getSplatIndex assumes that the first element of the mask contains the splat index

which is not always true if the mask contains undefs. Modified it to return
the first non undef value.

llvm-svn: 96621

2 files changed, 18 insertions, 6 deletions

diff --git a/llvm/include/llvm/CodeGen/SelectionDAGNodes.h b/llvm/include/llvm/CodeGen/SelectionDAGNodes.h
index 9a426a04773..90669136204 100644
--- a/llvm/include/llvm/CodeGen/SelectionDAGNodes.h
+++ b/llvm/include/llvm/CodeGen/SelectionDAGNodes.h
@@ -1764,7 +1764,12 @@ public:
   bool isSplat() const { return isSplatMask(Mask, getValueType(0)); }
   int  getSplatIndex() const { 
     assert(isSplat() && "Cannot get splat index for non-splat!");
-    return Mask[0];
+    EVT VT = getValueType(0);
+    for (unsigned i = 0, e = VT.getVectorNumElements(); i != e; ++i) {
+      if (Mask[i] != -1)
+        return Mask[i];
+    }
+    return -1;
   }
   static bool isSplatMask(const int *Mask, EVT VT);
 
diff --git a/llvm/test/CodeGen/X86/vec_shuffle-36.ll b/llvm/test/CodeGen/X86/vec_shuffle-36.ll
index 8a93a7eeee3..1ea37c881e7 100644
--- a/llvm/test/CodeGen/X86/vec_shuffle-36.ll
+++ b/llvm/test/CodeGen/X86/vec_shuffle-36.ll
@@ -1,9 +1,16 @@
-; RUN: llc < %s -march=x86 -mattr=sse41 -o %t
-; RUN: grep pshufb %t | count 1
-
+; RUN: llc < %s -march=x86-64 -mattr=sse41 | FileCheck %s
 
 define <8 x i16> @shuf6(<8 x i16> %T0, <8 x i16> %T1) nounwind readnone {
+; CHECK: pshufb
+; CHECK-NOT: pshufb
+; CHECK: ret
 entry:
-	%tmp9 = shufflevector <8 x i16> %T0, <8 x i16> %T1, <8 x i32> < i32 3, i32 2, i32 0, i32 2, i32 1, i32 5, i32 6 , i32 undef >
-	ret <8 x i16> %tmp9
+  %tmp9 = shufflevector <8 x i16> %T0, <8 x i16> %T1, <8 x i32> < i32 3, i32 2, i32 0, i32 2, i32 1, i32 5, i32 6 , i32 undef >
+  ret <8 x i16> %tmp9
 }
+
+define <8 x i16> @shuf7(<8 x i16> %t0) {
+; CHECK: pshufd
+  %tmp10 = shufflevector <8 x i16> %t0, <8 x i16> undef, <8 x i32> < i32 undef, i32 2, i32 2, i32 2, i32 2, i32 2, i32 undef, i32 undef >
+  ret <8 x i16> %tmp10
+}
\ No newline at end of file