//
// 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 "llvm/ADT/StringExtras.h"
+#include "StringToOffsetTable.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MathExtras.h"
#include <algorithm>
-#include <ostream>
using namespace llvm;
-static bool isIdentChar(char C) {
- return (C >= 'a' && C <= 'z') ||
- (C >= 'A' && C <= 'Z') ||
- (C >= '0' && C <= '9') ||
- C == '_';
-}
-
-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);
- }
-
- /// getCode - Return the code that prints this operand.
- std::string getCode() const;
- };
-}
-
-namespace llvm {
- 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));
- }
- };
-}
-
-
-std::string AsmWriterOperand::getCode() const {
- if (OperandType == isLiteralTextOperand)
- return "O << \"" + Str + "\"; ";
-
- std::string Result = Str + "(MI";
- if (MIOpNo != ~0U)
- Result += ", " + utostr(MIOpNo);
- if (!MiModifier.empty())
- Result += ", \"" + MiModifier + '"';
- return Result + "); ";
-}
-
-
-/// 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] == '$') {
- if (CurVariant == Variant || CurVariant == ~0U)
- AddLiteralString("$"); // "$$" -> $
- LastEmitted = DollarPos+2;
- } else {
- // Get the name of the variable.
- std::string::size_type VarEnd = DollarPos+1;
-
- // handle ${foo}bar as $foo by detecting whether the character following
- // the dollar sign is a curly brace. If so, advance VarEnd and DollarPos
- // so the variable name does not contain the leading curly brace.
- bool hasCurlyBraces = false;
- if (VarEnd < AsmString.size() && '{' == AsmString[VarEnd]) {
- hasCurlyBraces = true;
- ++DollarPos;
- ++VarEnd;
- }
-
- while (VarEnd < AsmString.size() && isIdentChar(AsmString[VarEnd]))
- ++VarEnd;
- std::string VarName(AsmString.begin()+DollarPos+1,
- AsmString.begin()+VarEnd);
-
- // Modifier - Support ${foo:modifier} syntax, where "modifier" is passed
- // into printOperand. Also support ${:feature}, which is passed into
- // PrintSpecial.
- std::string Modifier;
-
- // In order to avoid starting the next string at the terminating curly
- // brace, advance the end position past it if we found an opening curly
- // brace.
- if (hasCurlyBraces) {
- if (VarEnd >= AsmString.size())
- throw "Reached end of string before terminating curly brace in '"
- + CGI.TheDef->getName() + "'";
-
- // Look for a modifier string.
- if (AsmString[VarEnd] == ':') {
- ++VarEnd;
- if (VarEnd >= AsmString.size())
- throw "Reached end of string before terminating curly brace in '"
- + CGI.TheDef->getName() + "'";
-
- unsigned ModifierStart = VarEnd;
- while (VarEnd < AsmString.size() && isIdentChar(AsmString[VarEnd]))
- ++VarEnd;
- Modifier = std::string(AsmString.begin()+ModifierStart,
- AsmString.begin()+VarEnd);
- if (Modifier.empty())
- throw "Bad operand modifier name in '"+ CGI.TheDef->getName() + "'";
- }
-
- if (AsmString[VarEnd] != '}')
- throw "Variable name beginning with '{' did not end with '}' in '"
- + CGI.TheDef->getName() + "'";
- ++VarEnd;
- }
- if (VarName.empty() && Modifier.empty())
- throw "Stray '$' in '" + CGI.TheDef->getName() +
- "' asm string, maybe you want $$?";
-
- if (VarName.empty()) {
- // Just a modifier, pass this into PrintSpecial.
- Operands.push_back(AsmWriterOperand("PrintSpecial", ~0U, Modifier));
- } else {
- // Otherwise, normal operand.
- unsigned OpNo = CGI.getOperandNamed(VarName);
- CodeGenInstruction::OperandInfo OpInfo = CGI.OperandList[OpNo];
-
- // If this is a two-address instruction, verify the second operand isn't
- // used.
