1 //===- MIRParser.cpp - MIR serialization format parser implementation -----===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the class that parses the optional LLVM IR and machine
11 // functions that are stored in MIR files.
13 //===----------------------------------------------------------------------===//
15 #include "llvm/CodeGen/MIRParser/MIRParser.h"
17 #include "llvm/ADT/DenseMap.h"
18 #include "llvm/ADT/StringRef.h"
19 #include "llvm/ADT/StringMap.h"
20 #include "llvm/ADT/STLExtras.h"
21 #include "llvm/AsmParser/Parser.h"
22 #include "llvm/AsmParser/SlotMapping.h"
23 #include "llvm/CodeGen/MachineConstantPool.h"
24 #include "llvm/CodeGen/MachineFunction.h"
25 #include "llvm/CodeGen/MachineFrameInfo.h"
26 #include "llvm/CodeGen/MachineRegisterInfo.h"
27 #include "llvm/CodeGen/MIRYamlMapping.h"
28 #include "llvm/IR/BasicBlock.h"
29 #include "llvm/IR/DiagnosticInfo.h"
30 #include "llvm/IR/Instructions.h"
31 #include "llvm/IR/LLVMContext.h"
32 #include "llvm/IR/Module.h"
33 #include "llvm/IR/ValueSymbolTable.h"
34 #include "llvm/Support/LineIterator.h"
35 #include "llvm/Support/SMLoc.h"
36 #include "llvm/Support/SourceMgr.h"
37 #include "llvm/Support/MemoryBuffer.h"
38 #include "llvm/Support/YAMLTraits.h"
45 /// This class implements the parsing of LLVM IR that's embedded inside a MIR
51 StringMap<std::unique_ptr<yaml::MachineFunction>> Functions;
53 /// Maps from register class names to register classes.
54 StringMap<const TargetRegisterClass *> Names2RegClasses;
57 MIRParserImpl(std::unique_ptr<MemoryBuffer> Contents, StringRef Filename,
58 LLVMContext &Context);
60 void reportDiagnostic(const SMDiagnostic &Diag);
62 /// Report an error with the given message at unknown location.
64 /// Always returns true.
65 bool error(const Twine &Message);
67 /// Report an error with the given message at the given location.
69 /// Always returns true.
70 bool error(SMLoc Loc, const Twine &Message);
72 /// Report a given error with the location translated from the location in an
73 /// embedded string literal to a location in the MIR file.
75 /// Always returns true.
76 bool error(const SMDiagnostic &Error, SMRange SourceRange);
78 /// Try to parse the optional LLVM module and the machine functions in the MIR
81 /// Return null if an error occurred.
82 std::unique_ptr<Module> parse();
84 /// Parse the machine function in the current YAML document.
86 /// \param NoLLVMIR - set to true when the MIR file doesn't have LLVM IR.
87 /// A dummy IR function is created and inserted into the given module when
88 /// this parameter is true.
90 /// Return true if an error occurred.
91 bool parseMachineFunction(yaml::Input &In, Module &M, bool NoLLVMIR);
93 /// Initialize the machine function to the state that's described in the MIR
96 /// Return true if error occurred.
97 bool initializeMachineFunction(MachineFunction &MF);
99 /// Initialize the machine basic block using it's YAML representation.
101 /// Return true if an error occurred.
102 bool initializeMachineBasicBlock(MachineFunction &MF, MachineBasicBlock &MBB,
103 const yaml::MachineBasicBlock &YamlMBB,
104 const PerFunctionMIParsingState &PFS);
107 initializeRegisterInfo(const MachineFunction &MF,
108 MachineRegisterInfo &RegInfo,
109 const yaml::MachineFunction &YamlMF,
110 DenseMap<unsigned, unsigned> &VirtualRegisterSlots);
112 bool initializeFrameInfo(const Function &F, MachineFrameInfo &MFI,
113 const yaml::MachineFunction &YamlMF,
114 DenseMap<unsigned, int> &StackObjectSlots,
115 DenseMap<unsigned, int> &FixedStackObjectSlots);
117 bool initializeConstantPool(MachineConstantPool &ConstantPool,
118 const yaml::MachineFunction &YamlMF,
119 const MachineFunction &MF,
120 DenseMap<unsigned, unsigned> &ConstantPoolSlots);
122 bool initializeJumpTableInfo(MachineFunction &MF,
123 const yaml::MachineJumpTable &YamlJTI,
124 PerFunctionMIParsingState &PFS);
127 /// Return a MIR diagnostic converted from an MI string diagnostic.
