[AMDGPU] Assembler: Fix VOP3 only instructions

[nhaustov]

Summary

Separate methods to convert parsed instructions to MCInst:

• VOP3 only instructions (always create modifiers as operands in MCInst)
• VOP2 instrunctions with modifiers (create modifiers as operands in MCInst when e64 encoding is forced or modifiers are parsed)
• VOP2 instructions without modifiers (do not create modifiers as operands in MCInst)

Add VOP3Only flag. Pass HasMods flag to VOP3Common.
Simplify code that deals with modifiers (-1 is now same as 0). This is no longer needed.
Mark f16 type as allowing modifiers.
Add few tests (more will be added separately).
Update error message now correct.

[arsenm]

lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

1892 ↗

(On Diff #46555)

Variables are supposed start with a capital letter

lib/Target/AMDGPU/SIInstrInfo.td

1102 ↗

(On Diff #46555)

This comment should now be removed

[• tstellarAMD]

lib/Target/AMDGPU/SIInstrInfo.td

1104–1109 ↗

(On Diff #46555)

This change seems unrelated to the rest of this patch. Is this required to fix the VOP3 operands?

[nhaustov]

Fix variable names.
Remove f16 changes for now as they are not needed.

llvm/trunk/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

public:
   void addRegOrImmOperands(MCInst &Inst, unsigned N) const {
     if (isReg())
       addRegOperands(Inst, N);
     else
       addImmOperands(Inst, N);
   }
 
   void addRegWithInputModsOperands(MCInst &Inst, unsigned N) const {
-    Inst.addOperand(MCOperand::createImm(
-        Reg.Modifiers == -1 ? 0 : Reg.Modifiers));
+    Inst.addOperand(MCOperand::createImm(Reg.Modifiers));
     addRegOperands(Inst, N);
   }
 
   void addSoppBrTargetOperands(MCInst &Inst, unsigned N) const {
     if (isImm())
       addImmOperands(Inst, N);
     else {
       assert(isExpr());
enum ImmTy getImmTy() const {
     return Imm.Type;
   }
 
   bool isRegKind() const {
     return Kind == Register;
   }
 
   bool isReg() const override {
-    return Kind == Register && Reg.Modifiers == -1;
+    return Kind == Register && Reg.Modifiers == 0;
   }
 
   bool isRegWithInputMods() const {
-    return Kind == Register && (Reg.IsForcedVOP3 || Reg.Modifiers != -1);
+    return Kind == Register;
+  }
+
+  bool isClamp() const {
+    return isImm() && Imm.Type == ImmTyClamp;
+  }
+
+  bool isOMod() const {
+    return isImm() && Imm.Type == ImmTyOMod;
+  }
+
+  bool isMod() const {
+    return isClamp() || isOMod();
   }
 
   void setModifiers(unsigned Mods) {
     assert(isReg());
     Reg.Modifiers = Mods;
   }
 
   bool hasModifiers() const {
     assert(isRegKind());
-    return Reg.Modifiers != -1;
+    return Reg.Modifiers != 0;
   }
 
   unsigned getReg() const override {
     return Reg.RegNo;
   }
 
   bool isRegOrImm() const {
     return isReg() || isImm();
   }
 
   bool isRegClass(unsigned RCID) const {
-    return Reg.TRI->getRegClass(RCID).contains(getReg());
+    return isReg() && Reg.TRI->getRegClass(RCID).contains(getReg());
   }
 
   bool isSCSrc32() const {
     return isInlineImm() || (isReg() && isRegClass(AMDGPU::SReg_32RegClassID));
   }
 
   bool isSSrc32() const {
     return isImm() || (isReg() && isRegClass(AMDGPU::SReg_32RegClassID));
static std::unique_ptr<AMDGPUOperand> CreateReg(unsigned RegNo, SMLoc S,
                                                   SMLoc E,
                                                   const MCRegisterInfo *TRI,
                                                   const MCSubtargetInfo *STI,
                                                   bool ForceVOP3) {
     auto Op = llvm::make_unique<AMDGPUOperand>(Register);
     Op->Reg.RegNo = RegNo;
     Op->Reg.TRI = TRI;
     Op->Reg.STI = STI;
-    Op->Reg.Modifiers = -1;
+    Op->Reg.Modifiers = 0;
     Op->Reg.IsForcedVOP3 = ForceVOP3;
     Op->StartLoc = S;
     Op->EndLoc = E;
     return Op;
   }
 
   static std::unique_ptr<AMDGPUOperand> CreateExpr(const class MCExpr *Expr, SMLoc S) {
     auto Op = llvm::make_unique<AMDGPUOperand>(Expression);
public:
   OperandMatchResultTy parseGLC(OperandVector &Operands);
   OperandMatchResultTy parseSLC(OperandVector &Operands);
   OperandMatchResultTy parseTFE(OperandVector &Operands);
 
   OperandMatchResultTy parseDMask(OperandVector &Operands);
   OperandMatchResultTy parseUNorm(OperandVector &Operands);
   OperandMatchResultTy parseR128(OperandVector &Operands);
 
