projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: cd1bc5c) | patch
[X86] Prefer AND over PSHUFB for v64i8 when possible
authorCraig Topper <craig.topper@intel.com>
Mon, 29 Jun 2020 23:26:53 +0000 (16:26 -0700)
committerCraig Topper <craig.topper@intel.com>
Mon, 29 Jun 2020 23:26:53 +0000 (16:26 -0700)
commit9b04d69ccefb57ec202a0ecdb36a84ed0568aead
tree780a48ca52a387c997df2e74d91372f236f7d68dtree | snapshot
parentcd1bc5c15d4e58b574060c844917c8dfeb7a8f54commit | diff
[X86] Prefer AND over PSHUFB for v64i8 when possible

If the shuffle is a blend and one input is a 0 vector, we should prefer AND over PSHUFB since its available on more execution ports.

Differential Revision: https://reviews.llvm.org/D82798
llvm/lib/Target/X86/X86ISelLowering.cpp diff | blob | history
llvm/test/CodeGen/X86/avx512-mask-op.ll diff | blob | history
llvm/test/CodeGen/X86/vector-shuffle-512-v64.ll diff | blob | history
Domain: System / Toolchain;