Revert r122955. It seems using movups to lower memcpy can cause massive regression (even on Nehalem) in edge cases. I also didn't see any real performance benefit.

llvm-svn: 123015

1 files changed, 5 insertions, 1 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index ddec78bfff3..f871b5a7701 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1063,8 +1063,12 @@ X86TargetLowering::getOptimalMemOpType(uint64_t Size,
   // linux.  This is because the stack realignment code can't handle certain
   // cases like PR2962.  This should be removed when PR2962 is fixed.
   const Function *F = MF.getFunction();
-  if (NonScalarIntSafe && !F->hasFnAttr(Attribute::NoImplicitFloat)) {
+  if (NonScalarIntSafe &&
+      !F->hasFnAttr(Attribute::NoImplicitFloat)) {
     if (Size >= 16 &&
+        (Subtarget->isUnalignedMemAccessFast() ||
+         ((DstAlign == 0 || DstAlign >= 16) &&
+          (SrcAlign == 0 || SrcAlign >= 16))) &&
         Subtarget->getStackAlignment() >= 16) {
       if (Subtarget->hasSSE2())
         return MVT::v4i32;