err_dump.txt

/home/prashant/stable.mlir:793:11: error: failed to materialize conversion for result #0 of operation 'hal.interface.constant.load' that remained live after conversion
    %10 = linalg.generic {indexing_maps = [#map4, #map4], iterator_types = ["parallel", "parallel"]} ins(%9 : tensor<2x160xf32>) outs(%8 : tensor<2x160xf32>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:793:11: note: see current operation: %20 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
    %10 = linalg.generic {indexing_maps = [#map4, #map4], iterator_types = ["parallel", "parallel"]} ins(%9 : tensor<2x160xf32>) outs(%8 : tensor<2x160xf32>) {
          ^
/home/prashant/stable.mlir:793:11: note: see existing live user here: %28 = "spirv.UConvert"(%20) : (i32) -> i64
/home/prashant/stable.mlir:804:27: error: failed to run translation of source executable to target executable for backend #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>
    %inserted_slice_724 = tensor.insert_slice %10 into %12[0, 0] [2, 160] [1, 1] : tensor<2x160xf32> into tensor<2x320xf32>
                          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:804:27: note: see current operation: 
"hal.executable.variant"() ({
  "hal.executable.export"() ({
  ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index):
    %0 = "arith.constant"() {value = 5 : index} : () -> index
    %1 = "arith.constant"() {value = 2 : index} : () -> index
    %2 = "arith.constant"() {value = 1 : index} : () -> index
    "hal.return"(%0, %1, %2) : (index, index, index) -> ()
  }) {layout = #hal.pipeline.layout<push_constants = 2, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_3", translation_info = #iree_codegen.translation_info<SPIRVBaseDistribute>, workgroup_size = [32 : index, 1 : index, 1 : index]} : () -> ()
  "builtin.module"() ({
    "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
    "func.func"() ({
      %0 = "arith.constant"() {value = -1 : index} : () -> index
      %1 = "arith.constant"() {value = 4 : index} : () -> index
      %2 = "arith.constant"() {value = 32 : index} : () -> index
      %3 = "arith.constant"() {value = 160 : index} : () -> index
      %4 = "arith.constant"() {value = 0 : index} : () -> index
      %5 = "arith.constant"() {value = 640 : index} : () -> index
      %6 = "arith.constant"() {value = 320 : index} : () -> index
      %7 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
      %8 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
      %9 = "arith.index_castui"(%7) : (i32) -> index
      %10 = "arith.index_castui"(%8) : (i32) -> index
      %11 = "hal.interface.binding.subspan"(%9, %6) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %12 = "hal.interface.binding.subspan"(%4, %6) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %13 = "hal.interface.binding.subspan"(%10, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %14 = "hal.interface.binding.subspan"(%4, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %15 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
      %16 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
      %17 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
      %18 = "gpu.thread_id"() {dimension = #gpu<dim y>} : () -> index
      %19 = "arith.muli"(%16, %3) : (index, index) -> index
      %20 = "arith.muli"(%18, %3) : (index, index) -> index
      %21 = "arith.addi"(%19, %20) : (index, index) -> index
      %22 = "arith.muli"(%15, %2) : (index, index) -> index
      %23 = "arith.addi"(%21, %22) : (index, index) -> index
      %24 = "arith.addi"(%23, %17) : (index, index) -> index
      %25 = "arith.cmpi"(%9, %4) {predicate = 2 : i64} : (index, index) -> i1
      %26 = "arith.subi"(%0, %9) : (index, index) -> index
      %27 = "arith.select"(%25, %26, %9) : (i1, index, index) -> index
      %28 = "arith.divsi"(%27, %1) : (index, index) -> index
      %29 = "arith.subi"(%0, %28) : (index, index) -> index
      %30 = "arith.select"(%25, %29, %28) : (i1, index, index) -> index
      %31 = "arith.addi"(%24, %30) : (index, index) -> index
      %32 = "memref.load"(%12, %31) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
      %33 = "arith.muli"(%16, %6) : (index, index) -> index
      %34 = "arith.muli"(%18, %6) : (index, index) -> index
      %35 = "arith.addi"(%33, %34) : (index, index) -> index
      %36 = "arith.addi"(%35, %22) : (index, index) -> index
      %37 = "arith.addi"(%36, %17) : (index, index) -> index
      %38 = "arith.cmpi"(%10, %4) {predicate = 2 : i64} : (index, index) -> i1
      %39 = "arith.subi"(%0, %10) : (index, index) -> index
      %40 = "arith.select"(%38, %39, %10) : (i1, index, index) -> index
      %41 = "arith.divsi"(%40, %1) : (index, index) -> index
      %42 = "arith.subi"(%0, %41) : (index, index) -> index
      %43 = "arith.select"(%38, %42, %41) : (i1, index, index) -> index
      %44 = "arith.addi"(%37, %43) : (index, index) -> index
      "memref.store"(%32, %14, %44) : (f32, memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> ()
      "func.return"() : () -> ()
    }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [32, 1, 1]>, sym_name = "forward_dispatch_3"} : () -> ()
  }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
  "hal.executable.variant_end"() : () -> ()
}) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
    %inserted_slice_724 = tensor.insert_slice %10 into %12[0, 0] [2, 160] [1, 1] : tensor<2x160xf32> into tensor<2x320xf32>
                          ^
/home/prashant/stable.mlir:804:27: error: failed to serialize executables
    %inserted_slice_724 = tensor.insert_slice %10 into %12[0, 0] [2, 160] [1, 1] : tensor<2x160xf32> into tensor<2x320xf32>
                          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:804:27: note: see current operation: 
"hal.executable"() ({
  "hal.executable.variant"() ({
    "hal.executable.export"() ({
    ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index):
      %0 = "arith.constant"() {value = 5 : index} : () -> index
      %1 = "arith.constant"() {value = 2 : index} : () -> index
      %2 = "arith.constant"() {value = 1 : index} : () -> index
      "hal.return"(%0, %1, %2) : (index, index, index) -> ()
    }) {layout = #hal.pipeline.layout<push_constants = 2, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_3", translation_info = #iree_codegen.translation_info<SPIRVBaseDistribute>, workgroup_size = [32 : index, 1 : index, 1 : index]} : () -> ()
    "builtin.module"() ({
      "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
      "func.func"() ({
        %0 = "arith.constant"() {value = -1 : index} : () -> index
        %1 = "arith.constant"() {value = 4 : index} : () -> index
        %2 = "arith.constant"() {value = 32 : index} : () -> index
        %3 = "arith.constant"() {value = 160 : index} : () -> index
        %4 = "arith.constant"() {value = 0 : index} : () -> index
        %5 = "arith.constant"() {value = 640 : index} : () -> index
        %6 = "arith.constant"() {value = 320 : index} : () -> index
        %7 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
        %8 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
        %9 = "arith.index_castui"(%7) : (i32) -> index
        %10 = "arith.index_castui"(%8) : (i32) -> index
        %11 = "hal.interface.binding.subspan"(%9, %6) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %12 = "hal.interface.binding.subspan"(%4, %6) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %13 = "hal.interface.binding.subspan"(%10, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %14 = "hal.interface.binding.subspan"(%4, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %15 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
        %16 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
        %17 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
        %18 = "gpu.thread_id"() {dimension = #gpu<dim y>} : () -> index
        %19 = "arith.muli"(%16, %3) : (index, index) -> index
        %20 = "arith.muli"(%18, %3) : (index, index) -> index
        %21 = "arith.addi"(%19, %20) : (index, index) -> index
        %22 = "arith.muli"(%15, %2) : (index, index) -> index
        %23 = "arith.addi"(%21, %22) : (index, index) -> index
        %24 = "arith.addi"(%23, %17) : (index, index) -> index
        %25 = "arith.cmpi"(%9, %4) {predicate = 2 : i64} : (index, index) -> i1
        %26 = "arith.subi"(%0, %9) : (index, index) -> index
        %27 = "arith.select"(%25, %26, %9) : (i1, index, index) -> index
        %28 = "arith.divsi"(%27, %1) : (index, index) -> index
        %29 = "arith.subi"(%0, %28) : (index, index) -> index
        %30 = "arith.select"(%25, %29, %28) : (i1, index, index) -> index
        %31 = "arith.addi"(%24, %30) : (index, index) -> index
        %32 = "memref.load"(%12, %31) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
        %33 = "arith.muli"(%16, %6) : (index, index) -> index
        %34 = "arith.muli"(%18, %6) : (index, index) -> index
        %35 = "arith.addi"(%33, %34) : (index, index) -> index
        %36 = "arith.addi"(%35, %22) : (index, index) -> index
        %37 = "arith.addi"(%36, %17) : (index, index) -> index
        %38 = "arith.cmpi"(%10, %4) {predicate = 2 : i64} : (index, index) -> i1
        %39 = "arith.subi"(%0, %10) : (index, index) -> index
        %40 = "arith.select"(%38, %39, %10) : (i1, index, index) -> index
        %41 = "arith.divsi"(%40, %1) : (index, index) -> index
        %42 = "arith.subi"(%0, %41) : (index, index) -> index
        %43 = "arith.select"(%38, %42, %41) : (i1, index, index) -> index
        %44 = "arith.addi"(%37, %43) : (index, index) -> index
        "memref.store"(%32, %14, %44) : (f32, memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> ()
        "func.return"() : () -> ()
      }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [32, 1, 1]>, sym_name = "forward_dispatch_3"} : () -> ()
    }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
    "hal.executable.variant_end"() : () -> ()
  }) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
  "hal.executable_end"() : () -> ()
}) {sym_name = "forward_dispatch_3", sym_visibility = "private"} : () -> ()
    %inserted_slice_724 = tensor.insert_slice %10 into %12[0, 0] [2, 160] [1, 1] : tensor<2x160xf32> into tensor<2x320xf32>
                          ^
/home/prashant/stable.mlir:866:11: error: failed to materialize conversion for result #0 of operation 'hal.interface.constant.load' that remained live after conversion
    %32 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%29 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:866:11: note: see current operation: %51 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
    %32 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%29 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:866:11: note: see existing live user here: %59 = "spirv.UConvert"(%51) : (i32) -> i64
/home/prashant/stable.mlir:866:11: error: failed to run translation of source executable to target executable for backend #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>
    %32 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%29 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:866:11: note: see current operation: 
"hal.executable.variant"() ({
  "hal.executable.export"() ({
  ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index):
    %0 = "arith.constant"() {value = 64 : index} : () -> index
    %1 = "arith.constant"() {value = 1 : index} : () -> index
    "hal.return"(%0, %1, %1) : (index, index, index) -> ()
  }) {layout = #hal.pipeline.layout<push_constants = 2, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_10_generic_64x92160", translation_info = #iree_codegen.translation_info<SPIRVSubgroupReduce>, workgroup_size = [512 : index, 1 : index, 1 : index]} : () -> ()
  "builtin.module"() ({
    "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
    "func.func"() ({
      %0 = "arith.constant"() {value = 16 : index} : () -> index
      %1 = "arith.constant"() {value = 23040 : index} : () -> index
      %2 = "arith.constant"() {value = -1 : index} : () -> index
      %3 = "arith.constant"() {value = 4 : index} : () -> index
      %4 = "arith.constant"() {value = 0 : index} : () -> index
      %5 = "arith.constant"() {value = 64 : index} : () -> index
      %6 = "arith.constant"() {value = 1474560 : index} : () -> index
      %7 = "arith.constant"() {value = dense<0.000000e+00> : vector<1xf32>} : () -> vector<1xf32>
      %8 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
      %9 = "arith.constant"() {value = 32 : i32} : () -> i32
      %10 = "arith.constant"() {value = 1 : i32} : () -> i32
      %11 = "arith.constant"() {value = 2 : i32} : () -> i32
      %12 = "arith.constant"() {value = 4 : i32} : () -> i32
      %13 = "arith.constant"() {value = 8 : i32} : () -> i32
      %14 = "arith.constant"() {value = 16 : i32} : () -> i32
      %15 = "arith.constant"() {value = 32 : index} : () -> index
      %16 = "arith.constant"() {value = 15 : index} : () -> index
      %17 = "arith.constant"() {value = 0 : i32} : () -> i32
      %18 = "arith.constant"() {value = 2048 : index} : () -> index
      %19 = "arith.constant"() {value = 92160 : index} : () -> index
      %20 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
      %21 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
      %22 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
      %23 = "arith.index_castui"(%21) : (i32) -> index
      %24 = "arith.index_castui"(%22) : (i32) -> index
      %25 = "hal.interface.binding.subspan"(%23, %6) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %26 = "hal.interface.binding.subspan"(%4, %6) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %27 = "hal.interface.binding.subspan"(%24, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %28 = "hal.interface.binding.subspan"(%4, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %29 = "hal.interface.binding.subspan"(%4, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %30 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
      %31 = "scf.for"(%4, %19, %18, %8) ({
      ^bb0(%arg0: index, %arg1: vector<4xf32>):
        %70 = "arith.cmpi"(%arg0, %4) {predicate = 2 : i64} : (index, index) -> i1
        %71 = "arith.subi"(%2, %arg0) : (index, index) -> index
        %72 = "arith.select"(%70, %71, %arg0) : (i1, index, index) -> index
        %73 = "arith.divsi"(%72, %3) : (index, index) -> index
        %74 = "arith.subi"(%2, %73) : (index, index) -> index
        %75 = "arith.select"(%70, %74, %73) : (i1, index, index) -> index
        %76 = "arith.muli"(%30, %1) : (index, index) -> index
        %77 = "arith.addi"(%20, %76) : (index, index) -> index
        %78 = "arith.addi"(%75, %77) : (index, index) -> index
        %79 = "arith.cmpi"(%23, %4) {predicate = 2 : i64} : (index, index) -> i1
        %80 = "arith.subi"(%2, %23) : (index, index) -> index
        %81 = "arith.select"(%79, %80, %23) : (i1, index, index) -> index
        %82 = "arith.divsi"(%81, %0) : (index, index) -> index
        %83 = "arith.subi"(%2, %82) : (index, index) -> index
        %84 = "arith.select"(%79, %83, %82) : (i1, index, index) -> index
        %85 = "arith.addi"(%78, %84) : (index, index) -> index
        %86 = "memref.load"(%26, %85) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %87 = "arith.addf"(%86, %arg1) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        "scf.yield"(%87) : (vector<4xf32>) -> ()
      }) : (index, index, index, vector<4xf32>) -> vector<4xf32>
      %32 = "arith.cmpi"(%24, %4) {predicate = 2 : i64} : (index, index) -> i1
      %33 = "arith.subi"(%2, %24) : (index, index) -> index
      %34 = "arith.select"(%32, %33, %24) : (i1, index, index) -> index
      %35 = "arith.divsi"(%34, %3) : (index, index) -> index
      %36 = "arith.subi"(%2, %35) : (index, index) -> index
      %37 = "arith.select"(%32, %36, %35) : (i1, index, index) -> index
      %38 = "arith.addi"(%30, %37) : (index, index) -> index
      %39 = "memref.load"(%28, %38) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
      %40 = "vector.insert"(%39, %7) {position = [0]} : (f32, vector<1xf32>) -> vector<1xf32>
      %41 = "vector.extractelement"(%40, %4) : (vector<1xf32>, index) -> f32
      %42 = "vector.reduction"(%31) {kind = #vector.kind<add>} : (vector<4xf32>) -> f32
      %43:2 = "gpu.shuffle"(%42, %10, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %44 = "arith.addf"(%42, %43#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %45:2 = "gpu.shuffle"(%44, %11, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %46 = "arith.addf"(%44, %45#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %47:2 = "gpu.shuffle"(%46, %12, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %48 = "arith.addf"(%46, %47#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %49:2 = "gpu.shuffle"(%48, %13, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %50 = "arith.addf"(%48, %49#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %51:2 = "gpu.shuffle"(%50, %14, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %52 = "arith.addf"(%50, %51#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %53 = "memref.alloc"() {operand_segment_sizes = array<i32: 0, 0>} : () -> memref<16xf32, #spirv.storage_class<Workgroup>>
      %54 = "arith.divui"(%20, %15) : (index, index) -> index
      %55 = "arith.remui"(%20, %15) : (index, index) -> index
      %56 = "arith.cmpi"(%55, %4) {predicate = 0 : i64} : (index, index) -> i1
      "scf.if"(%56) ({
        "memref.store"(%52, %53, %54) {nontemporal = false} : (f32, memref<16xf32, #spirv.storage_class<Workgroup>>, index) -> ()
        "scf.yield"() : () -> ()
      }, {
      }) : (i1) -> ()
      "gpu.barrier"() : () -> ()
      %57 = "arith.minui"(%55, %16) : (index, index) -> index
      %58 = "memref.load"(%53, %57) {nontemporal = false} : (memref<16xf32, #spirv.storage_class<Workgroup>>, index) -> f32
      %59:2 = "gpu.shuffle"(%58, %10, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %60 = "arith.addf"(%58, %59#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %61:2 = "gpu.shuffle"(%60, %11, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %62 = "arith.addf"(%60, %61#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %63:2 = "gpu.shuffle"(%62, %12, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %64 = "arith.addf"(%62, %63#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %65:2 = "gpu.shuffle"(%64, %13, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %66 = "arith.addf"(%64, %65#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %67:2 = "gpu.shuffle"(%66, %17, %9) {mode = #gpu<shuffle_mode idx>} : (f32, i32, i32) -> (f32, i1)
      %68 = "arith.addf"(%67#0, %41) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %69 = "arith.cmpi"(%20, %4) {predicate = 0 : i64} : (index, index) -> i1
      "scf.if"(%69) ({
        "memref.store"(%68, %29, %38) : (f32, memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> ()
        "scf.yield"() : () -> ()
      }, {
      }) : (i1) -> ()
      "func.return"() : () -> ()
    }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [512, 1, 1]>, sym_name = "forward_dispatch_10_generic_64x92160"} : () -> ()
  }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
  "hal.executable.variant_end"() : () -> ()
}) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
    %32 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%29 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:866:11: error: failed to serialize executables
    %32 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%29 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:866:11: note: see current operation: 
"hal.executable"() ({
  "hal.executable.variant"() ({
    "hal.executable.export"() ({
    ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index):
      %0 = "arith.constant"() {value = 64 : index} : () -> index
      %1 = "arith.constant"() {value = 1 : index} : () -> index
      "hal.return"(%0, %1, %1) : (index, index, index) -> ()
    }) {layout = #hal.pipeline.layout<push_constants = 2, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_10_generic_64x92160", translation_info = #iree_codegen.translation_info<SPIRVSubgroupReduce>, workgroup_size = [512 : index, 1 : index, 1 : index]} : () -> ()
    "builtin.module"() ({
      "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
      "func.func"() ({
        %0 = "arith.constant"() {value = 16 : index} : () -> index
        %1 = "arith.constant"() {value = 23040 : index} : () -> index
        %2 = "arith.constant"() {value = -1 : index} : () -> index
        %3 = "arith.constant"() {value = 4 : index} : () -> index
        %4 = "arith.constant"() {value = 0 : index} : () -> index
        %5 = "arith.constant"() {value = 64 : index} : () -> index
        %6 = "arith.constant"() {value = 1474560 : index} : () -> index
        %7 = "arith.constant"() {value = dense<0.000000e+00> : vector<1xf32>} : () -> vector<1xf32>
        %8 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
        %9 = "arith.constant"() {value = 32 : i32} : () -> i32
        %10 = "arith.constant"() {value = 1 : i32} : () -> i32
        %11 = "arith.constant"() {value = 2 : i32} : () -> i32
        %12 = "arith.constant"() {value = 4 : i32} : () -> i32
        %13 = "arith.constant"() {value = 8 : i32} : () -> i32
        %14 = "arith.constant"() {value = 16 : i32} : () -> i32
        %15 = "arith.constant"() {value = 32 : index} : () -> index
        %16 = "arith.constant"() {value = 15 : index} : () -> index
        %17 = "arith.constant"() {value = 0 : i32} : () -> i32
        %18 = "arith.constant"() {value = 2048 : index} : () -> index
        %19 = "arith.constant"() {value = 92160 : index} : () -> index
        %20 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
        %21 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
        %22 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
        %23 = "arith.index_castui"(%21) : (i32) -> index
        %24 = "arith.index_castui"(%22) : (i32) -> index
        %25 = "hal.interface.binding.subspan"(%23, %6) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %26 = "hal.interface.binding.subspan"(%4, %6) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %27 = "hal.interface.binding.subspan"(%24, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %28 = "hal.interface.binding.subspan"(%4, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %29 = "hal.interface.binding.subspan"(%4, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %30 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
        %31 = "scf.for"(%4, %19, %18, %8) ({
        ^bb0(%arg0: index, %arg1: vector<4xf32>):
          %70 = "arith.cmpi"(%arg0, %4) {predicate = 2 : i64} : (index, index) -> i1
          %71 = "arith.subi"(%2, %arg0) : (index, index) -> index
          %72 = "arith.select"(%70, %71, %arg0) : (i1, index, index) -> index
          %73 = "arith.divsi"(%72, %3) : (index, index) -> index
          %74 = "arith.subi"(%2, %73) : (index, index) -> index
          %75 = "arith.select"(%70, %74, %73) : (i1, index, index) -> index
          %76 = "arith.muli"(%30, %1) : (index, index) -> index
          %77 = "arith.addi"(%20, %76) : (index, index) -> index
          %78 = "arith.addi"(%75, %77) : (index, index) -> index
          %79 = "arith.cmpi"(%23, %4) {predicate = 2 : i64} : (index, index) -> i1
          %80 = "arith.subi"(%2, %23) : (index, index) -> index
          %81 = "arith.select"(%79, %80, %23) : (i1, index, index) -> index
          %82 = "arith.divsi"(%81, %0) : (index, index) -> index
          %83 = "arith.subi"(%2, %82) : (index, index) -> index
          %84 = "arith.select"(%79, %83, %82) : (i1, index, index) -> index
          %85 = "arith.addi"(%78, %84) : (index, index) -> index
          %86 = "memref.load"(%26, %85) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
          %87 = "arith.addf"(%86, %arg1) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          "scf.yield"(%87) : (vector<4xf32>) -> ()
        }) : (index, index, index, vector<4xf32>) -> vector<4xf32>
        %32 = "arith.cmpi"(%24, %4) {predicate = 2 : i64} : (index, index) -> i1
        %33 = "arith.subi"(%2, %24) : (index, index) -> index
        %34 = "arith.select"(%32, %33, %24) : (i1, index, index) -> index
        %35 = "arith.divsi"(%34, %3) : (index, index) -> index
        %36 = "arith.subi"(%2, %35) : (index, index) -> index
        %37 = "arith.select"(%32, %36, %35) : (i1, index, index) -> index
        %38 = "arith.addi"(%30, %37) : (index, index) -> index
        %39 = "memref.load"(%28, %38) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
        %40 = "vector.insert"(%39, %7) {position = [0]} : (f32, vector<1xf32>) -> vector<1xf32>
        %41 = "vector.extractelement"(%40, %4) : (vector<1xf32>, index) -> f32
        %42 = "vector.reduction"(%31) {kind = #vector.kind<add>} : (vector<4xf32>) -> f32
        %43:2 = "gpu.shuffle"(%42, %10, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %44 = "arith.addf"(%42, %43#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %45:2 = "gpu.shuffle"(%44, %11, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %46 = "arith.addf"(%44, %45#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %47:2 = "gpu.shuffle"(%46, %12, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %48 = "arith.addf"(%46, %47#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %49:2 = "gpu.shuffle"(%48, %13, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %50 = "arith.addf"(%48, %49#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %51:2 = "gpu.shuffle"(%50, %14, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %52 = "arith.addf"(%50, %51#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %53 = "memref.alloc"() {operand_segment_sizes = array<i32: 0, 0>} : () -> memref<16xf32, #spirv.storage_class<Workgroup>>
        %54 = "arith.divui"(%20, %15) : (index, index) -> index
        %55 = "arith.remui"(%20, %15) : (index, index) -> index
        %56 = "arith.cmpi"(%55, %4) {predicate = 0 : i64} : (index, index) -> i1
        "scf.if"(%56) ({
          "memref.store"(%52, %53, %54) {nontemporal = false} : (f32, memref<16xf32, #spirv.storage_class<Workgroup>>, index) -> ()
          "scf.yield"() : () -> ()
        }, {
        }) : (i1) -> ()
        "gpu.barrier"() : () -> ()
        %57 = "arith.minui"(%55, %16) : (index, index) -> index
        %58 = "memref.load"(%53, %57) {nontemporal = false} : (memref<16xf32, #spirv.storage_class<Workgroup>>, index) -> f32
        %59:2 = "gpu.shuffle"(%58, %10, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %60 = "arith.addf"(%58, %59#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %61:2 = "gpu.shuffle"(%60, %11, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %62 = "arith.addf"(%60, %61#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %63:2 = "gpu.shuffle"(%62, %12, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %64 = "arith.addf"(%62, %63#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %65:2 = "gpu.shuffle"(%64, %13, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %66 = "arith.addf"(%64, %65#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %67:2 = "gpu.shuffle"(%66, %17, %9) {mode = #gpu<shuffle_mode idx>} : (f32, i32, i32) -> (f32, i1)
        %68 = "arith.addf"(%67#0, %41) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %69 = "arith.cmpi"(%20, %4) {predicate = 0 : i64} : (index, index) -> i1
        "scf.if"(%69) ({
          "memref.store"(%68, %29, %38) : (f32, memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> ()
          "scf.yield"() : () -> ()
        }, {
        }) : (i1) -> ()
        "func.return"() : () -> ()
      }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [512, 1, 1]>, sym_name = "forward_dispatch_10_generic_64x92160"} : () -> ()
    }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
    "hal.executable.variant_end"() : () -> ()
  }) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
  "hal.executable_end"() : () -> ()
}) {sym_name = "forward_dispatch_10", sym_visibility = "private"} : () -> ()
    %32 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%29 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:876:11: error: failed to materialize conversion for result #0 of operation 'hal.interface.constant.load' that remained live after conversion
    %34 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%29, %33 : tensor<2x32x10x9216xf64>, tensor<2x32x1x1xf64>) outs(%28 : tensor<2x32x10x9216xf64>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:876:11: note: see current operation: %30 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
    %34 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%29, %33 : tensor<2x32x10x9216xf64>, tensor<2x32x1x1xf64>) outs(%28 : tensor<2x32x10x9216xf64>) {
          ^
/home/prashant/stable.mlir:876:11: note: see existing live user here: %45 = "spirv.UConvert"(%30) : (i32) -> i64
/home/prashant/stable.mlir:876:11: error: failed to run translation of source executable to target executable for backend #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>
    %34 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%29, %33 : tensor<2x32x10x9216xf64>, tensor<2x32x1x1xf64>) outs(%28 : tensor<2x32x10x9216xf64>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:876:11: note: see current operation: 
"hal.executable.variant"() ({
  "hal.executable.export"() ({
  ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index, %arg3: index):
    %0 = "arith.constant"() {value = 72 : index} : () -> index
    %1 = "arith.constant"() {value = 10 : index} : () -> index
    %2 = "arith.constant"() {value = 64 : index} : () -> index
    "hal.return"(%0, %1, %2) : (index, index, index) -> ()
  }) {layout = #hal.pipeline.layout<push_constants = 3, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_11_generic_64x10x9216", translation_info = #iree_codegen.translation_info<SPIRVBaseVectorize>, workgroup_size = [32 : index, 1 : index, 1 : index]} : () -> ()
  "builtin.module"() ({
    "spirv.GlobalVariable"() {aliased, binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0__0", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {aliased, binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
    "func.func"() ({
      %0 = "arith.constant"() {value = 4 : index} : () -> index
      %1 = "arith.constant"() {value = -1 : index} : () -> index
      %2 = "arith.constant"() {value = 16 : index} : () -> index
      %3 = "arith.constant"() {value = 2304 : index} : () -> index
      %4 = "arith.constant"() {value = 23040 : index} : () -> index
      %5 = "arith.constant"() {value = 32 : index} : () -> index
      %6 = "arith.constant"() {value = 0 : index} : () -> index
      %7 = "arith.constant"() {value = 64 : index} : () -> index
      %8 = "arith.constant"() {value = 1474560 : index} : () -> index
      %9 = "arith.constant"() {value = dense<0.000000e+00> : vector<1xf32>} : () -> vector<1xf32>
      %10 = "arith.constant"() {value = dense<9.216000e+04> : vector<1xf32>} : () -> vector<1xf32>
      %11 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
      %12 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
      %13 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
      %14 = "hal.interface.constant.load"() {index = 2 : index} : () -> i32
      %15 = "arith.index_castui"(%12) : (i32) -> index
      %16 = "arith.index_castui"(%13) : (i32) -> index
      %17 = "arith.index_castui"(%14) : (i32) -> index
      %18 = "hal.interface.binding.subspan"(%15, %8) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %19 = "hal.interface.binding.subspan"(%6, %8) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %20 = "hal.interface.binding.subspan"(%16, %7) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %21 = "hal.interface.binding.subspan"(%6, %7) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %22 = "hal.interface.binding.subspan"(%17, %8) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %23 = "hal.interface.binding.subspan"(%6, %8) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %24 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
      %25 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
      %26 = "hal.interface.workgroup.id"() {dimension = 2 : index} : () -> index
      %27 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
      %28 = "arith.muli"(%24, %5) : (index, index) -> index
      %29 = "arith.addi"(%27, %28) : (index, index) -> index
      %30 = "arith.muli"(%26, %4) : (index, index) -> index
      %31 = "arith.addi"(%29, %30) : (index, index) -> index
      %32 = "arith.muli"(%25, %3) : (index, index) -> index
      %33 = "arith.addi"(%31, %32) : (index, index) -> index
      %34 = "arith.cmpi"(%15, %6) {predicate = 2 : i64} : (index, index) -> i1
      %35 = "arith.subi"(%1, %15) : (index, index) -> index
      %36 = "arith.select"(%34, %35, %15) : (i1, index, index) -> index
      %37 = "arith.divsi"(%36, %2) : (index, index) -> index
      %38 = "arith.subi"(%1, %37) : (index, index) -> index
      %39 = "arith.select"(%34, %38, %37) : (i1, index, index) -> index
      %40 = "arith.addi"(%33, %39) : (index, index) -> index
      %41 = "memref.load"(%19, %40) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
      %42 = "arith.cmpi"(%16, %6) {predicate = 2 : i64} : (index, index) -> i1
      %43 = "arith.subi"(%1, %16) : (index, index) -> index
      %44 = "arith.select"(%42, %43, %16) : (i1, index, index) -> index
      %45 = "arith.divsi"(%44, %0) : (index, index) -> index
      %46 = "arith.subi"(%1, %45) : (index, index) -> index
      %47 = "arith.select"(%42, %46, %45) : (i1, index, index) -> index
      %48 = "arith.addi"(%26, %47) : (index, index) -> index
      %49 = "memref.load"(%21, %48) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
      %50 = "vector.insert"(%49, %9) {position = [0]} : (f32, vector<1xf32>) -> vector<1xf32>
      %51 = "arith.divf"(%50, %10) {fastmath = #arith.fastmath<none>} : (vector<1xf32>, vector<1xf32>) -> vector<1xf32>
      %52 = "vector.extract"(%51) {position = [0]} : (vector<1xf32>) -> f32
      %53 = "vector.insert"(%52, %11) {position = [0]} : (f32, vector<4xf32>) -> vector<4xf32>
      %54 = "vector.insert"(%52, %53) {position = [1]} : (f32, vector<4xf32>) -> vector<4xf32>
      %55 = "vector.insert"(%52, %54) {position = [2]} : (f32, vector<4xf32>) -> vector<4xf32>
      %56 = "vector.insert"(%52, %55) {position = [3]} : (f32, vector<4xf32>) -> vector<4xf32>
      %57 = "arith.subf"(%41, %56) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %58 = "arith.cmpi"(%17, %6) {predicate = 2 : i64} : (index, index) -> i1
      %59 = "arith.subi"(%1, %17) : (index, index) -> index
      %60 = "arith.select"(%58, %59, %17) : (i1, index, index) -> index
      %61 = "arith.divsi"(%60, %2) : (index, index) -> index
      %62 = "arith.subi"(%1, %61) : (index, index) -> index
      %63 = "arith.select"(%58, %62, %61) : (i1, index, index) -> index
      %64 = "arith.addi"(%33, %63) : (index, index) -> index
      "memref.store"(%57, %23, %64) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
      "func.return"() : () -> ()
    }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [32, 1, 1]>, sym_name = "forward_dispatch_11_generic_64x10x9216"} : () -> ()
  }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
  "hal.executable.variant_end"() : () -> ()
}) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
    %34 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%29, %33 : tensor<2x32x10x9216xf64>, tensor<2x32x1x1xf64>) outs(%28 : tensor<2x32x10x9216xf64>) {
          ^
/home/prashant/stable.mlir:876:11: error: failed to serialize executables
    %34 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%29, %33 : tensor<2x32x10x9216xf64>, tensor<2x32x1x1xf64>) outs(%28 : tensor<2x32x10x9216xf64>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:876:11: note: see current operation: 
"hal.executable"() ({
  "hal.executable.variant"() ({
    "hal.executable.export"() ({
    ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index, %arg3: index):
      %0 = "arith.constant"() {value = 72 : index} : () -> index
      %1 = "arith.constant"() {value = 10 : index} : () -> index
      %2 = "arith.constant"() {value = 64 : index} : () -> index
      "hal.return"(%0, %1, %2) : (index, index, index) -> ()
    }) {layout = #hal.pipeline.layout<push_constants = 3, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_11_generic_64x10x9216", translation_info = #iree_codegen.translation_info<SPIRVBaseVectorize>, workgroup_size = [32 : index, 1 : index, 1 : index]} : () -> ()
    "builtin.module"() ({
      "spirv.GlobalVariable"() {aliased, binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0__0", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {aliased, binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
      "func.func"() ({
        %0 = "arith.constant"() {value = 4 : index} : () -> index
        %1 = "arith.constant"() {value = -1 : index} : () -> index
        %2 = "arith.constant"() {value = 16 : index} : () -> index
        %3 = "arith.constant"() {value = 2304 : index} : () -> index
        %4 = "arith.constant"() {value = 23040 : index} : () -> index
        %5 = "arith.constant"() {value = 32 : index} : () -> index
        %6 = "arith.constant"() {value = 0 : index} : () -> index
        %7 = "arith.constant"() {value = 64 : index} : () -> index
        %8 = "arith.constant"() {value = 1474560 : index} : () -> index
        %9 = "arith.constant"() {value = dense<0.000000e+00> : vector<1xf32>} : () -> vector<1xf32>
        %10 = "arith.constant"() {value = dense<9.216000e+04> : vector<1xf32>} : () -> vector<1xf32>
        %11 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
        %12 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
        %13 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
        %14 = "hal.interface.constant.load"() {index = 2 : index} : () -> i32
        %15 = "arith.index_castui"(%12) : (i32) -> index
        %16 = "arith.index_castui"(%13) : (i32) -> index
        %17 = "arith.index_castui"(%14) : (i32) -> index
        %18 = "hal.interface.binding.subspan"(%15, %8) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %19 = "hal.interface.binding.subspan"(%6, %8) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %20 = "hal.interface.binding.subspan"(%16, %7) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %21 = "hal.interface.binding.subspan"(%6, %7) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %22 = "hal.interface.binding.subspan"(%17, %8) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %23 = "hal.interface.binding.subspan"(%6, %8) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %24 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
        %25 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
        %26 = "hal.interface.workgroup.id"() {dimension = 2 : index} : () -> index
        %27 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
