[PATCH] D11370: Improved the interface of methods commuting operands, improved X86-FMA3 mem-folding&coalescing.

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Fri Sep 25 14:29:36 PDT 2015

arsenm accepted this revision.
arsenm added a reviewer: arsenm.
arsenm added a comment.


Comment at: llvm/lib/Target/AMDGPU/SIInstrInfo.cpp:779-780
@@ -769,4 +778,4 @@
   if (MI->getNumOperands() < 3)
     return nullptr;
I just removed this check a few days ago, so you probably will have a conflict when you apply this to trunk

Comment at: llvm/lib/Target/AMDGPU/SIInstrInfo.cpp:800-801
@@ -788,1 +799,4 @@
+        OpIdx1 == static_cast<unsigned>(Src1Idx)) &&
+      !(OpIdx0 == static_cast<unsigned>(Src1Idx) &&
+        OpIdx1 == static_cast<unsigned>(Src0Idx)))
     return nullptr;
These asserts can be removed. I decided that there's no point to checking if src0/src1 are valid operands


More information about the llvm-commits mailing list