[AMDGPU] Assembler: Custom converters for SDWA instructions. Support for _dpp and _sdwa suffixes in mnemonics.

[SamWot]

Summary

Added custom converters for SDWA instruction to support optional operands and modifiers.
Support for _dpp and _sdwa suffixes that allows to force DPP or SDWA encoding for instructions.

[arsenm]

lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

632

Should take StringRef and return new StringRef

2554

enum not necessary

2676–2681

Can you move around the IsVOP1 check to avoid having IsVOP1 and !IsVOP1 conditions

[SamWot]

Changes for arsenm review

[artem.tamazov]

Minor fixes required, otherwise fine.

test/MC/AMDGPU/vop_sdwa.s

14–52

Normally we put CHECK-lines *after* source lines.

[SamWot]

test/MC/AMDGPU/vop_sdwa.s

14–52

In all disassembler tests CHECK-lines are before source lines. Same for some assembler tests (vop1.s, vop2.s).

[SamWot]

Matt,
Are there any more issues?

lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

enum ImmTy {
     ImmTySLC,
     ImmTyTFE,
     ImmTyClampSI,
     ImmTyOModSI,
     ImmTyDppCtrl,
     ImmTyDppRowMask,
     ImmTyDppBankMask,
     ImmTyDppBoundCtrl,
-    ImmTySdwaSel,
+    ImmTySdwaDstSel,
+    ImmTySdwaSrc0Sel,
+    ImmTySdwaSrc1Sel,
     ImmTySdwaDstUnused,
     ImmTyDMask,
     ImmTyUNorm,
     ImmTyDA,
     ImmTyR128,
     ImmTyLWE,
     ImmTyHwreg,
     ImmTySendMsg,
public:
   bool isRowMask() const {
     return isImmTy(ImmTyDppRowMask);
   }
 
   bool isBoundCtrl() const {
     return isImmTy(ImmTyDppBoundCtrl);
   }
 
-  bool isSDWASel() const {
-    return isImmTy(ImmTySdwaSel);
+  bool isSDWADstSel() const {
+    return isImmTy(ImmTySdwaDstSel);
+  }
+
+  bool isSDWASrc0Sel() const {
+    return isImmTy(ImmTySdwaSrc0Sel);
+  }
+
+  bool isSDWASrc1Sel() const {
+    return isImmTy(ImmTySdwaSrc1Sel);
   }
 
   bool isSDWADstUnused() const {
     return isImmTy(ImmTySdwaDstUnused);
   }
 
   void setModifiers(unsigned Mods) {
     assert(isReg() || (isImm() && Imm.Modifiers == 0));
void printImmTy(raw_ostream& OS, ImmTy Type) const {
     case ImmTySLC: OS << "SLC"; break;
     case ImmTyTFE: OS << "TFE"; break;
     case ImmTyClampSI: OS << "ClampSI"; break;
     case ImmTyOModSI: OS << "OModSI"; break;
     case ImmTyDppCtrl: OS << "DppCtrl"; break;
     case ImmTyDppRowMask: OS << "DppRowMask"; break;
     case ImmTyDppBankMask: OS << "DppBankMask"; break;
     case ImmTyDppBoundCtrl: OS << "DppBoundCtrl"; break;
-    case ImmTySdwaSel: OS << "SdwaSel"; break;
+    case ImmTySdwaDstSel: OS << "SdwaDstSel"; break;
+    case ImmTySdwaSrc0Sel: OS << "SdwaSrc0Sel"; break;
+    case ImmTySdwaSrc1Sel: OS << "SdwaSrc1Sel"; break;
     case ImmTySdwaDstUnused: OS << "SdwaDstUnused"; break;
     case ImmTyDMask: OS << "DMask"; break;
     case ImmTyUNorm: OS << "UNorm"; break;
     case ImmTyDA: OS << "DA"; break;
     case ImmTyR128: OS << "R128"; break;
     case ImmTyLWE: OS << "LWE"; break;
     case ImmTyHwreg: OS << "Hwreg"; break;
     case ImmTySendMsg: OS << "SendMsg"; break;
public:
   bool isDPPCtrl() const;
 };
 
 class AMDGPUAsmParser : public MCTargetAsmParser {
   const MCInstrInfo &MII;
   MCAsmParser &Parser;
 
   unsigned ForcedEncodingSize;
+  bool ForcedDPP;
+  bool ForcedSDWA;
 
   bool isSI() const {
     return AMDGPU::isSI(getSTI());
   }
 
   bool isCI() const {
     return AMDGPU::isCI(getSTI());
   }
public:
   enum AMDGPUMatchResultTy {
     Match_PreferE32 = FIRST_TARGET_MATCH_RESULT_TY
   };
 
