//
//===----------------------------------------------------------------------===//
-#include "AsmWriterEmitter.h"
#include "AsmWriterInst.h"
#include "CodeGenTarget.h"
-#include "StringToOffsetTable.h"
+#include "SequenceToOffsetTable.h"
+#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/Twine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/TableGen/Error.h"
#include "llvm/TableGen/Record.h"
+#include "llvm/TableGen/TableGenBackend.h"
#include <algorithm>
+#include <cassert>
+#include <map>
+#include <vector>
using namespace llvm;
+namespace {
+class AsmWriterEmitter {
+ RecordKeeper &Records;
+ std::map<const CodeGenInstruction*, AsmWriterInst*> CGIAWIMap;
+ std::vector<const CodeGenInstruction*> NumberedInstructions;
+public:
+ AsmWriterEmitter(RecordKeeper &R) : Records(R) {}
+
+ void run(raw_ostream &o);
+
+private:
+ void EmitPrintInstruction(raw_ostream &o);
+ void EmitGetRegisterName(raw_ostream &o);
+ void EmitPrintAliasInstruction(raw_ostream &O);
+
+ AsmWriterInst *getAsmWriterInstByID(unsigned ID) const {
+ assert(ID < NumberedInstructions.size());
+ std::map<const CodeGenInstruction*, AsmWriterInst*>::const_iterator I =
+ CGIAWIMap.find(NumberedInstructions[ID]);
+ assert(I != CGIAWIMap.end() && "Didn't find inst!");
+ return I->second;
+ }
+ 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, raw_ostream &O) {
O << " case " << OpsToPrint.back().first << ": ";
CGIAWIMap.insert(std::make_pair(Instructions[i].CGI, &Instructions[i]));
// Build an aggregate string, and build a table of offsets into it.
- StringToOffsetTable StringTable;
+ SequenceToOffsetTable<std::string> StringTable;
/// 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;
+ /// To reduce the number of unhandled cases, we expand the size from 32-bit
+ /// to 32+16 = 48-bit.
+ std::vector<uint64_t> OpcodeInfo;
+
+ // Add all strings to the string table upfront so it can generate an optimized
+ // representation.
+ for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
+ AsmWriterInst *AWI = CGIAWIMap[NumberedInstructions[i]];
+ if (AWI != 0 &&
+ AWI->Operands[0].OperandType ==
+ AsmWriterOperand::isLiteralTextOperand &&
+ !AWI->Operands[0].Str.empty()) {
+ std::string Str = AWI->Operands[0].Str;
+ UnescapeString(Str);
+ StringTable.add(Str);
+ }
+ }
+
+ StringTable.layout();
unsigned MaxStringIdx = 0;
for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
AsmWriterOperand::isLiteralTextOperand ||
AWI->Operands[0].Str.empty()) {
// Something handled by the asmwriter printer, but with no leading string.
- Idx = StringTable.GetOrAddStringOffset("");
+ Idx = StringTable.get("");
} else {
std::string Str = AWI->Operands[0].Str;
UnescapeString(Str);
- Idx = StringTable.GetOrAddStringOffset(Str);
+ Idx = StringTable.get(Str);
MaxStringIdx = std::max(MaxStringIdx, Idx);
// Nuke the string from the operand list. It is now handled!
}
// Bias offset by one since we want 0 as a sentinel.
- assert((Idx+1) <= 0xffff && "String offset too large to fit in table");
OpcodeInfo.push_back(Idx+1);
}
// To reduce code size, we compactify common instructions into a few bits
// in the opcode-indexed table.
- unsigned BitsLeft = 32-AsmStrBits;
+ unsigned BitsLeft = 64-AsmStrBits;
std::vector<std::vector<std::string> > TableDrivenOperandPrinters;
}
// Otherwise, we can include this in the initial lookup table. Add it in.
