From d617a3015dbc00da9fbef6914da8f119ad84ef89 Mon Sep 17 00:00:00 2001
From: Michael Liao <michael.liao@intel.com>
Date: Tue, 26 Nov 2013 20:31:31 +0000
Subject: [PATCH] Fix PR18054

- Fix bug in (vsext (vzext x)) -> (vsext x) in SIGN_EXTEND_IN_REG
  lowering where we need to check whether x is a vector type (in-reg
  type) of i8, i16 or i32; otherwise, that optimization is not valid.

llvm-svn: 195779
---
 llvm/lib/Target/X86/X86ISelLowering.cpp | 22 +++++++++++++++-------
 llvm/test/CodeGen/X86/pr18054.ll        | 10 ++++++++++
 2 files changed, 25 insertions(+), 7 deletions(-)
 create mode 100644 llvm/test/CodeGen/X86/pr18054.ll

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index fc52c0cd6696..40d5f2b0c76f 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -13166,19 +13166,27 @@ SDValue X86TargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op,
       // fall through
     case MVT::v4i32:
     case MVT::v8i16: {
-      // (sext (vzext x)) -> (vsext x)
       SDValue Op0 = Op.getOperand(0);
       SDValue Op00 = Op0.getOperand(0);
       SDValue Tmp1;
       // Hopefully, this VECTOR_SHUFFLE is just a VZEXT.
       if (Op0.getOpcode() == ISD::BITCAST &&
-          Op00.getOpcode() == ISD::VECTOR_SHUFFLE)
+          Op00.getOpcode() == ISD::VECTOR_SHUFFLE) {
+        // (sext (vzext x)) -> (vsext x)
         Tmp1 = LowerVectorIntExtend(Op00, Subtarget, DAG);
-      if (Tmp1.getNode()) {
-        SDValue Tmp1Op0 = Tmp1.getOperand(0);
-        assert(Tmp1Op0.getOpcode() == X86ISD::VZEXT &&
-               "This optimization is invalid without a VZEXT.");
-        return DAG.getNode(X86ISD::VSEXT, dl, VT, Tmp1Op0.getOperand(0));
+        if (Tmp1.getNode()) {
+          EVT ExtraEltVT = ExtraVT.getVectorElementType();
+          // This folding is only valid when the in-reg type is a vector of i8,
+          // i16, or i32.
+          if (ExtraEltVT == MVT::i8 || ExtraEltVT == MVT::i16 ||
+              ExtraEltVT == MVT::i32) {
+            SDValue Tmp1Op0 = Tmp1.getOperand(0);
+            assert(Tmp1Op0.getOpcode() == X86ISD::VZEXT &&
+                   "This optimization is invalid without a VZEXT.");
+            return DAG.getNode(X86ISD::VSEXT, dl, VT, Tmp1Op0.getOperand(0));
+          }
+          Op0 = Tmp1;
+        }
       }
 
       // If the above didn't work, then just use Shift-Left + Shift-Right.
diff --git a/llvm/test/CodeGen/X86/pr18054.ll b/llvm/test/CodeGen/X86/pr18054.ll
new file mode 100644
index 000000000000..b7af51618047
--- /dev/null
+++ b/llvm/test/CodeGen/X86/pr18054.ll
@@ -0,0 +1,10 @@
+; RUN: llc < %s -mtriple=x86_64-pc-linux -mcpu=penryn | FileCheck %s
+
+define void @foo(<16 x i32>* %p, <16 x i1> %x) {
+  %ret = sext <16 x i1> %x to <16 x i32>
+  store <16 x i32> %ret, <16 x i32>* %p
+  ret void
+; CHECK: foo
+; CHECK-NOT: pmovsxbd
+; CHECK: ret
+}