3, %6 : index
    %c0 = arith.constant 0 : index
    %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%7, %8, %6, %6} : index
    %10 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%2, %3, %7, %8, %6](%5[%c0 to %4 for %4], %2, %3, %7, %8, %6) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%9}
    %11 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
    %12 = stream.tensor.clone on(#hal.device.affinity<@__device_0>) %10 : tensor<?x?x?x?xi32>{%c2, %c2, %c2, %c2} in !stream.resource<*>{%9} -> tensor<2x2x2x2xi32> in !stream.resource<*>{%11}
    %13 = stream.async.transfer %12 : !stream.resource<*>{%11} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%11}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%11} -> tensor<2x2x2x2xi32>
    %15 = stream.async.transfer %0 : !stream.resource<*>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
    %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_2x2x2x2xi32__size = util.global.load @__constant_tensor_2x2x2x2xi32__size : index
  %__constant_tensor_4x4xi32 = util.global.load @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__size = util.global.load @__constant_tensor_4x4xi32__size : index
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%1, %2} : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%5, %6, %4, %4} : index
  %8 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%0[%c0 to %3 for %3], %1, %2, %5, %6, %4) : (!stream.resource<*>{%3}, index, index, index, index, index) -> !stream.resource<*>{%7}
  %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
  %10 = stream.async.transfer %8 : !stream.resource<*>{%9} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%9}
  %11 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %10 : tensor<2x2x2x2xi32> in !stream.resource<external>{%9} -> tensor<2x2x2x2xi32>
  %12 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
  %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
  check.expect_eq(%11, %13) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
  %c0 = arith.constant 0 : index
  %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
  %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
  %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
  %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
  %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
  %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
  %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
  %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
  %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
  %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
  flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
  return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %0 = stream.resource.size %cst : !stream.resource<constant>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %0 = stream.resource.size %cst : !stream.resource<constant>
  util.global.store %cst, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %0, @__constant_tensor_4x4xi32__size : index
  util.return
}

// -----// IR Dump After Inliner (inline) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_2x2x2x2xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_4x4xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_4x4xi32__size : index
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32__size = util.global.load @__constant_tensor_2x2x2x2xi32__size : index
    %__constant_tensor_4x4xi32 = util.global.load @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32__size = util.global.load @__constant_tensor_4x4xi32__size : index
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%1, %2} : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%5, %6, %4, %4} : index
    %8 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%0[%c0 to %3 for %3], %1, %2, %5, %6, %4) : (!stream.resource<*>{%3}, index, index, index, index, index) -> !stream.resource<*>{%7}
    %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
    %10 = stream.async.transfer %8 : !stream.resource<*>{%9} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%9}
    %11 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %10 : tensor<2x2x2x2xi32> in !stream.resource<external>{%9} -> tensor<2x2x2x2xi32>
    %12 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
    %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
    check.expect_eq(%11, %13) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %0 = stream.resource.size %cst : !stream.resource<constant>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.initializer {
  %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %0 = stream.resource.size %cst : !stream.resource<constant>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %0 = stream.resource.size %cst : !stream.resource<constant>
  util.global.store %cst, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %0, @__constant_tensor_4x4xi32__size : index
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_2x2x2x2xi32__size = util.global.load @__constant_tensor_2x2x2x2xi32__size : index
  %__constant_tensor_4x4xi32 = util.global.load @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__size = util.global.load @__constant_tensor_4x4xi32__size : index
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%1, %2} : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%5, %6, %4, %4} : index
  %8 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%0[%c0 to %3 for %3], %1, %2, %5, %6, %4) : (!stream.resource<*>{%3}, index, index, index, index, index) -> !stream.resource<*>{%7}
  %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
  %10 = stream.async.transfer %8 : !stream.resource<*>{%9} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%9}
  %11 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %10 : tensor<2x2x2x2xi32> in !stream.resource<external>{%9} -> tensor<2x2x2x2xi32>
  %12 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
  %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
  check.expect_eq(%11, %13) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_2x2x2x2xi32__size = util.global.load @__constant_tensor_2x2x2x2xi32__size : index
  %__constant_tensor_4x4xi32 = util.global.load @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__size = util.global.load @__constant_tensor_4x4xi32__size : index
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%1, %2} : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%5, %6, %4, %4} : index
  %8 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%0[%c0 to %3 for %3], %1, %2, %5, %6, %4) : (!stream.resource<*>{%3}, index, index, index, index, index) -> !stream.resource<*>{%7}
  %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
  %10 = stream.async.transfer %8 : !stream.resource<*>{%9} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%9}
  %11 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %10 : tensor<2x2x2x2xi32> in !stream.resource<external>{%9} -> tensor<2x2x2x2xi32>
  %12 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
  %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
  check.expect_eq(%11, %13) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.initializer {
  %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %0 = stream.resource.size %cst : !stream.resource<constant>
  util.global.store %cst, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %0, @__constant_tensor_4x4xi32__size : index
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.initializer {
  %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  %0 = stream.resource.size %cst : !stream.resource<constant>
  util.global.store %cst, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %0, @__constant_tensor_4x4xi32__size : index
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_2x2x2x2xi32 = util.global.load @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_2x2x2x2xi32__size = util.global.load @__constant_tensor_2x2x2x2xi32__size : index
  %__constant_tensor_4x4xi32 = util.global.load @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__size = util.global.load @__constant_tensor_4x4xi32__size : index
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%1, %2} : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %1, %4 : index
  %6 = arith.ceildivui %2, %4 : index
  %7 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%5, %6, %4, %4} : index
  %8 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%0[%c0 to %3 for %3], %1, %2, %5, %6, %4) : (!stream.resource<*>{%3}, index, index, index, index, index) -> !stream.resource<*>{%7}
  %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
  %10 = stream.async.transfer %8 : !stream.resource<*>{%9} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%9}
  %11 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %10 : tensor<2x2x2x2xi32> in !stream.resource<external>{%9} -> tensor<2x2x2x2xi32>
  %12 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
  %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
  check.expect_eq(%11, %13) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.initializer {
  %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %0 = stream.resource.size %cst : !stream.resource<constant>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_2x2x2x2xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_4x4xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_4x4xi32__size : index
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32__size = util.global.load @__constant_tensor_2x2x2x2xi32__size : index
    %__constant_tensor_4x4xi32 = util.global.load @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32__size = util.global.load @__constant_tensor_4x4xi32__size : index
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%1, %2} : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%5, %6, %4, %4} : index
    %8 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%0[%c0 to %3 for %3], %1, %2, %5, %6, %4) : (!stream.resource<*>{%3}, index, index, index, index, index) -> !stream.resource<*>{%7}
    %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
    %10 = stream.async.transfer %8 : !stream.resource<*>{%9} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%9}
    %11 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %10 : tensor<2x2x2x2xi32> in !stream.resource<external>{%9} -> tensor<2x2x2x2xi32>
    %12 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
    %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
    check.expect_eq(%11, %13) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_2x2x2x2xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_4x4xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_4x4xi32__size : index
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32__size = util.global.load immutable @__constant_tensor_2x2x2x2xi32__size : index
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32__size = util.global.load immutable @__constant_tensor_4x4xi32__size : index
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%1, %2} : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%5, %6, %4, %4} : index
    %8 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%0[%c0 to %3 for %3], %1, %2, %5, %6, %4) : (!stream.resource<*>{%3}, index, index, index, index, index) -> !stream.resource<*>{%7}
    %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
    %10 = stream.async.transfer %8 : !stream.resource<*>{%9} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%9}
    %11 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %10 : tensor<2x2x2x2xi32> in !stream.resource<external>{%9} -> tensor<2x2x2x2xi32>
    %12 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
    %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
    check.expect_eq(%11, %13) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_2x2x2x2xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_4x4xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_4x4xi32__size : index
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32__size = util.global.load immutable @__constant_tensor_2x2x2x2xi32__size : index
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32__size = util.global.load immutable @__constant_tensor_4x4xi32__size : index
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%1, %2} : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%5, %6, %4, %4} : index
    %8 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%0[%c0 to %3 for %3], %1, %2, %5, %6, %4) : (!stream.resource<*>{%3}, index, index, index, index, index) -> !stream.resource<*>{%7}
    %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
    %10 = stream.async.transfer %8 : !stream.resource<*>{%9} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%9}
    %11 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %10 : tensor<2x2x2x2xi32> in !stream.resource<external>{%9} -> tensor<2x2x2x2xi32>
    %12 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
    %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
    check.expect_eq(%11, %13) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_2x2x2x2xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_4x4xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_4x4xi32__size : index
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32__size = util.global.load immutable @__constant_tensor_2x2x2x2xi32__size : index
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32__size = util.global.load immutable @__constant_tensor_4x4xi32__size : index
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%1, %2} : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%5, %6, %4, %4} : index
    %8 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%0[%c0 to %3 for %3], %1, %2, %5, %6, %4) : (!stream.resource<*>{%3}, index, index, index, index, index) -> !stream.resource<*>{%7}
    %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
    %10 = stream.async.transfer %8 : !stream.resource<*>{%9} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%9}
    %11 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %10 : tensor<2x2x2x2xi32> in !stream.resource<external>{%9} -> tensor<2x2x2x2xi32>
    %12 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
    %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
    check.expect_eq(%11, %13) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After CombineInitializers (iree-util-combine-initializers) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_2x2x2x2xi32__size : index
  util.initializer {
    %cst = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<2x2x2x2xi32> in !stream.resource<constant> = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %0 = stream.resource.size %cst : !stream.resource<constant>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %0, @__constant_tensor_2x2x2x2xi32__size : index
    %cst_0 = stream.tensor.constant on(#hal.device.affinity<@__device_0>) : tensor<4x4xi32> in !stream.resource<constant> = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    %1 = stream.resource.size %cst_0 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.global.store %1, @__constant_tensor_4x4xi32__size : index
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_4x4xi32__size : index
  util.func private @_fully_dynamic_pack_simple() {
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32__size = util.global.load immutable @__constant_tensor_2x2x2x2xi32__size : index
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32__size = util.global.load immutable @__constant_tensor_4x4xi32__size : index
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?xi32>{%1, %2} : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %1, %4 : index
    %6 = arith.ceildivui %2, %4 : index
    %7 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<?x?x?x?xi32>{%5, %6, %4, %4} : index
    %8 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %5, %6, %4](%0[%c0 to %3 for %3], %1, %2, %5, %6, %4) : (!stream.resource<*>{%3}, index, index, index, index, index) -> !stream.resource<*>{%7}
    %9 = stream.tensor.sizeof on(#hal.device.affinity<@__device_0>) tensor<2x2x2x2xi32> : index
    %10 = stream.async.transfer %8 : !stream.resource<*>{%9} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%9}
    %11 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %10 : tensor<2x2x2x2xi32> in !stream.resource<external>{%9} -> tensor<2x2x2x2xi32>
    %12 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
    %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
    check.expect_eq(%11, %13) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After EncodeHostTensorsPass (iree-stream-encode-host-tensors) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After EncodeHostTensorsPass (iree-stream-encode-host-tensors) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %c64, @__constant_tensor_2x2x2x2xi32__size : index
  %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %c64, @__constant_tensor_4x4xi32__size : index
  util.return
}

// -----// IR Dump After EncodeDeviceTensorsPass (iree-stream-encode-device-tensors) //----- //
stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
  stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
    %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
    stream.return %x, %y, %z : index, index, index
  }
  builtin.module {
    func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
      %c0 = arith.constant 0 : index
      %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
      %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
      %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
      %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
      %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
      %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
      %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
      %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
      %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
      %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
      flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
      return
    }
  }
}

// -----// IR Dump After EncodeHostTensorsPass (iree-stream-encode-host-tensors) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_2x2x2x2xi32__size = util.global.load immutable @__constant_tensor_2x2x2x2xi32__size : index
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__size = util.global.load immutable @__constant_tensor_4x4xi32__size : index
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = arith.muli %1, %c4 : index
