From ed6a0a817ffdd4078c25336b473094c60f98b509 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@intel.com>
Date: Sun, 4 Nov 2018 17:31:27 +0000
Subject: [X86] Add vector shift by immediate to
 SimplifyDemandedBitsForTargetNode.

Summary: This also enables some constant folding from KnownBits propagation. This helps on some cases vXi64 case in 32-bit mode where constant vectors appear as vXi32 and a bitcast. This can prevent getNode from constant folding sra/shl/srl.

Reviewers: RKSimon, spatel

Reviewed By: spatel

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D54069

llvm-svn: 346102
---
 llvm/lib/Target/X86/X86ISelLowering.cpp | 42 +++++++++++++++++++++++++++++++++
 1 file changed, 42 insertions(+)

(limited to 'llvm/lib/Target')

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 905b99590a6..891f4a4cbdf 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -31817,6 +31817,7 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
 bool X86TargetLowering::SimplifyDemandedBitsForTargetNode(
     SDValue Op, const APInt &OriginalDemandedBits, KnownBits &Known,
     TargetLoweringOpt &TLO, unsigned Depth) const {
+  unsigned BitWidth = OriginalDemandedBits.getBitWidth();
   unsigned Opc = Op.getOpcode();
   switch(Opc) {
   case X86ISD::PMULDQ:
@@ -31833,6 +31834,42 @@ bool X86TargetLowering::SimplifyDemandedBitsForTargetNode(
       return true;
     break;
   }
+  case X86ISD::VSHLI: {
+    if (auto *ShiftImm = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
+      if (ShiftImm->getAPIntValue().uge(BitWidth))
+        break;
+
+      KnownBits KnownOp;
+      unsigned ShAmt = ShiftImm->getZExtValue();
+      APInt DemandedMask = OriginalDemandedBits.lshr(ShAmt);
+      if (SimplifyDemandedBits(Op.getOperand(0), DemandedMask, KnownOp, TLO,
+                               Depth + 1))
+        return true;
+    }
+    break;
+  }
+  case X86ISD::VSRAI:
+  case X86ISD::VSRLI: {
+    if (auto *ShiftImm = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
+      if (ShiftImm->getAPIntValue().uge(BitWidth))
+        break;
+
+      KnownBits KnownOp;
+      unsigned ShAmt = ShiftImm->getZExtValue();
+      APInt DemandedMask = OriginalDemandedBits << ShAmt;
+
+      // If any of the demanded bits are produced by the sign extension, we also
+      // demand the input sign bit.
+      if (Opc == X86ISD::VSRAI &&
+          OriginalDemandedBits.countLeadingZeros() < ShAmt)
+        DemandedMask.setSignBit();
+
+      if (SimplifyDemandedBits(Op.getOperand(0), DemandedMask, KnownOp, TLO,
+                               Depth + 1))
+        return true;
+    }
+    break;
+  }
   }
 
   return TargetLowering::SimplifyDemandedBitsForTargetNode(
@@ -34861,6 +34898,11 @@ static SDValue combineVectorShiftImm(SDNode *N, SelectionDAG &DAG,
     return getConstVector(EltBits, UndefElts, VT.getSimpleVT(), DAG, SDLoc(N));
   }
 
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  if (TLI.SimplifyDemandedBits(SDValue(N, 0),
+                               APInt::getAllOnesValue(NumBitsPerElt), DCI))
+    return SDValue(N, 0);
+
   return SDValue();
 }
 
-- 
cgit v1.2.3