LLVM  10.0.0svn
SimplifyLibCalls.cpp
Go to the documentation of this file.
1 //===------ SimplifyLibCalls.cpp - Library calls simplifier ---------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file implements the library calls simplifier. It does not implement
10 // any pass, but can't be used by other passes to do simplifications.
11 //
12 //===----------------------------------------------------------------------===//
13 
15 #include "llvm/ADT/APSInt.h"
16 #include "llvm/ADT/SmallString.h"
17 #include "llvm/ADT/StringMap.h"
18 #include "llvm/ADT/Triple.h"
27 #include "llvm/Analysis/Loads.h"
28 #include "llvm/IR/DataLayout.h"
29 #include "llvm/IR/Function.h"
30 #include "llvm/IR/IRBuilder.h"
31 #include "llvm/IR/IntrinsicInst.h"
32 #include "llvm/IR/Intrinsics.h"
33 #include "llvm/IR/LLVMContext.h"
34 #include "llvm/IR/Module.h"
35 #include "llvm/IR/PatternMatch.h"
37 #include "llvm/Support/KnownBits.h"
40 
41 using namespace llvm;
42 using namespace PatternMatch;
43 
44 static cl::opt<bool>
45  EnableUnsafeFPShrink("enable-double-float-shrink", cl::Hidden,
46  cl::init(false),
47  cl::desc("Enable unsafe double to float "
48  "shrinking for math lib calls"));
49 
50 
51 //===----------------------------------------------------------------------===//
52 // Helper Functions
53 //===----------------------------------------------------------------------===//
54 
55 static bool ignoreCallingConv(LibFunc Func) {
56  return Func == LibFunc_abs || Func == LibFunc_labs ||
57  Func == LibFunc_llabs || Func == LibFunc_strlen;
58 }
59 
61  switch(CI->getCallingConv()) {
62  default:
63  return false;
65  return true;
69 
70  // The iOS ABI diverges from the standard in some cases, so for now don't
71  // try to simplify those calls.
72  if (Triple(CI->getModule()->getTargetTriple()).isiOS())
73  return false;
74 
75  auto *FuncTy = CI->getFunctionType();
76 
77  if (!FuncTy->getReturnType()->isPointerTy() &&
78  !FuncTy->getReturnType()->isIntegerTy() &&
79  !FuncTy->getReturnType()->isVoidTy())
80  return false;
81 
82  for (auto Param : FuncTy->params()) {
83  if (!Param->isPointerTy() && !Param->isIntegerTy())
84  return false;
85  }
86  return true;
87  }
88  }
89  return false;
90 }
91 
92 /// Return true if it is only used in equality comparisons with With.
93 static bool isOnlyUsedInEqualityComparison(Value *V, Value *With) {
94  for (User *U : V->users()) {
95  if (ICmpInst *IC = dyn_cast<ICmpInst>(U))
96  if (IC->isEquality() && IC->getOperand(1) == With)
97  continue;
98  // Unknown instruction.
99  return false;
100  }
101  return true;
102 }
103 
104 static bool callHasFloatingPointArgument(const CallInst *CI) {
105  return any_of(CI->operands(), [](const Use &OI) {
106  return OI->getType()->isFloatingPointTy();
107  });
108 }
109 
110 static bool callHasFP128Argument(const CallInst *CI) {
111  return any_of(CI->operands(), [](const Use &OI) {
112  return OI->getType()->isFP128Ty();
113  });
114 }
115 
116 static Value *convertStrToNumber(CallInst *CI, StringRef &Str, int64_t Base) {
117  if (Base < 2 || Base > 36)
118  // handle special zero base
119  if (Base != 0)
120  return nullptr;
121 
122  char *End;
123  std::string nptr = Str.str();
124  errno = 0;
125  long long int Result = strtoll(nptr.c_str(), &End, Base);
126  if (errno)
127  return nullptr;
128 
129  // if we assume all possible target locales are ASCII supersets,
130  // then if strtoll successfully parses a number on the host,
131  // it will also successfully parse the same way on the target
132  if (*End != '\0')
133  return nullptr;
134 
135  if (!isIntN(CI->getType()->getPrimitiveSizeInBits(), Result))
136  return nullptr;
137 
138  return ConstantInt::get(CI->getType(), Result);
139 }
140 
142  const TargetLibraryInfo *TLI) {
143  CallInst *FOpen = dyn_cast<CallInst>(File);
144  if (!FOpen)
145  return false;
146 
147  Function *InnerCallee = FOpen->getCalledFunction();
148  if (!InnerCallee)
149  return false;
150 
151  LibFunc Func;
152  if (!TLI->getLibFunc(*InnerCallee, Func) || !TLI->has(Func) ||
153  Func != LibFunc_fopen)
154  return false;
155 
157  if (PointerMayBeCaptured(File, true, true))
158  return false;
159 
160  return true;
161 }
162 
164  for (User *U : V->users()) {
165  if (ICmpInst *IC = dyn_cast<ICmpInst>(U))
166  if (Constant *C = dyn_cast<Constant>(IC->getOperand(1)))
167  if (C->isNullValue())
168  continue;
169  // Unknown instruction.
170  return false;
171  }
172  return true;
173 }
174 
175 static bool canTransformToMemCmp(CallInst *CI, Value *Str, uint64_t Len,
176  const DataLayout &DL) {
178  return false;
179 
180  if (!isDereferenceableAndAlignedPointer(Str, 1, APInt(64, Len), DL))
181  return false;
182 
183  if (CI->getFunction()->hasFnAttribute(Attribute::SanitizeMemory))
184  return false;
185 
186  return true;
187 }
188 
190  ArrayRef<unsigned> ArgNos,
191  uint64_t DereferenceableBytes) {
192  const Function *F = CI->getCaller();
193  if (!F)
194  return;
195  for (unsigned ArgNo : ArgNos) {
196  uint64_t DerefBytes = DereferenceableBytes;
197  unsigned AS = CI->getArgOperand(ArgNo)->getType()->getPointerAddressSpace();
198  if (!llvm::NullPointerIsDefined(F, AS) ||
199  CI->paramHasAttr(ArgNo, Attribute::NonNull))
200  DerefBytes = std::max(CI->getDereferenceableOrNullBytes(
202  DereferenceableBytes);
203 
205  DerefBytes) {
206  CI->removeParamAttr(ArgNo, Attribute::Dereferenceable);
207  if (!llvm::NullPointerIsDefined(F, AS) ||
208  CI->paramHasAttr(ArgNo, Attribute::NonNull))
209  CI->removeParamAttr(ArgNo, Attribute::DereferenceableOrNull);
211  CI->getContext(), DerefBytes));
212  }
213  }
214 }
215 
217  ArrayRef<unsigned> ArgNos) {
218  Function *F = CI->getCaller();
219  if (!F)
220  return;
221 
222  for (unsigned ArgNo : ArgNos) {
223  if (CI->paramHasAttr(ArgNo, Attribute::NonNull))
224  continue;
225  unsigned AS = CI->getArgOperand(ArgNo)->getType()->getPointerAddressSpace();
226  if (llvm::NullPointerIsDefined(F, AS))
227  continue;
228 
229  CI->addParamAttr(ArgNo, Attribute::NonNull);
230  annotateDereferenceableBytes(CI, ArgNo, 1);
231  }
232 }
233 
235  Value *Size, const DataLayout &DL) {
236  if (ConstantInt *LenC = dyn_cast<ConstantInt>(Size)) {
237  annotateNonNullBasedOnAccess(CI, ArgNos);
238  annotateDereferenceableBytes(CI, ArgNos, LenC->getZExtValue());
239  } else if (isKnownNonZero(Size, DL)) {
240  annotateNonNullBasedOnAccess(CI, ArgNos);
241  const APInt *X, *Y;
242  uint64_t DerefMin = 1;
243  if (match(Size, m_Select(m_Value(), m_APInt(X), m_APInt(Y)))) {
244  DerefMin = std::min(X->getZExtValue(), Y->getZExtValue());
245  annotateDereferenceableBytes(CI, ArgNos, DerefMin);
246  }
247  }
248 }
249 
250 //===----------------------------------------------------------------------===//
251 // String and Memory Library Call Optimizations
252 //===----------------------------------------------------------------------===//
253 
254 Value *LibCallSimplifier::optimizeStrCat(CallInst *CI, IRBuilder<> &B) {
255  // Extract some information from the instruction
256  Value *Dst = CI->getArgOperand(0);
257  Value *Src = CI->getArgOperand(1);
258  annotateNonNullBasedOnAccess(CI, {0, 1});
259 
260  // See if we can get the length of the input string.
261  uint64_t Len = GetStringLength(Src);
262  if (Len)
263  annotateDereferenceableBytes(CI, 1, Len);
264  else
265  return nullptr;
266  --Len; // Unbias length.
267 
268  // Handle the simple, do-nothing case: strcat(x, "") -> x
269  if (Len == 0)
270  return Dst;
271 
272  return emitStrLenMemCpy(Src, Dst, Len, B);
273 }
274 
275 Value *LibCallSimplifier::emitStrLenMemCpy(Value *Src, Value *Dst, uint64_t Len,
276  IRBuilder<> &B) {
277  // We need to find the end of the destination string. That's where the
278  // memory is to be moved to. We just generate a call to strlen.
279  Value *DstLen = emitStrLen(Dst, B, DL, TLI);
280  if (!DstLen)
281  return nullptr;
282 
283  // Now that we have the destination's length, we must index into the
284  // destination's pointer to get the actual memcpy destination (end of
285  // the string .. we're concatenating).
286  Value *CpyDst = B.CreateGEP(B.getInt8Ty(), Dst, DstLen, "endptr");
287 
288  // We have enough information to now generate the memcpy call to do the
289  // concatenation for us. Make a memcpy to copy the nul byte with align = 1.
290  B.CreateMemCpy(CpyDst, 1, Src, 1,
291  ConstantInt::get(DL.getIntPtrType(Src->getContext()), Len + 1));
292  return Dst;
293 }
294 
295 Value *LibCallSimplifier::optimizeStrNCat(CallInst *CI, IRBuilder<> &B) {
296  // Extract some information from the instruction.
297  Value *Dst = CI->getArgOperand(0);
298  Value *Src = CI->getArgOperand(1);
299  Value *Size = CI->getArgOperand(2);
300  uint64_t Len;
302  if (isKnownNonZero(Size, DL))
304 
305  // We don't do anything if length is not constant.
306  ConstantInt *LengthArg = dyn_cast<ConstantInt>(Size);
307  if (LengthArg) {
308  Len = LengthArg->getZExtValue();
309  // strncat(x, c, 0) -> x
310  if (!Len)
311  return Dst;
312  } else {
313  return nullptr;
314  }
315 
316  // See if we can get the length of the input string.
317  uint64_t SrcLen = GetStringLength(Src);
318  if (SrcLen) {
319  annotateDereferenceableBytes(CI, 1, SrcLen);
320  --SrcLen; // Unbias length.
321  } else {
322  return nullptr;
323  }
324 
325  // strncat(x, "", c) -> x
326  if (SrcLen == 0)
327  return Dst;
328 
329  // We don't optimize this case.
330  if (Len < SrcLen)
331  return nullptr;
332 
333  // strncat(x, s, c) -> strcat(x, s)
334  // s is constant so the strcat can be optimized further.
335  return emitStrLenMemCpy(Src, Dst, SrcLen, B);
336 }
337 
338 Value *LibCallSimplifier::optimizeStrChr(CallInst *CI, IRBuilder<> &B) {
340  FunctionType *FT = Callee->getFunctionType();
341  Value *SrcStr = CI->getArgOperand(0);
343 
344  // If the second operand is non-constant, see if we can compute the length
345  // of the input string and turn this into memchr.
346  ConstantInt *CharC = dyn_cast<ConstantInt>(CI->getArgOperand(1));
347  if (!CharC) {
348  uint64_t Len = GetStringLength(SrcStr);
349  if (Len)
350  annotateDereferenceableBytes(CI, 0, Len);
351  else
352  return nullptr;
353  if (!FT->getParamType(1)->isIntegerTy(32)) // memchr needs i32.
354  return nullptr;
355 
356  return emitMemChr(SrcStr, CI->getArgOperand(1), // include nul.
357  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len),
358  B, DL, TLI);
359  }
360 
361  // Otherwise, the character is a constant, see if the first argument is
362  // a string literal. If so, we can constant fold.
363  StringRef Str;
364  if (!getConstantStringInfo(SrcStr, Str)) {
365  if (CharC->isZero()) // strchr(p, 0) -> p + strlen(p)
366  return B.CreateGEP(B.getInt8Ty(), SrcStr, emitStrLen(SrcStr, B, DL, TLI),
367  "strchr");
368  return nullptr;
369  }
370 
371  // Compute the offset, make sure to handle the case when we're searching for
372  // zero (a weird way to spell strlen).
373  size_t I = (0xFF & CharC->getSExtValue()) == 0
374  ? Str.size()
375  : Str.find(CharC->getSExtValue());
376  if (I == StringRef::npos) // Didn't find the char. strchr returns null.
377  return Constant::getNullValue(CI->getType());
378 
379  // strchr(s+n,c) -> gep(s+n+i,c)
380  return B.CreateGEP(B.getInt8Ty(), SrcStr, B.getInt64(I), "strchr");
381 }
382 
383 Value *LibCallSimplifier::optimizeStrRChr(CallInst *CI, IRBuilder<> &B) {
384  Value *SrcStr = CI->getArgOperand(0);
385  ConstantInt *CharC = dyn_cast<ConstantInt>(CI->getArgOperand(1));
387 
388  // Cannot fold anything if we're not looking for a constant.
389  if (!CharC)
390  return nullptr;
391 
392  StringRef Str;
393  if (!getConstantStringInfo(SrcStr, Str)) {
394  // strrchr(s, 0) -> strchr(s, 0)
395  if (CharC->isZero())
396  return emitStrChr(SrcStr, '\0', B, TLI);
397  return nullptr;
398  }
399 
400  // Compute the offset.
401  size_t I = (0xFF & CharC->getSExtValue()) == 0
402  ? Str.size()
403  : Str.rfind(CharC->getSExtValue());
404  if (I == StringRef::npos) // Didn't find the char. Return null.