+  void cvtId(MCInst &Inst, const OperandVector &Operands);
+  void cvtVOP3_2_mod(MCInst &Inst, const OperandVector &Operands);
+  void cvtVOP3_2_nomod(MCInst &Inst, const OperandVector &Operands);
+  void cvtVOP3_only(MCInst &Inst, const OperandVector &Operands);
   void cvtVOP3(MCInst &Inst, const OperandVector &Operands);
   OperandMatchResultTy parseVOP3OptionalOps(OperandVector &Operands);
 };
 
 struct OptionalOperand {
   const char *Name;
   AMDGPUOperand::ImmTy Type;
   bool IsBit;
AMDGPUAsmParser::parseOperand(OperandVector &Operands, StringRef Mnemonic) {
   OperandMatchResultTy ResTy = MatchOperandParserImpl(Operands, Mnemonic);
 
   // If we successfully parsed the operand or if there as an error parsing,
   // we are done.
   //
   // If we are parsing after we reach EndOfStatement then this means we
   // are appending default values to the Operands list.  This is only done
   // by custom parser, so we shouldn't continue on to the generic parsing.
-  if (ResTy == MatchOperand_Success || ResTy == MatchOperand_ParseFail ||
+  if (ResTy == MatchOperand_Success || ResTy == MatchOperand_ParseFail||
       getLexer().is(AsmToken::EndOfStatement))
     return ResTy;
 
   bool Negate = false, Abs = false;
   if (getLexer().getKind()== AsmToken::Minus) {
     Parser.Lex();
     Negate = true;
   }
case AsmToken::Real: {
       Operands.push_back(
           AMDGPUOperand::CreateImm(F.bitcastToAPInt().getZExtValue(), S));
       return MatchOperand_Success;
     }
     case AsmToken::Identifier: {
       SMLoc S, E;
       unsigned RegNo;
       if (!ParseRegister(RegNo, S, E)) {
-
-        bool HasModifiers = operandsHaveModifiers(Operands);
         unsigned Modifiers = 0;
 
         if (Negate)
           Modifiers |= 0x1;
 
         if (Abs) {
           if (getLexer().getKind() != AsmToken::Pipe)
             return MatchOperand_ParseFail;
           Parser.Lex();
           Modifiers |= 0x2;
         }
 
-        if (Modifiers && !HasModifiers) {
-          // We are adding a modifier to src1 or src2 and previous sources
-          // don't have modifiers, so we need to go back and empty modifers
-          // for each previous source.
-          for (unsigned PrevRegIdx = Operands.size() - 1; PrevRegIdx > 1;
-               --PrevRegIdx) {
-
-            AMDGPUOperand &RegOp = ((AMDGPUOperand&)*Operands[PrevRegIdx]);
-            RegOp.setModifiers(0);
-          }
-        }
-
-
         Operands.push_back(AMDGPUOperand::CreateReg(
             RegNo, S, E, getContext().getRegisterInfo(), &getSTI(),
             isForcedVOP3()));
 
-        if (HasModifiers || Modifiers) {
+        if (Modifiers) {
           AMDGPUOperand &RegOp = ((AMDGPUOperand&)*Operands[Operands.size() - 1]);
           RegOp.setModifiers(Modifiers);
-
         }
-     }  else {
+      } else {
+        ResTy = parseVOP3OptionalOps(Operands);
+        if (ResTy == MatchOperand_NoMatch) {
           Operands.push_back(AMDGPUOperand::CreateToken(Parser.getTok().getString(),
                                                         S));
           Parser.Lex();
         }
+      }
       return MatchOperand_Success;
     }
     default:
       return MatchOperand_NoMatch;
   }
 }
 
 bool AMDGPUAsmParser::ParseInstruction(ParseInstructionInfo &Info,
                                        StringRef Name,
static const OptionalOperand VOP3OptionalOps [] = {
   {"mul",   AMDGPUOperand::ImmTyOMod, false, 1, ConvertOmodMul},
   {"div",   AMDGPUOperand::ImmTyOMod, false, 1, ConvertOmodDiv},
 };
 
 static bool isVOP3(OperandVector &Operands) {
   if (operandsHaveModifiers(Operands))
     return true;
 
+  if (Operands.size() >= 2) {
     AMDGPUOperand &DstOp = ((AMDGPUOperand&)*Operands[1]);
 
     if (DstOp.isReg() && DstOp.isRegClass(AMDGPU::SGPR_64RegClassID))
       return true;
+  }
 
   if (Operands.size() >= 5)
     return true;
 
   if (Operands.size() > 3) {
     AMDGPUOperand &Src1Op = ((AMDGPUOperand&)*Operands[3]);
     if (Src1Op.getReg() && (Src1Op.isRegClass(AMDGPU::SReg_32RegClassID) ||
                             Src1Op.isRegClass(AMDGPU::SReg_64RegClassID)))
if (!HasModifiers && Res == MatchOperand_Success) {
           Op.setModifiers(0);
       }
     }
     return Res;
   }
   return MatchOperand_NoMatch;
 }
 
