[llvm-commits] [llvm] r53666 - in /llvm/trunk: lib/Transforms/Scalar/InstructionCombining.cpp test/Transforms/InstCombine/2008-07-16-sse2_storel_dq.ll

[Evan Cheng]

Author: evancheng
Date: Wed Jul 16 02:28:14 2008
New Revision: 53666

URL: http://llvm.org/viewvc/llvm-project?rev=53666&view=rev
Log:
Fix PR2296. Do not transform x86_sse2_storel_dq into a full-width store.

Added:
    llvm/trunk/test/Transforms/InstCombine/2008-07-16-sse2_storel_dq.ll
Modified:
    llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp

Modified: llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp?rev=53666&r1=53665&r2=53666&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Wed Jul 16 02:28:14 2008
@@ -8646,7 +8646,6 @@
   case Intrinsic::x86_sse_storeu_ps:
   case Intrinsic::x86_sse2_storeu_pd:
   case Intrinsic::x86_sse2_storeu_dq:
-  case Intrinsic::x86_sse2_storel_dq:
     // Turn X86 storeu -> store if the pointer is known aligned.
     if (GetOrEnforceKnownAlignment(II->getOperand(1), 16) >= 16) {
       const Type *OpPtrTy = 

Added: llvm/trunk/test/Transforms/InstCombine/2008-07-16-sse2_storel_dq.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/2008-07-16-sse2_storel_dq.ll?rev=53666&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/2008-07-16-sse2_storel_dq.ll (added)
+++ llvm/trunk/test/Transforms/InstCombine/2008-07-16-sse2_storel_dq.ll Wed Jul 16 02:28:14 2008
@@ -0,0 +1,13 @@
+; RUN: llvm-as < %s | opt -instcombine | llvm-dis | not grep {store }
+; PR2296
+
+ at G = common global double 0.000000e+00, align 16
+
+define void @x(<2 x i64> %y) nounwind  {
+entry:
+	bitcast <2 x i64> %y to <4 x i32>
+	call void @llvm.x86.sse2.storel.dq( i8* bitcast (double* @G to i8*), <4 x i32> %0 ) nounwind 
+	ret void
+}
+
+declare void @llvm.x86.sse2.storel.dq(i8*, <4 x i32>) nounwind