[DAGCombiner] Support folding (mulhs/u X, 0)->0 for vectors.

We should probably also fold (mulhs/u X, 1) for vectors, but that's harder.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@320344 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 2170670a0b6156c183bd37e6a0ed3ee3491816c9..4cc86ed2d369c8a46c0f443664fa0323a890f1a3 100644 (file)
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -3099,6 +3099,14 @@ SDValue DAGCombiner::visitMULHS(SDNode *N) {
   EVT VT = N->getValueType(0);
   SDLoc DL(N);
 
+  if (VT.isVector()) {
+    // fold (mulhs x, 0) -> 0
+    if (ISD::isBuildVectorAllZeros(N1.getNode()))
+      return N1;
+    if (ISD::isBuildVectorAllZeros(N0.getNode()))
+      return N0;
+  }
+
   // fold (mulhs x, 0) -> 0
   if (isNullConstant(N1))
     return N1;
@@ -3138,6 +3146,14 @@ SDValue DAGCombiner::visitMULHU(SDNode *N) {
   EVT VT = N->getValueType(0);
   SDLoc DL(N);
 
+  if (VT.isVector()) {
+    // fold (mulhu x, 0) -> 0
+    if (ISD::isBuildVectorAllZeros(N1.getNode()))
+      return N1;
+    if (ISD::isBuildVectorAllZeros(N0.getNode()))
+      return N0;
+  }
+
   // fold (mulhu x, 0) -> 0
   if (isNullConstant(N1))
     return N1;

diff --git a/test/CodeGen/X86/pr34855.ll b/test/CodeGen/X86/pr34855.ll
index ee4428908a27c7543a27aeacd48c281ba3cee3e2..746d1ff56cca7dc56ec63935085de4e39e421cdf 100644 (file)
--- a/test/CodeGen/X86/pr34855.ll
+++ b/test/CodeGen/X86/pr34855.ll
@@ -8,18 +8,13 @@ define void @PR34855(<2 x i32> *%p0, <2 x i32> *%p1, <2 x i32> *%p2) {
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
-; X86-NEXT:    movlps %xmm0, (%eax)
+; X86-NEXT:    movsd %xmm0, (%eax)
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: PR34855:
 ; X64:       # %bb.0:
-; X64-NEXT:    movslq 4(%rdi), %rax
-; X64-NEXT:    movq %rax, %xmm0
-; X64-NEXT:    movslq (%rdi), %rax
-; X64-NEXT:    movq %rax, %xmm1
-; X64-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; X64-NEXT:    movq %xmm0, (%rdx)
+; X64-NEXT:    movq (%rdi), %rax
+; X64-NEXT:    movq %rax, (%rdx)
 ; X64-NEXT:    retq
   %tmp = load <2 x i32>, <2 x i32>* %p0, align 8
   %tmp1 = load <2 x i32>, <2 x i32>* %p1, align 8