6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-12-30 02:32:08 +00:00
Code Issues Projects Releases Wiki Activity
34ce687144
llvm-6502/utils/TableGen/AsmWriterEmitter.cpp
David Greene 8f6353fb2f Make a bunch of changes suggested by Chris and others to improve 
efficiency.  Fix a few formatting bugs along the way.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@76601 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-21 18:16:24 +00:00

939 lines
34 KiB
C++
Raw Blame History

//===- AsmWriterEmitter.cpp - Generate an assembly writer -----------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This tablegen backend is emits an assembly printer for the current target.
// Note that this is currently fairly skeletal, but will grow over time.
//
//===----------------------------------------------------------------------===//
#include "AsmWriterEmitter.h"
#include "CodeGenTarget.h"
#include "Record.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MathExtras.h"
#include <algorithm>
#include <sstream>
#include <iostream>
using namespace llvm;
static bool isIdentChar(char C) {
return (C >= 'a' && C <= 'z') ||
(C >= 'A' && C <= 'Z') ||
(C >= '0' && C <= '9') ||
C == '_';
}
// This should be an anon namespace, this works around a GCC warning.
namespace llvm {
struct AsmWriterOperand {
enum OpType {
isLiteralTextOperand,
isMachineInstrOperand,
isLiteralStatementOperand
} 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;
// To make VS STL happy
AsmWriterOperand(OpType op = isLiteralTextOperand):OperandType(op) {}
AsmWriterOperand(const std::string &LitStr,
OpType op = isLiteralTextOperand)
: OperandType(op), Str(LitStr) {}
AsmWriterOperand(const std::string &Printer, unsigned OpNo,
const std::string &Modifier,
OpType op = isMachineInstrOperand)
: OperandType(op), 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 {
class AsmWriterInst {
public:
std::vector<AsmWriterOperand> Operands;
const CodeGenInstruction *CGI;
/// MAX_GROUP_NESTING_LEVEL - The maximum number of group nesting
/// levels we ever expect to see in an asm operand.
static const int MAX_GROUP_NESTING_LEVEL = 10;
/// GroupLevel - The level of nesting of the current operand
/// group, such as [reg + (reg + offset)]. -1 means we are not in
/// a group.
int GroupLevel;
/// GroupDelim - Remember the delimeter for a group operand.
char GroupDelim[MAX_GROUP_NESTING_LEVEL];
/// ReadingWhitespace - Tell whether we just read some whitespace.
bool ReadingWhitespace;
/// InGroup - Determine whether we are in the middle of an
/// operand group.
bool InGroup() const { return GroupLevel != -1; }
/// InWhitespace - Determine whether we are in the middle of
/// emitting whitespace.
bool InWhitespace() const { return ReadingWhitespace; }
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 + "\"; ";
if (OperandType == isLiteralStatementOperand) {
return Str;
}
if (OperandType == isLiteralStatementOperand) {
return Str;
}
if (OperandType == isLiteralStatementOperand) {
return 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)
: GroupLevel(-1), ReadingWhitespace(false) {
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.
// TODO: Recognize an operand separator to determine when to pad
// to the next operator.
