//===- CodeEmitterGen.cpp - Code Emitter Generator ------------------------===//
//
-// FIXME: Document.
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// CodeEmitterGen uses the descriptions of instructions and their fields to
+// construct an automated code emitter: a function that, given a MachineInstr,
+// returns the (currently, 32-bit unsigned) value of the instruction.
//
//===----------------------------------------------------------------------===//
#include "CodeEmitterGen.h"
+#include "CodeGenTarget.h"
#include "Record.h"
-#include "Support/Statistic.h"
+#include "llvm/ADT/StringExtras.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+using namespace llvm;
-bool CodeEmitterGen::run(std::ostream &o) {
- std::vector<Record*> Insts;
+static cl::opt<bool>
+MCEmitter("mc-code-emitter",
+ cl::desc("Generate CodeEmitter for use with the MC library."),
+ cl::init(false));
- const std::map<std::string, Record*> &Defs = Records.getDefs();
- Record *Inst = Records.getClass("Instruction");
- assert(Inst && "Couldn't find Instruction class!");
+void CodeEmitterGen::reverseBits(std::vector<Record*> &Insts) {
+ for (std::vector<Record*>::iterator I = Insts.begin(), E = Insts.end();
+ I != E; ++I) {
+ Record *R = *I;
+ if (R->getValueAsString("Namespace") == "TargetOpcode")
+ continue;
- for (std::map<std::string, Record*>::const_iterator I = Defs.begin(),
- E = Defs.end(); I != E; ++I)
- if (I->second->isSubClassOf(Inst))
- Insts.push_back(I->second);
+ BitsInit *BI = R->getValueAsBitsInit("Inst");
- std::string Namespace = "V9::";
- std::string ClassName = "SparcV9CodeEmitter::";
+ unsigned numBits = BI->getNumBits();
+ BitsInit *NewBI = new BitsInit(numBits);
+ for (unsigned bit = 0, end = numBits / 2; bit != end; ++bit) {
+ unsigned bitSwapIdx = numBits - bit - 1;
+ Init *OrigBit = BI->getBit(bit);
+ Init *BitSwap = BI->getBit(bitSwapIdx);
+ NewBI->setBit(bit, BitSwap);
+ NewBI->setBit(bitSwapIdx, OrigBit);
+ }
+ if (numBits % 2) {
+ unsigned middle = (numBits + 1) / 2;
+ NewBI->setBit(middle, BI->getBit(middle));
+ }
- //const std::string &Namespace = Inst->getValue("Namespace")->getName();
- o << "unsigned " << ClassName
- << "getBinaryCodeForInstr(MachineInstr &MI) {\n"
- << " unsigned Value = 0;\n"
- << " DEBUG(std::cerr << MI);\n"
- << " switch (MI.getOpcode()) {\n";
- for (std::vector<Record*>::iterator I = Insts.begin(), E = Insts.end();
- I != E; ++I)
- {
- Record *R = *I;
- o << " case " << Namespace << R->getName() << ": {\n"
- << " DEBUG(std::cerr << \"Emitting " << R->getName() << "\\n\");\n";
+ // Update the bits in reversed order so that emitInstrOpBits will get the
+ // correct endianness.
+ R->getValue("Inst")->setValue(NewBI);
+ }
+}
- const RecordVal *InstVal = R->getValue("Inst");
- if (!InstVal) {
- std::cerr << "No 'Inst' record found in target description file!\n";
- return true;
+// If the VarBitInit at position 'bit' matches the specified variable then
+// return the variable bit position. Otherwise return -1.
