[llvm] r195779 - Fix PR18054

[Michael Liao]

Author: hliao
Date: Tue Nov 26 14:31:31 2013
New Revision: 195779

URL: http://llvm.org/viewvc/llvm-project?rev=195779&view=rev
Log:
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.


Added:
    llvm/trunk/test/CodeGen/X86/pr18054.ll
Modified:
    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=195779&r1=195778&r2=195779&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Tue Nov 26 14:31:31 2013
@@ -13166,19 +13166,27 @@ SDValue X86TargetLowering::LowerSIGN_EXT
       // 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.

Added: llvm/trunk/test/CodeGen/X86/pr18054.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pr18054.ll?rev=195779&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/pr18054.ll (added)
+++ llvm/trunk/test/CodeGen/X86/pr18054.ll Tue Nov 26 14:31:31 2013
@@ -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
+}