aboutsummaryrefslogtreecommitdiffstats
path: root/lib/CodeGen/Analysis.cpp
blob: 5162ad762e73112e0f6143d738d20a684a31b420 (plain)
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
//===-- Analysis.cpp - CodeGen LLVM IR Analysis Utilities -----------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines several CodeGen-specific LLVM IR analysis utilties.
//
//===----------------------------------------------------------------------===//

#include "llvm/CodeGen/Analysis.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Function.h"
#include "llvm/Instructions.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/LLVMContext.h"
#include "llvm/Module.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/DataLayout.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
using namespace llvm;

/// ComputeLinearIndex - Given an LLVM IR aggregate type and a sequence
/// of insertvalue or extractvalue indices that identify a member, return
/// the linearized index of the start of the member.
///
unsigned llvm::ComputeLinearIndex(Type *Ty,
                                  const unsigned *Indices,
                                  const unsigned *IndicesEnd,
                                  unsigned CurIndex) {
  // Base case: We're done.
  if (Indices && Indices == IndicesEnd)
    return CurIndex;

  // Given a struct type, recursively traverse the elements.
  if (StructType *STy = dyn_cast<StructType>(Ty)) {
    for (StructType::element_iterator EB = STy->element_begin(),
                                      EI = EB,
                                      EE = STy->element_end();
        EI != EE; ++EI) {
      if (Indices && *Indices == unsigned(EI - EB))
        return ComputeLinearIndex(*EI, Indices+1, IndicesEnd, CurIndex);
      CurIndex = ComputeLinearIndex(*EI, 0, 0, CurIndex);
    }
    return CurIndex;
  }
  // Given an array type, recursively traverse the elements.
  else if (ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
    Type *EltTy = ATy->getElementType();
    for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i) {
      if (Indices && *Indices == i)
        return ComputeLinearIndex(EltTy, Indices+1, IndicesEnd, CurIndex);
      CurIndex = ComputeLinearIndex(EltTy, 0, 0, CurIndex);
    }
    return CurIndex;
  }
  // We haven't found the type we're looking for, so keep searching.
  return CurIndex + 1;
}

/// ComputeValueVTs - Given an LLVM IR type, compute a sequence of
/// EVTs that represent all the individual underlying
/// non-aggregate types that comprise it.
///
/// If Offsets is non-null, it points to a vector to be filled in
/// with the in-memory offsets of each of the individual values.
///
void llvm::ComputeValueVTs(const TargetLowering &TLI, Type *Ty,
                           SmallVectorImpl<EVT> &ValueVTs,
                           SmallVectorImpl<uint64_t> *Offsets,
                           uint64_t StartingOffset) {
  // Given a struct type, recursively traverse the elements.
  if (StructType *STy = dyn_cast<StructType>(Ty)) {
    const StructLayout *SL = TLI.getDataLayout()->getStructLayout(STy);
    for (StructType::element_iterator EB = STy->element_begin(),
                                      EI = EB,
                                      EE = STy->element_end();
         EI != EE; ++EI)
      ComputeValueVTs(TLI, *EI, ValueVTs, Offsets,
                      StartingOffset + SL->getElementOffset(EI - EB));
    return;
  }
  // Given an array type, recursively traverse the elements.
  if (ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
    Type *EltTy = ATy->getElementType();
    uint64_t EltSize = TLI.getDataLayout()->getTypeAllocSize(EltTy);
    for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
      ComputeValueVTs(TLI, EltTy, ValueVTs, Offsets,
                      StartingOffset + i * EltSize);
    return;
  }
  // Interpret void as zero return values.
  if (Ty->isVoidTy())
    return;
  // Base case: we can get an EVT for this LLVM IR type.
  ValueVTs.push_back(TLI.getValueType(Ty));
  if (Offsets)
    Offsets->push_back(StartingOffset);
}