- BitsLeft -= NumBits;
for (unsigned i = 0, e = InstIdxs.size(); i != e; ++i)
- if (InstIdxs[i] != ~0U)
- OpcodeInfo[i] |= InstIdxs[i] << (BitsLeft+AsmStrBits);
+ if (InstIdxs[i] != ~0U) {
+ OpcodeInfo[i] |= (uint64_t)InstIdxs[i] << (64-BitsLeft);
+ }
+ BitsLeft -= NumBits;
// Remove the info about this operand.
for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
}
-
- O<<" static const unsigned OpInfo[] = {\n";
+ // We always emit at least one 32-bit table. A second table is emitted if
+ // more bits are needed.
+ O<<" static const uint32_t OpInfo[] = {\n";
for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
- O << " " << OpcodeInfo[i] << "U,\t// "
+ O << " " << (OpcodeInfo[i] & 0xffffffff) << "U,\t// "
<< NumberedInstructions[i]->TheDef->getName() << "\n";
}
// Add a dummy entry so the array init doesn't end with a comma.
O << " 0U\n";
O << " };\n\n";
+ if (BitsLeft < 32) {
+ // Add a second OpInfo table only when it is necessary.
+ // Adjust the type of the second table based on the number of bits needed.
+ O << " static const uint"
+ << ((BitsLeft < 16) ? "32" : (BitsLeft < 24) ? "16" : "8")
+ << "_t OpInfo2[] = {\n";
+ for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
+ O << " " << (OpcodeInfo[i] >> 32) << "U,\t// "
+ << NumberedInstructions[i]->TheDef->getName() << "\n";
+ }
+ // Add a dummy entry so the array init doesn't end with a comma.
+ O << " 0U\n";
+ O << " };\n\n";
+ }
+
// Emit the string itself.
- O << " const char *const AsmStrs = \n";
- StringTable.EmitString(O);
- O << ";\n\n";
+ O << " const char AsmStrs[] = {\n";
+ StringTable.emit(O, printChar);
+ O << " };\n\n";
O << " O << \"\\t\";\n\n";
- O << " // Emit the opcode for the instruction.\n"
- << " unsigned Bits = OpInfo[MI->getOpcode()];\n"
- << " assert(Bits != 0 && \"Cannot print this instruction.\");\n"
+ O << " // Emit the opcode for the instruction.\n";
+ if (BitsLeft < 32) {
+ // If we have two tables then we need to perform two lookups and combine
+ // the results into a single 64-bit value.
+ O << " uint64_t Bits1 = OpInfo[MI->getOpcode()];\n"
+ << " uint64_t Bits2 = OpInfo2[MI->getOpcode()];\n"
+ << " uint64_t Bits = (Bits2 << 32) | Bits1;\n";
+ } else {
+ // If only one table is used we just need to perform a single lookup.
+ O << " uint32_t Bits = OpInfo[MI->getOpcode()];\n";
+ }
+ 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 = 32-AsmStrBits;
+ BitsLeft = 64-AsmStrBits;
for (unsigned i = 0, e = TableDrivenOperandPrinters.size(); i != e; ++i) {
std::vector<std::string> &Commands = TableDrivenOperandPrinters[i];
assert(NumBits <= BitsLeft && "consistency error");
// Emit code to extract this field from Bits.
- BitsLeft -= NumBits;
-
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 >> " << (BitsLeft+AsmStrBits) << ") & "
+ O << " if ((Bits >> "
+ << (64-BitsLeft) << ") & "
<< ((1 << NumBits)-1) << ") {\n"
<< Commands[1]
<< " } else {\n"
// Emit a single possibility.
