[PATCH] D32416: [x86, SSE] AVX1 PR28129

[Dinar Temirbulatov via Phabricator via llvm-commits]

dtemirbulatov created this revision.

Further perf tests on Jaguar indicate that:

  vxorps  %ymm0, %ymm0, %ymm0
  vcmpps  $15, %ymm0, %ymm0, %ymm0

is consistently faster (by about 9%) than:

  vpcmpeqd  %xmm0, %xmm0, %xmm0
  vinsertf128  $1, %xmm0, %ymm0, %ymm0

Testing equivalent code on a SandyBridge (E5-2640) puts it slightly (~3%) faster as well.
OK to commit, this change is only related to AVX1?


https://reviews.llvm.org/D32416

Files:
  lib/Target/X86/X86InstrSSE.td
  test/CodeGen/X86/all-ones-vector.ll
  test/CodeGen/X86/avx-basic.ll
  test/CodeGen/X86/avx-cvt-3.ll
  test/CodeGen/X86/avx-intrinsics-fast-isel.ll
  test/CodeGen/X86/pr28129.ll
  test/CodeGen/X86/vector-pcmp.ll
  test/CodeGen/X86/xop-intrinsics-fast-isel.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D32416.96347.patch
Type: text/x-patch
Size: 18313 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170424/47d89bd3/attachment.bin>