[X86][XOP] Added VPERMIL2PD/VPERMIL2PS shuffle mask comment decoding

llvm-svn: 271809

3 files changed, 109 insertions, 0 deletions

diff --git a/llvm/lib/Target/X86/X86MCInstLower.cpp b/llvm/lib/Target/X86/X86MCInstLower.cpp
index 0111d7fbdc2..00debd849cd 100644
--- a/llvm/lib/Target/X86/X86MCInstLower.cpp
+++ b/llvm/lib/Target/X86/X86MCInstLower.cpp
@@ -1422,6 +1422,40 @@ void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
     }
     break;
   }
+
+  case X86::VPERMIL2PDrm:
+  case X86::VPERMIL2PSrm:
+  case X86::VPERMIL2PDrmY:
+  case X86::VPERMIL2PSrmY: {
+    if (!OutStreamer->isVerboseAsm())
+      break;
+    assert(MI->getNumOperands() > 7 &&
+      "We should always have at least 7 operands!");
+    const MachineOperand &DstOp = MI->getOperand(0);
+    const MachineOperand &SrcOp1 = MI->getOperand(1);
+    const MachineOperand &SrcOp2 = MI->getOperand(2);
+    const MachineOperand &MaskOp = MI->getOperand(6);
+    const MachineOperand &CtrlOp = MI->getOperand(MI->getNumOperands() - 1);
+
+    if (!CtrlOp.isImm())
+      break;
+
+    unsigned ElSize;
+    switch (MI->getOpcode()) {
+    default: llvm_unreachable("Invalid opcode");
+    case X86::VPERMIL2PSrm: case X86::VPERMIL2PSrmY: ElSize = 32; break;
+    case X86::VPERMIL2PDrm: case X86::VPERMIL2PDrmY: ElSize = 64; break;
+    }
+
+    if (auto *C = getConstantFromPool(*MI, MaskOp)) {
+      SmallVector<int, 16> Mask;
+      DecodeVPERMIL2PMask(C, (unsigned)CtrlOp.getImm(), ElSize, Mask);
+      if (!Mask.empty())
+        OutStreamer->AddComment(getShuffleComment(DstOp, SrcOp1, SrcOp2, Mask));
+    }
+    break;
+  }
+
   case X86::VPPERMrrm: {
     if (!OutStreamer->isVerboseAsm())
       break;
diff --git a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
index f24f38d4a21..abc05e43676 100644
--- a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
+++ b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
@@ -153,6 +153,77 @@ void DecodeVPERMILPMask(const Constant *C, unsigned ElSize,
   // TODO: Handle funny-looking vectors too.
 }
 
+void DecodeVPERMIL2PMask(const Constant *C, unsigned M2Z, unsigned ElSize,
+                         SmallVectorImpl<int> &ShuffleMask) {
+  Type *MaskTy = C->getType();
+
+  unsigned MaskTySize = MaskTy->getPrimitiveSizeInBits();
+  if (MaskTySize != 128 && MaskTySize != 256)
+    return;
+
+  // Only support vector types.
+  if (!MaskTy->isVectorTy())
+    return;
+
+  // Make sure its an integer type.
+  Type *VecEltTy = MaskTy->getVectorElementType();
+  if (!VecEltTy->isIntegerTy())
+    return;
+
+  // Support any element type from byte up to element size.
+  // This is necessary primarily because 64-bit elements get split to 32-bit
+  // in the constant pool on 32-bit target.
+  unsigned EltTySize = VecEltTy->getIntegerBitWidth();
+  if (EltTySize < 8 || EltTySize > ElSize)
+    return;
+
+  unsigned NumElements = MaskTySize / ElSize;
+  assert((NumElements == 2 || NumElements == 4 || NumElements == 8) &&
+         "Unexpected number of vector elements.");
+  ShuffleMask.reserve(NumElements);
+  unsigned NumElementsPerLane = 128 / ElSize;
+  unsigned Factor = ElSize / EltTySize;
+
+  for (unsigned i = 0; i < NumElements; ++i) {
+    Constant *COp = C->getAggregateElement(i * Factor);
+    if (!COp) {
+      ShuffleMask.clear();
+      return;
+    } else if (isa<UndefValue>(COp)) {
+      ShuffleMask.push_back(SM_SentinelUndef);
+      continue;
+    }
+
+    // VPERMIL2 Operation.
+    // Bits[3] - Match Bit.
+    // Bits[2:1] - (Per Lane) PD Shuffle Mask.
+    // Bits[2:0] - (Per Lane) PS Shuffle Mask.
+    uint64_t Selector = cast<ConstantInt>(COp)->getZExtValue();
+    int MatchBit = (Selector >> 3) & 0x1;
+
+    // M2Z[0:1]     MatchBit
+    //   0Xb           X        Source selected by Selector index.
+    //   10b           0        Source selected by Selector index.
+    //   10b           1        Zero.
+    //   11b           0        Zero.
+    //   11b           1        Source selected by Selector index.
+    if ((M2Z & 0x2) != 0 && MatchBit != (M2Z & 0x1)) {
+      ShuffleMask.push_back(SM_SentinelZero);
+      continue;
+    }
+
+    int Index = Selector & 0x3;
+    Index >>= (ElSize == 64 ? 1 : 0);
+    Index += (i / NumElementsPerLane) * NumElementsPerLane;
+
+    int Src = (Selector >> 2) & 0x1;
+    Index += Src * NumElements;
+    ShuffleMask.push_back(Index);
+  }
+
+  // TODO: Handle funny-looking vectors too.
+}
+
 void DecodeVPPERMMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask) {
   Type *MaskTy = C->getType();
   assert(MaskTy->getPrimitiveSizeInBits() == 128);
diff --git a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.h b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.h
index a07b26c04f7..d2565b84980 100644
--- a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.h
+++ b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.h
@@ -32,6 +32,10 @@ void DecodePSHUFBMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask);
 void DecodeVPERMILPMask(const Constant *C, unsigned ElSize,
                         SmallVectorImpl<int> &ShuffleMask);
 
+/// Decode a VPERMILP2 variable mask from an IR-level vector constant.
+void DecodeVPERMIL2PMask(const Constant *C, unsigned MatchImm, unsigned ElSize,
+                         SmallVectorImpl<int> &ShuffleMask);
+
 /// Decode a VPPERM variable mask from an IR-level vector constant.
 void DecodeVPPERMMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask);