search

nod-ai / iree-amd-aie

IREE plugin repository for the AMD AIE accelerator
Apache License 2.0
46 stars 23 forks source link

Add loop normalization pass and add to workgroup insertion #375

Closed jtuyls closed 1 month ago

MaheshRavishankar commented 1 month ago

@qedawkins was also looking for this. Maybe we can just move it into iree

Abhishek-Varma commented 1 month ago

@qedawkins was also looking for this. Maybe we can just move it into iree

Hi @MaheshRavishankar - I feel this is more suited to be upstreamed into LLVM instead of IREE.

qedawkins commented 1 month ago

I'd love to see this either upstream or in IREE!

MaheshRavishankar commented 1 month ago

@qedawkins was also looking for this. Maybe we can just move it into iree

Hi @MaheshRavishankar - I feel this is more suited to be upstreamed into LLVM instead of IREE.

Upstream would work too.. Should be as simple as just exposing this method https://github.com/llvm/llvm-project/blob/b17d44558ba4c30a3005089b334f68593d6a9c7c/mlir/lib/Dialect/SCF/Utils/Utils.cpp#L485

It unfortunately cant use the affine stuff, but does a reasonable job.

jtuyls commented 1 month ago

I created an upstream PR for a longer term solution: https://github.com/llvm/llvm-project/pull/93781, so let's move those discussions to that PR. But I would like to still merge this one as a temporary solution. Could you check again @Abhishek-Varma ?

cc @MaheshRavishankar