  %4 = arith.muli %3, %2 : index
  %5 = util.optimization_barrier %c2 : index
  %6 = arith.ceildivui %1, %5 : index
  %7 = arith.ceildivui %2, %5 : index
  %8 = arith.muli %6, %c4 : index
  %9 = arith.muli %8, %7 : index
  %10 = arith.muli %9, %5 : index
  %11 = arith.muli %10, %5 : index
  %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
  %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
  %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
  check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %c64, @__constant_tensor_2x2x2x2xi32__size : index
  %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %c64, @__constant_tensor_4x4xi32__size : index
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %c64, @__constant_tensor_2x2x2x2xi32__size : index
  %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %c64, @__constant_tensor_4x4xi32__size : index
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %c64, @__constant_tensor_2x2x2x2xi32__size : index
  util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %c64, @__constant_tensor_4x4xi32__size : index
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_2x2x2x2xi32__size = util.global.load immutable @__constant_tensor_2x2x2x2xi32__size : index
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__size = util.global.load immutable @__constant_tensor_4x4xi32__size : index
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = arith.muli %1, %c4 : index
  %4 = arith.muli %3, %2 : index
  %5 = util.optimization_barrier %c2 : index
  %6 = arith.ceildivui %1, %5 : index
  %7 = arith.ceildivui %2, %5 : index
  %8 = arith.muli %6, %c4 : index
  %9 = arith.muli %8, %7 : index
  %10 = arith.muli %9, %5 : index
  %11 = arith.muli %10, %5 : index
  %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
  %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
  %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
  check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_2x2x2x2xi32__size = util.global.load immutable @__constant_tensor_2x2x2x2xi32__size : index
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__size = util.global.load immutable @__constant_tensor_4x4xi32__size : index
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = arith.muli %1, %c4 : index
  %4 = arith.muli %3, %2 : index
  %5 = util.optimization_barrier %c2 : index
  %6 = arith.ceildivui %1, %5 : index
  %7 = arith.ceildivui %2, %5 : index
  %8 = arith.muli %6, %c4 : index
  %9 = arith.muli %8, %7 : index
  %10 = arith.muli %9, %5 : index
  %11 = arith.muli %10, %5 : index
  %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
  %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
  %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
  check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_2x2x2x2xi32__size = util.global.load immutable @__constant_tensor_2x2x2x2xi32__size : index
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__size = util.global.load immutable @__constant_tensor_4x4xi32__size : index
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = arith.muli %1, %c4 : index
  %4 = arith.muli %3, %2 : index
  %5 = util.optimization_barrier %c2 : index
  %6 = arith.ceildivui %1, %5 : index
  %7 = arith.ceildivui %2, %5 : index
  %8 = arith.muli %6, %c4 : index
  %9 = arith.muli %8, %7 : index
  %10 = arith.muli %9, %5 : index
  %11 = arith.muli %10, %5 : index
  %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
  %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
  %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
  check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_2x2x2x2xi32__size = 64 : index
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global private @__constant_tensor_4x4xi32__size = 64 : index
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32__size = util.global.load immutable @__constant_tensor_2x2x2x2xi32__size : index
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32__size = util.global.load immutable @__constant_tensor_4x4xi32__size : index
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%__constant_tensor_4x4xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%__constant_tensor_4x4xi32__size}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = arith.muli %1, %c4 : index
    %4 = arith.muli %3, %2 : index
    %5 = util.optimization_barrier %c2 : index
    %6 = arith.ceildivui %1, %5 : index
    %7 = arith.ceildivui %2, %5 : index
    %8 = arith.muli %6, %c4 : index
    %9 = arith.muli %8, %7 : index
    %10 = arith.muli %9, %5 : index
    %11 = arith.muli %10, %5 : index
    %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
    %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__size} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size}
    %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%__constant_tensor_2x2x2x2xi32__size} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%c64}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = arith.muli %1, %c4 : index
    %4 = arith.muli %3, %2 : index
    %5 = util.optimization_barrier %c2 : index
    %6 = arith.ceildivui %1, %5 : index
    %7 = arith.ceildivui %2, %5 : index
    %8 = arith.muli %6, %c4 : index
    %9 = arith.muli %8, %7 : index
    %10 = arith.muli %9, %5 : index
    %11 = arith.muli %10, %5 : index
    %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
    %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%c64}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = arith.muli %1, %c4 : index
    %4 = arith.muli %3, %2 : index
    %5 = util.optimization_barrier %c2 : index
    %6 = arith.ceildivui %1, %5 : index
    %7 = arith.ceildivui %2, %5 : index
    %8 = arith.muli %6, %c4 : index
    %9 = arith.muli %8, %7 : index
    %10 = arith.muli %9, %5 : index
    %11 = arith.muli %10, %5 : index
    %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
    %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%c64}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = arith.muli %1, %c4 : index
    %4 = arith.muli %3, %2 : index
    %5 = util.optimization_barrier %c2 : index
    %6 = arith.ceildivui %1, %5 : index
    %7 = arith.ceildivui %2, %5 : index
    %8 = arith.muli %6, %c4 : index
    %9 = arith.muli %8, %7 : index
    %10 = arith.muli %9, %5 : index
    %11 = arith.muli %10, %5 : index
    %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
    %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After VerifyLoweringToAsyncResourcesPass (iree-stream-verify-lowering-to-async-resources) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%c64}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = arith.muli %1, %c4 : index
    %4 = arith.muli %3, %2 : index
    %5 = util.optimization_barrier %c2 : index
    %6 = arith.ceildivui %1, %5 : index
    %7 = arith.ceildivui %2, %5 : index
    %8 = arith.muli %6, %c4 : index
    %9 = arith.muli %8, %7 : index
    %10 = arith.muli %9, %5 : index
    %11 = arith.muli %10, %5 : index
    %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
    %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After MaterializeCopyOnWritePass (iree-stream-materialize-copy-on-write) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After MaterializeCopyOnWritePass (iree-stream-materialize-copy-on-write) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After MaterializeCopyOnWritePass (iree-stream-materialize-copy-on-write) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%c64}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = arith.muli %1, %c4 : index
  %4 = arith.muli %3, %2 : index
  %5 = util.optimization_barrier %c2 : index
  %6 = arith.ceildivui %1, %5 : index
  %7 = arith.ceildivui %2, %5 : index
  %8 = arith.muli %6, %c4 : index
  %9 = arith.muli %8, %7 : index
  %10 = arith.muli %9, %5 : index
  %11 = arith.muli %10, %5 : index
  %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
  %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%c64}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = arith.muli %1, %c4 : index
  %4 = arith.muli %3, %2 : index
  %5 = util.optimization_barrier %c2 : index
  %6 = arith.ceildivui %1, %5 : index
  %7 = arith.ceildivui %2, %5 : index
  %8 = arith.muli %6, %c4 : index
  %9 = arith.muli %8, %7 : index
  %10 = arith.muli %9, %5 : index
  %11 = arith.muli %10, %5 : index
  %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
  %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After ElideAsyncCopiesPass (iree-stream-elide-async-copies) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%c64}
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c4 : index
    %3 = arith.muli %1, %c4 : index
    %4 = arith.muli %3, %2 : index
    %5 = util.optimization_barrier %c2 : index
    %6 = arith.ceildivui %1, %5 : index
    %7 = arith.ceildivui %2, %5 : index
    %8 = arith.muli %6, %c4 : index
    %9 = arith.muli %8, %7 : index
    %10 = arith.muli %9, %5 : index
    %11 = arith.muli %10, %5 : index
    %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
    %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After EmplaceAllocationsPass (iree-stream-emplace-allocations) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%c64}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = arith.muli %1, %c4 : index
  %4 = arith.muli %3, %2 : index
  %5 = util.optimization_barrier %c2 : index
  %6 = arith.ceildivui %1, %5 : index
  %7 = arith.ceildivui %2, %5 : index
  %8 = arith.muli %6, %c4 : index
  %9 = arith.muli %8, %7 : index
  %10 = arith.muli %9, %5 : index
  %11 = arith.muli %10, %5 : index
  %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
  %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After EmplaceAllocationsPass (iree-stream-emplace-allocations) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = stream.async.transfer %__constant_tensor_4x4xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<*>{%c64}
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c4 : index
  %3 = arith.muli %1, %c4 : index
  %4 = arith.muli %3, %2 : index
  %5 = util.optimization_barrier %c2 : index
  %6 = arith.ceildivui %1, %5 : index
  %7 = arith.ceildivui %2, %5 : index
  %8 = arith.muli %6, %c4 : index
  %9 = arith.muli %8, %7 : index
  %10 = arith.muli %9, %5 : index
  %11 = arith.muli %10, %5 : index
  %12 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%1, %2, %6, %7, %5](%0[%c0 to %4 for %4], %1, %2, %6, %7, %5) : (!stream.resource<*>{%4}, index, index, index, index, index) -> !stream.resource<*>{%11}
  %13 = stream.async.transfer %12 : !stream.resource<*>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %15 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%14, %16) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After EmplaceAllocationsPass (iree-stream-emplace-allocations) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After RefineUsagePass (iree-stream-refine-usage) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%__constant_tensor_4x4xi32[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
    %12 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %13 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%12, %14) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
  %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
  util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%__constant_tensor_4x4xi32[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
  %12 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %13 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%12, %14) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%__constant_tensor_4x4xi32[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
  %12 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %13 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%12, %14) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%__constant_tensor_4x4xi32[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
  %12 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %13 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%12, %14) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%__constant_tensor_4x4xi32[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
    %12 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %13 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%12, %14) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%__constant_tensor_4x4xi32[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
    %12 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %13 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%12, %14) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%__constant_tensor_4x4xi32[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
    %12 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %13 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%12, %14) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%__constant_tensor_4x4xi32[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
    %12 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %13 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%12, %14) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After VerifyAsyncAccessRangesPass (iree-stream-verify-async-access-ranges) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %cst = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    util.global.store %cst, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %cst_0, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.async.dispatch on(#hal.device.affinity<@__device_0>) @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%__constant_tensor_4x4xi32[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
    %12 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %13 = stream.async.transfer %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> to(#hal.device.affinity<@__device_0>) !stream.resource<external>{%c64}
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %13 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%12, %14) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ScheduleExecutionPass (iree-stream-schedule-execution) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After ScheduleConcurrencyPass (iree-stream-schedule-concurrency) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After ScheduleExecutionPass (iree-stream-schedule-execution) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
    %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
  } => !stream.timepoint
  %0:2 = stream.timepoint.await %result_timepoint => %results#0, %results#1 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
  util.global.store %0#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %0#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After ScheduleExecutionPass (iree-stream-schedule-execution) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
    %14 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%arg0[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
    %15 = stream.async.transfer %arg1 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
    stream.yield %14, %15 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
  } => !stream.timepoint
  %11:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
  %12 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%13, %12) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ScheduleConcurrencyPass (iree-stream-schedule-concurrency) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
    %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
  } => !stream.timepoint
  %0:2 = stream.timepoint.await %result_timepoint => %results#0, %results#1 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
  util.global.store %0#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %0#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After ScheduleConcurrencyPass (iree-stream-schedule-concurrency) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
    %14:2 = stream.async.concurrent with(%arg0 as %arg2: !stream.resource<constant>{%3}, %arg1 as %arg3: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
      %15 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%arg2[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
      %16 = stream.async.transfer %arg3 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
      stream.yield %15, %16 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
    }
    stream.yield %14#0, %14#1 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
  } => !stream.timepoint
  %11:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
  %12 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %11#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%13, %12) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After PropagateTimepointsPass (iree-stream-propagate-timepoints) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private mutable @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
      %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
      %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
      stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
    } => !stream.timepoint
    %0:2 = stream.timepoint.await %result_timepoint => %results#0, %results#1 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
    util.global.store %result_timepoint, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.global.store %results#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %result_timepoint, @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
    util.global.store %results#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private mutable @__constant_tensor_4x4xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = stream.timepoint.await %__constant_tensor_2x2x2x2xi32__timepoint => %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64}
    %__constant_tensor_4x4xi32__timepoint = util.global.load @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %1 = stream.resource.size %__constant_tensor_4x4xi32 : !stream.resource<constant>
    %2 = stream.timepoint.await %__constant_tensor_4x4xi32__timepoint => %__constant_tensor_4x4xi32 : !stream.resource<constant>{%1}
    %3 = util.optimization_barrier %c4 : index
    %4 = util.optimization_barrier %c4 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = util.optimization_barrier %c2 : index
    %8 = arith.ceildivui %3, %7 : index
    %9 = arith.ceildivui %4, %7 : index
    %10 = arith.muli %8, %c4 : index
    %11 = arith.muli %10, %9 : index
    %12 = arith.muli %11, %7 : index
    %13 = arith.muli %12, %7 : index