128 SMDiagnostic diagFromMIStringDiag(const SMDiagnostic &Error,
129 SMRange SourceRange);
131 /// Return a MIR diagnostic converted from an LLVM assembly diagnostic.
132 SMDiagnostic diagFromLLVMAssemblyDiag(const SMDiagnostic &Error,
133 SMRange SourceRange);
135 /// Create an empty function with the given name.
136 void createDummyFunction(StringRef Name, Module &M);
138 void initNames2RegClasses(const MachineFunction &MF);
140 /// Check if the given identifier is a name of a register class.
142 /// Return null if the name isn't a register class.
143 const TargetRegisterClass *getRegClass(const MachineFunction &MF,
147 } // end namespace llvm
149 MIRParserImpl::MIRParserImpl(std::unique_ptr<MemoryBuffer> Contents,
150 StringRef Filename, LLVMContext &Context)
151 : SM(), Filename(Filename), Context(Context) {
152 SM.AddNewSourceBuffer(std::move(Contents), SMLoc());
155 bool MIRParserImpl::error(const Twine &Message) {
156 Context.diagnose(DiagnosticInfoMIRParser(
157 DS_Error, SMDiagnostic(Filename, SourceMgr::DK_Error, Message.str())));
161 bool MIRParserImpl::error(SMLoc Loc, const Twine &Message) {
162 Context.diagnose(DiagnosticInfoMIRParser(
163 DS_Error, SM.GetMessage(Loc, SourceMgr::DK_Error, Message)));
167 bool MIRParserImpl::error(const SMDiagnostic &Error, SMRange SourceRange) {
168 assert(Error.getKind() == SourceMgr::DK_Error && "Expected an error");
169 reportDiagnostic(diagFromMIStringDiag(Error, SourceRange));
173 void MIRParserImpl::reportDiagnostic(const SMDiagnostic &Diag) {
174 DiagnosticSeverity Kind;
175 switch (Diag.getKind()) {
176 case SourceMgr::DK_Error:
179 case SourceMgr::DK_Warning:
182 case SourceMgr::DK_Note:
186 Context.diagnose(DiagnosticInfoMIRParser(Kind, Diag));
189 static void handleYAMLDiag(const SMDiagnostic &Diag, void *Context) {
190 reinterpret_cast<MIRParserImpl *>(Context)->reportDiagnostic(Diag);
193 std::unique_ptr<Module> MIRParserImpl::parse() {
194 yaml::Input In(SM.getMemoryBuffer(SM.getMainFileID())->getBuffer(),
195 /*Ctxt=*/nullptr, handleYAMLDiag, this);
198 if (!In.setCurrentDocument()) {
201 // Create an empty module when the MIR file is empty.
202 return llvm::make_unique<Module>(Filename, Context);
205 std::unique_ptr<Module> M;
206 bool NoLLVMIR = false;
207 // Parse the block scalar manually so that we can return unique pointer
208 // without having to go trough YAML traits.
209 if (const auto *BSN =
210 dyn_cast_or_null<yaml::BlockScalarNode>(In.getCurrentNode())) {
212 M = parseAssembly(MemoryBufferRef(BSN->getValue(), Filename), Error,
215 reportDiagnostic(diagFromLLVMAssemblyDiag(Error, BSN->getSourceRange()));
219 if (!In.setCurrentDocument())
222 // Create an new, empty module.
223 M = llvm::make_unique<Module>(Filename, Context);
227 // Parse the machine functions.