        %28 = "arith.muli"(%24, %5) : (index, index) -> index
        %29 = "arith.addi"(%27, %28) : (index, index) -> index
        %30 = "arith.muli"(%26, %4) : (index, index) -> index
        %31 = "arith.addi"(%29, %30) : (index, index) -> index
        %32 = "arith.muli"(%25, %3) : (index, index) -> index
        %33 = "arith.addi"(%31, %32) : (index, index) -> index
        %34 = "arith.cmpi"(%15, %6) {predicate = 2 : i64} : (index, index) -> i1
        %35 = "arith.subi"(%1, %15) : (index, index) -> index
        %36 = "arith.select"(%34, %35, %15) : (i1, index, index) -> index
        %37 = "arith.divsi"(%36, %2) : (index, index) -> index
        %38 = "arith.subi"(%1, %37) : (index, index) -> index
        %39 = "arith.select"(%34, %38, %37) : (i1, index, index) -> index
        %40 = "arith.addi"(%33, %39) : (index, index) -> index
        %41 = "memref.load"(%19, %40) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %42 = "arith.cmpi"(%16, %6) {predicate = 2 : i64} : (index, index) -> i1
        %43 = "arith.subi"(%1, %16) : (index, index) -> index
        %44 = "arith.select"(%42, %43, %16) : (i1, index, index) -> index
        %45 = "arith.divsi"(%44, %0) : (index, index) -> index
        %46 = "arith.subi"(%1, %45) : (index, index) -> index
        %47 = "arith.select"(%42, %46, %45) : (i1, index, index) -> index
        %48 = "arith.addi"(%26, %47) : (index, index) -> index
        %49 = "memref.load"(%21, %48) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
        %50 = "vector.insert"(%49, %9) {position = [0]} : (f32, vector<1xf32>) -> vector<1xf32>
        %51 = "arith.divf"(%50, %10) {fastmath = #arith.fastmath<none>} : (vector<1xf32>, vector<1xf32>) -> vector<1xf32>
        %52 = "vector.extract"(%51) {position = [0]} : (vector<1xf32>) -> f32
        %53 = "vector.insert"(%52, %11) {position = [0]} : (f32, vector<4xf32>) -> vector<4xf32>
        %54 = "vector.insert"(%52, %53) {position = [1]} : (f32, vector<4xf32>) -> vector<4xf32>
        %55 = "vector.insert"(%52, %54) {position = [2]} : (f32, vector<4xf32>) -> vector<4xf32>
        %56 = "vector.insert"(%52, %55) {position = [3]} : (f32, vector<4xf32>) -> vector<4xf32>
        %57 = "arith.subf"(%41, %56) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %58 = "arith.cmpi"(%17, %6) {predicate = 2 : i64} : (index, index) -> i1
        %59 = "arith.subi"(%1, %17) : (index, index) -> index
        %60 = "arith.select"(%58, %59, %17) : (i1, index, index) -> index
        %61 = "arith.divsi"(%60, %2) : (index, index) -> index
        %62 = "arith.subi"(%1, %61) : (index, index) -> index
        %63 = "arith.select"(%58, %62, %61) : (i1, index, index) -> index
        %64 = "arith.addi"(%33, %63) : (index, index) -> index
        "memref.store"(%57, %23, %64) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
        "func.return"() : () -> ()
      }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [32, 1, 1]>, sym_name = "forward_dispatch_11_generic_64x10x9216"} : () -> ()
    }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
    "hal.executable.variant_end"() : () -> ()
  }) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
  "hal.executable_end"() : () -> ()
}) {sym_name = "forward_dispatch_11", sym_visibility = "private"} : () -> ()
    %34 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%29, %33 : tensor<2x32x10x9216xf64>, tensor<2x32x1x1xf64>) outs(%28 : tensor<2x32x10x9216xf64>) {
          ^
/home/prashant/stable.mlir:886:11: error: failed to materialize conversion for result #0 of operation 'hal.interface.constant.load' that remained live after conversion
    %36 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%35 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:886:11: note: see current operation: %51 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
    %36 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%35 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:886:11: note: see existing live user here: %59 = "spirv.UConvert"(%51) : (i32) -> i64
/home/prashant/stable.mlir:886:11: error: failed to run translation of source executable to target executable for backend #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>
    %36 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%35 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:886:11: note: see current operation: 
"hal.executable.variant"() ({
  "hal.executable.export"() ({
  ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index):
    %0 = "arith.constant"() {value = 64 : index} : () -> index
    %1 = "arith.constant"() {value = 1 : index} : () -> index
    "hal.return"(%0, %1, %1) : (index, index, index) -> ()
  }) {layout = #hal.pipeline.layout<push_constants = 2, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_12_generic_64x92160", translation_info = #iree_codegen.translation_info<SPIRVSubgroupReduce>, workgroup_size = [512 : index, 1 : index, 1 : index]} : () -> ()
  "builtin.module"() ({
    "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
    "func.func"() ({
      %0 = "arith.constant"() {value = 16 : index} : () -> index
      %1 = "arith.constant"() {value = 23040 : index} : () -> index
      %2 = "arith.constant"() {value = -1 : index} : () -> index
      %3 = "arith.constant"() {value = 4 : index} : () -> index
      %4 = "arith.constant"() {value = 0 : index} : () -> index
      %5 = "arith.constant"() {value = 64 : index} : () -> index
      %6 = "arith.constant"() {value = 1474560 : index} : () -> index
      %7 = "arith.constant"() {value = dense<0.000000e+00> : vector<1xf32>} : () -> vector<1xf32>
      %8 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
      %9 = "arith.constant"() {value = 32 : i32} : () -> i32
      %10 = "arith.constant"() {value = 1 : i32} : () -> i32
      %11 = "arith.constant"() {value = 2 : i32} : () -> i32
      %12 = "arith.constant"() {value = 4 : i32} : () -> i32
      %13 = "arith.constant"() {value = 8 : i32} : () -> i32
      %14 = "arith.constant"() {value = 16 : i32} : () -> i32
      %15 = "arith.constant"() {value = 32 : index} : () -> index
      %16 = "arith.constant"() {value = 15 : index} : () -> index
      %17 = "arith.constant"() {value = 0 : i32} : () -> i32
      %18 = "arith.constant"() {value = 2048 : index} : () -> index
      %19 = "arith.constant"() {value = 92160 : index} : () -> index
      %20 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
      %21 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
      %22 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
      %23 = "arith.index_castui"(%21) : (i32) -> index
      %24 = "arith.index_castui"(%22) : (i32) -> index
      %25 = "hal.interface.binding.subspan"(%23, %6) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %26 = "hal.interface.binding.subspan"(%4, %6) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %27 = "hal.interface.binding.subspan"(%24, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %28 = "hal.interface.binding.subspan"(%4, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %29 = "hal.interface.binding.subspan"(%4, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %30 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
      %31 = "scf.for"(%4, %19, %18, %8) ({
      ^bb0(%arg0: index, %arg1: vector<4xf32>):
        %70 = "arith.cmpi"(%arg0, %4) {predicate = 2 : i64} : (index, index) -> i1
        %71 = "arith.subi"(%2, %arg0) : (index, index) -> index
        %72 = "arith.select"(%70, %71, %arg0) : (i1, index, index) -> index
        %73 = "arith.divsi"(%72, %3) : (index, index) -> index
        %74 = "arith.subi"(%2, %73) : (index, index) -> index
        %75 = "arith.select"(%70, %74, %73) : (i1, index, index) -> index
        %76 = "arith.muli"(%30, %1) : (index, index) -> index
        %77 = "arith.addi"(%20, %76) : (index, index) -> index
        %78 = "arith.addi"(%75, %77) : (index, index) -> index
        %79 = "arith.cmpi"(%23, %4) {predicate = 2 : i64} : (index, index) -> i1
        %80 = "arith.subi"(%2, %23) : (index, index) -> index
        %81 = "arith.select"(%79, %80, %23) : (i1, index, index) -> index
        %82 = "arith.divsi"(%81, %0) : (index, index) -> index
        %83 = "arith.subi"(%2, %82) : (index, index) -> index
        %84 = "arith.select"(%79, %83, %82) : (i1, index, index) -> index
        %85 = "arith.addi"(%78, %84) : (index, index) -> index
        %86 = "memref.load"(%26, %85) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %87 = "arith.mulf"(%86, %86) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %88 = "arith.addf"(%87, %arg1) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        "scf.yield"(%88) : (vector<4xf32>) -> ()
      }) : (index, index, index, vector<4xf32>) -> vector<4xf32>
      %32 = "arith.cmpi"(%24, %4) {predicate = 2 : i64} : (index, index) -> i1
      %33 = "arith.subi"(%2, %24) : (index, index) -> index
      %34 = "arith.select"(%32, %33, %24) : (i1, index, index) -> index
      %35 = "arith.divsi"(%34, %3) : (index, index) -> index
      %36 = "arith.subi"(%2, %35) : (index, index) -> index
      %37 = "arith.select"(%32, %36, %35) : (i1, index, index) -> index
      %38 = "arith.addi"(%30, %37) : (index, index) -> index
      %39 = "memref.load"(%28, %38) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
      %40 = "vector.insert"(%39, %7) {position = [0]} : (f32, vector<1xf32>) -> vector<1xf32>
      %41 = "vector.extractelement"(%40, %4) : (vector<1xf32>, index) -> f32
      %42 = "vector.reduction"(%31) {kind = #vector.kind<add>} : (vector<4xf32>) -> f32
      %43:2 = "gpu.shuffle"(%42, %10, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %44 = "arith.addf"(%42, %43#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %45:2 = "gpu.shuffle"(%44, %11, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %46 = "arith.addf"(%44, %45#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %47:2 = "gpu.shuffle"(%46, %12, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %48 = "arith.addf"(%46, %47#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %49:2 = "gpu.shuffle"(%48, %13, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %50 = "arith.addf"(%48, %49#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %51:2 = "gpu.shuffle"(%50, %14, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %52 = "arith.addf"(%50, %51#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %53 = "memref.alloc"() {operand_segment_sizes = array<i32: 0, 0>} : () -> memref<16xf32, #spirv.storage_class<Workgroup>>
      %54 = "arith.divui"(%20, %15) : (index, index) -> index
      %55 = "arith.remui"(%20, %15) : (index, index) -> index
      %56 = "arith.cmpi"(%55, %4) {predicate = 0 : i64} : (index, index) -> i1
      "scf.if"(%56) ({
        "memref.store"(%52, %53, %54) {nontemporal = false} : (f32, memref<16xf32, #spirv.storage_class<Workgroup>>, index) -> ()
        "scf.yield"() : () -> ()
      }, {
      }) : (i1) -> ()
      "gpu.barrier"() : () -> ()
      %57 = "arith.minui"(%55, %16) : (index, index) -> index
      %58 = "memref.load"(%53, %57) {nontemporal = false} : (memref<16xf32, #spirv.storage_class<Workgroup>>, index) -> f32
      %59:2 = "gpu.shuffle"(%58, %10, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %60 = "arith.addf"(%58, %59#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %61:2 = "gpu.shuffle"(%60, %11, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %62 = "arith.addf"(%60, %61#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %63:2 = "gpu.shuffle"(%62, %12, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %64 = "arith.addf"(%62, %63#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %65:2 = "gpu.shuffle"(%64, %13, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
      %66 = "arith.addf"(%64, %65#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %67:2 = "gpu.shuffle"(%66, %17, %9) {mode = #gpu<shuffle_mode idx>} : (f32, i32, i32) -> (f32, i1)
      %68 = "arith.addf"(%67#0, %41) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
      %69 = "arith.cmpi"(%20, %4) {predicate = 0 : i64} : (index, index) -> i1
      "scf.if"(%69) ({
        "memref.store"(%68, %29, %38) : (f32, memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> ()
        "scf.yield"() : () -> ()
      }, {
      }) : (i1) -> ()
      "func.return"() : () -> ()
    }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [512, 1, 1]>, sym_name = "forward_dispatch_12_generic_64x92160"} : () -> ()
  }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
  "hal.executable.variant_end"() : () -> ()
}) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
    %36 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%35 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:886:11: error: failed to serialize executables
    %36 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%35 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:886:11: note: see current operation: 
"hal.executable"() ({
  "hal.executable.variant"() ({
    "hal.executable.export"() ({
    ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index):
      %0 = "arith.constant"() {value = 64 : index} : () -> index
      %1 = "arith.constant"() {value = 1 : index} : () -> index
      "hal.return"(%0, %1, %1) : (index, index, index) -> ()
    }) {layout = #hal.pipeline.layout<push_constants = 2, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_12_generic_64x92160", translation_info = #iree_codegen.translation_info<SPIRVSubgroupReduce>, workgroup_size = [512 : index, 1 : index, 1 : index]} : () -> ()
    "builtin.module"() ({
      "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
      "func.func"() ({
        %0 = "arith.constant"() {value = 16 : index} : () -> index
        %1 = "arith.constant"() {value = 23040 : index} : () -> index
        %2 = "arith.constant"() {value = -1 : index} : () -> index
        %3 = "arith.constant"() {value = 4 : index} : () -> index
        %4 = "arith.constant"() {value = 0 : index} : () -> index
        %5 = "arith.constant"() {value = 64 : index} : () -> index
        %6 = "arith.constant"() {value = 1474560 : index} : () -> index
        %7 = "arith.constant"() {value = dense<0.000000e+00> : vector<1xf32>} : () -> vector<1xf32>
        %8 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
        %9 = "arith.constant"() {value = 32 : i32} : () -> i32
        %10 = "arith.constant"() {value = 1 : i32} : () -> i32
        %11 = "arith.constant"() {value = 2 : i32} : () -> i32
        %12 = "arith.constant"() {value = 4 : i32} : () -> i32
        %13 = "arith.constant"() {value = 8 : i32} : () -> i32
        %14 = "arith.constant"() {value = 16 : i32} : () -> i32
        %15 = "arith.constant"() {value = 32 : index} : () -> index
        %16 = "arith.constant"() {value = 15 : index} : () -> index
        %17 = "arith.constant"() {value = 0 : i32} : () -> i32
        %18 = "arith.constant"() {value = 2048 : index} : () -> index
        %19 = "arith.constant"() {value = 92160 : index} : () -> index
        %20 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
        %21 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
        %22 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
        %23 = "arith.index_castui"(%21) : (i32) -> index
        %24 = "arith.index_castui"(%22) : (i32) -> index
        %25 = "hal.interface.binding.subspan"(%23, %6) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %26 = "hal.interface.binding.subspan"(%4, %6) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %27 = "hal.interface.binding.subspan"(%24, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %28 = "hal.interface.binding.subspan"(%4, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %29 = "hal.interface.binding.subspan"(%4, %5) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %30 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
        %31 = "scf.for"(%4, %19, %18, %8) ({
        ^bb0(%arg0: index, %arg1: vector<4xf32>):
          %70 = "arith.cmpi"(%arg0, %4) {predicate = 2 : i64} : (index, index) -> i1
          %71 = "arith.subi"(%2, %arg0) : (index, index) -> index
          %72 = "arith.select"(%70, %71, %arg0) : (i1, index, index) -> index
          %73 = "arith.divsi"(%72, %3) : (index, index) -> index
          %74 = "arith.subi"(%2, %73) : (index, index) -> index
          %75 = "arith.select"(%70, %74, %73) : (i1, index, index) -> index
          %76 = "arith.muli"(%30, %1) : (index, index) -> index
          %77 = "arith.addi"(%20, %76) : (index, index) -> index
          %78 = "arith.addi"(%75, %77) : (index, index) -> index
          %79 = "arith.cmpi"(%23, %4) {predicate = 2 : i64} : (index, index) -> i1
          %80 = "arith.subi"(%2, %23) : (index, index) -> index
          %81 = "arith.select"(%79, %80, %23) : (i1, index, index) -> index
          %82 = "arith.divsi"(%81, %0) : (index, index) -> index
          %83 = "arith.subi"(%2, %82) : (index, index) -> index
          %84 = "arith.select"(%79, %83, %82) : (i1, index, index) -> index
          %85 = "arith.addi"(%78, %84) : (index, index) -> index
          %86 = "memref.load"(%26, %85) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
          %87 = "arith.mulf"(%86, %86) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %88 = "arith.addf"(%87, %arg1) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          "scf.yield"(%88) : (vector<4xf32>) -> ()
        }) : (index, index, index, vector<4xf32>) -> vector<4xf32>
        %32 = "arith.cmpi"(%24, %4) {predicate = 2 : i64} : (index, index) -> i1
        %33 = "arith.subi"(%2, %24) : (index, index) -> index
        %34 = "arith.select"(%32, %33, %24) : (i1, index, index) -> index
        %35 = "arith.divsi"(%34, %3) : (index, index) -> index
        %36 = "arith.subi"(%2, %35) : (index, index) -> index
        %37 = "arith.select"(%32, %36, %35) : (i1, index, index) -> index
        %38 = "arith.addi"(%30, %37) : (index, index) -> index
        %39 = "memref.load"(%28, %38) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
        %40 = "vector.insert"(%39, %7) {position = [0]} : (f32, vector<1xf32>) -> vector<1xf32>
        %41 = "vector.extractelement"(%40, %4) : (vector<1xf32>, index) -> f32
        %42 = "vector.reduction"(%31) {kind = #vector.kind<add>} : (vector<4xf32>) -> f32
        %43:2 = "gpu.shuffle"(%42, %10, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %44 = "arith.addf"(%42, %43#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %45:2 = "gpu.shuffle"(%44, %11, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %46 = "arith.addf"(%44, %45#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %47:2 = "gpu.shuffle"(%46, %12, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %48 = "arith.addf"(%46, %47#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %49:2 = "gpu.shuffle"(%48, %13, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %50 = "arith.addf"(%48, %49#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %51:2 = "gpu.shuffle"(%50, %14, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %52 = "arith.addf"(%50, %51#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %53 = "memref.alloc"() {operand_segment_sizes = array<i32: 0, 0>} : () -> memref<16xf32, #spirv.storage_class<Workgroup>>
        %54 = "arith.divui"(%20, %15) : (index, index) -> index
        %55 = "arith.remui"(%20, %15) : (index, index) -> index
        %56 = "arith.cmpi"(%55, %4) {predicate = 0 : i64} : (index, index) -> i1
        "scf.if"(%56) ({
          "memref.store"(%52, %53, %54) {nontemporal = false} : (f32, memref<16xf32, #spirv.storage_class<Workgroup>>, index) -> ()
          "scf.yield"() : () -> ()
        }, {
        }) : (i1) -> ()
        "gpu.barrier"() : () -> ()
        %57 = "arith.minui"(%55, %16) : (index, index) -> index
        %58 = "memref.load"(%53, %57) {nontemporal = false} : (memref<16xf32, #spirv.storage_class<Workgroup>>, index) -> f32
        %59:2 = "gpu.shuffle"(%58, %10, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %60 = "arith.addf"(%58, %59#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %61:2 = "gpu.shuffle"(%60, %11, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %62 = "arith.addf"(%60, %61#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %63:2 = "gpu.shuffle"(%62, %12, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %64 = "arith.addf"(%62, %63#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %65:2 = "gpu.shuffle"(%64, %13, %9) {mode = #gpu<shuffle_mode xor>} : (f32, i32, i32) -> (f32, i1)
        %66 = "arith.addf"(%64, %65#0) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %67:2 = "gpu.shuffle"(%66, %17, %9) {mode = #gpu<shuffle_mode idx>} : (f32, i32, i32) -> (f32, i1)
        %68 = "arith.addf"(%67#0, %41) {fastmath = #arith.fastmath<none>} : (f32, f32) -> f32
        %69 = "arith.cmpi"(%20, %4) {predicate = 0 : i64} : (index, index) -> i1
        "scf.if"(%69) ({
          "memref.store"(%68, %29, %38) : (f32, memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> ()
          "scf.yield"() : () -> ()
        }, {
        }) : (i1) -> ()
        "func.return"() : () -> ()
      }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [512, 1, 1]>, sym_name = "forward_dispatch_12_generic_64x92160"} : () -> ()
    }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
    "hal.executable.variant_end"() : () -> ()
  }) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
  "hal.executable_end"() : () -> ()
}) {sym_name = "forward_dispatch_12", sym_visibility = "private"} : () -> ()
    %36 = linalg.generic {indexing_maps = [#map8, #map9], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%35 : tensor<2x32x10x9216xf64>) outs(%31 : tensor<2x32x1x1xf64>) {
          ^
/home/prashant/stable.mlir:876:11: error: failed to materialize conversion for result #0 of operation 'hal.interface.constant.load' that remained live after conversion
    %34 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%29, %33 : tensor<2x32x10x9216xf64>, tensor<2x32x1x1xf64>) outs(%28 : tensor<2x32x10x9216xf64>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:876:11: note: see current operation: %30 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
    %34 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%29, %33 : tensor<2x32x10x9216xf64>, tensor<2x32x1x1xf64>) outs(%28 : tensor<2x32x10x9216xf64>) {
          ^
/home/prashant/stable.mlir:876:11: note: see existing live user here: %38 = "spirv.UConvert"(%30) : (i32) -> i64
/home/prashant/stable.mlir:930:11: error: failed to run translation of source executable to target executable for backend #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>
    %47 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%46, %44 : tensor<2x32x10x9216xf32>, tensor<2x32x1x1xf32>) outs(%45 : tensor<2x32x10x9216xf32>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:930:11: note: see current operation: 
"hal.executable.variant"() ({
  "hal.executable.export"() ({
  ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index, %arg3: index):
    %0 = "arith.constant"() {value = 72 : index} : () -> index
    %1 = "arith.constant"() {value = 10 : index} : () -> index
    %2 = "arith.constant"() {value = 64 : index} : () -> index
    "hal.return"(%0, %1, %2) : (index, index, index) -> ()
  }) {layout = #hal.pipeline.layout<push_constants = 2, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_13_generic_64x10x9216", translation_info = #iree_codegen.translation_info<SPIRVBaseVectorize>, workgroup_size = [32 : index, 1 : index, 1 : index]} : () -> ()
  "builtin.module"() ({
    "spirv.GlobalVariable"() {aliased, binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0__0", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {aliased, binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
    "func.func"() ({
      %0 = "arith.constant"() {value = -1 : index} : () -> index
      %1 = "arith.constant"() {value = 16 : index} : () -> index
      %2 = "arith.constant"() {value = 2304 : index} : () -> index
      %3 = "arith.constant"() {value = 23040 : index} : () -> index
      %4 = "arith.constant"() {value = 32 : index} : () -> index
      %5 = "arith.constant"() {value = 0 : index} : () -> index
      %6 = "arith.constant"() {value = 64 : index} : () -> index
      %7 = "arith.constant"() {value = 1474560 : index} : () -> index
      %8 = "arith.constant"() {value = dense<0.000000e+00> : vector<1xf32>} : () -> vector<1xf32>
      %9 = "arith.constant"() {value = dense<9.216000e+04> : vector<1xf32>} : () -> vector<1xf32>
      %10 = "arith.constant"() {value = dense<9.99999974E-6> : vector<1xf32>} : () -> vector<1xf32>
      %11 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
      %12 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
      %13 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
      %14 = "arith.index_castui"(%12) : (i32) -> index
      %15 = "arith.index_castui"(%13) : (i32) -> index
      %16 = "hal.interface.binding.subspan"(%14, %7) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %17 = "hal.interface.binding.subspan"(%5, %7) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %18 = "hal.interface.binding.subspan"(%5, %6) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %19 = "hal.interface.binding.subspan"(%15, %7) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %20 = "hal.interface.binding.subspan"(%5, %7) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %21 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
      %22 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
      %23 = "hal.interface.workgroup.id"() {dimension = 2 : index} : () -> index
      %24 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
      %25 = "arith.muli"(%21, %4) : (index, index) -> index
      %26 = "arith.addi"(%24, %25) : (index, index) -> index
      %27 = "arith.muli"(%23, %3) : (index, index) -> index
      %28 = "arith.addi"(%26, %27) : (index, index) -> index
      %29 = "arith.muli"(%22, %2) : (index, index) -> index
      %30 = "arith.addi"(%28, %29) : (index, index) -> index
      %31 = "arith.cmpi"(%14, %5) {predicate = 2 : i64} : (index, index) -> i1
      %32 = "arith.subi"(%0, %14) : (index, index) -> index
      %33 = "arith.select"(%31, %32, %14) : (i1, index, index) -> index
      %34 = "arith.divsi"(%33, %1) : (index, index) -> index
      %35 = "arith.subi"(%0, %34) : (index, index) -> index
      %36 = "arith.select"(%31, %35, %34) : (i1, index, index) -> index
      %37 = "arith.addi"(%30, %36) : (index, index) -> index
      %38 = "memref.load"(%17, %37) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
      %39 = "memref.load"(%18, %23) {nontemporal = false} : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
      %40 = "vector.insert"(%39, %8) {position = [0]} : (f32, vector<1xf32>) -> vector<1xf32>
      %41 = "arith.divf"(%40, %9) {fastmath = #arith.fastmath<none>} : (vector<1xf32>, vector<1xf32>) -> vector<1xf32>
      %42 = "arith.addf"(%41, %10) {fastmath = #arith.fastmath<none>} : (vector<1xf32>, vector<1xf32>) -> vector<1xf32>
      %43 = "math.rsqrt"(%42) {fastmath = #arith.fastmath<none>} : (vector<1xf32>) -> vector<1xf32>
      %44 = "vector.extract"(%43) {position = [0]} : (vector<1xf32>) -> f32
      %45 = "vector.insert"(%44, %11) {position = [0]} : (f32, vector<4xf32>) -> vector<4xf32>
      %46 = "vector.insert"(%44, %45) {position = [1]} : (f32, vector<4xf32>) -> vector<4xf32>
      %47 = "vector.insert"(%44, %46) {position = [2]} : (f32, vector<4xf32>) -> vector<4xf32>
      %48 = "vector.insert"(%44, %47) {position = [3]} : (f32, vector<4xf32>) -> vector<4xf32>
      %49 = "arith.mulf"(%38, %48) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %50 = "arith.cmpi"(%15, %5) {predicate = 2 : i64} : (index, index) -> i1
      %51 = "arith.subi"(%0, %15) : (index, index) -> index
      %52 = "arith.select"(%50, %51, %15) : (i1, index, index) -> index
      %53 = "arith.divsi"(%52, %1) : (index, index) -> index
      %54 = "arith.subi"(%0, %53) : (index, index) -> index
      %55 = "arith.select"(%50, %54, %53) : (i1, index, index) -> index
      %56 = "arith.addi"(%30, %55) : (index, index) -> index
      "memref.store"(%49, %20, %56) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
      "func.return"() : () -> ()
    }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [32, 1, 1]>, sym_name = "forward_dispatch_13_generic_64x10x9216"} : () -> ()
  }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
  "hal.executable.variant_end"() : () -> ()
}) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
    %47 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%46, %44 : tensor<2x32x10x9216xf32>, tensor<2x32x1x1xf32>) outs(%45 : tensor<2x32x10x9216xf32>) {
          ^
/home/prashant/stable.mlir:930:11: error: failed to serialize executables
    %47 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%46, %44 : tensor<2x32x10x9216xf32>, tensor<2x32x1x1xf32>) outs(%45 : tensor<2x32x10x9216xf32>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:930:11: note: see current operation: 
"hal.executable"() ({
  "hal.executable.variant"() ({
    "hal.executable.export"() ({
    ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index, %arg3: index):
      %0 = "arith.constant"() {value = 72 : index} : () -> index
      %1 = "arith.constant"() {value = 10 : index} : () -> index
      %2 = "arith.constant"() {value = 64 : index} : () -> index
      "hal.return"(%0, %1, %2) : (index, index, index) -> ()
    }) {layout = #hal.pipeline.layout<push_constants = 2, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_13_generic_64x10x9216", translation_info = #iree_codegen.translation_info<SPIRVBaseVectorize>, workgroup_size = [32 : index, 1 : index, 1 : index]} : () -> ()
    "builtin.module"() ({
      "spirv.GlobalVariable"() {aliased, binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0__0", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {aliased, binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
      "func.func"() ({
        %0 = "arith.constant"() {value = -1 : index} : () -> index
        %1 = "arith.constant"() {value = 16 : index} : () -> index
        %2 = "arith.constant"() {value = 2304 : index} : () -> index
        %3 = "arith.constant"() {value = 23040 : index} : () -> index
        %4 = "arith.constant"() {value = 32 : index} : () -> index
        %5 = "arith.constant"() {value = 0 : index} : () -> index
        %6 = "arith.constant"() {value = 64 : index} : () -> index
        %7 = "arith.constant"() {value = 1474560 : index} : () -> index
        %8 = "arith.constant"() {value = dense<0.000000e+00> : vector<1xf32>} : () -> vector<1xf32>
        %9 = "arith.constant"() {value = dense<9.216000e+04> : vector<1xf32>} : () -> vector<1xf32>
        %10 = "arith.constant"() {value = dense<9.99999974E-6> : vector<1xf32>} : () -> vector<1xf32>
        %11 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
        %12 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
        %13 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
        %14 = "arith.index_castui"(%12) : (i32) -> index
        %15 = "arith.index_castui"(%13) : (i32) -> index
        %16 = "hal.interface.binding.subspan"(%14, %7) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %17 = "hal.interface.binding.subspan"(%5, %7) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %18 = "hal.interface.binding.subspan"(%5, %6) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %19 = "hal.interface.binding.subspan"(%15, %7) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %20 = "hal.interface.binding.subspan"(%5, %7) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %21 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
        %22 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
        %23 = "hal.interface.workgroup.id"() {dimension = 2 : index} : () -> index
        %24 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
        %25 = "arith.muli"(%21, %4) : (index, index) -> index
        %26 = "arith.addi"(%24, %25) : (index, index) -> index
        %27 = "arith.muli"(%23, %3) : (index, index) -> index
        %28 = "arith.addi"(%26, %27) : (index, index) -> index
        %29 = "arith.muli"(%22, %2) : (index, index) -> index
        %30 = "arith.addi"(%28, %29) : (index, index) -> index
        %31 = "arith.cmpi"(%14, %5) {predicate = 2 : i64} : (index, index) -> i1
        %32 = "arith.subi"(%0, %14) : (index, index) -> index
        %33 = "arith.select"(%31, %32, %14) : (i1, index, index) -> index
        %34 = "arith.divsi"(%33, %1) : (index, index) -> index
        %35 = "arith.subi"(%0, %34) : (index, index) -> index
        %36 = "arith.select"(%31, %35, %34) : (i1, index, index) -> index
        %37 = "arith.addi"(%30, %36) : (index, index) -> index
        %38 = "memref.load"(%17, %37) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %39 = "memref.load"(%18, %23) {nontemporal = false} : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
        %40 = "vector.insert"(%39, %8) {position = [0]} : (f32, vector<1xf32>) -> vector<1xf32>
        %41 = "arith.divf"(%40, %9) {fastmath = #arith.fastmath<none>} : (vector<1xf32>, vector<1xf32>) -> vector<1xf32>
        %42 = "arith.addf"(%41, %10) {fastmath = #arith.fastmath<none>} : (vector<1xf32>, vector<1xf32>) -> vector<1xf32>
        %43 = "math.rsqrt"(%42) {fastmath = #arith.fastmath<none>} : (vector<1xf32>) -> vector<1xf32>
        %44 = "vector.extract"(%43) {position = [0]} : (vector<1xf32>) -> f32
        %45 = "vector.insert"(%44, %11) {position = [0]} : (f32, vector<4xf32>) -> vector<4xf32>
        %46 = "vector.insert"(%44, %45) {position = [1]} : (f32, vector<4xf32>) -> vector<4xf32>
        %47 = "vector.insert"(%44, %46) {position = [2]} : (f32, vector<4xf32>) -> vector<4xf32>
        %48 = "vector.insert"(%44, %47) {position = [3]} : (f32, vector<4xf32>) -> vector<4xf32>
        %49 = "arith.mulf"(%38, %48) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %50 = "arith.cmpi"(%15, %5) {predicate = 2 : i64} : (index, index) -> i1
        %51 = "arith.subi"(%0, %15) : (index, index) -> index
        %52 = "arith.select"(%50, %51, %15) : (i1, index, index) -> index
        %53 = "arith.divsi"(%52, %1) : (index, index) -> index
        %54 = "arith.subi"(%0, %53) : (index, index) -> index
        %55 = "arith.select"(%50, %54, %53) : (i1, index, index) -> index
        %56 = "arith.addi"(%30, %55) : (index, index) -> index
        "memref.store"(%49, %20, %56) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
        "func.return"() : () -> ()
      }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [32, 1, 1]>, sym_name = "forward_dispatch_13_generic_64x10x9216"} : () -> ()
    }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
    "hal.executable.variant_end"() : () -> ()
  }) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
  "hal.executable_end"() : () -> ()
}) {sym_name = "forward_dispatch_13", sym_visibility = "private"} : () -> ()
    %47 = linalg.generic {indexing_maps = [#map8, #map9, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%46, %44 : tensor<2x32x10x9216xf32>, tensor<2x32x1x1xf32>) outs(%45 : tensor<2x32x10x9216xf32>) {
          ^
/home/prashant/stable.mlir:936:21: error: failed to materialize conversion for result #0 of operation 'hal.interface.constant.load' that remained live after conversion
    %expanded_730 = tensor.expand_shape %collapsed_729 [[0], [1], [2, 3]] : tensor<2x320x9216xf32> into tensor<2x320x96x96xf32>
                    ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:936:21: note: see current operation: %58 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
    %expanded_730 = tensor.expand_shape %collapsed_729 [[0], [1], [2, 3]] : tensor<2x320x9216xf32> into tensor<2x320x96x96xf32>
                    ^
/home/prashant/stable.mlir:936:21: note: see existing live user here: %80 = "spirv.UConvert"(%58) : (i32) -> i64
/home/prashant/stable.mlir:957:11: error: failed to run translation of source executable to target executable for backend #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>
    %51 = linalg.generic {indexing_maps = [#map8, #map8, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%50, %49 : tensor<2x320x96x96xf32>, tensor<2x320x96x96xf32>) outs(%25 : tensor<2x320x96x96xf32>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:957:11: note: see current operation: 
"hal.executable.variant"() ({
  "hal.executable.export"() ({
  ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index, %arg3: index, %arg4: index):
    %0 = "arith.constant"() {value = 3 : index} : () -> index
    %1 = "arith.constant"() {value = 24 : index} : () -> index
    %2 = "arith.constant"() {value = 640 : index} : () -> index
    "hal.return"(%0, %1, %2) : (index, index, index) -> ()
  }) {layout = #hal.pipeline.layout<push_constants = 4, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer, ReadOnly>, <2, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_14_generic_2x320x96x96", translation_info = #iree_codegen.translation_info<SPIRVBaseVectorize>, workgroup_size = [8 : index, 4 : index, 1 : index]} : () -> ()
  "builtin.module"() ({
    "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {binding = 2 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_2_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
    "func.func"() ({
      %0 = "arith.constant"() {value = 4 : index} : () -> index
      %1 = "arith.constant"() {value = -1 : index} : () -> index
      %2 = "arith.constant"() {value = 16 : index} : () -> index
      %3 = "arith.constant"() {value = 8 : index} : () -> index
      %4 = "arith.constant"() {value = 24 : index} : () -> index
      %5 = "arith.constant"() {value = 96 : index} : () -> index
      %6 = "arith.constant"() {value = 2304 : index} : () -> index
      %7 = "arith.constant"() {value = 737280 : index} : () -> index
      %8 = "arith.constant"() {value = 0 : index} : () -> index
      %9 = "arith.constant"() {value = 1474560 : index} : () -> index
      %10 = "arith.constant"() {value = dense<0.693147182> : vector<4xf32>} : () -> vector<4xf32>
      %11 = "arith.constant"() {value = dense<1.44269502> : vector<4xf32>} : () -> vector<4xf32>
      %12 = "arith.constant"() {value = dense<0.499705136> : vector<4xf32>} : () -> vector<4xf32>
      %13 = "arith.constant"() {value = dense<0.168738902> : vector<4xf32>} : () -> vector<4xf32>
      %14 = "arith.constant"() {value = dense<0.0366896503> : vector<4xf32>} : () -> vector<4xf32>
      %15 = "arith.constant"() {value = dense<1.314350e-02> : vector<4xf32>} : () -> vector<4xf32>
      %16 = "arith.constant"() {value = dense<23> : vector<4xi32>} : () -> vector<4xi32>
      %17 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
      %18 = "arith.constant"() {value = dense<0x7F800000> : vector<4xf32>} : () -> vector<4xf32>
      %19 = "arith.constant"() {value = dense<0xFF800000> : vector<4xf32>} : () -> vector<4xf32>
      %20 = "arith.constant"() {value = dense<1.17549435E-38> : vector<4xf32>} : () -> vector<4xf32>
      %21 = "arith.constant"() {value = dense<127> : vector<4xi32>} : () -> vector<4xi32>
