1 //===- EDEmitter.cpp - Generate instruction descriptions for ED -*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This tablegen backend is responsible for emitting a description of each
11 // instruction in a format that the enhanced disassembler can use to tokenize
12 // and parse instructions.
14 //===----------------------------------------------------------------------===//
16 #include "EDEmitter.h"
18 #include "AsmWriterInst.h"
19 #include "CodeGenTarget.h"
22 #include "llvm/Support/ErrorHandling.h"
23 #include "llvm/Support/Format.h"
24 #include "llvm/Support/raw_ostream.h"
29 #define MAX_OPERANDS 5
30 #define MAX_SYNTAXES 2
34 ///////////////////////////////////////////////////////////
35 // Support classes for emitting nested C data structures //
36 ///////////////////////////////////////////////////////////
43 std::vector<std::string> Entries;
45 EnumEmitter(const char *N) : Name(N) {
47 int addEntry(const char *e) {
48 Entries.push_back(std::string(e));
49 return Entries.size() - 1;
51 void emit(raw_ostream &o, unsigned int &i) {
52 o.indent(i) << "enum " << Name.c_str() << " {" << "\n";
55 unsigned int index = 0;
56 unsigned int numEntries = Entries.size();
57 for(index = 0; index < numEntries; ++index) {
58 o.indent(i) << Entries[index];
59 if(index < (numEntries - 1))
65 o.indent(i) << "};" << "\n";
68 void emitAsFlags(raw_ostream &o, unsigned int &i) {
69 o.indent(i) << "enum " << Name.c_str() << " {" << "\n";
72 unsigned int index = 0;
73 unsigned int numEntries = Entries.size();
74 unsigned int flag = 1;
75 for (index = 0; index < numEntries; ++index) {
76 o.indent(i) << Entries[index] << " = " << format("0x%x", flag);
77 if (index < (numEntries - 1))
84 o.indent(i) << "};" << "\n";
91 std::vector<std::string> MemberTypes;
92 std::vector<std::string> MemberNames;
94 StructEmitter(const char *N) : Name(N) {
96 void addMember(const char *t, const char *n) {
97 MemberTypes.push_back(std::string(t));
98 MemberNames.push_back(std::string(n));
100 void emit(raw_ostream &o, unsigned int &i) {
101 o.indent(i) << "struct " << Name.c_str() << " {" << "\n";
104 unsigned int index = 0;
105 unsigned int numMembers = MemberTypes.size();
106 for (index = 0; index < numMembers; ++index) {
107 o.indent(i) << MemberTypes[index] << " " << MemberNames[index] << ";";
112 o.indent(i) << "};" << "\n";
116 class ConstantEmitter {
118 virtual ~ConstantEmitter() { }
119 virtual void emit(raw_ostream &o, unsigned int &i) = 0;
122 class LiteralConstantEmitter : public ConstantEmitter {
126 LiteralConstantEmitter(const char *literal) : Literal(literal) {
128 LiteralConstantEmitter(int literal) {
130 snprintf(buf, 256, "%d", literal);
133 void emit(raw_ostream &o, unsigned int &i) {
138 class CompoundConstantEmitter : public ConstantEmitter {
140 std::vector<ConstantEmitter*> Entries;
142 CompoundConstantEmitter() {
144 ~CompoundConstantEmitter() {
146 unsigned int numEntries = Entries.size();
147 for (index = 0; index < numEntries; ++index) {
148 delete Entries[index];
151 CompoundConstantEmitter &addEntry(ConstantEmitter *e) {
152 Entries.push_back(e);
155 void emit(raw_ostream &o, unsigned int &i) {
160 unsigned int numEntries = Entries.size();
161 for (index = 0; index < numEntries; ++index) {
163 Entries[index]->emit(o, i);
164 if (index < (numEntries - 1))
174 class FlagsConstantEmitter : public ConstantEmitter {
176 std::vector<std::string> Flags;
178 FlagsConstantEmitter() {
180 FlagsConstantEmitter &addEntry(const char *f) {
181 Flags.push_back(std::string(f));
184 void emit(raw_ostream &o, unsigned int &i) {
186 unsigned int numFlags = Flags.size();
190 for (index = 0; index < numFlags; ++index) {
191 o << Flags[index].c_str();
192 if (index < (numFlags - 1))
199 EDEmitter::EDEmitter(RecordKeeper &R) : Records(R) {
202 /// populateOperandOrder - Accepts a CodeGenInstruction and generates its
203 /// AsmWriterInst for the desired assembly syntax, giving an ordered list of
204 /// operands in the order they appear in the printed instruction. Then, for
205 /// each entry in that list, determines the index of the same operand in the
206 /// CodeGenInstruction, and emits the resulting mapping into an array, filling
207 /// in unused slots with -1.