+int CodeEmitterGen::getVariableBit(const std::string &VarName,
+ BitsInit *BI, int bit) {
+ if (VarBitInit *VBI = dynamic_cast<VarBitInit*>(BI->getBit(bit))) {
+ TypedInit *TI = VBI->getVariable();
+
+ if (VarInit *VI = dynamic_cast<VarInit*>(TI)) {
+ if (VI->getName() == VarName) return VBI->getBitNum();
}
+ }
- Init *InitVal = InstVal->getValue();
- assert(dynamic_cast<BitsInit*>(InitVal) &&
- "Can only handle undefined bits<> types!");
- BitsInit *BI = (BitsInit*)InitVal;
+ return -1;
+}
- unsigned Value = 0;
- const std::vector<RecordVal> &Vals = R->getValues();
+void CodeEmitterGen::run(raw_ostream &o) {
+ CodeGenTarget Target;
+ std::vector<Record*> Insts = Records.getAllDerivedDefinitions("Instruction");
+
+ // For little-endian instruction bit encodings, reverse the bit order
+ if (Target.isLittleEndianEncoding()) reverseBits(Insts);
+
+ EmitSourceFileHeader("Machine Code Emitter", o);
+ std::string Namespace = Insts[0]->getValueAsString("Namespace") + "::";
+
+ const std::vector<const CodeGenInstruction*> &NumberedInstructions =
+ Target.getInstructionsByEnumValue();
+
+ // Emit function declaration
+ o << "unsigned " << Target.getName() << "CodeEmitter::"
+ << "getBinaryCodeForInstr(const MachineInstr &MI) const {\n";
+
+ // Emit instruction base values
+ o << " static const unsigned InstBits[] = {\n";
+ for (std::vector<const CodeGenInstruction*>::const_iterator
+ IN = NumberedInstructions.begin(),
+ EN = NumberedInstructions.end();
+ IN != EN; ++IN) {
+ const CodeGenInstruction *CGI = *IN;
+ Record *R = CGI->TheDef;
+
+ if (R->getValueAsString("Namespace") == "TargetOpcode") {
+ o << " 0U,\n";
+ continue;
+ }
+
+ BitsInit *BI = R->getValueAsBitsInit("Inst");
- DEBUG(o << " // prefilling: ");
// Start by filling in fixed values...
+ unsigned Value = 0;
for (unsigned i = 0, e = BI->getNumBits(); i != e; ++i) {
if (BitInit *B = dynamic_cast<BitInit*>(BI->getBit(e-i-1))) {
Value |= B->getValue() << (e-i-1);
- DEBUG(o << B->getValue());
- } else {
- DEBUG(o << "0");
}
}
- DEBUG(o << "\n");
-
- DEBUG(o << " // " << *InstVal << "\n");
- o << " Value = " << Value << "U;\n\n";
-
- // Loop over all of the fields in the instruction adding in any
- // contributions to this value (due to bit references).
- //
- unsigned op = 0;
- std::map<const std::string,unsigned> OpOrder;
- std::map<const std::string,bool> OpContinuous;
- for (unsigned i = 0, e = Vals.size(); i != e; ++i) {
- if (Vals[i].getName() != "Inst" &&
- !Vals[i].getValue()->isComplete() &&
- /* ignore annul and predict bits since no one sets them yet */
- Vals[i].getName() != "annul" &&
- Vals[i].getName() != "predict")
- {
- o << " // op" << op << ": " << Vals[i].getName() << "\n"
- << " int64_t op" << op
- <<" = getMachineOpValue(MI, MI.getOperand("<<op<<"));\n";
- //<< " MachineOperand &op" << op <<" = MI.getOperand("<<op<<");\n";
- OpOrder[Vals[i].getName()] = op++;
+ o << " " << Value << "U," << '\t' << "// " << R->getName() << "\n";
+ }
+ o << " 0U\n };\n";
- // Is the operand continuous? If so, we can just mask and OR it in
- // instead of doing it bit-by-bit, saving a lot in runtime cost.
- const BitsInit *InstInit = BI;
- int beginBitInVar = -1, endBitInVar = -1,
- beginBitInInst = -1, endBitInInst = -1;
- bool continuous = true;
-
- for (int bit = InstInit->getNumBits()-1; bit >= 0; --bit) {
- if (VarBitInit *VBI =
- dynamic_cast<VarBitInit*>(InstInit->getBit(bit))) {
- TypedInit *TI = VBI->getVariable();
- if (VarInit *VI = dynamic_cast<VarInit*>(TI)) {
- // only process the current variable
- if (VI->getName() != Vals[i].getName())
- continue;
-
- if (beginBitInVar == -1)
- beginBitInVar = VBI->getBitNum();
-
- if (endBitInVar == -1)
- endBitInVar = VBI->getBitNum();
- else {
- if (endBitInVar == (int)VBI->getBitNum() + 1)
- endBitInVar = VBI->getBitNum();
- else {
- continuous = false;
- break;
- }
- }
+ // Map to accumulate all the cases.