229 if (parseMachineFunction(In, *M, NoLLVMIR))
232 } while (In.setCurrentDocument());
237 bool MIRParserImpl::parseMachineFunction(yaml::Input &In, Module &M,
239 auto MF = llvm::make_unique<yaml::MachineFunction>();
240 yaml::yamlize(In, *MF, false);
243 auto FunctionName = MF->Name;
244 if (Functions.find(FunctionName) != Functions.end())
245 return error(Twine("redefinition of machine function '") + FunctionName +
247 Functions.insert(std::make_pair(FunctionName, std::move(MF)));
249 createDummyFunction(FunctionName, M);
250 else if (!M.getFunction(FunctionName))
251 return error(Twine("function '") + FunctionName +
252 "' isn't defined in the provided LLVM IR");
256 void MIRParserImpl::createDummyFunction(StringRef Name, Module &M) {
257 auto &Context = M.getContext();
258 Function *F = cast<Function>(M.getOrInsertFunction(
259 Name, FunctionType::get(Type::getVoidTy(Context), false)));
260 BasicBlock *BB = BasicBlock::Create(Context, "entry", F);
261 new UnreachableInst(Context, BB);
264 bool MIRParserImpl::initializeMachineFunction(MachineFunction &MF) {
265 auto It = Functions.find(MF.getName());
266 if (It == Functions.end())
267 return error(Twine("no machine function information for function '") +
268 MF.getName() + "' in the MIR file");
269 // TODO: Recreate the machine function.
270 const yaml::MachineFunction &YamlMF = *It->getValue();
271 if (YamlMF.Alignment)
272 MF.setAlignment(YamlMF.Alignment);
273 MF.setExposesReturnsTwice(YamlMF.ExposesReturnsTwice);
274 MF.setHasInlineAsm(YamlMF.HasInlineAsm);
275 PerFunctionMIParsingState PFS;
276 if (initializeRegisterInfo(MF, MF.getRegInfo(), YamlMF,
277 PFS.VirtualRegisterSlots))
279 if (initializeFrameInfo(*MF.getFunction(), *MF.getFrameInfo(), YamlMF,
280 PFS.StackObjectSlots, PFS.FixedStackObjectSlots))
282 if (!YamlMF.Constants.empty()) {
283 auto *ConstantPool = MF.getConstantPool();
284 assert(ConstantPool && "Constant pool must be created");
285 if (initializeConstantPool(*ConstantPool, YamlMF, MF,
286 PFS.ConstantPoolSlots))
290 const auto &F = *MF.getFunction();
291 for (const auto &YamlMBB : YamlMF.BasicBlocks) {
292 const BasicBlock *BB = nullptr;
293 const yaml::StringValue &Name = YamlMBB.Name;
294 if (!Name.Value.empty()) {
295 BB = dyn_cast_or_null<BasicBlock>(
296 F.getValueSymbolTable().lookup(Name.Value));
298 return error(Name.SourceRange.Start,
299 Twine("basic block '") + Name.Value +
300 "' is not defined in the function '" + MF.getName() +
303 auto *MBB = MF.CreateMachineBasicBlock(BB);
304 MF.insert(MF.end(), MBB);
306 PFS.MBBSlots.insert(std::make_pair(YamlMBB.ID, MBB)).second;
308 return error(Twine("redefinition of machine basic block with id #") +
312 if (YamlMF.BasicBlocks.empty())
313 return error(Twine("machine function '") + Twine(MF.getName()) +
314 "' requires at least one machine basic block in its body");
315 // Initialize the jump table after creating all the MBBs so that the MBB
316 // references can be resolved.
317 if (!YamlMF.JumpTableInfo.Entries.empty() &&
318 initializeJumpTableInfo(MF, YamlMF.JumpTableInfo, PFS))
320 // Initialize the machine basic blocks after creating them all so that the
321 // machine instructions parser can resolve the MBB references.