209 /// @arg operandOrder - The array that will be populated with the operand
210 /// mapping. Each entry will contain -1 (invalid index
211 /// into the operands present in the AsmString) or a number
212 /// representing an index in the operand descriptor array.
213 /// @arg inst - The instruction to use when looking up the operands
214 /// @arg syntax - The syntax to use, according to LLVM's enumeration
215 void populateOperandOrder(CompoundConstantEmitter *operandOrder,
216 const CodeGenInstruction &inst,
218 unsigned int numArgs = 0;
220 AsmWriterInst awInst(inst, syntax, -1, -1);
222 std::vector<AsmWriterOperand>::iterator operandIterator;
224 for (operandIterator = awInst.Operands.begin();
225 operandIterator != awInst.Operands.end();
227 if (operandIterator->OperandType ==
228 AsmWriterOperand::isMachineInstrOperand) {
230 snprintf(buf, sizeof(buf), "%u", operandIterator->CGIOpNo);
231 operandOrder->addEntry(new LiteralConstantEmitter(buf));
236 for(; numArgs < MAX_OPERANDS; numArgs++) {
237 operandOrder->addEntry(new LiteralConstantEmitter("-1"));
241 /////////////////////////////////////////////////////
242 // Support functions for handling X86 instructions //
243 /////////////////////////////////////////////////////
245 #define ADDFLAG(flag) flags->addEntry(flag)
247 #define REG(str) if (name == str) { ADDFLAG("kOperandFlagRegister"); return 0; }
248 #define MEM(str) if (name == str) { ADDFLAG("kOperandFlagMemory"); return 0; }
249 #define LEA(str) if (name == str) { ADDFLAG("kOperandFlagEffectiveAddress"); \
251 #define IMM(str) if (name == str) { ADDFLAG("kOperandFlagImmediate"); \
253 #define PCR(str) if (name == str) { ADDFLAG("kOperandFlagMemory"); \
254 ADDFLAG("kOperandFlagPCRelative"); \
257 /// X86FlagFromOpName - Processes the name of a single X86 operand (which is
258 /// actually its type) and translates it into an operand flag
260 /// @arg flags - The flags object to add the flag to
261 /// @arg name - The name of the operand
262 static int X86FlagFromOpName(FlagsConstantEmitter *flags,
263 const std::string &name) {
282 REG("CONTROL_REG_32");
283 REG("CONTROL_REG_64");
317 IMM("i64i32imm_pcrel");
337 /// X86PopulateOperands - Handles all the operands in an X86 instruction, adding
338 /// the appropriate flags to their descriptors
340 /// @operandFlags - A reference the array of operand flag objects
341 /// @inst - The instruction to use as a source of information
342 static void X86PopulateOperands(
343 FlagsConstantEmitter *(&operandFlags)[MAX_OPERANDS],
344 const CodeGenInstruction &inst) {
345 if (!inst.TheDef->isSubClassOf("X86Inst"))
349 unsigned int numOperands = inst.OperandList.size();
351 for (index = 0; index < numOperands; ++index) {
352 const CodeGenInstruction::OperandInfo &operandInfo =
353 inst.OperandList[index];
354 Record &rec = *operandInfo.Rec;
356 if (X86FlagFromOpName(operandFlags[index], rec.getName())) {
357 errs() << "Operand type: " << rec.getName().c_str() << "\n";
358 errs() << "Operand name: " << operandInfo.Name.c_str() << "\n";
359 errs() << "Instruction mame: " << inst.TheDef->getName().c_str() << "\n";
360 llvm_unreachable("Unhandled type");
365 /// decorate1 - Decorates a named operand with a new flag
367 /// @operandFlags - The array of operand flag objects, which don't have names
368 /// @inst - The CodeGenInstruction, which provides a way to translate
369 /// between names and operand indices
370 /// @opName - The name of the operand
371 /// @flag - The name of the flag to add
372 static inline void decorate1(FlagsConstantEmitter *(&operandFlags)[MAX_OPERANDS],
373 const CodeGenInstruction &inst,
375 const char *opFlag) {
378 opIndex = inst.getOperandNamed(std::string(opName));
380 operandFlags[opIndex]->addEntry(opFlag);
