[TargetLowering][X86] TeachSimplifyDemandedBits to handle cases where only the sign bit is demanded from a SETCC and can be passed through

If we're doing a compare that only tests the sign bit and only the sign bit is demanded, we can just bypass the node. This removes one of the blend dependencies in our v2i64->v2f32 uint_to_fp codegen on pre-sse4.2 targets.

Differential Revision: https://reviews.llvm.org/D72356

1 files changed, 21 insertions, 0 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index ebbbc1550b9..a5ea7c121bd 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -726,6 +726,27 @@ SDValue TargetLowering::SimplifyMultipleUseDemandedBits(
       return Op.getOperand(1);
     break;
   }
+  case ISD::SETCC: {
+    SDValue Op0 = Op.getOperand(0);
+    SDValue Op1 = Op.getOperand(1);
+    ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
+    // If (1) we only need the sign-bit, (2) the setcc operands are the same
+    // width as the setcc result, and (3) the result of a setcc conforms to 0 or
+    // -1, we may be able to bypass the setcc.
+    if (DemandedBits.isSignMask() &&
+        Op0.getScalarValueSizeInBits() == DemandedBits.getBitWidth() &&
+        getBooleanContents(Op0.getValueType()) ==
+            BooleanContent::ZeroOrNegativeOneBooleanContent) {
+      // If we're testing X < 0, then this compare isn't needed - just use X!
+      // FIXME: We're limiting to integer types here, but this should also work
+      // if we don't care about FP signed-zero. The use of SETLT with FP means
+      // that we don't care about NaNs.
+      if (CC == ISD::SETLT && Op1.getValueType().isInteger() &&
+          (isNullConstant(Op1) || ISD::isBuildVectorAllZeros(Op1.getNode())))
+        return Op0;
+    }
+    break;
+  }
   case ISD::SIGN_EXTEND_INREG: {
     // If none of the extended bits are demanded, eliminate the sextinreg.
     EVT ExVT = cast<VTSDNode>(Op.getOperand(1))->getVT();