Created
February 28, 2024 21:28
-
-
Save AmosLewis/8618bc5e191674c30d03c50c4a921839 to your computer and use it in GitHub Desktop.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
module { | |
func.func @test_reduce_prod_default_axes_keepdims_random(%arg0: !torch.vtensor<[3,2,2],f32>) -> !torch.vtensor<[1,1,1],f32> attributes {torch.onnx_meta.ir_version = 8 : si64, torch.onnx_meta.opset_version = 18 : si64, torch.onnx_meta.producer_name = "backend-test", torch.onnx_meta.producer_version = ""} { | |
%int0 = torch.constant.int 0 | |
%int0_0 = torch.constant.int 0 | |
%int1 = torch.constant.int 1 | |
%int2 = torch.constant.int 2 | |
%0 = torch.aten.dim %arg0 : !torch.vtensor<[3,2,2],f32> -> !torch.int | |
%1 = torch.aten.lt.int %int0_0, %int0 : !torch.int, !torch.int -> !torch.bool | |
%2 = torch.aten.Int.bool %1 : !torch.bool -> !torch.int | |
%3 = torch.aten.mul.int %2, %0 : !torch.int, !torch.int -> !torch.int | |
%4 = torch.aten.add.int %int0_0, %3 : !torch.int, !torch.int -> !torch.int | |
%5 = torch.aten.lt.int %int1, %int0 : !torch.int, !torch.int -> !torch.bool | |
%6 = torch.aten.Int.bool %5 : !torch.bool -> !torch.int | |
%7 = torch.aten.mul.int %6, %0 : !torch.int, !torch.int -> !torch.int | |
%8 = torch.aten.add.int %int1, %7 : !torch.int, !torch.int -> !torch.int | |
%9 = torch.aten.lt.int %int2, %int0 : !torch.int, !torch.int -> !torch.bool | |
%10 = torch.aten.Int.bool %9 : !torch.bool -> !torch.int | |
%11 = torch.aten.mul.int %10, %0 : !torch.int, !torch.int -> !torch.int | |
%12 = torch.aten.add.int %int2, %11 : !torch.int, !torch.int -> !torch.int | |
%true = torch.constant.bool true | |
%none = torch.constant.none | |
%13 = torch.aten.prod.dim_int %arg0, %4, %true, %none : !torch.vtensor<[3,2,2],f32>, !torch.int, !torch.bool, !torch.none -> !torch.vtensor<[?,?,?],f32> | |
%14 = torch.aten.prod.dim_int %13, %8, %true, %none : !torch.vtensor<[?,?,?],f32>, !torch.int, !torch.bool, !torch.none -> !torch.vtensor<[?,?,?],f32> | |
%15 = torch.aten.prod.dim_int %14, %12, %true, %none : !torch.vtensor<[?,?,?],f32>, !torch.int, !torch.bool, !torch.none -> !torch.vtensor<[?,?,?],f32> | |
%int-1 = torch.constant.int -1 | |
%int-1_1 = torch.constant.int -1 | |
%int-1_2 = torch.constant.int -1 | |
%16 = torch.prim.ListConstruct %int-1, %int-1_1, %int-1_2 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int> | |
%17 = torch.aten.reshape %15, %16 : !torch.vtensor<[?,?,?],f32>, !torch.list<int> -> !torch.vtensor<[1,1,1],f32> | |
return %17 : !torch.vtensor<[1,1,1],f32> | |
} | |
} |
torch-mlir-opt --convert-torch-onnx-to-torch --torch-decompose-complex-ops --cse --canonicalize ../onnx_torch_import_20231121/node_test_reduce_prod_default_axes_keepdims_random_model.mlir
module {
func.func @test_reduce_prod_default_axes_keepdims_random(%arg0: !torch.vtensor<[3,2,2],f32>) -> !torch.vtensor<[1,1,1],f32> attributes {torch.onnx_meta.ir_version = 8 : si64, torch.onnx_meta.opset_version = 18 : si64, torch.onnx_meta.producer_name = "backend-test", torch.onnx_meta.producer_version = ""} {
%int-1 = torch.constant.int -1
%none = torch.constant.none
%true = torch.constant.bool true
%int0 = torch.constant.int 0
%int1 = torch.constant.int 1
%int2 = torch.constant.int 2
%0 = torch.aten.prod.dim_int %arg0, %int0, %true, %none : !torch.vtensor<[3,2,2],f32>, !torch.int, !torch.bool, !torch.none -> !torch.vtensor<[?,?,?],f32>
%1 = torch.aten.prod.dim_int %0, %int1, %true, %none : !torch.vtensor<[?,?,?],f32>, !torch.int, !torch.bool, !torch.none -> !torch.vtensor<[?,?,?],f32>
%2 = torch.aten.prod.dim_int %1, %int2, %true, %none : !torch.vtensor<[?,?,?],f32>, !torch.int, !torch.bool, !torch.none -> !torch.vtensor<[?,?,?],f32>
%3 = torch.prim.ListConstruct %int-1, %int-1, %int-1 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4 = torch.aten.view %2, %3 : !torch.vtensor<[?,?,?],f32>, !torch.list<int> -> !torch.vtensor<[1,1,1],f32>
return %4 : !torch.vtensor<[1,1,1],f32>
}
}
torch-mlir-opt --convert-torch-to-linalg ../test/prod/node_test_reduce_prod_default_axes_keepdims_random_model_torch.mlir
torch-mlir-opt --convert-torch-to-linalg ../test/prod/node_test_reduce_prod_default_axes_keepdims_random_model_torch.mlir
../test/prod/node_test_reduce_prod_default_axes_keepdims_random_model_torch.mlir:13:10: error: failed to legalize operation 'torch.aten.view' that was explicitly marked illegal
%4 = torch.aten.view %2, %3 : !torch.vtensor<[?,?,?],f32>, !torch.list<int> -> !torch.vtensor<[1,1,1],f32>
^
../test/prod/node_test_reduce_prod_default_axes_keepdims_random_model_torch.mlir:13:10: note: see current operation: %51 = "torch.aten.view"(%49, %50) : (!torch.vtensor<[?,?,?],f32>, !torch.list<int>) -> !torch.vtensor<[1,1,1],f32>
torch-mlir-opt: /home/chi/src/torch-mlir/externals/llvm-project/mlir/include/mlir/IR/UseDefLists.h:198: mlir::IRObjectWithUseList<mlir::OpOperand>::~IRObjectWithUseList() [OperandType = mlir::OpOperand]: Assertion `use_empty() && "Cannot destroy a value that still has uses!"' failed.
* Pattern : 'torch.aten.view -> ()' {
Trying to match "(anonymous namespace)::ConvertAtenViewOp"
** Insert : 'arith.constant'(0x55e7d0b01740)
ImplicitTypeIDRegistry::lookupOrInsert(mlir::OpTrait::OneTypedResult<mlir::TensorType>::Impl<Empty>)
ImplicitTypeIDRegistry::lookupOrInsert(mlir::OpTrait::OneOperand<Empty>)
ImplicitTypeIDRegistry::lookupOrInsert(mlir::CastOpInterface::Trait<Empty>)
** Insert : 'tensor.dim'(0x55e7d0b00860)
** Insert : 'arith.constant'(0x55e7d0b017b0)
** Insert : 'tensor.dim'(0x55e7d0b00910)
** Insert : 'arith.constant'(0x55e7d0b00e30)
** Insert : 'tensor.dim'(0x55e7d0b00ea0)
** Failure : at most one element in size list is allowed to be -1
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
FIXED ISSUE.
If we
op.getDim().dump();
before https://github.com/llvm/torch-mlir/blob/e48fe4588631e7a37a2899f9d4cd5c4cbc967481/lib/Conversion/TorchToLinalg/Reduction.cpp#L467we get the print
%5 = torch.aten.add.int %int0_0, %4 : !torch.int, !torch.int -> !torch.int
which won't jump into the
if (matchPattern(op.getDim(), m_TorchConstantInt(&dim)))
since the dim type istorch.int
nottorch.constant.int
, so it return fail.FIXED by add
--torch-decompose-complex-ops --cse --canonicalize