O << Commands[0] << "\n\n";
} else {
- O << " switch ((Bits >> " << (BitsLeft+AsmStrBits) << ") & "
+ O << " switch ((Bits >> "
+ << (64-BitsLeft) << ") & "
<< ((1 << NumBits)-1) << ") {\n"
<< " default: // unreachable.\n";
}
O << " }\n\n";
}
+ BitsLeft -= NumBits;
}
// Okay, delete instructions with no operand info left.
static void
emitRegisterNameString(raw_ostream &O, StringRef AltName,
- const std::vector<CodeGenRegister*> &Registers) {
- StringToOffsetTable StringTable;
- O << " static const unsigned RegAsmOffset" << AltName << "[] = {\n ";
+ const std::vector<CodeGenRegister*> &Registers) {
+ SequenceToOffsetTable<std::string> StringTable;
+ SmallVector<std::string, 4> AsmNames(Registers.size());
for (unsigned i = 0, e = Registers.size(); i != e; ++i) {
const CodeGenRegister &Reg = *Registers[i];
+ std::string &AsmName = AsmNames[i];
- std::string AsmName;
// "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") {
std::vector<std::string> AltNames =
Reg.TheDef->getValueAsListOfStrings("AltNames");
if (AltNames.size() <= Idx)
- throw TGError(Reg.TheDef->getLoc(),
- (Twine("Register definition missing alt name for '") +
- AltName + "'.").str());
+ PrintFatalError(Reg.TheDef->getLoc(),
+ (Twine("Register definition missing alt name for '") +
+ AltName + "'.").str());
AsmName = AltNames[Idx];
}
}
+ StringTable.add(AsmName);
+ }
- unsigned Idx = StringTable.GetOrAddStringOffset(AsmName);
- assert(Idx <= 0xffff && "String offset too large to fit in table");
- O << Idx;
- if (((i + 1) % 14) == 0)
- O << ",\n ";
- else
- O << ", ";
+ StringTable.layout();
+ O << " static const char AsmStrs" << AltName << "[] = {\n";
+ StringTable.emit(O, printChar);
+ O << " };\n\n";
+ O << " static const uint32_t RegAsmOffset" << AltName << "[] = {";
+ for (unsigned i = 0, e = Registers.size(); i != e; ++i) {
+ if ((i % 14) == 0)
+ O << "\n ";
+ O << StringTable.get(AsmNames[i]) << ", ";
}
- O << "0\n"
- << " };\n"
+ O << "\n };\n"
<< "\n";
-
- O << " const char *AsmStrs" << AltName << " =\n";
- StringTable.EmitString(O);
- O << ";\n";
}
void AsmWriterEmitter::EmitGetRegisterName(raw_ostream &O) {
emitRegisterNameString(O, "", Registers);
if (hasAltNames) {
- O << " const unsigned *RegAsmOffset;\n"
+ O << " const uint32_t *RegAsmOffset;\n"
<< " const char *AsmStrs;\n"
<< " switch(AltIdx) {\n"
<< " default: llvm_unreachable(\"Invalid register alt name index!\");\n";
<< "}\n";
}
-void AsmWriterEmitter::EmitGetInstructionName(raw_ostream &O) {
- CodeGenTarget Target(Records);
- 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 ";
-
- unsigned Idx = StringTable.GetOrAddStringOffset(AsmName);
- assert(Idx <= 0xffff && "String offset too large to fit in table");
- O << Idx << ", ";
- }
- 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";
-}
-
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
std::map<StringRef, unsigned> OpMap;
std::string Result;
std::string AsmString;
- std::vector<Record*> ReqFeatures;
+ SmallVector<Record*, 4> ReqFeatures;
public:
IAPrinter(std::string R, std::string AS)
: Result(R), AsmString(AS) {}
O << " I = OpMap.begin(), E = OpMap.end(); I != E; ++I)\n";
O << " if (I->first == Name)\n";
O << " return I->second;\n";
- O << " assert(false && \"Operand not in map!\");\n";
- O << " return 0;\n";
- O << "}\n\n";
-}
-
-void AsmWriterEmitter::EmitRegIsInRegClass(raw_ostream &O) {
- CodeGenTarget Target(Records);
-
- // Enumerate the register classes.
- ArrayRef<CodeGenRegisterClass*> RegisterClasses =
- Target.getRegBank().getRegClasses();
-
- O << "namespace { // Register classes\n";
- O << " enum RegClass {\n";
-
- // Emit the register enum value for each RegisterClass.
