foo

llvm-svn: 344873

4 files changed, 62 insertions, 45 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 9e431162083..26a7c004e1f 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -6046,7 +6046,7 @@ static bool getTargetShuffleMask(SDNode *N, MVT VT, bool AllowSentinelZero,
       break;
     }
     if (auto *C = getTargetConstantFromNode(MaskNode)) {
-      DecodeVPERMILPMask(C, MaskEltSize, Mask);
+      DecodeVPERMILPMask(C, MaskEltSize, VT.getSizeInBits(), Mask);
       break;
     }
     return false;
@@ -6063,7 +6063,7 @@ static bool getTargetShuffleMask(SDNode *N, MVT VT, bool AllowSentinelZero,
       break;
     }
     if (auto *C = getTargetConstantFromNode(MaskNode)) {
-      DecodePSHUFBMask(C, Mask);
+      DecodePSHUFBMask(C, VT.getSizeInBits(), Mask);
       break;
     }
     return false;
@@ -6128,7 +6128,7 @@ static bool getTargetShuffleMask(SDNode *N, MVT VT, bool AllowSentinelZero,
         break;
       }
       if (auto *C = getTargetConstantFromNode(MaskNode)) {
-        DecodeVPERMIL2PMask(C, CtrlImm, MaskEltSize, Mask);
+        DecodeVPERMIL2PMask(C, CtrlImm, MaskEltSize, VT.getSizeInBits(), Mask);
         break;
       }
     }
@@ -6145,7 +6145,7 @@ static bool getTargetShuffleMask(SDNode *N, MVT VT, bool AllowSentinelZero,
       break;
     }
     if (auto *C = getTargetConstantFromNode(MaskNode)) {
-      DecodeVPPERMMask(C, Mask);
+      DecodeVPPERMMask(C, VT.getSizeInBits(), Mask);
       break;
     }
     return false;
@@ -6163,7 +6163,7 @@ static bool getTargetShuffleMask(SDNode *N, MVT VT, bool AllowSentinelZero,
       break;
     }
     if (auto *C = getTargetConstantFromNode(MaskNode)) {
-      DecodeVPERMVMask(C, MaskEltSize, Mask);
+      DecodeVPERMVMask(C, MaskEltSize, VT.getSizeInBits(), Mask);
       break;
     }
     return false;
@@ -6178,7 +6178,7 @@ static bool getTargetShuffleMask(SDNode *N, MVT VT, bool AllowSentinelZero,
     SDValue MaskNode = N->getOperand(1);
     unsigned MaskEltSize = VT.getScalarSizeInBits();
     if (auto *C = getTargetConstantFromNode(MaskNode)) {
-      DecodeVPERMV3Mask(C, MaskEltSize, Mask);
+      DecodeVPERMV3Mask(C, MaskEltSize, VT.getSizeInBits(), Mask);
       break;
     }
     return false;
diff --git a/llvm/lib/Target/X86/X86MCInstLower.cpp b/llvm/lib/Target/X86/X86MCInstLower.cpp
index 58b1c505944..9c278116d7e 100644
--- a/llvm/lib/Target/X86/X86MCInstLower.cpp
+++ b/llvm/lib/Target/X86/X86MCInstLower.cpp
@@ -1594,6 +1594,18 @@ void X86AsmPrinter::EmitSEHInstruction(const MachineInstr *MI) {
   }
 }
 
+static unsigned getRegisterWidth(const MCOperandInfo &Info) {
+  if (Info.RegClass == X86::VR128RegClassID ||
+      Info.RegClass == X86::VR128XRegClassID)
+    return 128;
+  if (Info.RegClass == X86::VR256RegClassID ||
+      Info.RegClass == X86::VR256XRegClassID)
+    return 256;
+  if (Info.RegClass == X86::VR512RegClassID)
+    return 512;
+  llvm_unreachable("Unknown register class!");
+}
+
 void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
   X86MCInstLower MCInstLowering(*MF, *this);
   const X86RegisterInfo *RI =
@@ -1879,8 +1891,9 @@ void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
 