      %22 = "arith.constant"() {value = dense<-127> : vector<4xi32>} : () -> vector<4xi32>
      %23 = "arith.constant"() {value = dense<1.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
      %24 = "arith.constant"() {value = 320 : index} : () -> index
      %25 = "arith.constant"() {value = 2 : index} : () -> index
      %26 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
      %27 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
      %28 = "hal.interface.constant.load"() {index = 2 : index} : () -> i32
      %29 = "hal.interface.constant.load"() {index = 3 : index} : () -> i32
      %30 = "arith.index_castui"(%26) : (i32) -> index
      %31 = "arith.index_castui"(%27) : (i32) -> index
      %32 = "arith.index_castui"(%28) : (i32) -> index
      %33 = "arith.index_castui"(%29) : (i32) -> index
      %34 = "hal.interface.binding.subspan"(%30, %9) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %35 = "hal.interface.binding.subspan"(%8, %9) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %36 = "hal.interface.binding.subspan"(%31, %24) {alignment = 64 : index, binding = 1 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %37 = "hal.interface.binding.subspan"(%8, %24) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %38 = "hal.interface.binding.subspan"(%32, %24) {alignment = 64 : index, binding = 1 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %39 = "hal.interface.binding.subspan"(%8, %24) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %40 = "hal.interface.binding.subspan"(%33, %9) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %41 = "hal.interface.binding.subspan"(%8, %9) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %42 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
      %43 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
      %44 = "hal.interface.workgroup.id"() {dimension = 2 : index} : () -> index
      %45 = "arith.remui"(%44, %24) : (index, index) -> index
      %46 = "arith.divui"(%44, %24) : (index, index) -> index
      %47 = "gpu.thread_id"() {dimension = #gpu<dim y>} : () -> index
      %48 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
      "scf.for"(%46, %25, %25) ({
      ^bb0(%arg0: index):
        "scf.for"(%45, %24, %24) ({
        ^bb0(%arg1: index):
          %49 = "arith.muli"(%arg0, %7) : (index, index) -> index
          %50 = "arith.muli"(%arg1, %6) : (index, index) -> index
          %51 = "arith.addi"(%49, %50) : (index, index) -> index
          %52 = "arith.muli"(%43, %5) : (index, index) -> index
          %53 = "arith.addi"(%51, %52) : (index, index) -> index
          %54 = "arith.muli"(%47, %4) : (index, index) -> index
          %55 = "arith.addi"(%53, %54) : (index, index) -> index
          %56 = "arith.addi"(%55, %48) : (index, index) -> index
          %57 = "arith.muli"(%42, %3) : (index, index) -> index
          %58 = "arith.addi"(%56, %57) : (index, index) -> index
          %59 = "arith.cmpi"(%30, %8) {predicate = 2 : i64} : (index, index) -> i1
          %60 = "arith.subi"(%1, %30) : (index, index) -> index
          %61 = "arith.select"(%59, %60, %30) : (i1, index, index) -> index
          %62 = "arith.divsi"(%61, %2) : (index, index) -> index
          %63 = "arith.subi"(%1, %62) : (index, index) -> index
          %64 = "arith.select"(%59, %63, %62) : (i1, index, index) -> index
          %65 = "arith.addi"(%58, %64) : (index, index) -> index
          %66 = "memref.load"(%35, %65) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
          %67 = "arith.cmpi"(%31, %8) {predicate = 2 : i64} : (index, index) -> i1
          %68 = "arith.subi"(%1, %31) : (index, index) -> index
          %69 = "arith.select"(%67, %68, %31) : (i1, index, index) -> index
          %70 = "arith.divsi"(%69, %0) : (index, index) -> index
          %71 = "arith.subi"(%1, %70) : (index, index) -> index
          %72 = "arith.select"(%67, %71, %70) : (i1, index, index) -> index
          %73 = "arith.addi"(%arg1, %72) : (index, index) -> index
          %74 = "memref.load"(%37, %73) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
          %75 = "arith.cmpi"(%32, %8) {predicate = 2 : i64} : (index, index) -> i1
          %76 = "arith.subi"(%1, %32) : (index, index) -> index
          %77 = "arith.select"(%75, %76, %32) : (i1, index, index) -> index
          %78 = "arith.divsi"(%77, %0) : (index, index) -> index
          %79 = "arith.subi"(%1, %78) : (index, index) -> index
          %80 = "arith.select"(%75, %79, %78) : (i1, index, index) -> index
          %81 = "arith.addi"(%arg1, %80) : (index, index) -> index
          %82 = "memref.load"(%39, %81) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
          %83 = "vector.splat"(%74) : (f32) -> vector<4xf32>
          %84 = "arith.mulf"(%66, %83) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %85 = "vector.splat"(%82) : (f32) -> vector<4xf32>
          %86 = "arith.addf"(%84, %85) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %87 = "arith.negf"(%86) {fastmath = #arith.fastmath<none>} : (vector<4xf32>) -> vector<4xf32>
          %88 = "arith.cmpf"(%87, %87) {predicate = 14 : i64} : (vector<4xf32>, vector<4xf32>) -> vector<4xi1>
          %89 = "arith.mulf"(%87, %11) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %90 = "math.floor"(%89) {fastmath = #arith.fastmath<none>} : (vector<4xf32>) -> vector<4xf32>
          %91 = "arith.mulf"(%90, %10) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %92 = "arith.subf"(%87, %91) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %93 = "arith.mulf"(%92, %92) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %94 = "arith.mulf"(%93, %93) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %95 = "math.fma"(%23, %92, %23) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %96 = "math.fma"(%13, %92, %12) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %97 = "math.fma"(%15, %92, %14) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %98 = "math.fma"(%96, %93, %95) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %99 = "math.fma"(%97, %94, %98) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %100 = "arith.fptosi"(%90) : (vector<4xf32>) -> vector<4xi32>
          %101 = "arith.addi"(%100, %21) : (vector<4xi32>, vector<4xi32>) -> vector<4xi32>
          %102 = "arith.shli"(%101, %16) : (vector<4xi32>, vector<4xi32>) -> vector<4xi32>
          %103 = "arith.bitcast"(%102) : (vector<4xi32>) -> vector<4xf32>
          %104 = "arith.mulf"(%99, %103) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %105 = "arith.cmpi"(%100, %21) {predicate = 3 : i64} : (vector<4xi32>, vector<4xi32>) -> vector<4xi1>
          %106 = "arith.cmpi"(%100, %22) {predicate = 5 : i64} : (vector<4xi32>, vector<4xi32>) -> vector<4xi1>
          %107 = "arith.cmpf"(%87, %19) {predicate = 1 : i64} : (vector<4xf32>, vector<4xf32>) -> vector<4xi1>
          %108 = "arith.cmpf"(%87, %18) {predicate = 1 : i64} : (vector<4xf32>, vector<4xf32>) -> vector<4xi1>
          %109 = "arith.cmpf"(%87, %17) {predicate = 2 : i64} : (vector<4xf32>, vector<4xf32>) -> vector<4xi1>
          %110 = "arith.andi"(%105, %106) : (vector<4xi1>, vector<4xi1>) -> vector<4xi1>
          %111 = "arith.select"(%109, %18, %20) : (vector<4xi1>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %112 = "arith.select"(%110, %104, %111) : (vector<4xi1>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %113 = "arith.select"(%108, %18, %112) : (vector<4xi1>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %114 = "arith.select"(%107, %17, %113) : (vector<4xi1>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %115 = "arith.select"(%88, %87, %114) : (vector<4xi1>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %116 = "arith.addf"(%115, %23) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %117 = "arith.divf"(%23, %116) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %118 = "arith.mulf"(%117, %86) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %119 = "arith.cmpi"(%33, %8) {predicate = 2 : i64} : (index, index) -> i1
          %120 = "arith.subi"(%1, %33) : (index, index) -> index
          %121 = "arith.select"(%119, %120, %33) : (i1, index, index) -> index
          %122 = "arith.divsi"(%121, %2) : (index, index) -> index
          %123 = "arith.subi"(%1, %122) : (index, index) -> index
          %124 = "arith.select"(%119, %123, %122) : (i1, index, index) -> index
          %125 = "arith.addi"(%58, %124) : (index, index) -> index
          "memref.store"(%118, %41, %125) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
          "scf.yield"() : () -> ()
        }) : (index, index, index) -> ()
        "scf.yield"() : () -> ()
      }) : (index, index, index) -> ()
      "func.return"() : () -> ()
    }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [8, 4, 1]>, sym_name = "forward_dispatch_14_generic_2x320x96x96"} : () -> ()
  }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
  "hal.executable.variant_end"() : () -> ()
}) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
    %51 = linalg.generic {indexing_maps = [#map8, #map8, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%50, %49 : tensor<2x320x96x96xf32>, tensor<2x320x96x96xf32>) outs(%25 : tensor<2x320x96x96xf32>) {
          ^
/home/prashant/stable.mlir:957:11: error: failed to serialize executables
    %51 = linalg.generic {indexing_maps = [#map8, #map8, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%50, %49 : tensor<2x320x96x96xf32>, tensor<2x320x96x96xf32>) outs(%25 : tensor<2x320x96x96xf32>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:957:11: note: see current operation: 
"hal.executable"() ({
  "hal.executable.variant"() ({
    "hal.executable.export"() ({
    ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index, %arg3: index, %arg4: index):
      %0 = "arith.constant"() {value = 3 : index} : () -> index
      %1 = "arith.constant"() {value = 24 : index} : () -> index
      %2 = "arith.constant"() {value = 640 : index} : () -> index
      "hal.return"(%0, %1, %2) : (index, index, index) -> ()
    }) {layout = #hal.pipeline.layout<push_constants = 4, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer, ReadOnly>, <2, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_14_generic_2x320x96x96", translation_info = #iree_codegen.translation_info<SPIRVBaseVectorize>, workgroup_size = [8 : index, 4 : index, 1 : index]} : () -> ()
    "builtin.module"() ({
      "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {binding = 2 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_2_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
      "func.func"() ({
        %0 = "arith.constant"() {value = 4 : index} : () -> index
        %1 = "arith.constant"() {value = -1 : index} : () -> index
        %2 = "arith.constant"() {value = 16 : index} : () -> index
        %3 = "arith.constant"() {value = 8 : index} : () -> index
        %4 = "arith.constant"() {value = 24 : index} : () -> index
        %5 = "arith.constant"() {value = 96 : index} : () -> index
        %6 = "arith.constant"() {value = 2304 : index} : () -> index
        %7 = "arith.constant"() {value = 737280 : index} : () -> index
        %8 = "arith.constant"() {value = 0 : index} : () -> index
        %9 = "arith.constant"() {value = 1474560 : index} : () -> index
        %10 = "arith.constant"() {value = dense<0.693147182> : vector<4xf32>} : () -> vector<4xf32>
        %11 = "arith.constant"() {value = dense<1.44269502> : vector<4xf32>} : () -> vector<4xf32>
        %12 = "arith.constant"() {value = dense<0.499705136> : vector<4xf32>} : () -> vector<4xf32>
        %13 = "arith.constant"() {value = dense<0.168738902> : vector<4xf32>} : () -> vector<4xf32>
        %14 = "arith.constant"() {value = dense<0.0366896503> : vector<4xf32>} : () -> vector<4xf32>
        %15 = "arith.constant"() {value = dense<1.314350e-02> : vector<4xf32>} : () -> vector<4xf32>
        %16 = "arith.constant"() {value = dense<23> : vector<4xi32>} : () -> vector<4xi32>
        %17 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
        %18 = "arith.constant"() {value = dense<0x7F800000> : vector<4xf32>} : () -> vector<4xf32>
        %19 = "arith.constant"() {value = dense<0xFF800000> : vector<4xf32>} : () -> vector<4xf32>
        %20 = "arith.constant"() {value = dense<1.17549435E-38> : vector<4xf32>} : () -> vector<4xf32>
        %21 = "arith.constant"() {value = dense<127> : vector<4xi32>} : () -> vector<4xi32>
        %22 = "arith.constant"() {value = dense<-127> : vector<4xi32>} : () -> vector<4xi32>
        %23 = "arith.constant"() {value = dense<1.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
        %24 = "arith.constant"() {value = 320 : index} : () -> index
        %25 = "arith.constant"() {value = 2 : index} : () -> index
        %26 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
        %27 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
        %28 = "hal.interface.constant.load"() {index = 2 : index} : () -> i32
        %29 = "hal.interface.constant.load"() {index = 3 : index} : () -> i32
        %30 = "arith.index_castui"(%26) : (i32) -> index
        %31 = "arith.index_castui"(%27) : (i32) -> index
        %32 = "arith.index_castui"(%28) : (i32) -> index
        %33 = "arith.index_castui"(%29) : (i32) -> index
        %34 = "hal.interface.binding.subspan"(%30, %9) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %35 = "hal.interface.binding.subspan"(%8, %9) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %36 = "hal.interface.binding.subspan"(%31, %24) {alignment = 64 : index, binding = 1 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %37 = "hal.interface.binding.subspan"(%8, %24) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %38 = "hal.interface.binding.subspan"(%32, %24) {alignment = 64 : index, binding = 1 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %39 = "hal.interface.binding.subspan"(%8, %24) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %40 = "hal.interface.binding.subspan"(%33, %9) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %41 = "hal.interface.binding.subspan"(%8, %9) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %42 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
        %43 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
        %44 = "hal.interface.workgroup.id"() {dimension = 2 : index} : () -> index
        %45 = "arith.remui"(%44, %24) : (index, index) -> index
        %46 = "arith.divui"(%44, %24) : (index, index) -> index
        %47 = "gpu.thread_id"() {dimension = #gpu<dim y>} : () -> index
        %48 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
        "scf.for"(%46, %25, %25) ({
        ^bb0(%arg0: index):
          "scf.for"(%45, %24, %24) ({
          ^bb0(%arg1: index):
            %49 = "arith.muli"(%arg0, %7) : (index, index) -> index
            %50 = "arith.muli"(%arg1, %6) : (index, index) -> index
            %51 = "arith.addi"(%49, %50) : (index, index) -> index
            %52 = "arith.muli"(%43, %5) : (index, index) -> index
            %53 = "arith.addi"(%51, %52) : (index, index) -> index
            %54 = "arith.muli"(%47, %4) : (index, index) -> index
            %55 = "arith.addi"(%53, %54) : (index, index) -> index
            %56 = "arith.addi"(%55, %48) : (index, index) -> index
            %57 = "arith.muli"(%42, %3) : (index, index) -> index
            %58 = "arith.addi"(%56, %57) : (index, index) -> index
            %59 = "arith.cmpi"(%30, %8) {predicate = 2 : i64} : (index, index) -> i1
            %60 = "arith.subi"(%1, %30) : (index, index) -> index
            %61 = "arith.select"(%59, %60, %30) : (i1, index, index) -> index
            %62 = "arith.divsi"(%61, %2) : (index, index) -> index
            %63 = "arith.subi"(%1, %62) : (index, index) -> index
            %64 = "arith.select"(%59, %63, %62) : (i1, index, index) -> index
            %65 = "arith.addi"(%58, %64) : (index, index) -> index
            %66 = "memref.load"(%35, %65) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
            %67 = "arith.cmpi"(%31, %8) {predicate = 2 : i64} : (index, index) -> i1
            %68 = "arith.subi"(%1, %31) : (index, index) -> index
            %69 = "arith.select"(%67, %68, %31) : (i1, index, index) -> index
            %70 = "arith.divsi"(%69, %0) : (index, index) -> index
            %71 = "arith.subi"(%1, %70) : (index, index) -> index
            %72 = "arith.select"(%67, %71, %70) : (i1, index, index) -> index
            %73 = "arith.addi"(%arg1, %72) : (index, index) -> index
            %74 = "memref.load"(%37, %73) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
            %75 = "arith.cmpi"(%32, %8) {predicate = 2 : i64} : (index, index) -> i1
            %76 = "arith.subi"(%1, %32) : (index, index) -> index
            %77 = "arith.select"(%75, %76, %32) : (i1, index, index) -> index
            %78 = "arith.divsi"(%77, %0) : (index, index) -> index
            %79 = "arith.subi"(%1, %78) : (index, index) -> index
            %80 = "arith.select"(%75, %79, %78) : (i1, index, index) -> index
            %81 = "arith.addi"(%arg1, %80) : (index, index) -> index
            %82 = "memref.load"(%39, %81) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
            %83 = "vector.splat"(%74) : (f32) -> vector<4xf32>
            %84 = "arith.mulf"(%66, %83) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %85 = "vector.splat"(%82) : (f32) -> vector<4xf32>
            %86 = "arith.addf"(%84, %85) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %87 = "arith.negf"(%86) {fastmath = #arith.fastmath<none>} : (vector<4xf32>) -> vector<4xf32>
            %88 = "arith.cmpf"(%87, %87) {predicate = 14 : i64} : (vector<4xf32>, vector<4xf32>) -> vector<4xi1>
            %89 = "arith.mulf"(%87, %11) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %90 = "math.floor"(%89) {fastmath = #arith.fastmath<none>} : (vector<4xf32>) -> vector<4xf32>
            %91 = "arith.mulf"(%90, %10) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %92 = "arith.subf"(%87, %91) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %93 = "arith.mulf"(%92, %92) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %94 = "arith.mulf"(%93, %93) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %95 = "math.fma"(%23, %92, %23) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %96 = "math.fma"(%13, %92, %12) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %97 = "math.fma"(%15, %92, %14) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %98 = "math.fma"(%96, %93, %95) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %99 = "math.fma"(%97, %94, %98) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %100 = "arith.fptosi"(%90) : (vector<4xf32>) -> vector<4xi32>
            %101 = "arith.addi"(%100, %21) : (vector<4xi32>, vector<4xi32>) -> vector<4xi32>
            %102 = "arith.shli"(%101, %16) : (vector<4xi32>, vector<4xi32>) -> vector<4xi32>
            %103 = "arith.bitcast"(%102) : (vector<4xi32>) -> vector<4xf32>
            %104 = "arith.mulf"(%99, %103) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %105 = "arith.cmpi"(%100, %21) {predicate = 3 : i64} : (vector<4xi32>, vector<4xi32>) -> vector<4xi1>
            %106 = "arith.cmpi"(%100, %22) {predicate = 5 : i64} : (vector<4xi32>, vector<4xi32>) -> vector<4xi1>
            %107 = "arith.cmpf"(%87, %19) {predicate = 1 : i64} : (vector<4xf32>, vector<4xf32>) -> vector<4xi1>
            %108 = "arith.cmpf"(%87, %18) {predicate = 1 : i64} : (vector<4xf32>, vector<4xf32>) -> vector<4xi1>
            %109 = "arith.cmpf"(%87, %17) {predicate = 2 : i64} : (vector<4xf32>, vector<4xf32>) -> vector<4xi1>
            %110 = "arith.andi"(%105, %106) : (vector<4xi1>, vector<4xi1>) -> vector<4xi1>
            %111 = "arith.select"(%109, %18, %20) : (vector<4xi1>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %112 = "arith.select"(%110, %104, %111) : (vector<4xi1>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %113 = "arith.select"(%108, %18, %112) : (vector<4xi1>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %114 = "arith.select"(%107, %17, %113) : (vector<4xi1>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %115 = "arith.select"(%88, %87, %114) : (vector<4xi1>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %116 = "arith.addf"(%115, %23) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %117 = "arith.divf"(%23, %116) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %118 = "arith.mulf"(%117, %86) {fastmath = #arith.fastmath<none>} : (vector<4xf32>, vector<4xf32>) -> vector<4xf32>
            %119 = "arith.cmpi"(%33, %8) {predicate = 2 : i64} : (index, index) -> i1
            %120 = "arith.subi"(%1, %33) : (index, index) -> index
            %121 = "arith.select"(%119, %120, %33) : (i1, index, index) -> index
            %122 = "arith.divsi"(%121, %2) : (index, index) -> index
            %123 = "arith.subi"(%1, %122) : (index, index) -> index
            %124 = "arith.select"(%119, %123, %122) : (i1, index, index) -> index
            %125 = "arith.addi"(%58, %124) : (index, index) -> index
            "memref.store"(%118, %41, %125) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
            "scf.yield"() : () -> ()
          }) : (index, index, index) -> ()
          "scf.yield"() : () -> ()
        }) : (index, index, index) -> ()
        "func.return"() : () -> ()
      }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [8, 4, 1]>, sym_name = "forward_dispatch_14_generic_2x320x96x96"} : () -> ()
    }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
    "hal.executable.variant_end"() : () -> ()
  }) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
  "hal.executable_end"() : () -> ()
}) {sym_name = "forward_dispatch_14", sym_visibility = "private"} : () -> ()
    %51 = linalg.generic {indexing_maps = [#map8, #map8, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%50, %49 : tensor<2x320x96x96xf32>, tensor<2x320x96x96xf32>) outs(%25 : tensor<2x320x96x96xf32>) {
          ^
/home/prashant/stable.mlir:957:11: error: failed to materialize conversion for result #0 of operation 'hal.interface.constant.load' that remained live after conversion
    %51 = linalg.generic {indexing_maps = [#map8, #map8, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%50, %49 : tensor<2x320x96x96xf32>, tensor<2x320x96x96xf32>) outs(%25 : tensor<2x320x96x96xf32>) {
          ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:957:11: note: see current operation: %38 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
    %51 = linalg.generic {indexing_maps = [#map8, #map8, #map8], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%50, %49 : tensor<2x320x96x96xf32>, tensor<2x320x96x96xf32>) outs(%25 : tensor<2x320x96x96xf32>) {
          ^
/home/prashant/stable.mlir:957:11: note: see existing live user here: %39 = "spirv.UConvert"(%38) : (i32) -> i64
/home/prashant/stable.mlir:962:19: error: failed to run translation of source executable to target executable for backend #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>
    %padded_733 = tensor.pad %51 low[0, 0, 1, 1] high[0, 0, 1, 1] {
                  ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:962:19: note: see current operation: 
"hal.executable.variant"() ({
  "hal.executable.export"() ({
  ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index, %arg3: index, %arg4: index):
    %0 = "arith.constant"() {value = 3 : index} : () -> index
    %1 = "arith.constant"() {value = 96 : index} : () -> index
    %2 = "arith.constant"() {value = 640 : index} : () -> index
    "hal.return"(%0, %1, %2) : (index, index, index) -> ()
  }) {layout = #hal.pipeline.layout<push_constants = 1, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_15", translation_info = #iree_codegen.translation_info<SPIRVBaseDistribute>, workgroup_size = [32 : index, 1 : index, 1 : index]} : () -> ()
  "builtin.module"() ({
    "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
    "func.func"() ({
      %0 = "arith.constant"() {value = 17845699 : index} : () -> index
      %1 = "arith.constant"() {value = 98 : index} : () -> index
      %2 = "arith.constant"() {value = 9604 : index} : () -> index
      %3 = "arith.constant"() {value = 3073280 : index} : () -> index
      %4 = "arith.constant"() {value = -1 : index} : () -> index
      %5 = "arith.constant"() {value = 4 : index} : () -> index
      %6 = "arith.constant"() {value = 32 : index} : () -> index
      %7 = "arith.constant"() {value = 96 : index} : () -> index
      %8 = "arith.constant"() {value = 9216 : index} : () -> index
      %9 = "arith.constant"() {value = 2949120 : index} : () -> index
      %10 = "arith.constant"() {value = 0 : index} : () -> index
      %11 = "arith.constant"() {value = 6146560 : index} : () -> index
      %12 = "arith.constant"() {value = 5898240 : index} : () -> index
      %13 = "arith.constant"() {value = 320 : index} : () -> index
      %14 = "arith.constant"() {value = 2 : index} : () -> index
      %15 = "arith.constant"() {value = 71382400 : index} : () -> index
      %16 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
      %17 = "arith.index_castui"(%16) : (i32) -> index
      %18 = "hal.interface.binding.subspan"(%17, %12) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %19 = "hal.interface.binding.subspan"(%10, %12) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %20 = "hal.interface.binding.subspan"(%15, %11) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %21 = "hal.interface.binding.subspan"(%10, %11) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
      %22 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
      %23 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
      %24 = "hal.interface.workgroup.id"() {dimension = 2 : index} : () -> index
      %25 = "arith.remui"(%24, %13) : (index, index) -> index
      %26 = "arith.divui"(%24, %13) : (index, index) -> index
      %27 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
      %28 = "gpu.thread_id"() {dimension = #gpu<dim y>} : () -> index
      %29 = "gpu.thread_id"() {dimension = #gpu<dim z>} : () -> index
      "scf.for"(%26, %14, %14) ({
      ^bb0(%arg0: index):
        "scf.for"(%25, %13, %13) ({
        ^bb0(%arg1: index):
          %30 = "arith.muli"(%arg0, %9) : (index, index) -> index
          %31 = "arith.muli"(%arg1, %8) : (index, index) -> index
          %32 = "arith.addi"(%30, %31) : (index, index) -> index
          %33 = "arith.muli"(%29, %8) : (index, index) -> index
          %34 = "arith.addi"(%32, %33) : (index, index) -> index
          %35 = "arith.muli"(%23, %7) : (index, index) -> index
          %36 = "arith.addi"(%34, %35) : (index, index) -> index
          %37 = "arith.muli"(%28, %7) : (index, index) -> index
          %38 = "arith.addi"(%36, %37) : (index, index) -> index
          %39 = "arith.muli"(%22, %6) : (index, index) -> index
          %40 = "arith.addi"(%38, %39) : (index, index) -> index
          %41 = "arith.addi"(%40, %27) : (index, index) -> index
          %42 = "arith.cmpi"(%17, %10) {predicate = 2 : i64} : (index, index) -> i1
          %43 = "arith.subi"(%4, %17) : (index, index) -> index
          %44 = "arith.select"(%42, %43, %17) : (i1, index, index) -> index
          %45 = "arith.divsi"(%44, %5) : (index, index) -> index
          %46 = "arith.subi"(%4, %45) : (index, index) -> index
          %47 = "arith.select"(%42, %46, %45) : (i1, index, index) -> index
          %48 = "arith.addi"(%41, %47) : (index, index) -> index
          %49 = "memref.load"(%19, %48) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
          %50 = "arith.muli"(%arg0, %3) : (index, index) -> index
          %51 = "arith.muli"(%arg1, %2) : (index, index) -> index
          %52 = "arith.addi"(%50, %51) : (index, index) -> index
          %53 = "arith.muli"(%29, %2) : (index, index) -> index
          %54 = "arith.addi"(%52, %53) : (index, index) -> index
          %55 = "arith.muli"(%23, %1) : (index, index) -> index
          %56 = "arith.addi"(%54, %55) : (index, index) -> index
          %57 = "arith.muli"(%28, %1) : (index, index) -> index
          %58 = "arith.addi"(%56, %57) : (index, index) -> index
          %59 = "arith.addi"(%58, %39) : (index, index) -> index
          %60 = "arith.addi"(%59, %27) : (index, index) -> index
          %61 = "arith.addi"(%60, %0) : (index, index) -> index
          "memref.store"(%49, %21, %61) : (f32, memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> ()
          "scf.yield"() : () -> ()
        }) : (index, index, index) -> ()
        "scf.yield"() : () -> ()
      }) : (index, index, index) -> ()
      "func.return"() : () -> ()
    }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [32, 1, 1]>, sym_name = "forward_dispatch_15"} : () -> ()
  }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
  "hal.executable.variant_end"() : () -> ()
}) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
    %padded_733 = tensor.pad %51 low[0, 0, 1, 1] high[0, 0, 1, 1] {
                  ^
/home/prashant/stable.mlir:962:19: error: failed to serialize executables
    %padded_733 = tensor.pad %51 low[0, 0, 1, 1] high[0, 0, 1, 1] {
                  ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:962:19: note: see current operation: 
"hal.executable"() ({
  "hal.executable.variant"() ({
    "hal.executable.export"() ({
    ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index, %arg3: index, %arg4: index):
      %0 = "arith.constant"() {value = 3 : index} : () -> index
      %1 = "arith.constant"() {value = 96 : index} : () -> index
      %2 = "arith.constant"() {value = 640 : index} : () -> index
      "hal.return"(%0, %1, %2) : (index, index, index) -> ()
    }) {layout = #hal.pipeline.layout<push_constants = 1, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_15", translation_info = #iree_codegen.translation_info<SPIRVBaseDistribute>, workgroup_size = [32 : index, 1 : index, 1 : index]} : () -> ()
    "builtin.module"() ({
      "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<f32, stride=4> [0])>, StorageBuffer>} : () -> ()
      "func.func"() ({
        %0 = "arith.constant"() {value = 17845699 : index} : () -> index
        %1 = "arith.constant"() {value = 98 : index} : () -> index
        %2 = "arith.constant"() {value = 9604 : index} : () -> index
        %3 = "arith.constant"() {value = 3073280 : index} : () -> index
        %4 = "arith.constant"() {value = -1 : index} : () -> index
        %5 = "arith.constant"() {value = 4 : index} : () -> index
        %6 = "arith.constant"() {value = 32 : index} : () -> index
        %7 = "arith.constant"() {value = 96 : index} : () -> index
        %8 = "arith.constant"() {value = 9216 : index} : () -> index
        %9 = "arith.constant"() {value = 2949120 : index} : () -> index
        %10 = "arith.constant"() {value = 0 : index} : () -> index
        %11 = "arith.constant"() {value = 6146560 : index} : () -> index
        %12 = "arith.constant"() {value = 5898240 : index} : () -> index
        %13 = "arith.constant"() {value = 320 : index} : () -> index
        %14 = "arith.constant"() {value = 2 : index} : () -> index
        %15 = "arith.constant"() {value = 71382400 : index} : () -> index
        %16 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
        %17 = "arith.index_castui"(%16) : (i32) -> index
        %18 = "hal.interface.binding.subspan"(%17, %12) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %19 = "hal.interface.binding.subspan"(%10, %12) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %20 = "hal.interface.binding.subspan"(%15, %11) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %21 = "hal.interface.binding.subspan"(%10, %11) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xf32, #spirv.storage_class<StorageBuffer>>
        %22 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
        %23 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
        %24 = "hal.interface.workgroup.id"() {dimension = 2 : index} : () -> index
        %25 = "arith.remui"(%24, %13) : (index, index) -> index
        %26 = "arith.divui"(%24, %13) : (index, index) -> index
        %27 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
        %28 = "gpu.thread_id"() {dimension = #gpu<dim y>} : () -> index
        %29 = "gpu.thread_id"() {dimension = #gpu<dim z>} : () -> index
        "scf.for"(%26, %14, %14) ({
        ^bb0(%arg0: index):
          "scf.for"(%25, %13, %13) ({
          ^bb0(%arg1: index):
            %30 = "arith.muli"(%arg0, %9) : (index, index) -> index
            %31 = "arith.muli"(%arg1, %8) : (index, index) -> index
            %32 = "arith.addi"(%30, %31) : (index, index) -> index
            %33 = "arith.muli"(%29, %8) : (index, index) -> index
            %34 = "arith.addi"(%32, %33) : (index, index) -> index
            %35 = "arith.muli"(%23, %7) : (index, index) -> index
            %36 = "arith.addi"(%34, %35) : (index, index) -> index
            %37 = "arith.muli"(%28, %7) : (index, index) -> index
            %38 = "arith.addi"(%36, %37) : (index, index) -> index
            %39 = "arith.muli"(%22, %6) : (index, index) -> index
            %40 = "arith.addi"(%38, %39) : (index, index) -> index
            %41 = "arith.addi"(%40, %27) : (index, index) -> index
            %42 = "arith.cmpi"(%17, %10) {predicate = 2 : i64} : (index, index) -> i1
            %43 = "arith.subi"(%4, %17) : (index, index) -> index
            %44 = "arith.select"(%42, %43, %17) : (i1, index, index) -> index
            %45 = "arith.divsi"(%44, %5) : (index, index) -> index
            %46 = "arith.subi"(%4, %45) : (index, index) -> index
            %47 = "arith.select"(%42, %46, %45) : (i1, index, index) -> index
            %48 = "arith.addi"(%41, %47) : (index, index) -> index
            %49 = "memref.load"(%19, %48) : (memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> f32
            %50 = "arith.muli"(%arg0, %3) : (index, index) -> index
            %51 = "arith.muli"(%arg1, %2) : (index, index) -> index
            %52 = "arith.addi"(%50, %51) : (index, index) -> index
            %53 = "arith.muli"(%29, %2) : (index, index) -> index
            %54 = "arith.addi"(%52, %53) : (index, index) -> index
            %55 = "arith.muli"(%23, %1) : (index, index) -> index
            %56 = "arith.addi"(%54, %55) : (index, index) -> index
            %57 = "arith.muli"(%28, %1) : (index, index) -> index
            %58 = "arith.addi"(%56, %57) : (index, index) -> index
            %59 = "arith.addi"(%58, %39) : (index, index) -> index
            %60 = "arith.addi"(%59, %27) : (index, index) -> index
            %61 = "arith.addi"(%60, %0) : (index, index) -> index
            "memref.store"(%49, %21, %61) : (f32, memref<?xf32, #spirv.storage_class<StorageBuffer>>, index) -> ()
            "scf.yield"() : () -> ()
          }) : (index, index, index) -> ()
          "scf.yield"() : () -> ()
        }) : (index, index, index) -> ()
        "func.return"() : () -> ()
      }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [32, 1, 1]>, sym_name = "forward_dispatch_15"} : () -> ()
    }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
    "hal.executable.variant_end"() : () -> ()
  }) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
  "hal.executable_end"() : () -> ()
}) {sym_name = "forward_dispatch_15", sym_visibility = "private"} : () -> ()
    %padded_733 = tensor.pad %51 low[0, 0, 1, 1] high[0, 0, 1, 1] {
                  ^
/home/prashant/stable.mlir:1313:12: error: failed to materialize conversion for result #0 of operation 'hal.interface.constant.load' that remained live after conversion
    %130 = linalg.generic {indexing_maps = [#map4, #map5], iterator_types = ["parallel", "parallel"]} ins(%cst_662 : tensor<320x320xf32>) outs(%101 : tensor<320x320xf32>) {
           ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:1313:12: note: see current operation: %187 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
    %130 = linalg.generic {indexing_maps = [#map4, #map5], iterator_types = ["parallel", "parallel"]} ins(%cst_662 : tensor<320x320xf32>) outs(%101 : tensor<320x320xf32>) {
           ^
/home/prashant/stable.mlir:1313:12: note: see existing live user here: %195 = "spirv.UConvert"(%187) : (i32) -> i64
/home/prashant/stable.mlir:1320:12: error: failed to run translation of source executable to target executable for backend #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>
    %133 = linalg.matmul ins(%collapsed_749, %130 : tensor<18432x320xf32>, tensor<320x320xf32>) outs(%132 : tensor<18432x320xf32>) -> tensor<18432x320xf32>
           ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:1320:12: note: see current operation: 
"hal.executable.variant"() ({
  "hal.executable.export"() ({
  ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index, %arg3: index):
    %0 = "arith.constant"() {value = 5 : index} : () -> index
    %1 = "arith.constant"() {value = 288 : index} : () -> index
    %2 = "arith.constant"() {value = 1 : index} : () -> index
    "hal.return"(%0, %1, %2) : (index, index, index) -> ()
  }) {layout = #hal.pipeline.layout<push_constants = 2, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer, ReadOnly>, <2, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_35_matmul_18432x320x320", translation_info = #iree_codegen.translation_info<SPIRVMatmulPromoteVectorize pipeline_depth = 1>, workgroup_size = [16 : index, 16 : index, 1 : index]} : () -> ()
  "builtin.module"() ({
    "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
    "spirv.GlobalVariable"() {binding = 2 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_2_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
    "func.func"() ({
      %0 = "arith.constant"() {value = -33 : index} : () -> index
      %1 = "arith.constant"() {value = 527 : index} : () -> index
      %2 = "arith.constant"() {value = 510 : index} : () -> index
      %3 = "arith.constant"() {value = 493 : index} : () -> index
      %4 = "arith.constant"() {value = 476 : index} : () -> index
      %5 = "arith.constant"() {value = 459 : index} : () -> index
      %6 = "arith.constant"() {value = 442 : index} : () -> index
      %7 = "arith.constant"() {value = 425 : index} : () -> index
      %8 = "arith.constant"() {value = 408 : index} : () -> index
      %9 = "arith.constant"() {value = 391 : index} : () -> index
      %10 = "arith.constant"() {value = 374 : index} : () -> index
      %11 = "arith.constant"() {value = 357 : index} : () -> index
      %12 = "arith.constant"() {value = 340 : index} : () -> index
      %13 = "arith.constant"() {value = 323 : index} : () -> index
      %14 = "arith.constant"() {value = 306 : index} : () -> index
      %15 = "arith.constant"() {value = 289 : index} : () -> index
      %16 = "arith.constant"() {value = 255 : index} : () -> index
      %17 = "arith.constant"() {value = 238 : index} : () -> index
      %18 = "arith.constant"() {value = 221 : index} : () -> index
      %19 = "arith.constant"() {value = 204 : index} : () -> index
      %20 = "arith.constant"() {value = 187 : index} : () -> index
      %21 = "arith.constant"() {value = 170 : index} : () -> index
      %22 = "arith.constant"() {value = 153 : index} : () -> index
      %23 = "arith.constant"() {value = 136 : index} : () -> index
      %24 = "arith.constant"() {value = 119 : index} : () -> index
      %25 = "arith.constant"() {value = 102 : index} : () -> index
      %26 = "arith.constant"() {value = 85 : index} : () -> index
      %27 = "arith.constant"() {value = 68 : index} : () -> index
      %28 = "arith.constant"() {value = 51 : index} : () -> index
      %29 = "arith.constant"() {value = 34 : index} : () -> index
      %30 = "arith.constant"() {value = 33 : index} : () -> index
      %31 = "arith.constant"() {value = 31 : index} : () -> index
      %32 = "arith.constant"() {value = 30 : index} : () -> index
      %33 = "arith.constant"() {value = 29 : index} : () -> index
      %34 = "arith.constant"() {value = 28 : index} : () -> index
      %35 = "arith.constant"() {value = 27 : index} : () -> index
      %36 = "arith.constant"() {value = 25 : index} : () -> index
      %37 = "arith.constant"() {value = 24 : index} : () -> index
      %38 = "arith.constant"() {value = 23 : index} : () -> index
      %39 = "arith.constant"() {value = 22 : index} : () -> index
      %40 = "arith.constant"() {value = 21 : index} : () -> index
      %41 = "arith.constant"() {value = 20 : index} : () -> index
      %42 = "arith.constant"() {value = 19 : index} : () -> index
      %43 = "arith.constant"() {value = 15 : index} : () -> index
      %44 = "arith.constant"() {value = 14 : index} : () -> index
      %45 = "arith.constant"() {value = 13 : index} : () -> index
      %46 = "arith.constant"() {value = 12 : index} : () -> index
      %47 = "arith.constant"() {value = 11 : index} : () -> index
      %48 = "arith.constant"() {value = 10 : index} : () -> index
      %49 = "arith.constant"() {value = 9 : index} : () -> index
      %50 = "arith.constant"() {value = 7 : index} : () -> index
      %51 = "arith.constant"() {value = 6 : index} : () -> index
      %52 = "arith.constant"() {value = 5 : index} : () -> index
      %53 = "arith.constant"() {value = 4 : index} : () -> index
      %54 = "arith.constant"() {value = 3 : index} : () -> index
      %55 = "arith.constant"() {value = 2 : index} : () -> index
      %56 = "arith.constant"() {value = 1 : index} : () -> index
      %57 = "arith.constant"() {value = 36 : index} : () -> index
      %58 = "arith.constant"() {value = 272 : index} : () -> index
      %59 = "arith.constant"() {value = 17 : index} : () -> index
      %60 = "arith.constant"() {value = 18 : index} : () -> index
      %61 = "arith.constant"() {value = 64 : index} : () -> index
      %62 = "arith.constant"() {value = 1280 : index} : () -> index
      %63 = "arith.constant"() {value = 1477120 : index} : () -> index
      %64 = "arith.constant"() {value = 72 : index} : () -> index
      %65 = "arith.constant"() {value = 8 : index} : () -> index
      %66 = "arith.constant"() {value = 2560 : index} : () -> index
      %67 = "arith.constant"() {value = 240 : index} : () -> index
      %68 = "arith.constant"() {value = 160 : index} : () -> index
      %69 = "arith.constant"() {value = 80 : index} : () -> index
      %70 = "arith.constant"() {value = -1 : index} : () -> index
      %71 = "arith.constant"() {value = 16 : index} : () -> index
      %72 = "arith.constant"() {value = 320 : index} : () -> index
      %73 = "arith.constant"() {value = 5120 : index} : () -> index
      %74 = "arith.constant"() {value = 0 : index} : () -> index
      %75 = "arith.constant"() {value = 1474560 : index} : () -> index
      %76 = "arith.constant"() {value = 25600 : index} : () -> index
      %77 = "arith.constant"() {value = 288 : index} : () -> index
      %78 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
      %79 = "arith.constant"() {value = 23592960 : index} : () -> index
      %80 = "arith.constant"() {value = 32 : index} : () -> index
      %81 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
      %82 = "gpu.thread_id"() {dimension = #gpu<dim y>} : () -> index
      %83 = "gpu.thread_id"() {dimension = #gpu<dim z>} : () -> index
      %84 = "memref.alloc"() {operand_segment_sizes = array<i32: 0, 0>} : () -> memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>
      %85 = "memref.alloc"() {operand_segment_sizes = array<i32: 0, 0>} : () -> memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>
      %86 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
      %87 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
      %88 = "arith.index_castui"(%86) : (i32) -> index
      %89 = "arith.index_castui"(%87) : (i32) -> index
      %90 = "hal.interface.binding.subspan"(%79, %75) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %91 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %92 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %93 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %94 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %95 = "hal.interface.binding.subspan"(%88, %76) {alignment = 64 : index, binding = 1 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %96 = "hal.interface.binding.subspan"(%74, %76) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %97 = "hal.interface.binding.subspan"(%74, %76) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %98 = "hal.interface.binding.subspan"(%74, %76) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %99 = "hal.interface.binding.subspan"(%74, %76) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %100 = "hal.interface.binding.subspan"(%89, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %101 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %102 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %103 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %104 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %105 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %106 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %107 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %108 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %109 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %110 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %111 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %112 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
      %113 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
      %114 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
      %115 = "arith.muli"(%114, %73) : (index, index) -> index
      %116 = "arith.muli"(%82, %72) : (index, index) -> index
      %117 = "arith.addi"(%115, %116) : (index, index) -> index
      %118 = "arith.muli"(%113, %71) : (index, index) -> index
      %119 = "arith.addi"(%117, %118) : (index, index) -> index
      %120 = "arith.addi"(%119, %81) : (index, index) -> index
      %121 = "arith.cmpi"(%89, %74) {predicate = 2 : i64} : (index, index) -> i1
      %122 = "arith.subi"(%70, %89) : (index, index) -> index
      %123 = "arith.select"(%121, %122, %89) : (i1, index, index) -> index
      %124 = "arith.divsi"(%123, %71) : (index, index) -> index
      %125 = "arith.subi"(%70, %124) : (index, index) -> index
      %126 = "arith.select"(%121, %125, %124) : (i1, index, index) -> index
      %127 = "arith.addi"(%120, %126) : (index, index) -> index
      "memref.store"(%78, %101, %127) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
      %128 = "arith.addi"(%127, %69) : (index, index) -> index
      "memref.store"(%78, %102, %128) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
      %129 = "arith.addi"(%127, %68) : (index, index) -> index
      "memref.store"(%78, %103, %129) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
      %130 = "arith.addi"(%127, %67) : (index, index) -> index
      "memref.store"(%78, %104, %130) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
      %131 = "memref.load"(%105, %127) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
      %132 = "memref.load"(%106, %128) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
      %133 = "memref.load"(%107, %129) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
      %134 = "memref.load"(%108, %130) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
      %135 = "arith.addi"(%115, %81) : (index, index) -> index
      %136 = "arith.muli"(%82, %68) : (index, index) -> index
      %137 = "arith.addi"(%135, %136) : (index, index) -> index
      %138 = "arith.muli"(%83, %66) : (index, index) -> index
      %139 = "arith.addi"(%137, %138) : (index, index) -> index
      %140 = "arith.cmpi"(%81, %74) {predicate = 2 : i64} : (index, index) -> i1
      %141 = "arith.subi"(%70, %81) : (index, index) -> index
      %142 = "arith.select"(%140, %141, %81) : (i1, index, index) -> index
      %143 = "arith.divsi"(%142, %65) : (index, index) -> index
      %144 = "arith.subi"(%70, %143) : (index, index) -> index
      %145 = "arith.select"(%140, %144, %143) : (i1, index, index) -> index
      %146 = "arith.muli"(%145, %64) : (index, index) -> index
      %147 = "arith.addi"(%139, %146) : (index, index) -> index
      %148 = "arith.addi"(%147, %75) : (index, index) -> index
      %149 = "memref.load"(%91, %148) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
      %150 = "arith.addi"(%147, %63) : (index, index) -> index
      %151 = "memref.load"(%92, %150) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
      %152 = "arith.muli"(%82, %69) : (index, index) -> index
      %153 = "arith.addi"(%81, %152) : (index, index) -> index
      %154 = "arith.muli"(%83, %62) : (index, index) -> index
      %155 = "arith.addi"(%153, %154) : (index, index) -> index
      %156 = "arith.addi"(%155, %118) : (index, index) -> index
      %157 = "arith.divsi"(%142, %71) : (index, index) -> index
      %158 = "arith.subi"(%70, %157) : (index, index) -> index
      %159 = "arith.select"(%140, %158, %157) : (i1, index, index) -> index
      %160 = "arith.muli"(%159, %61) : (index, index) -> index
      %161 = "arith.addi"(%156, %160) : (index, index) -> index
      %162 = "arith.cmpi"(%88, %74) {predicate = 2 : i64} : (index, index) -> i1
      %163 = "arith.subi"(%70, %88) : (index, index) -> index
      %164 = "arith.select"(%162, %163, %88) : (i1, index, index) -> index
      %165 = "arith.divsi"(%164, %71) : (index, index) -> index
      %166 = "arith.subi"(%70, %165) : (index, index) -> index
      %167 = "arith.select"(%162, %166, %165) : (i1, index, index) -> index
      %168 = "arith.addi"(%161, %167) : (index, index) -> index
      %169 = "memref.load"(%96, %168) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
      %170 = "arith.addi"(%168, %62) : (index, index) -> index
      %171 = "memref.load"(%97, %170) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
      %172:8 = "scf.for"(%74, %77, %80, %131, %132, %133, %134, %149, %151, %169, %171) ({
      ^bb0(%arg0: index, %arg1: vector<4xf32>, %arg2: vector<4xf32>, %arg3: vector<4xf32>, %arg4: vector<4xf32>, %arg5: vector<4xf32>, %arg6: vector<4xf32>, %arg7: vector<4xf32>, %arg8: vector<4xf32>):
        "gpu.barrier"() : () -> ()
        %696 = "arith.muli"(%82, %60) : (index, index) -> index
        %697 = "arith.addi"(%81, %696) : (index, index) -> index
        %698 = "arith.muli"(%83, %77) : (index, index) -> index
        %699 = "arith.addi"(%697, %698) : (index, index) -> index
        %700 = "arith.addi"(%699, %145) : (index, index) -> index
        "memref.store"(%arg5, %84, %700) {nontemporal = false} : (vector<4xf32>, memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
        %701 = "arith.addi"(%700, %77) : (index, index) -> index
        "memref.store"(%arg6, %84, %701) {nontemporal = false} : (vector<4xf32>, memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
        %702 = "arith.muli"(%82, %59) : (index, index) -> index
        %703 = "arith.addi"(%81, %702) : (index, index) -> index
        %704 = "arith.muli"(%83, %58) : (index, index) -> index
        %705 = "arith.addi"(%703, %704) : (index, index) -> index
        %706 = "arith.addi"(%705, %159) : (index, index) -> index
        "memref.store"(%arg7, %85, %706) {nontemporal = false} : (vector<4xf32>, memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
        %707 = "arith.addi"(%706, %58) : (index, index) -> index
        "memref.store"(%arg8, %85, %707) {nontemporal = false} : (vector<4xf32>, memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
        "gpu.barrier"() : () -> ()
        %708 = "arith.muli"(%82, %57) : (index, index) -> index
        %709 = "memref.load"(%84, %708) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %710 = "arith.addi"(%708, %56) : (index, index) -> index
        %711 = "memref.load"(%84, %710) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %712 = "arith.addi"(%708, %55) : (index, index) -> index
        %713 = "memref.load"(%84, %712) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %714 = "arith.addi"(%708, %54) : (index, index) -> index
        %715 = "memref.load"(%84, %714) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %716 = "arith.addi"(%708, %53) : (index, index) -> index
        %717 = "memref.load"(%84, %716) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %718 = "arith.addi"(%708, %52) : (index, index) -> index
        %719 = "memref.load"(%84, %718) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %720 = "arith.addi"(%708, %51) : (index, index) -> index
        %721 = "memref.load"(%84, %720) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %722 = "arith.addi"(%708, %50) : (index, index) -> index
        %723 = "memref.load"(%84, %722) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %724 = "arith.addi"(%708, %49) : (index, index) -> index
        %725 = "memref.load"(%84, %724) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %726 = "arith.addi"(%708, %48) : (index, index) -> index
        %727 = "memref.load"(%84, %726) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %728 = "arith.addi"(%708, %47) : (index, index) -> index
        %729 = "memref.load"(%84, %728) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %730 = "arith.addi"(%708, %46) : (index, index) -> index
        %731 = "memref.load"(%84, %730) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %732 = "arith.addi"(%708, %45) : (index, index) -> index
        %733 = "memref.load"(%84, %732) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %734 = "arith.addi"(%708, %44) : (index, index) -> index
        %735 = "memref.load"(%84, %734) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %736 = "arith.addi"(%708, %43) : (index, index) -> index
        %737 = "memref.load"(%84, %736) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %738 = "arith.addi"(%708, %71) : (index, index) -> index
        %739 = "memref.load"(%84, %738) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %740 = "arith.addi"(%708, %60) : (index, index) -> index
        %741 = "memref.load"(%84, %740) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %742 = "arith.addi"(%708, %42) : (index, index) -> index
        %743 = "memref.load"(%84, %742) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %744 = "arith.addi"(%708, %41) : (index, index) -> index
        %745 = "memref.load"(%84, %744) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %746 = "arith.addi"(%708, %40) : (index, index) -> index
        %747 = "memref.load"(%84, %746) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %748 = "arith.addi"(%708, %39) : (index, index) -> index
        %749 = "memref.load"(%84, %748) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %750 = "arith.addi"(%708, %38) : (index, index) -> index
        %751 = "memref.load"(%84, %750) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %752 = "arith.addi"(%708, %37) : (index, index) -> index
        %753 = "memref.load"(%84, %752) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %754 = "arith.addi"(%708, %36) : (index, index) -> index
        %755 = "memref.load"(%84, %754) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %756 = "arith.addi"(%708, %35) : (index, index) -> index
        %757 = "memref.load"(%84, %756) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %758 = "arith.addi"(%708, %34) : (index, index) -> index
        %759 = "memref.load"(%84, %758) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %760 = "arith.addi"(%708, %33) : (index, index) -> index
        %761 = "memref.load"(%84, %760) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %762 = "arith.addi"(%708, %32) : (index, index) -> index
        %763 = "memref.load"(%84, %762) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %764 = "arith.addi"(%708, %31) : (index, index) -> index
        %765 = "memref.load"(%84, %764) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %766 = "arith.addi"(%708, %80) : (index, index) -> index
        %767 = "memref.load"(%84, %766) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %768 = "arith.addi"(%708, %30) : (index, index) -> index
        %769 = "memref.load"(%84, %768) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %770 = "arith.addi"(%708, %29) : (index, index) -> index
        %771 = "memref.load"(%84, %770) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %772 = "memref.load"(%85, %81) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %773 = "arith.addi"(%81, %59) : (index, index) -> index
        %774 = "memref.load"(%85, %773) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %775 = "arith.addi"(%81, %29) : (index, index) -> index
        %776 = "memref.load"(%85, %775) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %777 = "arith.addi"(%81, %28) : (index, index) -> index
        %778 = "memref.load"(%85, %777) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %779 = "arith.addi"(%81, %27) : (index, index) -> index
        %780 = "memref.load"(%85, %779) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %781 = "arith.addi"(%81, %26) : (index, index) -> index
        %782 = "memref.load"(%85, %781) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %783 = "arith.addi"(%81, %25) : (index, index) -> index
        %784 = "memref.load"(%85, %783) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %785 = "arith.addi"(%81, %24) : (index, index) -> index
        %786 = "memref.load"(%85, %785) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %787 = "arith.addi"(%81, %23) : (index, index) -> index
        %788 = "memref.load"(%85, %787) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %789 = "arith.addi"(%81, %22) : (index, index) -> index
        %790 = "memref.load"(%85, %789) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %791 = "arith.addi"(%81, %21) : (index, index) -> index
        %792 = "memref.load"(%85, %791) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %793 = "arith.addi"(%81, %20) : (index, index) -> index
        %794 = "memref.load"(%85, %793) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %795 = "arith.addi"(%81, %19) : (index, index) -> index
        %796 = "memref.load"(%85, %795) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %797 = "arith.addi"(%81, %18) : (index, index) -> index
        %798 = "memref.load"(%85, %797) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %799 = "arith.addi"(%81, %17) : (index, index) -> index
        %800 = "memref.load"(%85, %799) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %801 = "arith.addi"(%81, %16) : (index, index) -> index
        %802 = "memref.load"(%85, %801) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %803 = "arith.addi"(%81, %58) : (index, index) -> index
        %804 = "memref.load"(%85, %803) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %805 = "arith.addi"(%81, %15) : (index, index) -> index
        %806 = "memref.load"(%85, %805) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %807 = "arith.addi"(%81, %14) : (index, index) -> index
        %808 = "memref.load"(%85, %807) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %809 = "arith.addi"(%81, %13) : (index, index) -> index
        %810 = "memref.load"(%85, %809) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %811 = "arith.addi"(%81, %12) : (index, index) -> index
        %812 = "memref.load"(%85, %811) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %813 = "arith.addi"(%81, %11) : (index, index) -> index
        %814 = "memref.load"(%85, %813) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %815 = "arith.addi"(%81, %10) : (index, index) -> index
        %816 = "memref.load"(%85, %815) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %817 = "arith.addi"(%81, %9) : (index, index) -> index
        %818 = "memref.load"(%85, %817) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %819 = "arith.addi"(%81, %8) : (index, index) -> index
        %820 = "memref.load"(%85, %819) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %821 = "arith.addi"(%81, %7) : (index, index) -> index
        %822 = "memref.load"(%85, %821) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %823 = "arith.addi"(%81, %6) : (index, index) -> index
        %824 = "memref.load"(%85, %823) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %825 = "arith.addi"(%81, %5) : (index, index) -> index
        %826 = "memref.load"(%85, %825) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %827 = "arith.addi"(%81, %4) : (index, index) -> index
        %828 = "memref.load"(%85, %827) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %829 = "arith.addi"(%81, %3) : (index, index) -> index
        %830 = "memref.load"(%85, %829) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %831 = "arith.addi"(%81, %2) : (index, index) -> index
        %832 = "memref.load"(%85, %831) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %833 = "arith.addi"(%81, %1) : (index, index) -> index
        %834 = "memref.load"(%85, %833) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %835 = "vector.extract"(%709) {position = [0]} : (vector<4xf32>) -> f32
        %836 = "vector.splat"(%835) : (f32) -> vector<4xf32>
        %837 = "vector.fma"(%836, %772, %arg1) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %838 = "vector.extract"(%709) {position = [1]} : (vector<4xf32>) -> f32
        %839 = "vector.splat"(%838) : (f32) -> vector<4xf32>
        %840 = "vector.fma"(%839, %774, %837) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %841 = "vector.extract"(%709) {position = [2]} : (vector<4xf32>) -> f32
        %842 = "vector.splat"(%841) : (f32) -> vector<4xf32>
        %843 = "vector.fma"(%842, %776, %840) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %844 = "vector.extract"(%709) {position = [3]} : (vector<4xf32>) -> f32
        %845 = "vector.splat"(%844) : (f32) -> vector<4xf32>
        %846 = "vector.fma"(%845, %778, %843) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %847 = "vector.extract"(%711) {position = [0]} : (vector<4xf32>) -> f32
        %848 = "vector.splat"(%847) : (f32) -> vector<4xf32>
        %849 = "vector.fma"(%848, %780, %846) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %850 = "vector.extract"(%711) {position = [1]} : (vector<4xf32>) -> f32
        %851 = "vector.splat"(%850) : (f32) -> vector<4xf32>
        %852 = "vector.fma"(%851, %782, %849) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %853 = "vector.extract"(%711) {position = [2]} : (vector<4xf32>) -> f32
        %854 = "vector.splat"(%853) : (f32) -> vector<4xf32>
        %855 = "vector.fma"(%854, %784, %852) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %856 = "vector.extract"(%711) {position = [3]} : (vector<4xf32>) -> f32
        %857 = "vector.splat"(%856) : (f32) -> vector<4xf32>
        %858 = "vector.fma"(%857, %786, %855) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %859 = "vector.extract"(%713) {position = [0]} : (vector<4xf32>) -> f32
        %860 = "vector.splat"(%859) : (f32) -> vector<4xf32>
        %861 = "vector.fma"(%860, %788, %858) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %862 = "vector.extract"(%713) {position = [1]} : (vector<4xf32>) -> f32
        %863 = "vector.splat"(%862) : (f32) -> vector<4xf32>
        %864 = "vector.fma"(%863, %790, %861) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %865 = "vector.extract"(%713) {position = [2]} : (vector<4xf32>) -> f32
        %866 = "vector.splat"(%865) : (f32) -> vector<4xf32>
        %867 = "vector.fma"(%866, %792, %864) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %868 = "vector.extract"(%713) {position = [3]} : (vector<4xf32>) -> f32
        %869 = "vector.splat"(%868) : (f32) -> vector<4xf32>
        %870 = "vector.fma"(%869, %794, %867) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %871 = "vector.extract"(%715) {position = [0]} : (vector<4xf32>) -> f32
        %872 = "vector.splat"(%871) : (f32) -> vector<4xf32>
        %873 = "vector.fma"(%872, %796, %870) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %874 = "vector.extract"(%715) {position = [1]} : (vector<4xf32>) -> f32
        %875 = "vector.splat"(%874) : (f32) -> vector<4xf32>
        %876 = "vector.fma"(%875, %798, %873) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %877 = "vector.extract"(%715) {position = [2]} : (vector<4xf32>) -> f32
        %878 = "vector.splat"(%877) : (f32) -> vector<4xf32>
        %879 = "vector.fma"(%878, %800, %876) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %880 = "vector.extract"(%715) {position = [3]} : (vector<4xf32>) -> f32
        %881 = "vector.splat"(%880) : (f32) -> vector<4xf32>
        %882 = "vector.fma"(%881, %802, %879) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %883 = "vector.extract"(%717) {position = [0]} : (vector<4xf32>) -> f32
        %884 = "vector.splat"(%883) : (f32) -> vector<4xf32>
        %885 = "vector.fma"(%884, %804, %882) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %886 = "vector.extract"(%717) {position = [1]} : (vector<4xf32>) -> f32
        %887 = "vector.splat"(%886) : (f32) -> vector<4xf32>
        %888 = "vector.fma"(%887, %806, %885) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %889 = "vector.extract"(%717) {position = [2]} : (vector<4xf32>) -> f32
        %890 = "vector.splat"(%889) : (f32) -> vector<4xf32>
        %891 = "vector.fma"(%890, %808, %888) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %892 = "vector.extract"(%717) {position = [3]} : (vector<4xf32>) -> f32
        %893 = "vector.splat"(%892) : (f32) -> vector<4xf32>
        %894 = "vector.fma"(%893, %810, %891) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %895 = "vector.extract"(%719) {position = [0]} : (vector<4xf32>) -> f32
        %896 = "vector.splat"(%895) : (f32) -> vector<4xf32>
        %897 = "vector.fma"(%896, %812, %894) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %898 = "vector.extract"(%719) {position = [1]} : (vector<4xf32>) -> f32
        %899 = "vector.splat"(%898) : (f32) -> vector<4xf32>
        %900 = "vector.fma"(%899, %814, %897) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %901 = "vector.extract"(%719) {position = [2]} : (vector<4xf32>) -> f32
        %902 = "vector.splat"(%901) : (f32) -> vector<4xf32>
        %903 = "vector.fma"(%902, %816, %900) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %904 = "vector.extract"(%719) {position = [3]} : (vector<4xf32>) -> f32
        %905 = "vector.splat"(%904) : (f32) -> vector<4xf32>
        %906 = "vector.fma"(%905, %818, %903) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %907 = "vector.extract"(%721) {position = [0]} : (vector<4xf32>) -> f32
        %908 = "vector.splat"(%907) : (f32) -> vector<4xf32>
        %909 = "vector.fma"(%908, %820, %906) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %910 = "vector.extract"(%721) {position = [1]} : (vector<4xf32>) -> f32
        %911 = "vector.splat"(%910) : (f32) -> vector<4xf32>
        %912 = "vector.fma"(%911, %822, %909) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %913 = "vector.extract"(%721) {position = [2]} : (vector<4xf32>) -> f32
        %914 = "vector.splat"(%913) : (f32) -> vector<4xf32>
        %915 = "vector.fma"(%914, %824, %912) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %916 = "vector.extract"(%721) {position = [3]} : (vector<4xf32>) -> f32
        %917 = "vector.splat"(%916) : (f32) -> vector<4xf32>
        %918 = "vector.fma"(%917, %826, %915) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %919 = "vector.extract"(%723) {position = [0]} : (vector<4xf32>) -> f32
        %920 = "vector.splat"(%919) : (f32) -> vector<4xf32>
        %921 = "vector.fma"(%920, %828, %918) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %922 = "vector.extract"(%723) {position = [1]} : (vector<4xf32>) -> f32
        %923 = "vector.splat"(%922) : (f32) -> vector<4xf32>
        %924 = "vector.fma"(%923, %830, %921) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %925 = "vector.extract"(%723) {position = [2]} : (vector<4xf32>) -> f32
        %926 = "vector.splat"(%925) : (f32) -> vector<4xf32>
        %927 = "vector.fma"(%926, %832, %924) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %928 = "vector.extract"(%723) {position = [3]} : (vector<4xf32>) -> f32
        %929 = "vector.splat"(%928) : (f32) -> vector<4xf32>
        %930 = "vector.fma"(%929, %834, %927) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %931 = "vector.extract"(%725) {position = [0]} : (vector<4xf32>) -> f32
        %932 = "vector.splat"(%931) : (f32) -> vector<4xf32>
        %933 = "vector.fma"(%932, %772, %arg2) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %934 = "vector.extract"(%725) {position = [1]} : (vector<4xf32>) -> f32
        %935 = "vector.splat"(%934) : (f32) -> vector<4xf32>
        %936 = "vector.fma"(%935, %774, %933) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %937 = "vector.extract"(%725) {position = [2]} : (vector<4xf32>) -> f32
        %938 = "vector.splat"(%937) : (f32) -> vector<4xf32>
        %939 = "vector.fma"(%938, %776, %936) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %940 = "vector.extract"(%725) {position = [3]} : (vector<4xf32>) -> f32
        %941 = "vector.splat"(%940) : (f32) -> vector<4xf32>
        %942 = "vector.fma"(%941, %778, %939) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %943 = "vector.extract"(%727) {position = [0]} : (vector<4xf32>) -> f32
        %944 = "vector.splat"(%943) : (f32) -> vector<4xf32>
        %945 = "vector.fma"(%944, %780, %942) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %946 = "vector.extract"(%727) {position = [1]} : (vector<4xf32>) -> f32
        %947 = "vector.splat"(%946) : (f32) -> vector<4xf32>
        %948 = "vector.fma"(%947, %782, %945) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %949 = "vector.extract"(%727) {position = [2]} : (vector<4xf32>) -> f32
        %950 = "vector.splat"(%949) : (f32) -> vector<4xf32>
        %951 = "vector.fma"(%950, %784, %948) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %952 = "vector.extract"(%727) {position = [3]} : (vector<4xf32>) -> f32
        %953 = "vector.splat"(%952) : (f32) -> vector<4xf32>
        %954 = "vector.fma"(%953, %786, %951) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %955 = "vector.extract"(%729) {position = [0]} : (vector<4xf32>) -> f32
        %956 = "vector.splat"(%955) : (f32) -> vector<4xf32>
        %957 = "vector.fma"(%956, %788, %954) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %958 = "vector.extract"(%729) {position = [1]} : (vector<4xf32>) -> f32
        %959 = "vector.splat"(%958) : (f32) -> vector<4xf32>
        %960 = "vector.fma"(%959, %790, %957) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %961 = "vector.extract"(%729) {position = [2]} : (vector<4xf32>) -> f32
        %962 = "vector.splat"(%961) : (f32) -> vector<4xf32>
        %963 = "vector.fma"(%962, %792, %960) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %964 = "vector.extract"(%729) {position = [3]} : (vector<4xf32>) -> f32
        %965 = "vector.splat"(%964) : (f32) -> vector<4xf32>
        %966 = "vector.fma"(%965, %794, %963) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %967 = "vector.extract"(%731) {position = [0]} : (vector<4xf32>) -> f32
        %968 = "vector.splat"(%967) : (f32) -> vector<4xf32>
        %969 = "vector.fma"(%968, %796, %966) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %970 = "vector.extract"(%731) {position = [1]} : (vector<4xf32>) -> f32
        %971 = "vector.splat"(%970) : (f32) -> vector<4xf32>
        %972 = "vector.fma"(%971, %798, %969) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %973 = "vector.extract"(%731) {position = [2]} : (vector<4xf32>) -> f32
        %974 = "vector.splat"(%973) : (f32) -> vector<4xf32>
        %975 = "vector.fma"(%974, %800, %972) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %976 = "vector.extract"(%731) {position = [3]} : (vector<4xf32>) -> f32
        %977 = "vector.splat"(%976) : (f32) -> vector<4xf32>
        %978 = "vector.fma"(%977, %802, %975) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %979 = "vector.extract"(%733) {position = [0]} : (vector<4xf32>) -> f32
        %980 = "vector.splat"(%979) : (f32) -> vector<4xf32>
        %981 = "vector.fma"(%980, %804, %978) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %982 = "vector.extract"(%733) {position = [1]} : (vector<4xf32>) -> f32
        %983 = "vector.splat"(%982) : (f32) -> vector<4xf32>
        %984 = "vector.fma"(%983, %806, %981) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %985 = "vector.extract"(%733) {position = [2]} : (vector<4xf32>) -> f32
        %986 = "vector.splat"(%985) : (f32) -> vector<4xf32>
        %987 = "vector.fma"(%986, %808, %984) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %988 = "vector.extract"(%733) {position = [3]} : (vector<4xf32>) -> f32
        %989 = "vector.splat"(%988) : (f32) -> vector<4xf32>
        %990 = "vector.fma"(%989, %810, %987) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %991 = "vector.extract"(%735) {position = [0]} : (vector<4xf32>) -> f32
        %992 = "vector.splat"(%991) : (f32) -> vector<4xf32>
        %993 = "vector.fma"(%992, %812, %990) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %994 = "vector.extract"(%735) {position = [1]} : (vector<4xf32>) -> f32
        %995 = "vector.splat"(%994) : (f32) -> vector<4xf32>
        %996 = "vector.fma"(%995, %814, %993) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %997 = "vector.extract"(%735) {position = [2]} : (vector<4xf32>) -> f32
        %998 = "vector.splat"(%997) : (f32) -> vector<4xf32>
        %999 = "vector.fma"(%998, %816, %996) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1000 = "vector.extract"(%735) {position = [3]} : (vector<4xf32>) -> f32
        %1001 = "vector.splat"(%1000) : (f32) -> vector<4xf32>
        %1002 = "vector.fma"(%1001, %818, %999) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1003 = "vector.extract"(%737) {position = [0]} : (vector<4xf32>) -> f32
        %1004 = "vector.splat"(%1003) : (f32) -> vector<4xf32>
        %1005 = "vector.fma"(%1004, %820, %1002) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1006 = "vector.extract"(%737) {position = [1]} : (vector<4xf32>) -> f32
        %1007 = "vector.splat"(%1006) : (f32) -> vector<4xf32>
        %1008 = "vector.fma"(%1007, %822, %1005) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1009 = "vector.extract"(%737) {position = [2]} : (vector<4xf32>) -> f32
        %1010 = "vector.splat"(%1009) : (f32) -> vector<4xf32>
        %1011 = "vector.fma"(%1010, %824, %1008) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1012 = "vector.extract"(%737) {position = [3]} : (vector<4xf32>) -> f32
        %1013 = "vector.splat"(%1012) : (f32) -> vector<4xf32>
        %1014 = "vector.fma"(%1013, %826, %1011) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1015 = "vector.extract"(%739) {position = [0]} : (vector<4xf32>) -> f32
        %1016 = "vector.splat"(%1015) : (f32) -> vector<4xf32>
        %1017 = "vector.fma"(%1016, %828, %1014) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1018 = "vector.extract"(%739) {position = [1]} : (vector<4xf32>) -> f32
        %1019 = "vector.splat"(%1018) : (f32) -> vector<4xf32>
        %1020 = "vector.fma"(%1019, %830, %1017) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1021 = "vector.extract"(%739) {position = [2]} : (vector<4xf32>) -> f32
        %1022 = "vector.splat"(%1021) : (f32) -> vector<4xf32>
        %1023 = "vector.fma"(%1022, %832, %1020) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1024 = "vector.extract"(%739) {position = [3]} : (vector<4xf32>) -> f32
        %1025 = "vector.splat"(%1024) : (f32) -> vector<4xf32>
        %1026 = "vector.fma"(%1025, %834, %1023) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1027 = "vector.extract"(%741) {position = [0]} : (vector<4xf32>) -> f32
        %1028 = "vector.splat"(%1027) : (f32) -> vector<4xf32>
        %1029 = "vector.fma"(%1028, %772, %arg3) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1030 = "vector.extract"(%741) {position = [1]} : (vector<4xf32>) -> f32
        %1031 = "vector.splat"(%1030) : (f32) -> vector<4xf32>
        %1032 = "vector.fma"(%1031, %774, %1029) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1033 = "vector.extract"(%741) {position = [2]} : (vector<4xf32>) -> f32
        %1034 = "vector.splat"(%1033) : (f32) -> vector<4xf32>
        %1035 = "vector.fma"(%1034, %776, %1032) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1036 = "vector.extract"(%741) {position = [3]} : (vector<4xf32>) -> f32
        %1037 = "vector.splat"(%1036) : (f32) -> vector<4xf32>
        %1038 = "vector.fma"(%1037, %778, %1035) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1039 = "vector.extract"(%743) {position = [0]} : (vector<4xf32>) -> f32
        %1040 = "vector.splat"(%1039) : (f32) -> vector<4xf32>
        %1041 = "vector.fma"(%1040, %780, %1038) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1042 = "vector.extract"(%743) {position = [1]} : (vector<4xf32>) -> f32
        %1043 = "vector.splat"(%1042) : (f32) -> vector<4xf32>
        %1044 = "vector.fma"(%1043, %782, %1041) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1045 = "vector.extract"(%743) {position = [2]} : (vector<4xf32>) -> f32
        %1046 = "vector.splat"(%1045) : (f32) -> vector<4xf32>
        %1047 = "vector.fma"(%1046, %784, %1044) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1048 = "vector.extract"(%743) {position = [3]} : (vector<4xf32>) -> f32
        %1049 = "vector.splat"(%1048) : (f32) -> vector<4xf32>
        %1050 = "vector.fma"(%1049, %786, %1047) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1051 = "vector.extract"(%745) {position = [0]} : (vector<4xf32>) -> f32
        %1052 = "vector.splat"(%1051) : (f32) -> vector<4xf32>
        %1053 = "vector.fma"(%1052, %788, %1050) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1054 = "vector.extract"(%745) {position = [1]} : (vector<4xf32>) -> f32
        %1055 = "vector.splat"(%1054) : (f32) -> vector<4xf32>
        %1056 = "vector.fma"(%1055, %790, %1053) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1057 = "vector.extract"(%745) {position = [2]} : (vector<4xf32>) -> f32
