Fix a couple of logic bugs in TargetLowering::SimplifyDemandedBits. PR11514.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146219 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 8956651654048bc4c75efcde62c2917ee9cf2aa2..ee329dc671ea118e647b359a0aeff33d1c5de018 100644 (file)
--- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -1483,9 +1483,8 @@ bool TargetLowering::SimplifyDemandedBits(SDValue Op,
       if (InOp.getNode()->getOpcode() == ISD::ANY_EXTEND) {
         SDValue InnerOp = InOp.getNode()->getOperand(0);
         EVT InnerVT = InnerOp.getValueType();
-        if ((APInt::getHighBitsSet(BitWidth,
-                                   BitWidth - InnerVT.getSizeInBits()) &
-               DemandedMask) == 0 &&
+        unsigned InnerBits = InnerVT.getSizeInBits();
+        if (ShAmt < InnerBits && NewMask.lshr(InnerBits) == 0 &&
             isTypeDesirableForOp(ISD::SHL, InnerVT)) {
           EVT ShTy = getShiftAmountTy(InnerVT);
           if (!APInt(BitWidth, ShAmt).isIntN(ShTy.getSizeInBits()))
@@ -1555,7 +1554,7 @@ bool TargetLowering::SimplifyDemandedBits(SDValue Op,
     // always convert this into a logical shr, even if the shift amount is
     // variable.  The low bit of the shift cannot be an input sign bit unless
     // the shift amount is >= the size of the datatype, which is undefined.
-    if (DemandedMask == 1)
+    if (NewMask == 1)
       return TLO.CombineTo(Op,
                            TLO.DAG.getNode(ISD::SRL, dl, Op.getValueType(),
                                            Op.getOperand(0), Op.getOperand(1)));

diff --git a/test/CodeGen/PowerPC/2011-12-08-DemandedBitsMiscompile.ll b/test/CodeGen/PowerPC/2011-12-08-DemandedBitsMiscompile.ll
new file mode 100644 (file)
index 0000000..a18829e
--- /dev/null
+++ b/test/CodeGen/PowerPC/2011-12-08-DemandedBitsMiscompile.ll
@@ -0,0 +1,16 @@
+; RUN: llc < %s -mtriple=powerpc64-unknown-linux-gnu -mcpu=g4 | FileCheck %s
+
+define void @test(i32* nocapture %x, i64* %xx, i32* %yp) nounwind uwtable ssp {
+entry:
+  %yy = load i32* %yp
+  %y = add i32 %yy, 1
+  %z = zext i32 %y to i64
+  %z2 = shl i64 %z, 32 
+  store i64 %z2, i64* %xx, align 4
+  ret void
+
+; CHECK: test:
+; CHECK: sldi {{.*}}, {{.*}}, 32
+; Note: it's okay if someday CodeGen gets smart enough to optimize out
+; the shift.
+}