323 for (const auto &YamlMBB : YamlMF.BasicBlocks) {
324 if (initializeMachineBasicBlock(MF, *MF.getBlockNumbered(I++), YamlMBB,
331 bool MIRParserImpl::initializeMachineBasicBlock(
332 MachineFunction &MF, MachineBasicBlock &MBB,
333 const yaml::MachineBasicBlock &YamlMBB,
334 const PerFunctionMIParsingState &PFS) {
335 MBB.setAlignment(YamlMBB.Alignment);
336 if (YamlMBB.AddressTaken)
337 MBB.setHasAddressTaken();
338 MBB.setIsLandingPad(YamlMBB.IsLandingPad);
340 // Parse the successors.
341 for (const auto &MBBSource : YamlMBB.Successors) {
342 MachineBasicBlock *SuccMBB = nullptr;
343 if (parseMBBReference(SuccMBB, SM, MF, MBBSource.Value, PFS, IRSlots,
345 return error(Error, MBBSource.SourceRange);
346 // TODO: Report an error when adding the same successor more than once.
347 MBB.addSuccessor(SuccMBB);
349 // Parse the liveins.
350 for (const auto &LiveInSource : YamlMBB.LiveIns) {
352 if (parseNamedRegisterReference(Reg, SM, MF, LiveInSource.Value, PFS,
354 return error(Error, LiveInSource.SourceRange);
357 // Parse the instructions.
358 for (const auto &MISource : YamlMBB.Instructions) {
359 MachineInstr *MI = nullptr;
360 if (parseMachineInstr(MI, SM, MF, MISource.Value, PFS, IRSlots, Error))
361 return error(Error, MISource.SourceRange);
362 MBB.insert(MBB.end(), MI);
367 bool MIRParserImpl::initializeRegisterInfo(
368 const MachineFunction &MF, MachineRegisterInfo &RegInfo,
369 const yaml::MachineFunction &YamlMF,
370 DenseMap<unsigned, unsigned> &VirtualRegisterSlots) {
371 assert(RegInfo.isSSA());
374 assert(RegInfo.tracksLiveness());
375 if (!YamlMF.TracksRegLiveness)
376 RegInfo.invalidateLiveness();
377 RegInfo.enableSubRegLiveness(YamlMF.TracksSubRegLiveness);
379 // Parse the virtual register information.
380 for (const auto &VReg : YamlMF.VirtualRegisters) {
381 const auto *RC = getRegClass(MF, VReg.Class.Value);
383 return error(VReg.Class.SourceRange.Start,
384 Twine("use of undefined register class '") +
385 VReg.Class.Value + "'");
386 unsigned Reg = RegInfo.createVirtualRegister(RC);
387 // TODO: Report an error when the same virtual register with the same ID is
389 VirtualRegisterSlots.insert(std::make_pair(VReg.ID, Reg));
394 bool MIRParserImpl::initializeFrameInfo(
395 const Function &F, MachineFrameInfo &MFI,
396 const yaml::MachineFunction &YamlMF,
397 DenseMap<unsigned, int> &StackObjectSlots,
398 DenseMap<unsigned, int> &FixedStackObjectSlots) {
399 const yaml::MachineFrameInfo &YamlMFI = YamlMF.FrameInfo;
400 MFI.setFrameAddressIsTaken(YamlMFI.IsFrameAddressTaken);
401 MFI.setReturnAddressIsTaken(YamlMFI.IsReturnAddressTaken);
402 MFI.setHasStackMap(YamlMFI.HasStackMap);
403 MFI.setHasPatchPoint(YamlMFI.HasPatchPoint);
404 MFI.setStackSize(YamlMFI.StackSize);
405 MFI.setOffsetAdjustment(YamlMFI.OffsetAdjustment);
406 if (YamlMFI.MaxAlignment)
407 MFI.ensureMaxAlignment(YamlMFI.MaxAlignment);
408 MFI.setAdjustsStack(YamlMFI.AdjustsStack);
409 MFI.setHasCalls(YamlMFI.HasCalls);
410 MFI.setMaxCallFrameSize(YamlMFI.MaxCallFrameSize);
411 MFI.setHasOpaqueSPAdjustment(YamlMFI.HasOpaqueSPAdjustment);
412 MFI.setHasVAStart(YamlMFI.HasVAStart);
413 MFI.setHasMustTailInVarArgFunc(YamlMFI.HasMustTailInVarArgFunc);
415 // Initialize the fixed frame objects.