-void AMDGPUAsmParser::cvtVOP3(MCInst &Inst, const OperandVector &Operands) {
-
-  unsigned i = 1;
+void AMDGPUAsmParser::cvtId(MCInst &Inst, const OperandVector &Operands) {
+  unsigned I = 1;
   const MCInstrDesc &Desc = MII.get(Inst.getOpcode());
   if (Desc.getNumDefs() > 0) {
-    ((AMDGPUOperand &)*Operands[i++]).addRegOperands(Inst, 1);
-  }
+    ((AMDGPUOperand &)*Operands[I++]).addRegOperands(Inst, 1);
+  }
+  for (unsigned E = Operands.size(); I != E; ++I)
+    ((AMDGPUOperand &)*Operands[I]).addRegOrImmOperands(Inst, 1);
 
-  std::map<enum AMDGPUOperand::ImmTy, unsigned> OptionalIdx;
+void AMDGPUAsmParser::cvtVOP3_2_mod(MCInst &Inst, const OperandVector &Operands) {
+  if (operandsHaveModifiers(Operands) || isForcedVOP3()) {
+    cvtVOP3(Inst, Operands);
+  } else {
+    cvtId(Inst, Operands);
+  }
+}
 
+void AMDGPUAsmParser::cvtVOP3_2_nomod(MCInst &Inst, const OperandVector &Operands) {
   if (operandsHaveModifiers(Operands)) {
-    for (unsigned e = Operands.size(); i != e; ++i) {
-      AMDGPUOperand &Op = ((AMDGPUOperand &)*Operands[i]);
+    cvtVOP3(Inst, Operands);
+  } else {
+    cvtId(Inst, Operands);
+  }
+}
 
-      if (Op.isRegWithInputMods()) {
-        ((AMDGPUOperand &)*Operands[i]).addRegWithInputModsOperands(Inst, 2);
-        continue;
-      }
-      OptionalIdx[Op.getImmTy()] = i;
-    }
+void AMDGPUAsmParser::cvtVOP3_only(MCInst &Inst, const OperandVector &Operands) {
+  cvtVOP3(Inst, Operands);
+
+void AMDGPUAsmParser::cvtVOP3(MCInst &Inst, const OperandVector &Operands) {
+  unsigned I = 1;
+  const MCInstrDesc &Desc = MII.get(Inst.getOpcode());
+  if (Desc.getNumDefs() > 0) {
+    ((AMDGPUOperand &)*Operands[I++]).addRegOperands(Inst, 1);
 
-    unsigned ClampIdx = OptionalIdx[AMDGPUOperand::ImmTyClamp];
-    unsigned OModIdx = OptionalIdx[AMDGPUOperand::ImmTyOMod];
+  unsigned ClampIdx = 0, OModIdx = 0;
+  for (unsigned E = Operands.size(); I != E; ++I) {
+    AMDGPUOperand &Op = ((AMDGPUOperand &)*Operands[I]);
+    if (Op.isRegWithInputMods()) {
+      Op.addRegWithInputModsOperands(Inst, 2);
+    } else if (Op.isClamp()) {
+      ClampIdx = I;
+    } else if (Op.isOMod()) {
+      OModIdx = I;
+    } else if (Op.isImm()) {
+      Op.addImmOperands(Inst, 1);
+    } else {
+      assert(false);
+    }
+  }
 
-    ((AMDGPUOperand &)*Operands[ClampIdx]).addImmOperands(Inst, 1);
-    ((AMDGPUOperand &)*Operands[OModIdx]).addImmOperands(Inst, 1);
+  if (ClampIdx) {
+    AMDGPUOperand &Op = ((AMDGPUOperand &)*Operands[ClampIdx]);
+    Op.addImmOperands(Inst, 1);
+  } else {
+    Inst.addOperand(MCOperand::createImm(0));
+  }
+  if (OModIdx) {
+    AMDGPUOperand &Op = ((AMDGPUOperand &)*Operands[OModIdx]);
+    Op.addImmOperands(Inst, 1);
   } else {
-    for (unsigned e = Operands.size(); i != e; ++i)
-      ((AMDGPUOperand &)*Operands[i]).addRegOrImmOperands(Inst, 1);
+    Inst.addOperand(MCOperand::createImm(0));
   }
 }
 
 /// Force static initialization.
 extern "C" void LLVMInitializeAMDGPUAsmParser() {
   RegisterMCAsmParser<AMDGPUAsmParser> A(TheAMDGPUTarget);
   RegisterMCAsmParser<AMDGPUAsmParser> B(TheGCNTarget);
 }
 
 #define GET_REGISTER_MATCHER
 #define GET_MATCHER_IMPLEMENTATION
 #include "AMDGPUGenAsmMatcher.inc"
 

llvm/trunk/lib/Target/AMDGPU/SIInstrFormats.td

 
 class VOP2Common <dag outs, dag ins, string asm, list<dag> pattern> :
     VOPAnyCommon <outs, ins, asm, pattern> {
 
   let VOP2 = 1;
   let Size = 4;
 }
 
-class VOP3Common <dag outs, dag ins, string asm, list<dag> pattern> :
+class VOP3Common <dag outs, dag ins, string asm, list<dag> pattern, bit HasMods = 0, bit VOP3Only = 0> :
     VOPAnyCommon <outs, ins, asm, pattern> {
 
   // Using complex patterns gives VOP3 patterns a very high complexity rating,
   // but standalone patterns are almost always prefered, so we need to adjust the
   // priority lower.  The goal is to use a high number to reduce complexity to
   // zero (or less than zero).
   let AddedComplexity = -1000;
 
   let VOP3 = 1;
   let VALU = 1;
 
-  let AsmMatchConverter = "cvtVOP3";
+  let AsmMatchConverter =
+    !if(!eq(VOP3Only,1),
+        "cvtVOP3_only",
+        !if(!eq(HasMods,1), "cvtVOP3_2_mod", "cvtVOP3_2_nomod"));
   let isCodeGenOnly = 0;
 
   int Size = 8;
 
   // Because SGPRs may be allowed if there are multiple operands, we
   // need a post-isel hook to insert copies in order to avoid
   // violating constant bus requirements.
   let hasPostISelHook = 1;

llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td

 def TFEFlatMatchClass : TFEBaseMatchClass <"parseFlatOptionalOps">;
 def TFEFlatAtomicMatchClass : TFEBaseMatchClass <"parseFlatAtomicOptionalOps">;
 
 def OModMatchClass : AsmOperandClass {
   let Name = "OMod";
   let PredicateMethod = "isImm";
   let ParserMethod = "parseVOP3OptionalOps";
   let RenderMethod = "addImmOperands";
+  let IsOptional = 1;
 }
 
 def ClampMatchClass : AsmOperandClass {
   let Name = "Clamp";
   let PredicateMethod = "isImm";
   let ParserMethod = "parseVOP3OptionalOps";
   let RenderMethod = "addImmOperands";
+  let IsOptional = 1;
 }
 
 class SMRDOffsetBaseMatchClass <string predicate> : AsmOperandClass {
   let Name = "SMRDOffset"#predicate;
   let PredicateMethod = predicate;
   let RenderMethod = "addImmOperands";
 }
 
!if(!eq(VT.Size, 64),
           VCSrc_32
        )
     );
 }
 
 // Returns 1 if the source arguments have modifiers, 0 if they do not.
 // XXX - do f16 instructions?
 class hasModifiers<ValueType SrcVT> {
-  bit ret = !if(!eq(SrcVT.Value, f32.Value), 1,
-            !if(!eq(SrcVT.Value, f64.Value), 1, 0));
+  bit ret =
+    !if(!eq(SrcVT.Value, f32.Value), 1,
+    !if(!eq(SrcVT.Value, f64.Value), 1,
+    0));
 }
 
 // Returns the input arguments for VOP[12C] instructions for the given SrcVT.
 class getIns32 <RegisterOperand Src0RC, RegisterClass Src1RC, int NumSrcArgs> {
   dag ret = !if(!eq(NumSrcArgs, 1), (ins Src0RC:$src0),               // VOP1
             !if(!eq(NumSrcArgs, 2), (ins Src0RC:$src0, Src1RC:$src1), // VOP2
                                     (ins)));
 }
class VOP3DisableModFields <bit HasSrc0Mods,
                             bit HasOutputMods = 0> {
   bits<2> src0_modifiers = !if(HasSrc0Mods, ?, 0);
   bits<2> src1_modifiers = !if(HasSrc1Mods, ?, 0);
   bits<2> src2_modifiers = !if(HasSrc2Mods, ?, 0);
   bits<2> omod = !if(HasOutputMods, ?, 0);
   bits<1> clamp = !if(HasOutputMods, ?, 0);
 }
 
-class VOP3_Pseudo <dag outs, dag ins, list<dag> pattern, string opName> :
-  VOP3Common <outs, ins, "", pattern>,
+class VOP3_Pseudo <dag outs, dag ins, list<dag> pattern, string opName,
+                   bit HasMods = 0, bit VOP3Only = 0> :
+  VOP3Common <outs, ins, "", pattern, HasMods, VOP3Only>,
   VOP <opName>,
   SIMCInstr<opName#"_e64", SISubtarget.NONE>,
   MnemonicAlias<opName#"_e64", opName> {
   let isPseudo = 1;
   let isCodeGenOnly = 1;
 
   field bit vdst;
   field bit src0;
 }
 
-class VOP3_Real_si <bits<9> op, dag outs, dag ins, string asm, string opName> :
-  VOP3Common <outs, ins, asm, []>,
+class VOP3_Real_si <bits<9> op, dag outs, dag ins, string asm, string opName,
+                    bit HasMods = 0, bit VOP3Only = 0> :
+  VOP3Common <outs, ins, asm, [], HasMods, VOP3Only>,
   VOP3e <op>,
   SIMCInstr<opName#"_e64", SISubtarget.SI> {
   let AssemblerPredicates = [isSICI];
 }
 
-class VOP3_Real_vi <bits<10> op, dag outs, dag ins, string asm, string opName> :
-  VOP3Common <outs, ins, asm, []>,
+class VOP3_Real_vi <bits<10> op, dag outs, dag ins, string asm, string opName,
+                    bit HasMods = 0, bit VOP3Only = 0> :
+  VOP3Common <outs, ins, asm, [], HasMods, VOP3Only>,
   VOP3e_vi <op>,
   SIMCInstr <opName#"_e64", SISubtarget.VI> {
   let AssemblerPredicates = [isVI];
 }
 