383 #define DECORATE1(opName, opFlag) decorate1(operandFlags, inst, opName, opFlag)
385 #define MOV(source, target) { \
386 instFlags.addEntry("kInstructionFlagMove"); \
387 DECORATE1(source, "kOperandFlagSource"); \
388 DECORATE1(target, "kOperandFlagTarget"); \
391 #define BRANCH(target) { \
392 instFlags.addEntry("kInstructionFlagBranch"); \
393 DECORATE1(target, "kOperandFlagTarget"); \
396 #define PUSH(source) { \
397 instFlags.addEntry("kInstructionFlagPush"); \
398 DECORATE1(source, "kOperandFlagSource"); \
401 #define POP(target) { \
402 instFlags.addEntry("kInstructionFlagPop"); \
403 DECORATE1(target, "kOperandFlagTarget"); \
406 #define CALL(target) { \
407 instFlags.addEntry("kInstructionFlagCall"); \
408 DECORATE1(target, "kOperandFlagTarget"); \
412 instFlags.addEntry("kInstructionFlagReturn"); \
415 /// X86ExtractSemantics - Performs various checks on the name of an X86
416 /// instruction to determine what sort of an instruction it is and then adds
417 /// the appropriate flags to the instruction and its operands
419 /// @arg instFlags - A reference to the flags for the instruction as a whole
420 /// @arg operandFlags - A reference to the array of operand flag object pointers
421 /// @arg inst - A reference to the original instruction
422 static void X86ExtractSemantics(FlagsConstantEmitter &instFlags,
423 FlagsConstantEmitter *(&operandFlags)[MAX_OPERANDS],
424 const CodeGenInstruction &inst) {
425 const std::string &name = inst.TheDef->getName();
427 if (name.find("MOV") != name.npos) {
428 if (name.find("MOV_V") != name.npos) {
429 // ignore (this is a pseudoinstruction)
431 else if (name.find("MASK") != name.npos) {
432 // ignore (this is a masking move)
434 else if (name.find("r0") != name.npos) {
435 // ignore (this is a pseudoinstruction)
437 else if (name.find("PS") != name.npos ||
438 name.find("PD") != name.npos) {
439 // ignore (this is a shuffling move)
441 else if (name.find("MOVS") != name.npos) {
442 // ignore (this is a string move)
444 else if (name.find("_F") != name.npos) {
445 // TODO handle _F moves to ST(0)
447 else if (name.find("a") != name.npos) {
448 // TODO handle moves to/from %ax
450 else if (name.find("CMOV") != name.npos) {
453 else if (name.find("PC") != name.npos) {
461 if (name.find("JMP") != name.npos ||
462 name.find("J") == 0) {
463 if (name.find("FAR") != name.npos && name.find("i") != name.npos) {
471 if (name.find("PUSH") != name.npos) {
472 if (name.find("FS") != name.npos ||
473 name.find("GS") != name.npos) {
474 instFlags.addEntry("kInstructionFlagPush");
475 // TODO add support for fixed operands
477 else if (name.find("F") != name.npos) {
478 // ignore (this pushes onto the FP stack)
480 else if (name[name.length() - 1] == 'm') {
483 else if (name.find("i") != name.npos) {
491 if (name.find("POP") != name.npos) {
492 if (name.find("POPCNT") != name.npos) {
493 // ignore (not a real pop)
495 else if (name.find("FS") != name.npos ||
496 name.find("GS") != name.npos) {
497 instFlags.addEntry("kInstructionFlagPop");
498 // TODO add support for fixed operands
500 else if (name.find("F") != name.npos) {
501 // ignore (this pops from the FP stack)
503 else if (name[name.length() - 1] == 'm') {
511 if (name.find("CALL") != name.npos) {
512 if (name.find("ADJ") != name.npos) {
513 // ignore (not a call)
515 else if (name.find("SYSCALL") != name.npos) {
516 // ignore (doesn't go anywhere we know about)
518 else if (name.find("VMCALL") != name.npos) {
519 // ignore (rather different semantics than a regular call)
521 else if (name.find("FAR") != name.npos && name.find("i") != name.npos) {
529 if (name.find("RET") != name.npos) {
541 #undef COND_DECORATE_2
542 #undef COND_DECORATE_1
545 /// populateInstInfo - Fills an array of InstInfos with information about each
546 /// instruction in a target