     const MachineOperand &MaskOp = MI->getOperand(MaskIdx);
     if (auto *C = getConstantFromPool(*MI, MaskOp)) {
+      unsigned Width = getRegisterWidth(MI->getDesc().OpInfo[0]);
       SmallVector<int, 64> Mask;
-      DecodePSHUFBMask(C, Mask);
+      DecodePSHUFBMask(C, Width, Mask);
       if (!Mask.empty())
         OutStreamer->AddComment(getShuffleComment(MI, SrcIdx, SrcIdx, Mask),
                                 !EnablePrintSchedInfo);
@@ -1951,8 +1964,9 @@ void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
 
     const MachineOperand &MaskOp = MI->getOperand(MaskIdx);
     if (auto *C = getConstantFromPool(*MI, MaskOp)) {
+      unsigned Width = getRegisterWidth(MI->getDesc().OpInfo[0]);
       SmallVector<int, 16> Mask;
-      DecodeVPERMILPMask(C, ElSize, Mask);
+      DecodeVPERMILPMask(C, ElSize, Width, Mask);
       if (!Mask.empty())
         OutStreamer->AddComment(getShuffleComment(MI, SrcIdx, SrcIdx, Mask),
                                 !EnablePrintSchedInfo);
@@ -1982,8 +1996,9 @@ void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
 
     const MachineOperand &MaskOp = MI->getOperand(6);
     if (auto *C = getConstantFromPool(*MI, MaskOp)) {
+      unsigned Width = getRegisterWidth(MI->getDesc().OpInfo[0]);
       SmallVector<int, 16> Mask;
-      DecodeVPERMIL2PMask(C, (unsigned)CtrlOp.getImm(), ElSize, Mask);
+      DecodeVPERMIL2PMask(C, (unsigned)CtrlOp.getImm(), ElSize, Width, Mask);
       if (!Mask.empty())
         OutStreamer->AddComment(getShuffleComment(MI, 1, 2, Mask),
                                 !EnablePrintSchedInfo);
@@ -1999,8 +2014,9 @@ void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
 
     const MachineOperand &MaskOp = MI->getOperand(6);
     if (auto *C = getConstantFromPool(*MI, MaskOp)) {
+      unsigned Width = getRegisterWidth(MI->getDesc().OpInfo[0]);
       SmallVector<int, 16> Mask;
-      DecodeVPPERMMask(C, Mask);
+      DecodeVPPERMMask(C, Width, Mask);
       if (!Mask.empty())
         OutStreamer->AddComment(getShuffleComment(MI, 1, 2, Mask),
                                 !EnablePrintSchedInfo);
diff --git a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
index c7ddf93f8e8..720be8afa62 100644
--- a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
+++ b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
@@ -112,11 +112,10 @@ static bool extractConstantMask(const Constant *C, unsigned MaskEltSizeInBits,
   return true;
 }
 
-void DecodePSHUFBMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask) {
-  Type *MaskTy = C->getType();
-  unsigned MaskTySize = MaskTy->getPrimitiveSizeInBits();
-  (void)MaskTySize;
-  assert((MaskTySize == 128 || MaskTySize == 256 || MaskTySize == 512) &&
+void DecodePSHUFBMask(const Constant *C, unsigned Width,
+                      SmallVectorImpl<int> &ShuffleMask) {
+  assert((Width == 128 || Width == 256 || Width == 512) &&
+         C->getType()->getPrimitiveSizeInBits() >= Width &&
          "Unexpected vector size.");
 
   // The shuffle mask requires a byte vector.
@@ -125,7 +124,7 @@ void DecodePSHUFBMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask) {
   if (!extractConstantMask(C, 8, UndefElts, RawMask))
     return;
 
-  unsigned NumElts = RawMask.size();
+  unsigned NumElts = Width / 8;
   assert((NumElts == 16 || NumElts == 32 || NumElts == 64) &&
          "Unexpected number of vector elements.");
 
@@ -151,12 +150,10 @@ void DecodePSHUFBMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask) {
   }
 }
 
