[X86][SSE] MOVMSK only uses the sign bit from each vector element

[RKSimon]

Summary

Pass the input vector through SimplifyDemandedBits as we only need the sign bit from each vector element of MOVMSK

We'd probably get more hits if SimplifyDemandedBits was better at handling vectors...

[craig.topper]

LGTM

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

static SDValue combineSetCC(SDNode *N, SelectionDAG &DAG,
   // to avoid scalarization via legalization because v4i32 is not a legal type.
   if (Subtarget.hasSSE1() && !Subtarget.hasSSE2() && VT == MVT::v4i32 &&
       LHS.getValueType() == MVT::v4f32)
     return LowerVSETCC(SDValue(N, 0), Subtarget, DAG);
 
   return SDValue();
 }
 
+static SDValue combineMOVMSK(SDNode *N, SelectionDAG &DAG,
+                             TargetLowering::DAGCombinerInfo &DCI) {
+  SDValue Src = N->getOperand(0);
+  MVT SrcVT = Src.getSimpleValueType();
+
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  TargetLowering::TargetLoweringOpt TLO(DAG, !DCI.isBeforeLegalize(),
+                                        !DCI.isBeforeLegalizeOps());
+
+  // MOVMSK only uses the MSB from each vector element.
+  KnownBits Known;
+  APInt DemandedMask(APInt::getSignMask(SrcVT.getScalarSizeInBits()));
+  if (TLI.SimplifyDemandedBits(Src, DemandedMask, Known, TLO)) {
+    DCI.AddToWorklist(Src.getNode());
+    DCI.CommitTargetLoweringOpt(TLO);
+    return SDValue(N, 0);
+  }
+
+  return SDValue();
+}
+
 static SDValue combineGatherScatter(SDNode *N, SelectionDAG &DAG,
                                     TargetLowering::DAGCombinerInfo &DCI,
                                     const X86Subtarget &Subtarget) {
   SDLoc DL(N);
 
   // Pre-shrink oversized index elements to avoid triggering scalarization.
   if (DCI.isBeforeLegalize()) {
     SDValue Index = N->getOperand(4);
SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
   case X86ISD::FMADDS1:
   case X86ISD::FMADDS3:
   case X86ISD::FMADD4S:
   case ISD::FMA:            return combineFMA(N, DAG, Subtarget);
   case X86ISD::FMADDSUB_RND:
   case X86ISD::FMSUBADD_RND:
   case X86ISD::FMADDSUB:
   case X86ISD::FMSUBADD:    return combineFMADDSUB(N, DAG, Subtarget);
+  case X86ISD::MOVMSK:      return combineMOVMSK(N, DAG, DCI);
   case X86ISD::MGATHER:
   case X86ISD::MSCATTER:
   case ISD::MGATHER:
   case ISD::MSCATTER:       return combineGatherScatter(N, DAG, DCI, Subtarget);
   case X86ISD::TESTM:       return combineTestM(N, DAG, Subtarget);
   case X86ISD::PCMPEQ:
   case X86ISD::PCMPGT:      return combineVectorCompare(N, DAG, Subtarget);
   }

llvm/trunk/test/CodeGen/X86/setcc-lowering.ll

 ; AVX-NEXT:    xorl %eax, %eax
 ; AVX-NEXT:    cmpq $65536, %rdi # imm = 0x10000
 ; AVX-NEXT:    setl %al
 ; AVX-NEXT:    vmovd %eax, %xmm3
 ; AVX-NEXT:    vpshufb %xmm1, %xmm3, %xmm3
 ; AVX-NEXT:    vpand %xmm0, %xmm3, %xmm3
 ; AVX-NEXT:    vpsllw $7, %xmm3, %xmm3
 ; AVX-NEXT:    vpand %xmm2, %xmm3, %xmm3
-; AVX-NEXT:    vpcmpgtb %xmm3, %xmm1, %xmm3
 ; AVX-NEXT:    vpmovmskb %xmm3, %eax
 ; AVX-NEXT:    testw %ax, %ax
 ; AVX-NEXT:    jne .LBB1_1
 ; AVX-NEXT:  # %bb.2: # %for_exit600
 ; AVX-NEXT:    retq
 ;
 ; KNL-32-LABEL: pr26232:
 ; KNL-32:       # %bb.0: # %for_loop599.preheader