548 /// @arg infoArray - The array of InstInfo objects to populate
549 /// @arg target - The CodeGenTarget to use as a source of instructions
550 static void populateInstInfo(CompoundConstantEmitter &infoArray,
551 CodeGenTarget &target) {
552 const std::vector<const CodeGenInstruction*> &numberedInstructions =
553 target.getInstructionsByEnumValue();
556 unsigned int numInstructions = numberedInstructions.size();
558 for (index = 0; index < numInstructions; ++index) {
559 const CodeGenInstruction& inst = *numberedInstructions[index];
561 CompoundConstantEmitter *infoStruct = new CompoundConstantEmitter;
562 infoArray.addEntry(infoStruct);
564 FlagsConstantEmitter *instFlags = new FlagsConstantEmitter;
565 infoStruct->addEntry(instFlags);
567 LiteralConstantEmitter *numOperandsEmitter =
568 new LiteralConstantEmitter(inst.OperandList.size());
569 infoStruct->addEntry(numOperandsEmitter);
571 CompoundConstantEmitter *operandFlagArray = new CompoundConstantEmitter;
572 infoStruct->addEntry(operandFlagArray);
574 FlagsConstantEmitter *operandFlags[MAX_OPERANDS];
576 for (unsigned operandIndex = 0; operandIndex < MAX_OPERANDS; ++operandIndex) {
577 operandFlags[operandIndex] = new FlagsConstantEmitter;
578 operandFlagArray->addEntry(operandFlags[operandIndex]);
581 unsigned numSyntaxes = 0;
583 if (target.getName() == "X86") {
584 X86PopulateOperands(operandFlags, inst);
585 X86ExtractSemantics(*instFlags, operandFlags, inst);
589 CompoundConstantEmitter *operandOrderArray = new CompoundConstantEmitter;
590 infoStruct->addEntry(operandOrderArray);
592 for (unsigned syntaxIndex = 0; syntaxIndex < MAX_SYNTAXES; ++syntaxIndex) {
593 CompoundConstantEmitter *operandOrder = new CompoundConstantEmitter;
594 operandOrderArray->addEntry(operandOrder);
596 if (syntaxIndex < numSyntaxes) {
597 populateOperandOrder(operandOrder, inst, syntaxIndex);
600 for (unsigned operandIndex = 0;
601 operandIndex < MAX_OPERANDS;
603 operandOrder->addEntry(new LiteralConstantEmitter("-1"));
610 void EDEmitter::run(raw_ostream &o) {
613 CompoundConstantEmitter infoArray;
614 CodeGenTarget target;
616 populateInstInfo(infoArray, target);
618 o << "InstInfo instInfo" << target.getName().c_str() << "[] = ";
619 infoArray.emit(o, i);
623 void EDEmitter::runHeader(raw_ostream &o) {
624 EmitSourceFileHeader("Enhanced Disassembly Info Header", o);
626 o << "#ifndef EDInfo_" << "\n";
627 o << "#define EDInfo_" << "\n";
629 o << "#include <inttypes.h>" << "\n";
631 o << "#define MAX_OPERANDS " << format("%d", MAX_OPERANDS) << "\n";
632 o << "#define MAX_SYNTAXES " << format("%d", MAX_SYNTAXES) << "\n";
637 EnumEmitter operandFlags("OperandFlags");
638 operandFlags.addEntry("kOperandFlagImmediate");
639 operandFlags.addEntry("kOperandFlagRegister");
640 operandFlags.addEntry("kOperandFlagMemory");
641 operandFlags.addEntry("kOperandFlagEffectiveAddress");
642 operandFlags.addEntry("kOperandFlagPCRelative");
643 operandFlags.addEntry("kOperandFlagSource");
644 operandFlags.addEntry("kOperandFlagTarget");
645 operandFlags.emitAsFlags(o, i);
649 EnumEmitter instructionFlags("InstructionFlags");
650 instructionFlags.addEntry("kInstructionFlagMove");
651 instructionFlags.addEntry("kInstructionFlagBranch");
652 instructionFlags.addEntry("kInstructionFlagPush");
653 instructionFlags.addEntry("kInstructionFlagPop");
654 instructionFlags.addEntry("kInstructionFlagCall");
655 instructionFlags.addEntry("kInstructionFlagReturn");
656 instructionFlags.emitAsFlags(o, i);
660 StructEmitter instInfo("InstInfo");
661 instInfo.addMember("uint32_t", "instructionFlags");
662 instInfo.addMember("uint8_t", "numOperands");
663 instInfo.addMember("uint8_t", "operandFlags[MAX_OPERANDS]");
664 instInfo.addMember("const char", "operandOrders[MAX_SYNTAXES][MAX_OPERANDS]");
668 o << "#endif" << "\n";