    %14 = stream.timepoint.join max(%__constant_tensor_4x4xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint) => !stream.timepoint
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) await(%14) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%6}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%13}, !stream.resource<external>{%c64}) {
      %18:2 = stream.async.concurrent with(%arg0 as %arg2: !stream.resource<constant>{%6}, %arg1 as %arg3: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%13}, !stream.resource<external>{%c64}) {
        %19 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%3, %4, %8, %9, %7](%arg2[%c0 to %6 for %6], %3, %4, %8, %9, %7) : (!stream.resource<constant>{%6}, index, index, index, index, index) -> !stream.resource<external>{%13}
        %20 = stream.async.transfer %arg3 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
        stream.yield %19, %20 : !stream.resource<external>{%13}, !stream.resource<external>{%c64}
      }
      stream.yield %18#0, %18#1 : !stream.resource<external>{%13}, !stream.resource<external>{%c64}
    } => !stream.timepoint
    %15:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%13}
    %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%17, %16) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After MaterializeBuiltinsPass (iree-stream-materialize-builtins) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private mutable @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
      %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
      %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
      stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
    } => !stream.timepoint
    %0:2 = stream.timepoint.await %result_timepoint => %results#0, %results#1 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
    util.global.store %result_timepoint, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.global.store %results#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %result_timepoint, @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
    util.global.store %results#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private mutable @__constant_tensor_4x4xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = stream.timepoint.await %__constant_tensor_2x2x2x2xi32__timepoint => %__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>{%c64}
    %__constant_tensor_4x4xi32__timepoint = util.global.load @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %1 = stream.resource.size %__constant_tensor_4x4xi32 : !stream.resource<constant>
    %2 = stream.timepoint.await %__constant_tensor_4x4xi32__timepoint => %__constant_tensor_4x4xi32 : !stream.resource<constant>{%1}
    %3 = util.optimization_barrier %c4 : index
    %4 = util.optimization_barrier %c4 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = util.optimization_barrier %c2 : index
    %8 = arith.ceildivui %3, %7 : index
    %9 = arith.ceildivui %4, %7 : index
    %10 = arith.muli %8, %c4 : index
    %11 = arith.muli %10, %9 : index
    %12 = arith.muli %11, %7 : index
    %13 = arith.muli %12, %7 : index
    %14 = stream.timepoint.join max(%__constant_tensor_4x4xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint) => !stream.timepoint
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) await(%14) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%6}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%13}, !stream.resource<external>{%c64}) {
      %18:2 = stream.async.concurrent with(%arg0 as %arg2: !stream.resource<constant>{%6}, %arg1 as %arg3: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%13}, !stream.resource<external>{%c64}) {
        %19 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%3, %4, %8, %9, %7](%arg2[%c0 to %6 for %6], %3, %4, %8, %9, %7) : (!stream.resource<constant>{%6}, index, index, index, index, index) -> !stream.resource<external>{%13}
        %20 = stream.async.transfer %arg3 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
        stream.yield %19, %20 : !stream.resource<external>{%13}, !stream.resource<external>{%c64}
      }
      stream.yield %18#0, %18#1 : !stream.resource<external>{%13}, !stream.resource<external>{%c64}
    } => !stream.timepoint
    %15:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%13}
    %16 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%17, %16) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
    %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
  } => !stream.timepoint
  util.global.store %result_timepoint, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.global.store %results#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %result_timepoint, @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
  util.global.store %results#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
    %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
  } => !stream.timepoint
  util.global.store %result_timepoint, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.global.store %results#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %result_timepoint, @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
  util.global.store %results#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
    %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
    %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
    stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
  } => !stream.timepoint
  util.global.store %results#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %result_timepoint, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.global.store %results#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %result_timepoint, @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__timepoint = util.global.load @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = stream.timepoint.join max(%__constant_tensor_4x4xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint) => !stream.timepoint
  %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) await(%11) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
    %15:2 = stream.async.concurrent with(%arg0 as %arg2: !stream.resource<constant>{%3}, %arg1 as %arg3: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
      %16 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%arg2[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
      %17 = stream.async.transfer %arg3 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
      stream.yield %16, %17 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
    }
    stream.yield %15#0, %15#1 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
  } => !stream.timepoint
  %12:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
  %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%14, %13) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__timepoint = util.global.load @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = stream.timepoint.join max(%__constant_tensor_4x4xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint) => !stream.timepoint
  %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) await(%11) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
    %15:2 = stream.async.concurrent with(%arg0 as %arg2: !stream.resource<constant>{%3}, %arg1 as %arg3: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
      %16 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%arg2[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
      %17 = stream.async.transfer %arg3 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
      stream.yield %16, %17 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
    }
    stream.yield %15#0, %15#1 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
  } => !stream.timepoint
  %12:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
  %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%14, %13) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_4x4xi32__timepoint = util.global.load @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = stream.timepoint.join max(%__constant_tensor_4x4xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint) => !stream.timepoint
  %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) await(%11) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
    %15:2 = stream.async.concurrent with(%arg0 as %arg2: !stream.resource<constant>{%3}, %arg1 as %arg3: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
      %16 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%arg2[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
      %17 = stream.async.transfer %arg3 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
      stream.yield %16, %17 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
    }
    stream.yield %15#0, %15#1 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
  } => !stream.timepoint
  %12:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
  %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%14, %13) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private mutable @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
      %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
      %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
      stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
    } => !stream.timepoint
    util.global.store %results#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %result_timepoint, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.global.store %results#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.global.store %result_timepoint, @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
    util.return
  }
  util.global private mutable @__constant_tensor_4x4xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_4x4xi32__timepoint = util.global.load @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.timepoint.join max(%__constant_tensor_4x4xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint) => !stream.timepoint
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) await(%11) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
      %15:2 = stream.async.concurrent with(%arg0 as %arg2: !stream.resource<constant>{%3}, %arg1 as %arg3: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
        %16 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%arg2[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
        %17 = stream.async.transfer %arg3 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
        stream.yield %16, %17 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
      }
      stream.yield %15#0, %15#1 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
    } => !stream.timepoint
    %12:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
    %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %13) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
      %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
      %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
      stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
    } => !stream.timepoint
    util.global.store %results#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %result_timepoint, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.global.store %results#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.global.store %result_timepoint, @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
    util.return
  }
  util.global private @__constant_tensor_4x4xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_4x4xi32__timepoint = util.global.load immutable @__constant_tensor_4x4xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.timepoint.join max(%__constant_tensor_4x4xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint) => !stream.timepoint
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) await(%11) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
      %15:2 = stream.async.concurrent with(%arg0 as %arg2: !stream.resource<constant>{%3}, %arg1 as %arg3: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
        %16 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%arg2[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
        %17 = stream.async.transfer %arg3 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
        stream.yield %16, %17 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
      }
      stream.yield %15#0, %15#1 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
    } => !stream.timepoint
    %12:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
    %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %13) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
      %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
      %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
      stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
    } => !stream.timepoint
    util.global.store %results#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %result_timepoint, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.global.store %results#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32__timepoint_0 = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint_0) => !stream.timepoint
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) await(%11) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
      %15:2 = stream.async.concurrent with(%arg0 as %arg2: !stream.resource<constant>{%3}, %arg1 as %arg3: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
        %16 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%arg2[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
        %17 = stream.async.transfer %arg3 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
        stream.yield %16, %17 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
      }
      stream.yield %15#0, %15#1 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
    } => !stream.timepoint
    %12:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
    %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %13) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
      %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
      %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
      stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
    } => !stream.timepoint
    util.global.store %results#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %result_timepoint, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.global.store %results#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32__timepoint_0 = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint_0) => !stream.timepoint
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) await(%11) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
      %15:2 = stream.async.concurrent with(%arg0 as %arg2: !stream.resource<constant>{%3}, %arg1 as %arg3: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
        %16 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%arg2[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
        %17 = stream.async.transfer %arg3 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
        stream.yield %16, %17 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
      }
      stream.yield %15#0, %15#1 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
    } => !stream.timepoint
    %12:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
    %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %13) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After VerifyLoweringToAsyncPass (iree-stream-verify-lowering-to-async) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) with() -> (!stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}) {
      %cst = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>
      %cst_0 = stream.async.constant : !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
      stream.yield %cst, %cst_0 : !stream.resource<constant>{%c64}, !stream.resource<constant>{%c64}
    } => !stream.timepoint
    util.global.store %results#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %result_timepoint, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.global.store %results#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32__timepoint_0 = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint_0) => !stream.timepoint
    %results:2, %result_timepoint = stream.async.execute on(#hal.device.affinity<@__device_0>) await(%11) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
      %15:2 = stream.async.concurrent with(%arg0 as %arg2: !stream.resource<constant>{%3}, %arg1 as %arg3: !stream.resource<constant>{%c64}) -> (!stream.resource<external>{%10}, !stream.resource<external>{%c64}) {
        %16 = stream.async.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%arg2[%c0 to %3 for %3], %0, %1, %5, %6, %4) : (!stream.resource<constant>{%3}, index, index, index, index, index) -> !stream.resource<external>{%10}
        %17 = stream.async.transfer %arg3 : !stream.resource<constant>{%c64} from(#hal.device.affinity<@__device_0>) -> !stream.resource<external>{%c64}
        stream.yield %16, %17 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
      }
      stream.yield %15#0, %15#1 : !stream.resource<external>{%10}, !stream.resource<external>{%c64}
    } => !stream.timepoint
    %12:2 = stream.timepoint.await %result_timepoint => %results#1, %results#0 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
    %13 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %14 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %12#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%14, %13) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ScheduleAllocationPass (iree-stream-schedule-allocation) //----- //
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c64 = arith.constant 64 : index
    %results:2, %result_timepoint = stream.resource.constants on(#hal.device.affinity<@__device_0>) : 
      !stream.resource<constant>{%c64} = dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
      !stream.resource<constant>{%c64} = dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>
      => !stream.timepoint
    %0 = stream.cmd.execute once on(#hal.device.affinity<@__device_0>) with() {
    } => !stream.timepoint
    %1 = stream.timepoint.join max(%result_timepoint, %0) => !stream.timepoint
    util.global.store %results#0, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %1, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.global.store %results#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32__timepoint_0 = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint_0) => !stream.timepoint
