[PATCH] D41618: [x86] allow pairs of PCMPEQ for vector-sized integer equality comparisons (PR33325)

[Sanjay Patel via Phabricator via llvm-commits]

spatel added subscribers: craig.topper, hfinkel.
spatel added a comment.

cc @craig.topper and @hfinkel -- This might be an interesting case for the x86 preferred vector width effort (https://reviews.llvm.org/D41341) even for 128/256-bit vectors. Here, we're taking scalar code that is or would be produced by memcmp expansion and translating it to vectors based on the available ISA, but without accounting for the preferred vector width. I think we should add that predicate as a follow-up to prevent producing vector code if the user has requested we avoid it. There's no AVX512-specific codegen here or in memcmp expansion, so that's not a danger (yet).


https://reviews.llvm.org/D41618