if (DollarPos != LastEmitted) {
if (CurVariant == Variant || CurVariant == ~0U) {
for (; LastEmitted != DollarPos; ++LastEmitted)
switch (AsmString[LastEmitted]) {
case '\n':
assert(!InGroup() && "Missing matching group delimeter");
ReadingWhitespace = false;
AddLiteralString("\\n");
break;
case '\t':
if (!InGroup()) {
ReadingWhitespace = true;
}
AddLiteralString("\\t");
break;
case '"':
if (InWhitespace() && !InGroup())
Operands.push_back(
AsmWriterOperand(
"O.PadToColumn(TAI->getOperandColumn(OperandColumn++));\n",
AsmWriterOperand::isLiteralStatementOperand));
ReadingWhitespace = false;
AddLiteralString("\\\"");
break;
case '\\':
if (InWhitespace() && !InGroup())
Operands.push_back(
AsmWriterOperand(
"O.PadToColumn(TAI->getOperandColumn(OperandColumn++));\n",
AsmWriterOperand::isLiteralStatementOperand));
ReadingWhitespace = false;
AddLiteralString("\\\\");
break;
case '(': // Fallthrough
case '[':
if (InWhitespace() && !InGroup())
Operands.push_back(
AsmWriterOperand(
"O.PadToColumn(TAI->getOperandColumn(OperandColumn++));\n",
AsmWriterOperand::isLiteralStatementOperand));
ReadingWhitespace = false;
++GroupLevel;
assert(GroupLevel < MAX_GROUP_NESTING_LEVEL
&& "Exceeded maximum operand group nesting level");
GroupDelim[GroupLevel] = AsmString[LastEmitted];
AddLiteralString(std::string(1, AsmString[LastEmitted]));
break;
case ')': // Fallthrough
case ']':
if (InWhitespace() && !InGroup())
Operands.push_back(
AsmWriterOperand(
"O.PadToColumn(TAI->getOperandColumn(OperandColumn++));\n",
AsmWriterOperand::isLiteralStatementOperand));
ReadingWhitespace = false;
if (AsmString[LastEmitted] == ')')
assert(GroupDelim[GroupLevel] == '(' && "Mismatched delimeters");
else
assert(GroupDelim[GroupLevel] == '[' && "Mismatched delimeters");
--GroupLevel;
assert(GroupLevel > -2 && "Too many end delimeters!");
AddLiteralString(std::string(1, AsmString[LastEmitted]));
break;
default:
if (AsmString[LastEmitted] != ' ' &&
AsmString[LastEmitted] != '\t') {
if (!InGroup() && InWhitespace())
Operands.push_back(
AsmWriterOperand(
"O.PadToColumn(TAI->getOperandColumn(OperandColumn++));\n",
AsmWriterOperand::isLiteralStatementOperand));
ReadingWhitespace = false;
}
else
if (!InGroup())
ReadingWhitespace = true;
AddLiteralString(std::string(1, AsmString[LastEmitted]));
break;
}
} else {
LastEmitted = DollarPos;
}
} else if (AsmString[DollarPos] == '\\') {
if (DollarPos+1 != AsmString.size() &&
(CurVariant == Variant || CurVariant == ~0U)) {
if (AsmString[DollarPos+1] == 'n') {
assert(!InGroup() && "Missing matching group delimeter");
ReadingWhitespace = false;
AddLiteralString("\\n");
} else if (AsmString[DollarPos+1] == 't') {
if (!InGroup()) {
ReadingWhitespace = true;
}
AddLiteralString("\\t");
} else if (std::string("${|}\\").find(AsmString[DollarPos+1])
!= std::string::npos) {
if (InWhitespace() && !InGroup())
Operands.push_back(
AsmWriterOperand(
"O.PadToColumn(TAI->getOperandColumn(OperandColumn++));\n",
AsmWriterOperand::isLiteralStatementOperand));
ReadingWhitespace = false;
if (AsmString[DollarPos+1] == '{') {
++GroupLevel;
assert(GroupLevel < MAX_GROUP_NESTING_LEVEL
&& "Exceeded maximum operand group nesting level");
GroupDelim[GroupLevel] = AsmString[DollarPos+1];
} else if (AsmString[DollarPos+1] == '}') {
assert(GroupDelim[GroupLevel] == '{' && "Mismatched delimeters");
--GroupLevel;
assert(GroupLevel > -2 && "Too many end delimeters!");
}
AddLiteralString(std::string(1, AsmString[DollarPos+1]));
} else {
throw "Non-supported escaped character found in instruction '" +
CGI.TheDef->getName() + "'!";
}
LastEmitted = DollarPos+2;
continue;
}
} else if (AsmString[DollarPos] == '{') {
if (CurVariant != ~0U)
throw "Nested variants found for instruction '" +
CGI.TheDef->getName() + "'!";
LastEmitted = DollarPos+1;
CurVariant = 0; // We are now inside of the variant!