    %c0_1 = arith.constant 0 : index
    %12:3 = stream.resource.pack on(#hal.device.affinity<@__device_0>) slices({
      [0, 0] = %10,
      [0, 0] = %c64
    }) : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%11) => !stream.resource<external>{%12#0} => !stream.timepoint
    %13 = stream.resource.subview %result[%12#1] : !stream.resource<external>{%12#0} -> !stream.resource<external>{%10}
    %14 = stream.resource.subview %result[%12#2] : !stream.resource<external>{%12#0} -> !stream.resource<external>{%c64}
    %15 = stream.timepoint.join max(%11, %result_timepoint) => !stream.timepoint
    %16 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%15) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}, %13 as %arg2: !stream.resource<external>{%10}, %14 as %arg3: !stream.resource<external>{%c64}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c0 for %3] : !stream.resource<constant>{%3},
          wo %arg2[%c0_1 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg1[%c0_1], %arg3[%c0_1], %c64 : !stream.resource<constant>{%c64} -> !stream.resource<external>{%c64}
        stream.cmd.flush %arg3[%c0_1 for %c64] : !stream.resource<external>{%c64}
      }
    } => !stream.timepoint
    %17:2 = stream.timepoint.await %16 => %14, %13 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After PackConstantsPass (iree-stream-pack-constants) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After LayoutSlicesPass (iree-stream-layout-slices) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After PackConstantsPass (iree-stream-pack-constants) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  %c0_i64 = arith.constant 0 : i64
  %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
    scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
  } else {
    %6 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %7 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %6[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    scf.yield %7, %6 : !stream.timepoint, !stream.resource<constant>
  }
  %2 = stream.resource.subview %1#1[%c0] : !stream.resource<constant>{%c128} -> !stream.resource<constant>{%c64}
  %3 = stream.resource.subview %1#1[%c64] : !stream.resource<constant>{%c128} -> !stream.resource<constant>{%c64}
  %4 = stream.cmd.execute once on(#hal.device.affinity<@__device_0>) with() {
  } => !stream.timepoint
  %5 = stream.timepoint.join max(%1#0, %4) => !stream.timepoint
  util.global.store %2, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %5, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.global.store %3, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After LayoutSlicesPass (iree-stream-layout-slices) //----- //
util.initializer {
  %c64 = arith.constant 64 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  %c0_i64 = arith.constant 0 : i64
  %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
    scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
  } else {
    %6 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %7 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %6[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    scf.yield %7, %6 : !stream.timepoint, !stream.resource<constant>
  }
  %2 = stream.resource.subview %1#1[%c0] : !stream.resource<constant>{%c128} -> !stream.resource<constant>{%c64}
  %3 = stream.resource.subview %1#1[%c64] : !stream.resource<constant>{%c128} -> !stream.resource<constant>{%c64}
  %4 = stream.cmd.execute once on(#hal.device.affinity<@__device_0>) with() {
  } => !stream.timepoint
  %5 = stream.timepoint.join max(%1#0, %4) => !stream.timepoint
  util.global.store %2, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %5, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.global.store %3, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.return
}

// -----// IR Dump After PackConstantsPass (iree-stream-pack-constants) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32__timepoint_0 = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint_0) => !stream.timepoint
  %c0_1 = arith.constant 0 : index
  %12:3 = stream.resource.pack on(#hal.device.affinity<@__device_0>) slices({
    [0, 0] = %10,
    [0, 0] = %c64
  }) : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%11) => !stream.resource<external>{%12#0} => !stream.timepoint
  %13 = stream.resource.subview %result[%12#1] : !stream.resource<external>{%12#0} -> !stream.resource<external>{%10}
  %14 = stream.resource.subview %result[%12#2] : !stream.resource<external>{%12#0} -> !stream.resource<external>{%c64}
  %15 = stream.timepoint.join max(%11, %result_timepoint) => !stream.timepoint
  %16 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%15) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}, %13 as %arg2: !stream.resource<external>{%10}, %14 as %arg3: !stream.resource<external>{%c64}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c0 for %3] : !stream.resource<constant>{%3},
        wo %arg2[%c0_1 for %10] : !stream.resource<external>{%10}
      }
      stream.cmd.copy %arg1[%c0_1], %arg3[%c0_1], %c64 : !stream.resource<constant>{%c64} -> !stream.resource<external>{%c64}
      stream.cmd.flush %arg3[%c0_1 for %c64] : !stream.resource<external>{%c64}
    }
  } => !stream.timepoint
  %17:2 = stream.timepoint.await %16 => %14, %13 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
  %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After LayoutSlicesPass (iree-stream-layout-slices) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32__timepoint_0 = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint_0) => !stream.timepoint
  %c0_1 = arith.constant 0 : index
  %c0_2 = arith.constant 0 : index
  %c64_3 = arith.constant 64 : index
  %c64_4 = arith.constant 64 : index
  %c64_5 = arith.constant 64 : index
  %12 = util.align %10, %c64_5 : index
  %13 = arith.addi %12, %c64_4 : index
  %c64_6 = arith.constant 64 : index
  %c64_7 = arith.constant 64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%11) => !stream.resource<external>{%13} => !stream.timepoint
  %14 = stream.resource.subview %result[%c64_4] : !stream.resource<external>{%13} -> !stream.resource<external>{%10}
  %15 = stream.resource.subview %result[%c0_2] : !stream.resource<external>{%13} -> !stream.resource<external>{%c64}
  %16 = stream.timepoint.join max(%11, %result_timepoint) => !stream.timepoint
  %17 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%16) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%3}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c64}, %14 as %arg2: !stream.resource<external>{%10}, %15 as %arg3: !stream.resource<external>{%c64}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c0 for %3] : !stream.resource<constant>{%3},
        wo %arg2[%c0_1 for %10] : !stream.resource<external>{%10}
      }
      stream.cmd.copy %arg1[%c0_1], %arg3[%c0_1], %c64 : !stream.resource<constant>{%c64} -> !stream.resource<external>{%c64}
      stream.cmd.flush %arg3[%c0_1 for %c64] : !stream.resource<external>{%c64}
    }
  } => !stream.timepoint
  %18:2 = stream.timepoint.await %17 => %15, %14 : !stream.resource<external>{%c64}, !stream.resource<external>{%10}
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %18#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %20 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %18#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%20, %19) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After PropagateSubranges (iree-util-propagate-subranges) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global private mutable @__constant_tensor_2x2x2x2xi32__storage_size : index
  util.global private mutable @__constant_tensor_2x2x2x2xi32__offset : index
  util.global private mutable @__constant_tensor_2x2x2x2xi32__length : index
  util.initializer {
    %c0 = arith.constant 0 : index
    %c64 = arith.constant 64 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %c128 = arith.constant 128 : index
    %c0_0 = arith.constant 0 : index
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0_0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    %c0_i64 = arith.constant 0 : i64
    %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
      scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
    } else {
      %6 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
      %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0_0 for %c128] : !util.buffer{%c128} -> !stream.file
      %7 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %6[%c0_0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
      scf.yield %7, %6 : !stream.timepoint, !stream.resource<constant>
    }
    %2 = stream.resource.subview %1#1[%c0_0] : !stream.resource<constant>{%c128} -> !stream.resource<constant>{%c64}
    %3 = stream.resource.subview %1#1[%c64] : !stream.resource<constant>{%c128} -> !stream.resource<constant>{%c64}
    %4 = stream.cmd.execute once on(#hal.device.affinity<@__device_0>) with() {
    } => !stream.timepoint
    %5 = stream.timepoint.join max(%1#0, %4) => !stream.timepoint
    util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %c128, @__constant_tensor_2x2x2x2xi32__storage_size : index
    util.global.store %c0_0, @__constant_tensor_2x2x2x2xi32__offset : index
    util.global.store %c64, @__constant_tensor_2x2x2x2xi32__length : index
    util.global.store %5, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.global.store %1#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.global.store %c128, @__constant_tensor_4x4xi32__storage_size : index
    util.global.store %c64, @__constant_tensor_4x4xi32__offset : index
    util.global.store %c64, @__constant_tensor_4x4xi32__length : index
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global private mutable @__constant_tensor_4x4xi32__storage_size : index
  util.global private mutable @__constant_tensor_4x4xi32__offset : index
  util.global private mutable @__constant_tensor_4x4xi32__length : index
  util.func private @_fully_dynamic_pack_simple() {
    %c0 = arith.constant 0 : index
    %c0_0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32__timepoint_1 = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32__storage_size = util.global.load @__constant_tensor_2x2x2x2xi32__storage_size : index
    %__constant_tensor_2x2x2x2xi32__offset = util.global.load @__constant_tensor_2x2x2x2xi32__offset : index
    %__constant_tensor_2x2x2x2xi32__length = util.global.load @__constant_tensor_2x2x2x2xi32__length : index
    %0 = stream.resource.subview %__constant_tensor_2x2x2x2xi32[%__constant_tensor_2x2x2x2xi32__offset] : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__storage_size} -> !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__length}
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32__storage_size = util.global.load @__constant_tensor_4x4xi32__storage_size : index
    %__constant_tensor_4x4xi32__offset = util.global.load @__constant_tensor_4x4xi32__offset : index
    %__constant_tensor_4x4xi32__length = util.global.load @__constant_tensor_4x4xi32__length : index
    %1 = stream.resource.subview %__constant_tensor_4x4xi32[%__constant_tensor_4x4xi32__offset] : !stream.resource<constant>{%__constant_tensor_4x4xi32__storage_size} -> !stream.resource<constant>{%__constant_tensor_4x4xi32__length}
    %2 = util.optimization_barrier %c4 : index
    %3 = util.optimization_barrier %c4 : index
    %4 = arith.muli %2, %c4 : index
    %5 = arith.muli %4, %3 : index
    %6 = util.optimization_barrier %c2 : index
    %7 = arith.ceildivui %2, %6 : index
    %8 = arith.ceildivui %3, %6 : index
    %9 = arith.muli %7, %c4 : index
    %10 = arith.muli %9, %8 : index
    %11 = arith.muli %10, %6 : index
    %12 = arith.muli %11, %6 : index
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint_1) => !stream.timepoint
    %c0_2 = arith.constant 0 : index
    %c0_3 = arith.constant 0 : index
    %c64_4 = arith.constant 64 : index
    %c64_5 = arith.constant 64 : index
    %c64_6 = arith.constant 64 : index
    %14 = util.align %12, %c64_6 : index
    %15 = arith.addi %14, %c64_5 : index
    %c64_7 = arith.constant 64 : index
    %c64_8 = arith.constant 64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%13) => !stream.resource<external>{%15} => !stream.timepoint
    %16 = stream.resource.subview %result[%c64_5] : !stream.resource<external>{%15} -> !stream.resource<external>{%12}
    %17 = stream.resource.subview %result[%c0_3] : !stream.resource<external>{%15} -> !stream.resource<external>{%c64}
    %18 = stream.timepoint.join max(%13, %result_timepoint) => !stream.timepoint
    %19 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%18) => with(%1 as %arg0: !stream.resource<constant>{%5}, %0 as %arg1: !stream.resource<constant>{%c64}, %16 as %arg2: !stream.resource<external>{%12}, %17 as %arg3: !stream.resource<external>{%c64}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%2, %3, %7, %8, %6](%2, %3, %7, %8, %6 : index, index, index, index, index) {
          ro %arg0[%c0_0 for %5] : !stream.resource<constant>{%5},
          wo %arg2[%c0_2 for %12] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg1[%c0_2], %arg3[%c0_2], %c64 : !stream.resource<constant>{%c64} -> !stream.resource<external>{%c64}
        stream.cmd.flush %arg3[%c0_2 for %c64] : !stream.resource<external>{%c64}
      }
    } => !stream.timepoint
    %20:2 = stream.timepoint.await %19 => %17, %16 : !stream.resource<external>{%c64}, !stream.resource<external>{%12}
    %21 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %20#0 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %22 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %20#1 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%22, %21) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32__timepoint_0 = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_2x2x2x2xi32__storage_size = util.global.load @__constant_tensor_2x2x2x2xi32__storage_size : index
  %__constant_tensor_2x2x2x2xi32__offset = util.global.load @__constant_tensor_2x2x2x2xi32__offset : index
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__storage_size = util.global.load @__constant_tensor_4x4xi32__storage_size : index
  %__constant_tensor_4x4xi32__offset = util.global.load @__constant_tensor_4x4xi32__offset : index
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint_0) => !stream.timepoint
  %12 = util.align %10, %c64 : index
  %13 = arith.addi %12, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%11) => !stream.resource<external>{%13} => !stream.timepoint
  %14 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint_0, %result_timepoint) => !stream.timepoint
  %15 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%14) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%__constant_tensor_4x4xi32__storage_size}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__storage_size}, %result as %arg2: !stream.resource<external>{%13}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%__constant_tensor_4x4xi32__offset for %3] : !stream.resource<constant>{%__constant_tensor_4x4xi32__storage_size},
        wo %arg2[%c64 for %10] : !stream.resource<external>{%13}
      }
      stream.cmd.copy %arg1[%__constant_tensor_2x2x2x2xi32__offset], %arg2[%c0], %c64 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__storage_size} -> !stream.resource<external>{%13}
      stream.cmd.flush %arg2[%c0 for %c64] : !stream.resource<external>{%13}
    }
  } => !stream.timepoint
  %16 = stream.timepoint.await %15 => %result : !stream.resource<external>{%13}
  %17 = stream.resource.subview %16[%c0] : !stream.resource<external>{%13} -> !stream.resource<external>{%c64}
  %18 = stream.resource.subview %16[%c64] : !stream.resource<external>{%13} -> !stream.resource<external>{%10}
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %20 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %18 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%20, %19) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %c64 = arith.constant 64 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
    scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
  } else {
    %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
  }
  util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %c128, @__constant_tensor_2x2x2x2xi32__storage_size : index
  util.global.store %c0, @__constant_tensor_2x2x2x2xi32__offset : index
  util.global.store %c64, @__constant_tensor_2x2x2x2xi32__length : index
  util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.global.store %1#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %c128, @__constant_tensor_4x4xi32__storage_size : index
  util.global.store %c64, @__constant_tensor_4x4xi32__offset : index
  util.global.store %c64, @__constant_tensor_4x4xi32__length : index
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %c64 = arith.constant 64 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
    scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
  } else {
    %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
  }
  util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %c128, @__constant_tensor_2x2x2x2xi32__storage_size : index
  util.global.store %c0, @__constant_tensor_2x2x2x2xi32__offset : index
  util.global.store %c64, @__constant_tensor_2x2x2x2xi32__length : index
  util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.global.store %1#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %c128, @__constant_tensor_4x4xi32__storage_size : index
  util.global.store %c64, @__constant_tensor_4x4xi32__offset : index
  util.global.store %c64, @__constant_tensor_4x4xi32__length : index
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %c64 = arith.constant 64 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
    scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
  } else {
    %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
  }
  util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %c64, @__constant_tensor_2x2x2x2xi32__length : index
  util.global.store %c0, @__constant_tensor_2x2x2x2xi32__offset : index
  util.global.store %c128, @__constant_tensor_2x2x2x2xi32__storage_size : index
  util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.global.store %1#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global.store %c64, @__constant_tensor_4x4xi32__length : index
  util.global.store %c64, @__constant_tensor_4x4xi32__offset : index
  util.global.store %c128, @__constant_tensor_4x4xi32__storage_size : index
