//
// The LLVM Compiler Infrastructure
//
-// This file was developed by the LLVM research group and is distributed under
-// the University of Illinois Open Source License. See LICENSE.TXT for details.
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
//
//===----------------------------------------------------------------------===//
-#include "AsmWriterEmitter.h"
+#include "AsmWriterInst.h"
#include "CodeGenTarget.h"
-#include "Record.h"
+#include "SequenceToOffsetTable.h"
+#include "llvm/ADT/SmallString.h"
#include "llvm/ADT/StringExtras.h"
+#include "llvm/ADT/Twine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/Format.h"
+#include "llvm/Support/MathExtras.h"
+#include "llvm/TableGen/Error.h"
+#include "llvm/TableGen/Record.h"
+#include "llvm/TableGen/TableGenBackend.h"
#include <algorithm>
-#include <ostream>
+#include <cassert>
+#include <map>
+#include <vector>
using namespace llvm;
-static bool isIdentChar(char C) {
- return (C >= 'a' && C <= 'z') ||
- (C >= 'A' && C <= 'Z') ||
- (C >= '0' && C <= '9') ||
- C == '_';
-}
+#define DEBUG_TYPE "asm-writer-emitter"
namespace {
- struct AsmWriterOperand {
- enum { isLiteralTextOperand, isMachineInstrOperand } OperandType;
-
- /// Str - For isLiteralTextOperand, this IS the literal text. For
- /// isMachineInstrOperand, this is the PrinterMethodName for the operand.
- std::string Str;
-
- /// MiOpNo - For isMachineInstrOperand, this is the operand number of the
- /// machine instruction.
- unsigned MIOpNo;
-
- /// MiModifier - For isMachineInstrOperand, this is the modifier string for
- /// an operand, specified with syntax like ${opname:modifier}.
- std::string MiModifier;
-
- AsmWriterOperand(const std::string &LitStr)
- : OperandType(isLiteralTextOperand), Str(LitStr) {}
-
- AsmWriterOperand(const std::string &Printer, unsigned OpNo,
- const std::string &Modifier)
- : OperandType(isMachineInstrOperand), Str(Printer), MIOpNo(OpNo),
- MiModifier(Modifier) {}
-
- bool operator!=(const AsmWriterOperand &Other) const {
- if (OperandType != Other.OperandType || Str != Other.Str) return true;
- if (OperandType == isMachineInstrOperand)
- return MIOpNo != Other.MIOpNo || MiModifier != Other.MiModifier;
- return false;
- }
- bool operator==(const AsmWriterOperand &Other) const {
- return !operator!=(Other);
- }
- void EmitCode(std::ostream &OS) const;
- };
-
- struct AsmWriterInst {
- std::vector<AsmWriterOperand> Operands;
- const CodeGenInstruction *CGI;
-
- AsmWriterInst(const CodeGenInstruction &CGI, unsigned Variant);
-
- /// MatchesAllButOneOp - If this instruction is exactly identical to the
- /// specified instruction except for one differing operand, return the
- /// differing operand number. Otherwise return ~0.
- unsigned MatchesAllButOneOp(const AsmWriterInst &Other) const;
-
- private:
- void AddLiteralString(const std::string &Str) {
- // If the last operand was already a literal text string, append this to
- // it, otherwise add a new operand.
- if (!Operands.empty() &&
- Operands.back().OperandType == AsmWriterOperand::isLiteralTextOperand)
- Operands.back().Str.append(Str);
- else
- Operands.push_back(AsmWriterOperand(Str));
- }
- };
-}
-
-
-void AsmWriterOperand::EmitCode(std::ostream &OS) const {
- if (OperandType == isLiteralTextOperand)
- OS << "O << \"" << Str << "\"; ";
- else {
- OS << Str << "(MI, " << MIOpNo;
- if (!MiModifier.empty())
- OS << ", \"" << MiModifier << '"';
- OS << "); ";
- }
-}
-
-
-/// ParseAsmString - Parse the specified Instruction's AsmString into this
-/// AsmWriterInst.
-///
-AsmWriterInst::AsmWriterInst(const CodeGenInstruction &CGI, unsigned Variant) {
- this->CGI = &CGI;
- unsigned CurVariant = ~0U; // ~0 if we are outside a {.|.|.} region, other #.
-
- // NOTE: Any extensions to this code need to be mirrored in the
- // AsmPrinter::printInlineAsm code that executes as compile time (assuming
- // that inline asm strings should also get the new feature)!
- const std::string &AsmString = CGI.AsmString;
- std::string::size_type LastEmitted = 0;
- while (LastEmitted != AsmString.size()) {
- std::string::size_type DollarPos =
- AsmString.find_first_of("${|}", LastEmitted);
- if (DollarPos == std::string::npos) DollarPos = AsmString.size();
-
- // Emit a constant string fragment.
- if (DollarPos != LastEmitted) {
- // TODO: this should eventually handle escaping.
- if (CurVariant == Variant || CurVariant == ~0U)
- AddLiteralString(std::string(AsmString.begin()+LastEmitted,
- AsmString.begin()+DollarPos));
- LastEmitted = DollarPos;
- } else if (AsmString[DollarPos] == '{') {
- if (CurVariant != ~0U)
- throw "Nested variants found for instruction '" +
- CGI.TheDef->getName() + "'!";
- LastEmitted = DollarPos+1;
- CurVariant = 0; // We are now inside of the variant!
- } else if (AsmString[DollarPos] == '|') {
- if (CurVariant == ~0U)
- throw "'|' character found outside of a variant in instruction '"
- + CGI.TheDef->getName() + "'!";
- ++CurVariant;
- ++LastEmitted;
- } else if (AsmString[DollarPos] == '}') {
- if (CurVariant == ~0U)
- throw "'}' character found outside of a variant in instruction '"
- + CGI.TheDef->getName() + "'!";
- ++LastEmitted;
- CurVariant = ~0U;
- } else if (DollarPos+1 != AsmString.size() &&
- AsmString[DollarPos+1] == '$') {
- if (CurVariant == Variant || CurVariant == ~0U)
- AddLiteralString("$"); // "$$" -> $
- LastEmitted = DollarPos+2;
- } else {
- // Get the name of the variable.
- std::string::size_type VarEnd = DollarPos+1;
-
- // handle ${foo}bar as $foo by detecting whether the character following
- // the dollar sign is a curly brace. If so, advance VarEnd and DollarPos
- // so the variable name does not contain the leading curly brace.
- bool hasCurlyBraces = false;
- if (VarEnd < AsmString.size() && '{' == AsmString[VarEnd]) {
- hasCurlyBraces = true;
- ++DollarPos;
- ++VarEnd;
- }
-
- while (VarEnd < AsmString.size() && isIdentChar(AsmString[VarEnd]))
- ++VarEnd;
- std::string VarName(AsmString.begin()+DollarPos+1,
- AsmString.begin()+VarEnd);
-
- // Modifier - Support ${foo:modifier} syntax, where "modifier" is passed
- // into printOperand.
- std::string Modifier;
-
- // In order to avoid starting the next string at the terminating curly
- // brace, advance the end position past it if we found an opening curly
- // brace.
- if (hasCurlyBraces) {
- if (VarEnd >= AsmString.size())
- throw "Reached end of string before terminating curly brace in '"
- + CGI.TheDef->getName() + "'";
-
- // Look for a modifier string.
