[CI][XFAIL] test_shape_end_negative_1 - onnx.Shape compilation failure

[PhaneeshB]

What happened?

For Torch-mlir bump in IREE #18169 Compilation failure for onnx.Shape operator for CPU, GPU AMD (rocm + vulkan) and GPU Nvidia (cuda + vulkan)

CPU Error Log:

_ IREE compile and run: test_shape_end_negative_1::model.mlir::model.mlir::cpu_llvm_sync _
[gw1] linux -- Python 3.11.9 /home/runner/work/iree/iree/venv/bin/python
Error invoking iree-compile
Error code: 1
Stderr diagnostics:
model.mlir:4:10: error: failed to materialize conversion for result #0 of operation 'torch.operator' that remained live after conversion
    %0 = torch.operator "onnx.Shape"(%arg0) {torch.onnx.end = -1 : si64} : (!torch.vtensor<[3,4,5],f32>) -> !torch.vtensor<[2],si64> 
         ^
model.mlir:4:10: note: see current operation: %2 = "torch.operator"(%arg0) <{name = "onnx.Shape"}> {torch.onnx.end = -1 : si64} : (!torch.vtensor<[3,4,5],f32>) -> !torch.vtensor<[2],si64>
model.mlir:5:5: note: see existing live user here: func.return %1 : !torch.vtensor<[2],si64>
    return %0 : !torch.vtensor<[2],si64>
    ^

Invoked with:
  cd /home/runner/work/iree/iree/SHARK-TestSuite/iree_tests/onnx/node/generated/test_shape_end_negative_1 && iree-compile model.mlir --iree-hal-target-backends=llvm-cpu --iree-input-demote-f64-to-f32=false -o model_cpu_llvm_sync_.vmfb

  module {
  func.func @test_shape_end_negative_1(%arg0: !torch.vtensor<[3,4,5],f32>) -> !torch.vtensor<[2],si64> attributes {torch.onnx_meta.ir_version = 10 : si64, torch.onnx_meta.opset_version = 21 : si64, torch.onnx_meta.producer_name = "backend-test", torch.onnx_meta.producer_version = ""} {
    %none = torch.constant.none
    %0 = torch.operator "onnx.Shape"(%arg0) {torch.onnx.end = -1 : si64} : (!torch.vtensor<[3,4,5],f32>) -> !torch.vtensor<[2],si64>
    return %0 : !torch.vtensor<[2],si64>
    }
}

Compile command :

 iree-compile model.mlir --iree-hal-target-backends=llvm-cpu --iree-input-demote-f64-to-f32=false -o model_cpu_llvm_sync_.vmfb

Steps to reproduce your issue

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. See error

What component(s) does this issue relate to?

No response

Version information

No response

Additional context

No response

[IanWood1]

~Looks like torch-mlir doesn't support onnx's start or end arguments for slicing the shape.~

Edit: I was looking at an outdated version

[ScottTodd]

This may be related to https://github.com/iree-org/iree/issues/16814

[IanWood1]

This may be related to #16814

Yeah, I think they are the same

The onnx documentation says end=-1 results in truncating the last dim, but the logic in torch-mlir treats end=-1 as the default.

https://github.com/llvm/torch-mlir/blob/026dfade6406035bca7481071e2263e849f83b09/lib/Conversion/TorchOnnxToTorch/DefaultDomainQtoZ.cpp#L1669-L1672