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_2x2x2x2xi32__storage_size = util.global.load @__constant_tensor_2x2x2x2xi32__storage_size : index
  %__constant_tensor_2x2x2x2xi32__offset = util.global.load @__constant_tensor_2x2x2x2xi32__offset : index
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32__storage_size = util.global.load @__constant_tensor_4x4xi32__storage_size : index
  %__constant_tensor_4x4xi32__offset = util.global.load @__constant_tensor_4x4xi32__offset : index
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint) => !stream.timepoint
  %12 = util.align %10, %c64 : index
  %13 = arith.addi %12, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%11) => !stream.resource<external>{%13} => !stream.timepoint
  %14 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
  %15 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%14) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%__constant_tensor_4x4xi32__storage_size}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__storage_size}, %result as %arg2: !stream.resource<external>{%13}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%__constant_tensor_4x4xi32__offset for %3] : !stream.resource<constant>{%__constant_tensor_4x4xi32__storage_size},
        wo %arg2[%c64 for %10] : !stream.resource<external>{%13}
      }
      stream.cmd.copy %arg1[%__constant_tensor_2x2x2x2xi32__offset], %arg2[%c0], %c64 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__storage_size} -> !stream.resource<external>{%13}
      stream.cmd.flush %arg2[%c0 for %c64] : !stream.resource<external>{%13}
    }
  } => !stream.timepoint
  %16 = stream.timepoint.await %15 => %result : !stream.resource<external>{%13}
  %17 = stream.resource.subview %16[%c0] : !stream.resource<external>{%13} -> !stream.resource<external>{%c64}
  %18 = stream.resource.subview %16[%c64] : !stream.resource<external>{%13} -> !stream.resource<external>{%10}
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %20 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %18 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%20, %19) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_4x4xi32__storage_size = util.global.load @__constant_tensor_4x4xi32__storage_size : index
  %__constant_tensor_4x4xi32__offset = util.global.load @__constant_tensor_4x4xi32__offset : index
  %__constant_tensor_2x2x2x2xi32__storage_size = util.global.load @__constant_tensor_2x2x2x2xi32__storage_size : index
  %__constant_tensor_2x2x2x2xi32__offset = util.global.load @__constant_tensor_2x2x2x2xi32__offset : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint) => !stream.timepoint
  %12 = util.align %10, %c64 : index
  %13 = arith.addi %12, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%11) => !stream.resource<external>{%13} => !stream.timepoint
  %14 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
  %15 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%14) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%__constant_tensor_4x4xi32__storage_size}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__storage_size}, %result as %arg2: !stream.resource<external>{%13}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%__constant_tensor_4x4xi32__offset for %3] : !stream.resource<constant>{%__constant_tensor_4x4xi32__storage_size},
        wo %arg2[%c64 for %10] : !stream.resource<external>{%13}
      }
      stream.cmd.copy %arg1[%__constant_tensor_2x2x2x2xi32__offset], %arg2[%c0], %c64 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__storage_size} -> !stream.resource<external>{%13}
      stream.cmd.flush %arg2[%c0 for %c64] : !stream.resource<external>{%13}
    }
  } => !stream.timepoint
  %16 = stream.timepoint.await %15 => %result : !stream.resource<external>{%13}
  %17 = stream.resource.subview %16[%c0] : !stream.resource<external>{%13} -> !stream.resource<external>{%c64}
  %18 = stream.resource.subview %16[%c64] : !stream.resource<external>{%13} -> !stream.resource<external>{%10}
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %20 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %18 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%20, %19) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global private mutable @__constant_tensor_2x2x2x2xi32__storage_size = 128 : index
  util.global private mutable @__constant_tensor_2x2x2x2xi32__offset = 0 : index
  util.global private mutable @__constant_tensor_2x2x2x2xi32__length = 64 : index
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
      scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
    } else {
      %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
      %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
      %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
      scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
    }
    util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.global.store %1#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.global private mutable @__constant_tensor_4x4xi32__storage_size = 128 : index
  util.global private mutable @__constant_tensor_4x4xi32__offset = 64 : index
  util.global private mutable @__constant_tensor_4x4xi32__length = 64 : index
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %__constant_tensor_4x4xi32__storage_size = util.global.load @__constant_tensor_4x4xi32__storage_size : index
    %__constant_tensor_4x4xi32__offset = util.global.load @__constant_tensor_4x4xi32__offset : index
    %__constant_tensor_2x2x2x2xi32__storage_size = util.global.load @__constant_tensor_2x2x2x2xi32__storage_size : index
    %__constant_tensor_2x2x2x2xi32__offset = util.global.load @__constant_tensor_2x2x2x2xi32__offset : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%__constant_tensor_4x4xi32__storage_size}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__storage_size}, %result as %arg2: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%__constant_tensor_4x4xi32__offset for %3] : !stream.resource<constant>{%__constant_tensor_4x4xi32__storage_size},
          wo %arg2[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg1[%__constant_tensor_2x2x2x2xi32__offset], %arg2[%c0], %c64 : !stream.resource<constant>{%__constant_tensor_2x2x2x2xi32__storage_size} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg2[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
      scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
    } else {
      %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
      %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
      %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
      scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
    }
    util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.global.store %1#1, @__constant_tensor_4x4xi32 : !stream.resource<constant>
    util.return
  }
  util.global private @__constant_tensor_4x4xi32 : !stream.resource<constant>
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_4x4xi32 = util.global.load immutable @__constant_tensor_4x4xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_4x4xi32 as %arg0: !stream.resource<constant>{%c128}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c128}, %result as %arg2: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg2[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg1[%c0], %arg2[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg2[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
      scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
    } else {
      %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
      %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
      %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
      scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
    }
    util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32_0 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32_0 as %arg0: !stream.resource<constant>{%c128}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c128}, %result as %arg2: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg2[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg1[%c0], %arg2[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg2[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
      scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
    } else {
      %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
      %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
      %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
      scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
    }
    util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32_0 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32_0 as %arg0: !stream.resource<constant>{%c128}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c128}, %result as %arg2: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg2[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg1[%c0], %arg2[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg2[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After VerifyLoweringToCmdPass (iree-stream-verify-lowering-to-cmd) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
      scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
    } else {
      %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
      %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
      %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
      scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
    }
    util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %__constant_tensor_2x2x2x2xi32_0 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32_0 as %arg0: !stream.resource<constant>{%c128}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c128}, %result as %arg2: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg2[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg1[%c0], %arg2[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg2[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
    scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
  } else {
    %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
  }
  util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
    scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
  } else {
    %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
  }
  util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %c128 = arith.constant 128 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %__constant_tensor_2x2x2x2xi32_0 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = util.align %10, %c64 : index
  %12 = arith.addi %11, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
  %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
  %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32_0 as %arg0: !stream.resource<constant>{%c128}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c128}, %result as %arg2: !stream.resource<external>{%12}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
        wo %arg2[%c64 for %10] : !stream.resource<external>{%12}
      }
      stream.cmd.copy %arg1[%c0], %arg2[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
      stream.cmd.flush %arg2[%c0 for %c64] : !stream.resource<external>{%12}
    }
  } => !stream.timepoint
  %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
  %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
  %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
  %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
    scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
  } else {
    %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
  }
  util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %c128 = arith.constant 128 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = util.align %10, %c64 : index
  %12 = arith.addi %11, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
  %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
  %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c128}, %result as %arg2: !stream.resource<external>{%12}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
        wo %arg2[%c64 for %10] : !stream.resource<external>{%12}
      }
      stream.cmd.copy %arg1[%c0], %arg2[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
      stream.cmd.flush %arg2[%c0 for %c64] : !stream.resource<external>{%12}
    }
  } => !stream.timepoint
  %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
  %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
  %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
  %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %c128 = arith.constant 128 : index
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = util.align %10, %c64 : index
  %12 = arith.addi %11, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
  %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
  %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %__constant_tensor_2x2x2x2xi32 as %arg1: !stream.resource<constant>{%c128}, %result as %arg2: !stream.resource<external>{%12}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
        wo %arg2[%c64 for %10] : !stream.resource<external>{%12}
      }
      stream.cmd.copy %arg1[%c0], %arg2[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
      stream.cmd.flush %arg2[%c0 for %c64] : !stream.resource<external>{%12}
    }
  } => !stream.timepoint
  %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
  %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
  %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
  %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
      scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
    } else {
      %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
      %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
      %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
      scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
    }
    util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
      scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
    } else {
      %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
      %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
      %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
      scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
    }
    util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
      scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
    } else {
      %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
      %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
      %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
      scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
    }
    util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    %1:2 = scf.if %did_map -> (!stream.timepoint, !stream.resource<constant>) {
      scf.yield %0, %result : !stream.timepoint, !stream.resource<constant>
    } else {
      %2 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
      %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
      %3 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %2[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
      scf.yield %3, %2 : !stream.timepoint, !stream.resource<constant>
    }
    util.global.store %1#1, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %1#0, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After SCFToControlFlow (convert-scf-to-cf) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After SCFToControlFlow (convert-scf-to-cf) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = util.align %10, %c64 : index
  %12 = arith.addi %11, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
  %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
  %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
        wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
    }
  } => !stream.timepoint
  %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
  %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
  %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
  %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SCFToControlFlow (convert-scf-to-cf) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb1, ^bb2
^bb1:  // pred: ^bb0
  cf.br ^bb3(%0, %result : !stream.timepoint, !stream.resource<constant>)
^bb2:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb3(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb3(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb1, ^bb2
  cf.br ^bb4
^bb4:  // pred: ^bb3
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = util.align %10, %c64 : index
  %12 = arith.addi %11, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
  %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
  %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
        wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
    }
  } => !stream.timepoint
  %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
  %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
  %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
  %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = util.align %10, %c64 : index
  %12 = arith.addi %11, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
  %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
  %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
        wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
    }
  } => !stream.timepoint
  %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
  %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
  %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
  %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = util.align %10, %c64 : index
  %12 = arith.addi %11, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
  %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
  %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
        wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
    }
  } => !stream.timepoint
  %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
  %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
  %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
  %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.join max(%__constant_tensor_2x2x2x2xi32__timepoint, %result_timepoint) => !stream.timepoint
    %14 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%13) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %15 = stream.timepoint.await %14 => %result : !stream.resource<external>{%12}