- if (AsmString[VarEnd] == ':') {
- ++VarEnd;
- if (VarEnd >= AsmString.size())
- throw "Reached end of string before terminating curly brace in '"
- + CGI.TheDef->getName() + "'";
-
- unsigned ModifierStart = VarEnd;
- while (VarEnd < AsmString.size() && isIdentChar(AsmString[VarEnd]))
- ++VarEnd;
- Modifier = std::string(AsmString.begin()+ModifierStart,
- AsmString.begin()+VarEnd);
- if (Modifier.empty())
- throw "Bad operand modifier name in '"+ CGI.TheDef->getName() + "'";
- }
-
- if (AsmString[VarEnd] != '}')
- throw "Variable name beginning with '{' did not end with '}' in '"
- + CGI.TheDef->getName() + "'";
- ++VarEnd;
- }
- if (VarName.empty())
- throw "Stray '$' in '" + CGI.TheDef->getName() +
- "' asm string, maybe you want $$?";
-
- unsigned OpNo = CGI.getOperandNamed(VarName);
- CodeGenInstruction::OperandInfo OpInfo = CGI.OperandList[OpNo];
-
- // If this is a two-address instruction and we are not accessing the
- // 0th operand, remove an operand.
- unsigned MIOp = OpInfo.MIOperandNo;
- if (CGI.isTwoAddress && MIOp != 0) {
- if (MIOp == 1)
- throw "Should refer to operand #0 instead of #1 for two-address"
- " instruction '" + CGI.TheDef->getName() + "'!";
- --MIOp;
- }
+class AsmWriterEmitter {
+ RecordKeeper &Records;
+ CodeGenTarget Target;
+ std::map<const CodeGenInstruction*, AsmWriterInst*> CGIAWIMap;
+ const std::vector<const CodeGenInstruction*> *NumberedInstructions;
+ std::vector<AsmWriterInst> Instructions;
+ std::vector<std::string> PrintMethods;
+public:
+ AsmWriterEmitter(RecordKeeper &R);
- if (CurVariant == Variant || CurVariant == ~0U)
- Operands.push_back(AsmWriterOperand(OpInfo.PrinterMethodName, MIOp,
- Modifier));
- LastEmitted = VarEnd;
- }
- }
+ void run(raw_ostream &o);
- AddLiteralString("\\n");
-}
+private:
+ void EmitPrintInstruction(raw_ostream &o);
+ void EmitGetRegisterName(raw_ostream &o);
+ void EmitPrintAliasInstruction(raw_ostream &O);
-/// MatchesAllButOneOp - If this instruction is exactly identical to the
-/// specified instruction except for one differing operand, return the differing
-/// operand number. If more than one operand mismatches, return ~1, otherwise
-/// if the instructions are identical return ~0.
-unsigned AsmWriterInst::MatchesAllButOneOp(const AsmWriterInst &Other)const{
- if (Operands.size() != Other.Operands.size()) return ~1;
-
- unsigned MismatchOperand = ~0U;
- for (unsigned i = 0, e = Operands.size(); i != e; ++i) {
- if (Operands[i] != Other.Operands[i])
- if (MismatchOperand != ~0U) // Already have one mismatch?
- return ~1U;
- else
- MismatchOperand = i;
+ AsmWriterInst *getAsmWriterInstByID(unsigned ID) const {
+ assert(ID < NumberedInstructions->size());
+ std::map<const CodeGenInstruction*, AsmWriterInst*>::const_iterator I =
+ CGIAWIMap.find(NumberedInstructions->at(ID));
+ assert(I != CGIAWIMap.end() && "Didn't find inst!");
+ return I->second;
}
- return MismatchOperand;
-}
+ void FindUniqueOperandCommands(std::vector<std::string> &UOC,
+ std::vector<unsigned> &InstIdxs,
+ std::vector<unsigned> &InstOpsUsed) const;
+};
+} // end anonymous namespace
static void PrintCases(std::vector<std::pair<std::string,
- AsmWriterOperand> > &OpsToPrint, std::ostream &O) {
+ AsmWriterOperand> > &OpsToPrint, raw_ostream &O) {
O << " case " << OpsToPrint.back().first << ": ";
AsmWriterOperand TheOp = OpsToPrint.back().second;
OpsToPrint.pop_back();
}
// Finally, emit the code.
- TheOp.EmitCode(O);
+ O << TheOp.getCode();
O << "break;\n";
}
/// EmitInstructions - Emit the last instruction in the vector and any other
/// instructions that are suitably similar to it.
static void EmitInstructions(std::vector<AsmWriterInst> &Insts,
- std::ostream &O) {
+ raw_ostream &O) {
AsmWriterInst FirstInst = Insts.back();
Insts.pop_back();
O << " case " << FirstInst.CGI->Namespace << "::"
<< FirstInst.CGI->TheDef->getName() << ":\n";
- for (unsigned i = 0, e = SimilarInsts.size(); i != e; ++i)
- O << " case " << SimilarInsts[i].CGI->Namespace << "::"
- << SimilarInsts[i].CGI->TheDef->getName() << ":\n";
+ for (const AsmWriterInst &AWI : SimilarInsts)
+ O << " case " << AWI.CGI->Namespace << "::"
+ << AWI.CGI->TheDef->getName() << ":\n";
for (unsigned i = 0, e = FirstInst.Operands.size(); i != e; ++i) {
if (i != DifferingOperand) {
// If the operand is the same for all instructions, just print it.
- O << " ";
- FirstInst.Operands[i].EmitCode(O);
+ O << " " << FirstInst.Operands[i].getCode();
} else {
// If this is the operand that varies between all of the instructions,
// emit a switch for just this operand now.
FirstInst.CGI->TheDef->getName(),
FirstInst.Operands[i]));
- for (unsigned si = 0, e = SimilarInsts.size(); si != e; ++si) {
- AsmWriterInst &AWI = SimilarInsts[si];
+ for (const AsmWriterInst &AWI : SimilarInsts) {
OpsToPrint.push_back(std::make_pair(AWI.CGI->Namespace+"::"+
AWI.CGI->TheDef->getName(),
AWI.Operands[i]));
}
O << "\n";
}
-
O << " break;\n";
}
-void AsmWriterEmitter::run(std::ostream &O) {
- EmitSourceFileHeader("Assembly Writer Source Fragment", O);
+void AsmWriterEmitter::
+FindUniqueOperandCommands(std::vector<std::string> &UniqueOperandCommands,
+ std::vector<unsigned> &InstIdxs,
+ std::vector<unsigned> &InstOpsUsed) const {
+ InstIdxs.assign(NumberedInstructions->size(), ~0U);
- CodeGenTarget Target;
+ // This vector parallels UniqueOperandCommands, keeping track of which
+ // instructions each case are used for. It is a comma separated string of
+ // enums.
+ std::vector<std::string> InstrsForCase;
+ InstrsForCase.resize(UniqueOperandCommands.size());
+ InstOpsUsed.assign(UniqueOperandCommands.size(), 0);
+
+ for (unsigned i = 0, e = NumberedInstructions->size(); i != e; ++i) {
+ const AsmWriterInst *Inst = getAsmWriterInstByID(i);
+ if (!Inst)
+ continue; // PHI, INLINEASM, CFI_INSTRUCTION, etc.