+ std::map<std::string, std::vector<std::string> > CaseMap;
- if (beginBitInInst == -1)
- beginBitInInst = bit;
- if (endBitInInst == -1)
- endBitInInst = bit;
- else {
- if (endBitInInst == bit + 1)
- endBitInInst = bit;
- else {
- continuous = false;
- break;
- }
- }
+ // Construct all cases statement for each opcode
+ for (std::vector<Record*>::iterator IC = Insts.begin(), EC = Insts.end();
+ IC != EC; ++IC) {
+ Record *R = *IC;
+ if (R->getValueAsString("Namespace") == "TargetOpcode")
+ continue;
+ const std::string &InstName = R->getName();
+ std::string Case("");
- // maintain same distance between bits in field and bits in
- // instruction. if the relative distances stay the same
- // throughout,
- if ((beginBitInVar - (int)VBI->getBitNum()) !=
- (beginBitInInst - bit))
- {
- continuous = false;
- break;
- }
- }
- }
- }
+ BitsInit *BI = R->getValueAsBitsInit("Inst");
+ const std::vector<RecordVal> &Vals = R->getValues();
+ CodeGenInstruction &CGI = Target.getInstruction(R);
- DEBUG(o << " // Var: begin = " << beginBitInVar
- << ", end = " << endBitInVar
- << "; Inst: begin = " << beginBitInInst
- << ", end = " << endBitInInst << "\n");
-
- if (continuous) {
- DEBUG(o << " // continuous: op" << OpOrder[Vals[i].getName()]
- << "\n");
-
- // Mask off the right bits
- // Low mask (ie. shift, if necessary)
- if (endBitInVar != 0) {
- o << " op" << OpOrder[Vals[i].getName()]
- << " >>= " << endBitInVar << ";\n";
- beginBitInVar -= endBitInVar;
- endBitInVar = 0;
- }
+ // Loop over all of the fields in the instruction, determining which are the
+ // operands to the instruction.
+ unsigned NumberedOp = 0;
+ for (unsigned i = 0, e = Vals.size(); i != e; ++i) {
+ if (!Vals[i].getPrefix() && !Vals[i].getValue()->isComplete()) {
+ // Is the operand continuous? If so, we can just mask and OR it in
+ // instead of doing it bit-by-bit, saving a lot in runtime cost.
+ const std::string &VarName = Vals[i].getName();
+ bool gotOp = false;
- // High mask
- o << " op" << OpOrder[Vals[i].getName()]
- << " &= (1<<" << beginBitInVar+1 << ") - 1;\n";
+ for (int bit = BI->getNumBits()-1; bit >= 0; ) {
+ int varBit = getVariableBit(VarName, BI, bit);
- // Shift the value to the correct place (according to place in instr)
- if (endBitInInst != 0)
- o << " op" << OpOrder[Vals[i].getName()]
- << " <<= " << endBitInInst << ";\n";
+ if (varBit == -1) {
+ --bit;
+ } else {
+ int beginInstBit = bit;
+ int beginVarBit = varBit;
+ int N = 1;
- // Just OR in the result
- o << " Value |= op" << OpOrder[Vals[i].getName()] << ";\n";
- }
+ for (--bit; bit >= 0;) {
+ varBit = getVariableBit(VarName, BI, bit);
+ if (varBit == -1 || varBit != (beginVarBit - N)) break;
+ ++N;
+ --bit;
+ }
- // otherwise, will be taken care of in the loop below using this value:
- OpContinuous[Vals[i].getName()] = continuous;
- }
- }
+ if (!gotOp) {
+ // If the operand matches by name, reference according to that
+ // operand number. Non-matching operands are assumed to be in
+ // order.
+ unsigned OpIdx;
+ if (CGI.Operands.hasOperandNamed(VarName, OpIdx)) {
+ // Get the machine operand number for the indicated operand.
+ OpIdx = CGI.Operands[OpIdx].MIOperandNo;
+ assert (!CGI.Operands.isFlatOperandNotEmitted(OpIdx) &&
+ "Explicitly used operand also marked as not emitted!");
+ } else {
+ /// If this operand is not supposed to be emitted by the
+ /// generated emitter, skip it.
+ while (CGI.Operands.isFlatOperandNotEmitted(NumberedOp))
+ ++NumberedOp;
+ OpIdx = NumberedOp++;
+ }
+ std::pair<unsigned, unsigned> SO =
+ CGI.Operands.getSubOperandNumber(OpIdx);
+ std::string &EncoderMethodName =
+ CGI.Operands[SO.first].EncoderMethodName;
- for (unsigned f = 0, e = Vals.size(); f != e; ++f) {
- if (Vals[f].getPrefix()) {
- BitsInit *FieldInitializer = (BitsInit*)Vals[f].getValue();
-
- // Scan through the field looking for bit initializers of the current
- // variable...