    %16 = stream.resource.subview %15[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %17 = stream.resource.subview %15[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%19, %18) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ElideTimepointsPass (iree-stream-elide-timepoints) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 0 : index, iree.fixedpoint.modified, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.timepoint.immediate => !stream.timepoint
    %14 = stream.timepoint.join max(%13, %result_timepoint) => !stream.timepoint
    %15 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%14) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %16 = stream.timepoint.await %15 => %result : !stream.resource<external>{%12}
    %17 = stream.resource.subview %16[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %18 = stream.resource.subview %16[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %19 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %17 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %20 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %18 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%20, %19) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %c128 = arith.constant 128 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = util.align %10, %c64 : index
  %12 = arith.addi %11, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
  %13 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
        wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
    }
  } => !stream.timepoint
  %14 = stream.timepoint.await %13 => %result : !stream.resource<external>{%12}
  %15 = stream.resource.subview %14[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
  %16 = stream.resource.subview %14[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
  %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%18, %17) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %c128 = arith.constant 128 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = util.align %10, %c64 : index
  %12 = arith.addi %11, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
  %13 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
        wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
    }
  } => !stream.timepoint
  %14 = stream.timepoint.await %13 => %result : !stream.resource<external>{%12}
  %15 = stream.resource.subview %14[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
  %16 = stream.resource.subview %14[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
  %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%18, %17) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %c128 = arith.constant 128 : index
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = arith.muli %0, %c4 : index
  %3 = arith.muli %2, %1 : index
  %4 = util.optimization_barrier %c2 : index
  %5 = arith.ceildivui %0, %4 : index
  %6 = arith.ceildivui %1, %4 : index
  %7 = arith.muli %5, %c4 : index
  %8 = arith.muli %7, %6 : index
  %9 = arith.muli %8, %4 : index
  %10 = arith.muli %9, %4 : index
  %11 = util.align %10, %c64 : index
  %12 = arith.addi %11, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
  %13 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
        ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
        wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
    }
  } => !stream.timepoint
  %14 = stream.timepoint.await %13 => %result : !stream.resource<external>{%12}
  %15 = stream.resource.subview %14[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
  %16 = stream.resource.subview %14[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
  %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%18, %17) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 1 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %14 = stream.timepoint.await %13 => %result : !stream.resource<external>{%12}
    %15 = stream.resource.subview %14[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %16 = stream.resource.subview %14[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%18, %17) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 1 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %14 = stream.timepoint.await %13 => %result : !stream.resource<external>{%12}
    %15 = stream.resource.subview %14[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %16 = stream.resource.subview %14[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%18, %17) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 1 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %14 = stream.timepoint.await %13 => %result : !stream.resource<external>{%12}
    %15 = stream.resource.subview %14[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %16 = stream.resource.subview %14[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%18, %17) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 1 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %14 = stream.timepoint.await %13 => %result : !stream.resource<external>{%12}
    %15 = stream.resource.subview %14[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %16 = stream.resource.subview %14[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%18, %17) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ElideTimepointsPass (iree-stream-elide-timepoints) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {iree.fixedpoint.iteration = 1 : index, stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %14 = stream.timepoint.await %13 => %result : !stream.resource<external>{%12}
    %15 = stream.resource.subview %14[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %16 = stream.resource.subview %14[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%18, %17) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FixedPointIterator (iree-util-fixed-point-iterator) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: index, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: !stream.binding) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg1, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg2, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg3, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg4, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg5, 4 : index
        %5 = stream.binding.subspan %arg0[%c0] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg6[%c0] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %13 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%0, %1, %5, %6, %4 : index, index, index, index, index) {
          ro %arg0[%c64 for %3] : !stream.resource<constant>{%c128},
          wo %arg1[%c64 for %10] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %14 = stream.timepoint.await %13 => %result : !stream.resource<external>{%12}
    %15 = stream.resource.subview %14[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %16 = stream.resource.subview %14[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%18, %17) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseDispatchBindingsPass (iree-stream-fuse-dispatch-bindings) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding, %arg1: !stream.binding, %arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: index, %arg7: index, %arg8: index) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg4, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg5, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg6, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg7, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg8, 4 : index
        %5 = stream.binding.subspan %arg0[%arg2] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg1[%arg3] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %c0_0 = arith.constant 0 : index
    %13 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%c64, %c64, %0, %1, %5, %6, %4 : index, index, index, index, index, index, index) {
          ro %arg0[%c0_0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0_0 for %12] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %14 = stream.timepoint.await %13 => %result : !stream.resource<external>{%12}
    %15 = stream.resource.subview %14[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %16 = stream.resource.subview %14[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%18, %17) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After AnnotateDispatchArgumentsPass (iree-stream-annotate-dispatch-arguments) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: index {stream.alignment = 64 : index, stream.values = [64 : index]}, %arg3: index {stream.alignment = 64 : index, stream.values = [64 : index]}, %arg4: index, %arg5: index, %arg6: index, %arg7: index, %arg8: index) {
        %c0 = arith.constant 0 : index
        %0 = flow.dispatch.workload.ordinal %arg4, 0 : index
        %1 = flow.dispatch.workload.ordinal %arg5, 1 : index
        %2 = flow.dispatch.workload.ordinal %arg6, 2 : index
        %3 = flow.dispatch.workload.ordinal %arg7, 3 : index
        %4 = flow.dispatch.workload.ordinal %arg8, 4 : index
        %5 = stream.binding.subspan %arg0[%arg2] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1}
        %6 = stream.binding.subspan %arg1[%arg3] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        %7 = flow.dispatch.tensor.load %5, offsets = [0, 0], sizes = [%0, %1], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%0, %1} -> tensor<?x?xi32>
        %8 = tensor.empty(%2, %3, %4, %4) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %7 inner_dims_pos = [0, 1] inner_tiles = [%4, %4] into %8 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %6, offsets = [0, 0, 0, 0], sizes = [%2, %3, %4, %4], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%2, %3, %4, %4}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %c0_0 = arith.constant 0 : index
    %13 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%c64, %c64, %0, %1, %5, %6, %4 : index, index, index, index, index, index, index) {
          ro %arg0[%c0_0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0_0 for %12] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %14 = stream.timepoint.await %13 => %result : !stream.resource<external>{%12}
    %15 = stream.resource.subview %14[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %16 = stream.resource.subview %14[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %17 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %15 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %18 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %16 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%18, %17) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After PackDispatchOperandsPass (iree-stream-pack-dispatch-operands) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32, %arg12: i32, %arg13: i32, %arg14: i32, %arg15: i32) {
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %c32_i64 = arith.constant 32 : i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %c32_i64_0 = arith.constant 32 : i64
        %7 = arith.shli %6, %c32_i64_0 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %c32_i64_1 = arith.constant 32 : i64
        %12 = arith.shli %11, %c32_i64_1 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %c32_i64_2 = arith.constant 32 : i64
        %17 = arith.shli %16, %c32_i64_2 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %c32_i64_3 = arith.constant 32 : i64
        %22 = arith.shli %21, %c32_i64_3 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = arith.extui %arg12 : i32 to i64
        %26 = arith.extui %arg13 : i32 to i64
        %c32_i64_4 = arith.constant 32 : i64
        %27 = arith.shli %26, %c32_i64_4 : i64
        %28 = arith.ori %25, %27 : i64
        %29 = arith.index_castui %28 : i64 to index
        %30 = arith.extui %arg14 : i32 to i64
        %31 = arith.extui %arg15 : i32 to i64
        %c32_i64_5 = arith.constant 32 : i64
        %32 = arith.shli %31, %c32_i64_5 : i64
        %33 = arith.ori %30, %32 : i64
        %34 = arith.index_castui %33 : i64 to index
        %c0 = arith.constant 0 : index
        %35 = flow.dispatch.workload.ordinal %14, 0 : index
        %36 = flow.dispatch.workload.ordinal %19, 1 : index
        %37 = flow.dispatch.workload.ordinal %24, 2 : index
        %38 = flow.dispatch.workload.ordinal %29, 3 : index
        %39 = flow.dispatch.workload.ordinal %34, 4 : index
        %40 = stream.binding.subspan %arg0[%4] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36}
        %41 = stream.binding.subspan %arg1[%9] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        %42 = flow.dispatch.tensor.load %40, offsets = [0, 0], sizes = [%35, %36], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36} -> tensor<?x?xi32>
        %43 = tensor.empty(%37, %38, %39, %39) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %42 inner_dims_pos = [0, 1] inner_tiles = [%39, %39] into %43 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %41, offsets = [0, 0, 0, 0], sizes = [%37, %38, %39, %39], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = arith.muli %0, %c4 : index
    %3 = arith.muli %2, %1 : index
    %4 = util.optimization_barrier %c2 : index
    %5 = arith.ceildivui %0, %4 : index
    %6 = arith.ceildivui %1, %4 : index
    %7 = arith.muli %5, %c4 : index
    %8 = arith.muli %7, %6 : index
    %9 = arith.muli %8, %4 : index
    %10 = arith.muli %9, %4 : index
    %11 = util.align %10, %c64 : index
    %12 = arith.addi %11, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%12} => !stream.timepoint
    %c0_0 = arith.constant 0 : index
    %c64_i64 = arith.constant 64 : i64
    %c64_i32 = arith.constant 64 : i32
    %c32_i64 = arith.constant 32 : i64
    %c0_i64 = arith.constant 0 : i64
    %c0_i32 = arith.constant 0 : i32
    %c64_i64_1 = arith.constant 64 : i64
    %c64_i32_2 = arith.constant 64 : i32
    %c32_i64_3 = arith.constant 32 : i64
    %c0_i64_4 = arith.constant 0 : i64
    %c0_i32_5 = arith.constant 0 : i32
    %13 = arith.index_castui %0 : index to i64
    %14 = arith.trunci %13 : i64 to i32
    %c32_i64_6 = arith.constant 32 : i64
    %15 = arith.shrui %13, %c32_i64_6 : i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.index_castui %1 : index to i64
    %18 = arith.trunci %17 : i64 to i32
    %c32_i64_7 = arith.constant 32 : i64
    %19 = arith.shrui %17, %c32_i64_7 : i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.index_castui %5 : index to i64
    %22 = arith.trunci %21 : i64 to i32
    %c32_i64_8 = arith.constant 32 : i64
    %23 = arith.shrui %21, %c32_i64_8 : i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.index_castui %6 : index to i64
    %26 = arith.trunci %25 : i64 to i32
    %c32_i64_9 = arith.constant 32 : i64
    %27 = arith.shrui %25, %c32_i64_9 : i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.index_castui %4 : index to i64
    %30 = arith.trunci %29 : i64 to i32
    %c32_i64_10 = arith.constant 32 : i64
    %31 = arith.shrui %29, %c32_i64_10 : i64
    %32 = arith.trunci %31 : i64 to i32
    %33 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%12}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %5, %6, %4](%c64_i32, %c0_i32, %c64_i32_2, %c0_i32_5, %14, %16, %18, %20, %22, %24, %26, %28, %30, %32 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0_0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0_0 for %12] : !stream.resource<external>{%12}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%12}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%12}
      }
    } => !stream.timepoint
    %34 = stream.timepoint.await %33 => %result : !stream.resource<external>{%12}
    %35 = stream.resource.subview %34[%c0] : !stream.resource<external>{%12} -> !stream.resource<external>{%c64}
    %36 = stream.resource.subview %34[%c64] : !stream.resource<external>{%12} -> !stream.resource<external>{%10}
    %37 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %35 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %38 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %36 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%38, %37) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0_i32 = arith.constant 0 : i32
  %c32_i64 = arith.constant 32 : i64
  %c64_i32 = arith.constant 64 : i32
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c2 : index
  %3 = arith.ceildivui %0, %2 : index
  %4 = arith.ceildivui %1, %2 : index
  %5 = arith.muli %3, %c4 : index
  %6 = arith.muli %5, %4 : index
  %7 = arith.muli %6, %2 : index
  %8 = arith.muli %7, %2 : index
  %9 = util.align %8, %c64 : index
  %10 = arith.addi %9, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
  %11 = arith.index_castui %0 : index to i64
  %12 = arith.trunci %11 : i64 to i32
  %13 = arith.shrui %11, %c32_i64 : i64
  %14 = arith.trunci %13 : i64 to i32
  %15 = arith.index_castui %1 : index to i64
  %16 = arith.trunci %15 : i64 to i32
  %17 = arith.shrui %15, %c32_i64 : i64
  %18 = arith.trunci %17 : i64 to i32
  %19 = arith.index_castui %3 : index to i64
  %20 = arith.trunci %19 : i64 to i32
  %21 = arith.shrui %19, %c32_i64 : i64
  %22 = arith.trunci %21 : i64 to i32
  %23 = arith.index_castui %4 : index to i64
  %24 = arith.trunci %23 : i64 to i32
  %25 = arith.shrui %23, %c32_i64 : i64
  %26 = arith.trunci %25 : i64 to i32
  %27 = arith.index_castui %2 : index to i64
  %28 = arith.trunci %27 : i64 to i32
  %29 = arith.shrui %27, %c32_i64 : i64
  %30 = arith.trunci %29 : i64 to i32
  %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%c64_i32, %c0_i32, %c64_i32, %c0_i32, %12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
        ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
        wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
    }
  } => !stream.timepoint
  %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
  %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
  %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
  %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c0_i32 = arith.constant 0 : i32
  %c32_i64 = arith.constant 32 : i64
  %c64_i32 = arith.constant 64 : i32
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c2 : index
  %3 = arith.ceildivui %0, %2 : index
  %4 = arith.ceildivui %1, %2 : index
  %5 = arith.muli %3, %c4 : index
  %6 = arith.muli %5, %4 : index
  %7 = arith.muli %6, %2 : index
  %8 = arith.muli %7, %2 : index
  %9 = util.align %8, %c64 : index
  %10 = arith.addi %9, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
  %11 = arith.index_castui %0 : index to i64
  %12 = arith.trunci %11 : i64 to i32
  %13 = arith.shrui %11, %c32_i64 : i64
  %14 = arith.trunci %13 : i64 to i32
  %15 = arith.index_castui %1 : index to i64
  %16 = arith.trunci %15 : i64 to i32
  %17 = arith.shrui %15, %c32_i64 : i64