} else if (AsmString[DollarPos] == '|') {
if (CurVariant == ~0U)
throw "'|' character found outside of a variant in instruction '"
+ CGI.TheDef->getName() + "'!";
++CurVariant;
++LastEmitted;
} else if (AsmString[DollarPos] == '}') {
if (CurVariant == ~0U)
throw "'}' character found outside of a variant in instruction '"
+ CGI.TheDef->getName() + "'!";
++LastEmitted;
CurVariant = ~0U;
} else if (DollarPos+1 != AsmString.size() &&
AsmString[DollarPos+1] == '$') {
if (CurVariant == Variant || CurVariant == ~0U) {
if (InWhitespace() && !InGroup())
Operands.push_back(
AsmWriterOperand(
"O.PadToColumn(TAI->getOperandColumn(OperandColumn++));\n",
AsmWriterOperand::isLiteralStatementOperand));
ReadingWhitespace = false;
AddLiteralString("$"); // "$$" -> $
}
LastEmitted = DollarPos+2;
} else {
if (InWhitespace() && !InGroup())
Operands.push_back(
AsmWriterOperand(
"O.PadToColumn(TAI->getOperandColumn(OperandColumn++));\n",
AsmWriterOperand::isLiteralStatementOperand));
ReadingWhitespace = false;
// 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 (CurVariant == Variant || CurVariant == ~0U) {
unsigned MIOp = OpInfo.MIOperandNo;
Operands.push_back(AsmWriterOperand(OpInfo.PrinterMethodName, MIOp,
Modifier));
}
}
LastEmitted = VarEnd;
}
}
Operands.push_back(
AsmWriterOperand("EmitComments(*MI);\n",
AsmWriterOperand::isLiteralStatementOperand));
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, raw_ostream &O) {
O << " case " << OpsToPrint.back().first << ": ";
AsmWriterOperand TheOp = OpsToPrint.back().second;
OpsToPrint.pop_back();
// Check to see if any other operands are identical in this list, and if so,
// emit a case label for them.
for (unsigned i = OpsToPrint.size(); i != 0; --i)
if (OpsToPrint[i-1].second == TheOp) {
O << "\n case " << OpsToPrint[i-1].first << ": ";
OpsToPrint.erase(OpsToPrint.begin()+i-1);
}
// Finally, emit the code.
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,
raw_ostream &O) {
AsmWriterInst FirstInst = Insts.back();
Insts.pop_back();
std::vector<AsmWriterInst> SimilarInsts;
unsigned DifferingOperand = ~0;
for (unsigned i = Insts.size(); i != 0; --i) {
unsigned DiffOp = Insts[i-1].MatchesAllButOneOp(FirstInst);
if (DiffOp != ~1U) {
if (DifferingOperand == ~0U) // First match!
DifferingOperand = DiffOp;
// If this differs in the same operand as the rest of the instructions in
// this class, move it to the SimilarInsts list.
if (DifferingOperand == DiffOp || DiffOp == ~0U) {
SimilarInsts.push_back(Insts[i-1]);
Insts.erase(Insts.begin()+i-1);
}
}
}
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 (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].getCode();
} else {
// If this is the operand that varies between all of the instructions,
// emit a switch for just this operand now.
O << " switch (MI->getOpcode()) {\n";
std::vector<std::pair<std::string, AsmWriterOperand> > OpsToPrint;
OpsToPrint.push_back(std::make_pair(FirstInst.CGI->Namespace + "::" +
FirstInst.CGI->TheDef->getName(),
FirstInst.Operands[i]));
for (unsigned si = 0, e = SimilarInsts.size(); si != e; ++si) {
AsmWriterInst &AWI = SimilarInsts[si];
OpsToPrint.push_back(std::make_pair(AWI.CGI->Namespace+"::"+
AWI.CGI->TheDef->getName(),
AWI.Operands[i]));
}
std::reverse(OpsToPrint.begin(), OpsToPrint.end());
while (!OpsToPrint.empty())
PrintCases(OpsToPrint, O);
O << " }";
}
O << "\n";
}
O << " break;\n";
}
void AsmWriterEmitter::
FindUniqueOperandCommands(std::vector<std::string> &UniqueOperandCommands,
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, DBG_LABEL, etc.
std::string Command;
if (Inst->Operands.empty())
continue; // Instruction already done.
