We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
env -> OS: ubuntu 22.04
onnx model: git clone https://github.com/onnx/onnx.git
in onnx project, the model path is onnx/onnx/backend/test/data/node/test_roialign_mode_max/model.onnx
torch-mlir version: newest in main branch
run command: python -m torch_mlir.tools.import_onnx ./model.onnx -o onnx.mlir torch-mlir-opt onnx.mlir --mlir-print-debuginfo --torch-onnx-to-torch-backend-pipeline -o torch.mlir torch-mlir-opt torch.mlir --mlir-print-debuginfo --torch-backend-to-linalg-on-tensors-backend-pipeline -o linalg.mlir
onnx ir: module { func.func @test_roialign_aligned_false(%arg0: !torch.vtensor<[1,1,10,10],f32>, %arg1: !torch.vtensor<[3,4],f32>, %arg2: !torch.vtensor<[3],si64>) -> !torch.vtensor<[3,1,5,5],f32> attributes {torch.onnx_meta.ir_version = 10 : si64, torch.onnx_meta.opset_version = 22 : si64, torch.onnx_meta.producer_name = "backend-test", torch.onnx_meta.producer_version = ""} { %none = torch.constant.none %0 = torch.operator "onnx.RoiAlign"(%arg0, %arg1, %arg2) {torch.onnx.coordinate_transformation_mode = "output_half_pixel", torch.onnx.output_height = 5 : si64, torch.onnx.output_width = 5 : si64, torch.onnx.sampling_ratio = 2 : si64, torch.onnx.spatial_scale = 1.000000e+00 : f32} : (!torch.vtensor<[1,1,10,10],f32>, !torch.vtensor<[3,4],f32>, !torch.vtensor<[3],si64>) -> !torch.vtensor<[3,1,5,5],f32> return %0 : !torch.vtensor<[3,1,5,5],f32> } }
torch ir: module { func.func @test_roialign_aligned_false(%arg0: !torch.vtensor<[1,1,10,10],f32>, %arg1: !torch.vtensor<[3,4],f32>, %arg2: !torch.vtensor<[3],si64>) -> !torch.vtensor<[3,1,5,5],f32> attributes {torch.onnx_meta.ir_version = 10 : si64, torch.onnx_meta.opset_version = 22 : si64, torch.onnx_meta.producer_name = "backend-test", torch.onnx_meta.producer_version = ""} { %int2 = torch.constant.int 2 %int5 = torch.constant.int 5 %float1.000000e00 = torch.constant.float 1.000000e+00 %false = torch.constant.bool false %int6 = torch.constant.int 6 %none = torch.constant.none %int1 = torch.constant.int 1 %0 = torch.aten.unsqueeze %arg2, %int1 : !torch.vtensor<[3],si64>, !torch.int -> !torch.vtensor<[3,1],si64> %1 = torch.aten.to.dtype %0, %int6, %false, %false, %none : !torch.vtensor<[3,1],si64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[3,1],f32> %2 = torch.prim.ListConstruct %1, %arg1 : (!torch.vtensor<[3,1],f32>, !torch.vtensor<[3,4],f32>) -> !torch.list %3 = torch.aten.cat %2, %int1 : !torch.list, !torch.int -> !torch.vtensor<[3,5],f32> %4 = torch.torchvision.roi_align %arg0, %3, %float1.000000e00, %int5, %int5, %int2, %false : !torch.vtensor<[1,1,10,10],f32>, !torch.vtensor<[3,5],f32>, !torch.float, !torch.int, !torch.int, !torch.int, !torch.bool -> !torch.vtensor<[3,1,5,5],f32> return %4 : !torch.vtensor<[3,1,5,5],f32> } }
error: unknown :0: error: failed to legalize operation 'torch.constant.int' unknown :0: note: see current operation: %0 = "torch.constant.int"() <{value = 2 : i64}> : () -> !torch.int loc(unknown)
Is this an error or not supported? All ONNX models that include the RoiAlign operator produce errors when converting Torch IR to Linalg IR.