- unsigned MIOp = OpInfo.MIOperandNo;
- if (CGI.isTwoAddress && MIOp == 1)
- throw "Should refer to operand #0 instead of #1 for two-address"
- " instruction '" + CGI.TheDef->getName() + "'!";
-
- if (CurVariant == Variant || CurVariant == ~0U)
- Operands.push_back(AsmWriterOperand(OpInfo.PrinterMethodName, MIOp,
- Modifier));
- }
- LastEmitted = VarEnd;
- }
- }
-
- AddLiteralString("\\n");
-}
-
-/// 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;
- }
- return MismatchOperand;
-}
-
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();
/// 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 << "\n";
}
-
O << " break;\n";
}
for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
const AsmWriterInst *Inst = getAsmWriterInstByID(i);
- if (Inst == 0) continue; // PHI, INLINEASM, etc.
+ if (Inst == 0) continue; // PHI, INLINEASM, DBG_LABEL, etc.
std::string Command;
if (Inst->Operands.empty())
Command = " " + Inst->Operands[0].getCode() + "\n";
- // If this is the last operand, emit a return.
- if (Inst->Operands.size() == 1)
- Command += " return true;\n";
-
// Check to see if we already have 'Command' in UniqueOperandCommands.
// If not, add it.
bool FoundIt = false;
// Otherwise, scan to see if all of the other instructions in this command
// set share the operand.
bool AllSame = true;
-
+ // Keep track of the maximum, number of operands or any
+ // instruction we see in the group.
+ size_t MaxSize = FirstInst->Operands.size();
+
for (NIT = std::find(NIT+1, InstIdxs.end(), CommandIdx);
NIT != InstIdxs.end();
NIT = std::find(NIT+1, InstIdxs.end(), CommandIdx)) {
// matches, we're ok, otherwise bail out.
const AsmWriterInst *OtherInst =
getAsmWriterInstByID(NIT-InstIdxs.begin());
+
+ if (OtherInst &&
+ OtherInst->Operands.size() > FirstInst->Operands.size())
+ MaxSize = std::max(MaxSize, OtherInst->Operands.size());
+
if (!OtherInst || OtherInst->Operands.size() == Op ||
OtherInst->Operands[Op] != FirstInst->Operands[Op]) {
AllSame = false;
// to UniqueOperandCommands and remember that it was consumed.
std::string Command = " " + FirstInst->Operands[Op].getCode() + "\n";
- // If this is the last operand, emit a return after the code.
- if (FirstInst->Operands.size() == Op+1)
- Command += " return true;\n";
-
UniqueOperandCommands[CommandIdx] += Command;
InstOpsUsed[CommandIdx]++;
}
}
+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);
+ }
+ }
+}
-void AsmWriterEmitter::run(std::ostream &O) {
- EmitSourceFileHeader("Assembly Writer Source Fragment", O);
-
+/// EmitPrintInstruction - Generate the code for the "printInstruction" method
+/// implementation.
+void AsmWriterEmitter::EmitPrintInstruction(raw_ostream &O) {
CodeGenTarget Target;
Record *AsmWriter = Target.getAsmWriter();
std::string ClassName = AsmWriter->getValueAsString("AsmWriterClassName");
- unsigned Variant = AsmWriter->getValueAsInt("Variant");
-
+
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 MachineInstr *MI, raw_ostream &O) {\n";
std::vector<AsmWriterInst> Instructions;
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));
+ if (!(*I)->AsmString.empty() &&
+ (*I)->TheDef->getName() != "PHI")
+ Instructions.push_back(
+ AsmWriterInst(**I,
+ AsmWriter->getValueAsInt("Variant"),
+ AsmWriter->getValueAsInt("FirstOperandColumn"),
+ AsmWriter->getValueAsInt("OperandSpacing")));
// Get the instruction numbering.
