projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 57cef58) | patch
[X86] In CopyToFromAsymmetricReg, use VR128 instead of FR32 instructions for GR32...
authorCraig Topper <craig.topper@intel.com>
Wed, 17 Apr 2019 06:09:11 +0000 (06:09 +0000)
committerCraig Topper <craig.topper@intel.com>
Wed, 17 Apr 2019 06:09:11 +0000 (06:09 +0000)
commit6bf08027384200bf7ee45ccfd5b16fa7511566d0
tree899a2f7da102f87855ea6eaf426b58cc6ffeac97tree | snapshot
parent57cef586729526fc31b1ec184612c5573f94f1b3commit | diff
[X86] In CopyToFromAsymmetricReg, use VR128 instead of FR32 instructions for GR32<->XMM register copies.

We have two versions of some instructions, VR128 versions and FR32 versions that
are marked as CodeGenOnly.

This change switches to using the VR128 versions for these copies. It's after
register allocation so the class size no longer matters. This matches how GR64
works.

llvm-svn: 358555
llvm/lib/Target/X86/X86InstrInfo.cpp diff | blob | history
Domain: System / Toolchain;