405  return Constant::getNullValue(CI->getType());
406 
407  // strrchr(s+n,c) -> gep(s+n+i,c)
408  return B.CreateGEP(B.getInt8Ty(), SrcStr, B.getInt64(I), "strrchr");
409 }
410 
411 Value *LibCallSimplifier::optimizeStrCmp(CallInst *CI, IRBuilder<> &B) {
412  Value *Str1P = CI->getArgOperand(0), *Str2P = CI->getArgOperand(1);
413  if (Str1P == Str2P) // strcmp(x,x) -> 0
414  return ConstantInt::get(CI->getType(), 0);
415 
416  StringRef Str1, Str2;
417  bool HasStr1 = getConstantStringInfo(Str1P, Str1);
418  bool HasStr2 = getConstantStringInfo(Str2P, Str2);
419 
420  // strcmp(x, y) -> cnst (if both x and y are constant strings)
421  if (HasStr1 && HasStr2)
422  return ConstantInt::get(CI->getType(), Str1.compare(Str2));
423 
424  if (HasStr1 && Str1.empty()) // strcmp("", x) -> -*x
425  return B.CreateNeg(B.CreateZExt(
426  B.CreateLoad(B.getInt8Ty(), Str2P, "strcmpload"), CI->getType()));
427 
428  if (HasStr2 && Str2.empty()) // strcmp(x,"") -> *x
429  return B.CreateZExt(B.CreateLoad(B.getInt8Ty(), Str1P, "strcmpload"),
430  CI->getType());
431 
432  // strcmp(P, "x") -> memcmp(P, "x", 2)
433  uint64_t Len1 = GetStringLength(Str1P);
434  if (Len1)
435  annotateDereferenceableBytes(CI, 0, Len1);
436  uint64_t Len2 = GetStringLength(Str2P);
437  if (Len2)
438  annotateDereferenceableBytes(CI, 1, Len2);
439 
440  if (Len1 && Len2) {
441  return emitMemCmp(Str1P, Str2P,
442  ConstantInt::get(DL.getIntPtrType(CI->getContext()),
443  std::min(Len1, Len2)),
444  B, DL, TLI);
445  }
446 
447  // strcmp to memcmp
448  if (!HasStr1 && HasStr2) {
449  if (canTransformToMemCmp(CI, Str1P, Len2, DL))
450  return emitMemCmp(
451  Str1P, Str2P,
452  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len2), B, DL,
453  TLI);
454  } else if (HasStr1 && !HasStr2) {
455  if (canTransformToMemCmp(CI, Str2P, Len1, DL))
456  return emitMemCmp(
457  Str1P, Str2P,
458  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len1), B, DL,
459  TLI);
460  }
461 
462  annotateNonNullBasedOnAccess(CI, {0, 1});
463  return nullptr;
464 }
465 
466 Value *LibCallSimplifier::optimizeStrNCmp(CallInst *CI, IRBuilder<> &B) {
467  Value *Str1P = CI->getArgOperand(0);
468  Value *Str2P = CI->getArgOperand(1);
469  Value *Size = CI->getArgOperand(2);
470  if (Str1P == Str2P) // strncmp(x,x,n) -> 0
471  return ConstantInt::get(CI->getType(), 0);
472 
473  if (isKnownNonZero(Size, DL))
474  annotateNonNullBasedOnAccess(CI, {0, 1});
475  // Get the length argument if it is constant.
476  uint64_t Length;
477  if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(Size))
478  Length = LengthArg->getZExtValue();
479  else
480  return nullptr;
481 
482  if (Length == 0) // strncmp(x,y,0) -> 0
483  return ConstantInt::get(CI->getType(), 0);
484 
485  if (Length == 1) // strncmp(x,y,1) -> memcmp(x,y,1)
486  return emitMemCmp(Str1P, Str2P, Size, B, DL, TLI);
487 
488  StringRef Str1, Str2;
489  bool HasStr1 = getConstantStringInfo(Str1P, Str1);
490  bool HasStr2 = getConstantStringInfo(Str2P, Str2);
491 
492  // strncmp(x, y) -> cnst (if both x and y are constant strings)
493  if (HasStr1 && HasStr2) {
494  StringRef SubStr1 = Str1.substr(0, Length);
495  StringRef SubStr2 = Str2.substr(0, Length);
496  return ConstantInt::get(CI->getType(), SubStr1.compare(SubStr2));
497  }
498 
499  if (HasStr1 && Str1.empty()) // strncmp("", x, n) -> -*x
500  return B.CreateNeg(B.CreateZExt(
501  B.CreateLoad(B.getInt8Ty(), Str2P, "strcmpload"), CI->getType()));
502 
503  if (HasStr2 && Str2.empty()) // strncmp(x, "", n) -> *x
504  return B.CreateZExt(B.CreateLoad(B.getInt8Ty(), Str1P, "strcmpload"),
505  CI->getType());
506 
507  uint64_t Len1 = GetStringLength(Str1P);
508  if (Len1)
509  annotateDereferenceableBytes(CI, 0, Len1);
510  uint64_t Len2 = GetStringLength(Str2P);
511  if (Len2)
512  annotateDereferenceableBytes(CI, 1, Len2);
513 
514  // strncmp to memcmp
515  if (!HasStr1 && HasStr2) {
516  Len2 = std::min(Len2, Length);
517  if (canTransformToMemCmp(CI, Str1P, Len2, DL))
518  return emitMemCmp(
519  Str1P, Str2P,
520  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len2), B, DL,
521  TLI);
522  } else if (HasStr1 && !HasStr2) {
523  Len1 = std::min(Len1, Length);
524  if (canTransformToMemCmp(CI, Str2P, Len1, DL))
525  return emitMemCmp(
526  Str1P, Str2P,
527  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len1), B, DL,
528  TLI);
529  }
530 
531  return nullptr;
532 }
533 
534 Value *LibCallSimplifier::optimizeStrCpy(CallInst *CI, IRBuilder<> &B) {
535  Value *Dst = CI->getArgOperand(0), *Src = CI->getArgOperand(1);
536  if (Dst == Src) // strcpy(x,x) -> x
537  return Src;
538 
539  annotateNonNullBasedOnAccess(CI, {0, 1});
540  // See if we can get the length of the input string.
541  uint64_t Len = GetStringLength(Src);
542  if (Len)
543  annotateDereferenceableBytes(CI, 1, Len);
544  else
545  return nullptr;
546 
547  // We have enough information to now generate the memcpy call to do the
548  // copy for us. Make a memcpy to copy the nul byte with align = 1.
549  CallInst *NewCI =
550  B.CreateMemCpy(Dst, 1, Src, 1,
551  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len));
552  NewCI->setAttributes(CI->getAttributes());
553  return Dst;
554 }
555 
556 Value *LibCallSimplifier::optimizeStpCpy(CallInst *CI, IRBuilder<> &B) {
558  Value *Dst = CI->getArgOperand(0), *Src = CI->getArgOperand(1);
559  if (Dst == Src) { // stpcpy(x,x) -> x+strlen(x)
560  Value *StrLen = emitStrLen(Src, B, DL, TLI);
561  return StrLen ? B.CreateInBoundsGEP(B.getInt8Ty(), Dst, StrLen) : nullptr;
562  }
563 
564  // See if we can get the length of the input string.
565  uint64_t Len = GetStringLength(Src);
566  if (Len)
567  annotateDereferenceableBytes(CI, 1, Len);
568  else
569  return nullptr;
570 
571  Type *PT = Callee->getFunctionType()->getParamType(0);
572  Value *LenV = ConstantInt::get(DL.getIntPtrType(PT), Len);
573  Value *DstEnd = B.CreateGEP(B.getInt8Ty(), Dst,
574  ConstantInt::get(DL.getIntPtrType(PT), Len - 1));
575 
576  // We have enough information to now generate the memcpy call to do the
577  // copy for us. Make a memcpy to copy the nul byte with align = 1.
578  CallInst *NewCI = B.CreateMemCpy(Dst, 1, Src, 1, LenV);
579  NewCI->setAttributes(CI->getAttributes());
580  return DstEnd;
581 }
582 
583 Value *LibCallSimplifier::optimizeStrNCpy(CallInst *CI, IRBuilder<> &B) {
585  Value *Dst = CI->getArgOperand(0);
586  Value *Src = CI->getArgOperand(1);
587  Value *Size = CI->getArgOperand(2);
589  if (isKnownNonZero(Size, DL))
591 
592  uint64_t Len;
593  if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(Size))
594  Len = LengthArg->getZExtValue();
595  else
596  return nullptr;
597 
598  // strncpy(x, y, 0) -> x
599  if (Len == 0)
600  return Dst;
601 
602  // See if we can get the length of the input string.
603  uint64_t SrcLen = GetStringLength(Src);
604  if (SrcLen) {
605  annotateDereferenceableBytes(CI, 1, SrcLen);
606  --SrcLen; // Unbias length.
607  } else {
608  return nullptr;
609  }
610 
611  if (SrcLen == 0) {
612  // strncpy(x, "", y) -> memset(align 1 x, '\0', y)
613  CallInst *NewCI = B.CreateMemSet(Dst, B.getInt8('\0'), Size, 1);
614  AttrBuilder ArgAttrs(CI->getAttributes().getParamAttributes(0));
616  CI->getContext(), 0, ArgAttrs));
617  return Dst;
618  }
619 
620  // Let strncpy handle the zero padding
621  if (Len > SrcLen + 1)
622  return nullptr;
623 
624  Type *PT = Callee->getFunctionType()->getParamType(0);
625  // strncpy(x, s, c) -> memcpy(align 1 x, align 1 s, c) [s and c are constant]
626  CallInst *NewCI = B.CreateMemCpy(Dst, 1, Src, 1, ConstantInt::get(DL.getIntPtrType(PT), Len));
627  NewCI->setAttributes(CI->getAttributes());
628  return Dst;
629 }
630 
631 Value *LibCallSimplifier::optimizeStringLength(CallInst *CI, IRBuilder<> &B,
632  unsigned CharSize) {
633  Value *Src = CI->getArgOperand(0);
634 
635  // Constant folding: strlen("xyz") -> 3
636  if (uint64_t Len = GetStringLength(Src, CharSize))
637  return ConstantInt::get(CI->getType(), Len - 1);
638 
639  // If s is a constant pointer pointing to a string literal, we can fold
640  // strlen(s + x) to strlen(s) - x, when x is known to be in the range
641  // [0, strlen(s)] or the string has a single null terminator '\0' at the end.
642  // We only try to simplify strlen when the pointer s points to an array
643  // of i8. Otherwise, we would need to scale the offset x before doing the
644  // subtraction. This will make the optimization more complex, and it's not
645  // very useful because calling strlen for a pointer of other types is
646  // very uncommon.
647  if (GEPOperator *GEP = dyn_cast<GEPOperator>(Src)) {
648  if (!isGEPBasedOnPointerToString(GEP, CharSize))
649  return nullptr;
650 
652  if (getConstantDataArrayInfo(GEP->getOperand(0), Slice, CharSize)) {
653  uint64_t NullTermIdx;
654  if (Slice.Array == nullptr) {
655  NullTermIdx = 0;
656  } else {
657  NullTermIdx = ~((uint64_t)0);
658  for (uint64_t I = 0, E = Slice.Length; I < E; ++I) {
659  if (Slice.Array->getElementAsInteger(I + Slice.Offset) == 0) {
660  NullTermIdx = I;
661  break;
662  }
663  }
664  // If the string does not have '\0', leave it to strlen to compute
665  // its length.
666  if (NullTermIdx == ~((uint64_t)0))
667  return nullptr;
668  }
669 
670  Value *Offset = GEP->getOperand(2);
671  KnownBits Known = computeKnownBits(Offset, DL, 0, nullptr, CI, nullptr);
672  Known.Zero.flipAllBits();
673  uint64_t ArrSize =
674  cast<ArrayType>(GEP->getSourceElementType())->getNumElements();
675 
676  // KnownZero's bits are flipped, so zeros in KnownZero now represent
677  // bits known to be zeros in Offset, and ones in KnowZero represent
678  // bits unknown in Offset. Therefore, Offset is known to be in range
679  // [0, NullTermIdx] when the flipped KnownZero is non-negative and
680  // unsigned-less-than NullTermIdx.
681  //
682  // If Offset is not provably in the range [0, NullTermIdx], we can still
683  // optimize if we can prove that the program has undefined behavior when
684  // Offset is outside that range. That is the case when GEP->getOperand(0)
685  // is a pointer to an object whose memory extent is NullTermIdx+1.
686  if ((Known.Zero.isNonNegative() && Known.Zero.ule(NullTermIdx)) ||
687  (GEP->isInBounds() && isa<GlobalVariable>(GEP->getOperand(0)) &&
688  NullTermIdx == ArrSize - 1)) {
689  Offset = B.CreateSExtOrTrunc(Offset, CI->getType());
690  return B.CreateSub(ConstantInt::get(CI->getType(), NullTermIdx),
691  Offset);
692  }
693  }
694 
695  return nullptr;
696  }
697 
698  // strlen(x?"foo":"bars") --> x ? 3 : 4
699  if (SelectInst *SI = dyn_cast<SelectInst>(Src)) {
700  uint64_t LenTrue = GetStringLength(SI->getTrueValue(), CharSize);
701  uint64_t LenFalse = GetStringLength(SI->getFalseValue(), CharSize);
702  if (LenTrue && LenFalse) {
703  ORE.emit([&]() {
704  return OptimizationRemark("instcombine", "simplify-libcalls", CI)
705  << "folded strlen(select) to select of constants";
706  });
707  return B.CreateSelect(SI->getCondition(),
708  ConstantInt::get(CI->getType(), LenTrue - 1),
709  ConstantInt::get(CI->getType(), LenFalse - 1));
710  }
711  }
712 
713  // strlen(x) != 0 --> *x != 0
714  // strlen(x) == 0 --> *x == 0
716  return B.CreateZExt(B.CreateLoad(B.getIntNTy(CharSize), Src, "strlenfirst"),
717  CI->getType());
718 
719  return nullptr;
720 }
721 
722 Value *LibCallSimplifier::optimizeStrLen(CallInst *CI, IRBuilder<> &B) {
723  if (Value *V = optimizeStringLength(CI, B, 8))
724  return V;
726  return nullptr;
727 }
728 
729 Value *LibCallSimplifier::optimizeWcslen(CallInst *CI, IRBuilder<> &B) {
730  Module &M = *CI->getModule();
731  unsigned WCharSize = TLI->getWCharSize(M) * 8;
732  // We cannot perform this optimization without wchar_size metadata.
733  if (WCharSize == 0)
734  return nullptr;
735 
736  return optimizeStringLength(CI, B, WCharSize);
737 }
738 
739 Value *LibCallSimplifier::optimizeStrPBrk(CallInst *CI, IRBuilder<> &B) {
740  StringRef S1, S2;
741  bool HasS1 = getConstantStringInfo(CI->getArgOperand(0), S1);
742  bool HasS2 = getConstantStringInfo(CI->getArgOperand(1), S2);
743 
744  // strpbrk(s, "") -> nullptr
745  // strpbrk("", s) -> nullptr
746  if ((HasS1 && S1.empty()) || (HasS2 && S2.empty()))
747  return Constant::getNullValue(CI->getType());
748 
749  // Constant folding.
750  if (HasS1 && HasS2) {
751  size_t I = S1.find_first_of(S2);
752  if (I == StringRef::npos) // No match.
753  return Constant::getNullValue(CI->getType());
754 
755  return B.CreateGEP(B.getInt8Ty(), CI->getArgOperand(0), B.getInt64(I),
756  "strpbrk");
757  }
758 
759  // strpbrk(s, "a") -> strchr(s, 'a')
760  if (HasS2 && S2.size() == 1)
761  return emitStrChr(CI->getArgOperand(0), S2[0], B, TLI);
762 
763  return nullptr;
764 }
765 
766 Value *LibCallSimplifier::optimizeStrTo(CallInst *CI, IRBuilder<> &B) {
767  Value *EndPtr = CI->getArgOperand(1);
768  if (isa<ConstantPointerNull>(EndPtr)) {
769  // With a null EndPtr, this function won't capture the main argument.
770  // It would be readonly too, except that it still may write to errno.
771  CI->addParamAttr(0, Attribute::NoCapture);
772  }
773 
774  return nullptr;
775 }
776 
777 Value *LibCallSimplifier::optimizeStrSpn(CallInst *CI, IRBuilder<> &B) {
778  StringRef S1, S2;
779  bool HasS1 = getConstantStringInfo(CI->getArgOperand(0), S1);
780  bool HasS2 = getConstantStringInfo(CI->getArgOperand(1), S2);
781 
782  // strspn(s, "") -> 0
783  // strspn("", s) -> 0
784  if ((HasS1 && S1.empty()) || (HasS2 && S2.empty()))
785  return Constant::getNullValue(CI->getType());
786 
787  // Constant folding.