  %18 = arith.trunci %17 : i64 to i32
  %19 = arith.index_castui %3 : index to i64
  %20 = arith.trunci %19 : i64 to i32
  %21 = arith.shrui %19, %c32_i64 : i64
  %22 = arith.trunci %21 : i64 to i32
  %23 = arith.index_castui %4 : index to i64
  %24 = arith.trunci %23 : i64 to i32
  %25 = arith.shrui %23, %c32_i64 : i64
  %26 = arith.trunci %25 : i64 to i32
  %27 = arith.index_castui %2 : index to i64
  %28 = arith.trunci %27 : i64 to i32
  %29 = arith.shrui %27, %c32_i64 : i64
  %30 = arith.trunci %29 : i64 to i32
  %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%c64_i32, %c0_i32, %c64_i32, %c0_i32, %12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
        ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
        wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
    }
  } => !stream.timepoint
  %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
  %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
  %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
  %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %c0_i32 = arith.constant 0 : i32
  %c32_i64 = arith.constant 32 : i64
  %c64_i32 = arith.constant 64 : i32
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %c4 = arith.constant 4 : index
  %c2 = arith.constant 2 : index
  %c64 = arith.constant 64 : index
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c2 : index
  %3 = arith.ceildivui %0, %2 : index
  %4 = arith.ceildivui %1, %2 : index
  %5 = arith.muli %3, %c4 : index
  %6 = arith.muli %5, %4 : index
  %7 = arith.muli %6, %2 : index
  %8 = arith.muli %7, %2 : index
  %9 = util.align %8, %c64 : index
  %10 = arith.addi %9, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
  %11 = arith.index_castui %0 : index to i64
  %12 = arith.trunci %11 : i64 to i32
  %13 = arith.shrui %11, %c32_i64 : i64
  %14 = arith.trunci %13 : i64 to i32
  %15 = arith.index_castui %1 : index to i64
  %16 = arith.trunci %15 : i64 to i32
  %17 = arith.shrui %15, %c32_i64 : i64
  %18 = arith.trunci %17 : i64 to i32
  %19 = arith.index_castui %3 : index to i64
  %20 = arith.trunci %19 : i64 to i32
  %21 = arith.shrui %19, %c32_i64 : i64
  %22 = arith.trunci %21 : i64 to i32
  %23 = arith.index_castui %4 : index to i64
  %24 = arith.trunci %23 : i64 to i32
  %25 = arith.shrui %23, %c32_i64 : i64
  %26 = arith.trunci %25 : i64 to i32
  %27 = arith.index_castui %2 : index to i64
  %28 = arith.trunci %27 : i64 to i32
  %29 = arith.shrui %27, %c32_i64 : i64
  %30 = arith.trunci %29 : i64 to i32
  %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%c64_i32, %c0_i32, %c64_i32, %c0_i32, %12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
        ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
        wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
    }
  } => !stream.timepoint
  %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
  %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
  %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
  %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32, %arg12: i32, %arg13: i32, %arg14: i32, %arg15: i32) {
        %c32_i64 = arith.constant 32 : i64
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = arith.extui %arg12 : i32 to i64
        %26 = arith.extui %arg13 : i32 to i64
        %27 = arith.shli %26, %c32_i64 : i64
        %28 = arith.ori %25, %27 : i64
        %29 = arith.index_castui %28 : i64 to index
        %30 = arith.extui %arg14 : i32 to i64
        %31 = arith.extui %arg15 : i32 to i64
        %32 = arith.shli %31, %c32_i64 : i64
        %33 = arith.ori %30, %32 : i64
        %34 = arith.index_castui %33 : i64 to index
        %35 = flow.dispatch.workload.ordinal %14, 0 : index
        %36 = flow.dispatch.workload.ordinal %19, 1 : index
        %37 = flow.dispatch.workload.ordinal %24, 2 : index
        %38 = flow.dispatch.workload.ordinal %29, 3 : index
        %39 = flow.dispatch.workload.ordinal %34, 4 : index
        %40 = stream.binding.subspan %arg0[%4] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36}
        %41 = stream.binding.subspan %arg1[%9] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        %42 = flow.dispatch.tensor.load %40, offsets = [0, 0], sizes = [%35, %36], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36} -> tensor<?x?xi32>
        %43 = tensor.empty(%37, %38, %39, %39) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %42 inner_dims_pos = [0, 1] inner_tiles = [%39, %39] into %43 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %41, offsets = [0, 0, 0, 0], sizes = [%37, %38, %39, %39], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %c64_i32 = arith.constant 64 : i32
    %c32_i64 = arith.constant 32 : i64
    %c0_i32 = arith.constant 0 : i32
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%c64_i32, %c0_i32, %c64_i32, %c0_i32, %12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32, %arg12: i32, %arg13: i32, %arg14: i32, %arg15: i32) {
        %c32_i64 = arith.constant 32 : i64
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = arith.extui %arg12 : i32 to i64
        %26 = arith.extui %arg13 : i32 to i64
        %27 = arith.shli %26, %c32_i64 : i64
        %28 = arith.ori %25, %27 : i64
        %29 = arith.index_castui %28 : i64 to index
        %30 = arith.extui %arg14 : i32 to i64
        %31 = arith.extui %arg15 : i32 to i64
        %32 = arith.shli %31, %c32_i64 : i64
        %33 = arith.ori %30, %32 : i64
        %34 = arith.index_castui %33 : i64 to index
        %35 = flow.dispatch.workload.ordinal %14, 0 : index
        %36 = flow.dispatch.workload.ordinal %19, 1 : index
        %37 = flow.dispatch.workload.ordinal %24, 2 : index
        %38 = flow.dispatch.workload.ordinal %29, 3 : index
        %39 = flow.dispatch.workload.ordinal %34, 4 : index
        %40 = stream.binding.subspan %arg0[%4] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36}
        %41 = stream.binding.subspan %arg1[%9] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        %42 = flow.dispatch.tensor.load %40, offsets = [0, 0], sizes = [%35, %36], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36} -> tensor<?x?xi32>
        %43 = tensor.empty(%37, %38, %39, %39) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %42 inner_dims_pos = [0, 1] inner_tiles = [%39, %39] into %43 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %41, offsets = [0, 0, 0, 0], sizes = [%37, %38, %39, %39], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %c64_i32 = arith.constant 64 : i32
    %c32_i64 = arith.constant 32 : i64
    %c0_i32 = arith.constant 0 : i32
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%c64_i32, %c0_i32, %c64_i32, %c0_i32, %12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32, %arg12: i32, %arg13: i32, %arg14: i32, %arg15: i32) {
        %c32_i64 = arith.constant 32 : i64
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = arith.extui %arg12 : i32 to i64
        %26 = arith.extui %arg13 : i32 to i64
        %27 = arith.shli %26, %c32_i64 : i64
        %28 = arith.ori %25, %27 : i64
        %29 = arith.index_castui %28 : i64 to index
        %30 = arith.extui %arg14 : i32 to i64
        %31 = arith.extui %arg15 : i32 to i64
        %32 = arith.shli %31, %c32_i64 : i64
        %33 = arith.ori %30, %32 : i64
        %34 = arith.index_castui %33 : i64 to index
        %35 = flow.dispatch.workload.ordinal %14, 0 : index
        %36 = flow.dispatch.workload.ordinal %19, 1 : index
        %37 = flow.dispatch.workload.ordinal %24, 2 : index
        %38 = flow.dispatch.workload.ordinal %29, 3 : index
        %39 = flow.dispatch.workload.ordinal %34, 4 : index
        %40 = stream.binding.subspan %arg0[%4] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36}
        %41 = stream.binding.subspan %arg1[%9] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        %42 = flow.dispatch.tensor.load %40, offsets = [0, 0], sizes = [%35, %36], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36} -> tensor<?x?xi32>
        %43 = tensor.empty(%37, %38, %39, %39) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %42 inner_dims_pos = [0, 1] inner_tiles = [%39, %39] into %43 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %41, offsets = [0, 0, 0, 0], sizes = [%37, %38, %39, %39], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %c64_i32 = arith.constant 64 : i32
    %c32_i64 = arith.constant 32 : i64
    %c0_i32 = arith.constant 0 : i32
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%c64_i32, %c0_i32, %c64_i32, %c0_i32, %12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32, %arg12: i32, %arg13: i32, %arg14: i32, %arg15: i32) {
        %c32_i64 = arith.constant 32 : i64
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = arith.extui %arg12 : i32 to i64
        %26 = arith.extui %arg13 : i32 to i64
        %27 = arith.shli %26, %c32_i64 : i64
        %28 = arith.ori %25, %27 : i64
        %29 = arith.index_castui %28 : i64 to index
        %30 = arith.extui %arg14 : i32 to i64
        %31 = arith.extui %arg15 : i32 to i64
        %32 = arith.shli %31, %c32_i64 : i64
        %33 = arith.ori %30, %32 : i64
        %34 = arith.index_castui %33 : i64 to index
        %35 = flow.dispatch.workload.ordinal %14, 0 : index
        %36 = flow.dispatch.workload.ordinal %19, 1 : index
        %37 = flow.dispatch.workload.ordinal %24, 2 : index
        %38 = flow.dispatch.workload.ordinal %29, 3 : index
        %39 = flow.dispatch.workload.ordinal %34, 4 : index
        %40 = stream.binding.subspan %arg0[%4] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36}
        %41 = stream.binding.subspan %arg1[%9] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        %42 = flow.dispatch.tensor.load %40, offsets = [0, 0], sizes = [%35, %36], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36} -> tensor<?x?xi32>
        %43 = tensor.empty(%37, %38, %39, %39) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %42 inner_dims_pos = [0, 1] inner_tiles = [%39, %39] into %43 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %41, offsets = [0, 0, 0, 0], sizes = [%37, %38, %39, %39], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %c64_i32 = arith.constant 64 : i32
    %c32_i64 = arith.constant 32 : i64
    %c0_i32 = arith.constant 0 : i32
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%c64_i32, %c0_i32, %c64_i32, %c0_i32, %12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldUniformOperandsPass (iree-stream-fold-uniform-operands) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32) {
        %c64_i32 = arith.constant 64 : i32
        %c0_i32 = arith.constant 0 : i32
        %c32_i64 = arith.constant 32 : i64
        %0 = arith.extui %c64_i32 : i32 to i64
        %1 = arith.extui %c0_i32 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %5 = arith.extui %c64_i32 : i32 to i64
        %6 = arith.extui %c0_i32 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 {stream.alignment = 64 : index, stream.values = [64 : index]} : i64 to index
        %10 = arith.extui %arg2 : i32 to i64
        %11 = arith.extui %arg3 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg4 : i32 to i64
        %16 = arith.extui %arg5 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg6 : i32 to i64
        %21 = arith.extui %arg7 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = arith.extui %arg8 : i32 to i64
        %26 = arith.extui %arg9 : i32 to i64
        %27 = arith.shli %26, %c32_i64 : i64
        %28 = arith.ori %25, %27 : i64
        %29 = arith.index_castui %28 : i64 to index
        %30 = arith.extui %arg10 : i32 to i64
        %31 = arith.extui %arg11 : i32 to i64
        %32 = arith.shli %31, %c32_i64 : i64
        %33 = arith.ori %30, %32 : i64
        %34 = arith.index_castui %33 : i64 to index
        %35 = flow.dispatch.workload.ordinal %14, 0 : index
        %36 = flow.dispatch.workload.ordinal %19, 1 : index
        %37 = flow.dispatch.workload.ordinal %24, 2 : index
        %38 = flow.dispatch.workload.ordinal %29, 3 : index
        %39 = flow.dispatch.workload.ordinal %34, 4 : index
        %40 = stream.binding.subspan %arg0[%4] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36}
        %41 = stream.binding.subspan %arg1[%9] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        %42 = flow.dispatch.tensor.load %40, offsets = [0, 0], sizes = [%35, %36], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%35, %36} -> tensor<?x?xi32>
        %43 = tensor.empty(%37, %38, %39, %39) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %42 inner_dims_pos = [0, 1] inner_tiles = [%39, %39] into %43 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %41, offsets = [0, 0, 0, 0], sizes = [%37, %38, %39, %39], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%37, %38, %39, %39}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c64 = arith.constant 64 : index
    %c2 = arith.constant 2 : index
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c128 = arith.constant 128 : index
    %c64_i32 = arith.constant 64 : i32
    %c32_i64 = arith.constant 32 : i64
    %c0_i32 = arith.constant 0 : i32
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
  util.call @_fully_dynamic_pack_simple() : () -> ()
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.initializer {
  %c0_i64 = arith.constant 0 : i64
  %c128 = arith.constant 128 : index
  %c0 = arith.constant 0 : index
  %0 = stream.timepoint.immediate => !stream.timepoint
  %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
  %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
  cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
^bb1:  // pred: ^bb0
  %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
  %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
  %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
  cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
  util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  util.return
}

// -----// IR Dump After Canonicalizer (canonicalize) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %c128 = arith.constant 128 : index
  %c32_i64 = arith.constant 32 : i64
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c2 : index
  %3 = arith.ceildivui %0, %2 : index
  %4 = arith.ceildivui %1, %2 : index
  %5 = arith.muli %3, %c4 : index
  %6 = arith.muli %5, %4 : index
  %7 = arith.muli %6, %2 : index
  %8 = arith.muli %7, %2 : index
  %9 = util.align %8, %c64 : index
  %10 = arith.addi %9, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
  %11 = arith.index_castui %0 : index to i64
  %12 = arith.trunci %11 : i64 to i32
  %13 = arith.shrui %11, %c32_i64 : i64
  %14 = arith.trunci %13 : i64 to i32
  %15 = arith.index_castui %1 : index to i64
  %16 = arith.trunci %15 : i64 to i32
  %17 = arith.shrui %15, %c32_i64 : i64
  %18 = arith.trunci %17 : i64 to i32
  %19 = arith.index_castui %3 : index to i64
  %20 = arith.trunci %19 : i64 to i32
  %21 = arith.shrui %19, %c32_i64 : i64
  %22 = arith.trunci %21 : i64 to i32
  %23 = arith.index_castui %4 : index to i64
  %24 = arith.trunci %23 : i64 to i32
  %25 = arith.shrui %23, %c32_i64 : i64
  %26 = arith.trunci %25 : i64 to i32
  %27 = arith.index_castui %2 : index to i64
  %28 = arith.trunci %27 : i64 to i32
  %29 = arith.shrui %27, %c32_i64 : i64
  %30 = arith.trunci %29 : i64 to i32
  %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
        ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
        wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
    }
  } => !stream.timepoint
  %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
  %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
  %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
  %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After CSE (cse) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %c128 = arith.constant 128 : index
  %c32_i64 = arith.constant 32 : i64
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c2 : index
  %3 = arith.ceildivui %0, %2 : index
  %4 = arith.ceildivui %1, %2 : index
  %5 = arith.muli %3, %c4 : index
  %6 = arith.muli %5, %4 : index
  %7 = arith.muli %6, %2 : index
  %8 = arith.muli %7, %2 : index
  %9 = util.align %8, %c64 : index
  %10 = arith.addi %9, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
  %11 = arith.index_castui %0 : index to i64
  %12 = arith.trunci %11 : i64 to i32
  %13 = arith.shrui %11, %c32_i64 : i64
  %14 = arith.trunci %13 : i64 to i32
  %15 = arith.index_castui %1 : index to i64
  %16 = arith.trunci %15 : i64 to i32
  %17 = arith.shrui %15, %c32_i64 : i64
  %18 = arith.trunci %17 : i64 to i32
  %19 = arith.index_castui %3 : index to i64
  %20 = arith.trunci %19 : i64 to i32
  %21 = arith.shrui %19, %c32_i64 : i64
  %22 = arith.trunci %21 : i64 to i32
  %23 = arith.index_castui %4 : index to i64
  %24 = arith.trunci %23 : i64 to i32
  %25 = arith.shrui %23, %c32_i64 : i64
  %26 = arith.trunci %25 : i64 to i32
  %27 = arith.index_castui %2 : index to i64
  %28 = arith.trunci %27 : i64 to i32
  %29 = arith.shrui %27, %c32_i64 : i64
  %30 = arith.trunci %29 : i64 to i32
  %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
        ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
        wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
    }
  } => !stream.timepoint
  %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
  %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
  %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
  %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After SimplifyGlobalAccesses (iree-util-simplify-global-accesses) //----- //
util.func private @_fully_dynamic_pack_simple() {
  %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
  %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  %c64 = arith.constant 64 : index
  %c2 = arith.constant 2 : index
  %c4 = arith.constant 4 : index
  %c0 = arith.constant 0 : index
  %c128 = arith.constant 128 : index
  %c32_i64 = arith.constant 32 : i64
  %0 = util.optimization_barrier %c4 : index
  %1 = util.optimization_barrier %c4 : index
  %2 = util.optimization_barrier %c2 : index
  %3 = arith.ceildivui %0, %2 : index
  %4 = arith.ceildivui %1, %2 : index
  %5 = arith.muli %3, %c4 : index
  %6 = arith.muli %5, %4 : index
  %7 = arith.muli %6, %2 : index
  %8 = arith.muli %7, %2 : index
  %9 = util.align %8, %c64 : index
  %10 = arith.addi %9, %c64 : index
  %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
  %11 = arith.index_castui %0 : index to i64
  %12 = arith.trunci %11 : i64 to i32
  %13 = arith.shrui %11, %c32_i64 : i64
  %14 = arith.trunci %13 : i64 to i32