        %1058 = "vector.splat"(%1057) : (f32) -> vector<4xf32>
        %1059 = "vector.fma"(%1058, %792, %1056) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1060 = "vector.extract"(%745) {position = [3]} : (vector<4xf32>) -> f32
        %1061 = "vector.splat"(%1060) : (f32) -> vector<4xf32>
        %1062 = "vector.fma"(%1061, %794, %1059) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1063 = "vector.extract"(%747) {position = [0]} : (vector<4xf32>) -> f32
        %1064 = "vector.splat"(%1063) : (f32) -> vector<4xf32>
        %1065 = "vector.fma"(%1064, %796, %1062) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1066 = "vector.extract"(%747) {position = [1]} : (vector<4xf32>) -> f32
        %1067 = "vector.splat"(%1066) : (f32) -> vector<4xf32>
        %1068 = "vector.fma"(%1067, %798, %1065) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1069 = "vector.extract"(%747) {position = [2]} : (vector<4xf32>) -> f32
        %1070 = "vector.splat"(%1069) : (f32) -> vector<4xf32>
        %1071 = "vector.fma"(%1070, %800, %1068) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1072 = "vector.extract"(%747) {position = [3]} : (vector<4xf32>) -> f32
        %1073 = "vector.splat"(%1072) : (f32) -> vector<4xf32>
        %1074 = "vector.fma"(%1073, %802, %1071) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1075 = "vector.extract"(%749) {position = [0]} : (vector<4xf32>) -> f32
        %1076 = "vector.splat"(%1075) : (f32) -> vector<4xf32>
        %1077 = "vector.fma"(%1076, %804, %1074) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1078 = "vector.extract"(%749) {position = [1]} : (vector<4xf32>) -> f32
        %1079 = "vector.splat"(%1078) : (f32) -> vector<4xf32>
        %1080 = "vector.fma"(%1079, %806, %1077) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1081 = "vector.extract"(%749) {position = [2]} : (vector<4xf32>) -> f32
        %1082 = "vector.splat"(%1081) : (f32) -> vector<4xf32>
        %1083 = "vector.fma"(%1082, %808, %1080) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1084 = "vector.extract"(%749) {position = [3]} : (vector<4xf32>) -> f32
        %1085 = "vector.splat"(%1084) : (f32) -> vector<4xf32>
        %1086 = "vector.fma"(%1085, %810, %1083) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1087 = "vector.extract"(%751) {position = [0]} : (vector<4xf32>) -> f32
        %1088 = "vector.splat"(%1087) : (f32) -> vector<4xf32>
        %1089 = "vector.fma"(%1088, %812, %1086) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1090 = "vector.extract"(%751) {position = [1]} : (vector<4xf32>) -> f32
        %1091 = "vector.splat"(%1090) : (f32) -> vector<4xf32>
        %1092 = "vector.fma"(%1091, %814, %1089) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1093 = "vector.extract"(%751) {position = [2]} : (vector<4xf32>) -> f32
        %1094 = "vector.splat"(%1093) : (f32) -> vector<4xf32>
        %1095 = "vector.fma"(%1094, %816, %1092) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1096 = "vector.extract"(%751) {position = [3]} : (vector<4xf32>) -> f32
        %1097 = "vector.splat"(%1096) : (f32) -> vector<4xf32>
        %1098 = "vector.fma"(%1097, %818, %1095) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1099 = "vector.extract"(%753) {position = [0]} : (vector<4xf32>) -> f32
        %1100 = "vector.splat"(%1099) : (f32) -> vector<4xf32>
        %1101 = "vector.fma"(%1100, %820, %1098) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1102 = "vector.extract"(%753) {position = [1]} : (vector<4xf32>) -> f32
        %1103 = "vector.splat"(%1102) : (f32) -> vector<4xf32>
        %1104 = "vector.fma"(%1103, %822, %1101) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1105 = "vector.extract"(%753) {position = [2]} : (vector<4xf32>) -> f32
        %1106 = "vector.splat"(%1105) : (f32) -> vector<4xf32>
        %1107 = "vector.fma"(%1106, %824, %1104) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1108 = "vector.extract"(%753) {position = [3]} : (vector<4xf32>) -> f32
        %1109 = "vector.splat"(%1108) : (f32) -> vector<4xf32>
        %1110 = "vector.fma"(%1109, %826, %1107) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1111 = "vector.extract"(%755) {position = [0]} : (vector<4xf32>) -> f32
        %1112 = "vector.splat"(%1111) : (f32) -> vector<4xf32>
        %1113 = "vector.fma"(%1112, %828, %1110) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1114 = "vector.extract"(%755) {position = [1]} : (vector<4xf32>) -> f32
        %1115 = "vector.splat"(%1114) : (f32) -> vector<4xf32>
        %1116 = "vector.fma"(%1115, %830, %1113) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1117 = "vector.extract"(%755) {position = [2]} : (vector<4xf32>) -> f32
        %1118 = "vector.splat"(%1117) : (f32) -> vector<4xf32>
        %1119 = "vector.fma"(%1118, %832, %1116) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1120 = "vector.extract"(%755) {position = [3]} : (vector<4xf32>) -> f32
        %1121 = "vector.splat"(%1120) : (f32) -> vector<4xf32>
        %1122 = "vector.fma"(%1121, %834, %1119) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1123 = "vector.extract"(%757) {position = [0]} : (vector<4xf32>) -> f32
        %1124 = "vector.splat"(%1123) : (f32) -> vector<4xf32>
        %1125 = "vector.fma"(%1124, %772, %arg4) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1126 = "vector.extract"(%757) {position = [1]} : (vector<4xf32>) -> f32
        %1127 = "vector.splat"(%1126) : (f32) -> vector<4xf32>
        %1128 = "vector.fma"(%1127, %774, %1125) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1129 = "vector.extract"(%757) {position = [2]} : (vector<4xf32>) -> f32
        %1130 = "vector.splat"(%1129) : (f32) -> vector<4xf32>
        %1131 = "vector.fma"(%1130, %776, %1128) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1132 = "vector.extract"(%757) {position = [3]} : (vector<4xf32>) -> f32
        %1133 = "vector.splat"(%1132) : (f32) -> vector<4xf32>
        %1134 = "vector.fma"(%1133, %778, %1131) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1135 = "vector.extract"(%759) {position = [0]} : (vector<4xf32>) -> f32
        %1136 = "vector.splat"(%1135) : (f32) -> vector<4xf32>
        %1137 = "vector.fma"(%1136, %780, %1134) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1138 = "vector.extract"(%759) {position = [1]} : (vector<4xf32>) -> f32
        %1139 = "vector.splat"(%1138) : (f32) -> vector<4xf32>
        %1140 = "vector.fma"(%1139, %782, %1137) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1141 = "vector.extract"(%759) {position = [2]} : (vector<4xf32>) -> f32
        %1142 = "vector.splat"(%1141) : (f32) -> vector<4xf32>
        %1143 = "vector.fma"(%1142, %784, %1140) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1144 = "vector.extract"(%759) {position = [3]} : (vector<4xf32>) -> f32
        %1145 = "vector.splat"(%1144) : (f32) -> vector<4xf32>
        %1146 = "vector.fma"(%1145, %786, %1143) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1147 = "vector.extract"(%761) {position = [0]} : (vector<4xf32>) -> f32
        %1148 = "vector.splat"(%1147) : (f32) -> vector<4xf32>
        %1149 = "vector.fma"(%1148, %788, %1146) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1150 = "vector.extract"(%761) {position = [1]} : (vector<4xf32>) -> f32
        %1151 = "vector.splat"(%1150) : (f32) -> vector<4xf32>
        %1152 = "vector.fma"(%1151, %790, %1149) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1153 = "vector.extract"(%761) {position = [2]} : (vector<4xf32>) -> f32
        %1154 = "vector.splat"(%1153) : (f32) -> vector<4xf32>
        %1155 = "vector.fma"(%1154, %792, %1152) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1156 = "vector.extract"(%761) {position = [3]} : (vector<4xf32>) -> f32
        %1157 = "vector.splat"(%1156) : (f32) -> vector<4xf32>
        %1158 = "vector.fma"(%1157, %794, %1155) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1159 = "vector.extract"(%763) {position = [0]} : (vector<4xf32>) -> f32
        %1160 = "vector.splat"(%1159) : (f32) -> vector<4xf32>
        %1161 = "vector.fma"(%1160, %796, %1158) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1162 = "vector.extract"(%763) {position = [1]} : (vector<4xf32>) -> f32
        %1163 = "vector.splat"(%1162) : (f32) -> vector<4xf32>
        %1164 = "vector.fma"(%1163, %798, %1161) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1165 = "vector.extract"(%763) {position = [2]} : (vector<4xf32>) -> f32
        %1166 = "vector.splat"(%1165) : (f32) -> vector<4xf32>
        %1167 = "vector.fma"(%1166, %800, %1164) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1168 = "vector.extract"(%763) {position = [3]} : (vector<4xf32>) -> f32
        %1169 = "vector.splat"(%1168) : (f32) -> vector<4xf32>
        %1170 = "vector.fma"(%1169, %802, %1167) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1171 = "vector.extract"(%765) {position = [0]} : (vector<4xf32>) -> f32
        %1172 = "vector.splat"(%1171) : (f32) -> vector<4xf32>
        %1173 = "vector.fma"(%1172, %804, %1170) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1174 = "vector.extract"(%765) {position = [1]} : (vector<4xf32>) -> f32
        %1175 = "vector.splat"(%1174) : (f32) -> vector<4xf32>
        %1176 = "vector.fma"(%1175, %806, %1173) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1177 = "vector.extract"(%765) {position = [2]} : (vector<4xf32>) -> f32
        %1178 = "vector.splat"(%1177) : (f32) -> vector<4xf32>
        %1179 = "vector.fma"(%1178, %808, %1176) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1180 = "vector.extract"(%765) {position = [3]} : (vector<4xf32>) -> f32
        %1181 = "vector.splat"(%1180) : (f32) -> vector<4xf32>
        %1182 = "vector.fma"(%1181, %810, %1179) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1183 = "vector.extract"(%767) {position = [0]} : (vector<4xf32>) -> f32
        %1184 = "vector.splat"(%1183) : (f32) -> vector<4xf32>
        %1185 = "vector.fma"(%1184, %812, %1182) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1186 = "vector.extract"(%767) {position = [1]} : (vector<4xf32>) -> f32
        %1187 = "vector.splat"(%1186) : (f32) -> vector<4xf32>
        %1188 = "vector.fma"(%1187, %814, %1185) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1189 = "vector.extract"(%767) {position = [2]} : (vector<4xf32>) -> f32
        %1190 = "vector.splat"(%1189) : (f32) -> vector<4xf32>
        %1191 = "vector.fma"(%1190, %816, %1188) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1192 = "vector.extract"(%767) {position = [3]} : (vector<4xf32>) -> f32
        %1193 = "vector.splat"(%1192) : (f32) -> vector<4xf32>
        %1194 = "vector.fma"(%1193, %818, %1191) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1195 = "vector.extract"(%769) {position = [0]} : (vector<4xf32>) -> f32
        %1196 = "vector.splat"(%1195) : (f32) -> vector<4xf32>
        %1197 = "vector.fma"(%1196, %820, %1194) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1198 = "vector.extract"(%769) {position = [1]} : (vector<4xf32>) -> f32
        %1199 = "vector.splat"(%1198) : (f32) -> vector<4xf32>
        %1200 = "vector.fma"(%1199, %822, %1197) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1201 = "vector.extract"(%769) {position = [2]} : (vector<4xf32>) -> f32
        %1202 = "vector.splat"(%1201) : (f32) -> vector<4xf32>
        %1203 = "vector.fma"(%1202, %824, %1200) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1204 = "vector.extract"(%769) {position = [3]} : (vector<4xf32>) -> f32
        %1205 = "vector.splat"(%1204) : (f32) -> vector<4xf32>
        %1206 = "vector.fma"(%1205, %826, %1203) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1207 = "vector.extract"(%771) {position = [0]} : (vector<4xf32>) -> f32
        %1208 = "vector.splat"(%1207) : (f32) -> vector<4xf32>
        %1209 = "vector.fma"(%1208, %828, %1206) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1210 = "vector.extract"(%771) {position = [1]} : (vector<4xf32>) -> f32
        %1211 = "vector.splat"(%1210) : (f32) -> vector<4xf32>
        %1212 = "vector.fma"(%1211, %830, %1209) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1213 = "vector.extract"(%771) {position = [2]} : (vector<4xf32>) -> f32
        %1214 = "vector.splat"(%1213) : (f32) -> vector<4xf32>
        %1215 = "vector.fma"(%1214, %832, %1212) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1216 = "vector.extract"(%771) {position = [3]} : (vector<4xf32>) -> f32
        %1217 = "vector.splat"(%1216) : (f32) -> vector<4xf32>
        %1218 = "vector.fma"(%1217, %834, %1215) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %1219 = "arith.addi"(%arg0, %80) : (index, index) -> index
        %1220 = "arith.cmpi"(%1219, %74) {predicate = 2 : i64} : (index, index) -> i1
        %1221 = "arith.subi"(%0, %arg0) : (index, index) -> index
        %1222 = "arith.select"(%1220, %1221, %1219) : (i1, index, index) -> index
        %1223 = "arith.divsi"(%1222, %53) : (index, index) -> index
        %1224 = "arith.subi"(%70, %1223) : (index, index) -> index
        %1225 = "arith.select"(%1220, %1224, %1223) : (i1, index, index) -> index
        %1226 = "arith.addi"(%139, %1225) : (index, index) -> index
        %1227 = "arith.addi"(%1226, %146) : (index, index) -> index
        %1228 = "arith.addi"(%1227, %75) : (index, index) -> index
        %1229 = "memref.load"(%93, %1228) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %1230 = "arith.addi"(%1227, %63) : (index, index) -> index
        %1231 = "memref.load"(%94, %1230) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %1232 = "arith.muli"(%1219, %69) : (index, index) -> index
        %1233 = "arith.addi"(%1232, %81) : (index, index) -> index
        %1234 = "arith.addi"(%1233, %152) : (index, index) -> index
        %1235 = "arith.addi"(%1234, %154) : (index, index) -> index
        %1236 = "arith.addi"(%1235, %118) : (index, index) -> index
        %1237 = "arith.addi"(%1236, %160) : (index, index) -> index
        %1238 = "arith.addi"(%1237, %167) : (index, index) -> index
        %1239 = "memref.load"(%98, %1238) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %1240 = "arith.addi"(%1238, %62) : (index, index) -> index
        %1241 = "memref.load"(%99, %1240) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        "scf.yield"(%930, %1026, %1122, %1218, %1229, %1231, %1239, %1241) : (vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>) -> ()
      }) : (index, index, index, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>) -> (vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>)
      "gpu.barrier"() : () -> ()
      %173 = "arith.muli"(%82, %60) : (index, index) -> index
      %174 = "arith.addi"(%81, %173) : (index, index) -> index
      %175 = "arith.muli"(%83, %77) : (index, index) -> index
      %176 = "arith.addi"(%174, %175) : (index, index) -> index
      %177 = "arith.addi"(%176, %145) : (index, index) -> index
      "memref.store"(%172#4, %84, %177) {nontemporal = false} : (vector<4xf32>, memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
      %178 = "arith.addi"(%177, %77) : (index, index) -> index
      "memref.store"(%172#5, %84, %178) {nontemporal = false} : (vector<4xf32>, memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
      %179 = "arith.muli"(%82, %59) : (index, index) -> index
      %180 = "arith.addi"(%81, %179) : (index, index) -> index
      %181 = "arith.muli"(%83, %58) : (index, index) -> index
      %182 = "arith.addi"(%180, %181) : (index, index) -> index
      %183 = "arith.addi"(%182, %159) : (index, index) -> index
      "memref.store"(%172#6, %85, %183) {nontemporal = false} : (vector<4xf32>, memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
      %184 = "arith.addi"(%183, %58) : (index, index) -> index
      "memref.store"(%172#7, %85, %184) {nontemporal = false} : (vector<4xf32>, memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
      "gpu.barrier"() : () -> ()
      %185 = "arith.muli"(%82, %57) : (index, index) -> index
      %186 = "memref.load"(%84, %185) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %187 = "arith.addi"(%185, %56) : (index, index) -> index
      %188 = "memref.load"(%84, %187) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %189 = "arith.addi"(%185, %55) : (index, index) -> index
      %190 = "memref.load"(%84, %189) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %191 = "arith.addi"(%185, %54) : (index, index) -> index
      %192 = "memref.load"(%84, %191) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %193 = "arith.addi"(%185, %53) : (index, index) -> index
      %194 = "memref.load"(%84, %193) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %195 = "arith.addi"(%185, %52) : (index, index) -> index
      %196 = "memref.load"(%84, %195) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %197 = "arith.addi"(%185, %51) : (index, index) -> index
      %198 = "memref.load"(%84, %197) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %199 = "arith.addi"(%185, %50) : (index, index) -> index
      %200 = "memref.load"(%84, %199) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %201 = "arith.addi"(%185, %49) : (index, index) -> index
      %202 = "memref.load"(%84, %201) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %203 = "arith.addi"(%185, %48) : (index, index) -> index
      %204 = "memref.load"(%84, %203) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %205 = "arith.addi"(%185, %47) : (index, index) -> index
      %206 = "memref.load"(%84, %205) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %207 = "arith.addi"(%185, %46) : (index, index) -> index
      %208 = "memref.load"(%84, %207) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %209 = "arith.addi"(%185, %45) : (index, index) -> index
      %210 = "memref.load"(%84, %209) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %211 = "arith.addi"(%185, %44) : (index, index) -> index
      %212 = "memref.load"(%84, %211) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %213 = "arith.addi"(%185, %43) : (index, index) -> index
      %214 = "memref.load"(%84, %213) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %215 = "arith.addi"(%185, %71) : (index, index) -> index
      %216 = "memref.load"(%84, %215) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %217 = "arith.addi"(%185, %60) : (index, index) -> index
      %218 = "memref.load"(%84, %217) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %219 = "arith.addi"(%185, %42) : (index, index) -> index
      %220 = "memref.load"(%84, %219) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %221 = "arith.addi"(%185, %41) : (index, index) -> index
      %222 = "memref.load"(%84, %221) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %223 = "arith.addi"(%185, %40) : (index, index) -> index
      %224 = "memref.load"(%84, %223) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %225 = "arith.addi"(%185, %39) : (index, index) -> index
      %226 = "memref.load"(%84, %225) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %227 = "arith.addi"(%185, %38) : (index, index) -> index
      %228 = "memref.load"(%84, %227) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %229 = "arith.addi"(%185, %37) : (index, index) -> index
      %230 = "memref.load"(%84, %229) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %231 = "arith.addi"(%185, %36) : (index, index) -> index
      %232 = "memref.load"(%84, %231) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %233 = "arith.addi"(%185, %35) : (index, index) -> index
      %234 = "memref.load"(%84, %233) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %235 = "arith.addi"(%185, %34) : (index, index) -> index
      %236 = "memref.load"(%84, %235) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %237 = "arith.addi"(%185, %33) : (index, index) -> index
      %238 = "memref.load"(%84, %237) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %239 = "arith.addi"(%185, %32) : (index, index) -> index
      %240 = "memref.load"(%84, %239) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %241 = "arith.addi"(%185, %31) : (index, index) -> index
      %242 = "memref.load"(%84, %241) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %243 = "arith.addi"(%185, %80) : (index, index) -> index
      %244 = "memref.load"(%84, %243) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %245 = "arith.addi"(%185, %30) : (index, index) -> index
      %246 = "memref.load"(%84, %245) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %247 = "arith.addi"(%185, %29) : (index, index) -> index
      %248 = "memref.load"(%84, %247) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %249 = "memref.load"(%85, %81) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %250 = "arith.addi"(%81, %59) : (index, index) -> index
      %251 = "memref.load"(%85, %250) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %252 = "arith.addi"(%81, %29) : (index, index) -> index
      %253 = "memref.load"(%85, %252) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %254 = "arith.addi"(%81, %28) : (index, index) -> index
      %255 = "memref.load"(%85, %254) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %256 = "arith.addi"(%81, %27) : (index, index) -> index
      %257 = "memref.load"(%85, %256) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %258 = "arith.addi"(%81, %26) : (index, index) -> index
      %259 = "memref.load"(%85, %258) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %260 = "arith.addi"(%81, %25) : (index, index) -> index
      %261 = "memref.load"(%85, %260) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %262 = "arith.addi"(%81, %24) : (index, index) -> index
      %263 = "memref.load"(%85, %262) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %264 = "arith.addi"(%81, %23) : (index, index) -> index
      %265 = "memref.load"(%85, %264) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %266 = "arith.addi"(%81, %22) : (index, index) -> index
      %267 = "memref.load"(%85, %266) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %268 = "arith.addi"(%81, %21) : (index, index) -> index
      %269 = "memref.load"(%85, %268) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %270 = "arith.addi"(%81, %20) : (index, index) -> index
      %271 = "memref.load"(%85, %270) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %272 = "arith.addi"(%81, %19) : (index, index) -> index
      %273 = "memref.load"(%85, %272) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %274 = "arith.addi"(%81, %18) : (index, index) -> index
      %275 = "memref.load"(%85, %274) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %276 = "arith.addi"(%81, %17) : (index, index) -> index
      %277 = "memref.load"(%85, %276) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %278 = "arith.addi"(%81, %16) : (index, index) -> index
      %279 = "memref.load"(%85, %278) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %280 = "arith.addi"(%81, %58) : (index, index) -> index
      %281 = "memref.load"(%85, %280) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %282 = "arith.addi"(%81, %15) : (index, index) -> index
      %283 = "memref.load"(%85, %282) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %284 = "arith.addi"(%81, %14) : (index, index) -> index
      %285 = "memref.load"(%85, %284) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %286 = "arith.addi"(%81, %13) : (index, index) -> index
      %287 = "memref.load"(%85, %286) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %288 = "arith.addi"(%81, %12) : (index, index) -> index
      %289 = "memref.load"(%85, %288) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %290 = "arith.addi"(%81, %11) : (index, index) -> index
      %291 = "memref.load"(%85, %290) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %292 = "arith.addi"(%81, %10) : (index, index) -> index
      %293 = "memref.load"(%85, %292) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %294 = "arith.addi"(%81, %9) : (index, index) -> index
      %295 = "memref.load"(%85, %294) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %296 = "arith.addi"(%81, %8) : (index, index) -> index
      %297 = "memref.load"(%85, %296) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %298 = "arith.addi"(%81, %7) : (index, index) -> index
      %299 = "memref.load"(%85, %298) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %300 = "arith.addi"(%81, %6) : (index, index) -> index
      %301 = "memref.load"(%85, %300) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %302 = "arith.addi"(%81, %5) : (index, index) -> index
      %303 = "memref.load"(%85, %302) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %304 = "arith.addi"(%81, %4) : (index, index) -> index
      %305 = "memref.load"(%85, %304) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %306 = "arith.addi"(%81, %3) : (index, index) -> index
      %307 = "memref.load"(%85, %306) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %308 = "arith.addi"(%81, %2) : (index, index) -> index
      %309 = "memref.load"(%85, %308) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %310 = "arith.addi"(%81, %1) : (index, index) -> index
      %311 = "memref.load"(%85, %310) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
      %312 = "vector.extract"(%186) {position = [0]} : (vector<4xf32>) -> f32
      %313 = "vector.splat"(%312) : (f32) -> vector<4xf32>
      %314 = "vector.fma"(%313, %249, %172#0) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %315 = "vector.extract"(%186) {position = [1]} : (vector<4xf32>) -> f32
      %316 = "vector.splat"(%315) : (f32) -> vector<4xf32>
      %317 = "vector.fma"(%316, %251, %314) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %318 = "vector.extract"(%186) {position = [2]} : (vector<4xf32>) -> f32
      %319 = "vector.splat"(%318) : (f32) -> vector<4xf32>
      %320 = "vector.fma"(%319, %253, %317) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %321 = "vector.extract"(%186) {position = [3]} : (vector<4xf32>) -> f32
      %322 = "vector.splat"(%321) : (f32) -> vector<4xf32>
      %323 = "vector.fma"(%322, %255, %320) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %324 = "vector.extract"(%188) {position = [0]} : (vector<4xf32>) -> f32
      %325 = "vector.splat"(%324) : (f32) -> vector<4xf32>
      %326 = "vector.fma"(%325, %257, %323) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %327 = "vector.extract"(%188) {position = [1]} : (vector<4xf32>) -> f32
      %328 = "vector.splat"(%327) : (f32) -> vector<4xf32>
      %329 = "vector.fma"(%328, %259, %326) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %330 = "vector.extract"(%188) {position = [2]} : (vector<4xf32>) -> f32
      %331 = "vector.splat"(%330) : (f32) -> vector<4xf32>
      %332 = "vector.fma"(%331, %261, %329) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %333 = "vector.extract"(%188) {position = [3]} : (vector<4xf32>) -> f32
      %334 = "vector.splat"(%333) : (f32) -> vector<4xf32>
      %335 = "vector.fma"(%334, %263, %332) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %336 = "vector.extract"(%190) {position = [0]} : (vector<4xf32>) -> f32
      %337 = "vector.splat"(%336) : (f32) -> vector<4xf32>
      %338 = "vector.fma"(%337, %265, %335) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %339 = "vector.extract"(%190) {position = [1]} : (vector<4xf32>) -> f32
      %340 = "vector.splat"(%339) : (f32) -> vector<4xf32>
      %341 = "vector.fma"(%340, %267, %338) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %342 = "vector.extract"(%190) {position = [2]} : (vector<4xf32>) -> f32
      %343 = "vector.splat"(%342) : (f32) -> vector<4xf32>
      %344 = "vector.fma"(%343, %269, %341) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %345 = "vector.extract"(%190) {position = [3]} : (vector<4xf32>) -> f32
      %346 = "vector.splat"(%345) : (f32) -> vector<4xf32>
      %347 = "vector.fma"(%346, %271, %344) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %348 = "vector.extract"(%192) {position = [0]} : (vector<4xf32>) -> f32
      %349 = "vector.splat"(%348) : (f32) -> vector<4xf32>
      %350 = "vector.fma"(%349, %273, %347) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %351 = "vector.extract"(%192) {position = [1]} : (vector<4xf32>) -> f32
      %352 = "vector.splat"(%351) : (f32) -> vector<4xf32>
      %353 = "vector.fma"(%352, %275, %350) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %354 = "vector.extract"(%192) {position = [2]} : (vector<4xf32>) -> f32
      %355 = "vector.splat"(%354) : (f32) -> vector<4xf32>
      %356 = "vector.fma"(%355, %277, %353) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %357 = "vector.extract"(%192) {position = [3]} : (vector<4xf32>) -> f32
      %358 = "vector.splat"(%357) : (f32) -> vector<4xf32>
      %359 = "vector.fma"(%358, %279, %356) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %360 = "vector.extract"(%194) {position = [0]} : (vector<4xf32>) -> f32
      %361 = "vector.splat"(%360) : (f32) -> vector<4xf32>
      %362 = "vector.fma"(%361, %281, %359) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %363 = "vector.extract"(%194) {position = [1]} : (vector<4xf32>) -> f32
      %364 = "vector.splat"(%363) : (f32) -> vector<4xf32>
      %365 = "vector.fma"(%364, %283, %362) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %366 = "vector.extract"(%194) {position = [2]} : (vector<4xf32>) -> f32
      %367 = "vector.splat"(%366) : (f32) -> vector<4xf32>
      %368 = "vector.fma"(%367, %285, %365) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %369 = "vector.extract"(%194) {position = [3]} : (vector<4xf32>) -> f32
      %370 = "vector.splat"(%369) : (f32) -> vector<4xf32>
      %371 = "vector.fma"(%370, %287, %368) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %372 = "vector.extract"(%196) {position = [0]} : (vector<4xf32>) -> f32
      %373 = "vector.splat"(%372) : (f32) -> vector<4xf32>
      %374 = "vector.fma"(%373, %289, %371) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %375 = "vector.extract"(%196) {position = [1]} : (vector<4xf32>) -> f32
      %376 = "vector.splat"(%375) : (f32) -> vector<4xf32>
      %377 = "vector.fma"(%376, %291, %374) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %378 = "vector.extract"(%196) {position = [2]} : (vector<4xf32>) -> f32
      %379 = "vector.splat"(%378) : (f32) -> vector<4xf32>
      %380 = "vector.fma"(%379, %293, %377) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %381 = "vector.extract"(%196) {position = [3]} : (vector<4xf32>) -> f32
      %382 = "vector.splat"(%381) : (f32) -> vector<4xf32>
      %383 = "vector.fma"(%382, %295, %380) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %384 = "vector.extract"(%198) {position = [0]} : (vector<4xf32>) -> f32
      %385 = "vector.splat"(%384) : (f32) -> vector<4xf32>
      %386 = "vector.fma"(%385, %297, %383) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %387 = "vector.extract"(%198) {position = [1]} : (vector<4xf32>) -> f32
      %388 = "vector.splat"(%387) : (f32) -> vector<4xf32>
      %389 = "vector.fma"(%388, %299, %386) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %390 = "vector.extract"(%198) {position = [2]} : (vector<4xf32>) -> f32
      %391 = "vector.splat"(%390) : (f32) -> vector<4xf32>
      %392 = "vector.fma"(%391, %301, %389) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %393 = "vector.extract"(%198) {position = [3]} : (vector<4xf32>) -> f32
      %394 = "vector.splat"(%393) : (f32) -> vector<4xf32>
      %395 = "vector.fma"(%394, %303, %392) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %396 = "vector.extract"(%200) {position = [0]} : (vector<4xf32>) -> f32
      %397 = "vector.splat"(%396) : (f32) -> vector<4xf32>
      %398 = "vector.fma"(%397, %305, %395) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %399 = "vector.extract"(%200) {position = [1]} : (vector<4xf32>) -> f32
      %400 = "vector.splat"(%399) : (f32) -> vector<4xf32>
      %401 = "vector.fma"(%400, %307, %398) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %402 = "vector.extract"(%200) {position = [2]} : (vector<4xf32>) -> f32
      %403 = "vector.splat"(%402) : (f32) -> vector<4xf32>
      %404 = "vector.fma"(%403, %309, %401) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %405 = "vector.extract"(%200) {position = [3]} : (vector<4xf32>) -> f32
      %406 = "vector.splat"(%405) : (f32) -> vector<4xf32>
      %407 = "vector.fma"(%406, %311, %404) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %408 = "vector.extract"(%202) {position = [0]} : (vector<4xf32>) -> f32
      %409 = "vector.splat"(%408) : (f32) -> vector<4xf32>
      %410 = "vector.fma"(%409, %249, %172#1) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %411 = "vector.extract"(%202) {position = [1]} : (vector<4xf32>) -> f32
      %412 = "vector.splat"(%411) : (f32) -> vector<4xf32>
      %413 = "vector.fma"(%412, %251, %410) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %414 = "vector.extract"(%202) {position = [2]} : (vector<4xf32>) -> f32
      %415 = "vector.splat"(%414) : (f32) -> vector<4xf32>
      %416 = "vector.fma"(%415, %253, %413) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %417 = "vector.extract"(%202) {position = [3]} : (vector<4xf32>) -> f32
      %418 = "vector.splat"(%417) : (f32) -> vector<4xf32>
      %419 = "vector.fma"(%418, %255, %416) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %420 = "vector.extract"(%204) {position = [0]} : (vector<4xf32>) -> f32
      %421 = "vector.splat"(%420) : (f32) -> vector<4xf32>
      %422 = "vector.fma"(%421, %257, %419) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %423 = "vector.extract"(%204) {position = [1]} : (vector<4xf32>) -> f32
      %424 = "vector.splat"(%423) : (f32) -> vector<4xf32>
      %425 = "vector.fma"(%424, %259, %422) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %426 = "vector.extract"(%204) {position = [2]} : (vector<4xf32>) -> f32
      %427 = "vector.splat"(%426) : (f32) -> vector<4xf32>
      %428 = "vector.fma"(%427, %261, %425) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %429 = "vector.extract"(%204) {position = [3]} : (vector<4xf32>) -> f32
      %430 = "vector.splat"(%429) : (f32) -> vector<4xf32>
      %431 = "vector.fma"(%430, %263, %428) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %432 = "vector.extract"(%206) {position = [0]} : (vector<4xf32>) -> f32
      %433 = "vector.splat"(%432) : (f32) -> vector<4xf32>
      %434 = "vector.fma"(%433, %265, %431) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %435 = "vector.extract"(%206) {position = [1]} : (vector<4xf32>) -> f32
      %436 = "vector.splat"(%435) : (f32) -> vector<4xf32>
      %437 = "vector.fma"(%436, %267, %434) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %438 = "vector.extract"(%206) {position = [2]} : (vector<4xf32>) -> f32
      %439 = "vector.splat"(%438) : (f32) -> vector<4xf32>
      %440 = "vector.fma"(%439, %269, %437) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %441 = "vector.extract"(%206) {position = [3]} : (vector<4xf32>) -> f32
      %442 = "vector.splat"(%441) : (f32) -> vector<4xf32>
      %443 = "vector.fma"(%442, %271, %440) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %444 = "vector.extract"(%208) {position = [0]} : (vector<4xf32>) -> f32
      %445 = "vector.splat"(%444) : (f32) -> vector<4xf32>
      %446 = "vector.fma"(%445, %273, %443) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %447 = "vector.extract"(%208) {position = [1]} : (vector<4xf32>) -> f32
      %448 = "vector.splat"(%447) : (f32) -> vector<4xf32>
      %449 = "vector.fma"(%448, %275, %446) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %450 = "vector.extract"(%208) {position = [2]} : (vector<4xf32>) -> f32
      %451 = "vector.splat"(%450) : (f32) -> vector<4xf32>
      %452 = "vector.fma"(%451, %277, %449) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %453 = "vector.extract"(%208) {position = [3]} : (vector<4xf32>) -> f32
      %454 = "vector.splat"(%453) : (f32) -> vector<4xf32>
      %455 = "vector.fma"(%454, %279, %452) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %456 = "vector.extract"(%210) {position = [0]} : (vector<4xf32>) -> f32
      %457 = "vector.splat"(%456) : (f32) -> vector<4xf32>
      %458 = "vector.fma"(%457, %281, %455) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %459 = "vector.extract"(%210) {position = [1]} : (vector<4xf32>) -> f32
      %460 = "vector.splat"(%459) : (f32) -> vector<4xf32>
      %461 = "vector.fma"(%460, %283, %458) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %462 = "vector.extract"(%210) {position = [2]} : (vector<4xf32>) -> f32
      %463 = "vector.splat"(%462) : (f32) -> vector<4xf32>
      %464 = "vector.fma"(%463, %285, %461) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %465 = "vector.extract"(%210) {position = [3]} : (vector<4xf32>) -> f32
      %466 = "vector.splat"(%465) : (f32) -> vector<4xf32>
      %467 = "vector.fma"(%466, %287, %464) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %468 = "vector.extract"(%212) {position = [0]} : (vector<4xf32>) -> f32
      %469 = "vector.splat"(%468) : (f32) -> vector<4xf32>
      %470 = "vector.fma"(%469, %289, %467) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %471 = "vector.extract"(%212) {position = [1]} : (vector<4xf32>) -> f32
      %472 = "vector.splat"(%471) : (f32) -> vector<4xf32>
      %473 = "vector.fma"(%472, %291, %470) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %474 = "vector.extract"(%212) {position = [2]} : (vector<4xf32>) -> f32
      %475 = "vector.splat"(%474) : (f32) -> vector<4xf32>
      %476 = "vector.fma"(%475, %293, %473) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %477 = "vector.extract"(%212) {position = [3]} : (vector<4xf32>) -> f32
      %478 = "vector.splat"(%477) : (f32) -> vector<4xf32>
      %479 = "vector.fma"(%478, %295, %476) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %480 = "vector.extract"(%214) {position = [0]} : (vector<4xf32>) -> f32
      %481 = "vector.splat"(%480) : (f32) -> vector<4xf32>
      %482 = "vector.fma"(%481, %297, %479) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %483 = "vector.extract"(%214) {position = [1]} : (vector<4xf32>) -> f32
      %484 = "vector.splat"(%483) : (f32) -> vector<4xf32>
      %485 = "vector.fma"(%484, %299, %482) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %486 = "vector.extract"(%214) {position = [2]} : (vector<4xf32>) -> f32
      %487 = "vector.splat"(%486) : (f32) -> vector<4xf32>
      %488 = "vector.fma"(%487, %301, %485) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %489 = "vector.extract"(%214) {position = [3]} : (vector<4xf32>) -> f32
      %490 = "vector.splat"(%489) : (f32) -> vector<4xf32>
      %491 = "vector.fma"(%490, %303, %488) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %492 = "vector.extract"(%216) {position = [0]} : (vector<4xf32>) -> f32
      %493 = "vector.splat"(%492) : (f32) -> vector<4xf32>
      %494 = "vector.fma"(%493, %305, %491) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %495 = "vector.extract"(%216) {position = [1]} : (vector<4xf32>) -> f32