+
+ if (Inst->Operands.empty())
+ continue; // Instruction already done.
+
+ std::string Command = " " + Inst->Operands[0].getCode() + "\n";
+
+ // Check to see if we already have 'Command' in UniqueOperandCommands.
+ // If not, add it.
+ bool FoundIt = false;
+ for (unsigned idx = 0, e = UniqueOperandCommands.size(); idx != e; ++idx)
+ if (UniqueOperandCommands[idx] == Command) {
+ InstIdxs[i] = idx;
+ InstrsForCase[idx] += ", ";
+ InstrsForCase[idx] += Inst->CGI->TheDef->getName();
+ FoundIt = true;
+ break;
+ }
+ if (!FoundIt) {
+ InstIdxs[i] = UniqueOperandCommands.size();
+ UniqueOperandCommands.push_back(std::move(Command));
+ InstrsForCase.push_back(Inst->CGI->TheDef->getName());
+
+ // This command matches one operand so far.
+ InstOpsUsed.push_back(1);
+ }
+ }
+
+ // For each entry of UniqueOperandCommands, there is a set of instructions
+ // that uses it. If the next command of all instructions in the set are
+ // identical, fold it into the command.
+ for (unsigned CommandIdx = 0, e = UniqueOperandCommands.size();
+ CommandIdx != e; ++CommandIdx) {
+
+ for (unsigned Op = 1; ; ++Op) {
+ // Scan for the first instruction in the set.
+ std::vector<unsigned>::iterator NIT =
+ std::find(InstIdxs.begin(), InstIdxs.end(), CommandIdx);
+ if (NIT == InstIdxs.end()) break; // No commonality.
+
+ // If this instruction has no more operands, we isn't anything to merge
+ // into this command.
+ const AsmWriterInst *FirstInst =
+ getAsmWriterInstByID(NIT-InstIdxs.begin());
+ if (!FirstInst || FirstInst->Operands.size() == Op)
+ break;
+
+ // Otherwise, scan to see if all of the other instructions in this command
+ // set share the operand.
+ bool AllSame = true;
+
+ for (NIT = std::find(NIT+1, InstIdxs.end(), CommandIdx);
+ NIT != InstIdxs.end();
+ NIT = std::find(NIT+1, InstIdxs.end(), CommandIdx)) {
+ // Okay, found another instruction in this command set. If the operand
+ // matches, we're ok, otherwise bail out.
+ const AsmWriterInst *OtherInst =
+ getAsmWriterInstByID(NIT-InstIdxs.begin());
+
+ if (!OtherInst || OtherInst->Operands.size() == Op ||
+ OtherInst->Operands[Op] != FirstInst->Operands[Op]) {
+ AllSame = false;
+ break;
+ }
+ }
+ if (!AllSame) break;
+
+ // Okay, everything in this command set has the same next operand. Add it
+ // to UniqueOperandCommands and remember that it was consumed.
+ std::string Command = " " + FirstInst->Operands[Op].getCode() + "\n";
+
+ UniqueOperandCommands[CommandIdx] += Command;
+ InstOpsUsed[CommandIdx]++;
+ }
+ }
+
+ // Prepend some of the instructions each case is used for onto the case val.
+ for (unsigned i = 0, e = InstrsForCase.size(); i != e; ++i) {
+ std::string Instrs = InstrsForCase[i];
+ if (Instrs.size() > 70) {
+ Instrs.erase(Instrs.begin()+70, Instrs.end());
+ Instrs += "...";
+ }
+
+ if (!Instrs.empty())
+ UniqueOperandCommands[i] = " // " + Instrs + "\n" +
+ UniqueOperandCommands[i];
+ }
+}
+
+
+static void UnescapeString(std::string &Str) {
+ for (unsigned i = 0; i != Str.size(); ++i) {
+ if (Str[i] == '\\' && i != Str.size()-1) {
+ switch (Str[i+1]) {
+ default: continue; // Don't execute the code after the switch.
+ case 'a': Str[i] = '\a'; break;
+ case 'b': Str[i] = '\b'; break;
+ case 'e': Str[i] = 27; break;
+ case 'f': Str[i] = '\f'; break;
+ case 'n': Str[i] = '\n'; break;
+ case 'r': Str[i] = '\r'; break;
+ case 't': Str[i] = '\t'; break;
+ case 'v': Str[i] = '\v'; break;
+ case '"': Str[i] = '\"'; break;
+ case '\'': Str[i] = '\''; break;
+ case '\\': Str[i] = '\\'; break;
+ }
+ // Nuke the second character.
+ Str.erase(Str.begin()+i+1);
+ }
+ }
+}
+
+/// EmitPrintInstruction - Generate the code for the "printInstruction" method
+/// implementation. Destroys all instances of AsmWriterInst information, by
+/// clearing the Instructions vector.
+void AsmWriterEmitter::EmitPrintInstruction(raw_ostream &O) {
Record *AsmWriter = Target.getAsmWriter();
std::string ClassName = AsmWriter->getValueAsString("AsmWriterClassName");
- unsigned Variant = AsmWriter->getValueAsInt("Variant");
+ unsigned PassSubtarget = AsmWriter->getValueAsInt("PassSubtarget");
O <<
"/// printInstruction - This method is automatically generated by tablegen\n"
- "/// from the instruction set description. This method returns true if the\n"
- "/// machine instruction was sufficiently described to print it, otherwise\n"
- "/// it returns false.\n"
- "bool " << Target.getName() << ClassName
- << "::printInstruction(const MachineInstr *MI) {\n";
+ "/// from the instruction set description.\n"
+ "void " << Target.getName() << ClassName
+ << "::printInstruction(const MCInst *MI, "
+ << (PassSubtarget ? "const MCSubtargetInfo &STI, " : "")
+ << "raw_ostream &O) {\n";
- std::vector<AsmWriterInst> Instructions;
+ // Build an aggregate string, and build a table of offsets into it.
+ SequenceToOffsetTable<std::string> StringTable;
- for (CodeGenTarget::inst_iterator I = Target.inst_begin(),
- E = Target.inst_end(); I != E; ++I)
- if (!I->second.AsmString.empty())
- Instructions.push_back(AsmWriterInst(I->second, Variant));
+ /// OpcodeInfo - This encodes the index of the string to use for the first
+ /// chunk of the output as well as indices used for operand printing.
+ std::vector<uint64_t> OpcodeInfo;
+ const unsigned OpcodeInfoBits = 64;
- std::vector<const CodeGenInstruction*> NumberedInstructions;
- Target.getInstructionsByEnumValue(NumberedInstructions);
-
- // Compute the CodeGenInstruction -> AsmWriterInst mapping. Note that not
- // all machine instructions are necessarily being printed, so there may be
- // target instructions not in this map.
- std::map<const CodeGenInstruction*, AsmWriterInst*> CGIAWIMap;
- for (unsigned i = 0, e = Instructions.size(); i != e; ++i)
- CGIAWIMap.insert(std::make_pair(Instructions[i].CGI, &Instructions[i]));
+ // Add all strings to the string table upfront so it can generate an optimized
+ // representation.
