[PATCH] D116737: [DAG][AArch64][SVE] Fix VLS mulh code generation
Paul Walker via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Jan 10 04:17:55 PST 2022
paulwalker-arm added a comment.
Yes, that was what I was thinking, because it seemed odd that such a constant representation exists by this point in the pipeline. However, I can see that it's coming from the test itself and so I've changed my mind and am now wondering if there actually a problem that needs solving here?
The tests are using a non-standard immediate representation (aka "fixed vector shuffle of an inserted constant") that LLVM would have converted to a constant vector (aka `<i16 8, i16 8,....`) long before it get's to the code generator. This is backed up by running the test through clang, which results in `smul` when expected.
This makes me think that to protect the `MULS/MULU` functionality, you just need to update the test itself to remove the non-standard representation.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D116737/new/
https://reviews.llvm.org/D116737
More information about the llvm-commits
mailing list