-class VOP3b_Real_si <bits<9> op, dag outs, dag ins, string asm, string opName> :
-  VOP3Common <outs, ins, asm, []>,
+class VOP3b_Real_si <bits<9> op, dag outs, dag ins, string asm, string opName,
+                     bit HasMods = 0, bit VOP3Only = 0> :
+  VOP3Common <outs, ins, asm, [], HasMods, VOP3Only>,
   VOP3be <op>,
   SIMCInstr<opName#"_e64", SISubtarget.SI> {
   let AssemblerPredicates = [isSICI];
 }
 
-class VOP3b_Real_vi <bits<10> op, dag outs, dag ins, string asm, string opName> :
-  VOP3Common <outs, ins, asm, []>,
+class VOP3b_Real_vi <bits<10> op, dag outs, dag ins, string asm, string opName,
+                     bit HasMods = 0, bit VOP3Only = 0> :
+  VOP3Common <outs, ins, asm, [], HasMods, VOP3Only>,
   VOP3be_vi <op>,
   SIMCInstr <opName#"_e64", SISubtarget.VI> {
   let AssemblerPredicates = [isVI];
 }
 
 multiclass VOP3_m <vop op, dag outs, dag ins, string asm, list<dag> pattern,
-                   string opName, int NumSrcArgs, bit HasMods = 1> {
+                   string opName, int NumSrcArgs, bit HasMods = 1, bit VOP3Only = 0> {
 
   def "" : VOP3_Pseudo <outs, ins, pattern, opName>;
 
-  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName>,
+  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName, HasMods, VOP3Only>,
             VOP3DisableFields<!if(!eq(NumSrcArgs, 1), 0, 1),
                               !if(!eq(NumSrcArgs, 2), 0, 1),
                               HasMods>;
-  def _vi : VOP3_Real_vi <op.VI3, outs, ins, asm, opName>,
+  def _vi : VOP3_Real_vi <op.VI3, outs, ins, asm, opName, HasMods, VOP3Only>,
             VOP3DisableFields<!if(!eq(NumSrcArgs, 1), 0, 1),
                               !if(!eq(NumSrcArgs, 2), 0, 1),
                               HasMods>;
 }
 
 multiclass VOP3_1_m <vop op, dag outs, dag ins, string asm,
                      list<dag> pattern, string opName, bit HasMods = 1> {
 
-  def "" : VOP3_Pseudo <outs, ins, pattern, opName>;
+  def "" : VOP3_Pseudo <outs, ins, pattern, opName, HasMods>;
 
-  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName>,
+  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName, HasMods>,
             VOP3DisableFields<0, 0, HasMods>;
 
-  def _vi : VOP3_Real_vi <op.VI3, outs, ins, asm, opName>,
+  def _vi : VOP3_Real_vi <op.VI3, outs, ins, asm, opName, HasMods>,
             VOP3DisableFields<0, 0, HasMods>;
 }
 
 multiclass VOP3SI_1_m <vop op, dag outs, dag ins, string asm,
                      list<dag> pattern, string opName, bit HasMods = 1> {
 
-  def "" : VOP3_Pseudo <outs, ins, pattern, opName>;
+  def "" : VOP3_Pseudo <outs, ins, pattern, opName, HasMods>;
 
-  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName>,
+  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName, HasMods>,
             VOP3DisableFields<0, 0, HasMods>;
   // No VI instruction. This class is for SI only.
 }
 
 multiclass VOP3_2_m <vop op, dag outs, dag ins, string asm,
                      list<dag> pattern, string opName, string revOp,
                      bit HasMods = 1> {
 
-  def "" : VOP3_Pseudo <outs, ins, pattern, opName>,
+  def "" : VOP3_Pseudo <outs, ins, pattern, opName, HasMods>,
            VOP2_REV<revOp#"_e64", !eq(revOp, opName)>;
 
-  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName>,
+  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName, HasMods>,
             VOP3DisableFields<1, 0, HasMods>;
 
-  def _vi : VOP3_Real_vi <op.VI3, outs, ins, asm, opName>,
+  def _vi : VOP3_Real_vi <op.VI3, outs, ins, asm, opName, HasMods>,
             VOP3DisableFields<1, 0, HasMods>;
 }
 
 multiclass VOP3SI_2_m <vop op, dag outs, dag ins, string asm,
                      list<dag> pattern, string opName, string revOp,
                      bit HasMods = 1> {
 
-  def "" : VOP3_Pseudo <outs, ins, pattern, opName>,
+  def "" : VOP3_Pseudo <outs, ins, pattern, opName, HasMods>,
            VOP2_REV<revOp#"_e64", !eq(revOp, opName)>;
 
-  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName>,
+  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName, HasMods>,
             VOP3DisableFields<1, 0, HasMods>;
 
   // No VI instruction. This class is for SI only.
 }
 
 // Two operand VOP3b instruction that may have a 3rd SGPR bool operand
 // instead of an implicit VCC as in the VOP2b format.
 multiclass VOP3b_2_3_m <vop op, dag outs, dag ins, string asm,
                         list<dag> pattern, string opName, string revOp,
                         bit HasMods = 1, bit useSrc2Input = 0> {
   def "" : VOP3_Pseudo <outs, ins, pattern, opName>;
 
   def _si : VOP3b_Real_si <op.SI3, outs, ins, asm, opName>,
             VOP3DisableFields<1, useSrc2Input, HasMods>;
 
   def _vi : VOP3b_Real_vi <op.VI3, outs, ins, asm, opName>,
             VOP3DisableFields<1, useSrc2Input, HasMods>;
 }
 
 multiclass VOP3_C_m <vop op, dag outs, dag ins, string asm,
                      list<dag> pattern, string opName,
                      bit HasMods, bit defExec,
                      string revOp, list<SchedReadWrite> sched> {
 
