Fix pr11193.

SHL inserts zeros from the right, thus even when the original
sign_extend_inreg value was of 1-bit, we need to sra.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142724 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 96f04e729810e3331608f22aae422eab9e654c70..7dec7c4e336618e16a9a954f859da7d449ea60ac 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -10146,9 +10146,6 @@ SDValue X86TargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG)
                          DAG.getConstant(SHLIntrinsicsID, MVT::i32),
                          Node->getOperand(0), ShAmt);
 
-    // In case of 1 bit sext, no need to shr
-    if (ExtraVT.getScalarType().getSizeInBits() == 1) return Tmp1;
-
     return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT,
                        DAG.getConstant(SRAIntrinsicsID, MVT::i32),
                        Tmp1, ShAmt);

diff --git a/test/CodeGen/X86/2011-10-21-widen-cmp.ll b/test/CodeGen/X86/2011-10-21-widen-cmp.ll
index aa1a60026f43909e50fe4038da29eb80811a15bf..2fe645b078150cc441f450e70b3bc314dd6fd8b6 100644 (file)
--- a/test/CodeGen/X86/2011-10-21-widen-cmp.ll
+++ b/test/CodeGen/X86/2011-10-21-widen-cmp.ll
@@ -28,3 +28,18 @@ entry:
   store <2 x double> %1, <2 x double>* undef
   ret void
 }
+
+; CHECK: mp_11193
+; CHECK: psraw   $15
+; CHECK: ret
+define void @mp_11193(<8 x float> * nocapture %aFOO, <8 x float>* nocapture %RET)
+nounwind {
+allocas:
+  %bincmp = fcmp olt <8 x float> <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 9.000000e+00, float 1.000000e+00, float 9.000000e+00, float 1.000000e+00> , <float 3.000000e+00, float 3.000000e+00, float 3.000000e+00, float 3.000000e+00, float 3.000000e+00, float 3.000000e+00, float 3.000000e+00, float 3.000000e+00>
+  %t = extractelement <8 x i1> %bincmp, i32 0
+  %ft = sitofp i1 %t to float
+  %pp = bitcast <8 x float>* %RET to float*
+  store float %ft, float* %pp
+  ret void
+}
+