//
// 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 "llvm/Support/Debug.h"
#include "llvm/Support/MathExtras.h"
#include <algorithm>
-#include <ostream>
using namespace llvm;
static bool isIdentChar(char C) {
C == '_';
}
-namespace {
+// This should be an anon namespace, this works around a GCC warning.
+namespace llvm {
struct AsmWriterOperand {
enum { isLiteralTextOperand, isMachineInstrOperand } OperandType;
}
namespace llvm {
- struct AsmWriterInst {
+ class AsmWriterInst {
+ public:
std::vector<AsmWriterOperand> Operands;
const CodeGenInstruction *CGI;
if (OperandType == isLiteralTextOperand)
return "O << \"" + Str + "\"; ";
- std::string Result = Str + "(MI, " + utostr(MIOpNo);
+ std::string Result = Str + "(MI";
+ if (MIOpNo != ~0U)
+ Result += ", " + utostr(MIOpNo);
if (!MiModifier.empty())
Result += ", \"" + MiModifier + '"';
return Result + "); ";
std::string::size_type LastEmitted = 0;
while (LastEmitted != AsmString.size()) {
std::string::size_type DollarPos =
- AsmString.find_first_of("${|}", LastEmitted);
+ AsmString.find_first_of("${|}\\", LastEmitted);
if (DollarPos == std::string::npos) DollarPos = AsmString.size();
// Emit a constant string fragment.
AddLiteralString(std::string(AsmString.begin()+LastEmitted,
AsmString.begin()+DollarPos));
LastEmitted = DollarPos;
+ } else if (AsmString[DollarPos] == '\\') {
+ if (DollarPos+1 != AsmString.size() &&
+ (CurVariant == Variant || CurVariant == ~0U)) {
+ if (AsmString[DollarPos+1] == 'n') {
+ AddLiteralString("\\n");
+ } else if (AsmString[DollarPos+1] == 't') {
+ AddLiteralString("\\t");
+ } else if (std::string("${|}\\").find(AsmString[DollarPos+1])
+ != std::string::npos) {
+ AddLiteralString(std::string(1, AsmString[DollarPos+1]));
+ } else {
+ throw "Non-supported escaped character found in instruction '" +
+ CGI.TheDef->getName() + "'!";
+ }
+ LastEmitted = DollarPos+2;
+ continue;
+ }
} else if (AsmString[DollarPos] == '{') {
if (CurVariant != ~0U)
throw "Nested variants found for instruction '" +
AsmString.begin()+VarEnd);
// Modifier - Support ${foo:modifier} syntax, where "modifier" is passed
- // into printOperand.
+ // into printOperand. Also support ${:feature}, which is passed into
+ // PrintSpecial.
std::string Modifier;
// In order to avoid starting the next string at the terminating curly
+ CGI.TheDef->getName() + "'";
++VarEnd;
}
- if (VarName.empty())
+ if (VarName.empty() && Modifier.empty())
throw "Stray '$' 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;
+ 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 (CurVariant == Variant || CurVariant == ~0U) {
+ unsigned MIOp = OpInfo.MIOperandNo;
+ Operands.push_back(AsmWriterOperand(OpInfo.PrinterMethodName, MIOp,
+ Modifier));
+ }
}
-
- if (CurVariant == Variant || CurVariant == ~0U)
- Operands.push_back(AsmWriterOperand(OpInfo.PrinterMethodName, MIOp,
- Modifier));
LastEmitted = VarEnd;
}
}
unsigned MismatchOperand = ~0U;
for (unsigned i = 0, e = Operands.size(); i != e; ++i) {
- if (Operands[i] != Other.Operands[i])
+ if (Operands[i] != Other.Operands[i]) {
if (MismatchOperand != ~0U) // Already have one mismatch?
return ~1U;
else
MismatchOperand = i;
+ }
}
return MismatchOperand;
}
void AsmWriterEmitter::
FindUniqueOperandCommands(std::vector<std::string> &UniqueOperandCommands,
- std::vector<unsigned> &InstIdxs, unsigned Op) const {
- InstIdxs.clear();
- InstIdxs.resize(NumberedInstructions.size());
+ std::vector<unsigned> &InstIdxs,
+ std::vector<unsigned> &InstOpsUsed) const {
+ InstIdxs.assign(NumberedInstructions.size(), ~0U);
// 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 == 0) continue; // PHI, INLINEASM, etc.