- Target.getInstructionsByEnumValue(NumberedInstructions);
+ NumberedInstructions = Target.getInstructionsByEnumValue();
// Compute the CodeGenInstruction -> AsmWriterInst mapping. Note that not
// all machine instructions are necessarily being printed, so there may be
CGIAWIMap.insert(std::make_pair(Instructions[i].CGI, &Instructions[i]));
// Build an aggregate string, and build a table of offsets into it.
- std::map<std::string, unsigned> StringOffset;
- std::string AggregateString;
- AggregateString += "\0\0";
+ StringToOffsetTable StringTable;
- /// OpcodeInfo - Theis encodes the index of the string to use for the first
+ /// 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<unsigned> OpcodeInfo;
unsigned Idx;
if (AWI == 0) {
// 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 = 1;
+ Idx = StringTable.GetOrAddStringOffset("");
} else {
- unsigned &Entry = StringOffset[AWI->Operands[0].Str];
- if (Entry == 0) {
- // Add the string to the aggregate if this is the first time found.
- MaxStringIdx = Entry = AggregateString.size();
- std::string Str = AWI->Operands[0].Str;
- UnescapeString(Str);
- AggregateString += Str;
- AggregateString += '\0';
- }
- Idx = Entry;
-
+ std::string Str = AWI->Operands[0].Str;
+ UnescapeString(Str);
+ Idx = StringTable.GetOrAddStringOffset(Str);
+ MaxStringIdx = std::max(MaxStringIdx, Idx);
+
// Nuke the string from the operand list. It is now handled!
AWI->Operands.erase(AWI->Operands.begin());
}
- OpcodeInfo.push_back(Idx);
+
+ // 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);
+ unsigned AsmStrBits = Log2_32_Ceil(MaxStringIdx+2);
// To reduce code size, we compactify common instructions into a few bits
// in the opcode-indexed table.
std::vector<std::vector<std::string> > TableDrivenOperandPrinters;
- bool isFirst = true;
while (1) {
std::vector<std::string> UniqueOperandCommands;
-
- // For the first operand check, add a default value for instructions with
- // just opcode strings to use.
- if (isFirst) {
- UniqueOperandCommands.push_back(" return true;\n");
- isFirst = false;
- }
-
std::vector<unsigned> InstIdxs;
std::vector<unsigned> NumInstOpsHandled;
FindUniqueOperandCommands(UniqueOperandCommands, InstIdxs,
// If we don't have enough bits for this operand, don't include it.
if (NumBits > BitsLeft) {
- DEBUG(std::cerr << "Not enough bits to densely encode " << NumBits
- << " more bits\n");
+ DEBUG(errs() << "Not enough bits to densely encode " << NumBits
+ << " more bits\n");
break;
}
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;
- }
- 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";
+ O << " const char *AsmStrs = \n";
+ StringTable.EmitString(O);
+ O << ";\n\n";
+
+ O << " O << \"\\t\";\n\n";
- O << " if (MI->getOpcode() == TargetInstrInfo::INLINEASM) {\n"
- << " printInlineAsm(MI);\n"
- << " return true;\n"
- << " }\n\n";
-
O << " // Emit the opcode for the instruction.\n"
<< " unsigned Bits = OpInfo[MI->getOpcode()];\n"
- << " if (Bits == 0) return false;\n"
- << " O << AsmStrs+(Bits & " << (1 << AsmStrBits)-1 << ");\n\n";
+ << " assert(Bits != 0 && \"Cannot print this instruction.\");\n"
+ << " O << AsmStrs+(Bits & " << (1 << AsmStrBits)-1 << ")-1;\n\n";
// Output the table driven operand information.
BitsLeft = 32-AsmStrBits;
// elements in the vector.
std::reverse(Instructions.begin(), Instructions.end());
+
+ // Now that we've emitted all of the operand info that fit into 32 bits, emit
+ // information for those instructions that are left. This is a less dense
+ // encoding, but we expect the main 32-bit table to handle the majority of
+ // instructions.
if (!Instructions.empty()) {
// Find the opcode # of inline asm.
