call libc memcpy/memset if array size is bigger then threshold.

Coping 100MB array (after a warmup) shows that glibc 2.6.1 implementation on
x86-64 (core 2) is 30% faster (from 0.270917s to 0.188079s)



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@41479 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 6673c5f..75fbd44 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -3753,10 +3753,10 @@
   if (Align == 0) Align = 1;
 
   ConstantSDNode *I = dyn_cast<ConstantSDNode>(Op.getOperand(3));
-  // If not DWORD aligned, call memset if size is less than the threshold.
+  // If not DWORD aligned or size is more than the threshold, call memset.
   // It knows how to align to the right boundary first.
   if ((Align & 3) != 0 ||
-      (I && I->getValue() < Subtarget->getMinRepStrSizeThreshold())) {
+      (I && I->getValue() > Subtarget->getMinRepStrSizeThreshold())) {
     MVT::ValueType IntPtr = getPointerTy();
     const Type *IntPtrTy = getTargetData()->getIntPtrType();
     TargetLowering::ArgListTy Args; 
@@ -3909,10 +3909,10 @@
   if (Align == 0) Align = 1;
 
   ConstantSDNode *I = dyn_cast<ConstantSDNode>(Op.getOperand(3));
-  // If not DWORD aligned, call memcpy if size is less than the threshold.
+  // If not DWORD aligned or size is more than the threshold, call memcpy.
   // It knows how to align to the right boundary first.
   if ((Align & 3) != 0 ||
-      (I && I->getValue() < Subtarget->getMinRepStrSizeThreshold())) {
+      (I && I->getValue() > Subtarget->getMinRepStrSizeThreshold())) {
     MVT::ValueType IntPtr = getPointerTy();
     TargetLowering::ArgListTy Args;
     TargetLowering::ArgListEntry Entry;