   AMDGPUAsmParser(const MCSubtargetInfo &STI, MCAsmParser &_Parser,
                const MCInstrInfo &MII,
                const MCTargetOptions &Options)
       : MCTargetAsmParser(Options, STI), MII(MII), Parser(_Parser),
-        ForcedEncodingSize(0) {
+        ForcedEncodingSize(0),
+        ForcedDPP(false),
+        ForcedSDWA(false) {
     MCAsmParserExtension::Initialize(Parser);
 
     if (getSTI().getFeatureBits().none()) {
       // Set default features.
       copySTI().ToggleFeature("SOUTHERN_ISLANDS");
     }
 
     setAvailableFeatures(ComputeAvailableFeatures(getSTI().getFeatureBits()));
   }
 
   AMDGPUTargetStreamer &getTargetStreamer() {
     MCTargetStreamer &TS = *getParser().getStreamer().getTargetStreamer();
     return static_cast<AMDGPUTargetStreamer &>(TS);
   }
   
-  unsigned getForcedEncodingSize() const {
-    return ForcedEncodingSize;
-  }
-
-  void setForcedEncodingSize(unsigned Size) {
-    ForcedEncodingSize = Size;
-  }
-
-  bool isForcedVOP3() const {
-    return ForcedEncodingSize == 64;
-  }
+  void setForcedEncodingSize(unsigned Size) { ForcedEncodingSize = Size; }
+  void setForcedDPP(bool ForceDPP_) { ForcedDPP = ForceDPP_; }
+  void setForcedSDWA(bool ForceSDWA_) { ForcedSDWA = ForceSDWA_; }
+
+  unsigned getForcedEncodingSize() const { return ForcedEncodingSize; }
+  bool isForcedVOP3() const { return ForcedEncodingSize == 64; }
+  bool isForcedDPP() const { return ForcedDPP; }
+  bool isForcedSDWA() const { return ForcedSDWA; }
 
   std::unique_ptr<AMDGPUOperand> parseRegister();
   bool ParseRegister(unsigned &RegNo, SMLoc &StartLoc, SMLoc &EndLoc) override;
   unsigned checkTargetMatchPredicate(MCInst &Inst) override;
   unsigned validateTargetOperandClass(MCParsedAsmOperand &Op,
                                       unsigned Kind) override;
   bool MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
                                OperandVector &Operands, MCStreamer &Out,
                                uint64_t &ErrorInfo,
                                bool MatchingInlineAsm) override;
   bool ParseDirective(AsmToken DirectiveID) override;
   OperandMatchResultTy parseOperand(OperandVector &Operands, StringRef Mnemonic);
+  void parseMnemonicSuffix(StringRef& Name);
   bool ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
                         SMLoc NameLoc, OperandVector &Operands) override;
 
   OperandMatchResultTy parseIntWithPrefix(const char *Prefix, int64_t &Int);
   OperandMatchResultTy parseIntWithPrefix(const char *Prefix,
                                           OperandVector &Operands,
                                           enum AMDGPUOperand::ImmTy ImmTy = AMDGPUOperand::ImmTyNone,
                                           bool (*ConvertResult)(int64_t&) = 0);
   OperandMatchResultTy parseNamedBit(const char *Name, OperandVector &Operands,
                                      enum AMDGPUOperand::ImmTy ImmTy = AMDGPUOperand::ImmTyNone);
-  OperandMatchResultTy parseStringWithPrefix(const char *Prefix, StringRef &Value);
+  OperandMatchResultTy parseStringWithPrefix(StringRef Prefix, StringRef &Value);
 
   OperandMatchResultTy parseImm(OperandVector &Operands);
   OperandMatchResultTy parseRegOrImm(OperandVector &Operands);
   OperandMatchResultTy parseRegOrImmWithInputMods(OperandVector &Operands);
 
   void cvtDSOffset01(MCInst &Inst, const OperandVector &Operands);
   void cvtDS(MCInst &Inst, const OperandVector &Operands);
 
public:
   void cvtMIMGAtomic(MCInst &Inst, const OperandVector &Operands);
 
   OperandMatchResultTy parseDPPCtrl(OperandVector &Operands);
   AMDGPUOperand::Ptr defaultRowMask() const;
   AMDGPUOperand::Ptr defaultBankMask() const;
   AMDGPUOperand::Ptr defaultBoundCtrl() const;
   void cvtDPP(MCInst &Inst, const OperandVector &Operands);
 
-  OperandMatchResultTy parseSDWASel(OperandVector &Operands);
+  OperandMatchResultTy parseSDWASel(OperandVector &Operands, StringRef Prefix,
+                                    enum AMDGPUOperand::ImmTy Type);
   OperandMatchResultTy parseSDWADstUnused(OperandVector &Operands);
-  AMDGPUOperand::Ptr defaultSDWASel() const;
+  AMDGPUOperand::Ptr defaultSDWASel(enum AMDGPUOperand::ImmTy Type) const;
+  AMDGPUOperand::Ptr defaultSDWADstSel() const;
+  AMDGPUOperand::Ptr defaultSDWASrc0Sel() const;
+  AMDGPUOperand::Ptr defaultSDWASrc1Sel() const;
   AMDGPUOperand::Ptr defaultSDWADstUnused() const;
+  void cvtSdwaVop1_mod(MCInst &Inst, const OperandVector &Operands);
+  void cvtSdwaVop1_nomod(MCInst &Inst, const OperandVector &Operands);
+  void cvtSdwaVop2_mod(MCInst &Inst, const OperandVector &Operands);
+  void cvtSdwaVop2_nomod(MCInst &Inst, const OperandVector &Operands);
+  void cvtSDWA(MCInst &Inst, const OperandVector &Operands, bool HasMods,
+               bool IsVOP1);
 };
 
 struct OptionalOperand {
   const char *Name;
   AMDGPUOperand::ImmTy Type;
   bool IsBit;
   bool (*ConvertResult)(int64_t&);
 };

 }
 
 
 unsigned AMDGPUAsmParser::checkTargetMatchPredicate(MCInst &Inst) {
 
   uint64_t TSFlags = MII.get(Inst.getOpcode()).TSFlags;
 
   if ((getForcedEncodingSize() == 32 && (TSFlags & SIInstrFlags::VOP3)) ||
-      (getForcedEncodingSize() == 64 && !(TSFlags & SIInstrFlags::VOP3)))
+      (getForcedEncodingSize() == 64 && !(TSFlags & SIInstrFlags::VOP3)) ||
+      (isForcedDPP() && !(TSFlags & SIInstrFlags::DPP)) ||
+      (isForcedSDWA() && !(TSFlags & SIInstrFlags::SDWA)) )
     return Match_InvalidOperand;
 
   if ((TSFlags & SIInstrFlags::VOP3) &&
       (TSFlags & SIInstrFlags::VOPAsmPrefer32Bit) &&
       getForcedEncodingSize() != 64)
     return Match_PreferE32;
 
   return Match_Success;
if (getLexer().getKind() == AsmToken::Identifier) {
     const auto &Tok = Parser.getTok();
     Operands.push_back(AMDGPUOperand::CreateToken(Tok.getString(), Tok.getLoc()));
     Parser.Lex();
     return MatchOperand_Success;
   }
   return MatchOperand_NoMatch;
 }
 
-bool AMDGPUAsmParser::ParseInstruction(ParseInstructionInfo &Info,
-                                       StringRef Name,
-                                       SMLoc NameLoc, OperandVector &Operands) {
-
+void AMDGPUAsmParser::parseMnemonicSuffix(StringRef& Name) {
   // Clear any forced encodings from the previous instruction.
   setForcedEncodingSize(0);
+  setForcedDPP(false);
+  setForcedSDWA(false);
 
-  if (Name.endswith("_e64"))
+  if (Name.endswith("_e64")) {
     setForcedEncodingSize(64);
-  else if (Name.endswith("_e32"))
+    Name = Name.substr(0, Name.size() - 4);
+  } else if (Name.endswith("_e32")) {
     setForcedEncodingSize(32);
+    Name = Name.substr(0, Name.size() - 4);
+  } else if (Name.endswith("_dpp")) {
+    setForcedDPP(true);
+    Name = Name.substr(0, Name.size() - 4);
+  } else if (Name.endswith("_sdwa")) {
+    setForcedSDWA(true);
+    Name = Name.substr(0, Name.size() - 5);
+  }
+}
 
+bool AMDGPUAsmParser::ParseInstruction(ParseInstructionInfo &Info,
+                                       StringRef Name,
+                                       SMLoc NameLoc, OperandVector &Operands) {
   // Add the instruction mnemonic
+  parseMnemonicSuffix(Name);
   Operands.push_back(AMDGPUOperand::CreateToken(Name, NameLoc));
     
-
-  if (Name.endswith("_e64")) { Name = Name.substr(0, Name.size() - 4); }
-  if (Name.endswith("_e32")) { Name = Name.substr(0, Name.size() - 4); }
-
   while (!getLexer().is(AsmToken::EndOfStatement)) {
     AMDGPUAsmParser::OperandMatchResultTy Res = parseOperand(Operands, Name);
 
     // Eat the comma or space if there is one.
     if (getLexer().is(AsmToken::Comma))
       Parser.Lex();
     
     switch (Res) {
if (i != OptionalIdx.end()) {
     unsigned Idx = i->second;
     ((AMDGPUOperand &)*Operands[Idx]).addImmOperands(Inst, 1);
   } else {
     Inst.addOperand(MCOperand::createImm(Default));
   }
 }
 
 AMDGPUAsmParser::OperandMatchResultTy 
-AMDGPUAsmParser::parseStringWithPrefix(const char *Prefix, StringRef &Value) {
+AMDGPUAsmParser::parseStringWithPrefix(StringRef Prefix, StringRef &Value) {
   if (getLexer().isNot(AsmToken::Identifier)) {
     return MatchOperand_NoMatch;
   }
   StringRef Tok = Parser.getTok().getString();
   if (Tok != Prefix) {
     return MatchOperand_NoMatch;
   }
 
static const OptionalOperand AMDGPUOptionalOperandTable[] = {
   {"unorm",   AMDGPUOperand::ImmTyUNorm, true, nullptr},
   {"da",      AMDGPUOperand::ImmTyDA,    true, nullptr},
   {"r128",    AMDGPUOperand::ImmTyR128,  true, nullptr},
   {"lwe",     AMDGPUOperand::ImmTyLWE,   true, nullptr},
   {"dmask",   AMDGPUOperand::ImmTyDMask, false, nullptr},
   {"row_mask",   AMDGPUOperand::ImmTyDppRowMask, false, nullptr},
   {"bank_mask",  AMDGPUOperand::ImmTyDppBankMask, false, nullptr},
   {"bound_ctrl", AMDGPUOperand::ImmTyDppBoundCtrl, false, ConvertBoundCtrl},
-  {"sdwa_sel",   AMDGPUOperand::ImmTySdwaSel, false, nullptr},
+  {"dst_sel",    AMDGPUOperand::ImmTySdwaDstSel, false, nullptr},
+  {"src0_sel",   AMDGPUOperand::ImmTySdwaSrc0Sel, false, nullptr},
+  {"src1_sel",   AMDGPUOperand::ImmTySdwaSrc1Sel, false, nullptr},
   {"dst_unused", AMDGPUOperand::ImmTySdwaDstUnused, false, nullptr},
 };
 
 AMDGPUAsmParser::OperandMatchResultTy AMDGPUAsmParser::parseOptionalOperand(OperandVector &Operands) {
   OperandMatchResultTy res;
   for (const OptionalOperand &Op : AMDGPUOptionalOperandTable) {
     // try to parse any optional operand here
     if (Op.IsBit) {
       res = parseNamedBit(Op.Name, Operands, Op.Type);
     } else if (Op.Type == AMDGPUOperand::ImmTyOModSI) {
       res = parseOModOperand(Operands);
-    } else if (Op.Type == AMDGPUOperand::ImmTySdwaSel) {
-      res = parseSDWASel(Operands);
+    } else if (Op.Type == AMDGPUOperand::ImmTySdwaDstSel ||
+               Op.Type == AMDGPUOperand::ImmTySdwaSrc0Sel ||
+               Op.Type == AMDGPUOperand::ImmTySdwaSrc1Sel) {
+      res = parseSDWASel(Operands, Op.Name, Op.Type);
     } else if (Op.Type == AMDGPUOperand::ImmTySdwaDstUnused) {
       res = parseSDWADstUnused(Operands);
     } else {
       res = parseIntWithPrefix(Op.Name, Operands, Op.Type, Op.ConvertResult);
     }
     if (res != MatchOperand_NoMatch) {
       return res;
     }
for (unsigned E = Operands.size(); I != E; ++I) {
     } else if (Op.isImm()) {
       // Handle optional arguments
       OptionalIdx[Op.getImmTy()] = I;
     } else {
       llvm_unreachable("Invalid operand type");
     }
   }
 
-  // ToDo: fix default values for row_mask and bank_mask
   addOptionalImmOperand(Inst, Operands, OptionalIdx, AMDGPUOperand::ImmTyDppRowMask, 0xf);
   addOptionalImmOperand(Inst, Operands, OptionalIdx, AMDGPUOperand::ImmTyDppBankMask, 0xf);
   addOptionalImmOperand(Inst, Operands, OptionalIdx, AMDGPUOperand::ImmTyDppBoundCtrl);
 }
 
 //===----------------------------------------------------------------------===//
 // sdwa
 //===----------------------------------------------------------------------===//
 
 AMDGPUAsmParser::OperandMatchResultTy
-AMDGPUAsmParser::parseSDWASel(OperandVector &Operands) {
+AMDGPUAsmParser::parseSDWASel(OperandVector &Operands, StringRef Prefix,
+                              enum AMDGPUOperand::ImmTy Type) {
   SMLoc S = Parser.getTok().getLoc();
   StringRef Value;
   AMDGPUAsmParser::OperandMatchResultTy res;
   
-  res = parseStringWithPrefix("dst_sel", Value);
-  if (res == MatchOperand_ParseFail) {
-    return MatchOperand_ParseFail;
-  } else if (res == MatchOperand_NoMatch) {
-    res = parseStringWithPrefix("src0_sel", Value);
-    if (res == MatchOperand_ParseFail) {
-      return MatchOperand_ParseFail;
-    } else if (res == MatchOperand_NoMatch) {
-      res = parseStringWithPrefix("src1_sel", Value);
-      if (res != MatchOperand_Success) {
-        return res;
-      }
-    }
-  }
+  res = parseStringWithPrefix(Prefix, Value);
   
   int64_t Int;
   Int = StringSwitch<int64_t>(Value)
         .Case("BYTE_0", 0)
         .Case("BYTE_1", 1)
         .Case("BYTE_2", 2)
         .Case("BYTE_3", 3)
         .Case("WORD_0", 4)
         .Case("WORD_1", 5)
         .Case("DWORD", 6)
         .Default(0xffffffff);
   Parser.Lex(); // eat last token
 
   if (Int == 0xffffffff) {
     return MatchOperand_ParseFail;
   }
 
-  Operands.push_back(AMDGPUOperand::CreateImm(Int, S,
-                                              AMDGPUOperand::ImmTySdwaSel));
+  Operands.push_back(AMDGPUOperand::CreateImm(Int, S, Type));
   return MatchOperand_Success;
 }
 
 AMDGPUAsmParser::OperandMatchResultTy 
 AMDGPUAsmParser::parseSDWADstUnused(OperandVector &Operands) {
   SMLoc S = Parser.getTok().getLoc();
   StringRef Value;
   AMDGPUAsmParser::OperandMatchResultTy res;
if (Int == 0xffffffff) {
     return MatchOperand_ParseFail;
   }
 
   Operands.push_back(AMDGPUOperand::CreateImm(Int, S,
                                               AMDGPUOperand::ImmTySdwaDstUnused));
   return MatchOperand_Success;
 }
 
-AMDGPUOperand::Ptr AMDGPUAsmParser::defaultSDWASel() const {
-  return AMDGPUOperand::CreateImm(6, SMLoc(), AMDGPUOperand::ImmTySdwaSel);
+AMDGPUOperand::Ptr AMDGPUAsmParser::defaultSDWASel(enum AMDGPUOperand::ImmTy Type) const {
+  return AMDGPUOperand::CreateImm(6, SMLoc(), Type);
+}
+
+AMDGPUOperand::Ptr AMDGPUAsmParser::defaultSDWADstSel() const {
+  return defaultSDWASel(AMDGPUOperand::ImmTySdwaDstSel);
+}
+
+AMDGPUOperand::Ptr AMDGPUAsmParser::defaultSDWASrc0Sel() const {
+  return defaultSDWASel(AMDGPUOperand::ImmTySdwaSrc0Sel);
+}
+
+AMDGPUOperand::Ptr AMDGPUAsmParser::defaultSDWASrc1Sel() const {
+  return defaultSDWASel(AMDGPUOperand::ImmTySdwaSrc1Sel);
 }
 
 AMDGPUOperand::Ptr AMDGPUAsmParser::defaultSDWADstUnused() const {
   return AMDGPUOperand::CreateImm(0, SMLoc(), AMDGPUOperand::ImmTySdwaDstUnused);
 }
 
+void AMDGPUAsmParser::cvtSdwaVop1_mod(MCInst &Inst, const OperandVector &Operands) {
+  cvtSDWA(Inst, Operands, true, true);
+}
+
+void AMDGPUAsmParser::cvtSdwaVop1_nomod(MCInst &Inst, const OperandVector &Operands) {
+  cvtSDWA(Inst, Operands, false, true);
+}
+
+void AMDGPUAsmParser::cvtSdwaVop2_mod(MCInst &Inst, const OperandVector &Operands) {
+  cvtSDWA(Inst, Operands, true, false);
+}
+
+void AMDGPUAsmParser::cvtSdwaVop2_nomod(MCInst &Inst, const OperandVector &Operands) {
+  cvtSDWA(Inst, Operands, false, false);
+}
+
+void AMDGPUAsmParser::cvtSDWA(MCInst &Inst, const OperandVector &Operands,
+                              bool HasMods, bool IsVOP1) {
+  OptionalImmIndexMap OptionalIdx;
+
+  unsigned I = 1;
+  const MCInstrDesc &Desc = MII.get(Inst.getOpcode());
+  for (unsigned J = 0; J < Desc.getNumDefs(); ++J) {
+    ((AMDGPUOperand &)*Operands[I++]).addRegOperands(Inst, 1);
+  }
+
+  for (unsigned E = Operands.size(); I != E; ++I) {
+    AMDGPUOperand &Op = ((AMDGPUOperand &)*Operands[I]);
+    // Add the register arguments
+    if (!HasMods && Op.isReg()) {
+      Op.addRegOperands(Inst, 1);
+    } else if (HasMods && Op.isRegOrImmWithInputMods()) {
+      Op.addRegOrImmWithInputModsOperands(Inst, 2);
+    } else if (Op.isImm()) {
+      // Handle optional arguments
+      OptionalIdx[Op.getImmTy()] = I;
+    } else {
+      llvm_unreachable("Invalid operand type");
+    }
+  }
+
+  if (HasMods) {
+    addOptionalImmOperand(Inst, Operands, OptionalIdx, AMDGPUOperand::ImmTyClampSI, 0);
+  }
+  if (IsVOP1 &&
+      Inst.getOpcode() != AMDGPU::V_NOP_sdwa) { // V_NOP_sdwa has no optional sdwa arguments
+    addOptionalImmOperand(Inst, Operands, OptionalIdx, AMDGPUOperand::ImmTySdwaDstSel, 6);
+    addOptionalImmOperand(Inst, Operands, OptionalIdx, AMDGPUOperand::ImmTySdwaDstUnused, 2);
+    addOptionalImmOperand(Inst, Operands, OptionalIdx, AMDGPUOperand::ImmTySdwaSrc0Sel, 6);
+  } else if (!IsVOP1) { // VOP2
+    addOptionalImmOperand(Inst, Operands, OptionalIdx, AMDGPUOperand::ImmTySdwaDstSel, 6);
+    addOptionalImmOperand(Inst, Operands, OptionalIdx, AMDGPUOperand::ImmTySdwaDstUnused, 2);
+    addOptionalImmOperand(Inst, Operands, OptionalIdx, AMDGPUOperand::ImmTySdwaSrc0Sel, 6);
+    addOptionalImmOperand(Inst, Operands, OptionalIdx, AMDGPUOperand::ImmTySdwaSrc1Sel, 6);
+  }
+}
 
 /// Force static initialization.
 extern "C" void LLVMInitializeAMDGPUAsmParser() {
   RegisterMCAsmParser<AMDGPUAsmParser> A(TheAMDGPUTarget);
   RegisterMCAsmParser<AMDGPUAsmParser> B(TheGCNTarget);
 }
 
 #define GET_REGISTER_MATCHER

lib/Target/AMDGPU/SIInstrInfo.td

 include "VIInstrFormats.td"
 
 class NamedMatchClass<string CName, bit Optional = 1> : AsmOperandClass {
   let Name = "Imm"#CName;
   let PredicateMethod = "is"#CName;
   let ParserMethod = !if(Optional, "parseOptionalOperand", "parse"#CName);
   let RenderMethod = "addImmOperands";
   let IsOptional = Optional;
-  let DefaultMethod = "default"#CName;
+  let DefaultMethod = !if(Optional, "default"#CName, ?);
 }
 
-def sdwa_sel : NamedMatchClass<"SDWASel">;
-
 class NamedOperandBit<string Name, AsmOperandClass MatchClass> : Operand<i1> {
   let PrintMethod = "print"#Name;
   let ParserMatchClass = MatchClass;
 }
  
 class NamedOperandU8<string Name, AsmOperandClass MatchClass> : Operand<i8> {
   let PrintMethod = "print"#Name;
   let ParserMatchClass = MatchClass;

 
 def dmask : NamedOperandU16<"DMask", NamedMatchClass<"DMask">>;
 
 def dpp_ctrl : NamedOperandU32<"DPPCtrl", NamedMatchClass<"DPPCtrl", 0>>;
 def row_mask : NamedOperandU32<"RowMask", NamedMatchClass<"RowMask">>;
 def bank_mask : NamedOperandU32<"BankMask", NamedMatchClass<"BankMask">>;
 def bound_ctrl : NamedOperandBit<"BoundCtrl", NamedMatchClass<"BoundCtrl">>;
 
-def dst_sel : NamedOperandU32<"SDWADstSel", sdwa_sel>;
-def src0_sel : NamedOperandU32<"SDWASrc0Sel", sdwa_sel>;
-def src1_sel : NamedOperandU32<"SDWASrc1Sel", sdwa_sel>;
+def dst_sel : NamedOperandU32<"SDWADstSel", NamedMatchClass<"SDWADstSel">>;
+def src0_sel : NamedOperandU32<"SDWASrc0Sel", NamedMatchClass<"SDWASrc0Sel">>;
+def src1_sel : NamedOperandU32<"SDWASrc1Sel", NamedMatchClass<"SDWASrc1Sel">>;
 def dst_unused : NamedOperandU32<"SDWADstUnused", NamedMatchClass<"SDWADstUnused">>;
 
 def hwreg : NamedOperandU16<"Hwreg", NamedMatchClass<"Hwreg", 0>>;
 
 } // End OperandType = "OPERAND_IMMEDIATE"
 
 
 def VOPDstS64 : VOPDstOperand <SReg_64>;
string dst = !if(HasDst,
                        "$vdst"),
                     ""); // use $sdst for VOPC
   string src0 = !if(!eq(NumSrcArgs, 1), "$src0_modifiers", "$src0_modifiers,");
   string src1 = !if(!eq(NumSrcArgs, 1), "",
                    !if(!eq(NumSrcArgs, 2), " $src1_modifiers",
                                            " $src1_modifiers,"));
   string args = !if(!eq(HasModifiers, 0),
                      getAsm32<0, NumSrcArgs, DstVT>.ret,
-                     ", "#src0#src1#", $clamp");
+                     ", "#src0#src1#"$clamp");
   string sdwa = !if(!eq(NumSrcArgs, 0),
                     "",
                     !if(!eq(NumSrcArgs, 1),
                         " $dst_sel $dst_unused $src0_sel",
                         " $dst_sel $dst_unused $src0_sel $src1_sel"
                     )
                 );
   string ret = dst#args#sdwa;
class SDWADisableFields <VOPProfile p> {
   bits<3> src0_modifiers = !if(p.HasModifiers, ?, 0);
   bits<3> src1_sel = !if(!eq(p.NumSrcArgs, 0), 6, 
                          !if(!eq(p.NumSrcArgs, 1), 6,
                              ?));
   bits<3> src1_modifiers = !if(!eq(p.NumSrcArgs, 0), 0, 
                                !if(!eq(p.NumSrcArgs, 1), 0,
                                    !if(p.HasModifiers, ?, 0)));
   bits<3> dst_sel = !if(p.HasDst, ?, 6);
-  bits<2> dst_unused = !if(p.HasDst, ?, 0);
+  bits<2> dst_unused = !if(p.HasDst, ?, 2);
   bits<1> clamp = !if(p.HasModifiers, ?, 0);
 }
 
 class VOP1_SDWA <vop1 op, string opName, VOPProfile p> :
   VOP1_SDWAe <op.VI>,
   VOP_SDWA <p.OutsSDWA, p.InsSDWA, opName#p.AsmSDWA, [], p.HasModifiers>,
   SDWADisableFields <p> {
+  let AsmMatchConverter = !if(!eq(p.HasModifiers,1), "cvtSdwaVop1_mod", "cvtSdwaVop1_nomod");
   let AssemblerPredicates = !if(p.HasExt, [isVI], [DisableInst]);
   let DecoderNamespace = "SDWA";
   let DisableDecoder = DisableVIDecoder;
 }
 
 multiclass VOP1SI_m <vop1 op, string opName, VOPProfile p, list<dag> pattern,
                      string asm = opName#p.Asm32> {
 
class VOP2_DPP <vop2 op, string opName, VOPProfile p> :
   let src0_modifiers = !if(p.HasModifiers, ?, 0);
   let src1_modifiers = !if(p.HasModifiers, ?, 0);
 }
 
 class VOP2_SDWA <vop2 op, string opName, VOPProfile p> :
   VOP2_SDWAe <op.VI>,
   VOP_SDWA <p.OutsSDWA, p.InsSDWA, opName#p.AsmSDWA, [], p.HasModifiers>,
   SDWADisableFields <p> {
+  let AsmMatchConverter = !if(!eq(p.HasModifiers,1), "cvtSdwaVop2_mod", "cvtSdwaVop2_nomod");
   let AssemblerPredicates = !if(p.HasExt, [isVI], [DisableInst]);
   let DecoderNamespace = "SDWA";
   let DisableDecoder = DisableVIDecoder;
 }
 