416 for (const auto &Object : YamlMF.FixedStackObjects) {
418 if (Object.Type != yaml::FixedMachineStackObject::SpillSlot)
419 ObjectIdx = MFI.CreateFixedObject(Object.Size, Object.Offset,
420 Object.IsImmutable, Object.IsAliased);
422 ObjectIdx = MFI.CreateFixedSpillStackObject(Object.Size, Object.Offset);
423 MFI.setObjectAlignment(ObjectIdx, Object.Alignment);
424 // TODO: Report an error when objects are redefined.
425 FixedStackObjectSlots.insert(std::make_pair(Object.ID, ObjectIdx));
428 // Initialize the ordinary frame objects.
429 for (const auto &Object : YamlMF.StackObjects) {
431 const AllocaInst *Alloca = nullptr;
432 const yaml::StringValue &Name = Object.Name;
433 if (!Name.Value.empty()) {
434 Alloca = dyn_cast_or_null<AllocaInst>(
435 F.getValueSymbolTable().lookup(Name.Value));
437 return error(Name.SourceRange.Start,
438 "alloca instruction named '" + Name.Value +
439 "' isn't defined in the function '" + F.getName() +
442 if (Object.Type == yaml::MachineStackObject::VariableSized)
443 ObjectIdx = MFI.CreateVariableSizedObject(Object.Alignment, Alloca);
445 ObjectIdx = MFI.CreateStackObject(
446 Object.Size, Object.Alignment,
447 Object.Type == yaml::MachineStackObject::SpillSlot, Alloca);
448 MFI.setObjectOffset(ObjectIdx, Object.Offset);
449 // TODO: Report an error when objects are redefined.
450 StackObjectSlots.insert(std::make_pair(Object.ID, ObjectIdx));
455 bool MIRParserImpl::initializeConstantPool(
456 MachineConstantPool &ConstantPool, const yaml::MachineFunction &YamlMF,
457 const MachineFunction &MF,
458 DenseMap<unsigned, unsigned> &ConstantPoolSlots) {
459 const auto &M = *MF.getFunction()->getParent();
461 for (const auto &YamlConstant : YamlMF.Constants) {
462 const Constant *Value = dyn_cast_or_null<Constant>(
463 parseConstantValue(YamlConstant.Value.Value, Error, M));
465 return error(Error, YamlConstant.Value.SourceRange);
467 YamlConstant.Alignment
468 ? YamlConstant.Alignment
469 : M.getDataLayout().getPrefTypeAlignment(Value->getType());
470 // TODO: Report an error when the same constant pool value ID is redefined.
471 ConstantPoolSlots.insert(std::make_pair(
472 YamlConstant.ID, ConstantPool.getConstantPoolIndex(Value, Alignment)));
477 bool MIRParserImpl::initializeJumpTableInfo(
478 MachineFunction &MF, const yaml::MachineJumpTable &YamlJTI,
479 PerFunctionMIParsingState &PFS) {
480 MachineJumpTableInfo *JTI = MF.getOrCreateJumpTableInfo(YamlJTI.Kind);
482 for (const auto &Entry : YamlJTI.Entries) {
483 std::vector<MachineBasicBlock *> Blocks;
484 for (const auto &MBBSource : Entry.Blocks) {
485 MachineBasicBlock *MBB = nullptr;
486 if (parseMBBReference(MBB, SM, MF, MBBSource.Value, PFS, IRSlots, Error))
487 return error(Error, MBBSource.SourceRange);
488 Blocks.push_back(MBB);
490 unsigned Index = JTI->createJumpTableIndex(Blocks);
491 // TODO: Report an error when the same jump table slot ID is redefined.
