[llvm-commits] [llvm] r156539 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/avx-vbroadcast.ll

[Nadav Rotem]

Author: nadav
Date: Thu May 10 07:22:05 2012
New Revision: 156539

URL: http://llvm.org/viewvc/llvm-project?rev=156539&view=rev
Log:
Generate AVX/AVX2 shuffles even when there is a memory op somewhere else in the program.
Starting r155461 we are able to select patterns for vbroadcast even when the load op is used by other users.

Fix PR11900.



Modified:
    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
    llvm/trunk/test/CodeGen/X86/avx-vbroadcast.ll

Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=156539&r1=156538&r2=156539&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Thu May 10 07:22:05 2012
@@ -5029,10 +5029,6 @@
   if (!ISD::isNormalLoad(Ld.getNode()))
     return SDValue();
 
-  // Reject loads that have uses of the chain result
-  if (Ld->hasAnyUseOfValue(1))
-    return SDValue();
-
   unsigned ScalarSize = Ld.getValueType().getSizeInBits();
 
   if (ScalarSize == 32 || (Is256 && ScalarSize == 64))

Modified: llvm/trunk/test/CodeGen/X86/avx-vbroadcast.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx-vbroadcast.ll?rev=156539&r1=156538&r2=156539&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/avx-vbroadcast.ll (original)
+++ llvm/trunk/test/CodeGen/X86/avx-vbroadcast.ll Thu May 10 07:22:05 2012
@@ -112,3 +112,20 @@
   %vecinit2.i = insertelement <2 x double> %vecinit.i, double %q, i32 1
   ret <2 x double> %vecinit2.i
 }
+
+; CHECK: _RR
+; CHECK: vbroadcastss (%
+; CHECK: ret
+define <4 x float> @_RR(float* %ptr, i32* %k) nounwind uwtable readnone ssp {
+entry:
+  %q = load float* %ptr, align 4
+  %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
+  %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
+  %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
+  %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
+  ; force a chain
+  %j = load i32* %k, align 4
+  store i32 %j, i32* undef
+  ret <4 x float> %vecinit6.i
+}
+