+ for (const CodeGenInstruction *Inst : *NumberedInstructions) {
+ AsmWriterInst *AWI = CGIAWIMap[Inst];
+ if (AWI &&
+ AWI->Operands[0].OperandType ==
+ AsmWriterOperand::isLiteralTextOperand &&
+ !AWI->Operands[0].Str.empty()) {
+ std::string Str = AWI->Operands[0].Str;
+ UnescapeString(Str);
+ StringTable.add(Str);
+ }
+ }
- // Build an aggregate string, and build a table of offsets into it.
- std::map<std::string, unsigned> StringOffset;
- std::string AggregateString;
- AggregateString += '\0';
-
- O << " static unsigned short OpStrIdxs[] = {\n";
- for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
- AsmWriterInst *AWI = CGIAWIMap[NumberedInstructions[i]];
+ StringTable.layout();
+
+ unsigned MaxStringIdx = 0;
+ for (const CodeGenInstruction *Inst : *NumberedInstructions) {
+ AsmWriterInst *AWI = CGIAWIMap[Inst];
unsigned Idx;
- if (AWI == 0 || AWI->Operands[0].Str.empty()) {
+ if (!AWI) {
// Something not handled by the asmwriter printer.
- Idx = 0;
+ Idx = ~0U;
+ } else if (AWI->Operands[0].OperandType !=
+ AsmWriterOperand::isLiteralTextOperand ||
+ AWI->Operands[0].Str.empty()) {
+ // Something handled by the asmwriter printer, but with no leading string.
+ Idx = StringTable.get("");
} else {
- unsigned &Entry = StringOffset[AWI->Operands[0].Str];
- if (Entry == 0) {
- // Add the string to the aggregate if this is the first time found.
- Entry = AggregateString.size();
- std::string Str = AWI->Operands[0].Str;
- UnescapeString(Str);
- AggregateString += Str;
- AggregateString += '\0';
- }
- Idx = Entry;
- assert(Entry < 65536 && "Must not use unsigned short for table idx!");
+ std::string Str = AWI->Operands[0].Str;
+ UnescapeString(Str);
+ Idx = StringTable.get(Str);
+ MaxStringIdx = std::max(MaxStringIdx, Idx);
// Nuke the string from the operand list. It is now handled!
AWI->Operands.erase(AWI->Operands.begin());
}
- O << " " << Idx << ",\t// " << NumberedInstructions[i]->TheDef->getName()
- << "\n";
+
+ // Bias offset by one since we want 0 as a sentinel.
+ OpcodeInfo.push_back(Idx+1);
+ }
+
+ // Figure out how many bits we used for the string index.
+ unsigned AsmStrBits = Log2_32_Ceil(MaxStringIdx+2);
+
+ // To reduce code size, we compactify common instructions into a few bits
+ // in the opcode-indexed table.
+ unsigned BitsLeft = OpcodeInfoBits-AsmStrBits;
+
+ std::vector<std::vector<std::string>> TableDrivenOperandPrinters;
+
+ while (1) {
+ std::vector<std::string> UniqueOperandCommands;
+ std::vector<unsigned> InstIdxs;
+ std::vector<unsigned> NumInstOpsHandled;
+ FindUniqueOperandCommands(UniqueOperandCommands, InstIdxs,
+ NumInstOpsHandled);
+
+ // If we ran out of operands to print, we're done.
+ if (UniqueOperandCommands.empty()) break;
+
+ // Compute the number of bits we need to represent these cases, this is
+ // ceil(log2(numentries)).
+ unsigned NumBits = Log2_32_Ceil(UniqueOperandCommands.size());
+
+ // If we don't have enough bits for this operand, don't include it.
+ if (NumBits > BitsLeft) {
+ DEBUG(errs() << "Not enough bits to densely encode " << NumBits
+ << " more bits\n");
+ break;
+ }
+
+ // Otherwise, we can include this in the initial lookup table. Add it in.
+ for (unsigned i = 0, e = InstIdxs.size(); i != e; ++i)
+ if (InstIdxs[i] != ~0U) {
+ OpcodeInfo[i] |= (uint64_t)InstIdxs[i] << (OpcodeInfoBits-BitsLeft);
+ }
+ BitsLeft -= NumBits;
+
+ // Remove the info about this operand.
+ for (unsigned i = 0, e = NumberedInstructions->size(); i != e; ++i) {
+ if (AsmWriterInst *Inst = getAsmWriterInstByID(i))
+ if (!Inst->Operands.empty()) {
+ unsigned NumOps = NumInstOpsHandled[InstIdxs[i]];
+ assert(NumOps <= Inst->Operands.size() &&
+ "Can't remove this many ops!");
+ Inst->Operands.erase(Inst->Operands.begin(),
+ Inst->Operands.begin()+NumOps);
+ }
+ }
+
+ // Remember the handlers for this set of operands.
+ TableDrivenOperandPrinters.push_back(std::move(UniqueOperandCommands));
}
+
+ // Emit the string table itself.
+ O << " static const char AsmStrs[] = {\n";
+ StringTable.emit(O, printChar);
O << " };\n\n";
-
- // Emit the string itself.
- O << " const char *AsmStrs = \n \"";
- unsigned CharsPrinted = 0;
- EscapeString(AggregateString);
- for (unsigned i = 0, e = AggregateString.size(); i != e; ++i) {
- if (CharsPrinted > 70) {
- O << "\"\n \"";
- CharsPrinted = 0;
+
+ // Emit the lookup tables in pieces to minimize wasted bytes.
+ unsigned BytesNeeded = ((OpcodeInfoBits - BitsLeft) + 7) / 8;
+ unsigned Table = 0, Shift = 0;
+ SmallString<128> BitsString;
+ raw_svector_ostream BitsOS(BitsString);
+ // If the total bits is more than 32-bits we need to use a 64-bit type.
+ BitsOS << " uint" << ((BitsLeft < (OpcodeInfoBits - 32)) ? 64 : 32)
+ << "_t Bits = 0;\n";
+ while (BytesNeeded != 0) {
+ // Figure out how big this table section needs to be, but no bigger than 4.
+ unsigned TableSize = std::min(1 << Log2_32(BytesNeeded), 4);
+ BytesNeeded -= TableSize;
+ TableSize *= 8; // Convert to bits;
+ uint64_t Mask = (1ULL << TableSize) - 1;
+ O << " static const uint" << TableSize << "_t OpInfo" << Table
+ << "[] = {\n";
+ for (unsigned i = 0, e = NumberedInstructions->size(); i != e; ++i) {
+ O << " " << ((OpcodeInfo[i] >> Shift) & Mask) << "U,\t// "
+ << NumberedInstructions->at(i)->TheDef->getName() << "\n";
}
- O << AggregateString[i];
- ++CharsPrinted;
-
- // Print escape sequences all together.
- if (AggregateString[i] == '\\') {
- assert(i+1 < AggregateString.size() && "Incomplete escape sequence!");
- if (isdigit(AggregateString[i+1])) {
- assert(isdigit(AggregateString[i+2]) && isdigit(AggregateString[i+3]) &&
- "Expected 3 digit octal escape!");
- O << AggregateString[++i];
- O << AggregateString[++i];
- O << AggregateString[++i];
- CharsPrinted += 3;
- } else {
- O << AggregateString[++i];
- ++CharsPrinted;
+ O << " };\n\n";
+ // Emit string to combine the individual table lookups.
