[Objectfifo] Create a new pass to flatten vectorized ops

nod-ai / iree-amd-aie

IREE plugin repository for the AMD AIE accelerator

Apache License 2.0

69 stars 30 forks source link

[Objectfifo] Create a new pass to flatten vectorized ops #823

Closed Abhishek-Varma closed 1 month ago

Abhishek-Varma commented 1 month ago

-- This commit creates a new pass iree-amdaie-flatten-vectorized-ops which currently linearizes ONLY arith.truncf op. -- This is the first commit that needs to go in to enable "Matmul + truncf".

Signed-off-by: Abhishek Varma abhvarma@amd.com

yzhang93 commented 1 month ago

One minor title suggestion. It's not really meaningful to use tag [objectfifo], please change it to vectorization / matmul-elemenwise. Same for your other opening PRs.

newling commented 1 month ago

@Abhishek-Varma What does this PR fix? I'm wondering if this should be done in https://github.com/nod-ai/iree-amd-aie/blob/main/compiler/plugins/target/AMD-AIE/aievec/VectorToVectorConversions.cpp where there are patterns for flattening vectors