-  def "" : VOP3_Pseudo <outs, ins, pattern, opName>,
+  def "" : VOP3_Pseudo <outs, ins, pattern, opName, HasMods>,
            VOP2_REV<revOp#"_e64", !eq(revOp, opName)> {
     let Defs = !if(defExec, [EXEC], []);
     let SchedRW = sched;
   }
 
-  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName>,
+  def _si : VOP3_Real_si <op.SI3, outs, ins, asm, opName, HasMods>,
             VOP3DisableFields<1, 0, HasMods> {
     let Defs = !if(defExec, [EXEC], []);
     let SchedRW = sched;
   }
 
-  def _vi : VOP3_Real_vi <op.VI3, outs, ins, asm, opName>,
+  def _vi : VOP3_Real_vi <op.VI3, outs, ins, asm, opName, HasMods>,
             VOP3DisableFields<1, 0, HasMods> {
     let Defs = !if(defExec, [EXEC], []);
     let SchedRW = sched;
   }
 }
 
 // An instruction that is VOP2 on SI and VOP3 on VI, no modifiers.
 multiclass VOP2SI_3VI_m <vop3 op, string opName, dag outs, dag ins,

 
 multiclass VOPCX_I32 <vopc op, string opName, string revOp = opName> :
   VOPCX <op, opName, VOPC_I1_I32_I32, COND_NULL, [Write32Bit], revOp>;
 
 multiclass VOPCX_I64 <vopc op, string opName, string revOp = opName> :
   VOPCX <op, opName, VOPC_I1_I64_I64, COND_NULL, [Write64Bit], revOp>;
 
 multiclass VOP3_Helper <vop3 op, string opName, dag outs, dag ins, string asm,
-                        list<dag> pat, int NumSrcArgs, bit HasMods> : VOP3_m <
-    op, outs, ins, opName#" "#asm, pat, opName, NumSrcArgs, HasMods
+                        list<dag> pat, int NumSrcArgs, bit HasMods,
+                        bit VOP3Only = 0> : VOP3_m <
+    op, outs, ins, opName#" "#asm, pat, opName, NumSrcArgs, HasMods, VOP3Only
 >;
 
 multiclass VOPC_CLASS_F32 <vopc op, string opName> :
   VOPCClassInst <op, opName, VOPC_I1_F32_I32, 0, [Write32Bit]>;
 
 multiclass VOPCX_CLASS_F32 <vopc op, string opName> :
   VOPCClassInst <op, opName, VOPC_I1_F32_I32, 1, [Write32Bit]>;
 
 multiclass VOPC_CLASS_F64 <vopc op, string opName> :
   VOPCClassInst <op, opName, VOPC_I1_F64_I32, 0, [WriteDoubleAdd]>;
 
 multiclass VOPCX_CLASS_F64 <vopc op, string opName> :
   VOPCClassInst <op, opName, VOPC_I1_F64_I32, 1, [WriteDoubleAdd]>;
 
 multiclass VOP3Inst <vop3 op, string opName, VOPProfile P,
-                     SDPatternOperator node = null_frag> : VOP3_Helper <
+                     SDPatternOperator node = null_frag, bit VOP3Only = 0> :
+  VOP3_Helper <
   op, opName, (outs P.DstRC.RegClass:$dst), P.Ins64, P.Asm64,
   !if(!eq(P.NumSrcArgs, 3),
     !if(P.HasModifiers,
         [(set P.DstVT:$dst,
             (node (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers,
                                        i1:$clamp, i32:$omod)),
                   (P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers)),
                   (P.Src2VT (VOP3Mods P.Src2VT:$src2, i32:$src2_modifiers))))],
         [(set P.DstVT:$dst, (node P.Src0VT:$src0, P.Src1VT:$src1,
                                   P.Src2VT:$src2))]),
   !if(!eq(P.NumSrcArgs, 2),
     !if(P.HasModifiers,
         [(set P.DstVT:$dst,
             (node (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers,
                                        i1:$clamp, i32:$omod)),
                   (P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers))))],
         [(set P.DstVT:$dst, (node P.Src0VT:$src0, P.Src1VT:$src1))])
   /* P.NumSrcArgs == 1 */,
     !if(P.HasModifiers,
         [(set P.DstVT:$dst,
             (node (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers,
                                        i1:$clamp, i32:$omod))))],
         [(set P.DstVT:$dst, (node P.Src0VT:$src0))]))),
-  P.NumSrcArgs, P.HasModifiers
+  P.NumSrcArgs, P.HasModifiers, VOP3Only
 >;
 
