I have used func.func(affine-super-vectorize=virtual-vector-size=2) after https://github.com/llvm/torch-mlir/blob/main/python/torch_mlir_e2e_test/linalg_on_tensors_backends/refbackend.py#L159 .
It is giving some error. I have tried with Compiler Explorer , and it is giving the expected result.
So I don’t understand why is not working lowering pipeline pass?
The torch-mlir reference backend lowers directly to scf.for which are not targeted by vectorization.
The godbolt example lower to affine.for which is.
Understood, Thank you …