O << " switch (MI->getOpcode()) {\n";
EmitInstructions(Instructions, O);
O << " }\n";
- O << " return true;\n";
+ O << " return;\n";
}
-
+
O << "}\n";
}
+
+
+void AsmWriterEmitter::EmitGetRegisterName(raw_ostream &O) {
+ CodeGenTarget Target;
+ Record *AsmWriter = Target.getAsmWriter();
+ std::string ClassName = AsmWriter->getValueAsString("AsmWriterClassName");
+ const std::vector<CodeGenRegister> &Registers = Target.getRegisters();
+
+ StringToOffsetTable StringTable;
+ 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
+ << "::getRegisterName(unsigned RegNo) {\n"
+ << " assert(RegNo && RegNo < " << (Registers.size()+1)
+ << " && \"Invalid register number!\");\n"
+ << "\n"
+ << " static const unsigned RegAsmOffset[] = {";
+ for (unsigned i = 0, e = Registers.size(); i != e; ++i) {
+ const CodeGenRegister &Reg = Registers[i];
+
+ std::string AsmName = Reg.TheDef->getValueAsString("AsmName");
+ if (AsmName.empty())
+ AsmName = Reg.getName();
+
+
+ if ((i % 14) == 0)
+ O << "\n ";
+
+ O << StringTable.GetOrAddStringOffset(AsmName) << ", ";
+ }
+ O << "0\n"
+ << " };\n"
+ << "\n";
+
+ O << " const char *AsmStrs =\n";
+ StringTable.EmitString(O);
+ O << ";\n";
+
+ O << " return AsmStrs+RegAsmOffset[RegNo-1];\n"
+ << "}\n";
+}
+
+void AsmWriterEmitter::EmitGetInstructionName(raw_ostream &O) {
+ CodeGenTarget Target;
+ Record *AsmWriter = Target.getAsmWriter();
+ std::string ClassName = AsmWriter->getValueAsString("AsmWriterClassName");
+
+ const std::vector<const CodeGenInstruction*> &NumberedInstructions =
+ Target.getInstructionsByEnumValue();
+
+ StringToOffsetTable StringTable;
+ O <<
+"\n\n#ifdef GET_INSTRUCTION_NAME\n"
+"#undef GET_INSTRUCTION_NAME\n\n"
+"/// getInstructionName: This method is automatically generated by tblgen\n"
+"/// from the instruction set description. This returns the enum name of the\n"
+"/// specified instruction.\n"
+ "const char *" << Target.getName() << ClassName
+ << "::getInstructionName(unsigned Opcode) {\n"
+ << " assert(Opcode < " << NumberedInstructions.size()
+ << " && \"Invalid instruction number!\");\n"
+ << "\n"
+ << " static const unsigned InstAsmOffset[] = {";
+ for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
+ const CodeGenInstruction &Inst = *NumberedInstructions[i];
+
+ std::string AsmName = Inst.TheDef->getName();
+ if ((i % 14) == 0)
+ O << "\n ";
+
+ O << StringTable.GetOrAddStringOffset(AsmName) << ", ";
+ }
+ O << "0\n"
+ << " };\n"
+ << "\n";
+
+ O << " const char *Strs =\n";
+ StringTable.EmitString(O);
+ O << ";\n";
+
+ O << " return Strs+InstAsmOffset[Opcode];\n"
+ << "}\n\n#endif\n";
+}
+
+
+
+void AsmWriterEmitter::run(raw_ostream &O) {
+ EmitSourceFileHeader("Assembly Writer Source Fragment", O);
+
+ EmitPrintInstruction(O);
+ EmitGetRegisterName(O);
+ EmitGetInstructionName(O);
+}
+