 // Special case for v_div_fmas_{f32|f64}, since it seems to be the
 // only VOP instruction that implicitly reads VCC.
 multiclass VOP3_VCC_Inst <vop3 op, string opName,
                           VOPProfile P,
                           SDPatternOperator node = null_frag> : VOP3_Helper <
   op, opName,

llvm/trunk/lib/Target/AMDGPU/SIInstructions.td

 >;
 
 } // End SchedRW = [WriteDouble]
 
 let SchedRW = [WriteDoubleAdd] in {
 let isCommutable = 1 in {
 
 defm V_ADD_F64 : VOP3Inst <vop3<0x164, 0x280>, "v_add_f64",
-  VOP_F64_F64_F64, fadd
+  VOP_F64_F64_F64, fadd, 1
 >;
 defm V_MUL_F64 : VOP3Inst <vop3<0x165, 0x281>, "v_mul_f64",
-  VOP_F64_F64_F64, fmul
+  VOP_F64_F64_F64, fmul, 1
 >;
 
 defm V_MIN_F64 : VOP3Inst <vop3<0x166, 0x282>, "v_min_f64",
-  VOP_F64_F64_F64, fminnum
+  VOP_F64_F64_F64, fminnum, 1
 >;
 defm V_MAX_F64 : VOP3Inst <vop3<0x167, 0x283>, "v_max_f64",
-  VOP_F64_F64_F64, fmaxnum
+  VOP_F64_F64_F64, fmaxnum, 1
 >;
 
 } // End isCommutable = 1
 
 defm V_LDEXP_F64 : VOP3Inst <vop3<0x168, 0x284>, "v_ldexp_f64",
-  VOP_F64_F64_I32, AMDGPUldexp
+  VOP_F64_F64_I32, AMDGPUldexp, 1
 >;
 
 } // End let SchedRW = [WriteDoubleAdd]
 
 let isCommutable = 1, SchedRW = [WriteQuarterRate32] in {
 
 defm V_MUL_LO_U32 : VOP3Inst <vop3<0x169, 0x285>, "v_mul_lo_u32",
   VOP_I32_I32_I32

llvm/trunk/test/MC/AMDGPU/vop3.s

 // VI:   v_sub_f32_e64 v1, v3, s5 ; encoding: [0x01,0x00,0x02,0xd1,0x03,0x0b,0x00,0x00]
 
 v_subrev_f32 v1, v3, s5
 // SICI: v_subrev_f32_e64 v1, v3, s5 ; encoding: [0x01,0x00,0x0a,0xd2,0x03,0x0b,0x00,0x00]
 // VI:   v_subrev_f32_e64 v1, v3, s5 ; encoding: [0x01,0x00,0x03,0xd1,0x03,0x0b,0x00,0x00]
 
 v_mac_legacy_f32 v1, v3, s5
 // SICI: v_mac_legacy_f32_e64 v1, v3, s5 ; encoding: [0x01,0x00,0x0c,0xd2,0x03,0x0b,0x00,0x00]
-// FIXME: The error message should be: error: instruction not supported on this GPU
-// NOVI: error: invalid operand for instruction
+// NOVI: error: instruction not supported on this GPU
 
 v_mul_legacy_f32 v1, v3, s5
 // SICI: v_mul_legacy_f32_e64 v1, v3, s5 ; encoding: [0x01,0x00,0x0e,0xd2,0x03,0x0b,0x00,0x00]
 // VI:   v_mul_legacy_f32_e64 v1, v3, s5 ; encoding: [0x01,0x00,0x04,0xd1,0x03,0x0b,0x00,0x00]
 
 v_mul_f32 v1, v3, s5
 // SICI: v_mul_f32_e64 v1, v3, s5 ; encoding: [0x01,0x00,0x10,0xd2,0x03,0x0b,0x00,0x00]
 // VI:   v_mul_f32_e64 v1, v3, s5 ; encoding: [0x01,0x00,0x05,0xd1,0x03,0x0b,0x00,0x00]
 
 v_mul_i32_i24 v1, v3, s5
 // SICI: v_mul_i32_i24_e64 v1, v3, s5 ; encoding: [0x01,0x00,0x12,0xd2,0x03,0x0b,0x00,0x00]
 // VI:   v_mul_i32_i24_e64 v1, v3, s5 ; encoding: [0x01,0x00,0x06,0xd1,0x03,0x0b,0x00,0x00]
 
 ///===---------------------------------------------------------------------===//
 // VOP3 Instructions
 ///===---------------------------------------------------------------------===//
 
 // TODO: Modifier tests
 
 v_mad_legacy_f32 v2, v4, v6, v8
 // SICI: v_mad_legacy_f32 v2, v4, v6, v8 ; encoding: [0x02,0x00,0x80,0xd2,0x04,0x0d,0x22,0x04]
 // VI:   v_mad_legacy_f32 v2, v4, v6, v8 ; encoding: [0x02,0x00,0xc0,0xd1,0x04,0x0d,0x22,0x04]
 