/// ExtractTypeInfo - Returns the type info, possibly bitcast, encoded in V.
GlobalVariable *llvm::ExtractTypeInfo(Value *V) {
  V = V->stripPointerCasts();
  GlobalVariable *GV = dyn_cast<GlobalVariable>(V);

  if (GV && GV->getName() == "llvm.eh.catch.all.value") {
    assert(GV->hasInitializer() &&
           "The EH catch-all value must have an initializer");
    Value *Init = GV->getInitializer();
    GV = dyn_cast<GlobalVariable>(Init);
    if (!GV) V = cast<ConstantPointerNull>(Init);
  }

  assert((GV || isa<ConstantPointerNull>(V)) &&
         "TypeInfo must be a global variable or NULL");
  return GV;
}

/// hasInlineAsmMemConstraint - Return true if the inline asm instruction being
/// processed uses a memory 'm' constraint.
bool
llvm::hasInlineAsmMemConstraint(InlineAsm::ConstraintInfoVector &CInfos,
                                const TargetLowering &TLI) {
  for (unsigned i = 0, e = CInfos.size(); i != e; ++i) {
    InlineAsm::ConstraintInfo &CI = CInfos[i];
    for (unsigned j = 0, ee = CI.Codes.size(); j != ee; ++j) {
      TargetLowering::ConstraintType CType = TLI.getConstraintType(CI.Codes[j]);
      if (CType == TargetLowering::C_Memory)
        return true;
    }

    // Indirect operand accesses access memory.
    if (CI.isIndirect)
      return true;
  }

  return false;
}

/// getFCmpCondCode - Return the ISD condition code corresponding to
/// the given LLVM IR floating-point condition code.  This includes
/// consideration of global floating-point math flags.
///
ISD::CondCode llvm::getFCmpCondCode(FCmpInst::Predicate Pred) {
  switch (Pred) {
  case FCmpInst::FCMP_FALSE: return ISD::SETFALSE;
  case FCmpInst::FCMP_OEQ:   return ISD::SETOEQ;
  case FCmpInst::FCMP_OGT:   return ISD::SETOGT;
  case FCmpInst::FCMP_OGE:   return ISD::SETOGE;
  case FCmpInst::FCMP_OLT:   return ISD::SETOLT;
  case FCmpInst::FCMP_OLE:   return ISD::SETOLE;
  case FCmpInst::FCMP_ONE:   return ISD::SETONE;
  case FCmpInst::FCMP_ORD:   return ISD::SETO;
  case FCmpInst::FCMP_UNO:   return ISD::SETUO;
  case FCmpInst::FCMP_UEQ:   return ISD::SETUEQ;
  case FCmpInst::FCMP_UGT:   return ISD::SETUGT;
  case FCmpInst::FCMP_UGE:   return ISD::SETUGE;
  case FCmpInst::FCMP_ULT:   return ISD::SETULT;
  case FCmpInst::FCMP_ULE:   return ISD::SETULE;
  case FCmpInst::FCMP_UNE:   return ISD::SETUNE;
  case FCmpInst::FCMP_TRUE:  return ISD::SETTRUE;
  default: llvm_unreachable("Invalid FCmp predicate opcode!");
  }
}

ISD::CondCode llvm::getFCmpCodeWithoutNaN(ISD::CondCode CC) {
  switch (CC) {
    case ISD::SETOEQ: case ISD::SETUEQ: return ISD::SETEQ;
    case ISD::SETONE: case ISD::SETUNE: return ISD::SETNE;
    case ISD::SETOLT: case ISD::SETULT: return ISD::SETLT;
    case ISD::SETOLE: case ISD::SETULE: return ISD::SETLE;
    case ISD::SETOGT: case ISD::SETUGT: return ISD::SETGT;
    case ISD::SETOGE: case ISD::SETUGE: return ISD::SETGE;
    default: return CC;
  }
}

/// getICmpCondCode - Return the ISD condition code corresponding to
/// the given LLVM IR integer condition code.
///
ISD::CondCode llvm::getICmpCondCode(ICmpInst::Predicate Pred) {
  switch (Pred) {
  case ICmpInst::ICMP_EQ:  return ISD::SETEQ;
  case ICmpInst::ICMP_NE:  return ISD::SETNE;
  case ICmpInst::ICMP_SLE: return ISD::SETLE;
  case ICmpInst::ICMP_ULE: return ISD::SETULE;
  case ICmpInst::ICMP_SGE: return ISD::SETGE;
  case ICmpInst::ICMP_UGE: return ISD::SETUGE;
  case ICmpInst::ICMP_SLT: return ISD::SETLT;
  case ICmpInst::ICMP_ULT: return ISD::SETULT;
  case ICmpInst::ICMP_SGT: return ISD::SETGT;
  case ICmpInst::ICMP_UGT: return ISD::SETUGT;
  default:
    llvm_unreachable("Invalid ICmp predicate opcode!");
  }
}


/// getNoopInput - If V is a noop (i.e., lowers to no machine code), look
/// through it (and any transitive noop operands to it) and return its input
/// value.  This is used to determine if a tail call can be formed.
///
static const Value *getNoopInput(const Value *V, const TargetLowering &TLI) {
  // If V is not an instruction, it can't be looked through.
  const Instruction *I = dyn_cast<Instruction>(V);
  if (I == 0 || !I->hasOneUse() || I->getNumOperands() == 0) return V;
  
  Value *Op = I->getOperand(0);

  // Look through truly no-op truncates.
  if (isa<TruncInst>(I) &&
      TLI.isTruncateFree(I->getOperand(0)->getType(), I->getType()))
    return getNoopInput(I->getOperand(0), TLI);
  
  // Look through truly no-op bitcasts.
  if (isa<BitCastInst>(I)) {
    // No type change at all.
    if (Op->getType() == I->getType())
      return getNoopInput(Op, TLI);

    // Pointer to pointer cast.
    if (Op->getType()->isPointerTy() && I->getType()->isPointerTy())
      return getNoopInput(Op, TLI);
    
    if (isa<VectorType>(Op->getType()) && isa<VectorType>(I->getType()) &&
        TLI.isTypeLegal(EVT::getEVT(Op->getType())) &&
        TLI.isTypeLegal(EVT::getEVT(I->getType())))
      return getNoopInput(Op, TLI);
  }
  
  // Look through inttoptr.
  if (isa<IntToPtrInst>(I) && !isa<VectorType>(I->getType())) {
    // Make sure this isn't a truncating or extending cast.  We could support
    // this eventually, but don't bother for now.
    if (TLI.getPointerTy().getSizeInBits() == 
          cast<IntegerType>(Op->getType())->getBitWidth())
      return getNoopInput(Op, TLI);
  }

  // Look through ptrtoint.
  if (isa<PtrToIntInst>(I) && !isa<VectorType>(I->getType())) {
    // Make sure this isn't a truncating or extending cast.  We could support
    // this eventually, but don't bother for now.
    if (TLI.getPointerTy().getSizeInBits() == 
        cast<IntegerType>(I->getType())->getBitWidth())
      return getNoopInput(Op, TLI);
  }


  // Otherwise it's not something we can look through.
  return V;
}


/// Test if the given instruction is in a position to be optimized
/// with a tail-call. This roughly means that it's in a block with
/// a return and there's nothing that needs to be scheduled
/// between it and the return.
///
/// This function only tests target-independent requirements.
bool llvm::isInTailCallPosition(ImmutableCallSite CS, Attributes CalleeRetAttr,
                                const TargetLowering &TLI) {
  const Instruction *I = CS.getInstruction();
  const BasicBlock *ExitBB = I->getParent();
  const TerminatorInst *Term = ExitBB->getTerminator();
  const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);

  // The block must end in a return statement or unreachable.
  //
  // FIXME: Decline tailcall if it's not guaranteed and if the block ends in
  // an unreachable, for now. The way tailcall optimization is currently
  // implemented means it will add an epilogue followed by a jump. That is
  // not profitable. Also, if the callee is a special function (e.g.
  // longjmp on x86), it can end up causing miscompilation that has not
  // been fully understood.
  if (!Ret &&
      (!TLI.getTargetMachine().Options.GuaranteedTailCallOpt ||
       !isa<UnreachableInst>(Term)))
    return false;

  // If I will have a chain, make sure no other instruction that will have a
  // chain interposes between I and the return.
  if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
      !isSafeToSpeculativelyExecute(I))
    for (BasicBlock::const_iterator BBI = prior(prior(ExitBB->end())); ;
         --BBI) {
      if (&*BBI == I)
        break;
      // Debug info intrinsics do not get in the way of tail call optimization.
      if (isa<DbgInfoIntrinsic>(BBI))
        continue;
      if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
          !isSafeToSpeculativelyExecute(BBI))
        return false;
    }

  // If the block ends with a void return or unreachable, it doesn't matter
  // what the call's return type is.
  if (!Ret || Ret->getNumOperands() == 0) return true;

  // If the return value is undef, it doesn't matter what the call's
  // return type is.
  if (isa<UndefValue>(Ret->getOperand(0))) return true;

  // Conservatively require the attributes of the call to match those of
  // the return. Ignore noalias because it doesn't affect the call sequence.
  const Function *F = ExitBB->getParent();
  Attributes CallerRetAttr = F->getAttributes().getRetAttributes();
  if (AttrBuilder(CalleeRetAttr).removeAttribute(Attributes::NoAlias) !=
      AttrBuilder(CallerRetAttr).removeAttribute(Attributes::NoAlias))
    return false;

  // It's not safe to eliminate the sign / zero extension of the return value.
  if (CallerRetAttr.hasAttribute(Attributes::ZExt) ||
      CallerRetAttr.hasAttribute(Attributes::SExt))
    return false;

  // Otherwise, make sure the unmodified return value of I is the return value.
  // We handle two cases: multiple return values + scalars.
  Value *RetVal = Ret->getOperand(0);
  if (!isa<InsertValueInst>(RetVal) || !isa<StructType>(RetVal->getType()))
    // Handle scalars first.
    return getNoopInput(Ret->getOperand(0), TLI) == I;
  
  // If this is an aggregate return, look through the insert/extract values and
  // see if each is transparent.
  for (unsigned i = 0, e =cast<StructType>(RetVal->getType())->getNumElements();
       i != e; ++i) {
    const Value *InScalar = FindInsertedValue(RetVal, i);
    if (InScalar == 0) return false;
    InScalar = getNoopInput(InScalar, TLI);
    
    // If the scalar value being inserted is an extractvalue of the right index
    // from the call, then everything is good.
    const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(InScalar);
    if (EVI == 0 || EVI->getOperand(0) != I || EVI->getNumIndices() != 1 ||
        EVI->getIndices()[0] != i)
      return false;
  }
  
  return true;
}

bool llvm::isInTailCallPosition(SelectionDAG &DAG, SDNode *Node,
                                SDValue &Chain, const TargetLowering &TLI) {
  const Function *F = DAG.getMachineFunction().getFunction();

  // Conservatively require the attributes of the call to match those of
  // the return. Ignore noalias because it doesn't affect the call sequence.
  Attributes CallerRetAttr = F->getAttributes().getRetAttributes();
  if (AttrBuilder(CallerRetAttr)
      .removeAttribute(Attributes::NoAlias).hasAttributes())
    return false;

  // It's not safe to eliminate the sign / zero extension of the return value.
  if (CallerRetAttr.hasAttribute(Attributes::ZExt) ||
      CallerRetAttr.hasAttribute(Attributes::SExt))
    return false;

  // Check if the only use is a function return node.
  return TLI.isUsedByReturnOnly(Node, Chain);
}