  %15 = arith.index_castui %1 : index to i64
  %16 = arith.trunci %15 : i64 to i32
  %17 = arith.shrui %15, %c32_i64 : i64
  %18 = arith.trunci %17 : i64 to i32
  %19 = arith.index_castui %3 : index to i64
  %20 = arith.trunci %19 : i64 to i32
  %21 = arith.shrui %19, %c32_i64 : i64
  %22 = arith.trunci %21 : i64 to i32
  %23 = arith.index_castui %4 : index to i64
  %24 = arith.trunci %23 : i64 to i32
  %25 = arith.shrui %23, %c32_i64 : i64
  %26 = arith.trunci %25 : i64 to i32
  %27 = arith.index_castui %2 : index to i64
  %28 = arith.trunci %27 : i64 to i32
  %29 = arith.shrui %27, %c32_i64 : i64
  %30 = arith.trunci %29 : i64 to i32
  %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
    stream.cmd.concurrent {
      stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
        ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
        wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
      }
      stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
      stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
    }
  } => !stream.timepoint
  %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
  %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
  %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
  %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
  check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
  util.return
}

// -----// IR Dump After ApplyPatterns (iree-util-apply-patterns) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32) {
        %c32_i64 = arith.constant 32 : i64
        %c64 = arith.constant 64 : index
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = flow.dispatch.workload.ordinal %4, 0 : index
        %26 = flow.dispatch.workload.ordinal %9, 1 : index
        %27 = flow.dispatch.workload.ordinal %14, 2 : index
        %28 = flow.dispatch.workload.ordinal %19, 3 : index
        %29 = flow.dispatch.workload.ordinal %24, 4 : index
        %30 = stream.binding.subspan %arg0[%c64] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26}
        %31 = stream.binding.subspan %arg1[%c64] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        %32 = flow.dispatch.tensor.load %30, offsets = [0, 0], sizes = [%25, %26], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26} -> tensor<?x?xi32>
        %33 = tensor.empty(%27, %28, %29, %29) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %32 inner_dims_pos = [0, 1] inner_tiles = [%29, %29] into %33 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %31, offsets = [0, 0, 0, 0], sizes = [%27, %28, %29, %29], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c32_i64 = arith.constant 32 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FoldGlobals (iree-util-fold-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32) {
        %c32_i64 = arith.constant 32 : i64
        %c64 = arith.constant 64 : index
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = flow.dispatch.workload.ordinal %4, 0 : index
        %26 = flow.dispatch.workload.ordinal %9, 1 : index
        %27 = flow.dispatch.workload.ordinal %14, 2 : index
        %28 = flow.dispatch.workload.ordinal %19, 3 : index
        %29 = flow.dispatch.workload.ordinal %24, 4 : index
        %30 = stream.binding.subspan %arg0[%c64] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26}
        %31 = stream.binding.subspan %arg1[%c64] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        %32 = flow.dispatch.tensor.load %30, offsets = [0, 0], sizes = [%25, %26], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26} -> tensor<?x?xi32>
        %33 = tensor.empty(%27, %28, %29, %29) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %32 inner_dims_pos = [0, 1] inner_tiles = [%29, %29] into %33 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %31, offsets = [0, 0, 0, 0], sizes = [%27, %28, %29, %29], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c32_i64 = arith.constant 32 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After FuseGlobals (iree-util-fuse-globals) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32) {
        %c32_i64 = arith.constant 32 : i64
        %c64 = arith.constant 64 : index
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = flow.dispatch.workload.ordinal %4, 0 : index
        %26 = flow.dispatch.workload.ordinal %9, 1 : index
        %27 = flow.dispatch.workload.ordinal %14, 2 : index
        %28 = flow.dispatch.workload.ordinal %19, 3 : index
        %29 = flow.dispatch.workload.ordinal %24, 4 : index
        %30 = stream.binding.subspan %arg0[%c64] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26}
        %31 = stream.binding.subspan %arg1[%c64] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        %32 = flow.dispatch.tensor.load %30, offsets = [0, 0], sizes = [%25, %26], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26} -> tensor<?x?xi32>
        %33 = tensor.empty(%27, %28, %29, %29) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %32 inner_dims_pos = [0, 1] inner_tiles = [%29, %29] into %33 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %31, offsets = [0, 0, 0, 0], sizes = [%27, %28, %29, %29], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c32_i64 = arith.constant 32 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After IPO (iree-util-ipo) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32) {
        %c32_i64 = arith.constant 32 : i64
        %c64 = arith.constant 64 : index
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = flow.dispatch.workload.ordinal %4, 0 : index
        %26 = flow.dispatch.workload.ordinal %9, 1 : index
        %27 = flow.dispatch.workload.ordinal %14, 2 : index
        %28 = flow.dispatch.workload.ordinal %19, 3 : index
        %29 = flow.dispatch.workload.ordinal %24, 4 : index
        %30 = stream.binding.subspan %arg0[%c64] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26}
        %31 = stream.binding.subspan %arg1[%c64] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        %32 = flow.dispatch.tensor.load %30, offsets = [0, 0], sizes = [%25, %26], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26} -> tensor<?x?xi32>
        %33 = tensor.empty(%27, %28, %29, %29) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %32 inner_dims_pos = [0, 1] inner_tiles = [%29, %29] into %33 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %31, offsets = [0, 0, 0, 0], sizes = [%27, %28, %29, %29], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c32_i64 = arith.constant 32 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After SymbolDCE (symbol-dce) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32) {
        %c32_i64 = arith.constant 32 : i64
        %c64 = arith.constant 64 : index
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = flow.dispatch.workload.ordinal %4, 0 : index
        %26 = flow.dispatch.workload.ordinal %9, 1 : index
        %27 = flow.dispatch.workload.ordinal %14, 2 : index
        %28 = flow.dispatch.workload.ordinal %19, 3 : index
        %29 = flow.dispatch.workload.ordinal %24, 4 : index
        %30 = stream.binding.subspan %arg0[%c64] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26}
        %31 = stream.binding.subspan %arg1[%c64] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        %32 = flow.dispatch.tensor.load %30, offsets = [0, 0], sizes = [%25, %26], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26} -> tensor<?x?xi32>
        %33 = tensor.empty(%27, %28, %29, %29) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %32 inner_dims_pos = [0, 1] inner_tiles = [%29, %29] into %33 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %31, offsets = [0, 0, 0, 0], sizes = [%27, %28, %29, %29], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c32_i64 = arith.constant 32 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After AssignLegacyTargetDevicesPass (iree-hal-assign-legacy-target-devices) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32) {
        %c32_i64 = arith.constant 32 : i64
        %c64 = arith.constant 64 : index
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = flow.dispatch.workload.ordinal %4, 0 : index
        %26 = flow.dispatch.workload.ordinal %9, 1 : index
        %27 = flow.dispatch.workload.ordinal %14, 2 : index
        %28 = flow.dispatch.workload.ordinal %19, 3 : index
        %29 = flow.dispatch.workload.ordinal %24, 4 : index
        %30 = stream.binding.subspan %arg0[%c64] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26}
        %31 = stream.binding.subspan %arg1[%c64] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        %32 = flow.dispatch.tensor.load %30, offsets = [0, 0], sizes = [%25, %26], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26} -> tensor<?x?xi32>
        %33 = tensor.empty(%27, %28, %29, %29) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %32 inner_dims_pos = [0, 1] inner_tiles = [%29, %29] into %33 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %31, offsets = [0, 0, 0, 0], sizes = [%27, %28, %29, %29], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c32_i64 = arith.constant 32 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After MaterializeTargetDevicesPass (iree-hal-materialize-target-devices) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [12, 13]], [[10, 11], [14, 15]]]]> : tensor<2x2x2x2xi32>,
    dense<[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15]]> : tensor<4x4xi32>,
]>
#executable_target_embedded_elf_x86_64_ = #hal.executable.target<"llvm-cpu", "embedded-elf-x86_64", {cpu = "generic", cpu_features = "", data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", native_vector_size = 16 : i64, target_triple = "x86_64-unknown-unknown-eabi-elf"}>
#device_target_local = #hal.device.target<"local", [#executable_target_embedded_elf_x86_64_]> : !hal.device
module attributes {stream.affinity.default = #hal.device.affinity<@__device_0>} {
  util.global private @__device_0 = #device_target_local
  util.func public @fully_dynamic_pack_simple() attributes {iree.abi.stub, iree.reflection = {iree.abi.declaration = "sync func @fully_dynamic_pack_simple() -> ()"}} {
    util.call @_fully_dynamic_pack_simple() : () -> ()
    util.return
  }
  stream.executable private @_fully_dynamic_pack_simple_dispatch_0 {
    stream.executable.export public @_fully_dynamic_pack_simple_dispatch_0_pack_i32 workgroups(%arg0: index, %arg1: index, %arg2: index, %arg3: index, %arg4: index) -> (index, index, index) {
      %x, %y, %z = flow.dispatch.workgroup_count_from_slice %arg0, %arg1, %arg2, %arg3, %arg4
      stream.return %x, %y, %z : index, index, index
    }
    builtin.module {
      func.func @_fully_dynamic_pack_simple_dispatch_0_pack_i32(%arg0: !stream.binding {stream.alignment = 64 : index}, %arg1: !stream.binding {stream.alignment = 64 : index}, %arg2: i32, %arg3: i32, %arg4: i32, %arg5: i32, %arg6: i32, %arg7: i32, %arg8: i32, %arg9: i32, %arg10: i32, %arg11: i32) {
        %c32_i64 = arith.constant 32 : i64
        %c64 = arith.constant 64 : index
        %0 = arith.extui %arg2 : i32 to i64
        %1 = arith.extui %arg3 : i32 to i64
        %2 = arith.shli %1, %c32_i64 : i64
        %3 = arith.ori %0, %2 : i64
        %4 = arith.index_castui %3 : i64 to index
        %5 = arith.extui %arg4 : i32 to i64
        %6 = arith.extui %arg5 : i32 to i64
        %7 = arith.shli %6, %c32_i64 : i64
        %8 = arith.ori %5, %7 : i64
        %9 = arith.index_castui %8 : i64 to index
        %10 = arith.extui %arg6 : i32 to i64
        %11 = arith.extui %arg7 : i32 to i64
        %12 = arith.shli %11, %c32_i64 : i64
        %13 = arith.ori %10, %12 : i64
        %14 = arith.index_castui %13 : i64 to index
        %15 = arith.extui %arg8 : i32 to i64
        %16 = arith.extui %arg9 : i32 to i64
        %17 = arith.shli %16, %c32_i64 : i64
        %18 = arith.ori %15, %17 : i64
        %19 = arith.index_castui %18 : i64 to index
        %20 = arith.extui %arg10 : i32 to i64
        %21 = arith.extui %arg11 : i32 to i64
        %22 = arith.shli %21, %c32_i64 : i64
        %23 = arith.ori %20, %22 : i64
        %24 = arith.index_castui %23 : i64 to index
        %25 = flow.dispatch.workload.ordinal %4, 0 : index
        %26 = flow.dispatch.workload.ordinal %9, 1 : index
        %27 = flow.dispatch.workload.ordinal %14, 2 : index
        %28 = flow.dispatch.workload.ordinal %19, 3 : index
        %29 = flow.dispatch.workload.ordinal %24, 4 : index
        %30 = stream.binding.subspan %arg0[%c64] : !stream.binding -> !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26}
        %31 = stream.binding.subspan %arg1[%c64] : !stream.binding -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        %32 = flow.dispatch.tensor.load %30, offsets = [0, 0], sizes = [%25, %26], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?xi32>>{%25, %26} -> tensor<?x?xi32>
        %33 = tensor.empty(%27, %28, %29, %29) : tensor<?x?x?x?xi32>
        %pack = tensor.pack %32 inner_dims_pos = [0, 1] inner_tiles = [%29, %29] into %33 : tensor<?x?xi32> -> tensor<?x?x?x?xi32>
        flow.dispatch.tensor.store %pack, %31, offsets = [0, 0, 0, 0], sizes = [%27, %28, %29, %29], strides = [1, 1, 1, 1] : tensor<?x?x?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?x?x?xi32>>{%27, %28, %29, %29}
        return
      }
    }
  }
  util.global private @__constant_tensor_2x2x2x2xi32__timepoint = #stream.timepoint<immediate> : !stream.timepoint
  util.global private @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
  util.initializer {
    %c0_i64 = arith.constant 0 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %0 = stream.timepoint.immediate => !stream.timepoint
    %buffer_cst = util.buffer.constant {alignment = 64 : index} : !util.buffer = #composite_of_128b
    %did_map, %result = stream.resource.try_map on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0] : !util.buffer -> i1, !stream.resource<constant>{%c128}
    cf.cond_br %did_map, ^bb2(%0, %result : !stream.timepoint, !stream.resource<constant>), ^bb1
  ^bb1:  // pred: ^bb0
    %1 = stream.resource.alloc uninitialized on(#hal.device.affinity<@__device_0>) : !stream.resource<constant>{%c128}
    %file = stream.file.constant on(#hal.device.affinity<@__device_0>) %buffer_cst[%c0 for %c128] : !util.buffer{%c128} -> !stream.file
    %2 = stream.file.read on(#hal.device.affinity<@__device_0>) await(%0) => %file[%c0_i64], %1[%c0], %c128 : !stream.file -> !stream.resource<constant>{%c128} => !stream.timepoint
    cf.br ^bb2(%2, %1 : !stream.timepoint, !stream.resource<constant>)
  ^bb2(%3: !stream.timepoint, %4: !stream.resource<constant>):  // 2 preds: ^bb0, ^bb1
    util.global.store %4, @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    util.global.store %3, @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    util.return
  }
  util.func private @_fully_dynamic_pack_simple() {
    %c32_i64 = arith.constant 32 : i64
    %c128 = arith.constant 128 : index
    %c0 = arith.constant 0 : index
    %c4 = arith.constant 4 : index
    %c2 = arith.constant 2 : index
    %c64 = arith.constant 64 : index
    %__constant_tensor_2x2x2x2xi32__timepoint = util.global.load immutable @__constant_tensor_2x2x2x2xi32__timepoint : !stream.timepoint
    %__constant_tensor_2x2x2x2xi32 = util.global.load immutable @__constant_tensor_2x2x2x2xi32 : !stream.resource<constant>
    %0 = util.optimization_barrier %c4 : index
    %1 = util.optimization_barrier %c4 : index
    %2 = util.optimization_barrier %c2 : index
    %3 = arith.ceildivui %0, %2 : index
    %4 = arith.ceildivui %1, %2 : index
    %5 = arith.muli %3, %c4 : index
    %6 = arith.muli %5, %4 : index
    %7 = arith.muli %6, %2 : index
    %8 = arith.muli %7, %2 : index
    %9 = util.align %8, %c64 : index
    %10 = arith.addi %9, %c64 : index
    %result, %result_timepoint = stream.resource.alloca uninitialized on(#hal.device.affinity<@__device_0>) await(%__constant_tensor_2x2x2x2xi32__timepoint) => !stream.resource<external>{%10} => !stream.timepoint
    %11 = arith.index_castui %0 : index to i64
    %12 = arith.trunci %11 : i64 to i32
    %13 = arith.shrui %11, %c32_i64 : i64
    %14 = arith.trunci %13 : i64 to i32
    %15 = arith.index_castui %1 : index to i64
    %16 = arith.trunci %15 : i64 to i32
    %17 = arith.shrui %15, %c32_i64 : i64
    %18 = arith.trunci %17 : i64 to i32
    %19 = arith.index_castui %3 : index to i64
    %20 = arith.trunci %19 : i64 to i32
    %21 = arith.shrui %19, %c32_i64 : i64
    %22 = arith.trunci %21 : i64 to i32
    %23 = arith.index_castui %4 : index to i64
    %24 = arith.trunci %23 : i64 to i32
    %25 = arith.shrui %23, %c32_i64 : i64
    %26 = arith.trunci %25 : i64 to i32
    %27 = arith.index_castui %2 : index to i64
    %28 = arith.trunci %27 : i64 to i32
    %29 = arith.shrui %27, %c32_i64 : i64
    %30 = arith.trunci %29 : i64 to i32
    %31 = stream.cmd.execute on(#hal.device.affinity<@__device_0>) await(%result_timepoint) => with(%__constant_tensor_2x2x2x2xi32 as %arg0: !stream.resource<constant>{%c128}, %result as %arg1: !stream.resource<external>{%10}) {
      stream.cmd.concurrent {
        stream.cmd.dispatch @_fully_dynamic_pack_simple_dispatch_0::@_fully_dynamic_pack_simple_dispatch_0_pack_i32[%0, %1, %3, %4, %2](%12, %14, %16, %18, %20, %22, %24, %26, %28, %30 : i32, i32, i32, i32, i32, i32, i32, i32, i32, i32) {
          ro %arg0[%c0 for %c128] : !stream.resource<constant>{%c128},
          wo %arg1[%c0 for %10] : !stream.resource<external>{%10}
        }
        stream.cmd.copy %arg0[%c0], %arg1[%c0], %c64 : !stream.resource<constant>{%c128} -> !stream.resource<external>{%10}
        stream.cmd.flush %arg1[%c0 for %c64] : !stream.resource<external>{%10}
      }
    } => !stream.timepoint
    %32 = stream.timepoint.await %31 => %result : !stream.resource<external>{%10}
    %33 = stream.resource.subview %32[%c0] : !stream.resource<external>{%10} -> !stream.resource<external>{%c64}
    %34 = stream.resource.subview %32[%c64] : !stream.resource<external>{%10} -> !stream.resource<external>{%8}
    %35 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %33 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    %36 = stream.tensor.export on(#hal.device.affinity<@__device_0>) %34 : tensor<2x2x2x2xi32> in !stream.resource<external>{%c64} -> tensor<2x2x2x2xi32>
    check.expect_eq(%36, %35) : tensor<2x2x2x2xi32>
    util.return
  }
}


// -----// IR Dump After ResolveDevicePromisesPass (iree-hal-resolve-device-promises) //----- //
#composite_of_128b = #util.composite<128xi8, [
    dense<[[[[0, 1], [4, 5]], [[2, 3], [6, 7]]], [[[8, 9], [1