      %496 = "vector.splat"(%495) : (f32) -> vector<4xf32>
      %497 = "vector.fma"(%496, %307, %494) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %498 = "vector.extract"(%216) {position = [2]} : (vector<4xf32>) -> f32
      %499 = "vector.splat"(%498) : (f32) -> vector<4xf32>
      %500 = "vector.fma"(%499, %309, %497) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %501 = "vector.extract"(%216) {position = [3]} : (vector<4xf32>) -> f32
      %502 = "vector.splat"(%501) : (f32) -> vector<4xf32>
      %503 = "vector.fma"(%502, %311, %500) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %504 = "vector.extract"(%218) {position = [0]} : (vector<4xf32>) -> f32
      %505 = "vector.splat"(%504) : (f32) -> vector<4xf32>
      %506 = "vector.fma"(%505, %249, %172#2) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %507 = "vector.extract"(%218) {position = [1]} : (vector<4xf32>) -> f32
      %508 = "vector.splat"(%507) : (f32) -> vector<4xf32>
      %509 = "vector.fma"(%508, %251, %506) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %510 = "vector.extract"(%218) {position = [2]} : (vector<4xf32>) -> f32
      %511 = "vector.splat"(%510) : (f32) -> vector<4xf32>
      %512 = "vector.fma"(%511, %253, %509) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %513 = "vector.extract"(%218) {position = [3]} : (vector<4xf32>) -> f32
      %514 = "vector.splat"(%513) : (f32) -> vector<4xf32>
      %515 = "vector.fma"(%514, %255, %512) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %516 = "vector.extract"(%220) {position = [0]} : (vector<4xf32>) -> f32
      %517 = "vector.splat"(%516) : (f32) -> vector<4xf32>
      %518 = "vector.fma"(%517, %257, %515) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %519 = "vector.extract"(%220) {position = [1]} : (vector<4xf32>) -> f32
      %520 = "vector.splat"(%519) : (f32) -> vector<4xf32>
      %521 = "vector.fma"(%520, %259, %518) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %522 = "vector.extract"(%220) {position = [2]} : (vector<4xf32>) -> f32
      %523 = "vector.splat"(%522) : (f32) -> vector<4xf32>
      %524 = "vector.fma"(%523, %261, %521) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %525 = "vector.extract"(%220) {position = [3]} : (vector<4xf32>) -> f32
      %526 = "vector.splat"(%525) : (f32) -> vector<4xf32>
      %527 = "vector.fma"(%526, %263, %524) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %528 = "vector.extract"(%222) {position = [0]} : (vector<4xf32>) -> f32
      %529 = "vector.splat"(%528) : (f32) -> vector<4xf32>
      %530 = "vector.fma"(%529, %265, %527) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %531 = "vector.extract"(%222) {position = [1]} : (vector<4xf32>) -> f32
      %532 = "vector.splat"(%531) : (f32) -> vector<4xf32>
      %533 = "vector.fma"(%532, %267, %530) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %534 = "vector.extract"(%222) {position = [2]} : (vector<4xf32>) -> f32
      %535 = "vector.splat"(%534) : (f32) -> vector<4xf32>
      %536 = "vector.fma"(%535, %269, %533) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %537 = "vector.extract"(%222) {position = [3]} : (vector<4xf32>) -> f32
      %538 = "vector.splat"(%537) : (f32) -> vector<4xf32>
      %539 = "vector.fma"(%538, %271, %536) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %540 = "vector.extract"(%224) {position = [0]} : (vector<4xf32>) -> f32
      %541 = "vector.splat"(%540) : (f32) -> vector<4xf32>
      %542 = "vector.fma"(%541, %273, %539) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %543 = "vector.extract"(%224) {position = [1]} : (vector<4xf32>) -> f32
      %544 = "vector.splat"(%543) : (f32) -> vector<4xf32>
      %545 = "vector.fma"(%544, %275, %542) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %546 = "vector.extract"(%224) {position = [2]} : (vector<4xf32>) -> f32
      %547 = "vector.splat"(%546) : (f32) -> vector<4xf32>
      %548 = "vector.fma"(%547, %277, %545) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %549 = "vector.extract"(%224) {position = [3]} : (vector<4xf32>) -> f32
      %550 = "vector.splat"(%549) : (f32) -> vector<4xf32>
      %551 = "vector.fma"(%550, %279, %548) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %552 = "vector.extract"(%226) {position = [0]} : (vector<4xf32>) -> f32
      %553 = "vector.splat"(%552) : (f32) -> vector<4xf32>
      %554 = "vector.fma"(%553, %281, %551) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %555 = "vector.extract"(%226) {position = [1]} : (vector<4xf32>) -> f32
      %556 = "vector.splat"(%555) : (f32) -> vector<4xf32>
      %557 = "vector.fma"(%556, %283, %554) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %558 = "vector.extract"(%226) {position = [2]} : (vector<4xf32>) -> f32
      %559 = "vector.splat"(%558) : (f32) -> vector<4xf32>
      %560 = "vector.fma"(%559, %285, %557) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %561 = "vector.extract"(%226) {position = [3]} : (vector<4xf32>) -> f32
      %562 = "vector.splat"(%561) : (f32) -> vector<4xf32>
      %563 = "vector.fma"(%562, %287, %560) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %564 = "vector.extract"(%228) {position = [0]} : (vector<4xf32>) -> f32
      %565 = "vector.splat"(%564) : (f32) -> vector<4xf32>
      %566 = "vector.fma"(%565, %289, %563) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %567 = "vector.extract"(%228) {position = [1]} : (vector<4xf32>) -> f32
      %568 = "vector.splat"(%567) : (f32) -> vector<4xf32>
      %569 = "vector.fma"(%568, %291, %566) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %570 = "vector.extract"(%228) {position = [2]} : (vector<4xf32>) -> f32
      %571 = "vector.splat"(%570) : (f32) -> vector<4xf32>
      %572 = "vector.fma"(%571, %293, %569) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %573 = "vector.extract"(%228) {position = [3]} : (vector<4xf32>) -> f32
      %574 = "vector.splat"(%573) : (f32) -> vector<4xf32>
      %575 = "vector.fma"(%574, %295, %572) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %576 = "vector.extract"(%230) {position = [0]} : (vector<4xf32>) -> f32
      %577 = "vector.splat"(%576) : (f32) -> vector<4xf32>
      %578 = "vector.fma"(%577, %297, %575) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %579 = "vector.extract"(%230) {position = [1]} : (vector<4xf32>) -> f32
      %580 = "vector.splat"(%579) : (f32) -> vector<4xf32>
      %581 = "vector.fma"(%580, %299, %578) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %582 = "vector.extract"(%230) {position = [2]} : (vector<4xf32>) -> f32
      %583 = "vector.splat"(%582) : (f32) -> vector<4xf32>
      %584 = "vector.fma"(%583, %301, %581) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %585 = "vector.extract"(%230) {position = [3]} : (vector<4xf32>) -> f32
      %586 = "vector.splat"(%585) : (f32) -> vector<4xf32>
      %587 = "vector.fma"(%586, %303, %584) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %588 = "vector.extract"(%232) {position = [0]} : (vector<4xf32>) -> f32
      %589 = "vector.splat"(%588) : (f32) -> vector<4xf32>
      %590 = "vector.fma"(%589, %305, %587) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %591 = "vector.extract"(%232) {position = [1]} : (vector<4xf32>) -> f32
      %592 = "vector.splat"(%591) : (f32) -> vector<4xf32>
      %593 = "vector.fma"(%592, %307, %590) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %594 = "vector.extract"(%232) {position = [2]} : (vector<4xf32>) -> f32
      %595 = "vector.splat"(%594) : (f32) -> vector<4xf32>
      %596 = "vector.fma"(%595, %309, %593) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %597 = "vector.extract"(%232) {position = [3]} : (vector<4xf32>) -> f32
      %598 = "vector.splat"(%597) : (f32) -> vector<4xf32>
      %599 = "vector.fma"(%598, %311, %596) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %600 = "vector.extract"(%234) {position = [0]} : (vector<4xf32>) -> f32
      %601 = "vector.splat"(%600) : (f32) -> vector<4xf32>
      %602 = "vector.fma"(%601, %249, %172#3) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %603 = "vector.extract"(%234) {position = [1]} : (vector<4xf32>) -> f32
      %604 = "vector.splat"(%603) : (f32) -> vector<4xf32>
      %605 = "vector.fma"(%604, %251, %602) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %606 = "vector.extract"(%234) {position = [2]} : (vector<4xf32>) -> f32
      %607 = "vector.splat"(%606) : (f32) -> vector<4xf32>
      %608 = "vector.fma"(%607, %253, %605) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %609 = "vector.extract"(%234) {position = [3]} : (vector<4xf32>) -> f32
      %610 = "vector.splat"(%609) : (f32) -> vector<4xf32>
      %611 = "vector.fma"(%610, %255, %608) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %612 = "vector.extract"(%236) {position = [0]} : (vector<4xf32>) -> f32
      %613 = "vector.splat"(%612) : (f32) -> vector<4xf32>
      %614 = "vector.fma"(%613, %257, %611) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %615 = "vector.extract"(%236) {position = [1]} : (vector<4xf32>) -> f32
      %616 = "vector.splat"(%615) : (f32) -> vector<4xf32>
      %617 = "vector.fma"(%616, %259, %614) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %618 = "vector.extract"(%236) {position = [2]} : (vector<4xf32>) -> f32
      %619 = "vector.splat"(%618) : (f32) -> vector<4xf32>
      %620 = "vector.fma"(%619, %261, %617) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %621 = "vector.extract"(%236) {position = [3]} : (vector<4xf32>) -> f32
      %622 = "vector.splat"(%621) : (f32) -> vector<4xf32>
      %623 = "vector.fma"(%622, %263, %620) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %624 = "vector.extract"(%238) {position = [0]} : (vector<4xf32>) -> f32
      %625 = "vector.splat"(%624) : (f32) -> vector<4xf32>
      %626 = "vector.fma"(%625, %265, %623) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %627 = "vector.extract"(%238) {position = [1]} : (vector<4xf32>) -> f32
      %628 = "vector.splat"(%627) : (f32) -> vector<4xf32>
      %629 = "vector.fma"(%628, %267, %626) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %630 = "vector.extract"(%238) {position = [2]} : (vector<4xf32>) -> f32
      %631 = "vector.splat"(%630) : (f32) -> vector<4xf32>
      %632 = "vector.fma"(%631, %269, %629) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %633 = "vector.extract"(%238) {position = [3]} : (vector<4xf32>) -> f32
      %634 = "vector.splat"(%633) : (f32) -> vector<4xf32>
      %635 = "vector.fma"(%634, %271, %632) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %636 = "vector.extract"(%240) {position = [0]} : (vector<4xf32>) -> f32
      %637 = "vector.splat"(%636) : (f32) -> vector<4xf32>
      %638 = "vector.fma"(%637, %273, %635) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %639 = "vector.extract"(%240) {position = [1]} : (vector<4xf32>) -> f32
      %640 = "vector.splat"(%639) : (f32) -> vector<4xf32>
      %641 = "vector.fma"(%640, %275, %638) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %642 = "vector.extract"(%240) {position = [2]} : (vector<4xf32>) -> f32
      %643 = "vector.splat"(%642) : (f32) -> vector<4xf32>
      %644 = "vector.fma"(%643, %277, %641) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %645 = "vector.extract"(%240) {position = [3]} : (vector<4xf32>) -> f32
      %646 = "vector.splat"(%645) : (f32) -> vector<4xf32>
      %647 = "vector.fma"(%646, %279, %644) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %648 = "vector.extract"(%242) {position = [0]} : (vector<4xf32>) -> f32
      %649 = "vector.splat"(%648) : (f32) -> vector<4xf32>
      %650 = "vector.fma"(%649, %281, %647) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %651 = "vector.extract"(%242) {position = [1]} : (vector<4xf32>) -> f32
      %652 = "vector.splat"(%651) : (f32) -> vector<4xf32>
      %653 = "vector.fma"(%652, %283, %650) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %654 = "vector.extract"(%242) {position = [2]} : (vector<4xf32>) -> f32
      %655 = "vector.splat"(%654) : (f32) -> vector<4xf32>
      %656 = "vector.fma"(%655, %285, %653) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %657 = "vector.extract"(%242) {position = [3]} : (vector<4xf32>) -> f32
      %658 = "vector.splat"(%657) : (f32) -> vector<4xf32>
      %659 = "vector.fma"(%658, %287, %656) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %660 = "vector.extract"(%244) {position = [0]} : (vector<4xf32>) -> f32
      %661 = "vector.splat"(%660) : (f32) -> vector<4xf32>
      %662 = "vector.fma"(%661, %289, %659) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %663 = "vector.extract"(%244) {position = [1]} : (vector<4xf32>) -> f32
      %664 = "vector.splat"(%663) : (f32) -> vector<4xf32>
      %665 = "vector.fma"(%664, %291, %662) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %666 = "vector.extract"(%244) {position = [2]} : (vector<4xf32>) -> f32
      %667 = "vector.splat"(%666) : (f32) -> vector<4xf32>
      %668 = "vector.fma"(%667, %293, %665) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %669 = "vector.extract"(%244) {position = [3]} : (vector<4xf32>) -> f32
      %670 = "vector.splat"(%669) : (f32) -> vector<4xf32>
      %671 = "vector.fma"(%670, %295, %668) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %672 = "vector.extract"(%246) {position = [0]} : (vector<4xf32>) -> f32
      %673 = "vector.splat"(%672) : (f32) -> vector<4xf32>
      %674 = "vector.fma"(%673, %297, %671) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %675 = "vector.extract"(%246) {position = [1]} : (vector<4xf32>) -> f32
      %676 = "vector.splat"(%675) : (f32) -> vector<4xf32>
      %677 = "vector.fma"(%676, %299, %674) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %678 = "vector.extract"(%246) {position = [2]} : (vector<4xf32>) -> f32
      %679 = "vector.splat"(%678) : (f32) -> vector<4xf32>
      %680 = "vector.fma"(%679, %301, %677) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %681 = "vector.extract"(%246) {position = [3]} : (vector<4xf32>) -> f32
      %682 = "vector.splat"(%681) : (f32) -> vector<4xf32>
      %683 = "vector.fma"(%682, %303, %680) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %684 = "vector.extract"(%248) {position = [0]} : (vector<4xf32>) -> f32
      %685 = "vector.splat"(%684) : (f32) -> vector<4xf32>
      %686 = "vector.fma"(%685, %305, %683) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %687 = "vector.extract"(%248) {position = [1]} : (vector<4xf32>) -> f32
      %688 = "vector.splat"(%687) : (f32) -> vector<4xf32>
      %689 = "vector.fma"(%688, %307, %686) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %690 = "vector.extract"(%248) {position = [2]} : (vector<4xf32>) -> f32
      %691 = "vector.splat"(%690) : (f32) -> vector<4xf32>
      %692 = "vector.fma"(%691, %309, %689) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      %693 = "vector.extract"(%248) {position = [3]} : (vector<4xf32>) -> f32
      %694 = "vector.splat"(%693) : (f32) -> vector<4xf32>
      %695 = "vector.fma"(%694, %311, %692) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
      "memref.store"(%695, %109, %130) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
      "memref.store"(%599, %110, %129) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
      "memref.store"(%503, %111, %128) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
      "memref.store"(%407, %112, %127) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
      "func.return"() : () -> ()
    }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [16, 16, 1]>, sym_name = "forward_dispatch_35_matmul_18432x320x320"} : () -> ()
  }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
  "hal.executable.variant_end"() : () -> ()
}) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
    %133 = linalg.matmul ins(%collapsed_749, %130 : tensor<18432x320xf32>, tensor<320x320xf32>) outs(%132 : tensor<18432x320xf32>) -> tensor<18432x320xf32>
           ^
/home/prashant/stable.mlir:1320:12: error: failed to serialize executables
    %133 = linalg.matmul ins(%collapsed_749, %130 : tensor<18432x320xf32>, tensor<320x320xf32>) outs(%132 : tensor<18432x320xf32>) -> tensor<18432x320xf32>
           ^
/home/prashant/stable.mlir:24:3: note: called from
  func.func @forward(%arg0: tensor<1x4x96x96xf32>, %arg1: tensor<1xf32>, %arg2: tensor<2x64x1024xf32>, %arg3: tensor<f32>) -> tensor<1x4x96x96xf32> {
  ^
/home/prashant/stable.mlir:1320:12: note: see current operation: 
"hal.executable"() ({
  "hal.executable.variant"() ({
    "hal.executable.export"() ({
    ^bb0(%arg0: !hal.device, %arg1: index, %arg2: index, %arg3: index):
      %0 = "arith.constant"() {value = 5 : index} : () -> index
      %1 = "arith.constant"() {value = 288 : index} : () -> index
      %2 = "arith.constant"() {value = 1 : index} : () -> index
      "hal.return"(%0, %1, %2) : (index, index, index) -> ()
    }) {layout = #hal.pipeline.layout<push_constants = 2, sets = [<0, bindings = [<0, storage_buffer, ReadOnly>, <1, storage_buffer, ReadOnly>, <2, storage_buffer>]>]>, ordinal = 0 : index, sym_name = "forward_dispatch_35_matmul_18432x320x320", translation_info = #iree_codegen.translation_info<SPIRVMatmulPromoteVectorize pipeline_depth = 1>, workgroup_size = [16 : index, 16 : index, 1 : index]} : () -> ()
    "builtin.module"() ({
      "spirv.GlobalVariable"() {binding = 0 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_0_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {binding = 1 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_1_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
      "spirv.GlobalVariable"() {binding = 2 : i32, descriptor_set = 0 : i32, sym_name = "__resource_var_0_2_", type = !spirv.ptr<!spirv.struct<(!spirv.rtarray<vector<4xf32>, stride=16> [0])>, StorageBuffer>} : () -> ()
      "func.func"() ({
        %0 = "arith.constant"() {value = -33 : index} : () -> index
        %1 = "arith.constant"() {value = 527 : index} : () -> index
        %2 = "arith.constant"() {value = 510 : index} : () -> index
        %3 = "arith.constant"() {value = 493 : index} : () -> index
        %4 = "arith.constant"() {value = 476 : index} : () -> index
        %5 = "arith.constant"() {value = 459 : index} : () -> index
        %6 = "arith.constant"() {value = 442 : index} : () -> index
        %7 = "arith.constant"() {value = 425 : index} : () -> index
        %8 = "arith.constant"() {value = 408 : index} : () -> index
        %9 = "arith.constant"() {value = 391 : index} : () -> index
        %10 = "arith.constant"() {value = 374 : index} : () -> index
        %11 = "arith.constant"() {value = 357 : index} : () -> index
        %12 = "arith.constant"() {value = 340 : index} : () -> index
        %13 = "arith.constant"() {value = 323 : index} : () -> index
        %14 = "arith.constant"() {value = 306 : index} : () -> index
        %15 = "arith.constant"() {value = 289 : index} : () -> index
        %16 = "arith.constant"() {value = 255 : index} : () -> index
        %17 = "arith.constant"() {value = 238 : index} : () -> index
        %18 = "arith.constant"() {value = 221 : index} : () -> index
        %19 = "arith.constant"() {value = 204 : index} : () -> index
        %20 = "arith.constant"() {value = 187 : index} : () -> index
        %21 = "arith.constant"() {value = 170 : index} : () -> index
        %22 = "arith.constant"() {value = 153 : index} : () -> index
        %23 = "arith.constant"() {value = 136 : index} : () -> index
        %24 = "arith.constant"() {value = 119 : index} : () -> index
        %25 = "arith.constant"() {value = 102 : index} : () -> index
        %26 = "arith.constant"() {value = 85 : index} : () -> index
        %27 = "arith.constant"() {value = 68 : index} : () -> index
        %28 = "arith.constant"() {value = 51 : index} : () -> index
        %29 = "arith.constant"() {value = 34 : index} : () -> index
        %30 = "arith.constant"() {value = 33 : index} : () -> index
        %31 = "arith.constant"() {value = 31 : index} : () -> index
        %32 = "arith.constant"() {value = 30 : index} : () -> index
        %33 = "arith.constant"() {value = 29 : index} : () -> index
        %34 = "arith.constant"() {value = 28 : index} : () -> index
        %35 = "arith.constant"() {value = 27 : index} : () -> index
        %36 = "arith.constant"() {value = 25 : index} : () -> index
        %37 = "arith.constant"() {value = 24 : index} : () -> index
        %38 = "arith.constant"() {value = 23 : index} : () -> index
        %39 = "arith.constant"() {value = 22 : index} : () -> index
        %40 = "arith.constant"() {value = 21 : index} : () -> index
        %41 = "arith.constant"() {value = 20 : index} : () -> index
        %42 = "arith.constant"() {value = 19 : index} : () -> index
        %43 = "arith.constant"() {value = 15 : index} : () -> index
        %44 = "arith.constant"() {value = 14 : index} : () -> index
        %45 = "arith.constant"() {value = 13 : index} : () -> index
        %46 = "arith.constant"() {value = 12 : index} : () -> index
        %47 = "arith.constant"() {value = 11 : index} : () -> index
        %48 = "arith.constant"() {value = 10 : index} : () -> index
        %49 = "arith.constant"() {value = 9 : index} : () -> index
        %50 = "arith.constant"() {value = 7 : index} : () -> index
        %51 = "arith.constant"() {value = 6 : index} : () -> index
        %52 = "arith.constant"() {value = 5 : index} : () -> index
        %53 = "arith.constant"() {value = 4 : index} : () -> index
        %54 = "arith.constant"() {value = 3 : index} : () -> index
        %55 = "arith.constant"() {value = 2 : index} : () -> index
        %56 = "arith.constant"() {value = 1 : index} : () -> index
        %57 = "arith.constant"() {value = 36 : index} : () -> index
        %58 = "arith.constant"() {value = 272 : index} : () -> index
        %59 = "arith.constant"() {value = 17 : index} : () -> index
        %60 = "arith.constant"() {value = 18 : index} : () -> index
        %61 = "arith.constant"() {value = 64 : index} : () -> index
        %62 = "arith.constant"() {value = 1280 : index} : () -> index
        %63 = "arith.constant"() {value = 1477120 : index} : () -> index
        %64 = "arith.constant"() {value = 72 : index} : () -> index
        %65 = "arith.constant"() {value = 8 : index} : () -> index
        %66 = "arith.constant"() {value = 2560 : index} : () -> index
        %67 = "arith.constant"() {value = 240 : index} : () -> index
        %68 = "arith.constant"() {value = 160 : index} : () -> index
        %69 = "arith.constant"() {value = 80 : index} : () -> index
        %70 = "arith.constant"() {value = -1 : index} : () -> index
        %71 = "arith.constant"() {value = 16 : index} : () -> index
        %72 = "arith.constant"() {value = 320 : index} : () -> index
        %73 = "arith.constant"() {value = 5120 : index} : () -> index
        %74 = "arith.constant"() {value = 0 : index} : () -> index
        %75 = "arith.constant"() {value = 1474560 : index} : () -> index
        %76 = "arith.constant"() {value = 25600 : index} : () -> index
        %77 = "arith.constant"() {value = 288 : index} : () -> index
        %78 = "arith.constant"() {value = dense<0.000000e+00> : vector<4xf32>} : () -> vector<4xf32>
        %79 = "arith.constant"() {value = 23592960 : index} : () -> index
        %80 = "arith.constant"() {value = 32 : index} : () -> index
        %81 = "gpu.thread_id"() {dimension = #gpu<dim x>} : () -> index
        %82 = "gpu.thread_id"() {dimension = #gpu<dim y>} : () -> index
        %83 = "gpu.thread_id"() {dimension = #gpu<dim z>} : () -> index
        %84 = "memref.alloc"() {operand_segment_sizes = array<i32: 0, 0>} : () -> memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>
        %85 = "memref.alloc"() {operand_segment_sizes = array<i32: 0, 0>} : () -> memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>
        %86 = "hal.interface.constant.load"() {index = 0 : index} : () -> i32
        %87 = "hal.interface.constant.load"() {index = 1 : index} : () -> i32
        %88 = "arith.index_castui"(%86) : (i32) -> index
        %89 = "arith.index_castui"(%87) : (i32) -> index
        %90 = "hal.interface.binding.subspan"(%79, %75) {alignment = 64 : index, binding = 0 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %91 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %92 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %93 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %94 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 0 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %95 = "hal.interface.binding.subspan"(%88, %76) {alignment = 64 : index, binding = 1 : index, descriptor_flags = 1 : i32, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %96 = "hal.interface.binding.subspan"(%74, %76) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %97 = "hal.interface.binding.subspan"(%74, %76) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %98 = "hal.interface.binding.subspan"(%74, %76) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %99 = "hal.interface.binding.subspan"(%74, %76) {alignment = 64 : index, binding = 1 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %100 = "hal.interface.binding.subspan"(%89, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %101 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %102 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %103 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %104 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %105 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %106 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %107 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %108 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %109 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %110 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %111 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %112 = "hal.interface.binding.subspan"(%74, %75) {alignment = 64 : index, binding = 2 : index, descriptor_type = #hal.descriptor_type<storage_buffer>, operand_segment_sizes = array<i32: 1, 1>, set = 0 : index} : (index, index) -> memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>
        %113 = "hal.interface.workgroup.id"() {dimension = 0 : index} : () -> index
        %114 = "hal.interface.workgroup.id"() {dimension = 1 : index} : () -> index
        %115 = "arith.muli"(%114, %73) : (index, index) -> index
        %116 = "arith.muli"(%82, %72) : (index, index) -> index
        %117 = "arith.addi"(%115, %116) : (index, index) -> index
        %118 = "arith.muli"(%113, %71) : (index, index) -> index
        %119 = "arith.addi"(%117, %118) : (index, index) -> index
        %120 = "arith.addi"(%119, %81) : (index, index) -> index
        %121 = "arith.cmpi"(%89, %74) {predicate = 2 : i64} : (index, index) -> i1
        %122 = "arith.subi"(%70, %89) : (index, index) -> index
        %123 = "arith.select"(%121, %122, %89) : (i1, index, index) -> index
        %124 = "arith.divsi"(%123, %71) : (index, index) -> index
        %125 = "arith.subi"(%70, %124) : (index, index) -> index
        %126 = "arith.select"(%121, %125, %124) : (i1, index, index) -> index
        %127 = "arith.addi"(%120, %126) : (index, index) -> index
        "memref.store"(%78, %101, %127) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
        %128 = "arith.addi"(%127, %69) : (index, index) -> index
        "memref.store"(%78, %102, %128) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
        %129 = "arith.addi"(%127, %68) : (index, index) -> index
        "memref.store"(%78, %103, %129) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
        %130 = "arith.addi"(%127, %67) : (index, index) -> index
        "memref.store"(%78, %104, %130) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
        %131 = "memref.load"(%105, %127) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %132 = "memref.load"(%106, %128) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %133 = "memref.load"(%107, %129) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %134 = "memref.load"(%108, %130) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %135 = "arith.addi"(%115, %81) : (index, index) -> index
        %136 = "arith.muli"(%82, %68) : (index, index) -> index
        %137 = "arith.addi"(%135, %136) : (index, index) -> index
        %138 = "arith.muli"(%83, %66) : (index, index) -> index
        %139 = "arith.addi"(%137, %138) : (index, index) -> index
        %140 = "arith.cmpi"(%81, %74) {predicate = 2 : i64} : (index, index) -> i1
        %141 = "arith.subi"(%70, %81) : (index, index) -> index
        %142 = "arith.select"(%140, %141, %81) : (i1, index, index) -> index
        %143 = "arith.divsi"(%142, %65) : (index, index) -> index
        %144 = "arith.subi"(%70, %143) : (index, index) -> index
        %145 = "arith.select"(%140, %144, %143) : (i1, index, index) -> index
        %146 = "arith.muli"(%145, %64) : (index, index) -> index
        %147 = "arith.addi"(%139, %146) : (index, index) -> index
        %148 = "arith.addi"(%147, %75) : (index, index) -> index
        %149 = "memref.load"(%91, %148) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %150 = "arith.addi"(%147, %63) : (index, index) -> index
        %151 = "memref.load"(%92, %150) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %152 = "arith.muli"(%82, %69) : (index, index) -> index
        %153 = "arith.addi"(%81, %152) : (index, index) -> index
        %154 = "arith.muli"(%83, %62) : (index, index) -> index
        %155 = "arith.addi"(%153, %154) : (index, index) -> index
        %156 = "arith.addi"(%155, %118) : (index, index) -> index
        %157 = "arith.divsi"(%142, %71) : (index, index) -> index
        %158 = "arith.subi"(%70, %157) : (index, index) -> index
        %159 = "arith.select"(%140, %158, %157) : (i1, index, index) -> index
        %160 = "arith.muli"(%159, %61) : (index, index) -> index
        %161 = "arith.addi"(%156, %160) : (index, index) -> index
        %162 = "arith.cmpi"(%88, %74) {predicate = 2 : i64} : (index, index) -> i1
        %163 = "arith.subi"(%70, %88) : (index, index) -> index
        %164 = "arith.select"(%162, %163, %88) : (i1, index, index) -> index
        %165 = "arith.divsi"(%164, %71) : (index, index) -> index
        %166 = "arith.subi"(%70, %165) : (index, index) -> index
        %167 = "arith.select"(%162, %166, %165) : (i1, index, index) -> index
        %168 = "arith.addi"(%161, %167) : (index, index) -> index
        %169 = "memref.load"(%96, %168) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %170 = "arith.addi"(%168, %62) : (index, index) -> index
        %171 = "memref.load"(%97, %170) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
        %172:8 = "scf.for"(%74, %77, %80, %131, %132, %133, %134, %149, %151, %169, %171) ({
        ^bb0(%arg0: index, %arg1: vector<4xf32>, %arg2: vector<4xf32>, %arg3: vector<4xf32>, %arg4: vector<4xf32>, %arg5: vector<4xf32>, %arg6: vector<4xf32>, %arg7: vector<4xf32>, %arg8: vector<4xf32>):
          "gpu.barrier"() : () -> ()
          %696 = "arith.muli"(%82, %60) : (index, index) -> index
          %697 = "arith.addi"(%81, %696) : (index, index) -> index
          %698 = "arith.muli"(%83, %77) : (index, index) -> index
          %699 = "arith.addi"(%697, %698) : (index, index) -> index
          %700 = "arith.addi"(%699, %145) : (index, index) -> index
          "memref.store"(%arg5, %84, %700) {nontemporal = false} : (vector<4xf32>, memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
          %701 = "arith.addi"(%700, %77) : (index, index) -> index
          "memref.store"(%arg6, %84, %701) {nontemporal = false} : (vector<4xf32>, memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
          %702 = "arith.muli"(%82, %59) : (index, index) -> index
          %703 = "arith.addi"(%81, %702) : (index, index) -> index
          %704 = "arith.muli"(%83, %58) : (index, index) -> index
          %705 = "arith.addi"(%703, %704) : (index, index) -> index
          %706 = "arith.addi"(%705, %159) : (index, index) -> index
          "memref.store"(%arg7, %85, %706) {nontemporal = false} : (vector<4xf32>, memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
          %707 = "arith.addi"(%706, %58) : (index, index) -> index
          "memref.store"(%arg8, %85, %707) {nontemporal = false} : (vector<4xf32>, memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
          "gpu.barrier"() : () -> ()
          %708 = "arith.muli"(%82, %57) : (index, index) -> index
          %709 = "memref.load"(%84, %708) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %710 = "arith.addi"(%708, %56) : (index, index) -> index
          %711 = "memref.load"(%84, %710) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %712 = "arith.addi"(%708, %55) : (index, index) -> index
          %713 = "memref.load"(%84, %712) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %714 = "arith.addi"(%708, %54) : (index, index) -> index
          %715 = "memref.load"(%84, %714) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %716 = "arith.addi"(%708, %53) : (index, index) -> index
          %717 = "memref.load"(%84, %716) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %718 = "arith.addi"(%708, %52) : (index, index) -> index
          %719 = "memref.load"(%84, %718) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %720 = "arith.addi"(%708, %51) : (index, index) -> index
          %721 = "memref.load"(%84, %720) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %722 = "arith.addi"(%708, %50) : (index, index) -> index
          %723 = "memref.load"(%84, %722) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %724 = "arith.addi"(%708, %49) : (index, index) -> index
          %725 = "memref.load"(%84, %724) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %726 = "arith.addi"(%708, %48) : (index, index) -> index
          %727 = "memref.load"(%84, %726) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %728 = "arith.addi"(%708, %47) : (index, index) -> index
          %729 = "memref.load"(%84, %728) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %730 = "arith.addi"(%708, %46) : (index, index) -> index
          %731 = "memref.load"(%84, %730) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %732 = "arith.addi"(%708, %45) : (index, index) -> index
          %733 = "memref.load"(%84, %732) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %734 = "arith.addi"(%708, %44) : (index, index) -> index
          %735 = "memref.load"(%84, %734) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %736 = "arith.addi"(%708, %43) : (index, index) -> index
          %737 = "memref.load"(%84, %736) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %738 = "arith.addi"(%708, %71) : (index, index) -> index
          %739 = "memref.load"(%84, %738) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %740 = "arith.addi"(%708, %60) : (index, index) -> index
          %741 = "memref.load"(%84, %740) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %742 = "arith.addi"(%708, %42) : (index, index) -> index
          %743 = "memref.load"(%84, %742) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %744 = "arith.addi"(%708, %41) : (index, index) -> index
          %745 = "memref.load"(%84, %744) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %746 = "arith.addi"(%708, %40) : (index, index) -> index
          %747 = "memref.load"(%84, %746) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %748 = "arith.addi"(%708, %39) : (index, index) -> index
          %749 = "memref.load"(%84, %748) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %750 = "arith.addi"(%708, %38) : (index, index) -> index
          %751 = "memref.load"(%84, %750) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %752 = "arith.addi"(%708, %37) : (index, index) -> index
          %753 = "memref.load"(%84, %752) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %754 = "arith.addi"(%708, %36) : (index, index) -> index
          %755 = "memref.load"(%84, %754) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %756 = "arith.addi"(%708, %35) : (index, index) -> index
          %757 = "memref.load"(%84, %756) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %758 = "arith.addi"(%708, %34) : (index, index) -> index
          %759 = "memref.load"(%84, %758) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %760 = "arith.addi"(%708, %33) : (index, index) -> index
          %761 = "memref.load"(%84, %760) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %762 = "arith.addi"(%708, %32) : (index, index) -> index
          %763 = "memref.load"(%84, %762) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %764 = "arith.addi"(%708, %31) : (index, index) -> index
          %765 = "memref.load"(%84, %764) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %766 = "arith.addi"(%708, %80) : (index, index) -> index
          %767 = "memref.load"(%84, %766) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %768 = "arith.addi"(%708, %30) : (index, index) -> index
          %769 = "memref.load"(%84, %768) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %770 = "arith.addi"(%708, %29) : (index, index) -> index
          %771 = "memref.load"(%84, %770) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %772 = "memref.load"(%85, %81) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %773 = "arith.addi"(%81, %59) : (index, index) -> index
          %774 = "memref.load"(%85, %773) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %775 = "arith.addi"(%81, %29) : (index, index) -> index
          %776 = "memref.load"(%85, %775) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %777 = "arith.addi"(%81, %28) : (index, index) -> index
          %778 = "memref.load"(%85, %777) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %779 = "arith.addi"(%81, %27) : (index, index) -> index
          %780 = "memref.load"(%85, %779) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %781 = "arith.addi"(%81, %26) : (index, index) -> index
          %782 = "memref.load"(%85, %781) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %783 = "arith.addi"(%81, %25) : (index, index) -> index
          %784 = "memref.load"(%85, %783) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %785 = "arith.addi"(%81, %24) : (index, index) -> index
          %786 = "memref.load"(%85, %785) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %787 = "arith.addi"(%81, %23) : (index, index) -> index
          %788 = "memref.load"(%85, %787) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %789 = "arith.addi"(%81, %22) : (index, index) -> index
          %790 = "memref.load"(%85, %789) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %791 = "arith.addi"(%81, %21) : (index, index) -> index
          %792 = "memref.load"(%85, %791) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %793 = "arith.addi"(%81, %20) : (index, index) -> index
          %794 = "memref.load"(%85, %793) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %795 = "arith.addi"(%81, %19) : (index, index) -> index
