[X86] SimplifyDemandedBits - add X86ISD::BLENDV SimplifyMultipleUseDemandedBits handling

Lets us see through multiple use operands

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 378eb97..12d3607 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -40906,6 +40906,28 @@ bool X86TargetLowering::SimplifyDemandedBitsForTargetNode(
       Known.One.setHighBits(ShAmt);
     return false;
   }
+  case X86ISD::BLENDV: {
+    SDValue Sel = Op.getOperand(0);
+    SDValue LHS = Op.getOperand(1);
+    SDValue RHS = Op.getOperand(2);
+
+    APInt SignMask = APInt::getSignMask(BitWidth);
+    SDValue NewSel = SimplifyMultipleUseDemandedBits(
+        Sel, SignMask, OriginalDemandedElts, TLO.DAG, Depth + 1);
+    SDValue NewLHS = SimplifyMultipleUseDemandedBits(
+        LHS, OriginalDemandedBits, OriginalDemandedElts, TLO.DAG, Depth + 1);
+    SDValue NewRHS = SimplifyMultipleUseDemandedBits(
+        RHS, OriginalDemandedBits, OriginalDemandedElts, TLO.DAG, Depth + 1);
+
+    if (NewSel || NewLHS || NewRHS) {
+      NewSel = NewSel ? NewSel : Sel;
+      NewLHS = NewLHS ? NewLHS : LHS;
+      NewRHS = NewRHS ? NewRHS : RHS;
+      return TLO.CombineTo(Op, TLO.DAG.getNode(X86ISD::BLENDV, SDLoc(Op), VT,
+                                               NewSel, NewLHS, NewRHS));
+    }
+    break;
+  }
   case X86ISD::PEXTRB:
   case X86ISD::PEXTRW: {
     SDValue Vec = Op.getOperand(0);

diff --git a/llvm/test/CodeGen/X86/vselect.ll b/llvm/test/CodeGen/X86/vselect.ll
index adeeb87..c7439e1 100644 (file)
--- a/llvm/test/CodeGen/X86/vselect.ll
+++ b/llvm/test/CodeGen/X86/vselect.ll
@@ -535,8 +535,8 @@ define <2 x i64> @shrunkblend_nonvselectuse(<2 x i1> %cond, <2 x i64> %a, <2 x i
 ; SSE41-LABEL: shrunkblend_nonvselectuse:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    psllq $63, %xmm0
-; SSE41-NEXT:    psrad $31, %xmm0
 ; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm2
+; SSE41-NEXT:    psrad $31, %xmm0
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
 ; SSE41-NEXT:    paddq %xmm2, %xmm0
 ; SSE41-NEXT:    retq