788  if (HasS1 && HasS2) {
789  size_t Pos = S1.find_first_not_of(S2);
790  if (Pos == StringRef::npos)
791  Pos = S1.size();
792  return ConstantInt::get(CI->getType(), Pos);
793  }
794 
795  return nullptr;
796 }
797 
798 Value *LibCallSimplifier::optimizeStrCSpn(CallInst *CI, IRBuilder<> &B) {
799  StringRef S1, S2;
800  bool HasS1 = getConstantStringInfo(CI->getArgOperand(0), S1);
801  bool HasS2 = getConstantStringInfo(CI->getArgOperand(1), S2);
802 
803  // strcspn("", s) -> 0
804  if (HasS1 && S1.empty())
805  return Constant::getNullValue(CI->getType());
806 
807  // Constant folding.
808  if (HasS1 && HasS2) {
809  size_t Pos = S1.find_first_of(S2);
810  if (Pos == StringRef::npos)
811  Pos = S1.size();
812  return ConstantInt::get(CI->getType(), Pos);
813  }
814 
815  // strcspn(s, "") -> strlen(s)
816  if (HasS2 && S2.empty())
817  return emitStrLen(CI->getArgOperand(0), B, DL, TLI);
818 
819  return nullptr;
820 }
821 
822 Value *LibCallSimplifier::optimizeStrStr(CallInst *CI, IRBuilder<> &B) {
823  // fold strstr(x, x) -> x.
824  if (CI->getArgOperand(0) == CI->getArgOperand(1))
825  return B.CreateBitCast(CI->getArgOperand(0), CI->getType());
826 
827  // fold strstr(a, b) == a -> strncmp(a, b, strlen(b)) == 0
829  Value *StrLen = emitStrLen(CI->getArgOperand(1), B, DL, TLI);
830  if (!StrLen)
831  return nullptr;
832  Value *StrNCmp = emitStrNCmp(CI->getArgOperand(0), CI->getArgOperand(1),
833  StrLen, B, DL, TLI);
834  if (!StrNCmp)
835  return nullptr;
836  for (auto UI = CI->user_begin(), UE = CI->user_end(); UI != UE;) {
837  ICmpInst *Old = cast<ICmpInst>(*UI++);
838  Value *Cmp =
839  B.CreateICmp(Old->getPredicate(), StrNCmp,
840  ConstantInt::getNullValue(StrNCmp->getType()), "cmp");
841  replaceAllUsesWith(Old, Cmp);
842  }
843  return CI;
844  }
845 
846  // See if either input string is a constant string.
847  StringRef SearchStr, ToFindStr;
848  bool HasStr1 = getConstantStringInfo(CI->getArgOperand(0), SearchStr);
849  bool HasStr2 = getConstantStringInfo(CI->getArgOperand(1), ToFindStr);
850 
851  // fold strstr(x, "") -> x.
852  if (HasStr2 && ToFindStr.empty())
853  return B.CreateBitCast(CI->getArgOperand(0), CI->getType());
854 
855  // If both strings are known, constant fold it.
856  if (HasStr1 && HasStr2) {
857  size_t Offset = SearchStr.find(ToFindStr);
858 
859  if (Offset == StringRef::npos) // strstr("foo", "bar") -> null
860  return Constant::getNullValue(CI->getType());
861 
862  // strstr("abcd", "bc") -> gep((char*)"abcd", 1)
863  Value *Result = castToCStr(CI->getArgOperand(0), B);
864  Result =
865  B.CreateConstInBoundsGEP1_64(B.getInt8Ty(), Result, Offset, "strstr");
866  return B.CreateBitCast(Result, CI->getType());
867  }
868 
869  // fold strstr(x, "y") -> strchr(x, 'y').
870  if (HasStr2 && ToFindStr.size() == 1) {
871  Value *StrChr = emitStrChr(CI->getArgOperand(0), ToFindStr[0], B, TLI);
872  return StrChr ? B.CreateBitCast(StrChr, CI->getType()) : nullptr;
873  }
874 
875  annotateNonNullBasedOnAccess(CI, {0, 1});
876  return nullptr;
877 }
878 
879 Value *LibCallSimplifier::optimizeMemRChr(CallInst *CI, IRBuilder<> &B) {
880  if (isKnownNonZero(CI->getOperand(2), DL))
882  return nullptr;
883 }
884 
885 Value *LibCallSimplifier::optimizeMemChr(CallInst *CI, IRBuilder<> &B) {
886  Value *SrcStr = CI->getArgOperand(0);
887  Value *Size = CI->getArgOperand(2);
888  annotateNonNullAndDereferenceable(CI, 0, Size, DL);
889  ConstantInt *CharC = dyn_cast<ConstantInt>(CI->getArgOperand(1));
891 
892  // memchr(x, y, 0) -> null
893  if (LenC) {
894  if (LenC->isZero())
895  return Constant::getNullValue(CI->getType());
896  } else {
897  // From now on we need at least constant length and string.
898  return nullptr;
899  }
900 
901  StringRef Str;
902  if (!getConstantStringInfo(SrcStr, Str, 0, /*TrimAtNul=*/false))
903  return nullptr;
904 
905  // Truncate the string to LenC. If Str is smaller than LenC we will still only
906  // scan the string, as reading past the end of it is undefined and we can just
907  // return null if we don't find the char.
908  Str = Str.substr(0, LenC->getZExtValue());
909 
910  // If the char is variable but the input str and length are not we can turn
911  // this memchr call into a simple bit field test. Of course this only works
912  // when the return value is only checked against null.
913  //
914  // It would be really nice to reuse switch lowering here but we can't change
915  // the CFG at this point.
916  //
917  // memchr("\r\n", C, 2) != nullptr -> (1 << C & ((1 << '\r') | (1 << '\n')))
918  // != 0
919  // after bounds check.
920  if (!CharC && !Str.empty() && isOnlyUsedInZeroEqualityComparison(CI)) {
921  unsigned char Max =
922  *std::max_element(reinterpret_cast<const unsigned char *>(Str.begin()),
923  reinterpret_cast<const unsigned char *>(Str.end()));
924 
925  // Make sure the bit field we're about to create fits in a register on the
926  // target.
927  // FIXME: On a 64 bit architecture this prevents us from using the
928  // interesting range of alpha ascii chars. We could do better by emitting
929  // two bitfields or shifting the range by 64 if no lower chars are used.
930  if (!DL.fitsInLegalInteger(Max + 1))
931  return nullptr;
932 
933  // For the bit field use a power-of-2 type with at least 8 bits to avoid
934  // creating unnecessary illegal types.
935  unsigned char Width = NextPowerOf2(std::max((unsigned char)7, Max));
936 
937  // Now build the bit field.
938  APInt Bitfield(Width, 0);
939  for (char C : Str)
940  Bitfield.setBit((unsigned char)C);
941  Value *BitfieldC = B.getInt(Bitfield);
942 
943  // Adjust width of "C" to the bitfield width, then mask off the high bits.
944  Value *C = B.CreateZExtOrTrunc(CI->getArgOperand(1), BitfieldC->getType());
945  C = B.CreateAnd(C, B.getIntN(Width, 0xFF));
946 
947  // First check that the bit field access is within bounds.
948  Value *Bounds = B.CreateICmp(ICmpInst::ICMP_ULT, C, B.getIntN(Width, Width),
949  "memchr.bounds");
950 
951  // Create code that checks if the given bit is set in the field.
952  Value *Shl = B.CreateShl(B.getIntN(Width, 1ULL), C);
953  Value *Bits = B.CreateIsNotNull(B.CreateAnd(Shl, BitfieldC), "memchr.bits");
954 
955  // Finally merge both checks and cast to pointer type. The inttoptr
956  // implicitly zexts the i1 to intptr type.
957  return B.CreateIntToPtr(B.CreateAnd(Bounds, Bits, "memchr"), CI->getType());
958  }
959 
960  // Check if all arguments are constants. If so, we can constant fold.
961  if (!CharC)
962  return nullptr;
963 
964  // Compute the offset.
965  size_t I = Str.find(CharC->getSExtValue() & 0xFF);
966  if (I == StringRef::npos) // Didn't find the char. memchr returns null.
967  return Constant::getNullValue(CI->getType());
968 
969  // memchr(s+n,c,l) -> gep(s+n+i,c)
970  return B.CreateGEP(B.getInt8Ty(), SrcStr, B.getInt64(I), "memchr");
971 }
972 
974  uint64_t Len, IRBuilder<> &B,
975  const DataLayout &DL) {
976  if (Len == 0) // memcmp(s1,s2,0) -> 0
977  return Constant::getNullValue(CI->getType());
978 
979  // memcmp(S1,S2,1) -> *(unsigned char*)LHS - *(unsigned char*)RHS
980  if (Len == 1) {
981  Value *LHSV =
982  B.CreateZExt(B.CreateLoad(B.getInt8Ty(), castToCStr(LHS, B), "lhsc"),
983  CI->getType(), "lhsv");
984  Value *RHSV =
985  B.CreateZExt(B.CreateLoad(B.getInt8Ty(), castToCStr(RHS, B), "rhsc"),
986  CI->getType(), "rhsv");
987  return B.CreateSub(LHSV, RHSV, "chardiff");
988  }
989 
990  // memcmp(S1,S2,N/8)==0 -> (*(intN_t*)S1 != *(intN_t*)S2)==0
991  // TODO: The case where both inputs are constants does not need to be limited
992  // to legal integers or equality comparison. See block below this.
993  if (DL.isLegalInteger(Len * 8) && isOnlyUsedInZeroEqualityComparison(CI)) {
994  IntegerType *IntType = IntegerType::get(CI->getContext(), Len * 8);
995  unsigned PrefAlignment = DL.getPrefTypeAlignment(IntType);
996 
997  // First, see if we can fold either argument to a constant.
998  Value *LHSV = nullptr;
999  if (auto *LHSC = dyn_cast<Constant>(LHS)) {
1000  LHSC = ConstantExpr::getBitCast(LHSC, IntType->getPointerTo());
1001  LHSV = ConstantFoldLoadFromConstPtr(LHSC, IntType, DL);
1002  }
1003  Value *RHSV = nullptr;
1004  if (auto *RHSC = dyn_cast<Constant>(RHS)) {
1005  RHSC = ConstantExpr::getBitCast(RHSC, IntType->getPointerTo());
1006  RHSV = ConstantFoldLoadFromConstPtr(RHSC, IntType, DL);
1007  }
1008 
1009  // Don't generate unaligned loads. If either source is constant data,
1010  // alignment doesn't matter for that source because there is no load.
1011  if ((LHSV || getKnownAlignment(LHS, DL, CI) >= PrefAlignment) &&
1012  (RHSV || getKnownAlignment(RHS, DL, CI) >= PrefAlignment)) {
1013  if (!LHSV) {
1014  Type *LHSPtrTy =
1015  IntType->getPointerTo(LHS->getType()->getPointerAddressSpace());
1016  LHSV = B.CreateLoad(IntType, B.CreateBitCast(LHS, LHSPtrTy), "lhsv");
1017  }
1018  if (!RHSV) {
1019  Type *RHSPtrTy =
1020  IntType->getPointerTo(RHS->getType()->getPointerAddressSpace());
1021  RHSV = B.CreateLoad(IntType, B.CreateBitCast(RHS, RHSPtrTy), "rhsv");
1022  }
1023  return B.CreateZExt(B.CreateICmpNE(LHSV, RHSV), CI->getType(), "memcmp");
1024  }
1025  }
1026 
1027  // Constant folding: memcmp(x, y, Len) -> constant (all arguments are const).
1028  // TODO: This is limited to i8 arrays.
1029  StringRef LHSStr, RHSStr;
1030  if (getConstantStringInfo(LHS, LHSStr) &&
1031  getConstantStringInfo(RHS, RHSStr)) {
1032  // Make sure we're not reading out-of-bounds memory.
1033  if (Len > LHSStr.size() || Len > RHSStr.size())
1034  return nullptr;
1035  // Fold the memcmp and normalize the result. This way we get consistent
1036  // results across multiple platforms.
1037  uint64_t Ret = 0;
1038  int Cmp = memcmp(LHSStr.data(), RHSStr.data(), Len);
1039  if (Cmp < 0)
1040  Ret = -1;
1041  else if (Cmp > 0)
1042  Ret = 1;
1043  return ConstantInt::get(CI->getType(), Ret);
1044  }
1045 
1046  return nullptr;
1047 }
1048 
1049 // Most simplifications for memcmp also apply to bcmp.
1050 Value *LibCallSimplifier::optimizeMemCmpBCmpCommon(CallInst *CI,
1051  IRBuilder<> &B) {
1052  Value *LHS = CI->getArgOperand(0), *RHS = CI->getArgOperand(1);
1053  Value *Size = CI->getArgOperand(2);
1054 
1055  if (LHS == RHS) // memcmp(s,s,x) -> 0
1056  return Constant::getNullValue(CI->getType());
1057 
1058  annotateNonNullAndDereferenceable(CI, {0, 1}, Size, DL);
1059  // Handle constant lengths.
1061  if (!LenC)
1062  return nullptr;
1063 
1064  // memcmp(d,s,0) -> 0
1065  if (LenC->getZExtValue() == 0)
1066  return Constant::getNullValue(CI->getType());
1067 
1068  if (Value *Res =
1069  optimizeMemCmpConstantSize(CI, LHS, RHS, LenC->getZExtValue(), B, DL))
1070  return Res;
1071  return nullptr;
1072 }
1073 
1074 Value *LibCallSimplifier::optimizeMemCmp(CallInst *CI, IRBuilder<> &B) {
1075  if (Value *V = optimizeMemCmpBCmpCommon(CI, B))
1076  return V;
1077 
1078  // memcmp(x, y, Len) == 0 -> bcmp(x, y, Len) == 0
1079  // bcmp can be more efficient than memcmp because it only has to know that
1080  // there is a difference, not how different one is to the other.
1081  if (TLI->has(LibFunc_bcmp) && isOnlyUsedInZeroEqualityComparison(CI)) {
1082  Value *LHS = CI->getArgOperand(0);
1083  Value *RHS = CI->getArgOperand(1);
1084  Value *Size = CI->getArgOperand(2);
1085  return emitBCmp(LHS, RHS, Size, B, DL, TLI);
1086  }
1087 
1088  return nullptr;
1089 }
1090 
1091 Value *LibCallSimplifier::optimizeBCmp(CallInst *CI, IRBuilder<> &B) {
1092  return optimizeMemCmpBCmpCommon(CI, B);
1093 }
1094 
1095 Value *LibCallSimplifier::optimizeMemCpy(CallInst *CI, IRBuilder<> &B) {
1096  Value *Size = CI->getArgOperand(2);
1097  annotateNonNullAndDereferenceable(CI, {0, 1}, Size, DL);
1098  if (isa<IntrinsicInst>(CI))
1099  return nullptr;
1100 
1101  // memcpy(x, y, n) -> llvm.memcpy(align 1 x, align 1 y, n)
1102  CallInst *NewCI =
1103  B.CreateMemCpy(CI->getArgOperand(0), 1, CI->getArgOperand(1), 1, Size);
1104  NewCI->setAttributes(CI->getAttributes());
1105  return CI->getArgOperand(0);
1106 }
1107 
1108 Value *LibCallSimplifier::optimizeMemPCpy(CallInst *CI, IRBuilder<> &B) {
1109  Value *Dst = CI->getArgOperand(0);
1110  Value *N = CI->getArgOperand(2);
1111  // mempcpy(x, y, n) -> llvm.memcpy(align 1 x, align 1 y, n), x + n
1112  CallInst *NewCI = B.CreateMemCpy(Dst, 1, CI->getArgOperand(1), 1, N);
1113  NewCI->setAttributes(CI->getAttributes());
1114  return B.CreateInBoundsGEP(B.getInt8Ty(), Dst, N);
1115 }
1116 
1117 Value *LibCallSimplifier::optimizeMemMove(CallInst *CI, IRBuilder<> &B) {
1118  Value *Size = CI->getArgOperand(2);
1119  annotateNonNullAndDereferenceable(CI, {0, 1}, Size, DL);
1120  if (isa<IntrinsicInst>(CI))
1121  return nullptr;
1122 
1123  // memmove(x, y, n) -> llvm.memmove(align 1 x, align 1 y, n)
1124  CallInst *NewCI =
1125  B.CreateMemMove(CI->getArgOperand(0), 1, CI->getArgOperand(1), 1, Size);
1126  NewCI->setAttributes(CI->getAttributes());
1127  return CI->getArgOperand(0);
1128 }
1129 
1130 /// Fold memset[_chk](malloc(n), 0, n) --> calloc(1, n).