- for (unsigned I = 0, E = RegisterClasses.size(); I != E; ++I) {
- if (I != 0) O << ",\n";
- O << " RC_" << RegisterClasses[I]->getName();
- }
-
- O << "\n };\n";
- O << "} // end anonymous namespace\n\n";
-
- // Emit a function that returns 'true' if a regsiter is part of a particular
- // register class. I.e., RAX is part of GR64 on X86.
- O << "static bool regIsInRegisterClass"
- << "(unsigned RegClass, unsigned Reg) {\n";
-
- // Emit the switch that checks if a register belongs to a particular register
- // class.
- O << " switch (RegClass) {\n";
- O << " default: break;\n";
-
- for (unsigned I = 0, E = RegisterClasses.size(); I != E; ++I) {
- const CodeGenRegisterClass &RC = *RegisterClasses[I];
-
- // Give the register class a legal C name if it's anonymous.
- std::string Name = RC.getName();
- O << " case RC_" << Name << ":\n";
-
- // Emit the register list now.
- unsigned IE = RC.getOrder().size();
- if (IE == 1) {
- O << " if (Reg == " << getQualifiedName(RC.getOrder()[0]) << ")\n";
- O << " return true;\n";
- } else {
- O << " switch (Reg) {\n";
- O << " default: break;\n";
-
- for (unsigned II = 0; II != IE; ++II) {
- Record *Reg = RC.getOrder()[II];
- O << " case " << getQualifiedName(Reg) << ":\n";
- }
-
- O << " return true;\n";
- O << " }\n";
- }
-
- O << " break;\n";
- }
-
- O << " }\n\n";
- O << " return false;\n";
+ O << " llvm_unreachable(\"Operand not in map!\");\n";
O << "}\n\n";
}
O << "\n#ifdef PRINT_ALIAS_INSTR\n";
O << "#undef PRINT_ALIAS_INSTR\n\n";
- EmitRegIsInRegClass(O);
-
// Emit the method that prints the alias instruction.
std::string ClassName = AsmWriter->getValueAsString("AsmWriterClassName");
if (!R->getValueAsBit("EmitAlias"))
continue; // We were told not to emit the alias, but to emit the aliasee.
const DagInit *DI = R->getValueAsDag("ResultInst");
- const DefInit *Op = dynamic_cast<const DefInit*>(DI->getOperator());
+ const DefInit *Op = cast<DefInit>(DI->getOperator());
AliasMap[getQualifiedName(Op->getDef())].push_back(Alias);
}
if (!IAP->isOpMapped(ROName)) {
IAP->addOperand(ROName, i);
- Cond = std::string("regIsInRegisterClass(RC_") +
- CGA->ResultOperands[i].getRecord()->getName() +
- ", MI->getOperand(" + llvm::utostr(i) + ").getReg())";
+ Cond = std::string("MRI.getRegClass(") + Target.getName() + "::" +
+ CGA->ResultOperands[i].getRecord()->getName() + "RegClassID)"
+ ".contains(MI->getOperand(" + llvm::utostr(i) + ").getReg())";
IAP->addCond(Cond);
} else {
Cond = std::string("MI->getOperand(") +
break;
}
- case CodeGenInstAlias::ResultOperand::K_Imm:
- Cond = std::string("MI->getOperand(") +
- llvm::utostr(i) + ").getImm() == " +
- llvm::utostr(CGA->ResultOperands[i].getImm());
+ case CodeGenInstAlias::ResultOperand::K_Imm: {
+ std::string Op = "MI->getOperand(" + llvm::utostr(i) + ")";
+
+ // 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.
O << " }\n";
O << " }\n";
O << " }\n\n";
-
+
O << " return true;\n";
O << "}\n\n";
}
void AsmWriterEmitter::run(raw_ostream &O) {
- EmitSourceFileHeader("Assembly Writer Source Fragment", O);
-
EmitPrintInstruction(O);
EmitGetRegisterName(O);
- EmitGetInstructionName(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