[X86] computeKnownBitsForTargetNode - add generic handling of PSHUFB

When PSHUFB is used as a LUT (for CTPOP, BITREVERSE etc.), its the source operand that is constant and the index operand the variable. As long as the indices don't set the MSB (which zeros the output element), then the common known bits from the source operand can be used directly, even though the shuffle mask isn't constant.

Further helps to improve CTPOP reduction codegen

Author: RKSimon

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -36707,6 +36707,19 @@ void X86TargetLowering::computeKnownBitsForTargetNode(const SDValue Op,
     Known = Known.trunc(BitWidth);
     break;
   }
+  case X86ISD::PSHUFB: {
+    SDValue Src = Op.getOperand(0);
+    SDValue Idx = Op.getOperand(1);
+
+    // If the index vector is never negative (MSB is zero), then all elements
+    // come from the source vector. This is useful for cases where
+    // PSHUFB is being used as a LUT (ctpop etc.) - the target shuffle handling
+    // below will handle the more common constant shuffle mask case.
+    KnownBits KnownIdx = DAG.computeKnownBits(Idx, DemandedElts, Depth + 1);
+    if (KnownIdx.isNonNegative())
+      Known = DAG.computeKnownBits(Src, Depth + 1);
+    break;
+  }
   case X86ISD::VBROADCAST: {
     SDValue Src = Op.getOperand(0);
     if (!Src.getSimpleValueType().isVector()) {