1131 Value *LibCallSimplifier::foldMallocMemset(CallInst *Memset, IRBuilder<> &B) {
1132  // This has to be a memset of zeros (bzero).
1133  auto *FillValue = dyn_cast<ConstantInt>(Memset->getArgOperand(1));
1134  if (!FillValue || FillValue->getZExtValue() != 0)
1135  return nullptr;
1136 
1137  // TODO: We should handle the case where the malloc has more than one use.
1138  // This is necessary to optimize common patterns such as when the result of
1139  // the malloc is checked against null or when a memset intrinsic is used in
1140  // place of a memset library call.
1141  auto *Malloc = dyn_cast<CallInst>(Memset->getArgOperand(0));
1142  if (!Malloc || !Malloc->hasOneUse())
1143  return nullptr;
1144 
1145  // Is the inner call really malloc()?
1146  Function *InnerCallee = Malloc->getCalledFunction();
1147  if (!InnerCallee)
1148  return nullptr;
1149 
1150  LibFunc Func;
1151  if (!TLI->getLibFunc(*InnerCallee, Func) || !TLI->has(Func) ||
1152  Func != LibFunc_malloc)
1153  return nullptr;
1154 
1155  // The memset must cover the same number of bytes that are malloc'd.
1156  if (Memset->getArgOperand(2) != Malloc->getArgOperand(0))
1157  return nullptr;
1158 
1159  // Replace the malloc with a calloc. We need the data layout to know what the
1160  // actual size of a 'size_t' parameter is.
1161  B.SetInsertPoint(Malloc->getParent(), ++Malloc->getIterator());
1162  const DataLayout &DL = Malloc->getModule()->getDataLayout();
1163  IntegerType *SizeType = DL.getIntPtrType(B.GetInsertBlock()->getContext());
1164  if (Value *Calloc = emitCalloc(ConstantInt::get(SizeType, 1),
1165  Malloc->getArgOperand(0),
1166  Malloc->getAttributes(), B, *TLI)) {
1167  substituteInParent(Malloc, Calloc);
1168  return Calloc;
1169  }
1170 
1171  return nullptr;
1172 }
1173 
1174 Value *LibCallSimplifier::optimizeMemSet(CallInst *CI, IRBuilder<> &B) {
1175  Value *Size = CI->getArgOperand(2);
1176  annotateNonNullAndDereferenceable(CI, 0, Size, DL);
1177  if (isa<IntrinsicInst>(CI))
1178  return nullptr;
1179 
1180  if (auto *Calloc = foldMallocMemset(CI, B))
1181  return Calloc;
1182 
1183  // memset(p, v, n) -> llvm.memset(align 1 p, v, n)
1184  Value *Val = B.CreateIntCast(CI->getArgOperand(1), B.getInt8Ty(), false);
1185  CallInst *NewCI = B.CreateMemSet(CI->getArgOperand(0), Val, Size, 1);
1186  NewCI->setAttributes(CI->getAttributes());
1187  return CI->getArgOperand(0);
1188 }
1189 
1190 Value *LibCallSimplifier::optimizeRealloc(CallInst *CI, IRBuilder<> &B) {
1191  if (isa<ConstantPointerNull>(CI->getArgOperand(0)))
1192  return emitMalloc(CI->getArgOperand(1), B, DL, TLI);
1193 
1194  return nullptr;
1195 }
1196 
1197 //===----------------------------------------------------------------------===//
1198 // Math Library Optimizations
1199 //===----------------------------------------------------------------------===//
1200 
1201 // Replace a libcall \p CI with a call to intrinsic \p IID
1203  // Propagate fast-math flags from the existing call to the new call.
1206 
1207  Module *M = CI->getModule();
1208  Value *V = CI->getArgOperand(0);
1209  Function *F = Intrinsic::getDeclaration(M, IID, CI->getType());
1210  CallInst *NewCall = B.CreateCall(F, V);
1211  NewCall->takeName(CI);
1212  return NewCall;
1213 }
1214 
1215 /// Return a variant of Val with float type.
1216 /// Currently this works in two cases: If Val is an FPExtension of a float
1217 /// value to something bigger, simply return the operand.
1218 /// If Val is a ConstantFP but can be converted to a float ConstantFP without
1219 /// loss of precision do so.
1221  if (FPExtInst *Cast = dyn_cast<FPExtInst>(Val)) {
1222  Value *Op = Cast->getOperand(0);
1223  if (Op->getType()->isFloatTy())
1224  return Op;
1225  }
1226  if (ConstantFP *Const = dyn_cast<ConstantFP>(Val)) {
1227  APFloat F = Const->getValueAPF();
1228  bool losesInfo;
1230  &losesInfo);
1231  if (!losesInfo)
1232  return ConstantFP::get(Const->getContext(), F);
1233  }
1234  return nullptr;
1235 }
1236 
1237 /// Shrink double -> float functions.
1239  bool isBinary, bool isPrecise = false) {
1240  Function *CalleeFn = CI->getCalledFunction();
1241  if (!CI->getType()->isDoubleTy() || !CalleeFn)
1242  return nullptr;
1243 
1244  // If not all the uses of the function are converted to float, then bail out.
1245  // This matters if the precision of the result is more important than the
1246  // precision of the arguments.
1247  if (isPrecise)
1248  for (User *U : CI->users()) {
1249  FPTruncInst *Cast = dyn_cast<FPTruncInst>(U);
1250  if (!Cast || !Cast->getType()->isFloatTy())
1251  return nullptr;
1252  }
1253 
1254  // If this is something like 'g((double) float)', convert to 'gf(float)'.
1255  Value *V[2];
1256  V[0] = valueHasFloatPrecision(CI->getArgOperand(0));
1257  V[1] = isBinary ? valueHasFloatPrecision(CI->getArgOperand(1)) : nullptr;
1258  if (!V[0] || (isBinary && !V[1]))
1259  return nullptr;
1260 
1261  // If call isn't an intrinsic, check that it isn't within a function with the
1262  // same name as the float version of this call, otherwise the result is an
1263  // infinite loop. For example, from MinGW-w64:
1264  //
1265  // float expf(float val) { return (float) exp((double) val); }
1266  StringRef CalleeName = CalleeFn->getName();
1267  bool IsIntrinsic = CalleeFn->isIntrinsic();
1268  if (!IsIntrinsic) {
1269  StringRef CallerName = CI->getFunction()->getName();
1270  if (!CallerName.empty() && CallerName.back() == 'f' &&
1271  CallerName.size() == (CalleeName.size() + 1) &&
1272  CallerName.startswith(CalleeName))
1273  return nullptr;
1274  }
1275 
1276  // Propagate the math semantics from the current function to the new function.
1279 
1280  // g((double) float) -> (double) gf(float)
1281  Value *R;
1282  if (IsIntrinsic) {
1283  Module *M = CI->getModule();
1284  Intrinsic::ID IID = CalleeFn->getIntrinsicID();
1285  Function *Fn = Intrinsic::getDeclaration(M, IID, B.getFloatTy());
1286  R = isBinary ? B.CreateCall(Fn, V) : B.CreateCall(Fn, V[0]);
1287  } else {
1288  AttributeList CalleeAttrs = CalleeFn->getAttributes();
1289  R = isBinary ? emitBinaryFloatFnCall(V[0], V[1], CalleeName, B, CalleeAttrs)
1290  : emitUnaryFloatFnCall(V[0], CalleeName, B, CalleeAttrs);
1291  }
1292  return B.CreateFPExt(R, B.getDoubleTy());
1293 }
1294 
1295 /// Shrink double -> float for unary functions.
1297  bool isPrecise = false) {
1298  return optimizeDoubleFP(CI, B, false, isPrecise);
1299 }
1300 
1301 /// Shrink double -> float for binary functions.
1303  bool isPrecise = false) {
1304  return optimizeDoubleFP(CI, B, true, isPrecise);
1305 }
1306 
1307 // cabs(z) -> sqrt((creal(z)*creal(z)) + (cimag(z)*cimag(z)))
1308 Value *LibCallSimplifier::optimizeCAbs(CallInst *CI, IRBuilder<> &B) {
1309  if (!CI->isFast())
1310  return nullptr;
1311 
1312  // Propagate fast-math flags from the existing call to new instructions.
1315 
1316  Value *Real, *Imag;
1317  if (CI->getNumArgOperands() == 1) {
1318  Value *Op = CI->getArgOperand(0);
1319  assert(Op->getType()->isArrayTy() && "Unexpected signature for cabs!");
1320  Real = B.CreateExtractValue(Op, 0, "real");
1321  Imag = B.CreateExtractValue(Op, 1, "imag");
1322  } else {
1323  assert(CI->getNumArgOperands() == 2 && "Unexpected signature for cabs!");
1324  Real = CI->getArgOperand(0);
1325  Imag = CI->getArgOperand(1);
1326  }
1327 
1328  Value *RealReal = B.CreateFMul(Real, Real);
1329  Value *ImagImag = B.CreateFMul(Imag, Imag);
1330 
1331  Function *FSqrt = Intrinsic::getDeclaration(CI->getModule(), Intrinsic::sqrt,
1332  CI->getType());
1333  return B.CreateCall(FSqrt, B.CreateFAdd(RealReal, ImagImag), "cabs");
1334 }
1335 
1337  IRBuilder<> &B) {
1338  if (!isa<FPMathOperator>(Call))
1339  return nullptr;
1340 
1343 
1344  // TODO: Can this be shared to also handle LLVM intrinsics?
1345  Value *X;
1346  switch (Func) {
1347  case LibFunc_sin:
1348  case LibFunc_sinf:
1349  case LibFunc_sinl:
1350  case LibFunc_tan:
1351  case LibFunc_tanf:
1352  case LibFunc_tanl:
1353  // sin(-X) --> -sin(X)
1354  // tan(-X) --> -tan(X)
1355  if (match(Call->getArgOperand(0), m_OneUse(m_FNeg(m_Value(X)))))
1356  return B.CreateFNeg(B.CreateCall(Call->getCalledFunction(), X));
1357  break;
1358  case LibFunc_cos:
1359  case LibFunc_cosf:
1360  case LibFunc_cosl:
1361  // cos(-X) --> cos(X)
1362  if (match(Call->getArgOperand(0), m_FNeg(m_Value(X))))
1363  return B.CreateCall(Call->getCalledFunction(), X, "cos");
1364  break;
1365  default:
1366  break;
1367  }
1368  return nullptr;
1369 }
1370 
1371 static Value *getPow(Value *InnerChain[33], unsigned Exp, IRBuilder<> &B) {
1372  // Multiplications calculated using Addition Chains.
1373  // Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html
1374 
1375  assert(Exp != 0 && "Incorrect exponent 0 not handled");
1376 
1377  if (InnerChain[Exp])
1378  return InnerChain[Exp];
1379 
1380  static const unsigned AddChain[33][2] = {
1381  {0, 0}, // Unused.
1382  {0, 0}, // Unused (base case = pow1).
1383  {1, 1}, // Unused (pre-computed).
1384  {1, 2}, {2, 2}, {2, 3}, {3, 3}, {2, 5}, {4, 4},
1385  {1, 8}, {5, 5}, {1, 10}, {6, 6}, {4, 9}, {7, 7},
1386  {3, 12}, {8, 8}, {8, 9}, {2, 16}, {1, 18}, {10, 10},
1387  {6, 15}, {11, 11}, {3, 20}, {12, 12}, {8, 17}, {13, 13},
1388  {3, 24}, {14, 14}, {4, 25}, {15, 15}, {3, 28}, {16, 16},
1389  };
1390 
1391  InnerChain[Exp] = B.CreateFMul(getPow(InnerChain, AddChain[Exp][0], B),
1392  getPow(InnerChain, AddChain[Exp][1], B));
1393  return InnerChain[Exp];
1394 }
1395 
1396 // Return a properly extended 32-bit integer if the operation is an itofp.
1398  if (isa<SIToFPInst>(I2F) || isa<UIToFPInst>(I2F)) {
1399  Value *Op = cast<Instruction>(I2F)->getOperand(0);
1400  // Make sure that the exponent fits inside an int32_t,
1401  // thus avoiding any range issues that FP has not.
1402  unsigned BitWidth = Op->getType()->getPrimitiveSizeInBits();
1403  if (BitWidth < 32 ||
1404  (BitWidth == 32 && isa<SIToFPInst>(I2F)))
1405  return isa<SIToFPInst>(I2F) ? B.CreateSExt(Op, B.getInt32Ty())
1406  : B.CreateZExt(Op, B.getInt32Ty());
1407  }
1408 
1409  return nullptr;
1410 }
1411 
1412 /// Use exp{,2}(x * y) for pow(exp{,2}(x), y);
1413 /// ldexp(1.0, x) for pow(2.0, itofp(x)); exp2(n * x) for pow(2.0 ** n, x);
1414 /// exp10(x) for pow(10.0, x); exp2(log2(n) * x) for pow(n, x).
1415 Value *LibCallSimplifier::replacePowWithExp(CallInst *Pow, IRBuilder<> &B) {
1416  Value *Base = Pow->getArgOperand(0), *Expo = Pow->getArgOperand(1);
1418  Module *Mod = Pow->getModule();
1419  Type *Ty = Pow->getType();
1420  bool Ignored;
1421 
1422  // Evaluate special cases related to a nested function as the base.
1423 
1424  // pow(exp(x), y) -> exp(x * y)
1425  // pow(exp2(x), y) -> exp2(x * y)
1426  // If exp{,2}() is used only once, it is better to fold two transcendental
1427  // math functions into one. If used again, exp{,2}() would still have to be
1428  // called with the original argument, then keep both original transcendental
1429  // functions. However, this transformation is only safe with fully relaxed
1430  // math semantics, since, besides rounding differences, it changes overflow
1431  // and underflow behavior quite dramatically. For example:
1432  // pow(exp(1000), 0.001) = pow(inf, 0.001) = inf
1433  // Whereas:
1434  // exp(1000 * 0.001) = exp(1)
1435  // TODO: Loosen the requirement for fully relaxed math semantics.
1436  // TODO: Handle exp10() when more targets have it available.
1437  CallInst *BaseFn = dyn_cast<CallInst>(Base);
1438  if (BaseFn && BaseFn->hasOneUse() && BaseFn->isFast() && Pow->isFast()) {
1439  LibFunc LibFn;
1440 
1441  Function *CalleeFn = BaseFn->getCalledFunction();
1442  if (CalleeFn &&
1443  TLI->getLibFunc(CalleeFn->getName(), LibFn) && TLI->has(LibFn)) {
1444  StringRef ExpName;
1445  Intrinsic::ID ID;
1446  Value *ExpFn;
1447  LibFunc LibFnFloat;
1448  LibFunc LibFnDouble;
1449  LibFunc LibFnLongDouble;
1450 
1451  switch (LibFn) {
1452  default:
1453  return nullptr;
1454  case LibFunc_expf: case LibFunc_exp: case LibFunc_expl:
1455  ExpName = TLI->getName(LibFunc_exp);
1456  ID = Intrinsic::exp;
1457  LibFnFloat = LibFunc_expf;
1458  LibFnDouble = LibFunc_exp;
1459  LibFnLongDouble = LibFunc_expl;
1460  break;
1461  case LibFunc_exp2f: case LibFunc_exp2: case LibFunc_exp2l:
1462  ExpName = TLI->getName(LibFunc_exp2);
1463  ID = Intrinsic::exp2;
1464  LibFnFloat = LibFunc_exp2f;
1465  LibFnDouble = LibFunc_exp2;
1466  LibFnLongDouble = LibFunc_exp2l;
1467  break;
1468  }
1469 
1470  // Create new exp{,2}() with the product as its argument.
1471  Value *FMul = B.CreateFMul(BaseFn->getArgOperand(0), Expo, "mul");
1472  ExpFn = BaseFn->doesNotAccessMemory()
1473  ? B.CreateCall(Intrinsic::getDeclaration(Mod, ID, Ty),
1474  FMul, ExpName)
1475  : emitUnaryFloatFnCall(FMul, TLI, LibFnDouble, LibFnFloat,
1476  LibFnLongDouble, B,
1477  BaseFn->getAttributes());
1478 
1479  // Since the new exp{,2}() is different from the original one, dead code
1480  // elimination cannot be trusted to remove it, since it may have side
1481  // effects (e.g., errno). When the only consumer for the original
1482  // exp{,2}() is pow(), then it has to be explicitly erased.
1483  substituteInParent(BaseFn, ExpFn);
1484  return ExpFn;
1485  }
1486  }
1487 
1488  // Evaluate special cases related to a constant base.
1489 
1490  const APFloat *BaseF;
1491  if (!match(Pow->getArgOperand(0), m_APFloat(BaseF)))
1492  return nullptr;
1493 
1494  // pow(2.0, itofp(x)) -> ldexp(1.0, x)
1495  if (match(Base, m_SpecificFP(2.0)) &&
1496  (isa<SIToFPInst>(Expo) || isa<UIToFPInst>(Expo)) &&
1497  hasFloatFn(TLI, Ty, LibFunc_ldexp, LibFunc_ldexpf, LibFunc_ldexpl)) {
1498  if (Value *ExpoI = getIntToFPVal(Expo, B))
1499  return emitBinaryFloatFnCall(ConstantFP::get(Ty, 1.0), ExpoI, TLI,
1500  LibFunc_ldexp, LibFunc_ldexpf, LibFunc_ldexpl,
1501  B, Attrs);
1502  }
1503 
1504  // pow(2.0 ** n, x) -> exp2(n * x)
1505  if (hasFloatFn(TLI, Ty, LibFunc_exp2, LibFunc_exp2f, LibFunc_exp2l)) {
1506  APFloat BaseR = APFloat(1.0);
1507  BaseR.convert(BaseF->getSemantics(), APFloat::rmTowardZero, &Ignored);
1508  BaseR = BaseR / *BaseF;
1509  bool IsInteger = BaseF->isInteger(), IsReciprocal = BaseR.isInteger();
1510  const APFloat *NF = IsReciprocal ? &BaseR : BaseF;
1511  APSInt NI(64, false);
1512  if ((IsInteger || IsReciprocal) &&
1513  NF->convertToInteger(NI, APFloat::rmTowardZero, &Ignored) ==
1514  APFloat::opOK &&
1515  NI > 1 && NI.isPowerOf2()) {
1516  double N = NI.logBase2() * (IsReciprocal ? -1.0 : 1.0);
1517  Value *FMul = B.CreateFMul(Expo, ConstantFP::get(Ty, N), "mul");
1518  if (Pow->doesNotAccessMemory())
1519  return B.CreateCall(Intrinsic::getDeclaration(Mod, Intrinsic::exp2, Ty),
1520  FMul, "exp2");
1521  else
1522  return emitUnaryFloatFnCall(FMul, TLI, LibFunc_exp2, LibFunc_exp2f,
1523  LibFunc_exp2l, B, Attrs);
1524  }
1525  }
1526 
1527  // pow(10.0, x) -> exp10(x)
1528  // TODO: There is no exp10() intrinsic yet, but some day there shall be one.
1529  if (match(Base, m_SpecificFP(10.0)) &&
1530  hasFloatFn(TLI, Ty, LibFunc_exp10, LibFunc_exp10f, LibFunc_exp10l))
1531  return emitUnaryFloatFnCall(Expo, TLI, LibFunc_exp10, LibFunc_exp10f,
1532  LibFunc_exp10l, B, Attrs);
1533 
1534  // pow(n, x) -> exp2(log2(n) * x)
1535  if (Pow->hasOneUse() && Pow->hasApproxFunc() && Pow->hasNoNaNs() &&
1536  Pow->hasNoInfs() && BaseF->isNormal() && !BaseF->isNegative()) {
1537  Value *Log = nullptr;
1538  if (Ty->isFloatTy())
1539  Log = ConstantFP::get(Ty, std::log2(BaseF->convertToFloat()));
1540  else if (Ty->isDoubleTy())
1541  Log = ConstantFP::get(Ty, std::log2(BaseF->convertToDouble()));
1542 
1543  if (Log) {
1544  Value *FMul = B.CreateFMul(Log, Expo, "mul");
1545  if (Pow->doesNotAccessMemory())
1546  return B.CreateCall(Intrinsic::getDeclaration(Mod, Intrinsic::exp2, Ty),
1547  FMul, "exp2");
1548  else if (hasFloatFn(TLI, Ty, LibFunc_exp2, LibFunc_exp2f, LibFunc_exp2l))
1549  return emitUnaryFloatFnCall(FMul, TLI, LibFunc_exp2, LibFunc_exp2f,
1550  LibFunc_exp2l, B, Attrs);
1551  }
1552  }
1553 
1554  return nullptr;
1555 }
1556 
1557 static Value *getSqrtCall(Value *V, AttributeList Attrs, bool NoErrno,
1558  Module *M, IRBuilder<> &B,
1559  const TargetLibraryInfo *TLI) {
1560  // If errno is never set, then use the intrinsic for sqrt().
1561  if (NoErrno) {
1562  Function *SqrtFn =
1563  Intrinsic::getDeclaration(M, Intrinsic::sqrt, V->getType());
1564  return B.CreateCall(SqrtFn, V, "sqrt");
1565  }
1566 
1567  // Otherwise, use the libcall for sqrt().
1568  if (hasFloatFn(TLI, V->getType(), LibFunc_sqrt, LibFunc_sqrtf, LibFunc_sqrtl))
1569  // TODO: We also should check that the target can in fact lower the sqrt()
1570  // libcall. We currently have no way to ask this question, so we ask if
1571  // the target has a sqrt() libcall, which is not exactly the same.
1572  return emitUnaryFloatFnCall(V, TLI, LibFunc_sqrt, LibFunc_sqrtf,
1573  LibFunc_sqrtl, B, Attrs);
1574 
1575  return nullptr;
1576 }
1577 
1578 /// Use square root in place of pow(x, +/-0.5).
1579 Value *LibCallSimplifier::replacePowWithSqrt(CallInst *Pow, IRBuilder<> &B) {
1580  Value *Sqrt, *Base = Pow->getArgOperand(0), *Expo = Pow->getArgOperand(1);
1582  Module *Mod = Pow->getModule();
1583  Type *Ty = Pow->getType();
1584 
1585  const APFloat *ExpoF;
1586  if (!match(Expo, m_APFloat(ExpoF)) ||
1587  (!ExpoF->isExactlyValue(0.5) && !ExpoF->isExactlyValue(-0.5)))
1588  return nullptr;
1589 
1590  Sqrt = getSqrtCall(Base, Attrs, Pow->doesNotAccessMemory(), Mod, B, TLI);
1591  if (!Sqrt)
1592  return nullptr;
1593 
1594  // Handle signed zero base by expanding to fabs(sqrt(x)).
1595  if (!Pow->hasNoSignedZeros()) {
1596  Function *FAbsFn = Intrinsic::getDeclaration(Mod, Intrinsic::fabs, Ty);
1597  Sqrt = B.CreateCall(FAbsFn, Sqrt, "abs");
1598  }
1599 
1600  // Handle non finite base by expanding to
1601  // (x == -infinity ? +infinity : sqrt(x)).
1602  if (!Pow->hasNoInfs()) {
1603  Value *PosInf = ConstantFP::getInfinity(Ty),
1604  *NegInf = ConstantFP::getInfinity(Ty, true);
1605  Value *FCmp = B.CreateFCmpOEQ(Base, NegInf, "isinf");
1606  Sqrt = B.CreateSelect(FCmp, PosInf, Sqrt);
1607  }
1608 
1609  // If the exponent is negative, then get the reciprocal.
1610  if (ExpoF->isNegative())
1611  Sqrt = B.CreateFDiv(ConstantFP::get(Ty, 1.0), Sqrt, "reciprocal");
1612 
1613  return Sqrt;
1614 }
1615 
1617  IRBuilder<> &B) {
1618  Value *Args[] = {Base, Expo};
1619  Function *F = Intrinsic::getDeclaration(M, Intrinsic::powi, Base->getType());
1620  return B.CreateCall(F, Args);
1621 }
1622 
1623 Value *LibCallSimplifier::optimizePow(CallInst *Pow, IRBuilder<> &B) {
1624  Value *Base = Pow->getArgOperand(0);
1625  Value *Expo = Pow->getArgOperand(1);
1626  Function *Callee = Pow->getCalledFunction();
1627  StringRef Name = Callee->getName();
1628  Type *Ty = Pow->getType();
1629  Module *M = Pow->getModule();
1630  Value *Shrunk = nullptr;
1631  bool AllowApprox = Pow->hasApproxFunc();
1632  bool Ignored;
1633 
1634  // Bail out if simplifying libcalls to pow() is disabled.
1635  if (!hasFloatFn(TLI, Ty, LibFunc_pow, LibFunc_powf, LibFunc_powl))
1636  return nullptr;
1637 
1638  // Propagate the math semantics from the call to any created instructions.
1641 
1642  // Shrink pow() to powf() if the arguments are single precision,
1643  // unless the result is expected to be double precision.
1644  if (UnsafeFPShrink && Name == TLI->getName(LibFunc_pow) &&
1645  hasFloatVersion(Name))
1646  Shrunk = optimizeBinaryDoubleFP(Pow, B, true);
1647 
1648  // Evaluate special cases related to the base.
1649 
1650  // pow(1.0, x) -> 1.0
1651  if (match(Base, m_FPOne()))
1652  return Base;
1653 
1654  if (Value *Exp = replacePowWithExp(Pow, B))
1655  return Exp;
1656 
1657  // Evaluate special cases related to the exponent.
1658 
1659  // pow(x, -1.0) -> 1.0 / x
1660  if (match(Expo, m_SpecificFP(-1.0)))
1661  return B.CreateFDiv(ConstantFP::get(Ty, 1.0), Base, "reciprocal");
1662 
1663  // pow(x, +/-0.0) -> 1.0
1664  if (match(Expo, m_AnyZeroFP()))
1665  return ConstantFP::get(Ty, 1.0);
1666 
1667  // pow(x, 1.0) -> x
1668  if (match(Expo, m_FPOne()))
1669  return Base;
1670 
1671  // pow(x, 2.0) -> x * x
1672  if (match(Expo, m_SpecificFP(2.0)))
1673  return B.CreateFMul(Base, Base, "square");
1674 
1675  if (Value *Sqrt = replacePowWithSqrt(Pow, B))
1676  return Sqrt;
1677 
1678  // pow(x, n) -> x * x * x * ...
1679  const APFloat *ExpoF;
1680  if (AllowApprox && match(Expo, m_APFloat(ExpoF))) {
1681  // We limit to a max of 7 multiplications, thus the maximum exponent is 32.
1682  // If the exponent is an integer+0.5 we generate a call to sqrt and an
1683  // additional fmul.
1684  // TODO: This whole transformation should be backend specific (e.g. some
1685  // backends might prefer libcalls or the limit for the exponent might
1686  // be different) and it should also consider optimizing for size.
1687  APFloat LimF(ExpoF->getSemantics(), 33.0),
1688  ExpoA(abs(*ExpoF));
1689  if (ExpoA.compare(LimF) == APFloat::cmpLessThan) {
1690  // This transformation applies to integer or integer+0.5 exponents only.
1691  // For integer+0.5, we create a sqrt(Base) call.
1692  Value *Sqrt = nullptr;
1693  if (!ExpoA.isInteger()) {
1694  APFloat Expo2 = ExpoA;
1695  // To check if ExpoA is an integer + 0.5, we add it to itself. If there
1696  // is no floating point exception and the result is an integer, then
1697  // ExpoA == integer + 0.5
1698  if (Expo2.add(ExpoA, APFloat::rmNearestTiesToEven) != APFloat::opOK)
1699  return nullptr;
1700 
1701  if (!Expo2.isInteger())
1702  return nullptr;
1703 
1704  Sqrt = getSqrtCall(Base, Pow->getCalledFunction()->getAttributes(),
1705  Pow->doesNotAccessMemory(), M, B, TLI);
1706  }
1707 
1708  // We will memoize intermediate products of the Addition Chain.
1709  Value *InnerChain[33] = {nullptr};
1710  InnerChain[1] = Base;
1711  InnerChain[2] = B.CreateFMul(Base, Base, "square");
1712 
1713  // We cannot readily convert a non-double type (like float) to a double.
1714  // So we first convert it to something which could be converted to double.
1716  Value *FMul = getPow(InnerChain, ExpoA.convertToDouble(), B);
1717 
1718  // Expand pow(x, y+0.5) to pow(x, y) * sqrt(x).
1719  if (Sqrt)
1720  FMul = B.CreateFMul(FMul, Sqrt);
1721 
1722  // If the exponent is negative, then get the reciprocal.
1723  if (ExpoF->isNegative())
1724  FMul = B.CreateFDiv(ConstantFP::get(Ty, 1.0), FMul, "reciprocal");
1725 
1726  return FMul;
1727  }
1728 
1729  APSInt IntExpo(32, /*isUnsigned=*/false);
1730  // powf(x, n) -> powi(x, n) if n is a constant signed integer value
1731  if (ExpoF->isInteger() &&
1732  ExpoF->convertToInteger(IntExpo, APFloat::rmTowardZero, &Ignored) ==
1733  APFloat::opOK) {
1735  Base, ConstantInt::get(B.getInt32Ty(), IntExpo), M, B);
1736  }
1737  }
1738 
1739  // powf(x, itofp(y)) -> powi(x, y)
1740  if (AllowApprox && (isa<SIToFPInst>(Expo) || isa<UIToFPInst>(Expo))) {
1741  if (Value *ExpoI = getIntToFPVal(Expo, B))
1742  return createPowWithIntegerExponent(Base, ExpoI, M, B);
1743  }
1744 
1745  return Shrunk;
1746 }
1747 
1748 Value *LibCallSimplifier::optimizeExp2(CallInst *CI, IRBuilder<> &B) {
1750  StringRef Name = Callee->getName();
1751  Value *Ret = nullptr;
1752  if (UnsafeFPShrink && Name == TLI->getName(LibFunc_exp2) &&
1753  hasFloatVersion(Name))
1754  Ret = optimizeUnaryDoubleFP(CI, B, true);
1755 
1756  Type *Ty = CI->getType();
1757  Value *Op = CI->getArgOperand(0);
1758 
1759  // Turn exp2(sitofp(x)) -> ldexp(1.0, sext(x)) if sizeof(x) <= 32
1760  // Turn exp2(uitofp(x)) -> ldexp(1.0, zext(x)) if sizeof(x) < 32
1761  if ((isa<SIToFPInst>(Op) || isa<UIToFPInst>(Op)) &&
1762  hasFloatFn(TLI, Ty, LibFunc_ldexp, LibFunc_ldexpf, LibFunc_ldexpl)) {
1763  if (Value *Exp = getIntToFPVal(Op, B))
1764  return emitBinaryFloatFnCall(ConstantFP::get(Ty, 1.0), Exp, TLI,
1765  LibFunc_ldexp, LibFunc_ldexpf, LibFunc_ldexpl,
1766  B, CI->getCalledFunction()->getAttributes());
1767  }
1768 
1769  return Ret;
1770 }
1771 
1772 Value *LibCallSimplifier::optimizeFMinFMax(CallInst *CI, IRBuilder<> &B) {
1773  // If we can shrink the call to a float function rather than a double
1774  // function, do that first.