+ BitsOS << " Bits |= ";
+ // If the total bits is more than 32-bits we need to use a 64-bit type.
+ if (BitsLeft < (OpcodeInfoBits - 32))
+ BitsOS << "(uint64_t)";
+ BitsOS << "OpInfo" << Table << "[MI->getOpcode()] << " << Shift << ";\n";
+ // Prepare the shift for the next iteration and increment the table count.
+ Shift += TableSize;
+ ++Table;
+ }
+
+ // Emit the initial tab character.
+ O << " O << \"\\t\";\n\n";
+
+ O << " // Emit the opcode for the instruction.\n";
+ O << BitsString;
+
+ // Emit the starting string.
+ O << " assert(Bits != 0 && \"Cannot print this instruction.\");\n"
+ << " O << AsmStrs+(Bits & " << (1 << AsmStrBits)-1 << ")-1;\n\n";
+
+ // Output the table driven operand information.
+ BitsLeft = OpcodeInfoBits-AsmStrBits;
+ for (unsigned i = 0, e = TableDrivenOperandPrinters.size(); i != e; ++i) {
+ std::vector<std::string> &Commands = TableDrivenOperandPrinters[i];
+
+ // Compute the number of bits we need to represent these cases, this is
+ // ceil(log2(numentries)).
+ unsigned NumBits = Log2_32_Ceil(Commands.size());
+ assert(NumBits <= BitsLeft && "consistency error");
+
+ // Emit code to extract this field from Bits.
+ O << "\n // Fragment " << i << " encoded into " << NumBits
+ << " bits for " << Commands.size() << " unique commands.\n";
+
+ if (Commands.size() == 2) {
+ // Emit two possibilitys with if/else.
+ O << " if ((Bits >> "
+ << (OpcodeInfoBits-BitsLeft) << ") & "
+ << ((1 << NumBits)-1) << ") {\n"
+ << Commands[1]
+ << " } else {\n"
+ << Commands[0]
+ << " }\n\n";
+ } else if (Commands.size() == 1) {
+ // Emit a single possibility.
+ O << Commands[0] << "\n\n";
+ } else {
+ O << " switch ((Bits >> "
+ << (OpcodeInfoBits-BitsLeft) << ") & "
+ << ((1 << NumBits)-1) << ") {\n"
+ << " default: llvm_unreachable(\"Invalid command number.\");\n";
+
+ // Print out all the cases.
+ for (unsigned j = 0, e = Commands.size(); j != e; ++j) {
+ O << " case " << j << ":\n";
+ O << Commands[j];
+ O << " break;\n";
}
+ O << " }\n\n";
}
+ BitsLeft -= NumBits;
}
- O << "\";\n\n";
- O << " // Emit the opcode for the instruction.\n"
- << " O << AsmStrs+OpStrIdxs[MI->getOpcode()];\n\n";
+ // Okay, delete instructions with no operand info left.
+ auto I = std::remove_if(Instructions.begin(), Instructions.end(),
+ [](AsmWriterInst &Inst) {
+ return Inst.Operands.empty();
+ });
+ Instructions.erase(I, Instructions.end());
- // Because this is a vector we want to emit from the end. Reverse all of the
+
+ // Because this is a vector, we want to emit from the end. Reverse all of the
// elements in the vector.
std::reverse(Instructions.begin(), Instructions.end());
- // Find the opcode # of inline asm
- O << " switch (MI->getOpcode()) {\n"
- " default: return false;\n"
- " case " << NumberedInstructions.back()->Namespace
- << "::INLINEASM: printInlineAsm(MI); break;\n";
- while (!Instructions.empty())
- EmitInstructions(Instructions, O);
+ // Now that we've emitted all of the operand info that fit into 64 bits, emit
+ // information for those instructions that are left. This is a less dense
+ // encoding, but we expect the main 64-bit table to handle the majority of
+ // instructions.
+ if (!Instructions.empty()) {
+ // Find the opcode # of inline asm.
+ O << " switch (MI->getOpcode()) {\n";
+ while (!Instructions.empty())
+ EmitInstructions(Instructions, O);
+
+ O << " }\n";
+ O << " return;\n";
+ }
+
+ O << "}\n";
+}
+
+static const char *getMinimalTypeForRange(uint64_t Range) {
+ assert(Range < 0xFFFFFFFFULL && "Enum too large");
+ if (Range > 0xFFFF)
+ return "uint32_t";
+ if (Range > 0xFF)
+ return "uint16_t";
+ return "uint8_t";
+}
+
+static void
+emitRegisterNameString(raw_ostream &O, StringRef AltName,
+ const std::deque<CodeGenRegister> &Registers) {
+ SequenceToOffsetTable<std::string> StringTable;
+ SmallVector<std::string, 4> AsmNames(Registers.size());
+ unsigned i = 0;
+ for (const auto &Reg : Registers) {
+ std::string &AsmName = AsmNames[i++];
+
+ // "NoRegAltName" is special. We don't need to do a lookup for that,
+ // as it's just a reference to the default register name.
+ if (AltName == "" || AltName == "NoRegAltName") {
+ AsmName = Reg.TheDef->getValueAsString("AsmName");
+ if (AsmName.empty())
+ AsmName = Reg.getName();
+ } else {
+ // Make sure the register has an alternate name for this index.
+ std::vector<Record*> AltNameList =
+ Reg.TheDef->getValueAsListOfDefs("RegAltNameIndices");
+ unsigned Idx = 0, e;
+ for (e = AltNameList.size();
+ Idx < e && (AltNameList[Idx]->getName() != AltName);
+ ++Idx)
+ ;
+ // If the register has an alternate name for this index, use it.
+ // Otherwise, leave it empty as an error flag.