-void DecodeVPERMILPMask(const Constant *C, unsigned ElSize,
+void DecodeVPERMILPMask(const Constant *C, unsigned ElSize, unsigned Width,
                         SmallVectorImpl<int> &ShuffleMask) {
-  Type *MaskTy = C->getType();
-  unsigned MaskTySize = MaskTy->getPrimitiveSizeInBits();
-  (void)MaskTySize;
-  assert((MaskTySize == 128 || MaskTySize == 256 || MaskTySize == 512) &&
+  assert((Width == 128 || Width == 256 || Width == 512) &&
+         C->getType()->getPrimitiveSizeInBits() >= Width &&
          "Unexpected vector size.");
   assert((ElSize == 32 || ElSize == 64) && "Unexpected vector element size.");
 
@@ -166,7 +163,7 @@ void DecodeVPERMILPMask(const Constant *C, unsigned ElSize,
   if (!extractConstantMask(C, ElSize, UndefElts, RawMask))
     return;
 
-  unsigned NumElts = RawMask.size();
+  unsigned NumElts = Width / ElSize;
   unsigned NumEltsPerLane = 128 / ElSize;
   assert((NumElts == 2 || NumElts == 4 || NumElts == 8 || NumElts == 16) &&
          "Unexpected number of vector elements.");
@@ -189,11 +186,13 @@ void DecodeVPERMILPMask(const Constant *C, unsigned ElSize,
 }
 
 void DecodeVPERMIL2PMask(const Constant *C, unsigned M2Z, unsigned ElSize,
+                         unsigned Width,
                          SmallVectorImpl<int> &ShuffleMask) {
   Type *MaskTy = C->getType();
   unsigned MaskTySize = MaskTy->getPrimitiveSizeInBits();
   (void)MaskTySize;
-  assert((MaskTySize == 128 || MaskTySize == 256) && "Unexpected vector size.");
+  assert((MaskTySize == 128 || MaskTySize == 256) &&
+         Width >= MaskTySize && "Unexpected vector size.");
 
   // The shuffle mask requires elements the same size as the target.
   APInt UndefElts;
@@ -201,7 +200,7 @@ void DecodeVPERMIL2PMask(const Constant *C, unsigned M2Z, unsigned ElSize,
   if (!extractConstantMask(C, ElSize, UndefElts, RawMask))
     return;
 
-  unsigned NumElts = RawMask.size();
+  unsigned NumElts = Width / ElSize;
   unsigned NumEltsPerLane = 128 / ElSize;
   assert((NumElts == 2 || NumElts == 4 || NumElts == 8) &&
          "Unexpected number of vector elements.");
@@ -242,9 +241,12 @@ void DecodeVPERMIL2PMask(const Constant *C, unsigned M2Z, unsigned ElSize,
   }
 }
 
-void DecodeVPPERMMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask) {
-  assert(C->getType()->getPrimitiveSizeInBits() == 128 &&
-         "Unexpected vector size.");
+void DecodeVPPERMMask(const Constant *C, unsigned Width,
+                      SmallVectorImpl<int> &ShuffleMask) {
+  Type *MaskTy = C->getType();
+  unsigned MaskTySize = MaskTy->getPrimitiveSizeInBits();
+  (void)MaskTySize;
+  assert(Width == 128 && Width >= MaskTySize && "Unexpected vector size.");
 
   // The shuffle mask requires a byte vector.
   APInt UndefElts;
@@ -252,7 +254,7 @@ void DecodeVPPERMMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask) {
   if (!extractConstantMask(C, 8, UndefElts, RawMask))
     return;
 
-  unsigned NumElts = RawMask.size();
+  unsigned NumElts = Width / 8;
   assert(NumElts == 16 && "Unexpected number of vector elements.");
 
   for (unsigned i = 0; i != NumElts; ++i) {
@@ -291,12 +293,10 @@ void DecodeVPPERMMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask) {
   }
 }
 