          %796 = "memref.load"(%85, %795) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %797 = "arith.addi"(%81, %18) : (index, index) -> index
          %798 = "memref.load"(%85, %797) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %799 = "arith.addi"(%81, %17) : (index, index) -> index
          %800 = "memref.load"(%85, %799) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %801 = "arith.addi"(%81, %16) : (index, index) -> index
          %802 = "memref.load"(%85, %801) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %803 = "arith.addi"(%81, %58) : (index, index) -> index
          %804 = "memref.load"(%85, %803) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %805 = "arith.addi"(%81, %15) : (index, index) -> index
          %806 = "memref.load"(%85, %805) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %807 = "arith.addi"(%81, %14) : (index, index) -> index
          %808 = "memref.load"(%85, %807) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %809 = "arith.addi"(%81, %13) : (index, index) -> index
          %810 = "memref.load"(%85, %809) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %811 = "arith.addi"(%81, %12) : (index, index) -> index
          %812 = "memref.load"(%85, %811) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %813 = "arith.addi"(%81, %11) : (index, index) -> index
          %814 = "memref.load"(%85, %813) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %815 = "arith.addi"(%81, %10) : (index, index) -> index
          %816 = "memref.load"(%85, %815) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %817 = "arith.addi"(%81, %9) : (index, index) -> index
          %818 = "memref.load"(%85, %817) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %819 = "arith.addi"(%81, %8) : (index, index) -> index
          %820 = "memref.load"(%85, %819) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %821 = "arith.addi"(%81, %7) : (index, index) -> index
          %822 = "memref.load"(%85, %821) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %823 = "arith.addi"(%81, %6) : (index, index) -> index
          %824 = "memref.load"(%85, %823) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %825 = "arith.addi"(%81, %5) : (index, index) -> index
          %826 = "memref.load"(%85, %825) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %827 = "arith.addi"(%81, %4) : (index, index) -> index
          %828 = "memref.load"(%85, %827) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %829 = "arith.addi"(%81, %3) : (index, index) -> index
          %830 = "memref.load"(%85, %829) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %831 = "arith.addi"(%81, %2) : (index, index) -> index
          %832 = "memref.load"(%85, %831) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %833 = "arith.addi"(%81, %1) : (index, index) -> index
          %834 = "memref.load"(%85, %833) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
          %835 = "vector.extract"(%709) {position = [0]} : (vector<4xf32>) -> f32
          %836 = "vector.splat"(%835) : (f32) -> vector<4xf32>
          %837 = "vector.fma"(%836, %772, %arg1) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %838 = "vector.extract"(%709) {position = [1]} : (vector<4xf32>) -> f32
          %839 = "vector.splat"(%838) : (f32) -> vector<4xf32>
          %840 = "vector.fma"(%839, %774, %837) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %841 = "vector.extract"(%709) {position = [2]} : (vector<4xf32>) -> f32
          %842 = "vector.splat"(%841) : (f32) -> vector<4xf32>
          %843 = "vector.fma"(%842, %776, %840) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %844 = "vector.extract"(%709) {position = [3]} : (vector<4xf32>) -> f32
          %845 = "vector.splat"(%844) : (f32) -> vector<4xf32>
          %846 = "vector.fma"(%845, %778, %843) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %847 = "vector.extract"(%711) {position = [0]} : (vector<4xf32>) -> f32
          %848 = "vector.splat"(%847) : (f32) -> vector<4xf32>
          %849 = "vector.fma"(%848, %780, %846) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %850 = "vector.extract"(%711) {position = [1]} : (vector<4xf32>) -> f32
          %851 = "vector.splat"(%850) : (f32) -> vector<4xf32>
          %852 = "vector.fma"(%851, %782, %849) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %853 = "vector.extract"(%711) {position = [2]} : (vector<4xf32>) -> f32
          %854 = "vector.splat"(%853) : (f32) -> vector<4xf32>
          %855 = "vector.fma"(%854, %784, %852) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %856 = "vector.extract"(%711) {position = [3]} : (vector<4xf32>) -> f32
          %857 = "vector.splat"(%856) : (f32) -> vector<4xf32>
          %858 = "vector.fma"(%857, %786, %855) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %859 = "vector.extract"(%713) {position = [0]} : (vector<4xf32>) -> f32
          %860 = "vector.splat"(%859) : (f32) -> vector<4xf32>
          %861 = "vector.fma"(%860, %788, %858) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %862 = "vector.extract"(%713) {position = [1]} : (vector<4xf32>) -> f32
          %863 = "vector.splat"(%862) : (f32) -> vector<4xf32>
          %864 = "vector.fma"(%863, %790, %861) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %865 = "vector.extract"(%713) {position = [2]} : (vector<4xf32>) -> f32
          %866 = "vector.splat"(%865) : (f32) -> vector<4xf32>
          %867 = "vector.fma"(%866, %792, %864) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %868 = "vector.extract"(%713) {position = [3]} : (vector<4xf32>) -> f32
          %869 = "vector.splat"(%868) : (f32) -> vector<4xf32>
          %870 = "vector.fma"(%869, %794, %867) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %871 = "vector.extract"(%715) {position = [0]} : (vector<4xf32>) -> f32
          %872 = "vector.splat"(%871) : (f32) -> vector<4xf32>
          %873 = "vector.fma"(%872, %796, %870) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %874 = "vector.extract"(%715) {position = [1]} : (vector<4xf32>) -> f32
          %875 = "vector.splat"(%874) : (f32) -> vector<4xf32>
          %876 = "vector.fma"(%875, %798, %873) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %877 = "vector.extract"(%715) {position = [2]} : (vector<4xf32>) -> f32
          %878 = "vector.splat"(%877) : (f32) -> vector<4xf32>
          %879 = "vector.fma"(%878, %800, %876) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %880 = "vector.extract"(%715) {position = [3]} : (vector<4xf32>) -> f32
          %881 = "vector.splat"(%880) : (f32) -> vector<4xf32>
          %882 = "vector.fma"(%881, %802, %879) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %883 = "vector.extract"(%717) {position = [0]} : (vector<4xf32>) -> f32
          %884 = "vector.splat"(%883) : (f32) -> vector<4xf32>
          %885 = "vector.fma"(%884, %804, %882) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %886 = "vector.extract"(%717) {position = [1]} : (vector<4xf32>) -> f32
          %887 = "vector.splat"(%886) : (f32) -> vector<4xf32>
          %888 = "vector.fma"(%887, %806, %885) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %889 = "vector.extract"(%717) {position = [2]} : (vector<4xf32>) -> f32
          %890 = "vector.splat"(%889) : (f32) -> vector<4xf32>
          %891 = "vector.fma"(%890, %808, %888) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %892 = "vector.extract"(%717) {position = [3]} : (vector<4xf32>) -> f32
          %893 = "vector.splat"(%892) : (f32) -> vector<4xf32>
          %894 = "vector.fma"(%893, %810, %891) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %895 = "vector.extract"(%719) {position = [0]} : (vector<4xf32>) -> f32
          %896 = "vector.splat"(%895) : (f32) -> vector<4xf32>
          %897 = "vector.fma"(%896, %812, %894) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %898 = "vector.extract"(%719) {position = [1]} : (vector<4xf32>) -> f32
          %899 = "vector.splat"(%898) : (f32) -> vector<4xf32>
          %900 = "vector.fma"(%899, %814, %897) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %901 = "vector.extract"(%719) {position = [2]} : (vector<4xf32>) -> f32
          %902 = "vector.splat"(%901) : (f32) -> vector<4xf32>
          %903 = "vector.fma"(%902, %816, %900) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %904 = "vector.extract"(%719) {position = [3]} : (vector<4xf32>) -> f32
          %905 = "vector.splat"(%904) : (f32) -> vector<4xf32>
          %906 = "vector.fma"(%905, %818, %903) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %907 = "vector.extract"(%721) {position = [0]} : (vector<4xf32>) -> f32
          %908 = "vector.splat"(%907) : (f32) -> vector<4xf32>
          %909 = "vector.fma"(%908, %820, %906) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %910 = "vector.extract"(%721) {position = [1]} : (vector<4xf32>) -> f32
          %911 = "vector.splat"(%910) : (f32) -> vector<4xf32>
          %912 = "vector.fma"(%911, %822, %909) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %913 = "vector.extract"(%721) {position = [2]} : (vector<4xf32>) -> f32
          %914 = "vector.splat"(%913) : (f32) -> vector<4xf32>
          %915 = "vector.fma"(%914, %824, %912) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %916 = "vector.extract"(%721) {position = [3]} : (vector<4xf32>) -> f32
          %917 = "vector.splat"(%916) : (f32) -> vector<4xf32>
          %918 = "vector.fma"(%917, %826, %915) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %919 = "vector.extract"(%723) {position = [0]} : (vector<4xf32>) -> f32
          %920 = "vector.splat"(%919) : (f32) -> vector<4xf32>
          %921 = "vector.fma"(%920, %828, %918) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %922 = "vector.extract"(%723) {position = [1]} : (vector<4xf32>) -> f32
          %923 = "vector.splat"(%922) : (f32) -> vector<4xf32>
          %924 = "vector.fma"(%923, %830, %921) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %925 = "vector.extract"(%723) {position = [2]} : (vector<4xf32>) -> f32
          %926 = "vector.splat"(%925) : (f32) -> vector<4xf32>
          %927 = "vector.fma"(%926, %832, %924) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %928 = "vector.extract"(%723) {position = [3]} : (vector<4xf32>) -> f32
          %929 = "vector.splat"(%928) : (f32) -> vector<4xf32>
          %930 = "vector.fma"(%929, %834, %927) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %931 = "vector.extract"(%725) {position = [0]} : (vector<4xf32>) -> f32
          %932 = "vector.splat"(%931) : (f32) -> vector<4xf32>
          %933 = "vector.fma"(%932, %772, %arg2) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %934 = "vector.extract"(%725) {position = [1]} : (vector<4xf32>) -> f32
          %935 = "vector.splat"(%934) : (f32) -> vector<4xf32>
          %936 = "vector.fma"(%935, %774, %933) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %937 = "vector.extract"(%725) {position = [2]} : (vector<4xf32>) -> f32
          %938 = "vector.splat"(%937) : (f32) -> vector<4xf32>
          %939 = "vector.fma"(%938, %776, %936) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %940 = "vector.extract"(%725) {position = [3]} : (vector<4xf32>) -> f32
          %941 = "vector.splat"(%940) : (f32) -> vector<4xf32>
          %942 = "vector.fma"(%941, %778, %939) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %943 = "vector.extract"(%727) {position = [0]} : (vector<4xf32>) -> f32
          %944 = "vector.splat"(%943) : (f32) -> vector<4xf32>
          %945 = "vector.fma"(%944, %780, %942) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %946 = "vector.extract"(%727) {position = [1]} : (vector<4xf32>) -> f32
          %947 = "vector.splat"(%946) : (f32) -> vector<4xf32>
          %948 = "vector.fma"(%947, %782, %945) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %949 = "vector.extract"(%727) {position = [2]} : (vector<4xf32>) -> f32
          %950 = "vector.splat"(%949) : (f32) -> vector<4xf32>
          %951 = "vector.fma"(%950, %784, %948) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %952 = "vector.extract"(%727) {position = [3]} : (vector<4xf32>) -> f32
          %953 = "vector.splat"(%952) : (f32) -> vector<4xf32>
          %954 = "vector.fma"(%953, %786, %951) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %955 = "vector.extract"(%729) {position = [0]} : (vector<4xf32>) -> f32
          %956 = "vector.splat"(%955) : (f32) -> vector<4xf32>
          %957 = "vector.fma"(%956, %788, %954) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %958 = "vector.extract"(%729) {position = [1]} : (vector<4xf32>) -> f32
          %959 = "vector.splat"(%958) : (f32) -> vector<4xf32>
          %960 = "vector.fma"(%959, %790, %957) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %961 = "vector.extract"(%729) {position = [2]} : (vector<4xf32>) -> f32
          %962 = "vector.splat"(%961) : (f32) -> vector<4xf32>
          %963 = "vector.fma"(%962, %792, %960) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %964 = "vector.extract"(%729) {position = [3]} : (vector<4xf32>) -> f32
          %965 = "vector.splat"(%964) : (f32) -> vector<4xf32>
          %966 = "vector.fma"(%965, %794, %963) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %967 = "vector.extract"(%731) {position = [0]} : (vector<4xf32>) -> f32
          %968 = "vector.splat"(%967) : (f32) -> vector<4xf32>
          %969 = "vector.fma"(%968, %796, %966) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %970 = "vector.extract"(%731) {position = [1]} : (vector<4xf32>) -> f32
          %971 = "vector.splat"(%970) : (f32) -> vector<4xf32>
          %972 = "vector.fma"(%971, %798, %969) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %973 = "vector.extract"(%731) {position = [2]} : (vector<4xf32>) -> f32
          %974 = "vector.splat"(%973) : (f32) -> vector<4xf32>
          %975 = "vector.fma"(%974, %800, %972) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %976 = "vector.extract"(%731) {position = [3]} : (vector<4xf32>) -> f32
          %977 = "vector.splat"(%976) : (f32) -> vector<4xf32>
          %978 = "vector.fma"(%977, %802, %975) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %979 = "vector.extract"(%733) {position = [0]} : (vector<4xf32>) -> f32
          %980 = "vector.splat"(%979) : (f32) -> vector<4xf32>
          %981 = "vector.fma"(%980, %804, %978) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %982 = "vector.extract"(%733) {position = [1]} : (vector<4xf32>) -> f32
          %983 = "vector.splat"(%982) : (f32) -> vector<4xf32>
          %984 = "vector.fma"(%983, %806, %981) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %985 = "vector.extract"(%733) {position = [2]} : (vector<4xf32>) -> f32
          %986 = "vector.splat"(%985) : (f32) -> vector<4xf32>
          %987 = "vector.fma"(%986, %808, %984) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %988 = "vector.extract"(%733) {position = [3]} : (vector<4xf32>) -> f32
          %989 = "vector.splat"(%988) : (f32) -> vector<4xf32>
          %990 = "vector.fma"(%989, %810, %987) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %991 = "vector.extract"(%735) {position = [0]} : (vector<4xf32>) -> f32
          %992 = "vector.splat"(%991) : (f32) -> vector<4xf32>
          %993 = "vector.fma"(%992, %812, %990) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %994 = "vector.extract"(%735) {position = [1]} : (vector<4xf32>) -> f32
          %995 = "vector.splat"(%994) : (f32) -> vector<4xf32>
          %996 = "vector.fma"(%995, %814, %993) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %997 = "vector.extract"(%735) {position = [2]} : (vector<4xf32>) -> f32
          %998 = "vector.splat"(%997) : (f32) -> vector<4xf32>
          %999 = "vector.fma"(%998, %816, %996) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1000 = "vector.extract"(%735) {position = [3]} : (vector<4xf32>) -> f32
          %1001 = "vector.splat"(%1000) : (f32) -> vector<4xf32>
          %1002 = "vector.fma"(%1001, %818, %999) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1003 = "vector.extract"(%737) {position = [0]} : (vector<4xf32>) -> f32
          %1004 = "vector.splat"(%1003) : (f32) -> vector<4xf32>
          %1005 = "vector.fma"(%1004, %820, %1002) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1006 = "vector.extract"(%737) {position = [1]} : (vector<4xf32>) -> f32
          %1007 = "vector.splat"(%1006) : (f32) -> vector<4xf32>
          %1008 = "vector.fma"(%1007, %822, %1005) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1009 = "vector.extract"(%737) {position = [2]} : (vector<4xf32>) -> f32
          %1010 = "vector.splat"(%1009) : (f32) -> vector<4xf32>
          %1011 = "vector.fma"(%1010, %824, %1008) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1012 = "vector.extract"(%737) {position = [3]} : (vector<4xf32>) -> f32
          %1013 = "vector.splat"(%1012) : (f32) -> vector<4xf32>
          %1014 = "vector.fma"(%1013, %826, %1011) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1015 = "vector.extract"(%739) {position = [0]} : (vector<4xf32>) -> f32
          %1016 = "vector.splat"(%1015) : (f32) -> vector<4xf32>
          %1017 = "vector.fma"(%1016, %828, %1014) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1018 = "vector.extract"(%739) {position = [1]} : (vector<4xf32>) -> f32
          %1019 = "vector.splat"(%1018) : (f32) -> vector<4xf32>
          %1020 = "vector.fma"(%1019, %830, %1017) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1021 = "vector.extract"(%739) {position = [2]} : (vector<4xf32>) -> f32
          %1022 = "vector.splat"(%1021) : (f32) -> vector<4xf32>
          %1023 = "vector.fma"(%1022, %832, %1020) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1024 = "vector.extract"(%739) {position = [3]} : (vector<4xf32>) -> f32
          %1025 = "vector.splat"(%1024) : (f32) -> vector<4xf32>
          %1026 = "vector.fma"(%1025, %834, %1023) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1027 = "vector.extract"(%741) {position = [0]} : (vector<4xf32>) -> f32
          %1028 = "vector.splat"(%1027) : (f32) -> vector<4xf32>
          %1029 = "vector.fma"(%1028, %772, %arg3) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1030 = "vector.extract"(%741) {position = [1]} : (vector<4xf32>) -> f32
          %1031 = "vector.splat"(%1030) : (f32) -> vector<4xf32>
          %1032 = "vector.fma"(%1031, %774, %1029) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1033 = "vector.extract"(%741) {position = [2]} : (vector<4xf32>) -> f32
          %1034 = "vector.splat"(%1033) : (f32) -> vector<4xf32>
          %1035 = "vector.fma"(%1034, %776, %1032) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1036 = "vector.extract"(%741) {position = [3]} : (vector<4xf32>) -> f32
          %1037 = "vector.splat"(%1036) : (f32) -> vector<4xf32>
          %1038 = "vector.fma"(%1037, %778, %1035) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1039 = "vector.extract"(%743) {position = [0]} : (vector<4xf32>) -> f32
          %1040 = "vector.splat"(%1039) : (f32) -> vector<4xf32>
          %1041 = "vector.fma"(%1040, %780, %1038) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1042 = "vector.extract"(%743) {position = [1]} : (vector<4xf32>) -> f32
          %1043 = "vector.splat"(%1042) : (f32) -> vector<4xf32>
          %1044 = "vector.fma"(%1043, %782, %1041) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1045 = "vector.extract"(%743) {position = [2]} : (vector<4xf32>) -> f32
          %1046 = "vector.splat"(%1045) : (f32) -> vector<4xf32>
          %1047 = "vector.fma"(%1046, %784, %1044) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1048 = "vector.extract"(%743) {position = [3]} : (vector<4xf32>) -> f32
          %1049 = "vector.splat"(%1048) : (f32) -> vector<4xf32>
          %1050 = "vector.fma"(%1049, %786, %1047) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1051 = "vector.extract"(%745) {position = [0]} : (vector<4xf32>) -> f32
          %1052 = "vector.splat"(%1051) : (f32) -> vector<4xf32>
          %1053 = "vector.fma"(%1052, %788, %1050) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1054 = "vector.extract"(%745) {position = [1]} : (vector<4xf32>) -> f32
          %1055 = "vector.splat"(%1054) : (f32) -> vector<4xf32>
          %1056 = "vector.fma"(%1055, %790, %1053) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1057 = "vector.extract"(%745) {position = [2]} : (vector<4xf32>) -> f32
          %1058 = "vector.splat"(%1057) : (f32) -> vector<4xf32>
          %1059 = "vector.fma"(%1058, %792, %1056) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1060 = "vector.extract"(%745) {position = [3]} : (vector<4xf32>) -> f32
          %1061 = "vector.splat"(%1060) : (f32) -> vector<4xf32>
          %1062 = "vector.fma"(%1061, %794, %1059) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1063 = "vector.extract"(%747) {position = [0]} : (vector<4xf32>) -> f32
          %1064 = "vector.splat"(%1063) : (f32) -> vector<4xf32>
          %1065 = "vector.fma"(%1064, %796, %1062) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1066 = "vector.extract"(%747) {position = [1]} : (vector<4xf32>) -> f32
          %1067 = "vector.splat"(%1066) : (f32) -> vector<4xf32>
          %1068 = "vector.fma"(%1067, %798, %1065) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1069 = "vector.extract"(%747) {position = [2]} : (vector<4xf32>) -> f32
          %1070 = "vector.splat"(%1069) : (f32) -> vector<4xf32>
          %1071 = "vector.fma"(%1070, %800, %1068) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1072 = "vector.extract"(%747) {position = [3]} : (vector<4xf32>) -> f32
          %1073 = "vector.splat"(%1072) : (f32) -> vector<4xf32>
          %1074 = "vector.fma"(%1073, %802, %1071) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1075 = "vector.extract"(%749) {position = [0]} : (vector<4xf32>) -> f32
          %1076 = "vector.splat"(%1075) : (f32) -> vector<4xf32>
          %1077 = "vector.fma"(%1076, %804, %1074) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1078 = "vector.extract"(%749) {position = [1]} : (vector<4xf32>) -> f32
          %1079 = "vector.splat"(%1078) : (f32) -> vector<4xf32>
          %1080 = "vector.fma"(%1079, %806, %1077) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1081 = "vector.extract"(%749) {position = [2]} : (vector<4xf32>) -> f32
          %1082 = "vector.splat"(%1081) : (f32) -> vector<4xf32>
          %1083 = "vector.fma"(%1082, %808, %1080) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1084 = "vector.extract"(%749) {position = [3]} : (vector<4xf32>) -> f32
          %1085 = "vector.splat"(%1084) : (f32) -> vector<4xf32>
          %1086 = "vector.fma"(%1085, %810, %1083) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1087 = "vector.extract"(%751) {position = [0]} : (vector<4xf32>) -> f32
          %1088 = "vector.splat"(%1087) : (f32) -> vector<4xf32>
          %1089 = "vector.fma"(%1088, %812, %1086) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1090 = "vector.extract"(%751) {position = [1]} : (vector<4xf32>) -> f32
          %1091 = "vector.splat"(%1090) : (f32) -> vector<4xf32>
          %1092 = "vector.fma"(%1091, %814, %1089) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1093 = "vector.extract"(%751) {position = [2]} : (vector<4xf32>) -> f32
          %1094 = "vector.splat"(%1093) : (f32) -> vector<4xf32>
          %1095 = "vector.fma"(%1094, %816, %1092) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1096 = "vector.extract"(%751) {position = [3]} : (vector<4xf32>) -> f32
          %1097 = "vector.splat"(%1096) : (f32) -> vector<4xf32>
          %1098 = "vector.fma"(%1097, %818, %1095) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1099 = "vector.extract"(%753) {position = [0]} : (vector<4xf32>) -> f32
          %1100 = "vector.splat"(%1099) : (f32) -> vector<4xf32>
          %1101 = "vector.fma"(%1100, %820, %1098) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1102 = "vector.extract"(%753) {position = [1]} : (vector<4xf32>) -> f32
          %1103 = "vector.splat"(%1102) : (f32) -> vector<4xf32>
          %1104 = "vector.fma"(%1103, %822, %1101) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1105 = "vector.extract"(%753) {position = [2]} : (vector<4xf32>) -> f32
          %1106 = "vector.splat"(%1105) : (f32) -> vector<4xf32>
          %1107 = "vector.fma"(%1106, %824, %1104) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1108 = "vector.extract"(%753) {position = [3]} : (vector<4xf32>) -> f32
          %1109 = "vector.splat"(%1108) : (f32) -> vector<4xf32>
          %1110 = "vector.fma"(%1109, %826, %1107) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1111 = "vector.extract"(%755) {position = [0]} : (vector<4xf32>) -> f32
          %1112 = "vector.splat"(%1111) : (f32) -> vector<4xf32>
          %1113 = "vector.fma"(%1112, %828, %1110) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1114 = "vector.extract"(%755) {position = [1]} : (vector<4xf32>) -> f32
          %1115 = "vector.splat"(%1114) : (f32) -> vector<4xf32>
          %1116 = "vector.fma"(%1115, %830, %1113) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1117 = "vector.extract"(%755) {position = [2]} : (vector<4xf32>) -> f32
          %1118 = "vector.splat"(%1117) : (f32) -> vector<4xf32>
          %1119 = "vector.fma"(%1118, %832, %1116) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1120 = "vector.extract"(%755) {position = [3]} : (vector<4xf32>) -> f32
          %1121 = "vector.splat"(%1120) : (f32) -> vector<4xf32>
          %1122 = "vector.fma"(%1121, %834, %1119) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1123 = "vector.extract"(%757) {position = [0]} : (vector<4xf32>) -> f32
          %1124 = "vector.splat"(%1123) : (f32) -> vector<4xf32>
          %1125 = "vector.fma"(%1124, %772, %arg4) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1126 = "vector.extract"(%757) {position = [1]} : (vector<4xf32>) -> f32
          %1127 = "vector.splat"(%1126) : (f32) -> vector<4xf32>
          %1128 = "vector.fma"(%1127, %774, %1125) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1129 = "vector.extract"(%757) {position = [2]} : (vector<4xf32>) -> f32
          %1130 = "vector.splat"(%1129) : (f32) -> vector<4xf32>
          %1131 = "vector.fma"(%1130, %776, %1128) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1132 = "vector.extract"(%757) {position = [3]} : (vector<4xf32>) -> f32
          %1133 = "vector.splat"(%1132) : (f32) -> vector<4xf32>
          %1134 = "vector.fma"(%1133, %778, %1131) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1135 = "vector.extract"(%759) {position = [0]} : (vector<4xf32>) -> f32
          %1136 = "vector.splat"(%1135) : (f32) -> vector<4xf32>
          %1137 = "vector.fma"(%1136, %780, %1134) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1138 = "vector.extract"(%759) {position = [1]} : (vector<4xf32>) -> f32
          %1139 = "vector.splat"(%1138) : (f32) -> vector<4xf32>
          %1140 = "vector.fma"(%1139, %782, %1137) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1141 = "vector.extract"(%759) {position = [2]} : (vector<4xf32>) -> f32
          %1142 = "vector.splat"(%1141) : (f32) -> vector<4xf32>
          %1143 = "vector.fma"(%1142, %784, %1140) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1144 = "vector.extract"(%759) {position = [3]} : (vector<4xf32>) -> f32
          %1145 = "vector.splat"(%1144) : (f32) -> vector<4xf32>
          %1146 = "vector.fma"(%1145, %786, %1143) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1147 = "vector.extract"(%761) {position = [0]} : (vector<4xf32>) -> f32
          %1148 = "vector.splat"(%1147) : (f32) -> vector<4xf32>
          %1149 = "vector.fma"(%1148, %788, %1146) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1150 = "vector.extract"(%761) {position = [1]} : (vector<4xf32>) -> f32
          %1151 = "vector.splat"(%1150) : (f32) -> vector<4xf32>
          %1152 = "vector.fma"(%1151, %790, %1149) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1153 = "vector.extract"(%761) {position = [2]} : (vector<4xf32>) -> f32
          %1154 = "vector.splat"(%1153) : (f32) -> vector<4xf32>
          %1155 = "vector.fma"(%1154, %792, %1152) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1156 = "vector.extract"(%761) {position = [3]} : (vector<4xf32>) -> f32
          %1157 = "vector.splat"(%1156) : (f32) -> vector<4xf32>
          %1158 = "vector.fma"(%1157, %794, %1155) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1159 = "vector.extract"(%763) {position = [0]} : (vector<4xf32>) -> f32
          %1160 = "vector.splat"(%1159) : (f32) -> vector<4xf32>
          %1161 = "vector.fma"(%1160, %796, %1158) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1162 = "vector.extract"(%763) {position = [1]} : (vector<4xf32>) -> f32
          %1163 = "vector.splat"(%1162) : (f32) -> vector<4xf32>
          %1164 = "vector.fma"(%1163, %798, %1161) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1165 = "vector.extract"(%763) {position = [2]} : (vector<4xf32>) -> f32
          %1166 = "vector.splat"(%1165) : (f32) -> vector<4xf32>
          %1167 = "vector.fma"(%1166, %800, %1164) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1168 = "vector.extract"(%763) {position = [3]} : (vector<4xf32>) -> f32
          %1169 = "vector.splat"(%1168) : (f32) -> vector<4xf32>
          %1170 = "vector.fma"(%1169, %802, %1167) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1171 = "vector.extract"(%765) {position = [0]} : (vector<4xf32>) -> f32
          %1172 = "vector.splat"(%1171) : (f32) -> vector<4xf32>
          %1173 = "vector.fma"(%1172, %804, %1170) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1174 = "vector.extract"(%765) {position = [1]} : (vector<4xf32>) -> f32
          %1175 = "vector.splat"(%1174) : (f32) -> vector<4xf32>
          %1176 = "vector.fma"(%1175, %806, %1173) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1177 = "vector.extract"(%765) {position = [2]} : (vector<4xf32>) -> f32
          %1178 = "vector.splat"(%1177) : (f32) -> vector<4xf32>
          %1179 = "vector.fma"(%1178, %808, %1176) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1180 = "vector.extract"(%765) {position = [3]} : (vector<4xf32>) -> f32
          %1181 = "vector.splat"(%1180) : (f32) -> vector<4xf32>
          %1182 = "vector.fma"(%1181, %810, %1179) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1183 = "vector.extract"(%767) {position = [0]} : (vector<4xf32>) -> f32
          %1184 = "vector.splat"(%1183) : (f32) -> vector<4xf32>
          %1185 = "vector.fma"(%1184, %812, %1182) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1186 = "vector.extract"(%767) {position = [1]} : (vector<4xf32>) -> f32
          %1187 = "vector.splat"(%1186) : (f32) -> vector<4xf32>
          %1188 = "vector.fma"(%1187, %814, %1185) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1189 = "vector.extract"(%767) {position = [2]} : (vector<4xf32>) -> f32
          %1190 = "vector.splat"(%1189) : (f32) -> vector<4xf32>
          %1191 = "vector.fma"(%1190, %816, %1188) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1192 = "vector.extract"(%767) {position = [3]} : (vector<4xf32>) -> f32
          %1193 = "vector.splat"(%1192) : (f32) -> vector<4xf32>
          %1194 = "vector.fma"(%1193, %818, %1191) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1195 = "vector.extract"(%769) {position = [0]} : (vector<4xf32>) -> f32
          %1196 = "vector.splat"(%1195) : (f32) -> vector<4xf32>
          %1197 = "vector.fma"(%1196, %820, %1194) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1198 = "vector.extract"(%769) {position = [1]} : (vector<4xf32>) -> f32
          %1199 = "vector.splat"(%1198) : (f32) -> vector<4xf32>
          %1200 = "vector.fma"(%1199, %822, %1197) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1201 = "vector.extract"(%769) {position = [2]} : (vector<4xf32>) -> f32
          %1202 = "vector.splat"(%1201) : (f32) -> vector<4xf32>
          %1203 = "vector.fma"(%1202, %824, %1200) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1204 = "vector.extract"(%769) {position = [3]} : (vector<4xf32>) -> f32
          %1205 = "vector.splat"(%1204) : (f32) -> vector<4xf32>
          %1206 = "vector.fma"(%1205, %826, %1203) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1207 = "vector.extract"(%771) {position = [0]} : (vector<4xf32>) -> f32
          %1208 = "vector.splat"(%1207) : (f32) -> vector<4xf32>
          %1209 = "vector.fma"(%1208, %828, %1206) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1210 = "vector.extract"(%771) {position = [1]} : (vector<4xf32>) -> f32
          %1211 = "vector.splat"(%1210) : (f32) -> vector<4xf32>
          %1212 = "vector.fma"(%1211, %830, %1209) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1213 = "vector.extract"(%771) {position = [2]} : (vector<4xf32>) -> f32
          %1214 = "vector.splat"(%1213) : (f32) -> vector<4xf32>
          %1215 = "vector.fma"(%1214, %832, %1212) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1216 = "vector.extract"(%771) {position = [3]} : (vector<4xf32>) -> f32
          %1217 = "vector.splat"(%1216) : (f32) -> vector<4xf32>
          %1218 = "vector.fma"(%1217, %834, %1215) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
          %1219 = "arith.addi"(%arg0, %80) : (index, index) -> index
          %1220 = "arith.cmpi"(%1219, %74) {predicate = 2 : i64} : (index, index) -> i1
          %1221 = "arith.subi"(%0, %arg0) : (index, index) -> index
          %1222 = "arith.select"(%1220, %1221, %1219) : (i1, index, index) -> index
          %1223 = "arith.divsi"(%1222, %53) : (index, index) -> index
          %1224 = "arith.subi"(%70, %1223) : (index, index) -> index
          %1225 = "arith.select"(%1220, %1224, %1223) : (i1, index, index) -> index
          %1226 = "arith.addi"(%139, %1225) : (index, index) -> index
          %1227 = "arith.addi"(%1226, %146) : (index, index) -> index
          %1228 = "arith.addi"(%1227, %75) : (index, index) -> index
          %1229 = "memref.load"(%93, %1228) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
          %1230 = "arith.addi"(%1227, %63) : (index, index) -> index
          %1231 = "memref.load"(%94, %1230) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
          %1232 = "arith.muli"(%1219, %69) : (index, index) -> index
          %1233 = "arith.addi"(%1232, %81) : (index, index) -> index
          %1234 = "arith.addi"(%1233, %152) : (index, index) -> index
          %1235 = "arith.addi"(%1234, %154) : (index, index) -> index
          %1236 = "arith.addi"(%1235, %118) : (index, index) -> index
          %1237 = "arith.addi"(%1236, %160) : (index, index) -> index
          %1238 = "arith.addi"(%1237, %167) : (index, index) -> index
          %1239 = "memref.load"(%98, %1238) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
          %1240 = "arith.addi"(%1238, %62) : (index, index) -> index
          %1241 = "memref.load"(%99, %1240) : (memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> vector<4xf32>
          "scf.yield"(%930, %1026, %1122, %1218, %1229, %1231, %1239, %1241) : (vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>) -> ()
        }) : (index, index, index, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>) -> (vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>, vector<4xf32>)
        "gpu.barrier"() : () -> ()
        %173 = "arith.muli"(%82, %60) : (index, index) -> index
        %174 = "arith.addi"(%81, %173) : (index, index) -> index
        %175 = "arith.muli"(%83, %77) : (index, index) -> index
        %176 = "arith.addi"(%174, %175) : (index, index) -> index
        %177 = "arith.addi"(%176, %145) : (index, index) -> index
        "memref.store"(%172#4, %84, %177) {nontemporal = false} : (vector<4xf32>, memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
        %178 = "arith.addi"(%177, %77) : (index, index) -> index
        "memref.store"(%172#5, %84, %178) {nontemporal = false} : (vector<4xf32>, memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
        %179 = "arith.muli"(%82, %59) : (index, index) -> index
        %180 = "arith.addi"(%81, %179) : (index, index) -> index
        %181 = "arith.muli"(%83, %58) : (index, index) -> index
        %182 = "arith.addi"(%180, %181) : (index, index) -> index
        %183 = "arith.addi"(%182, %159) : (index, index) -> index
        "memref.store"(%172#6, %85, %183) {nontemporal = false} : (vector<4xf32>, memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
        %184 = "arith.addi"(%183, %58) : (index, index) -> index
        "memref.store"(%172#7, %85, %184) {nontemporal = false} : (vector<4xf32>, memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> ()
        "gpu.barrier"() : () -> ()
        %185 = "arith.muli"(%82, %57) : (index, index) -> index
        %186 = "memref.load"(%84, %185) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %187 = "arith.addi"(%185, %56) : (index, index) -> index
        %188 = "memref.load"(%84, %187) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %189 = "arith.addi"(%185, %55) : (index, index) -> index
        %190 = "memref.load"(%84, %189) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %191 = "arith.addi"(%185, %54) : (index, index) -> index
        %192 = "memref.load"(%84, %191) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %193 = "arith.addi"(%185, %53) : (index, index) -> index
        %194 = "memref.load"(%84, %193) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %195 = "arith.addi"(%185, %52) : (index, index) -> index
        %196 = "memref.load"(%84, %195) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %197 = "arith.addi"(%185, %51) : (index, index) -> index
        %198 = "memref.load"(%84, %197) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %199 = "arith.addi"(%185, %50) : (index, index) -> index
        %200 = "memref.load"(%84, %199) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %201 = "arith.addi"(%185, %49) : (index, index) -> index
        %202 = "memref.load"(%84, %201) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %203 = "arith.addi"(%185, %48) : (index, index) -> index
        %204 = "memref.load"(%84, %203) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %205 = "arith.addi"(%185, %47) : (index, index) -> index
        %206 = "memref.load"(%84, %205) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %207 = "arith.addi"(%185, %46) : (index, index) -> index
        %208 = "memref.load"(%84, %207) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %209 = "arith.addi"(%185, %45) : (index, index) -> index
        %210 = "memref.load"(%84, %209) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %211 = "arith.addi"(%185, %44) : (index, index) -> index
        %212 = "memref.load"(%84, %211) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %213 = "arith.addi"(%185, %43) : (index, index) -> index
        %214 = "memref.load"(%84, %213) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %215 = "arith.addi"(%185, %71) : (index, index) -> index