 class VOP3DisableFields <bit HasSrc1, bit HasSrc2, bit HasModifiers> {
 
   bits<2> src0_modifiers = !if(HasModifiers, ?, 0);

test/MC/AMDGPU/vop_sdwa.s

 // RUN: llvm-mc -arch=amdgcn -mcpu=tonga -show-encoding %s | FileCheck %s --check-prefix=GCN --check-prefix=CIVI --check-prefix=VI
 // RUN: not llvm-mc -arch=amdgcn -show-encoding %s 2>&1 | FileCheck %s --check-prefix=NOSI --check-prefix=NOSICI
 // RUN: not llvm-mc -arch=amdgcn -mcpu=SI -show-encoding %s 2>&1 | FileCheck %s --check-prefix=NOSI --check-prefix=NOSICI
 // RUN: not llvm-mc -arch=amdgcn -mcpu=bonaire -show-encoding %s 2>&1 | FileCheck %s --check-prefix=NOSICI
 
 // ToDo: converters
 // ToDo: VOPC
 // ToDo: VOP2b (see vop_dpp.s)
 // ToDo: V_MAC_F32 (see vop_dpp.s)
 // ToDo: sext()
 // ToDo: intrinsics
 
 
 // NOSICI: error:
 // VI: v_mov_b32_sdwa v1, v2 dst_sel:BYTE_0 dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x02,0x02,0x7e,0x02,0x10,0x06,0x06]
 v_mov_b32 v1, v2 dst_sel:BYTE_0 dst_unused:UNUSED_PRESERVE src0_sel:DWORD
 
 // NOSICI: error:
 // VI: v_mov_b32_sdwa v3, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 ; encoding: [0xf9,0x02,0x06,0x7e,0x04,0x11,0x05,0x06]
 v_mov_b32 v3, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_1
 
 // NOSICI: error:
 // VI: v_mov_b32_sdwa v15, v99 dst_sel:BYTE_2 dst_unused:UNUSED_SEXT src0_sel:WORD_0 ; encoding: [0xf9,0x02,0x1e,0x7e,0x63,0x0a,0x04,0x06]
 v_mov_b32 v15, v99 dst_sel:BYTE_2 dst_unused:UNUSED_SEXT src0_sel:WORD_0
 
 // NOSICI: error:
 // VI: v_min_u32_sdwa v194, v13, v1 dst_sel:BYTE_3 dst_unused:UNUSED_SEXT src0_sel:BYTE_3 src1_sel:BYTE_2 ; encoding: [0xf9,0x02,0x84,0x1d,0x0d,0x0b,0x03,0x02]
 v_min_u32 v194, v13, v1 dst_sel:BYTE_3 dst_unused:UNUSED_SEXT src0_sel:BYTE_3 src1_sel:BYTE_2
 
 // NOSICI: error:
 // VI: v_min_u32_sdwa v255, v4, v1 dst_sel:WORD_0 dst_unused:UNUSED_PAD src0_sel:BYTE_2 src1_sel:WORD_1 ; encoding: [0xf9,0x02,0xfe,0x1d,0x04,0x04,0x02,0x05]
 v_min_u32 v255, v4, v1 dst_sel:WORD_0 dst_unused:UNUSED_PAD src0_sel:BYTE_2 src1_sel:WORD_1
 
 // NOSICI: error:
 // VI: v_min_u32_sdwa v200, v200, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD ; encoding: [0xf9,0x02,0x90,0x1d,0xc8,0x05,0x01,0x06]
 v_min_u32 v200, v200, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
 
 // NOSICI: error:
 // VI: v_min_u32_sdwa v1, v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD ; encoding: [0xf9,0x02,0x02,0x1c,0x01,0x06,0x00,0x06]
 v_min_u32 v1, v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+
+// NOSICI: error:
+// VI: v_mov_b32_sdwa v0, v0 dst_sel:BYTE_0 dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x02,0x00,0x7e,0x00,0x10,0x06,0x06]
+v_mov_b32 v0, v0 dst_sel:BYTE_0 src0_sel:DWORD
+
+// NOSICI: error:
+// VI: v_mov_b32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x02,0x00,0x7e,0x00,0x16,0x06,0x06]
+v_mov_b32 v0, v0 src0_sel:DWORD
+
+// NOSICI: error:
+// VI: v_mov_b32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x02,0x00,0x7e,0x00,0x16,0x06,0x06]
+v_mov_b32_sdwa v0, v0