+ if (Idx < e) {
+ std::vector<std::string> AltNames =
+ Reg.TheDef->getValueAsListOfStrings("AltNames");
+ if (AltNames.size() <= Idx)
+ PrintFatalError(Reg.TheDef->getLoc(),
+ "Register definition missing alt name for '" +
+ AltName + "'.");
+ AsmName = AltNames[Idx];
+ }
+ }
+ StringTable.add(AsmName);
+ }
+
+ StringTable.layout();
+ O << " static const char AsmStrs" << AltName << "[] = {\n";
+ StringTable.emit(O, printChar);
+ O << " };\n\n";
+
+ O << " static const " << getMinimalTypeForRange(StringTable.size()-1)
+ << " RegAsmOffset" << AltName << "[] = {";
+ for (unsigned i = 0, e = Registers.size(); i != e; ++i) {
+ if ((i % 14) == 0)
+ O << "\n ";
+ O << StringTable.get(AsmNames[i]) << ", ";
+ }
+ O << "\n };\n"
+ << "\n";
+}
+
+void AsmWriterEmitter::EmitGetRegisterName(raw_ostream &O) {
+ Record *AsmWriter = Target.getAsmWriter();
+ std::string ClassName = AsmWriter->getValueAsString("AsmWriterClassName");
+ const auto &Registers = Target.getRegBank().getRegisters();
+ std::vector<Record*> AltNameIndices = Target.getRegAltNameIndices();
+ bool hasAltNames = AltNameIndices.size() > 1;
+ std::string Namespace =
+ Registers.front().TheDef->getValueAsString("Namespace");
+
+ O <<
+ "\n\n/// getRegisterName - This method is automatically generated by tblgen\n"
+ "/// from the register set description. This returns the assembler name\n"
+ "/// for the specified register.\n"
+ "const char *" << Target.getName() << ClassName << "::";
+ if (hasAltNames)
+ O << "\ngetRegisterName(unsigned RegNo, unsigned AltIdx) {\n";
+ else
+ O << "getRegisterName(unsigned RegNo) {\n";
+ O << " assert(RegNo && RegNo < " << (Registers.size()+1)
+ << " && \"Invalid register number!\");\n"
+ << "\n";
+
+ if (hasAltNames) {
+ for (const Record *R : AltNameIndices)
+ emitRegisterNameString(O, R->getName(), Registers);
+ } else
+ emitRegisterNameString(O, "", Registers);
- O << " }\n"
- " return true;\n"
- "}\n";
+ if (hasAltNames) {
+ O << " switch(AltIdx) {\n"
+ << " default: llvm_unreachable(\"Invalid register alt name index!\");\n";
+ for (const Record *R : AltNameIndices) {
+ std::string AltName(R->getName());
+ std::string Prefix = !Namespace.empty() ? Namespace + "::" : "";
+ O << " case " << Prefix << AltName << ":\n"
+ << " assert(*(AsmStrs" << AltName << "+RegAsmOffset"
+ << AltName << "[RegNo-1]) &&\n"
+ << " \"Invalid alt name index for register!\");\n"
+ << " return AsmStrs" << AltName << "+RegAsmOffset"
+ << AltName << "[RegNo-1];\n";
+ }
+ O << " }\n";
+ } else {
+ O << " assert (*(AsmStrs+RegAsmOffset[RegNo-1]) &&\n"
+ << " \"Invalid alt name index for register!\");\n"
+ << " return AsmStrs+RegAsmOffset[RegNo-1];\n";
+ }
+ O << "}\n";
}
+
+namespace {
+// IAPrinter - Holds information about an InstAlias. Two InstAliases match if
+// they both have the same conditionals. In which case, we cannot print out the
+// alias for that pattern.
+class IAPrinter {
+ std::vector<std::string> Conds;
+ std::map<StringRef, std::pair<int, int>> OpMap;
+ SmallVector<Record*, 4> ReqFeatures;
+
+ std::string Result;
+ std::string AsmString;
+public:
+ IAPrinter(std::string R, std::string AS) : Result(R), AsmString(AS) {}
+
+ void addCond(const std::string &C) { Conds.push_back(C); }
+
+ void addOperand(StringRef Op, int OpIdx, int PrintMethodIdx = -1) {
+ assert(OpIdx >= 0 && OpIdx < 0xFE && "Idx out of range");
+ assert(PrintMethodIdx >= -1 && PrintMethodIdx < 0xFF &&
+ "Idx out of range");
+ OpMap[Op] = std::make_pair(OpIdx, PrintMethodIdx);
+ }
+
+ bool isOpMapped(StringRef Op) { return OpMap.find(Op) != OpMap.end(); }
+ int getOpIndex(StringRef Op) { return OpMap[Op].first; }
+ std::pair<int, int> &getOpData(StringRef Op) { return OpMap[Op]; }
+
+ std::pair<StringRef, StringRef::iterator> parseName(StringRef::iterator Start,
+ StringRef::iterator End) {
+ StringRef::iterator I = Start;
+ StringRef::iterator Next;
+ if (*I == '{') {
+ // ${some_name}
+ Start = ++I;
+ while (I != End && *I != '}')
+ ++I;
+ Next = I;
+ // eat the final '}'
+ if (Next != End)
+ ++Next;
+ } else {
+ // $name, just eat the usual suspects.
+ while (I != End &&
+ ((*I >= 'a' && *I <= 'z') || (*I >= 'A' && *I <= 'Z') ||
+ (*I >= '0' && *I <= '9') || *I == '_'))
+ ++I;
+ Next = I;
+ }
+
+ return std::make_pair(StringRef(Start, I - Start), Next);
+ }
+
+ void print(raw_ostream &O) {
+ if (Conds.empty() && ReqFeatures.empty()) {
+ O.indent(6) << "return true;\n";
+ return;
+ }
+
+ O << "if (";
+
+ for (std::vector<std::string>::iterator
+ I = Conds.begin(), E = Conds.end(); I != E; ++I) {
+ if (I != Conds.begin()) {
+ O << " &&\n";
+ O.indent(8);
+ }
+
+ O << *I;
+ }
+
+ O << ") {\n";
+ O.indent(6) << "// " << Result << "\n";
+
+ // Directly mangle mapped operands into the string. Each operand is
+ // identified by a '$' sign followed by a byte identifying the number of the
+ // operand. We add one to the index to avoid zero bytes.
+ StringRef ASM(AsmString);
+ SmallString<128> OutString;
+ raw_svector_ostream OS(OutString);
+ for (StringRef::iterator I = ASM.begin(), E = ASM.end(); I != E;) {
+ OS << *I;
+ if (*I == '$') {
+ StringRef Name;
+ std::tie(Name, I) = parseName(++I, E);
+ assert(isOpMapped(Name) && "Unmapped operand!");
+
+ int OpIndex, PrintIndex;
+ std::tie(OpIndex, PrintIndex) = getOpData(Name);
+ if (PrintIndex == -1) {
+ // Can use the default printOperand route.
+ OS << format("\\x%02X", (unsigned char)OpIndex + 1);
+ } else
+ // 3 bytes if a PrintMethod is needed: 0xFF, the MCInst operand
+ // number, and which of our pre-detected Methods to call.
+ OS << format("\\xFF\\x%02X\\x%02X", OpIndex + 1, PrintIndex + 1);
+ } else {
+ ++I;
+ }
+ }
+
+ // Emit the string.
+ O.indent(6) << "AsmString = \"" << OutString << "\";\n";
+
+ O.indent(6) << "break;\n";
+ O.indent(4) << '}';
+ }
+
+ bool operator==(const IAPrinter &RHS) const {
+ if (Conds.size() != RHS.Conds.size())
+ return false;
+
+ unsigned Idx = 0;
+ for (const auto &str : Conds)
+ if (str != RHS.Conds[Idx++])
+ return false;
+
+ return true;
+ }
+};
+
+} // end anonymous namespace
+
+static unsigned CountNumOperands(StringRef AsmString, unsigned Variant) {
+ std::string FlatAsmString =
+ CodeGenInstruction::FlattenAsmStringVariants(AsmString, Variant);
+ AsmString = FlatAsmString;
+
+ return AsmString.count(' ') + AsmString.count('\t');
+}
+
+namespace {
+struct AliasPriorityComparator {
+ typedef std::pair<CodeGenInstAlias, int> ValueType;
+ bool operator()(const ValueType &LHS, const ValueType &RHS) {
+ if (LHS.second == RHS.second) {
+ // We don't actually care about the order, but for consistency it
+ // shouldn't depend on pointer comparisons.
+ return LHS.first.TheDef->getName() < RHS.first.TheDef->getName();
+ }
+
+ // Aliases with larger priorities should be considered first.
