projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: d7ec4d0) | patch
[X86][AVX512] Use WriteVPMOV256 sched class for all truncations/extensions.
authorSimon Pilgrim <llvm-dev@redking.me.uk>
Sun, 12 Dec 2021 13:24:26 +0000 (13:24 +0000)
committerSimon Pilgrim <llvm-dev@redking.me.uk>
Sun, 12 Dec 2021 13:24:26 +0000 (13:24 +0000)
commit8e833d081b1541d2d644d93e482960c7a6d7b321
tree45c0d5384873de2c26cfd36092318044dd97a958tree | snapshot
parentd7ec4d0be34f4d99cedff1a06e12f0a664d039d9commit | diff
[X86][AVX512] Use WriteVPMOV256 sched class for all truncations/extensions.

At the moment these are identical to WriteShuffle256 (which we were using), but it should be WriteVPMOV256 to match the AVX2 instruction, plus it will help us remove some unnecessary overrides by tweaking the WriteVPMOV256 class.

Also, as D115547 shows, we still need to split off 128-bit extensions/truncations sched classes to remove some other overrides.
llvm/lib/Target/X86/X86InstrAVX512.td diff | blob | history
Domain: System / Toolchain;