        %216 = "memref.load"(%84, %215) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %217 = "arith.addi"(%185, %60) : (index, index) -> index
        %218 = "memref.load"(%84, %217) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %219 = "arith.addi"(%185, %42) : (index, index) -> index
        %220 = "memref.load"(%84, %219) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %221 = "arith.addi"(%185, %41) : (index, index) -> index
        %222 = "memref.load"(%84, %221) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %223 = "arith.addi"(%185, %40) : (index, index) -> index
        %224 = "memref.load"(%84, %223) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %225 = "arith.addi"(%185, %39) : (index, index) -> index
        %226 = "memref.load"(%84, %225) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %227 = "arith.addi"(%185, %38) : (index, index) -> index
        %228 = "memref.load"(%84, %227) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %229 = "arith.addi"(%185, %37) : (index, index) -> index
        %230 = "memref.load"(%84, %229) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %231 = "arith.addi"(%185, %36) : (index, index) -> index
        %232 = "memref.load"(%84, %231) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %233 = "arith.addi"(%185, %35) : (index, index) -> index
        %234 = "memref.load"(%84, %233) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %235 = "arith.addi"(%185, %34) : (index, index) -> index
        %236 = "memref.load"(%84, %235) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %237 = "arith.addi"(%185, %33) : (index, index) -> index
        %238 = "memref.load"(%84, %237) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %239 = "arith.addi"(%185, %32) : (index, index) -> index
        %240 = "memref.load"(%84, %239) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %241 = "arith.addi"(%185, %31) : (index, index) -> index
        %242 = "memref.load"(%84, %241) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %243 = "arith.addi"(%185, %80) : (index, index) -> index
        %244 = "memref.load"(%84, %243) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %245 = "arith.addi"(%185, %30) : (index, index) -> index
        %246 = "memref.load"(%84, %245) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %247 = "arith.addi"(%185, %29) : (index, index) -> index
        %248 = "memref.load"(%84, %247) {nontemporal = false} : (memref<576xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %249 = "memref.load"(%85, %81) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %250 = "arith.addi"(%81, %59) : (index, index) -> index
        %251 = "memref.load"(%85, %250) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %252 = "arith.addi"(%81, %29) : (index, index) -> index
        %253 = "memref.load"(%85, %252) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %254 = "arith.addi"(%81, %28) : (index, index) -> index
        %255 = "memref.load"(%85, %254) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %256 = "arith.addi"(%81, %27) : (index, index) -> index
        %257 = "memref.load"(%85, %256) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %258 = "arith.addi"(%81, %26) : (index, index) -> index
        %259 = "memref.load"(%85, %258) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %260 = "arith.addi"(%81, %25) : (index, index) -> index
        %261 = "memref.load"(%85, %260) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %262 = "arith.addi"(%81, %24) : (index, index) -> index
        %263 = "memref.load"(%85, %262) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %264 = "arith.addi"(%81, %23) : (index, index) -> index
        %265 = "memref.load"(%85, %264) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %266 = "arith.addi"(%81, %22) : (index, index) -> index
        %267 = "memref.load"(%85, %266) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %268 = "arith.addi"(%81, %21) : (index, index) -> index
        %269 = "memref.load"(%85, %268) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %270 = "arith.addi"(%81, %20) : (index, index) -> index
        %271 = "memref.load"(%85, %270) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %272 = "arith.addi"(%81, %19) : (index, index) -> index
        %273 = "memref.load"(%85, %272) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %274 = "arith.addi"(%81, %18) : (index, index) -> index
        %275 = "memref.load"(%85, %274) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %276 = "arith.addi"(%81, %17) : (index, index) -> index
        %277 = "memref.load"(%85, %276) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %278 = "arith.addi"(%81, %16) : (index, index) -> index
        %279 = "memref.load"(%85, %278) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %280 = "arith.addi"(%81, %58) : (index, index) -> index
        %281 = "memref.load"(%85, %280) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %282 = "arith.addi"(%81, %15) : (index, index) -> index
        %283 = "memref.load"(%85, %282) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %284 = "arith.addi"(%81, %14) : (index, index) -> index
        %285 = "memref.load"(%85, %284) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %286 = "arith.addi"(%81, %13) : (index, index) -> index
        %287 = "memref.load"(%85, %286) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %288 = "arith.addi"(%81, %12) : (index, index) -> index
        %289 = "memref.load"(%85, %288) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %290 = "arith.addi"(%81, %11) : (index, index) -> index
        %291 = "memref.load"(%85, %290) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %292 = "arith.addi"(%81, %10) : (index, index) -> index
        %293 = "memref.load"(%85, %292) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %294 = "arith.addi"(%81, %9) : (index, index) -> index
        %295 = "memref.load"(%85, %294) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %296 = "arith.addi"(%81, %8) : (index, index) -> index
        %297 = "memref.load"(%85, %296) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %298 = "arith.addi"(%81, %7) : (index, index) -> index
        %299 = "memref.load"(%85, %298) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %300 = "arith.addi"(%81, %6) : (index, index) -> index
        %301 = "memref.load"(%85, %300) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %302 = "arith.addi"(%81, %5) : (index, index) -> index
        %303 = "memref.load"(%85, %302) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %304 = "arith.addi"(%81, %4) : (index, index) -> index
        %305 = "memref.load"(%85, %304) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %306 = "arith.addi"(%81, %3) : (index, index) -> index
        %307 = "memref.load"(%85, %306) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %308 = "arith.addi"(%81, %2) : (index, index) -> index
        %309 = "memref.load"(%85, %308) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %310 = "arith.addi"(%81, %1) : (index, index) -> index
        %311 = "memref.load"(%85, %310) {nontemporal = false} : (memref<544xvector<4xf32>, #spirv.storage_class<Workgroup>>, index) -> vector<4xf32>
        %312 = "vector.extract"(%186) {position = [0]} : (vector<4xf32>) -> f32
        %313 = "vector.splat"(%312) : (f32) -> vector<4xf32>
        %314 = "vector.fma"(%313, %249, %172#0) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %315 = "vector.extract"(%186) {position = [1]} : (vector<4xf32>) -> f32
        %316 = "vector.splat"(%315) : (f32) -> vector<4xf32>
        %317 = "vector.fma"(%316, %251, %314) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %318 = "vector.extract"(%186) {position = [2]} : (vector<4xf32>) -> f32
        %319 = "vector.splat"(%318) : (f32) -> vector<4xf32>
        %320 = "vector.fma"(%319, %253, %317) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %321 = "vector.extract"(%186) {position = [3]} : (vector<4xf32>) -> f32
        %322 = "vector.splat"(%321) : (f32) -> vector<4xf32>
        %323 = "vector.fma"(%322, %255, %320) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %324 = "vector.extract"(%188) {position = [0]} : (vector<4xf32>) -> f32
        %325 = "vector.splat"(%324) : (f32) -> vector<4xf32>
        %326 = "vector.fma"(%325, %257, %323) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %327 = "vector.extract"(%188) {position = [1]} : (vector<4xf32>) -> f32
        %328 = "vector.splat"(%327) : (f32) -> vector<4xf32>
        %329 = "vector.fma"(%328, %259, %326) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %330 = "vector.extract"(%188) {position = [2]} : (vector<4xf32>) -> f32
        %331 = "vector.splat"(%330) : (f32) -> vector<4xf32>
        %332 = "vector.fma"(%331, %261, %329) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %333 = "vector.extract"(%188) {position = [3]} : (vector<4xf32>) -> f32
        %334 = "vector.splat"(%333) : (f32) -> vector<4xf32>
        %335 = "vector.fma"(%334, %263, %332) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %336 = "vector.extract"(%190) {position = [0]} : (vector<4xf32>) -> f32
        %337 = "vector.splat"(%336) : (f32) -> vector<4xf32>
        %338 = "vector.fma"(%337, %265, %335) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %339 = "vector.extract"(%190) {position = [1]} : (vector<4xf32>) -> f32
        %340 = "vector.splat"(%339) : (f32) -> vector<4xf32>
        %341 = "vector.fma"(%340, %267, %338) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %342 = "vector.extract"(%190) {position = [2]} : (vector<4xf32>) -> f32
        %343 = "vector.splat"(%342) : (f32) -> vector<4xf32>
        %344 = "vector.fma"(%343, %269, %341) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %345 = "vector.extract"(%190) {position = [3]} : (vector<4xf32>) -> f32
        %346 = "vector.splat"(%345) : (f32) -> vector<4xf32>
        %347 = "vector.fma"(%346, %271, %344) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %348 = "vector.extract"(%192) {position = [0]} : (vector<4xf32>) -> f32
        %349 = "vector.splat"(%348) : (f32) -> vector<4xf32>
        %350 = "vector.fma"(%349, %273, %347) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %351 = "vector.extract"(%192) {position = [1]} : (vector<4xf32>) -> f32
        %352 = "vector.splat"(%351) : (f32) -> vector<4xf32>
        %353 = "vector.fma"(%352, %275, %350) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %354 = "vector.extract"(%192) {position = [2]} : (vector<4xf32>) -> f32
        %355 = "vector.splat"(%354) : (f32) -> vector<4xf32>
        %356 = "vector.fma"(%355, %277, %353) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %357 = "vector.extract"(%192) {position = [3]} : (vector<4xf32>) -> f32
        %358 = "vector.splat"(%357) : (f32) -> vector<4xf32>
        %359 = "vector.fma"(%358, %279, %356) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %360 = "vector.extract"(%194) {position = [0]} : (vector<4xf32>) -> f32
        %361 = "vector.splat"(%360) : (f32) -> vector<4xf32>
        %362 = "vector.fma"(%361, %281, %359) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %363 = "vector.extract"(%194) {position = [1]} : (vector<4xf32>) -> f32
        %364 = "vector.splat"(%363) : (f32) -> vector<4xf32>
        %365 = "vector.fma"(%364, %283, %362) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %366 = "vector.extract"(%194) {position = [2]} : (vector<4xf32>) -> f32
        %367 = "vector.splat"(%366) : (f32) -> vector<4xf32>
        %368 = "vector.fma"(%367, %285, %365) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %369 = "vector.extract"(%194) {position = [3]} : (vector<4xf32>) -> f32
        %370 = "vector.splat"(%369) : (f32) -> vector<4xf32>
        %371 = "vector.fma"(%370, %287, %368) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %372 = "vector.extract"(%196) {position = [0]} : (vector<4xf32>) -> f32
        %373 = "vector.splat"(%372) : (f32) -> vector<4xf32>
        %374 = "vector.fma"(%373, %289, %371) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %375 = "vector.extract"(%196) {position = [1]} : (vector<4xf32>) -> f32
        %376 = "vector.splat"(%375) : (f32) -> vector<4xf32>
        %377 = "vector.fma"(%376, %291, %374) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %378 = "vector.extract"(%196) {position = [2]} : (vector<4xf32>) -> f32
        %379 = "vector.splat"(%378) : (f32) -> vector<4xf32>
        %380 = "vector.fma"(%379, %293, %377) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %381 = "vector.extract"(%196) {position = [3]} : (vector<4xf32>) -> f32
        %382 = "vector.splat"(%381) : (f32) -> vector<4xf32>
        %383 = "vector.fma"(%382, %295, %380) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %384 = "vector.extract"(%198) {position = [0]} : (vector<4xf32>) -> f32
        %385 = "vector.splat"(%384) : (f32) -> vector<4xf32>
        %386 = "vector.fma"(%385, %297, %383) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %387 = "vector.extract"(%198) {position = [1]} : (vector<4xf32>) -> f32
        %388 = "vector.splat"(%387) : (f32) -> vector<4xf32>
        %389 = "vector.fma"(%388, %299, %386) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %390 = "vector.extract"(%198) {position = [2]} : (vector<4xf32>) -> f32
        %391 = "vector.splat"(%390) : (f32) -> vector<4xf32>
        %392 = "vector.fma"(%391, %301, %389) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %393 = "vector.extract"(%198) {position = [3]} : (vector<4xf32>) -> f32
        %394 = "vector.splat"(%393) : (f32) -> vector<4xf32>
        %395 = "vector.fma"(%394, %303, %392) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %396 = "vector.extract"(%200) {position = [0]} : (vector<4xf32>) -> f32
        %397 = "vector.splat"(%396) : (f32) -> vector<4xf32>
        %398 = "vector.fma"(%397, %305, %395) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %399 = "vector.extract"(%200) {position = [1]} : (vector<4xf32>) -> f32
        %400 = "vector.splat"(%399) : (f32) -> vector<4xf32>
        %401 = "vector.fma"(%400, %307, %398) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %402 = "vector.extract"(%200) {position = [2]} : (vector<4xf32>) -> f32
        %403 = "vector.splat"(%402) : (f32) -> vector<4xf32>
        %404 = "vector.fma"(%403, %309, %401) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %405 = "vector.extract"(%200) {position = [3]} : (vector<4xf32>) -> f32
        %406 = "vector.splat"(%405) : (f32) -> vector<4xf32>
        %407 = "vector.fma"(%406, %311, %404) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %408 = "vector.extract"(%202) {position = [0]} : (vector<4xf32>) -> f32
        %409 = "vector.splat"(%408) : (f32) -> vector<4xf32>
        %410 = "vector.fma"(%409, %249, %172#1) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %411 = "vector.extract"(%202) {position = [1]} : (vector<4xf32>) -> f32
        %412 = "vector.splat"(%411) : (f32) -> vector<4xf32>
        %413 = "vector.fma"(%412, %251, %410) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %414 = "vector.extract"(%202) {position = [2]} : (vector<4xf32>) -> f32
        %415 = "vector.splat"(%414) : (f32) -> vector<4xf32>
        %416 = "vector.fma"(%415, %253, %413) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %417 = "vector.extract"(%202) {position = [3]} : (vector<4xf32>) -> f32
        %418 = "vector.splat"(%417) : (f32) -> vector<4xf32>
        %419 = "vector.fma"(%418, %255, %416) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %420 = "vector.extract"(%204) {position = [0]} : (vector<4xf32>) -> f32
        %421 = "vector.splat"(%420) : (f32) -> vector<4xf32>
        %422 = "vector.fma"(%421, %257, %419) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %423 = "vector.extract"(%204) {position = [1]} : (vector<4xf32>) -> f32
        %424 = "vector.splat"(%423) : (f32) -> vector<4xf32>
        %425 = "vector.fma"(%424, %259, %422) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %426 = "vector.extract"(%204) {position = [2]} : (vector<4xf32>) -> f32
        %427 = "vector.splat"(%426) : (f32) -> vector<4xf32>
        %428 = "vector.fma"(%427, %261, %425) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %429 = "vector.extract"(%204) {position = [3]} : (vector<4xf32>) -> f32
        %430 = "vector.splat"(%429) : (f32) -> vector<4xf32>
        %431 = "vector.fma"(%430, %263, %428) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %432 = "vector.extract"(%206) {position = [0]} : (vector<4xf32>) -> f32
        %433 = "vector.splat"(%432) : (f32) -> vector<4xf32>
        %434 = "vector.fma"(%433, %265, %431) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %435 = "vector.extract"(%206) {position = [1]} : (vector<4xf32>) -> f32
        %436 = "vector.splat"(%435) : (f32) -> vector<4xf32>
        %437 = "vector.fma"(%436, %267, %434) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %438 = "vector.extract"(%206) {position = [2]} : (vector<4xf32>) -> f32
        %439 = "vector.splat"(%438) : (f32) -> vector<4xf32>
        %440 = "vector.fma"(%439, %269, %437) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %441 = "vector.extract"(%206) {position = [3]} : (vector<4xf32>) -> f32
        %442 = "vector.splat"(%441) : (f32) -> vector<4xf32>
        %443 = "vector.fma"(%442, %271, %440) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %444 = "vector.extract"(%208) {position = [0]} : (vector<4xf32>) -> f32
        %445 = "vector.splat"(%444) : (f32) -> vector<4xf32>
        %446 = "vector.fma"(%445, %273, %443) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %447 = "vector.extract"(%208) {position = [1]} : (vector<4xf32>) -> f32
        %448 = "vector.splat"(%447) : (f32) -> vector<4xf32>
        %449 = "vector.fma"(%448, %275, %446) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %450 = "vector.extract"(%208) {position = [2]} : (vector<4xf32>) -> f32
        %451 = "vector.splat"(%450) : (f32) -> vector<4xf32>
        %452 = "vector.fma"(%451, %277, %449) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %453 = "vector.extract"(%208) {position = [3]} : (vector<4xf32>) -> f32
        %454 = "vector.splat"(%453) : (f32) -> vector<4xf32>
        %455 = "vector.fma"(%454, %279, %452) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %456 = "vector.extract"(%210) {position = [0]} : (vector<4xf32>) -> f32
        %457 = "vector.splat"(%456) : (f32) -> vector<4xf32>
        %458 = "vector.fma"(%457, %281, %455) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %459 = "vector.extract"(%210) {position = [1]} : (vector<4xf32>) -> f32
        %460 = "vector.splat"(%459) : (f32) -> vector<4xf32>
        %461 = "vector.fma"(%460, %283, %458) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %462 = "vector.extract"(%210) {position = [2]} : (vector<4xf32>) -> f32
        %463 = "vector.splat"(%462) : (f32) -> vector<4xf32>
        %464 = "vector.fma"(%463, %285, %461) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %465 = "vector.extract"(%210) {position = [3]} : (vector<4xf32>) -> f32
        %466 = "vector.splat"(%465) : (f32) -> vector<4xf32>
        %467 = "vector.fma"(%466, %287, %464) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %468 = "vector.extract"(%212) {position = [0]} : (vector<4xf32>) -> f32
        %469 = "vector.splat"(%468) : (f32) -> vector<4xf32>
        %470 = "vector.fma"(%469, %289, %467) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %471 = "vector.extract"(%212) {position = [1]} : (vector<4xf32>) -> f32
        %472 = "vector.splat"(%471) : (f32) -> vector<4xf32>
        %473 = "vector.fma"(%472, %291, %470) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %474 = "vector.extract"(%212) {position = [2]} : (vector<4xf32>) -> f32
        %475 = "vector.splat"(%474) : (f32) -> vector<4xf32>
        %476 = "vector.fma"(%475, %293, %473) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %477 = "vector.extract"(%212) {position = [3]} : (vector<4xf32>) -> f32
        %478 = "vector.splat"(%477) : (f32) -> vector<4xf32>
        %479 = "vector.fma"(%478, %295, %476) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %480 = "vector.extract"(%214) {position = [0]} : (vector<4xf32>) -> f32
        %481 = "vector.splat"(%480) : (f32) -> vector<4xf32>
        %482 = "vector.fma"(%481, %297, %479) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %483 = "vector.extract"(%214) {position = [1]} : (vector<4xf32>) -> f32
        %484 = "vector.splat"(%483) : (f32) -> vector<4xf32>
        %485 = "vector.fma"(%484, %299, %482) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %486 = "vector.extract"(%214) {position = [2]} : (vector<4xf32>) -> f32
        %487 = "vector.splat"(%486) : (f32) -> vector<4xf32>
        %488 = "vector.fma"(%487, %301, %485) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %489 = "vector.extract"(%214) {position = [3]} : (vector<4xf32>) -> f32
        %490 = "vector.splat"(%489) : (f32) -> vector<4xf32>
        %491 = "vector.fma"(%490, %303, %488) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %492 = "vector.extract"(%216) {position = [0]} : (vector<4xf32>) -> f32
        %493 = "vector.splat"(%492) : (f32) -> vector<4xf32>
        %494 = "vector.fma"(%493, %305, %491) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %495 = "vector.extract"(%216) {position = [1]} : (vector<4xf32>) -> f32
        %496 = "vector.splat"(%495) : (f32) -> vector<4xf32>
        %497 = "vector.fma"(%496, %307, %494) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %498 = "vector.extract"(%216) {position = [2]} : (vector<4xf32>) -> f32
        %499 = "vector.splat"(%498) : (f32) -> vector<4xf32>
        %500 = "vector.fma"(%499, %309, %497) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %501 = "vector.extract"(%216) {position = [3]} : (vector<4xf32>) -> f32
        %502 = "vector.splat"(%501) : (f32) -> vector<4xf32>
        %503 = "vector.fma"(%502, %311, %500) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %504 = "vector.extract"(%218) {position = [0]} : (vector<4xf32>) -> f32
        %505 = "vector.splat"(%504) : (f32) -> vector<4xf32>
        %506 = "vector.fma"(%505, %249, %172#2) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %507 = "vector.extract"(%218) {position = [1]} : (vector<4xf32>) -> f32
        %508 = "vector.splat"(%507) : (f32) -> vector<4xf32>
        %509 = "vector.fma"(%508, %251, %506) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %510 = "vector.extract"(%218) {position = [2]} : (vector<4xf32>) -> f32
        %511 = "vector.splat"(%510) : (f32) -> vector<4xf32>
        %512 = "vector.fma"(%511, %253, %509) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %513 = "vector.extract"(%218) {position = [3]} : (vector<4xf32>) -> f32
        %514 = "vector.splat"(%513) : (f32) -> vector<4xf32>
        %515 = "vector.fma"(%514, %255, %512) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %516 = "vector.extract"(%220) {position = [0]} : (vector<4xf32>) -> f32
        %517 = "vector.splat"(%516) : (f32) -> vector<4xf32>
        %518 = "vector.fma"(%517, %257, %515) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %519 = "vector.extract"(%220) {position = [1]} : (vector<4xf32>) -> f32
        %520 = "vector.splat"(%519) : (f32) -> vector<4xf32>
        %521 = "vector.fma"(%520, %259, %518) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %522 = "vector.extract"(%220) {position = [2]} : (vector<4xf32>) -> f32
        %523 = "vector.splat"(%522) : (f32) -> vector<4xf32>
        %524 = "vector.fma"(%523, %261, %521) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %525 = "vector.extract"(%220) {position = [3]} : (vector<4xf32>) -> f32
        %526 = "vector.splat"(%525) : (f32) -> vector<4xf32>
        %527 = "vector.fma"(%526, %263, %524) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %528 = "vector.extract"(%222) {position = [0]} : (vector<4xf32>) -> f32
        %529 = "vector.splat"(%528) : (f32) -> vector<4xf32>
        %530 = "vector.fma"(%529, %265, %527) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %531 = "vector.extract"(%222) {position = [1]} : (vector<4xf32>) -> f32
        %532 = "vector.splat"(%531) : (f32) -> vector<4xf32>
        %533 = "vector.fma"(%532, %267, %530) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %534 = "vector.extract"(%222) {position = [2]} : (vector<4xf32>) -> f32
        %535 = "vector.splat"(%534) : (f32) -> vector<4xf32>
        %536 = "vector.fma"(%535, %269, %533) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %537 = "vector.extract"(%222) {position = [3]} : (vector<4xf32>) -> f32
        %538 = "vector.splat"(%537) : (f32) -> vector<4xf32>
        %539 = "vector.fma"(%538, %271, %536) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %540 = "vector.extract"(%224) {position = [0]} : (vector<4xf32>) -> f32
        %541 = "vector.splat"(%540) : (f32) -> vector<4xf32>
        %542 = "vector.fma"(%541, %273, %539) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %543 = "vector.extract"(%224) {position = [1]} : (vector<4xf32>) -> f32
        %544 = "vector.splat"(%543) : (f32) -> vector<4xf32>
        %545 = "vector.fma"(%544, %275, %542) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %546 = "vector.extract"(%224) {position = [2]} : (vector<4xf32>) -> f32
        %547 = "vector.splat"(%546) : (f32) -> vector<4xf32>
        %548 = "vector.fma"(%547, %277, %545) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %549 = "vector.extract"(%224) {position = [3]} : (vector<4xf32>) -> f32
        %550 = "vector.splat"(%549) : (f32) -> vector<4xf32>
        %551 = "vector.fma"(%550, %279, %548) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %552 = "vector.extract"(%226) {position = [0]} : (vector<4xf32>) -> f32
        %553 = "vector.splat"(%552) : (f32) -> vector<4xf32>
        %554 = "vector.fma"(%553, %281, %551) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %555 = "vector.extract"(%226) {position = [1]} : (vector<4xf32>) -> f32
        %556 = "vector.splat"(%555) : (f32) -> vector<4xf32>
        %557 = "vector.fma"(%556, %283, %554) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %558 = "vector.extract"(%226) {position = [2]} : (vector<4xf32>) -> f32
        %559 = "vector.splat"(%558) : (f32) -> vector<4xf32>
        %560 = "vector.fma"(%559, %285, %557) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %561 = "vector.extract"(%226) {position = [3]} : (vector<4xf32>) -> f32
        %562 = "vector.splat"(%561) : (f32) -> vector<4xf32>
        %563 = "vector.fma"(%562, %287, %560) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %564 = "vector.extract"(%228) {position = [0]} : (vector<4xf32>) -> f32
        %565 = "vector.splat"(%564) : (f32) -> vector<4xf32>
        %566 = "vector.fma"(%565, %289, %563) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %567 = "vector.extract"(%228) {position = [1]} : (vector<4xf32>) -> f32
        %568 = "vector.splat"(%567) : (f32) -> vector<4xf32>
        %569 = "vector.fma"(%568, %291, %566) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %570 = "vector.extract"(%228) {position = [2]} : (vector<4xf32>) -> f32
        %571 = "vector.splat"(%570) : (f32) -> vector<4xf32>
        %572 = "vector.fma"(%571, %293, %569) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %573 = "vector.extract"(%228) {position = [3]} : (vector<4xf32>) -> f32
        %574 = "vector.splat"(%573) : (f32) -> vector<4xf32>
        %575 = "vector.fma"(%574, %295, %572) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %576 = "vector.extract"(%230) {position = [0]} : (vector<4xf32>) -> f32
        %577 = "vector.splat"(%576) : (f32) -> vector<4xf32>
        %578 = "vector.fma"(%577, %297, %575) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %579 = "vector.extract"(%230) {position = [1]} : (vector<4xf32>) -> f32
        %580 = "vector.splat"(%579) : (f32) -> vector<4xf32>
        %581 = "vector.fma"(%580, %299, %578) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %582 = "vector.extract"(%230) {position = [2]} : (vector<4xf32>) -> f32
        %583 = "vector.splat"(%582) : (f32) -> vector<4xf32>
        %584 = "vector.fma"(%583, %301, %581) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %585 = "vector.extract"(%230) {position = [3]} : (vector<4xf32>) -> f32
        %586 = "vector.splat"(%585) : (f32) -> vector<4xf32>
        %587 = "vector.fma"(%586, %303, %584) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %588 = "vector.extract"(%232) {position = [0]} : (vector<4xf32>) -> f32
        %589 = "vector.splat"(%588) : (f32) -> vector<4xf32>
        %590 = "vector.fma"(%589, %305, %587) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %591 = "vector.extract"(%232) {position = [1]} : (vector<4xf32>) -> f32
        %592 = "vector.splat"(%591) : (f32) -> vector<4xf32>
        %593 = "vector.fma"(%592, %307, %590) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %594 = "vector.extract"(%232) {position = [2]} : (vector<4xf32>) -> f32
        %595 = "vector.splat"(%594) : (f32) -> vector<4xf32>
        %596 = "vector.fma"(%595, %309, %593) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %597 = "vector.extract"(%232) {position = [3]} : (vector<4xf32>) -> f32
        %598 = "vector.splat"(%597) : (f32) -> vector<4xf32>
        %599 = "vector.fma"(%598, %311, %596) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %600 = "vector.extract"(%234) {position = [0]} : (vector<4xf32>) -> f32
        %601 = "vector.splat"(%600) : (f32) -> vector<4xf32>
        %602 = "vector.fma"(%601, %249, %172#3) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %603 = "vector.extract"(%234) {position = [1]} : (vector<4xf32>) -> f32
        %604 = "vector.splat"(%603) : (f32) -> vector<4xf32>
        %605 = "vector.fma"(%604, %251, %602) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %606 = "vector.extract"(%234) {position = [2]} : (vector<4xf32>) -> f32
        %607 = "vector.splat"(%606) : (f32) -> vector<4xf32>
        %608 = "vector.fma"(%607, %253, %605) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %609 = "vector.extract"(%234) {position = [3]} : (vector<4xf32>) -> f32
        %610 = "vector.splat"(%609) : (f32) -> vector<4xf32>
        %611 = "vector.fma"(%610, %255, %608) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %612 = "vector.extract"(%236) {position = [0]} : (vector<4xf32>) -> f32
        %613 = "vector.splat"(%612) : (f32) -> vector<4xf32>
        %614 = "vector.fma"(%613, %257, %611) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %615 = "vector.extract"(%236) {position = [1]} : (vector<4xf32>) -> f32
        %616 = "vector.splat"(%615) : (f32) -> vector<4xf32>
        %617 = "vector.fma"(%616, %259, %614) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %618 = "vector.extract"(%236) {position = [2]} : (vector<4xf32>) -> f32
        %619 = "vector.splat"(%618) : (f32) -> vector<4xf32>
        %620 = "vector.fma"(%619, %261, %617) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %621 = "vector.extract"(%236) {position = [3]} : (vector<4xf32>) -> f32
        %622 = "vector.splat"(%621) : (f32) -> vector<4xf32>
        %623 = "vector.fma"(%622, %263, %620) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %624 = "vector.extract"(%238) {position = [0]} : (vector<4xf32>) -> f32
        %625 = "vector.splat"(%624) : (f32) -> vector<4xf32>
        %626 = "vector.fma"(%625, %265, %623) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %627 = "vector.extract"(%238) {position = [1]} : (vector<4xf32>) -> f32
        %628 = "vector.splat"(%627) : (f32) -> vector<4xf32>
        %629 = "vector.fma"(%628, %267, %626) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %630 = "vector.extract"(%238) {position = [2]} : (vector<4xf32>) -> f32
        %631 = "vector.splat"(%630) : (f32) -> vector<4xf32>
        %632 = "vector.fma"(%631, %269, %629) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %633 = "vector.extract"(%238) {position = [3]} : (vector<4xf32>) -> f32
        %634 = "vector.splat"(%633) : (f32) -> vector<4xf32>
        %635 = "vector.fma"(%634, %271, %632) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %636 = "vector.extract"(%240) {position = [0]} : (vector<4xf32>) -> f32
        %637 = "vector.splat"(%636) : (f32) -> vector<4xf32>
        %638 = "vector.fma"(%637, %273, %635) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %639 = "vector.extract"(%240) {position = [1]} : (vector<4xf32>) -> f32
        %640 = "vector.splat"(%639) : (f32) -> vector<4xf32>
        %641 = "vector.fma"(%640, %275, %638) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %642 = "vector.extract"(%240) {position = [2]} : (vector<4xf32>) -> f32
        %643 = "vector.splat"(%642) : (f32) -> vector<4xf32>
        %644 = "vector.fma"(%643, %277, %641) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %645 = "vector.extract"(%240) {position = [3]} : (vector<4xf32>) -> f32
        %646 = "vector.splat"(%645) : (f32) -> vector<4xf32>
        %647 = "vector.fma"(%646, %279, %644) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %648 = "vector.extract"(%242) {position = [0]} : (vector<4xf32>) -> f32
        %649 = "vector.splat"(%648) : (f32) -> vector<4xf32>
        %650 = "vector.fma"(%649, %281, %647) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %651 = "vector.extract"(%242) {position = [1]} : (vector<4xf32>) -> f32
        %652 = "vector.splat"(%651) : (f32) -> vector<4xf32>
        %653 = "vector.fma"(%652, %283, %650) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %654 = "vector.extract"(%242) {position = [2]} : (vector<4xf32>) -> f32
        %655 = "vector.splat"(%654) : (f32) -> vector<4xf32>
        %656 = "vector.fma"(%655, %285, %653) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %657 = "vector.extract"(%242) {position = [3]} : (vector<4xf32>) -> f32
        %658 = "vector.splat"(%657) : (f32) -> vector<4xf32>
        %659 = "vector.fma"(%658, %287, %656) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %660 = "vector.extract"(%244) {position = [0]} : (vector<4xf32>) -> f32
        %661 = "vector.splat"(%660) : (f32) -> vector<4xf32>
        %662 = "vector.fma"(%661, %289, %659) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %663 = "vector.extract"(%244) {position = [1]} : (vector<4xf32>) -> f32
        %664 = "vector.splat"(%663) : (f32) -> vector<4xf32>
        %665 = "vector.fma"(%664, %291, %662) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %666 = "vector.extract"(%244) {position = [2]} : (vector<4xf32>) -> f32
        %667 = "vector.splat"(%666) : (f32) -> vector<4xf32>
        %668 = "vector.fma"(%667, %293, %665) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %669 = "vector.extract"(%244) {position = [3]} : (vector<4xf32>) -> f32
        %670 = "vector.splat"(%669) : (f32) -> vector<4xf32>
        %671 = "vector.fma"(%670, %295, %668) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %672 = "vector.extract"(%246) {position = [0]} : (vector<4xf32>) -> f32
        %673 = "vector.splat"(%672) : (f32) -> vector<4xf32>
        %674 = "vector.fma"(%673, %297, %671) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %675 = "vector.extract"(%246) {position = [1]} : (vector<4xf32>) -> f32
        %676 = "vector.splat"(%675) : (f32) -> vector<4xf32>
        %677 = "vector.fma"(%676, %299, %674) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %678 = "vector.extract"(%246) {position = [2]} : (vector<4xf32>) -> f32
        %679 = "vector.splat"(%678) : (f32) -> vector<4xf32>
        %680 = "vector.fma"(%679, %301, %677) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %681 = "vector.extract"(%246) {position = [3]} : (vector<4xf32>) -> f32
        %682 = "vector.splat"(%681) : (f32) -> vector<4xf32>
        %683 = "vector.fma"(%682, %303, %680) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %684 = "vector.extract"(%248) {position = [0]} : (vector<4xf32>) -> f32
        %685 = "vector.splat"(%684) : (f32) -> vector<4xf32>
        %686 = "vector.fma"(%685, %305, %683) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %687 = "vector.extract"(%248) {position = [1]} : (vector<4xf32>) -> f32
        %688 = "vector.splat"(%687) : (f32) -> vector<4xf32>
        %689 = "vector.fma"(%688, %307, %686) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %690 = "vector.extract"(%248) {position = [2]} : (vector<4xf32>) -> f32
        %691 = "vector.splat"(%690) : (f32) -> vector<4xf32>
        %692 = "vector.fma"(%691, %309, %689) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        %693 = "vector.extract"(%248) {position = [3]} : (vector<4xf32>) -> f32
        %694 = "vector.splat"(%693) : (f32) -> vector<4xf32>
        %695 = "vector.fma"(%694, %311, %692) : (vector<4xf32>, vector<4xf32>, vector<4xf32>) -> vector<4xf32>
        "memref.store"(%695, %109, %130) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
        "memref.store"(%599, %110, %129) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
        "memref.store"(%503, %111, %128) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
        "memref.store"(%407, %112, %127) : (vector<4xf32>, memref<?xvector<4xf32>, #spirv.storage_class<StorageBuffer>>, index) -> ()
        "func.return"() : () -> ()
      }) {function_type = () -> (), spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [16, 16, 1]>, sym_name = "forward_dispatch_35_matmul_18432x320x320"} : () -> ()
    }) {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>} : () -> ()
    "hal.executable.variant_end"() : () -> ()
  }) {sym_name = "vulkan_spirv_fb", target = #hal.executable.target<"vulkan", "vulkan-spirv-fb", {spirv.target_env = #spirv.target_env<#spirv.vce<v1.6, [Shader, Float64, Float16, Int64, Int16, Int8, StorageBuffer16BitAccess, StorageUniform16, StoragePushConstant16, StorageBuffer8BitAccess, UniformAndStorageBuffer8BitAccess, StoragePushConstant8, GroupNonUniform, GroupNonUniformVote, GroupNonUniformArithmetic, GroupNonUniformBallot, GroupNonUniformShuffle, GroupNonUniformShuffleRelative, GroupNonUniformClustered, GroupNonUniformQuad, VariablePointers, VariablePointersStorageBuffer, CooperativeMatrixNV], [SPV_KHR_16bit_storage, SPV_KHR_8bit_storage, SPV_KHR_storage_buffer_storage_class, SPV_KHR_variable_pointers, SPV_NV_cooperative_matrix]>, api=Vulkan, NVIDIA:DiscreteGPU, #spirv.resource_limits<max_compute_shared_memory_size = 49152, max_compute_workgroup_invocations = 1024, max_compute_workgroup_size = [1024, 1024, 64], min_subgroup_size = 32, max_subgroup_size = 32, cooperative_matrix_properties_nv = [#spirv.coop_matrix_props<m_size = 8, n_size = 8, k_size = 32, a_type = i8, b_type = i8, c_type = i32, result_type = i32, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f16, result_type = f16, scope = <Subgroup>>, #spirv.coop_matrix_props<m_size = 16, n_size = 16, k_size = 16, a_type = f16, b_type = f16, c_type = f32, result_type = f32, scope = <Subgroup>>]>>}>} : () -> ()
  "hal.executable_end"() : () -> ()
}) {sym_name = "forward_dispatch_35", sym_visibility = "private"} : () -> ()
    %133 = linalg.matmul ins(%collapsed_749, %130 : tensor<18432x320xf32>, tensor<320x320xf32>) outs(%132 : tensor<18432x320xf32>) -> tensor<18432x320xf32>
           ^