+ return LHS.second > RHS.second;
+ }
+};
+}
+
+
+void AsmWriterEmitter::EmitPrintAliasInstruction(raw_ostream &O) {
+ Record *AsmWriter = Target.getAsmWriter();
+
+ O << "\n#ifdef PRINT_ALIAS_INSTR\n";
+ O << "#undef PRINT_ALIAS_INSTR\n\n";
+
+ //////////////////////////////
+ // Gather information about aliases we need to print
+ //////////////////////////////
+
+ // Emit the method that prints the alias instruction.
+ std::string ClassName = AsmWriter->getValueAsString("AsmWriterClassName");
+ unsigned Variant = AsmWriter->getValueAsInt("Variant");
+ unsigned PassSubtarget = AsmWriter->getValueAsInt("PassSubtarget");
+
+ std::vector<Record*> AllInstAliases =
+ Records.getAllDerivedDefinitions("InstAlias");
+
+ // Create a map from the qualified name to a list of potential matches.
+ typedef std::set<std::pair<CodeGenInstAlias, int>, AliasPriorityComparator>
+ AliasWithPriority;
+ std::map<std::string, AliasWithPriority> AliasMap;
+ for (Record *R : AllInstAliases) {
+ int Priority = R->getValueAsInt("EmitPriority");
+ if (Priority < 1)
+ continue; // Aliases with priority 0 are never emitted.
+
+ const DagInit *DI = R->getValueAsDag("ResultInst");
+ const DefInit *Op = cast<DefInit>(DI->getOperator());
+ AliasMap[getQualifiedName(Op->getDef())].insert(
+ std::make_pair(CodeGenInstAlias(R, Variant, Target), Priority));
+ }
+
+ // A map of which conditions need to be met for each instruction operand
+ // before it can be matched to the mnemonic.
+ std::map<std::string, std::vector<IAPrinter>> IAPrinterMap;
+
+ // A list of MCOperandPredicates for all operands in use, and the reverse map
+ std::vector<const Record*> MCOpPredicates;
+ DenseMap<const Record*, unsigned> MCOpPredicateMap;
+
+ for (auto &Aliases : AliasMap) {
+ for (auto &Alias : Aliases.second) {
+ const CodeGenInstAlias &CGA = Alias.first;
+ unsigned LastOpNo = CGA.ResultInstOperandIndex.size();
+ unsigned NumResultOps =
+ CountNumOperands(CGA.ResultInst->AsmString, Variant);
+
+ // Don't emit the alias if it has more operands than what it's aliasing.
+ if (NumResultOps < CountNumOperands(CGA.AsmString, Variant))
+ continue;
+
+ IAPrinter IAP(CGA.Result->getAsString(), CGA.AsmString);
+
+ unsigned NumMIOps = 0;
+ for (auto &Operand : CGA.ResultOperands)
+ NumMIOps += Operand.getMINumOperands();
+
+ std::string Cond;
+ Cond = std::string("MI->getNumOperands() == ") + llvm::utostr(NumMIOps);
+ IAP.addCond(Cond);
+
+ bool CantHandle = false;
+
+ unsigned MIOpNum = 0;
+ for (unsigned i = 0, e = LastOpNo; i != e; ++i) {
+ std::string Op = "MI->getOperand(" + llvm::utostr(MIOpNum) + ")";
+
+ const CodeGenInstAlias::ResultOperand &RO = CGA.ResultOperands[i];
+
+ switch (RO.Kind) {
+ case CodeGenInstAlias::ResultOperand::K_Record: {
+ const Record *Rec = RO.getRecord();
+ StringRef ROName = RO.getName();
+ int PrintMethodIdx = -1;
+
+ // These two may have a PrintMethod, which we want to record (if it's
+ // the first time we've seen it) and provide an index for the aliasing
+ // code to use.
+ if (Rec->isSubClassOf("RegisterOperand") ||
+ Rec->isSubClassOf("Operand")) {
+ std::string PrintMethod = Rec->getValueAsString("PrintMethod");
+ if (PrintMethod != "" && PrintMethod != "printOperand") {
+ PrintMethodIdx = std::find(PrintMethods.begin(),
+ PrintMethods.end(), PrintMethod) -
+ PrintMethods.begin();
+ if (static_cast<unsigned>(PrintMethodIdx) == PrintMethods.size())
+ PrintMethods.push_back(PrintMethod);
+ }
+ }
+
+ if (Rec->isSubClassOf("RegisterOperand"))
+ Rec = Rec->getValueAsDef("RegClass");
+ if (Rec->isSubClassOf("RegisterClass")) {
+ IAP.addCond(Op + ".isReg()");
+
+ if (!IAP.isOpMapped(ROName)) {
+ IAP.addOperand(ROName, MIOpNum, PrintMethodIdx);
+ Record *R = CGA.ResultOperands[i].getRecord();
+ if (R->isSubClassOf("RegisterOperand"))
+ R = R->getValueAsDef("RegClass");
+ Cond = std::string("MRI.getRegClass(") + Target.getName() + "::" +
+ R->getName() + "RegClassID)"
+ ".contains(" + Op + ".getReg())";
+ } else {
+ Cond = Op + ".getReg() == MI->getOperand(" +
+ llvm::utostr(IAP.getOpIndex(ROName)) + ").getReg()";
+ }
+ } else {
+ // Assume all printable operands are desired for now. This can be
+ // overridden in the InstAlias instantiation if necessary.
+ IAP.addOperand(ROName, MIOpNum, PrintMethodIdx);
+
+ // There might be an additional predicate on the MCOperand
+ unsigned Entry = MCOpPredicateMap[Rec];
+ if (!Entry) {
+ if (!Rec->isValueUnset("MCOperandPredicate")) {
+ MCOpPredicates.push_back(Rec);
+ Entry = MCOpPredicates.size();
+ MCOpPredicateMap[Rec] = Entry;
+ } else
+ break; // No conditions on this operand at all
+ }
+ Cond = Target.getName() + ClassName + "ValidateMCOperand(" +
+ Op + ", STI, " + llvm::utostr(Entry) + ")";
+ }
+ // for all subcases of ResultOperand::K_Record:
+ IAP.addCond(Cond);
+ break;
+ }
+ case CodeGenInstAlias::ResultOperand::K_Imm: {
+ // Just because the alias has an immediate result, doesn't mean the
+ // MCInst will. An MCExpr could be present, for example.
+ IAP.addCond(Op + ".isImm()");
+
+ Cond = Op + ".getImm() == " +
+ llvm::utostr(CGA.ResultOperands[i].getImm());
+ IAP.addCond(Cond);
+ break;
+ }
+ case CodeGenInstAlias::ResultOperand::K_Reg:
+ // If this is zero_reg, something's playing tricks we're not
+ // equipped to handle.