- for (int i = FieldInitializer->getNumBits()-1; i >= 0; --i) {
- if (BitInit *BI=dynamic_cast<BitInit*>(FieldInitializer->getBit(i)))
- {
- DEBUG(o << " // bit init: f: " << f << ", i: " << i << "\n");
- } else if (UnsetInit *UI =
- dynamic_cast<UnsetInit*>(FieldInitializer->getBit(i))) {
- DEBUG(o << " // unset init: f: " << f << ", i: " << i << "\n");
- } else if (VarBitInit *VBI =
- dynamic_cast<VarBitInit*>(FieldInitializer->getBit(i))) {
- TypedInit *TI = VBI->getVariable();
- if (VarInit *VI = dynamic_cast<VarInit*>(TI)) {
- // If the bits of the field are laid out consecutively in the
- // instruction, then instead of separately ORing in bits, just
- // mask and shift the entire field for efficiency.
- if (OpContinuous[VI->getName()]) {
- // already taken care of in the loop above, thus there is no
- // need to individually OR in the bits
-
- // for debugging, output the regular version anyway, commented
- DEBUG(o << " // Value |= getValueBit(op"
- << OpOrder[VI->getName()] << ", " << VBI->getBitNum()
- << ")" << " << " << i << ";\n");
+ // If the source operand has a custom encoder, use it. This will
+ // get the encoding for all of the suboperands.
+ if (!EncoderMethodName.empty()) {
+ // A custom encoder has all of the information for the
+ // sub-operands, if there are more than one, so only
+ // query the encoder once per source operand.
+ if (SO.second == 0) {
+ Case += " // op: " + VarName + "\n"
+ + " op = " + EncoderMethodName + "(MI, "
+ + utostr(OpIdx) + ");\n";
+ }
} else {
- o << " Value |= getValueBit(op" << OpOrder[VI->getName()]
- << ", " << VBI->getBitNum()
- << ")" << " << " << i << ";\n";
+ Case += " // op: " + VarName + "\n"
+ + " op = getMachineOpValue(MI, MI.getOperand("
+ + utostr(OpIdx) + "));\n";
}
- } else if (FieldInit *FI = dynamic_cast<FieldInit*>(TI)) {
- // FIXME: implement this!
- o << "FIELD INIT not implemented yet!\n";
+ gotOp = true;
+ }
+
+ unsigned opMask = ~0U >> (32-N);
+ int opShift = beginVarBit - N + 1;
+ opMask <<= opShift;
+ opShift = beginInstBit - beginVarBit;
+
+ if (opShift > 0) {
+ Case += " Value |= (op & " + utostr(opMask) + "U) << "
+ + itostr(opShift) + ";\n";
+ } else if (opShift < 0) {
+ Case += " Value |= (op & " + utostr(opMask) + "U) >> "
+ + itostr(-opShift) + ";\n";
} else {
- o << "Error: UNIMPLEMENTED\n";
+ Case += " Value |= op & " + utostr(opMask) + "U;\n";
}
}
}
- } else {
- // ignore annul and predict bits since no one sets them yet
- if (Vals[f].getName() == "annul" || Vals[f].getName() == "predict") {
- o << " // found " << Vals[f].getName() << "\n";
- }
}
}
+ std::vector<std::string> &InstList = CaseMap[Case];
+ InstList.push_back(InstName);
+ }
+
+ // Emit initial function code
+ o << " const unsigned opcode = MI.getOpcode();\n"
+ << " unsigned Value = InstBits[opcode];\n"
+ << " unsigned op = 0;\n"
+ << " op = op; // suppress warning\n"
+ << " switch (opcode) {\n";
+
+ // Emit each case statement
+ std::map<std::string, std::vector<std::string> >::iterator IE, EE;
+ for (IE = CaseMap.begin(), EE = CaseMap.end(); IE != EE; ++IE) {
+ const std::string &Case = IE->first;
+ std::vector<std::string> &InstList = IE->second;
+
+ for (int i = 0, N = InstList.size(); i < N; i++) {
+ if (i) o << "\n";
+ o << " case " << Namespace << InstList[i] << ":";
+ }
+ o << " {\n";
+ o << Case;
o << " break;\n"
<< " }\n";
}
+ // Default case: unhandled opcode
o << " default:\n"
- << " DEBUG(std::cerr << \"Not supported instr: \" << MI << \"\\n\");\n"
- << " abort();\n"
+ << " std::string msg;\n"
+ << " raw_string_ostream Msg(msg);\n"
+ << " Msg << \"Not supported instr: \" << MI;\n"
+ << " report_fatal_error(Msg.str());\n"
<< " }\n"
<< " return Value;\n"
- << "}\n";
- return false;
+ << "}\n\n";
}
projects / oota-llvm.git / blobdiff