+ if (Inst == 0) continue; // PHI, INLINEASM, LABEL, etc.
std::string Command;
- if (Op > Inst->Operands.size())
+ if (Inst->Operands.empty())
continue; // Instruction already done.
- else if (Op == Inst->Operands.size())
- Command = " return true;\n";
- else
- Command = " " + Inst->Operands[Op].getCode() + "\n";
+
+ 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.
InstIdxs[i] = UniqueOperandCommands.size();
UniqueOperandCommands.push_back(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";
+
+ // 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]++;
}
}
// Build an aggregate string, and build a table of offsets into it.
std::map<std::string, unsigned> StringOffset;
std::string AggregateString;
- AggregateString += '\0';
+ AggregateString.push_back(0); // "\0"
+ AggregateString.push_back(0); // "\0"
- /// OpcodeInfo - The first value in the pair is the index into the string, the
- /// second is an index used for operand printing information.
- std::vector<std::pair<unsigned short, unsigned short> > OpcodeInfo;
+ /// 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 MaxStringIdx = 0;
for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
AsmWriterInst *AWI = CGIAWIMap[NumberedInstructions[i]];
unsigned Idx;
- if (AWI == 0 || AWI->Operands[0].Str.empty()) {
+ if (AWI == 0) {
// Something not handled by the asmwriter printer.
Idx = 0;
+ } 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;
} 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();
+ MaxStringIdx = 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!");
// Nuke the string from the operand list. It is now handled!
AWI->Operands.erase(AWI->Operands.begin());
}
- OpcodeInfo.push_back(std::pair<unsigned short, unsigned short>(Idx,0));
+ OpcodeInfo.push_back(Idx);
}
+ // Figure out how many bits we used for the string index.
+ unsigned AsmStrBits = Log2_32_Ceil(MaxStringIdx);
+
// To reduce code size, we compactify common instructions into a few bits
// in the opcode-indexed table.
- // 16 bits to play with.
- unsigned BitsLeft = 16;
+ unsigned BitsLeft = 32-AsmStrBits;
std::vector<std::vector<std::string> > TableDrivenOperandPrinters;
- for (unsigned i = 0; ; ++i) {
+ bool isFirst = true;
+ while (1) {
std::vector<std::string> UniqueOperandCommands;
- // For the first operand check, add a default value that unhandled
- // instructions will use.
- if (i == 0)
- UniqueOperandCommands.push_back(" return false;\n");
+ // 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;
- FindUniqueOperandCommands(UniqueOperandCommands, InstIdxs, i);
+ std::vector<unsigned> NumInstOpsHandled;
+ FindUniqueOperandCommands(UniqueOperandCommands, InstIdxs,
+ NumInstOpsHandled);
// If we ran out of operands to print, we're done.
if (UniqueOperandCommands.empty()) break;
- // FIXME: GROW THEM MAXIMALLY.
-
// 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(std::cerr << "Not enough bits to densely encode " << NumBits
- << " more bits\n");
+ DOUT << "Not enough bits to densely encode " << NumBits
+ << " more bits\n";
break;
}
// 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)
- OpcodeInfo[i].second |= InstIdxs[i] << BitsLeft;
+ if (InstIdxs[i] != ~0U)
+ OpcodeInfo[i] |= InstIdxs[i] << (BitsLeft+AsmStrBits);
+ // 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(UniqueOperandCommands);
}
- O<<" static const struct { unsigned short StrIdx, Bits; } OpInfo[] = {\n";
+ O<<" static const unsigned OpInfo[] = {\n";
for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
- O << " { " << OpcodeInfo[i].first << ", " << OpcodeInfo[i].second
- << " },\t// " << NumberedInstructions[i]->TheDef->getName() << "\n";
+ O << " " << OpcodeInfo[i] << "U,\t// "
+ << NumberedInstructions[i]->TheDef->getName() << "\n";
}
// Add a dummy entry so the array init doesn't end with a comma.
- O << " { 65535, 65535 }\n";
+ O << " 0U\n";
O << " };\n\n";
// Emit the string itself.