+ if (!CGA.ResultOperands[i].getRegister()) {
+ CantHandle = true;
+ break;
+ }
+
+ Cond = Op + ".getReg() == " + Target.getName() + "::" +
+ CGA.ResultOperands[i].getRegister()->getName();
+ IAP.addCond(Cond);
+ break;
+ }
+
+ MIOpNum += RO.getMINumOperands();
+ }
+
+ if (CantHandle) continue;
+ IAPrinterMap[Aliases.first].push_back(std::move(IAP));
+ }
+ }
+
+ //////////////////////////////
+ // Write out the printAliasInstr function
+ //////////////////////////////
+
+ std::string Header;
+ raw_string_ostream HeaderO(Header);
+
+ HeaderO << "bool " << Target.getName() << ClassName
+ << "::printAliasInstr(const MCInst"
+ << " *MI, " << (PassSubtarget ? "const MCSubtargetInfo &STI, " : "")
+ << "raw_ostream &OS) {\n";
+
+ std::string Cases;
+ raw_string_ostream CasesO(Cases);
+
+ for (auto &Entry : IAPrinterMap) {
+ std::vector<IAPrinter> &IAPs = Entry.second;
+ std::vector<IAPrinter*> UniqueIAPs;
+
+ for (auto &LHS : IAPs) {
+ bool IsDup = false;
+ for (const auto &RHS : IAPs) {
+ if (&LHS != &RHS && LHS == RHS) {
+ IsDup = true;
+ break;
+ }
+ }
+
+ if (!IsDup)
+ UniqueIAPs.push_back(&LHS);
+ }
+
+ if (UniqueIAPs.empty()) continue;
+
+ CasesO.indent(2) << "case " << Entry.first << ":\n";
+
+ for (IAPrinter *IAP : UniqueIAPs) {
+ CasesO.indent(4);
+ IAP->print(CasesO);
+ CasesO << '\n';
+ }
+
+ CasesO.indent(4) << "return false;\n";
+ }
+
+ if (CasesO.str().empty()) {
+ O << HeaderO.str();
+ O << " return false;\n";
+ O << "}\n\n";
+ O << "#endif // PRINT_ALIAS_INSTR\n";
+ return;
+ }
+
+ if (!MCOpPredicates.empty())
+ O << "static bool " << Target.getName() << ClassName
+ << "ValidateMCOperand(const MCOperand &MCOp,\n"
+ << " const MCSubtargetInfo &STI,\n"
+ << " unsigned PredicateIndex);\n";
+
+ O << HeaderO.str();
+ O.indent(2) << "const char *AsmString;\n";
+ O.indent(2) << "switch (MI->getOpcode()) {\n";
+ O.indent(2) << "default: return false;\n";
+ O << CasesO.str();
+ O.indent(2) << "}\n\n";
+
+ // Code that prints the alias, replacing the operands with the ones from the
+ // MCInst.
+ O << " unsigned I = 0;\n";
+ O << " while (AsmString[I] != ' ' && AsmString[I] != '\t' &&\n";
+ O << " AsmString[I] != '\\0')\n";
+ O << " ++I;\n";
+ O << " OS << '\\t' << StringRef(AsmString, I);\n";
+
+ O << " if (AsmString[I] != '\\0') {\n";
+ O << " OS << '\\t';\n";
+ O << " do {\n";
+ O << " if (AsmString[I] == '$') {\n";
+ O << " ++I;\n";
+ O << " if (AsmString[I] == (char)0xff) {\n";
+ O << " ++I;\n";
+ O << " int OpIdx = AsmString[I++] - 1;\n";
+ O << " int PrintMethodIdx = AsmString[I++] - 1;\n";
+ O << " printCustomAliasOperand(MI, OpIdx, PrintMethodIdx, ";
+ O << (PassSubtarget ? "STI, " : "");
+ O << "OS);\n";
+ O << " } else\n";
+ O << " printOperand(MI, unsigned(AsmString[I++]) - 1, ";
+ O << (PassSubtarget ? "STI, " : "");
+ O << "OS);\n";
+ O << " } else {\n";
+ O << " OS << AsmString[I++];\n";
+ O << " }\n";
+ O << " } while (AsmString[I] != '\\0');\n";
+ O << " }\n\n";
+
+ O << " return true;\n";
+ O << "}\n\n";
+
+ //////////////////////////////
+ // Write out the printCustomAliasOperand function
+ //////////////////////////////
+
+ O << "void " << Target.getName() << ClassName << "::"
+ << "printCustomAliasOperand(\n"
+ << " const MCInst *MI, unsigned OpIdx,\n"
+ << " unsigned PrintMethodIdx,\n"
+ << (PassSubtarget ? " const MCSubtargetInfo &STI,\n" : "")
+ << " raw_ostream &OS) {\n";
+ if (PrintMethods.empty())
+ O << " llvm_unreachable(\"Unknown PrintMethod kind\");\n";
+ else {
+ O << " switch (PrintMethodIdx) {\n"
+ << " default:\n"
+ << " llvm_unreachable(\"Unknown PrintMethod kind\");\n"
+ << " break;\n";
+
+ for (unsigned i = 0; i < PrintMethods.size(); ++i) {
+ O << " case " << i << ":\n"
+ << " " << PrintMethods[i] << "(MI, OpIdx, "
+ << (PassSubtarget ? "STI, " : "") << "OS);\n"
+ << " break;\n";
+ }
+ O << " }\n";
+ }
+ O << "}\n\n";
+
+ if (!MCOpPredicates.empty()) {
+ O << "static bool " << Target.getName() << ClassName
+ << "ValidateMCOperand(const MCOperand &MCOp,\n"
+ << " const MCSubtargetInfo &STI,\n"
+ << " unsigned PredicateIndex) {\n"
+ << " switch (PredicateIndex) {\n"
+ << " default:\n"
+ << " llvm_unreachable(\"Unknown MCOperandPredicate kind\");\n"
+ << " break;\n";
+
+ for (unsigned i = 0; i < MCOpPredicates.size(); ++i) {
+ Init *MCOpPred = MCOpPredicates[i]->getValueInit("MCOperandPredicate");
+ if (StringInit *SI = dyn_cast<StringInit>(MCOpPred)) {
+ O << " case " << i + 1 << ": {\n"
+ << SI->getValue() << "\n"
+ << " }\n";
+ } else
+ llvm_unreachable("Unexpected MCOperandPredicate field!");
+ }
+ O << " }\n"
+ << "}\n\n";
+ }
+
+ O << "#endif // PRINT_ALIAS_INSTR\n";
+}
+
+AsmWriterEmitter::AsmWriterEmitter(RecordKeeper &R) : Records(R), Target(R) {
+ Record *AsmWriter = Target.getAsmWriter();
+ unsigned Variant = AsmWriter->getValueAsInt("Variant");
+ unsigned PassSubtarget = AsmWriter->getValueAsInt("PassSubtarget");
+ for (const CodeGenInstruction *I : Target.instructions())
+ if (!I->AsmString.empty() && I->TheDef->getName() != "PHI")
+ Instructions.emplace_back(*I, Variant, PassSubtarget);
+
+ // Get the instruction numbering.
+ NumberedInstructions = &Target.getInstructionsByEnumValue();
+
+ // Compute the CodeGenInstruction -> AsmWriterInst mapping. Note that not
+ // all machine instructions are necessarily being printed, so there may be
+ // target instructions not in this map.
+ for (AsmWriterInst &AWI : Instructions)
+ CGIAWIMap.insert(std::make_pair(AWI.CGI, &AWI));
+}
+
+void AsmWriterEmitter::run(raw_ostream &O) {
+ EmitPrintInstruction(O);
+ EmitGetRegisterName(O);
+ EmitPrintAliasInstruction(O);
+}
+
+
+namespace llvm {
+
+void EmitAsmWriter(RecordKeeper &RK, raw_ostream &OS) {
+ emitSourceFileHeader("Assembly Writer Source Fragment", OS);
+ AsmWriterEmitter(RK).run(OS);
+}
+
+} // End llvm namespace