[X86][SSE] Add computeKnownBitsForTargetNode support for (V)PSLL/(V)PSRL instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@298806 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index fb432a000f00ba6e4871f0130511301ca118d12b..1cb81d338bdced3fbc1de4a8470b469ecf399a00 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -26591,6 +26591,7 @@ void X86TargetLowering::computeKnownBitsForTargetNode(const SDValue Op,
                                                       unsigned Depth) const {
   unsigned BitWidth = KnownZero.getBitWidth();
   unsigned Opc = Op.getOpcode();
+  EVT VT = Op.getValueType();
   assert((Opc >= ISD::BUILTIN_OP_END ||
           Opc == ISD::INTRINSIC_WO_CHAIN ||
           Opc == ISD::INTRINSIC_W_CHAIN ||
@@ -26624,9 +26625,33 @@ void X86TargetLowering::computeKnownBitsForTargetNode(const SDValue Op,
     KnownZero.setBits(NumLoBits, BitWidth);
     break;
   }
+  case X86ISD::VSHLI:
+  case X86ISD::VSRLI: {
+    if (auto *ShiftImm = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
+      if (ShiftImm->getAPIntValue().uge(VT.getScalarSizeInBits())) {
+        KnownZero = APInt::getAllOnesValue(BitWidth);
+        break;
+      }
+
+      DAG.computeKnownBits(Op.getOperand(0), KnownZero, KnownOne, Depth + 1);
+      unsigned ShAmt = ShiftImm->getZExtValue();
+      if (Opc == X86ISD::VSHLI) {
+        KnownZero = KnownZero << ShAmt;
+        KnownOne = KnownOne << ShAmt;
+        // Low bits are known zero.
+        KnownZero.setLowBits(ShAmt);
+      } else {
+        KnownZero = KnownZero.lshr(ShAmt);
+        KnownOne = KnownOne.lshr(ShAmt);
+        // High bits are known zero.
+        KnownZero.setHighBits(ShAmt);
+      }
+    }
+    break;
+  }
   case X86ISD::VZEXT: {
     SDValue N0 = Op.getOperand(0);
-    unsigned NumElts = Op.getValueType().getVectorNumElements();
+    unsigned NumElts = VT.getVectorNumElements();
 
     EVT SrcVT = N0.getValueType();
     unsigned InNumElts = SrcVT.getVectorNumElements();

diff --git a/test/CodeGen/X86/combine-abs.ll b/test/CodeGen/X86/combine-abs.ll
index 2d71e4f2f2734eaca492c1b2d279e8e8a2c682c9..ac8f790a2ead6b1718a255e9b7955925af2ec0b7 100644 (file)
--- a/test/CodeGen/X86/combine-abs.ll
+++ b/test/CodeGen/X86/combine-abs.ll
@@ -84,7 +84,6 @@ define <8 x i32> @combine_v8i32_abs_pos(<8 x i32> %a) {
 ; CHECK-LABEL: combine_v8i32_abs_pos:
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    vpsrld $1, %ymm0, %ymm0
-; CHECK-NEXT:    vpabsd %ymm0, %ymm0
 ; CHECK-NEXT:    retq
   %1 = lshr <8 x i32> %a, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
   %2 = call <8 x i32> @llvm.x86.avx2.pabs.d(<8 x i32> %1)