-void DecodeVPERMVMask(const Constant *C, unsigned ElSize,
+void DecodeVPERMVMask(const Constant *C, unsigned ElSize, unsigned Width,
                       SmallVectorImpl<int> &ShuffleMask) {
-  Type *MaskTy = C->getType();
-  unsigned MaskTySize = MaskTy->getPrimitiveSizeInBits();
-  (void)MaskTySize;
-  assert((MaskTySize == 128 || MaskTySize == 256 || MaskTySize == 512) &&
+  assert((Width == 128 || Width == 256 || Width == 512) &&
+         C->getType()->getPrimitiveSizeInBits() >= Width &&
          "Unexpected vector size.");
   assert((ElSize == 8 || ElSize == 16 || ElSize == 32 || ElSize == 64) &&
          "Unexpected vector element size.");
@@ -307,7 +307,7 @@ void DecodeVPERMVMask(const Constant *C, unsigned ElSize,
   if (!extractConstantMask(C, ElSize, UndefElts, RawMask))
     return;
 
-  unsigned NumElts = RawMask.size();
+  unsigned NumElts = Width / ElSize;
 
   for (unsigned i = 0; i != NumElts; ++i) {
     if (UndefElts[i]) {
@@ -319,12 +319,10 @@ void DecodeVPERMVMask(const Constant *C, unsigned ElSize,
   }
 }
 
-void DecodeVPERMV3Mask(const Constant *C, unsigned ElSize,
+void DecodeVPERMV3Mask(const Constant *C, unsigned ElSize, unsigned Width,
                        SmallVectorImpl<int> &ShuffleMask) {
-  Type *MaskTy = C->getType();
-  unsigned MaskTySize = MaskTy->getPrimitiveSizeInBits();
-  (void)MaskTySize;
-  assert((MaskTySize == 128 || MaskTySize == 256 || MaskTySize == 512) &&
+  assert((Width == 128 || Width == 256 || Width == 512) &&
+         C->getType()->getPrimitiveSizeInBits() >= Width &&
          "Unexpected vector size.");
   assert((ElSize == 8 || ElSize == 16 || ElSize == 32 || ElSize == 64) &&
          "Unexpected vector element size.");
@@ -335,7 +333,7 @@ void DecodeVPERMV3Mask(const Constant *C, unsigned ElSize,
   if (!extractConstantMask(C, ElSize, UndefElts, RawMask))
     return;
 
-  unsigned NumElts = RawMask.size();
+  unsigned NumElts = Width / ElSize;
 
   for (unsigned i = 0; i != NumElts; ++i) {
     if (UndefElts[i]) {
diff --git a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.h b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.h
index b703cbbd2b2..b08c31935d2 100644
--- a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.h
+++ b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.h
@@ -26,25 +26,28 @@ class Constant;
 class MVT;
 
 /// Decode a PSHUFB mask from an IR-level vector constant.
-void DecodePSHUFBMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask);
+void DecodePSHUFBMask(const Constant *C, unsigned Width,
+                      SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a VPERMILP variable mask from an IR-level vector constant.
-void DecodeVPERMILPMask(const Constant *C, unsigned ElSize,
+void DecodeVPERMILPMask(const Constant *C, unsigned ElSize, unsigned Width,
                         SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a VPERMILP2 variable mask from an IR-level vector constant.
 void DecodeVPERMIL2PMask(const Constant *C, unsigned MatchImm, unsigned ElSize,
+                         unsigned Width,
                          SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a VPPERM variable mask from an IR-level vector constant.
-void DecodeVPPERMMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask);
+void DecodeVPPERMMask(const Constant *C, unsigned Width,
+                      SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a VPERM W/D/Q/PS/PD mask from an IR-level vector constant.
-void DecodeVPERMVMask(const Constant *C, unsigned ElSize,
+void DecodeVPERMVMask(const Constant *C, unsigned ElSize, unsigned Width,
                       SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a VPERMT2 W/D/Q/PS/PD mask from an IR-level vector constant.
-void DecodeVPERMV3Mask(const Constant *C, unsigned ElSize,
+void DecodeVPERMV3Mask(const Constant *C, unsigned ElSize, unsigned Width,
                        SmallVectorImpl<int> &ShuffleMask);
 
 } // llvm namespace