The text was updated successfully, but these errors were encountered:
No branches or pull requests
env ->
OS: ubuntu 22.04
onnx model:
git clone https://github.com/onnx/onnx.git
in onnx project, the model path is
onnx/onnx/backend/test/data/node/test_roialign_mode_max/model.onnx
torch-mlir version: newest in main branch
run command:
python -m torch_mlir.tools.import_onnx ./model.onnx -o onnx.mlir
torch-mlir-opt onnx.mlir --mlir-print-debuginfo --torch-onnx-to-torch-backend-pipeline -o torch.mlir
torch-mlir-opt torch.mlir --mlir-print-debuginfo --torch-backend-to-linalg-on-tensors-backend-pipeline -o linalg.mlir
onnx ir:
module {
func.func @test_roialign_aligned_false(%arg0: !torch.vtensor<[1,1,10,10],f32>, %arg1: !torch.vtensor<[3,4],f32>, %arg2: !torch.vtensor<[3],si64>) -> !torch.vtensor<[3,1,5,5],f32> attributes {torch.onnx_meta.ir_version = 10 : si64, torch.onnx_meta.opset_version = 22 : si64, torch.onnx_meta.producer_name = "backend-test", torch.onnx_meta.producer_version = ""} {
%none = torch.constant.none
%0 = torch.operator "onnx.RoiAlign"(%arg0, %arg1, %arg2) {torch.onnx.coordinate_transformation_mode = "output_half_pixel", torch.onnx.output_height = 5 : si64, torch.onnx.output_width = 5 : si64, torch.onnx.sampling_ratio = 2 : si64, torch.onnx.spatial_scale = 1.000000e+00 : f32} : (!torch.vtensor<[1,1,10,10],f32>, !torch.vtensor<[3,4],f32>, !torch.vtensor<[3],si64>) -> !torch.vtensor<[3,1,5,5],f32>
return %0 : !torch.vtensor<[3,1,5,5],f32>
}
}
torch ir:
module {
func.func @test_roialign_aligned_false(%arg0: !torch.vtensor<[1,1,10,10],f32>, %arg1: !torch.vtensor<[3,4],f32>, %arg2: !torch.vtensor<[3],si64>) -> !torch.vtensor<[3,1,5,5],f32> attributes {torch.onnx_meta.ir_version = 10 : si64, torch.onnx_meta.opset_version = 22 : si64, torch.onnx_meta.producer_name = "backend-test", torch.onnx_meta.producer_version = ""} {
%int2 = torch.constant.int 2
%int5 = torch.constant.int 5
%float1.000000e00 = torch.constant.float 1.000000e+00
%false = torch.constant.bool false
%int6 = torch.constant.int 6
%none = torch.constant.none
%int1 = torch.constant.int 1
%0 = torch.aten.unsqueeze %arg2, %int1 : !torch.vtensor<[3],si64>, !torch.int -> !torch.vtensor<[3,1],si64>
%1 = torch.aten.to.dtype %0, %int6, %false, %false, %none : !torch.vtensor<[3,1],si64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[3,1],f32>
%2 = torch.prim.ListConstruct %1, %arg1 : (!torch.vtensor<[3,1],f32>, !torch.vtensor<[3,4],f32>) -> !torch.list
%3 = torch.aten.cat %2, %int1 : !torch.list, !torch.int -> !torch.vtensor<[3,5],f32>
%4 = torch.torchvision.roi_align %arg0, %3, %float1.000000e00, %int5, %int5, %int2, %false : !torch.vtensor<[1,1,10,10],f32>, !torch.vtensor<[3,5],f32>, !torch.float, !torch.int, !torch.int, !torch.int, !torch.bool -> !torch.vtensor<[3,1,5,5],f32>
return %4 : !torch.vtensor<[3,1,5,5],f32>
}
}
error:
unknown :0: error: failed to legalize operation 'torch.constant.int'
unknown :0: note: see current operation: %0 = "torch.constant.int"() <{value = 2 : i64}> : () -> !torch.int loc(unknown)
Is this an error or not supported? All ONNX models that include the RoiAlign operator produce errors when converting Torch IR to Linalg IR.
The text was updated successfully, but these errors were encountered: