llvm.org GIT mirror llvm / 502f20b lib / ExecutionEngine / JIT / JIT.cpp
502f20b

Tree @502f20b (Download .tar.gz)

JIT.cpp @502f20braw · history · blame

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
//===-- JIT.cpp - LLVM Just in Time Compiler ------------------------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This tool implements a just-in-time compiler for LLVM, allowing direct
// execution of LLVM bitcode in an efficient manner.
//
//===----------------------------------------------------------------------===//

#include "JIT.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Function.h"
#include "llvm/GlobalVariable.h"
#include "llvm/Instructions.h"
#include "llvm/ModuleProvider.h"
#include "llvm/CodeGen/MachineCodeEmitter.h"
#include "llvm/ExecutionEngine/GenericValue.h"
#include "llvm/Support/MutexGuard.h"
#include "llvm/System/DynamicLibrary.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetJITInfo.h"

#include "llvm/Config/config.h"

using namespace llvm;

#ifdef __APPLE__ 
// Apple gcc defaults to -fuse-cxa-atexit (i.e. calls __cxa_atexit instead
// of atexit). It passes the address of linker generated symbol __dso_handle
// to the function.
// This configuration change happened at version 5330.
# include <AvailabilityMacros.h>
# if defined(MAC_OS_X_VERSION_10_4) && \
     ((MAC_OS_X_VERSION_MIN_REQUIRED > MAC_OS_X_VERSION_10_4) || \
      (MAC_OS_X_VERSION_MIN_REQUIRED == MAC_OS_X_VERSION_10_4 && \
       __APPLE_CC__ >= 5330))
#  ifndef HAVE___DSO_HANDLE
#   define HAVE___DSO_HANDLE 1
#  endif
# endif
#endif

#if HAVE___DSO_HANDLE
extern void *__dso_handle __attribute__ ((__visibility__ ("hidden")));
#endif

namespace {

static struct RegisterJIT {
  RegisterJIT() { JIT::Register(); }
} JITRegistrator;

}

namespace llvm {
  void LinkInJIT() {
  }
}

#if defined (__GNUC__)
extern "C" void __register_frame(void*);
#endif

/// createJIT - This is the factory method for creating a JIT for the current
/// machine, it does not fall back to the interpreter.  This takes ownership
/// of the module provider.
ExecutionEngine *ExecutionEngine::createJIT(ModuleProvider *MP,
                                            std::string *ErrorStr,
                                            JITMemoryManager *JMM,
                                            bool Fast) {
  ExecutionEngine *EE = JIT::createJIT(MP, ErrorStr, JMM, Fast);
  if (!EE) return 0;
  
  // Register routine for informing unwinding runtime about new EH frames
#if defined(__GNUC__)
  EE->InstallExceptionTableRegister(__register_frame);
#endif

  // Make sure we can resolve symbols in the program as well. The zero arg
  // to the function tells DynamicLibrary to load the program, not a library.
  sys::DynamicLibrary::LoadLibraryPermanently(0, ErrorStr);
  return EE;
}

JIT::JIT(ModuleProvider *MP, TargetMachine &tm, TargetJITInfo &tji,
         JITMemoryManager *JMM, bool Fast)
  : ExecutionEngine(MP), TM(tm), TJI(tji) {
  setTargetData(TM.getTargetData());

  jitstate = new JITState(MP);

  // Initialize MCE
  MCE = createEmitter(*this, JMM);

  // Add target data
  MutexGuard locked(lock);
  FunctionPassManager &PM = jitstate->getPM(locked);
  PM.add(new TargetData(*TM.getTargetData()));

  // Turn the machine code intermediate representation into bytes in memory that
  // may be executed.
  if (TM.addPassesToEmitMachineCode(PM, *MCE, Fast)) {
    cerr << "Target does not support machine code emission!\n";
    abort();
  }
  
  // Initialize passes.
  PM.doInitialization();
}

JIT::~JIT() {
  delete jitstate;
  delete MCE;
  delete &TM;
}

/// addModuleProvider - Add a new ModuleProvider to the JIT.  If we previously
/// removed the last ModuleProvider, we need re-initialize jitstate with a valid
/// ModuleProvider.
void JIT::addModuleProvider(ModuleProvider *MP) {
  MutexGuard locked(lock);

  if (Modules.empty()) {
    assert(!jitstate && "jitstate should be NULL if Modules vector is empty!");

    jitstate = new JITState(MP);

    FunctionPassManager &PM = jitstate->getPM(locked);
    PM.add(new TargetData(*TM.getTargetData()));

    // Turn the machine code intermediate representation into bytes in memory
    // that may be executed.
    if (TM.addPassesToEmitMachineCode(PM, *MCE, false /*fast*/)) {
      cerr << "Target does not support machine code emission!\n";
      abort();
    }
    
    // Initialize passes.
    PM.doInitialization();
  }
  
  ExecutionEngine::addModuleProvider(MP);
}

/// removeModuleProvider - If we are removing the last ModuleProvider, 
/// invalidate the jitstate since the PassManager it contains references a
/// released ModuleProvider.
Module *JIT::removeModuleProvider(ModuleProvider *MP, std::string *E) {
  Module *result = ExecutionEngine::removeModuleProvider(MP, E);
  
  MutexGuard locked(lock);
  if (Modules.empty()) {
    delete jitstate;
    jitstate = 0;
  }
  
  return result;
}

/// run - Start execution with the specified function and arguments.
///
GenericValue JIT::runFunction(Function *F,
                              const std::vector<GenericValue> &ArgValues) {
  assert(F && "Function *F was null at entry to run()");

  void *FPtr = getPointerToFunction(F);
  assert(FPtr && "Pointer to fn's code was null after getPointerToFunction");
  const FunctionType *FTy = F->getFunctionType();
  const Type *RetTy = FTy->getReturnType();

  assert((FTy->getNumParams() <= ArgValues.size() || FTy->isVarArg()) &&
         "Too many arguments passed into function!");
  assert(FTy->getNumParams() == ArgValues.size() &&
         "This doesn't support passing arguments through varargs (yet)!");

  // Handle some common cases first.  These cases correspond to common `main'
  // prototypes.
  if (RetTy == Type::Int32Ty || RetTy == Type::VoidTy) {
    switch (ArgValues.size()) {
    case 3:
      if (FTy->getParamType(0) == Type::Int32Ty &&
          isa<PointerType>(FTy->getParamType(1)) &&
          isa<PointerType>(FTy->getParamType(2))) {
        int (*PF)(int, char **, const char **) =
          (int(*)(int, char **, const char **))(intptr_t)FPtr;

        // Call the function.
        GenericValue rv;
        rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue(), 
                                 (char **)GVTOP(ArgValues[1]),
                                 (const char **)GVTOP(ArgValues[2])));
        return rv;
      }
      break;
    case 2:
      if (FTy->getParamType(0) == Type::Int32Ty &&
          isa<PointerType>(FTy->getParamType(1))) {
        int (*PF)(int, char **) = (int(*)(int, char **))(intptr_t)FPtr;

        // Call the function.
        GenericValue rv;
        rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue(), 
                                 (char **)GVTOP(ArgValues[1])));
        return rv;
      }
      break;
    case 1:
      if (FTy->getNumParams() == 1 &&
          FTy->getParamType(0) == Type::Int32Ty) {
        GenericValue rv;
        int (*PF)(int) = (int(*)(int))(intptr_t)FPtr;
        rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue()));
        return rv;
      }
      break;
    }
  }

  // Handle cases where no arguments are passed first.
  if (ArgValues.empty()) {
    GenericValue rv;
    switch (RetTy->getTypeID()) {
    default: assert(0 && "Unknown return type for function call!");
    case Type::IntegerTyID: {
      unsigned BitWidth = cast<IntegerType>(RetTy)->getBitWidth();
      if (BitWidth == 1)
        rv.IntVal = APInt(BitWidth, ((bool(*)())(intptr_t)FPtr)());
      else if (BitWidth <= 8)
        rv.IntVal = APInt(BitWidth, ((char(*)())(intptr_t)FPtr)());
      else if (BitWidth <= 16)
        rv.IntVal = APInt(BitWidth, ((short(*)())(intptr_t)FPtr)());
      else if (BitWidth <= 32)
        rv.IntVal = APInt(BitWidth, ((int(*)())(intptr_t)FPtr)());
      else if (BitWidth <= 64)
        rv.IntVal = APInt(BitWidth, ((int64_t(*)())(intptr_t)FPtr)());
      else 
        assert(0 && "Integer types > 64 bits not supported");
      return rv;
    }
    case Type::VoidTyID:
      rv.IntVal = APInt(32, ((int(*)())(intptr_t)FPtr)());
      return rv;
    case Type::FloatTyID:
      rv.FloatVal = ((float(*)())(intptr_t)FPtr)();
      return rv;
    case Type::DoubleTyID:
      rv.DoubleVal = ((double(*)())(intptr_t)FPtr)();
      return rv;
    case Type::X86_FP80TyID:
    case Type::FP128TyID:
    case Type::PPC_FP128TyID:
      assert(0 && "long double not supported yet");
      return rv;
    case Type::PointerTyID:
      return PTOGV(((void*(*)())(intptr_t)FPtr)());
    }
  }

  // Okay, this is not one of our quick and easy cases.  Because we don't have a
  // full FFI, we have to codegen a nullary stub function that just calls the
  // function we are interested in, passing in constants for all of the
  // arguments.  Make this function and return.

  // First, create the function.
  FunctionType *STy=FunctionType::get(RetTy, std::vector<const Type*>(), false);
  Function *Stub = Function::Create(STy, Function::InternalLinkage, "",
                                    F->getParent());

  // Insert a basic block.
  BasicBlock *StubBB = BasicBlock::Create("", Stub);

  // Convert all of the GenericValue arguments over to constants.  Note that we
  // currently don't support varargs.
  SmallVector<Value*, 8> Args;
  for (unsigned i = 0, e = ArgValues.size(); i != e; ++i) {
    Constant *C = 0;
    const Type *ArgTy = FTy->getParamType(i);
    const GenericValue &AV = ArgValues[i];
    switch (ArgTy->getTypeID()) {
    default: assert(0 && "Unknown argument type for function call!");
    case Type::IntegerTyID:
        C = ConstantInt::get(AV.IntVal);
        break;
    case Type::FloatTyID:
        C = ConstantFP::get(APFloat(AV.FloatVal));
        break;
    case Type::DoubleTyID:
        C = ConstantFP::get(APFloat(AV.DoubleVal));
        break;
    case Type::PPC_FP128TyID:
    case Type::X86_FP80TyID:
    case Type::FP128TyID:
        C = ConstantFP::get(APFloat(AV.IntVal));
        break;
    case Type::PointerTyID:
      void *ArgPtr = GVTOP(AV);
      if (sizeof(void*) == 4)
        C = ConstantInt::get(Type::Int32Ty, (int)(intptr_t)ArgPtr);
      else
        C = ConstantInt::get(Type::Int64Ty, (intptr_t)ArgPtr);
      C = ConstantExpr::getIntToPtr(C, ArgTy);  // Cast the integer to pointer
      break;
    }
    Args.push_back(C);
  }

  CallInst *TheCall = CallInst::Create(F, Args.begin(), Args.end(),
                                       "", StubBB);
  TheCall->setTailCall();
  if (TheCall->getType() != Type::VoidTy)
    ReturnInst::Create(TheCall, StubBB);    // Return result of the call.
  else
    ReturnInst::Create(StubBB);             // Just return void.

  // Finally, return the value returned by our nullary stub function.
  return runFunction(Stub, std::vector<GenericValue>());
}

/// runJITOnFunction - Run the FunctionPassManager full of
/// just-in-time compilation passes on F, hopefully filling in
/// GlobalAddress[F] with the address of F's machine code.
///
void JIT::runJITOnFunction(Function *F) {
  static bool isAlreadyCodeGenerating = false;

  MutexGuard locked(lock);
  assert(!isAlreadyCodeGenerating && "Error: Recursive compilation detected!");

  // JIT the function
  isAlreadyCodeGenerating = true;
  jitstate->getPM(locked).run(*F);
  isAlreadyCodeGenerating = false;

  // If the function referred to a global variable that had not yet been
  // emitted, it allocates memory for the global, but doesn't emit it yet.  Emit
  // all of these globals now.
  while (!jitstate->getPendingGlobals(locked).empty()) {
    const GlobalVariable *GV = jitstate->getPendingGlobals(locked).back();
    jitstate->getPendingGlobals(locked).pop_back();
    EmitGlobalVariable(GV);
  }
}

/// getPointerToFunction - This method is used to get the address of the
/// specified function, compiling it if neccesary.
///
void *JIT::getPointerToFunction(Function *F) {

  if (void *Addr = getPointerToGlobalIfAvailable(F))
    return Addr;   // Check if function already code gen'd

  // Make sure we read in the function if it exists in this Module.
  if (F->hasNotBeenReadFromBitcode()) {
    // Determine the module provider this function is provided by.
    Module *M = F->getParent();
    ModuleProvider *MP = 0;
    for (unsigned i = 0, e = Modules.size(); i != e; ++i) {
      if (Modules[i]->getModule() == M) {
        MP = Modules[i];
        break;
      }
    }
    assert(MP && "Function isn't in a module we know about!");
    
    std::string ErrorMsg;
    if (MP->materializeFunction(F, &ErrorMsg)) {
      cerr << "Error reading function '" << F->getName()
           << "' from bitcode file: " << ErrorMsg << "\n";
      abort();
    }
  }
  
  if (void *Addr = getPointerToGlobalIfAvailable(F)) {
    return Addr;
  }

  MutexGuard locked(lock);
  
  if (F->isDeclaration()) {
    void *Addr = getPointerToNamedFunction(F->getName());
    addGlobalMapping(F, Addr);
    return Addr;
  }

  runJITOnFunction(F);

  void *Addr = getPointerToGlobalIfAvailable(F);
  assert(Addr && "Code generation didn't add function to GlobalAddress table!");
  return Addr;
}

/// getOrEmitGlobalVariable - Return the address of the specified global
/// variable, possibly emitting it to memory if needed.  This is used by the
/// Emitter.
void *JIT::getOrEmitGlobalVariable(const GlobalVariable *GV) {
  MutexGuard locked(lock);

  void *Ptr = getPointerToGlobalIfAvailable(GV);
  if (Ptr) return Ptr;

  // If the global is external, just remember the address.
  if (GV->isDeclaration()) {
#if HAVE___DSO_HANDLE
    if (GV->getName() == "__dso_handle")
      return (void*)&__dso_handle;
#endif
    Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(GV->getName().c_str());
    if (Ptr == 0) {
      cerr << "Could not resolve external global address: "
           << GV->getName() << "\n";
      abort();
    addGlobalMapping(GV, Ptr);
    }
  } else {
    // If the global hasn't been emitted to memory yet, allocate space and
    // emit it into memory.  It goes in the same array as the generated
    // code, jump tables, etc.
    const Type *GlobalType = GV->getType()->getElementType();
    size_t S = getTargetData()->getABITypeSize(GlobalType);
    size_t A = getTargetData()->getPreferredAlignment(GV);
    Ptr = MCE->allocateSpace(S, A);
    addGlobalMapping(GV, Ptr);
    EmitGlobalVariable(GV);
  }
  return Ptr;
}

/// recompileAndRelinkFunction - This method is used to force a function
/// which has already been compiled, to be compiled again, possibly
/// after it has been modified. Then the entry to the old copy is overwritten
/// with a branch to the new copy. If there was no old copy, this acts
/// just like JIT::getPointerToFunction().
///
void *JIT::recompileAndRelinkFunction(Function *F) {
  void *OldAddr = getPointerToGlobalIfAvailable(F);

  // If it's not already compiled there is no reason to patch it up.
  if (OldAddr == 0) { return getPointerToFunction(F); }

  // Delete the old function mapping.
  addGlobalMapping(F, 0);

  // Recodegen the function
  runJITOnFunction(F);

  // Update state, forward the old function to the new function.
  void *Addr = getPointerToGlobalIfAvailable(F);
  assert(Addr && "Code generation didn't add function to GlobalAddress table!");
  TJI.replaceMachineCodeForFunction(OldAddr, Addr);
  return Addr;
}