1776  StringRef Name = Callee->getName();
1777  if ((Name == "fmin" || Name == "fmax") && hasFloatVersion(Name))
1778  if (Value *Ret = optimizeBinaryDoubleFP(CI, B))
1779  return Ret;
1780 
1781  // The LLVM intrinsics minnum/maxnum correspond to fmin/fmax. Canonicalize to
1782  // the intrinsics for improved optimization (for example, vectorization).
1783  // No-signed-zeros is implied by the definitions of fmax/fmin themselves.
1784  // From the C standard draft WG14/N1256:
1785  // "Ideally, fmax would be sensitive to the sign of zero, for example
1786  // fmax(-0.0, +0.0) would return +0; however, implementation in software
1787  // might be impractical."
1789  FastMathFlags FMF = CI->getFastMathFlags();
1790  FMF.setNoSignedZeros();
1791  B.setFastMathFlags(FMF);
1792 
1793  Intrinsic::ID IID = Callee->getName().startswith("fmin") ? Intrinsic::minnum
1795  Function *F = Intrinsic::getDeclaration(CI->getModule(), IID, CI->getType());
1796  return B.CreateCall(F, { CI->getArgOperand(0), CI->getArgOperand(1) });
1797 }
1798 
1799 Value *LibCallSimplifier::optimizeLog(CallInst *CI, IRBuilder<> &B) {
1801  Value *Ret = nullptr;
1802  StringRef Name = Callee->getName();
1803  if (UnsafeFPShrink && hasFloatVersion(Name))
1804  Ret = optimizeUnaryDoubleFP(CI, B, true);
1805 
1806  if (!CI->isFast())
1807  return Ret;
1808  Value *Op1 = CI->getArgOperand(0);
1809  auto *OpC = dyn_cast<CallInst>(Op1);
1810 
1811  // The earlier call must also be 'fast' in order to do these transforms.
1812  if (!OpC || !OpC->isFast())
1813  return Ret;
1814 
1815  // log(pow(x,y)) -> y*log(x)
1816  // This is only applicable to log, log2, log10.
1817  if (Name != "log" && Name != "log2" && Name != "log10")
1818  return Ret;
1819 
1821  FastMathFlags FMF;
1822  FMF.setFast();
1823  B.setFastMathFlags(FMF);
1824 
1825  LibFunc Func;
1826  Function *F = OpC->getCalledFunction();
1827  if (F && ((TLI->getLibFunc(F->getName(), Func) && TLI->has(Func) &&
1828  Func == LibFunc_pow) || F->getIntrinsicID() == Intrinsic::pow))
1829  return B.CreateFMul(OpC->getArgOperand(1),
1830  emitUnaryFloatFnCall(OpC->getOperand(0), Callee->getName(), B,
1831  Callee->getAttributes()), "mul");
1832 
1833  // log(exp2(y)) -> y*log(2)
1834  if (F && Name == "log" && TLI->getLibFunc(F->getName(), Func) &&
1835  TLI->has(Func) && Func == LibFunc_exp2)
1836  return B.CreateFMul(
1837  OpC->getArgOperand(0),
1839  Callee->getName(), B, Callee->getAttributes()),
1840  "logmul");
1841  return Ret;
1842 }
1843 
1844 Value *LibCallSimplifier::optimizeSqrt(CallInst *CI, IRBuilder<> &B) {
1846  Value *Ret = nullptr;
1847  // TODO: Once we have a way (other than checking for the existince of the
1848  // libcall) to tell whether our target can lower @llvm.sqrt, relax the
1849  // condition below.
1850  if (TLI->has(LibFunc_sqrtf) && (Callee->getName() == "sqrt" ||
1851  Callee->getIntrinsicID() == Intrinsic::sqrt))
1852  Ret = optimizeUnaryDoubleFP(CI, B, true);
1853 
1854  if (!CI->isFast())
1855  return Ret;
1856 
1858  if (!I || I->getOpcode() != Instruction::FMul || !I->isFast())
1859  return Ret;
1860 
1861  // We're looking for a repeated factor in a multiplication tree,
1862  // so we can do this fold: sqrt(x * x) -> fabs(x);
1863  // or this fold: sqrt((x * x) * y) -> fabs(x) * sqrt(y).
1864  Value *Op0 = I->getOperand(0);
1865  Value *Op1 = I->getOperand(1);
1866  Value *RepeatOp = nullptr;
1867  Value *OtherOp = nullptr;
1868  if (Op0 == Op1) {
1869  // Simple match: the operands of the multiply are identical.
1870  RepeatOp = Op0;
1871  } else {
1872  // Look for a more complicated pattern: one of the operands is itself
1873  // a multiply, so search for a common factor in that multiply.
1874  // Note: We don't bother looking any deeper than this first level or for
1875  // variations of this pattern because instcombine's visitFMUL and/or the
1876  // reassociation pass should give us this form.
1877  Value *OtherMul0, *OtherMul1;
1878  if (match(Op0, m_FMul(m_Value(OtherMul0), m_Value(OtherMul1)))) {
1879  // Pattern: sqrt((x * y) * z)
1880  if (OtherMul0 == OtherMul1 && cast<Instruction>(Op0)->isFast()) {
1881  // Matched: sqrt((x * x) * z)
1882  RepeatOp = OtherMul0;
1883  OtherOp = Op1;
1884  }
1885  }
1886  }
1887  if (!RepeatOp)
1888  return Ret;
1889 
1890  // Fast math flags for any created instructions should match the sqrt
1891  // and multiply.
1894 
1895  // If we found a repeated factor, hoist it out of the square root and
1896  // replace it with the fabs of that factor.
1897  Module *M = Callee->getParent();
1898  Type *ArgType = I->getType();
1899  Function *Fabs = Intrinsic::getDeclaration(M, Intrinsic::fabs, ArgType);
1900  Value *FabsCall = B.CreateCall(Fabs, RepeatOp, "fabs");
1901  if (OtherOp) {
1902  // If we found a non-repeated factor, we still need to get its square
1903  // root. We then multiply that by the value that was simplified out
1904  // of the square root calculation.
1905  Function *Sqrt = Intrinsic::getDeclaration(M, Intrinsic::sqrt, ArgType);
1906  Value *SqrtCall = B.CreateCall(Sqrt, OtherOp, "sqrt");
1907  return B.CreateFMul(FabsCall, SqrtCall);
1908  }
1909  return FabsCall;
1910 }
1911 
1912 // TODO: Generalize to handle any trig function and its inverse.
1913 Value *LibCallSimplifier::optimizeTan(CallInst *CI, IRBuilder<> &B) {
1915  Value *Ret = nullptr;
1916  StringRef Name = Callee->getName();
1917  if (UnsafeFPShrink && Name == "tan" && hasFloatVersion(Name))
1918  Ret = optimizeUnaryDoubleFP(CI, B, true);
1919 
1920  Value *Op1 = CI->getArgOperand(0);
1921  auto *OpC = dyn_cast<CallInst>(Op1);
1922  if (!OpC)
1923  return Ret;
1924 
1925  // Both calls must be 'fast' in order to remove them.
1926  if (!CI->isFast() || !OpC->isFast())
1927  return Ret;
1928 
1929  // tan(atan(x)) -> x
1930  // tanf(atanf(x)) -> x
1931  // tanl(atanl(x)) -> x
1932  LibFunc Func;
1933  Function *F = OpC->getCalledFunction();
1934  if (F && TLI->getLibFunc(F->getName(), Func) && TLI->has(Func) &&
1935  ((Func == LibFunc_atan && Callee->getName() == "tan") ||
1936  (Func == LibFunc_atanf && Callee->getName() == "tanf") ||
1937  (Func == LibFunc_atanl && Callee->getName() == "tanl")))
1938  Ret = OpC->getArgOperand(0);
1939  return Ret;
1940 }
1941 
1942 static bool isTrigLibCall(CallInst *CI) {
1943  // We can only hope to do anything useful if we can ignore things like errno
1944  // and floating-point exceptions.
1945  // We already checked the prototype.
1946  return CI->hasFnAttr(Attribute::NoUnwind) &&
1947  CI->hasFnAttr(Attribute::ReadNone);
1948 }
1949 
1950 static void insertSinCosCall(IRBuilder<> &B, Function *OrigCallee, Value *Arg,
1951  bool UseFloat, Value *&Sin, Value *&Cos,
1952  Value *&SinCos) {
1953  Type *ArgTy = Arg->getType();
1954  Type *ResTy;
1955  StringRef Name;
1956 
1957  Triple T(OrigCallee->getParent()->getTargetTriple());
1958  if (UseFloat) {
1959  Name = "__sincospif_stret";
1960 
1961  assert(T.getArch() != Triple::x86 && "x86 messy and unsupported for now");
1962  // x86_64 can't use {float, float} since that would be returned in both
1963  // xmm0 and xmm1, which isn't what a real struct would do.
1964  ResTy = T.getArch() == Triple::x86_64
1965  ? static_cast<Type *>(VectorType::get(ArgTy, 2))
1966  : static_cast<Type *>(StructType::get(ArgTy, ArgTy));
1967  } else {
1968  Name = "__sincospi_stret";
1969  ResTy = StructType::get(ArgTy, ArgTy);
1970  }
1971 
1972  Module *M = OrigCallee->getParent();
1974  M->getOrInsertFunction(Name, OrigCallee->getAttributes(), ResTy, ArgTy);
1975 
1976  if (Instruction *ArgInst = dyn_cast<Instruction>(Arg)) {
1977  // If the argument is an instruction, it must dominate all uses so put our
1978  // sincos call there.
1979  B.SetInsertPoint(ArgInst->getParent(), ++ArgInst->getIterator());
1980  } else {
1981  // Otherwise (e.g. for a constant) the beginning of the function is as
1982  // good a place as any.
1983  BasicBlock &EntryBB = B.GetInsertBlock()->getParent()->getEntryBlock();
1984  B.SetInsertPoint(&EntryBB, EntryBB.begin());
1985  }
1986 
1987  SinCos = B.CreateCall(Callee, Arg, "sincospi");
1988 
1989  if (SinCos->getType()->isStructTy()) {
1990  Sin = B.CreateExtractValue(SinCos, 0, "sinpi");
1991  Cos = B.CreateExtractValue(SinCos, 1, "cospi");
1992  } else {
1993  Sin = B.CreateExtractElement(SinCos, ConstantInt::get(B.getInt32Ty(), 0),
1994  "sinpi");
1995  Cos = B.CreateExtractElement(SinCos, ConstantInt::get(B.getInt32Ty(), 1),
1996  "cospi");
1997  }
1998 }
1999 
2000 Value *LibCallSimplifier::optimizeSinCosPi(CallInst *CI, IRBuilder<> &B) {
2001  // Make sure the prototype is as expected, otherwise the rest of the
2002  // function is probably invalid and likely to abort.
2003  if (!isTrigLibCall(CI))
2004  return nullptr;
2005 
2006  Value *Arg = CI->getArgOperand(0);
2007  SmallVector<CallInst *, 1> SinCalls;
2008  SmallVector<CallInst *, 1> CosCalls;
2009  SmallVector<CallInst *, 1> SinCosCalls;
2010 
2011  bool IsFloat = Arg->getType()->isFloatTy();
2012 
2013  // Look for all compatible sinpi, cospi and sincospi calls with the same
2014  // argument. If there are enough (in some sense) we can make the
2015  // substitution.
2016  Function *F = CI->getFunction();
2017  for (User *U : Arg->users())
2018  classifyArgUse(U, F, IsFloat, SinCalls, CosCalls, SinCosCalls);
2019 
2020  // It's only worthwhile if both sinpi and cospi are actually used.
2021  if (SinCosCalls.empty() && (SinCalls.empty() || CosCalls.empty()))
2022  return nullptr;
2023 
2024  Value *Sin, *Cos, *SinCos;
2025  insertSinCosCall(B, CI->getCalledFunction(), Arg, IsFloat, Sin, Cos, SinCos);
2026 
2027  auto replaceTrigInsts = [this](SmallVectorImpl<CallInst *> &Calls,
2028  Value *Res) {
2029  for (CallInst *C : Calls)
2030  replaceAllUsesWith(C, Res);
2031  };
2032 
2033  replaceTrigInsts(SinCalls, Sin);
2034  replaceTrigInsts(CosCalls, Cos);
2035  replaceTrigInsts(SinCosCalls, SinCos);
2036 
2037  return nullptr;
2038 }
2039 
2040 void LibCallSimplifier::classifyArgUse(
2041  Value *Val, Function *F, bool IsFloat,
2042  SmallVectorImpl<CallInst *> &SinCalls,
2043  SmallVectorImpl<CallInst *> &CosCalls,
2044  SmallVectorImpl<CallInst *> &SinCosCalls) {
2045  CallInst *CI = dyn_cast<CallInst>(Val);
2046 
2047  if (!CI)
2048  return;
2049 
2050  // Don't consider calls in other functions.
2051  if (CI->getFunction() != F)
2052  return;
2053 
2055  LibFunc Func;
2056  if (!Callee || !TLI->getLibFunc(*Callee, Func) || !TLI->has(Func) ||
2057  !isTrigLibCall(CI))
2058  return;
2059 
2060  if (IsFloat) {
2061  if (Func == LibFunc_sinpif)
2062  SinCalls.push_back(CI);
2063  else if (Func == LibFunc_cospif)
2064  CosCalls.push_back(CI);
2065  else if (Func == LibFunc_sincospif_stret)
2066  SinCosCalls.push_back(CI);
2067  } else {
2068  if (Func == LibFunc_sinpi)
2069  SinCalls.push_back(CI);
2070  else if (Func == LibFunc_cospi)
2071  CosCalls.push_back(CI);
2072  else if (Func == LibFunc_sincospi_stret)
2073  SinCosCalls.push_back(CI);
2074  }
2075 }
2076 
2077 //===----------------------------------------------------------------------===//
2078 // Integer Library Call Optimizations
2079 //===----------------------------------------------------------------------===//
2080 
2081 Value *LibCallSimplifier::optimizeFFS(CallInst *CI, IRBuilder<> &B) {
2082  // ffs(x) -> x != 0 ? (i32)llvm.cttz(x)+1 : 0
2083  Value *Op = CI->getArgOperand(0);
2084  Type *ArgType = Op->getType();
2086  Intrinsic::cttz, ArgType);
2087  Value *V = B.CreateCall(F, {Op, B.getTrue()}, "cttz");
2088  V = B.CreateAdd(V, ConstantInt::get(V->getType(), 1));
2089  V = B.CreateIntCast(V, B.getInt32Ty(), false);
2090 
2091  Value *Cond = B.CreateICmpNE(Op, Constant::getNullValue(ArgType));
2092  return B.CreateSelect(Cond, V, B.getInt32(0));
2093 }
2094 
2095 Value *LibCallSimplifier::optimizeFls(CallInst *CI, IRBuilder<> &B) {
2096  // fls(x) -> (i32)(sizeInBits(x) - llvm.ctlz(x, false))
2097  Value *Op = CI->getArgOperand(0);
2098  Type *ArgType = Op->getType();
2100  Intrinsic::ctlz, ArgType);
2101  Value *V = B.CreateCall(F, {Op, B.getFalse()}, "ctlz");
2102  V = B.CreateSub(ConstantInt::get(V->getType(), ArgType->getIntegerBitWidth()),
2103  V);
2104  return B.CreateIntCast(V, CI->getType(), false);
2105 }
2106 
2107 Value *LibCallSimplifier::optimizeAbs(CallInst *CI, IRBuilder<> &B) {
2108  // abs(x) -> x <s 0 ? -x : x
2109  // The negation has 'nsw' because abs of INT_MIN is undefined.