O << "\";\n\n";
O << " if (MI->getOpcode() == TargetInstrInfo::INLINEASM) {\n"
+ << " O << \"\\t\";\n"
<< " printInlineAsm(MI);\n"
<< " return true;\n"
+ << " } else if (MI->getOpcode() == TargetInstrInfo::LABEL) {\n"
+ << " printLabel(MI);\n"
+ << " return true;\n"
+ << " } else if (MI->getOpcode() == TargetInstrInfo::DECLARE) {\n"
+ << " printDeclare(MI);\n"
+ << " return true;\n"
+ << " } else if (MI->getOpcode() == TargetInstrInfo::IMPLICIT_DEF) {\n"
+ << " printImplicitDef(MI);\n"
+ << " return true;\n"
<< " }\n\n";
+ O << " O << \"\\t\";\n\n";
+
O << " // Emit the opcode for the instruction.\n"
- << " O << AsmStrs+OpInfo[MI->getOpcode()].StrIdx;\n\n";
+ << " unsigned Bits = OpInfo[MI->getOpcode()];\n"
+ << " if (Bits == 0) return false;\n"
+ << " O << AsmStrs+(Bits & " << (1 << AsmStrBits)-1 << ");\n\n";
// Output the table driven operand information.
- O << " unsigned short Bits = OpInfo[MI->getOpcode()].Bits;\n";
-
- BitsLeft = 16;
+ BitsLeft = 32-AsmStrBits;
for (unsigned i = 0, e = TableDrivenOperandPrinters.size(); i != e; ++i) {
std::vector<std::string> &Commands = TableDrivenOperandPrinters[i];
BitsLeft -= NumBits;
O << "\n // Fragment " << i << " encoded into " << NumBits
- << " bits for " << Commands.size() << " unique commands.\n"
- << " switch ((Bits >> " << BitsLeft << ") & " << ((1 << NumBits)-1)
- << ") {\n"
- << " default: // unreachable.\n";
+ << " bits for " << Commands.size() << " unique commands.\n";
- // Print out all the cases.
- for (unsigned i = 0, e = Commands.size(); i != e; ++i) {
- O << " case " << i << ":\n";
- O << Commands[i];
- O << " break;\n";
+ if (Commands.size() == 2) {
+ // Emit two possibilitys with if/else.
+ O << " if ((Bits >> " << (BitsLeft+AsmStrBits) << ") & "
+ << ((1 << NumBits)-1) << ") {\n"
+ << Commands[1]
+ << " } else {\n"
+ << Commands[0]
+ << " }\n\n";
+ } else {
+ O << " switch ((Bits >> " << (BitsLeft+AsmStrBits) << ") & "
+ << ((1 << NumBits)-1) << ") {\n"
+ << " default: // unreachable.\n";
+
+ // Print out all the cases.
+ for (unsigned i = 0, e = Commands.size(); i != e; ++i) {
+ O << " case " << i << ":\n";
+ O << Commands[i];
+ O << " break;\n";
+ }
+ O << " }\n\n";
}
- O << " }\n\n";
}
- // Okay, go through and strip out the operand information that we just
- // emitted.
- unsigned NumOpsToRemove = TableDrivenOperandPrinters.size();
+ // Okay, delete instructions with no operand info left.
for (unsigned i = 0, e = Instructions.size(); i != e; ++i) {
// Entire instruction has been emitted?
AsmWriterInst &Inst = Instructions[i];
- if (Inst.Operands.size() <= NumOpsToRemove) {
+ if (Inst.Operands.empty()) {
Instructions.erase(Instructions.begin()+i);
- --i; --e;
- } else {
- Inst.Operands.erase(Inst.Operands.begin(),
- Inst.Operands.begin()+NumOpsToRemove);
+ --i; --e;
}
}
// elements in the vector.
std::reverse(Instructions.begin(), Instructions.end());
- // Find the opcode # of inline asm
- O << " switch (MI->getOpcode()) {\n";
- while (!Instructions.empty())
- EmitInstructions(Instructions, O);
-
- O << " }\n"
- " return true;\n"
- "}\n";
+ if (!Instructions.empty()) {
+ // Find the opcode # of inline asm.
+ O << " switch (MI->getOpcode()) {\n";
+ while (!Instructions.empty())
+ EmitInstructions(Instructions, O);
+
+ O << " }\n";
+ O << " return true;\n";
+ }
+
+ O << "}\n";
}