-
-
-
+v_add_f64 v[0:1], v[2:3], v[5:6]
+// SICI: v_add_f64 v[0:1], v[2:3], v[5:6] ; encoding: [0x00,0x00,0xc8,0xd2,0x02,0x0b,0x02,0x00]
+// VI:   v_add_f64 v[0:1], v[2:3], v[5:6] ; encoding: [0x00,0x00,0x80,0xd2,0x02,0x0b,0x02,0x00]
+
+v_add_f64_e64 v[0:1], v[2:3], v[5:6]
+// SICI: v_add_f64 v[0:1], v[2:3], v[5:6] ; encoding: [0x00,0x00,0xc8,0xd2,0x02,0x0b,0x02,0x00]
+// VI:   v_add_f64 v[0:1], v[2:3], v[5:6] ; encoding: [0x00,0x00,0x80,0xd2,0x02,0x0b,0x02,0x00]
+
+v_add_f64 v[0:1], -v[2:3], v[5:6]
+// SICI: v_add_f64 v[0:1], -v[2:3], v[5:6] ; encoding: [0x00,0x00,0xc8,0xd2,0x02,0x0b,0x02,0x20]
+// VI:   v_add_f64 v[0:1], -v[2:3], v[5:6] ; encoding: [0x00,0x00,0x80,0xd2,0x02,0x0b,0x02,0x20]
+
+v_add_f64_e64 v[0:1], -v[2:3], v[5:6]
+// SICI: v_add_f64 v[0:1], -v[2:3], v[5:6] ; encoding: [0x00,0x00,0xc8,0xd2,0x02,0x0b,0x02,0x20]
+// VI:   v_add_f64 v[0:1], -v[2:3], v[5:6] ; encoding: [0x00,0x00,0x80,0xd2,0x02,0x0b,0x02,0x20]
+
+v_add_f64 v[0:1], v[2:3], -v[5:6]
+// SICI: v_add_f64 v[0:1], v[2:3], -v[5:6] ; encoding: [0x00,0x00,0xc8,0xd2,0x02,0x0b,0x02,0x40]
+// VI:   v_add_f64 v[0:1], v[2:3], -v[5:6] ; encoding: [0x00,0x00,0x80,0xd2,0x02,0x0b,0x02,0x40]
+
+v_add_f64_e64 v[0:1], v[2:3], -v[5:6]
+// SICI: v_add_f64 v[0:1], v[2:3], -v[5:6] ; encoding: [0x00,0x00,0xc8,0xd2,0x02,0x0b,0x02,0x40]
+// VI:   v_add_f64 v[0:1], v[2:3], -v[5:6] ; encoding: [0x00,0x00,0x80,0xd2,0x02,0x0b,0x02,0x40]
+
+v_add_f64 v[0:1], |v[2:3]|, v[5:6]
+// SICI: v_add_f64 v[0:1], |v[2:3]|, v[5:6] ; encoding: [0x00,0x01,0xc8,0xd2,0x02,0x0b,0x02,0x00]
+// VI:   v_add_f64 v[0:1], |v[2:3]|, v[5:6] ; encoding: [0x00,0x01,0x80,0xd2,0x02,0x0b,0x02,0x00]
+
+v_add_f64_e64 v[0:1], |v[2:3]|, v[5:6]
+// SICI: v_add_f64 v[0:1], |v[2:3]|, v[5:6] ; encoding: [0x00,0x01,0xc8,0xd2,0x02,0x0b,0x02,0x00]
+// VI:   v_add_f64 v[0:1], |v[2:3]|, v[5:6] ; encoding: [0x00,0x01,0x80,0xd2,0x02,0x0b,0x02,0x00]
+
+v_add_f64 v[0:1], v[2:3], |v[5:6]|
+// SICI: v_add_f64 v[0:1], v[2:3], |v[5:6]| ; encoding: [0x00,0x02,0xc8,0xd2,0x02,0x0b,0x02,0x00]
+// VI:   v_add_f64 v[0:1], v[2:3], |v[5:6]| ; encoding: [0x00,0x02,0x80,0xd2,0x02,0x0b,0x02,0x00]
+
+v_add_f64_e64 v[0:1], v[2:3], |v[5:6]|
+// SICI: v_add_f64 v[0:1], v[2:3], |v[5:6]| ; encoding: [0x00,0x02,0xc8,0xd2,0x02,0x0b,0x02,0x00]
+// VI:   v_add_f64 v[0:1], v[2:3], |v[5:6]| ; encoding: [0x00,0x02,0x80,0xd2,0x02,0x0b,0x02,0x00]
+
+v_add_f64 v[0:1], -v[2:3], |v[5:6]| clamp mul:4
+// SICI: v_add_f64 v[0:1], -v[2:3], |v[5:6]| clamp mul:4 ; encoding: [0x00,0x0a,0xc8,0xd2,0x02,0x0b,0x02,0x30]
+// VI:   v_add_f64 v[0:1], -v[2:3], |v[5:6]| clamp mul:4 ; encoding: [0x00,0x82,0x80,0xd2,0x02,0x0b,0x02,0x30]
+
+v_add_f64_e64 v[0:1], -v[2:3], |v[5:6]| clamp mul:4
+// SICI: v_add_f64 v[0:1], -v[2:3], |v[5:6]| clamp mul:4 ; encoding: [0x00,0x0a,0xc8,0xd2,0x02,0x0b,0x02,0x30]
+// VI:   v_add_f64 v[0:1], -v[2:3], |v[5:6]| clamp mul:4 ; encoding: [0x00,0x82,0x80,0xd2,0x02,0x0b,0x02,0x30]