2110  Value *X = CI->getArgOperand(0);
2111  Value *IsNeg = B.CreateICmpSLT(X, Constant::getNullValue(X->getType()));
2112  Value *NegX = B.CreateNSWNeg(X, "neg");
2113  return B.CreateSelect(IsNeg, NegX, X);
2114 }
2115 
2116 Value *LibCallSimplifier::optimizeIsDigit(CallInst *CI, IRBuilder<> &B) {
2117  // isdigit(c) -> (c-'0') <u 10
2118  Value *Op = CI->getArgOperand(0);
2119  Op = B.CreateSub(Op, B.getInt32('0'), "isdigittmp");
2120  Op = B.CreateICmpULT(Op, B.getInt32(10), "isdigit");
2121  return B.CreateZExt(Op, CI->getType());
2122 }
2123 
2124 Value *LibCallSimplifier::optimizeIsAscii(CallInst *CI, IRBuilder<> &B) {
2125  // isascii(c) -> c <u 128
2126  Value *Op = CI->getArgOperand(0);
2127  Op = B.CreateICmpULT(Op, B.getInt32(128), "isascii");
2128  return B.CreateZExt(Op, CI->getType());
2129 }
2130 
2131 Value *LibCallSimplifier::optimizeToAscii(CallInst *CI, IRBuilder<> &B) {
2132  // toascii(c) -> c & 0x7f
2133  return B.CreateAnd(CI->getArgOperand(0),
2134  ConstantInt::get(CI->getType(), 0x7F));
2135 }
2136 
2137 Value *LibCallSimplifier::optimizeAtoi(CallInst *CI, IRBuilder<> &B) {
2138  StringRef Str;
2139  if (!getConstantStringInfo(CI->getArgOperand(0), Str))
2140  return nullptr;
2141 
2142  return convertStrToNumber(CI, Str, 10);
2143 }
2144 
2145 Value *LibCallSimplifier::optimizeStrtol(CallInst *CI, IRBuilder<> &B) {
2146  StringRef Str;
2147  if (!getConstantStringInfo(CI->getArgOperand(0), Str))
2148  return nullptr;
2149 
2150  if (!isa<ConstantPointerNull>(CI->getArgOperand(1)))
2151  return nullptr;
2152 
2153  if (ConstantInt *CInt = dyn_cast<ConstantInt>(CI->getArgOperand(2))) {
2154  return convertStrToNumber(CI, Str, CInt->getSExtValue());
2155  }
2156 
2157  return nullptr;
2158 }
2159 
2160 //===----------------------------------------------------------------------===//
2161 // Formatting and IO Library Call Optimizations
2162 //===----------------------------------------------------------------------===//
2163 
2164 static bool isReportingError(Function *Callee, CallInst *CI, int StreamArg);
2165 
2166 Value *LibCallSimplifier::optimizeErrorReporting(CallInst *CI, IRBuilder<> &B,
2167  int StreamArg) {
2168  Function *Callee = CI->getCalledFunction();
2169  // Error reporting calls should be cold, mark them as such.
2170  // This applies even to non-builtin calls: it is only a hint and applies to
2171  // functions that the frontend might not understand as builtins.
2172 
2173  // This heuristic was suggested in:
2174  // Improving Static Branch Prediction in a Compiler
2175  // Brian L. Deitrich, Ben-Chung Cheng, Wen-mei W. Hwu
2176  // Proceedings of PACT'98, Oct. 1998, IEEE
2177  if (!CI->hasFnAttr(Attribute::Cold) &&
2178  isReportingError(Callee, CI, StreamArg)) {
2180  }
2181 
2182  return nullptr;
2183 }
2184 
2185 static bool isReportingError(Function *Callee, CallInst *CI, int StreamArg) {
2186  if (!Callee || !Callee->isDeclaration())
2187  return false;
2188 
2189  if (StreamArg < 0)
2190  return true;
2191 
2192  // These functions might be considered cold, but only if their stream
2193  // argument is stderr.
2194 
2195  if (StreamArg >= (int)CI->getNumArgOperands())
2196  return false;
2197  LoadInst *LI = dyn_cast<LoadInst>(CI->getArgOperand(StreamArg));
2198  if (!LI)
2199  return false;
2201  if (!GV || !GV->isDeclaration())
2202  return false;
2203  return GV->getName() == "stderr";
2204 }
2205 
2206 Value *LibCallSimplifier::optimizePrintFString(CallInst *CI, IRBuilder<> &B) {
2207  // Check for a fixed format string.
2208  StringRef FormatStr;
2209  if (!getConstantStringInfo(CI->getArgOperand(0), FormatStr))
2210  return nullptr;
2211 
2212  // Empty format string -> noop.
2213  if (FormatStr.empty()) // Tolerate printf's declared void.
2214  return CI->use_empty() ? (Value *)CI : ConstantInt::get(CI->getType(), 0);
2215 
2216  // Do not do any of the following transformations if the printf return value
2217  // is used, in general the printf return value is not compatible with either
2218  // putchar() or puts().
2219  if (!CI->use_empty())
2220  return nullptr;
2221 
2222  // printf("x") -> putchar('x'), even for "%" and "%%".
2223  if (FormatStr.size() == 1 || FormatStr == "%%")
2224  return emitPutChar(B.getInt32(FormatStr[0]), B, TLI);
2225 
2226  // printf("%s", "a") --> putchar('a')
2227  if (FormatStr == "%s" && CI->getNumArgOperands() > 1) {
2228  StringRef ChrStr;
2229  if (!getConstantStringInfo(CI->getOperand(1), ChrStr))
2230  return nullptr;
2231  if (ChrStr.size() != 1)
2232  return nullptr;
2233  return emitPutChar(B.getInt32(ChrStr[0]), B, TLI);
2234  }
2235 
2236  // printf("foo\n") --> puts("foo")
2237  if (FormatStr[FormatStr.size() - 1] == '\n' &&
2238  FormatStr.find('%') == StringRef::npos) { // No format characters.
2239  // Create a string literal with no \n on it. We expect the constant merge
2240  // pass to be run after this pass, to merge duplicate strings.
2241  FormatStr = FormatStr.drop_back();
2242  Value *GV = B.CreateGlobalString(FormatStr, "str");
2243  return emitPutS(GV, B, TLI);
2244  }
2245 
2246  // Optimize specific format strings.
2247  // printf("%c", chr) --> putchar(chr)
2248  if (FormatStr == "%c" && CI->getNumArgOperands() > 1 &&
2249  CI->getArgOperand(1)->getType()->isIntegerTy())
2250  return emitPutChar(CI->getArgOperand(1), B, TLI);
2251 
2252  // printf("%s\n", str) --> puts(str)
2253  if (FormatStr == "%s\n" && CI->getNumArgOperands() > 1 &&
2254  CI->getArgOperand(1)->getType()->isPointerTy())
2255  return emitPutS(CI->getArgOperand(1), B, TLI);
2256  return nullptr;
2257 }
2258 
2259 Value *LibCallSimplifier::optimizePrintF(CallInst *CI, IRBuilder<> &B) {
2260 
2261  Function *Callee = CI->getCalledFunction();
2262  FunctionType *FT = Callee->getFunctionType();
2263  if (Value *V = optimizePrintFString(CI, B)) {
2264  return V;
2265  }
2266 
2267  // printf(format, ...) -> iprintf(format, ...) if no floating point
2268  // arguments.
2269  if (TLI->has(LibFunc_iprintf) && !callHasFloatingPointArgument(CI)) {
2270  Module *M = B.GetInsertBlock()->getParent()->getParent();
2271  FunctionCallee IPrintFFn =
2272  M->getOrInsertFunction("iprintf", FT, Callee->getAttributes());
2273  CallInst *New = cast<CallInst>(CI->clone());
2274  New->setCalledFunction(IPrintFFn);
2275  B.Insert(New);
2276  return New;
2277  }
2278 
2279  // printf(format, ...) -> __small_printf(format, ...) if no 128-bit floating point
2280  // arguments.
2281  if (TLI->has(LibFunc_small_printf) && !callHasFP128Argument(CI)) {
2282  Module *M = B.GetInsertBlock()->getParent()->getParent();
2283  auto SmallPrintFFn =
2284  M->getOrInsertFunction(TLI->getName(LibFunc_small_printf),
2285  FT, Callee->getAttributes());
2286  CallInst *New = cast<CallInst>(CI->clone());
2287  New->setCalledFunction(SmallPrintFFn);
2288  B.Insert(New);
2289  return New;
2290  }
2291 
2293  return nullptr;
2294 }
2295 
2296 Value *LibCallSimplifier::optimizeSPrintFString(CallInst *CI, IRBuilder<> &B) {
2297  // Check for a fixed format string.
2298  StringRef FormatStr;
2299  if (!getConstantStringInfo(CI->getArgOperand(1), FormatStr))
2300  return nullptr;
2301 
2302  // If we just have a format string (nothing else crazy) transform it.
2303  if (CI->getNumArgOperands() == 2) {
2304  // Make sure there's no % in the constant array. We could try to handle
2305  // %% -> % in the future if we cared.
2306  if (FormatStr.find('%') != StringRef::npos)
2307  return nullptr; // we found a format specifier, bail out.
2308 
2309  // sprintf(str, fmt) -> llvm.memcpy(align 1 str, align 1 fmt, strlen(fmt)+1)
2310  B.CreateMemCpy(CI->getArgOperand(0), 1, CI->getArgOperand(1), 1,
2311  ConstantInt::get(DL.getIntPtrType(CI->getContext()),
2312  FormatStr.size() + 1)); // Copy the null byte.
2313  return ConstantInt::get(CI->getType(), FormatStr.size());
2314  }
2315 
2316  // The remaining optimizations require the format string to be "%s" or "%c"
2317  // and have an extra operand.
2318  if (FormatStr.size() != 2 || FormatStr[0] != '%' ||
2319  CI->getNumArgOperands() < 3)
2320  return nullptr;
2321 
2322  // Decode the second character of the format string.
2323  if (FormatStr[1] == 'c') {
2324  // sprintf(dst, "%c", chr) --> *(i8*)dst = chr; *((i8*)dst+1) = 0
2325  if (!CI->getArgOperand(2)->getType()->isIntegerTy())
2326  return nullptr;
2327  Value *V = B.CreateTrunc(CI->getArgOperand(2), B.getInt8Ty(), "char");
2328  Value *Ptr = castToCStr(CI->getArgOperand(0), B);
2329  B.CreateStore(V, Ptr);
2330  Ptr = B.CreateGEP(B.getInt8Ty(), Ptr, B.getInt32(1), "nul");
2331  B.CreateStore(B.getInt8(0), Ptr);
2332 
2333  return ConstantInt::get(CI->getType(), 1);
2334  }
2335 
2336  if (FormatStr[1] == 's') {
2337  // sprintf(dest, "%s", str) -> llvm.memcpy(align 1 dest, align 1 str,
2338  // strlen(str)+1)
2339  if (!CI->getArgOperand(2)->getType()->isPointerTy())
2340  return nullptr;
2341 
2342  Value *Len = emitStrLen(CI->getArgOperand(2), B, DL, TLI);
2343  if (!Len)
2344  return nullptr;
2345  Value *IncLen =
2346  B.CreateAdd(Len, ConstantInt::get(Len->getType(), 1), "leninc");
2347  B.CreateMemCpy(CI->getArgOperand(0), 1, CI->getArgOperand(2), 1, IncLen);
2348 
2349  // The sprintf result is the unincremented number of bytes in the string.
2350  return B.CreateIntCast(Len, CI->getType(), false);
2351  }
2352  return nullptr;
2353 }
2354 
2355 Value *LibCallSimplifier::optimizeSPrintF(CallInst *CI, IRBuilder<> &B) {
2356  Function *Callee = CI->getCalledFunction();
2357  FunctionType *FT = Callee->getFunctionType();
2358  if (Value *V = optimizeSPrintFString(CI, B)) {
2359  return V;
2360  }
2361 
2362  // sprintf(str, format, ...) -> siprintf(str, format, ...) if no floating
2363  // point arguments.
2364  if (TLI->has(LibFunc_siprintf) && !callHasFloatingPointArgument(CI)) {
2365  Module *M = B.GetInsertBlock()->getParent()->getParent();
2366  FunctionCallee SIPrintFFn =
2367  M->getOrInsertFunction("siprintf", FT, Callee->getAttributes());
2368  CallInst *New = cast<CallInst>(CI->clone());
2369  New->setCalledFunction(SIPrintFFn);
2370  B.Insert(New);
2371  return New;
2372  }
2373 
2374  // sprintf(str, format, ...) -> __small_sprintf(str, format, ...) if no 128-bit
2375  // floating point arguments.
2376  if (TLI->has(LibFunc_small_sprintf) && !callHasFP128Argument(CI)) {
2377  Module *M = B.GetInsertBlock()->getParent()->getParent();
2378  auto SmallSPrintFFn =
2379  M->getOrInsertFunction(TLI->getName(LibFunc_small_sprintf),
2380  FT, Callee->getAttributes());
2381  CallInst *New = cast<CallInst>(CI->clone());
2382  New->setCalledFunction(SmallSPrintFFn);
2383  B.Insert(New);
2384  return New;
2385  }
2386 
2387  annotateNonNullBasedOnAccess(CI, {0, 1});
2388  return nullptr;
2389 }
2390 
2391 Value *LibCallSimplifier::optimizeSnPrintFString(CallInst *CI, IRBuilder<> &B) {
2392  // Check for size
2394  if (!Size)
2395  return nullptr;
2396 
2397  uint64_t N = Size->getZExtValue();
2398  // Check for a fixed format string.
2399  StringRef FormatStr;
2400  if (!getConstantStringInfo(CI->getArgOperand(2), FormatStr))
2401  return nullptr;
2402 
2403  // If we just have a format string (nothing else crazy) transform it.
2404  if (CI->getNumArgOperands() == 3) {
2405  // Make sure there's no % in the constant array. We could try to handle
2406  // %% -> % in the future if we cared.
2407  if (FormatStr.find('%') != StringRef::npos)
2408  return nullptr; // we found a format specifier, bail out.
2409 
2410  if (N == 0)
2411  return ConstantInt::get(CI->getType(), FormatStr.size());
2412  else if (N < FormatStr.size() + 1)
2413  return nullptr;
2414 
2415  // snprintf(dst, size, fmt) -> llvm.memcpy(align 1 dst, align 1 fmt,
2416  // strlen(fmt)+1)
2417  B.CreateMemCpy(
2418  CI->getArgOperand(0), 1, CI->getArgOperand(2), 1,
2419  ConstantInt::get(DL.getIntPtrType(CI->getContext()),
2420  FormatStr.size() + 1)); // Copy the null byte.
2421  return ConstantInt::get(CI->getType(), FormatStr.size());
2422  }
2423 
2424  // The remaining optimizations require the format string to be "%s" or "%c"
2425  // and have an extra operand.
