Stella Laurenzo

As of https://github.com/llvm/torch-mlir/pull/3319, upstream now sets the strict attribute on functions produced via fx/dynamo, so we don't have to do it locally. The following can be removed (and all impl/support):...

Tracking bug to tie workarounds together for future work.

See Pass.cpp: ``` // Actually run the pass manager. PyMlirContext::ErrorCapture errors(op.getOperation().getContext()); MlirLogicalResult status = mlirPassManagerRunOnOp( passManager.get(), op.getOperation().get()); if (mlirLogicalResultIsFailure(status)) throw MLIRError("Failure while executing pass pipeline", errors.take()); ``` In a debug...

[onnx] Build real onnx frontend cli/API

1

The current ONNX compilation flow is serviceable but somewhat user hostile, directly requiring too much low-level manipulation of the system. Recommend creating a real ONNX frontend API and CL which...

Add a pass for simplifying numeric/index arithmetic.

2

The new `iree-util-simplify-index-arithmetic` pass uses the new range annotations to do range analysis and simplification. Currently it is looping to: * Rewrite redundant comparison ops to constant true/false. * Performing...

Compiling for llvm-cpu without targeting a specific CPU is a bad experience

33

I've seen multiple people falling down this hole: they run iree-compile on their model, targeting CPU. Then they get performance that is 10x-100x off of any reasonable expectation. Then they...

Skip clock reset if rocm-smi is not available.

1

* Pinning clocks was already conditioned on this, so the cleanup logic should also be. * Also removes a windows.h `#define` that was causing trouble.