492 PFS.JumpTableSlots.insert(std::make_pair(Entry.ID, Index));
497 SMDiagnostic MIRParserImpl::diagFromMIStringDiag(const SMDiagnostic &Error,
498 SMRange SourceRange) {
499 assert(SourceRange.isValid() && "Invalid source range");
500 SMLoc Loc = SourceRange.Start;
501 bool HasQuote = Loc.getPointer() < SourceRange.End.getPointer() &&
502 *Loc.getPointer() == '\'';
503 // Translate the location of the error from the location in the MI string to
504 // the corresponding location in the MIR file.
505 Loc = Loc.getFromPointer(Loc.getPointer() + Error.getColumnNo() +
508 // TODO: Translate any source ranges as well.
509 return SM.GetMessage(Loc, Error.getKind(), Error.getMessage(), None,
513 SMDiagnostic MIRParserImpl::diagFromLLVMAssemblyDiag(const SMDiagnostic &Error,
514 SMRange SourceRange) {
515 assert(SourceRange.isValid());
517 // Translate the location of the error from the location in the llvm IR string
518 // to the corresponding location in the MIR file.
519 auto LineAndColumn = SM.getLineAndColumn(SourceRange.Start);
520 unsigned Line = LineAndColumn.first + Error.getLineNo() - 1;
521 unsigned Column = Error.getColumnNo();
522 StringRef LineStr = Error.getLineContents();
523 SMLoc Loc = Error.getLoc();
525 // Get the full line and adjust the column number by taking the indentation of
526 // LLVM IR into account.
527 for (line_iterator L(*SM.getMemoryBuffer(SM.getMainFileID()), false), E;
529 if (L.line_number() == Line) {
531 Loc = SMLoc::getFromPointer(LineStr.data());
532 auto Indent = LineStr.find(Error.getLineContents());
533 if (Indent != StringRef::npos)
539 return SMDiagnostic(SM, Loc, Filename, Line, Column, Error.getKind(),
540 Error.getMessage(), LineStr, Error.getRanges(),
544 void MIRParserImpl::initNames2RegClasses(const MachineFunction &MF) {
545 if (!Names2RegClasses.empty())
547 const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
548 for (unsigned I = 0, E = TRI->getNumRegClasses(); I < E; ++I) {
549 const auto *RC = TRI->getRegClass(I);
550 Names2RegClasses.insert(
551 std::make_pair(StringRef(TRI->getRegClassName(RC)).lower(), RC));
555 const TargetRegisterClass *MIRParserImpl::getRegClass(const MachineFunction &MF,
557 initNames2RegClasses(MF);
558 auto RegClassInfo = Names2RegClasses.find(Name);
559 if (RegClassInfo == Names2RegClasses.end())
561 return RegClassInfo->getValue();
564 MIRParser::MIRParser(std::unique_ptr<MIRParserImpl> Impl)
565 : Impl(std::move(Impl)) {}
567 MIRParser::~MIRParser() {}
569 std::unique_ptr<Module> MIRParser::parseLLVMModule() { return Impl->parse(); }
571 bool MIRParser::initializeMachineFunction(MachineFunction &MF) {
572 return Impl->initializeMachineFunction(MF);
575 std::unique_ptr<MIRParser> llvm::createMIRParserFromFile(StringRef Filename,
577 LLVMContext &Context) {
578 auto FileOrErr = MemoryBuffer::getFile(Filename);
579 if (std::error_code EC = FileOrErr.getError()) {
580 Error = SMDiagnostic(Filename, SourceMgr::DK_Error,
581 "Could not open input file: " + EC.message());
584 return createMIRParser(std::move(FileOrErr.get()), Context);
587 std::unique_ptr<MIRParser>
588 llvm::createMIRParser(std::unique_ptr<MemoryBuffer> Contents,
589 LLVMContext &Context) {
590 auto Filename = Contents->getBufferIdentifier();
591 return llvm::make_unique<MIRParser>(
592 llvm::make_unique<MIRParserImpl>(std::move(Contents), Filename, Context));