Reconcile PyTorch pass pipelines with those upstream in torch-mlir (if needed)

Open ScottTodd opened this issue 1 year ago • 2 comments

We have some people running "e2e" test workflows in torch-mlir that end up differing when running through IREE. I suspect the pipelines may have drifted a bit.

linalg:

IREE: https://github.com/openxla/iree/blob/main/compiler/plugins/input/Torch/InputConversion/Passes.cpp
torch-mlir: https://github.com/llvm/torch-mlir/blob/main/lib/Dialect/Torch/Transforms/Passes.cpp (maybe other files too?)

onnx:

IREE: https://github.com/openxla/iree/blob/main/compiler/plugins/input/Torch/PluginRegistration.cpp

Apr 11 '24 19:04 ScottTodd

Related: https://github.com/nod-ai/SHARK-TestSuite/issues/47

May 04 '24 01:05 ScottTodd

Related https://github.com/nod-ai/SHARK-TestSuite/issues/225

May 08 '24 19:05 AmosLewis