[PATCH] D104127: [Matrix] In transpose opts, handle a^t * a^t
Florian Hahn via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Fri Jun 11 09:22:09 PDT 2021
fhahn accepted this revision.
fhahn added a comment.
This revision is now accepted and ready to land.
LGTM, thanks!
> Without the fix the testcase crashes.
Perhaps worth expanding the message saying that the test crashes because we remove the same instruction twice?
================
Comment at: llvm/test/Transforms/LowerMatrixIntrinsics/transpose-opts.ll:1087
+ %t = call <9 x float> @llvm.matrix.transpose.v9f32(<9 x float> %a, i32 3, i32 3)
+ %m = call <9 x float> @llvm.matrix.multiply.v9f32.v9f32.v9f32(<9 x float> %t, <9 x float> %t, i32 3, i32 3, i32 3)
+ ret <9 x float> %m
----------------
Can we use smaller dimensions to check for the issue? IIUC we mostly want to make sure to not remove the transpose twice, so something like 2x2 should also work with the benefit of having a few less check lines.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D104127/new/
https://reviews.llvm.org/D104127
More information about the llvm-commits
mailing list