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;
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(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;
// 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)) {
// 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() > FirstInst->Operands.size())
MaxSize = std::max(MaxSize, OtherInst->Operands.size());
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 &&
// Don't early-out too soon. Other instructions in this
// group may have more operands.
FirstInst->Operands.size() == MaxSize) {
Command += " return true;\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];
}
}
void AsmWriterEmitter::run(raw_ostream &O) {
EmitSourceFileHeader("Assembly Writer Source Fragment", 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";
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));
// Get the instruction numbering.
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.
for (unsigned i = 0, e = Instructions.size(); i != e; ++i)
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.push_back(0); // "\0"
AggregateString.push_back(0); // "\0"
/// 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) {
// 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.
MaxStringIdx = Entry = AggregateString.size();
std::string Str = AWI->Operands[0].Str;
UnescapeString(Str);
AggregateString += Str;
AggregateString += '\0';
}
Idx = Entry;
// Nuke the string from the operand list. It is now handled!
AWI->Operands.erase(AWI->Operands.begin());
}
OpcodeInfo.push_back(Idx);
}
// Figure out how many bits we used for the string index.
unsigned AsmStrBits = Log2_32_Ceil(MaxStringIdx+1);
// To reduce code size, we compactify common instructions into a few bits
// in the opcode-indexed table.
unsigned BitsLeft = 32-AsmStrBits;
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,
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) {
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)
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 unsigned OpInfo[] = {\n";
for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
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 << " 0U\n";
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 << " processDebugLoc(MI->getDebugLoc());\n\n";
O << "\n#ifndef NO_ASM_WRITER_BOILERPLATE\n";
O << " if (MI->getOpcode() == TargetInstrInfo::INLINEASM) {\n"
<< " O << \"\\t\";\n"
<< " printInlineAsm(MI);\n"
<< " return true;\n"
<< " } else if (MI->isLabel()) {\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 << "\n#endif\n";
O << " O << \"\\t\";\n\n";
O << " // Emit the opcode for the instruction.\n"
<< " unsigned Bits = OpInfo[MI->getOpcode()];\n"
<< " if (Bits == 0) return false;\n\n";
O << " unsigned OperandColumn = 1;\n\n"
<< " if (TAI->getOperandColumn(1) > 0) {\n"
<< " // Don't emit trailing whitespace, let the column padding do it. This\n"
<< " // guarantees that a stray long opcode + tab won't upset the alignment.\n"
<< " unsigned OpLength = std::strlen(AsmStrs+(Bits & " << (1 << AsmStrBits)-1 << "));\n"
<< " if (OpLength > 0 &&\n"
<< " ((AsmStrs+(Bits & " << (1 << AsmStrBits)-1 << "))[OpLength-1] == ' ' ||\n"
<< " (AsmStrs+(Bits & " << (1 << AsmStrBits)-1 << "))[OpLength-1] == '\\t')) {\n"
<< " do {\n"
<< " --OpLength;\n"
<< " } while ((AsmStrs+(Bits & " << (1 << AsmStrBits)-1 << "))[OpLength-1] == ' ' ||\n"
<< " (AsmStrs+(Bits & " << (1 << AsmStrBits)-1 << "))[OpLength-1] == '\\t');\n"
<< " for (unsigned Idx = 0; Idx < OpLength; ++Idx)\n"
<< " O << (AsmStrs+(Bits & " << (1 << AsmStrBits)-1 << "))[Idx];\n"
<< " O.PadToColumn(TAI->getOperandColumn(OperandColumn++), 1);\n"
<< " }\n"
<< " } else {\n"
<< " O << AsmStrs+(Bits & " << (1 << AsmStrBits)-1 << ");\n"
<< " }\n\n";
// Output the table driven operand information.
BitsLeft = 32-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.
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) << ") & "
<< ((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";
}
}
// 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.empty()) {
Instructions.erase(Instructions.begin()+i);
--i; --e;
}
}
// 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());
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 << " return true;\n";
O << "}\n";
}
Reference in New Issue View Git Blame Copy Permalink