Use movups to lower memcpy and memset even if it's not fast (like corei7).

The theory is it's still faster than a pair of movq / a quad of movl. This
will probably hurt older chips like P4 but should run faster on current
and future Intel processors. rdar://8817010

llvm-svn: 122955

1 files changed, 1 insertions, 5 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index f871b5a7701..ddec78bfff3 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1063,12 +1063,8 @@ 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;