[llvm-commits] [llvm] r48136 - in /llvm/trunk: lib/CodeGen/SelectionDAG/DAGCombiner.cpp test/CodeGen/X86/vec_zero-2.ll
Evan Cheng
evan.cheng at apple.com
Mon Mar 10 00:19:13 PDT 2008
Author: evancheng
Date: Mon Mar 10 02:19:13 2008
New Revision: 48136
URL: http://llvm.org/viewvc/llvm-project?rev=48136&view=rev
Log:
Avoid creating BUILD_VECTOR of all zero elements of "non-normalized" type (e.g. v8i16 on x86) after legalizer. Instruction selection does not expect to see them. In all likelihood this can only be an issue in a bugpoint reduced test case.
Added:
llvm/trunk/test/CodeGen/X86/vec_zero-2.ll
Modified:
llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=48136&r1=48135&r2=48136&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Mon Mar 10 02:19:13 2008
@@ -1103,6 +1103,10 @@
// fold (sub x, x) -> 0
if (N0 == N1)
+ if (ISD::isBuildVectorAllZeros(N0.Val))
+ // Zero vectors might be normalized to a particular vector type to ensure
+ // they are CSE'd. Return it as it is.
+ return N0;
return DAG.getConstant(0, N->getValueType(0));
// fold (sub c1, c2) -> c1-c2
if (N0C && N1C)
Added: llvm/trunk/test/CodeGen/X86/vec_zero-2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vec_zero-2.ll?rev=48136&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vec_zero-2.ll (added)
+++ llvm/trunk/test/CodeGen/X86/vec_zero-2.ll Mon Mar 10 02:19:13 2008
@@ -0,0 +1,24 @@
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2
+
+define i32 @t() {
+entry:
+ br i1 true, label %bb4743, label %bb1656
+bb1656: ; preds = %entry
+ ret i32 0
+bb1664: ; preds = %entry
+ br i1 false, label %bb5310, label %bb4743
+bb4743: ; preds = %bb1664
+ %tmp5256 = bitcast <2 x i64> zeroinitializer to <8 x i16> ; <<8 x i16>> [#uses=1]
+ %tmp5257 = sub <8 x i16> %tmp5256, zeroinitializer ; <<8 x i16>> [#uses=1]
+ %tmp5258 = bitcast <8 x i16> %tmp5257 to <2 x i64> ; <<2 x i64>> [#uses=1]
+ %tmp5265 = bitcast <2 x i64> %tmp5258 to <8 x i16> ; <<8 x i16>> [#uses=1]
+ %tmp5266 = call <8 x i16> @llvm.x86.sse2.packuswb.128( <8 x i16> %tmp5265, <8 x i16> zeroinitializer ) nounwind readnone ; <<8 x i16>> [#uses=1]
+ %tmp5267 = bitcast <8 x i16> %tmp5266 to <2 x i64> ; <<2 x i64>> [#uses=1]
+ %tmp5294 = and <2 x i64> zeroinitializer, %tmp5267 ; <<2 x i64>> [#uses=1]
+ br label %bb5310
+bb5310: ; preds = %bb4743, %bb1664
+ %tmp5294.pn = phi <2 x i64> [ %tmp5294, %bb4743 ], [ zeroinitializer, %bb1664 ] ; <<2 x i64>> [#uses=0]
+ ret i32 0
+}
+
+declare <8 x i16> @llvm.x86.sse2.packuswb.128(<8 x i16>, <8 x i16>) nounwind readnone
More information about the llvm-commits
mailing list