2426  if (FormatStr.size() == 2 && FormatStr[0] == '%' &&
2427  CI->getNumArgOperands() == 4) {
2428 
2429  // Decode the second character of the format string.
2430  if (FormatStr[1] == 'c') {
2431  if (N == 0)
2432  return ConstantInt::get(CI->getType(), 1);
2433  else if (N == 1)
2434  return nullptr;
2435 
2436  // snprintf(dst, size, "%c", chr) --> *(i8*)dst = chr; *((i8*)dst+1) = 0
2437  if (!CI->getArgOperand(3)->getType()->isIntegerTy())
2438  return nullptr;
2439  Value *V = B.CreateTrunc(CI->getArgOperand(3), B.getInt8Ty(), "char");
2440  Value *Ptr = castToCStr(CI->getArgOperand(0), B);
2441  B.CreateStore(V, Ptr);
2442  Ptr = B.CreateGEP(B.getInt8Ty(), Ptr, B.getInt32(1), "nul");
2443  B.CreateStore(B.getInt8(0), Ptr);
2444 
2445  return ConstantInt::get(CI->getType(), 1);
2446  }
2447 
2448  if (FormatStr[1] == 's') {
2449  // snprintf(dest, size, "%s", str) to llvm.memcpy(dest, str, len+1, 1)
2450  StringRef Str;
2451  if (!getConstantStringInfo(CI->getArgOperand(3), Str))
2452  return nullptr;
2453 
2454  if (N == 0)
2455  return ConstantInt::get(CI->getType(), Str.size());
2456  else if (N < Str.size() + 1)
2457  return nullptr;
2458 
2459  B.CreateMemCpy(CI->getArgOperand(0), 1, CI->getArgOperand(3), 1,
2460  ConstantInt::get(CI->getType(), Str.size() + 1));
2461 
2462  // The snprintf result is the unincremented number of bytes in the string.
2463  return ConstantInt::get(CI->getType(), Str.size());
2464  }
2465  }
2466  return nullptr;
2467 }
2468 
2469 Value *LibCallSimplifier::optimizeSnPrintF(CallInst *CI, IRBuilder<> &B) {
2470  if (Value *V = optimizeSnPrintFString(CI, B)) {
2471  return V;
2472  }
2473 
2474  if (isKnownNonZero(CI->getOperand(1), DL))
2476  return nullptr;
2477 }
2478 
2479 Value *LibCallSimplifier::optimizeFPrintFString(CallInst *CI, IRBuilder<> &B) {
2480  optimizeErrorReporting(CI, B, 0);
2481 
2482  // All the optimizations depend on the format string.
2483  StringRef FormatStr;
2484  if (!getConstantStringInfo(CI->getArgOperand(1), FormatStr))
2485  return nullptr;
2486 
2487  // Do not do any of the following transformations if the fprintf return
2488  // value is used, in general the fprintf return value is not compatible
2489  // with fwrite(), fputc() or fputs().
2490  if (!CI->use_empty())
2491  return nullptr;
2492 
2493  // fprintf(F, "foo") --> fwrite("foo", 3, 1, F)
2494  if (CI->getNumArgOperands() == 2) {
2495  // Could handle %% -> % if we cared.
2496  if (FormatStr.find('%') != StringRef::npos)
2497  return nullptr; // We found a format specifier.
2498 
2499  return emitFWrite(
2500  CI->getArgOperand(1),
2501  ConstantInt::get(DL.getIntPtrType(CI->getContext()), FormatStr.size()),
2502  CI->getArgOperand(0), B, DL, TLI);
2503  }
2504 
2505  // The remaining optimizations require the format string to be "%s" or "%c"
2506  // and have an extra operand.
2507  if (FormatStr.size() != 2 || FormatStr[0] != '%' ||
2508  CI->getNumArgOperands() < 3)
2509  return nullptr;
2510 
2511  // Decode the second character of the format string.
2512  if (FormatStr[1] == 'c') {
2513  // fprintf(F, "%c", chr) --> fputc(chr, F)
2514  if (!CI->getArgOperand(2)->getType()->isIntegerTy())
2515  return nullptr;
2516  return emitFPutC(CI->getArgOperand(2), CI->getArgOperand(0), B, TLI);
2517  }
2518 
2519  if (FormatStr[1] == 's') {
2520  // fprintf(F, "%s", str) --> fputs(str, F)
2521  if (!CI->getArgOperand(2)->getType()->isPointerTy())
2522  return nullptr;
2523  return emitFPutS(CI->getArgOperand(2), CI->getArgOperand(0), B, TLI);
2524  }
2525  return nullptr;
2526 }
2527 
2528 Value *LibCallSimplifier::optimizeFPrintF(CallInst *CI, IRBuilder<> &B) {
2529  Function *Callee = CI->getCalledFunction();
2530  FunctionType *FT = Callee->getFunctionType();
2531  if (Value *V = optimizeFPrintFString(CI, B)) {
2532  return V;
2533  }
2534 
2535  // fprintf(stream, format, ...) -> fiprintf(stream, format, ...) if no
2536  // floating point arguments.
2537  if (TLI->has(LibFunc_fiprintf) && !callHasFloatingPointArgument(CI)) {
2538  Module *M = B.GetInsertBlock()->getParent()->getParent();
2539  FunctionCallee FIPrintFFn =
2540  M->getOrInsertFunction("fiprintf", FT, Callee->getAttributes());
2541  CallInst *New = cast<CallInst>(CI->clone());
2542  New->setCalledFunction(FIPrintFFn);
2543  B.Insert(New);
2544  return New;
2545  }
2546 
2547  // fprintf(stream, format, ...) -> __small_fprintf(stream, format, ...) if no
2548  // 128-bit floating point arguments.
2549  if (TLI->has(LibFunc_small_fprintf) && !callHasFP128Argument(CI)) {
2550  Module *M = B.GetInsertBlock()->getParent()->getParent();
2551  auto SmallFPrintFFn =
2552  M->getOrInsertFunction(TLI->getName(LibFunc_small_fprintf),
2553  FT, Callee->getAttributes());
2554  CallInst *New = cast<CallInst>(CI->clone());
2555  New->setCalledFunction(SmallFPrintFFn);
2556  B.Insert(New);
2557  return New;
2558  }
2559 
2560  return nullptr;
2561 }
2562 
2563 Value *LibCallSimplifier::optimizeFWrite(CallInst *CI, IRBuilder<> &B) {
2564  optimizeErrorReporting(CI, B, 3);
2565 
2566  // Get the element size and count.
2567  ConstantInt *SizeC = dyn_cast<ConstantInt>(CI->getArgOperand(1));
2568  ConstantInt *CountC = dyn_cast<ConstantInt>(CI->getArgOperand(2));
2569  if (SizeC && CountC) {
2570  uint64_t Bytes = SizeC->getZExtValue() * CountC->getZExtValue();
2571 
2572  // If this is writing zero records, remove the call (it's a noop).
2573  if (Bytes == 0)
2574  return ConstantInt::get(CI->getType(), 0);
2575 
2576  // If this is writing one byte, turn it into fputc.
2577  // This optimisation is only valid, if the return value is unused.
2578  if (Bytes == 1 && CI->use_empty()) { // fwrite(S,1,1,F) -> fputc(S[0],F)
2579  Value *Char = B.CreateLoad(B.getInt8Ty(),
2580  castToCStr(CI->getArgOperand(0), B), "char");
2581  Value *NewCI = emitFPutC(Char, CI->getArgOperand(3), B, TLI);
2582  return NewCI ? ConstantInt::get(CI->getType(), 1) : nullptr;
2583  }
2584  }
2585 
2586  if (isLocallyOpenedFile(CI->getArgOperand(3), CI, B, TLI))
2587  return emitFWriteUnlocked(CI->getArgOperand(0), CI->getArgOperand(1),
2588  CI->getArgOperand(2), CI->getArgOperand(3), B, DL,
2589  TLI);
2590 
2591  return nullptr;
2592 }
2593 
2594 Value *LibCallSimplifier::optimizeFPuts(CallInst *CI, IRBuilder<> &B) {
2595  optimizeErrorReporting(CI, B, 1);
2596 
2597  // Don't rewrite fputs to fwrite when optimising for size because fwrite
2598  // requires more arguments and thus extra MOVs are required.
2599  bool OptForSize = CI->getFunction()->hasOptSize() ||
2601  if (OptForSize)
2602  return nullptr;
2603 
2604  // Check if has any use
2605  if (!CI->use_empty()) {
2606  if (isLocallyOpenedFile(CI->getArgOperand(1), CI, B, TLI))
2607  return emitFPutSUnlocked(CI->getArgOperand(0), CI->getArgOperand(1), B,
2608  TLI);
2609  else
2610  // We can't optimize if return value is used.
2611  return nullptr;
2612  }
2613 
2614  // fputs(s,F) --> fwrite(s,strlen(s),1,F)
2615  uint64_t Len = GetStringLength(CI->getArgOperand(0));
2616  if (!Len)
2617  return nullptr;
2618 
2619  // Known to have no uses (see above).
2620  return emitFWrite(
2621  CI->getArgOperand(0),
2622  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len - 1),
2623  CI->getArgOperand(1), B, DL, TLI);
2624 }
2625 
2626 Value *LibCallSimplifier::optimizeFPutc(CallInst *CI, IRBuilder<> &B) {
2627  optimizeErrorReporting(CI, B, 1);
2628 
2629  if (isLocallyOpenedFile(CI->getArgOperand(1), CI, B, TLI))
2630  return emitFPutCUnlocked(CI->getArgOperand(0), CI->getArgOperand(1), B,
2631  TLI);
2632 
2633  return nullptr;
2634 }
2635 
2636 Value *LibCallSimplifier::optimizeFGetc(CallInst *CI, IRBuilder<> &B) {
2637  if (isLocallyOpenedFile(CI->getArgOperand(0), CI, B, TLI))
2638  return emitFGetCUnlocked(CI->getArgOperand(0), B, TLI);
2639 
2640  return nullptr;
2641 }
2642 
2643 Value *LibCallSimplifier::optimizeFGets(CallInst *CI, IRBuilder<> &B) {
2644  if (isLocallyOpenedFile(CI->getArgOperand(2), CI, B, TLI))
2645  return emitFGetSUnlocked(CI->getArgOperand(0), CI->getArgOperand(1),
2646  CI->getArgOperand(2), B, TLI);
2647 
2648  return nullptr;
2649 }
2650 
2651 Value *LibCallSimplifier::optimizeFRead(CallInst *CI, IRBuilder<> &B) {
2652  if (isLocallyOpenedFile(CI->getArgOperand(3), CI, B, TLI))
2653  return emitFReadUnlocked(CI->getArgOperand(0), CI->getArgOperand(1),
2654  CI->getArgOperand(2), CI->getArgOperand(3), B, DL,
2655  TLI);
2656 
2657  return nullptr;
2658 }
2659 
2660 Value *LibCallSimplifier::optimizePuts(CallInst *CI, IRBuilder<> &B) {
2662  if (!CI->use_empty())
2663  return nullptr;
2664 
2665  // Check for a constant string.
2666  // puts("") -> putchar('\n')
2667  StringRef Str;
2668  if (getConstantStringInfo(CI->getArgOperand(0), Str) && Str.empty())
2669  return emitPutChar(B.getInt32('\n'), B, TLI);
2670 
2671  return nullptr;
2672 }
2673 
2674 bool LibCallSimplifier::hasFloatVersion(StringRef FuncName) {
2675  LibFunc Func;
2676  SmallString<20> FloatFuncName = FuncName;
2677  FloatFuncName += 'f';
2678  if (TLI->getLibFunc(FloatFuncName, Func))
2679  return TLI->has(Func);
2680  return false;
2681 }
2682 
2683 Value *LibCallSimplifier::optimizeStringMemoryLibCall(CallInst *CI,
2684  IRBuilder<> &Builder) {
2685  LibFunc Func;
2686  Function *Callee = CI->getCalledFunction();
2687  // Check for string/memory library functions.
2688  if (TLI->getLibFunc(*Callee, Func) && TLI->has(Func)) {
2689  // Make sure we never change the calling convention.
2690  assert((ignoreCallingConv(Func) ||
2691  isCallingConvCCompatible(CI)) &&
2692  "Optimizing string/memory libcall would change the calling convention");
2693  switch (Func) {
2694  case LibFunc_strcat:
2695  return optimizeStrCat(CI, Builder);
2696  case LibFunc_strncat:
2697  return optimizeStrNCat(CI, Builder);
2698  case LibFunc_strchr:
2699  return optimizeStrChr(CI, Builder);
2700  case LibFunc_strrchr:
2701  return optimizeStrRChr(CI, Builder);
2702  case LibFunc_strcmp:
2703  return optimizeStrCmp(CI, Builder);
2704  case LibFunc_strncmp:
2705  return optimizeStrNCmp(CI, Builder);
2706  case LibFunc_strcpy:
2707  return optimizeStrCpy(CI, Builder);
2708  case LibFunc_stpcpy:
2709  return optimizeStpCpy(CI, Builder);
2710  case LibFunc_strncpy:
2711  return optimizeStrNCpy(CI, Builder);
2712  case LibFunc_strlen:
2713  return optimizeStrLen(CI, Builder);
2714  case LibFunc_strpbrk:
2715  return optimizeStrPBrk(CI, Builder);
2716  case LibFunc_strtol:
2717  case LibFunc_strtod:
2718  case LibFunc_strtof:
2719  case LibFunc_strtoul:
2720  case LibFunc_strtoll:
2721  case LibFunc_strtold:
2722  case LibFunc_strtoull:
2723  return optimizeStrTo(CI, Builder);
2724  case LibFunc_strspn:
2725  return optimizeStrSpn(CI, Builder);
2726  case LibFunc_strcspn:
2727  return optimizeStrCSpn(CI, Builder);
2728  case LibFunc_strstr:
2729  return optimizeStrStr(CI, Builder);
2730  case LibFunc_memchr:
2731  return optimizeMemChr(CI, Builder);
2732  case LibFunc_memrchr:
2733  return optimizeMemRChr(CI, Builder);
2734  case LibFunc_bcmp:
2735  return optimizeBCmp(CI, Builder);
2736  case LibFunc_memcmp:
2737  return optimizeMemCmp(CI, Builder);
2738  case LibFunc_memcpy:
2739  return optimizeMemCpy(CI, Builder);
2740  case LibFunc_mempcpy:
2741  return optimizeMemPCpy(CI, Builder);
2742  case LibFunc_memmove:
2743  return optimizeMemMove(CI, Builder);
2744  case LibFunc_memset:
2745  return optimizeMemSet(CI, Builder);
2746  case LibFunc_realloc:
2747  return optimizeRealloc(CI, Builder);
2748  case LibFunc_wcslen:
2749  return optimizeWcslen(CI, Builder);
2750  default:
2751  break;
2752  }
2753  }
2754  return nullptr;
2755 }
2756 
2757 Value *LibCallSimplifier::optimizeFloatingPointLibCall(CallInst *CI,
2758  LibFunc Func,
2759  IRBuilder<> &Builder) {
2760  // Don't optimize calls that require strict floating point semantics.
2761  if (CI->isStrictFP())
2762  return nullptr;
2763 
2764  if (Value *V = optimizeTrigReflections(CI, Func, Builder))
2765  return V;
2766 
2767  switch (Func) {
2768  case LibFunc_sinpif:
2769  case LibFunc_sinpi:
2770  case LibFunc_cospif:
2771  case LibFunc_cospi:
2772  return optimizeSinCosPi(CI, Builder);
2773  case LibFunc_powf:
2774  case LibFunc_pow:
2775  case LibFunc_powl:
2776  return optimizePow(CI, Builder);
2777  case LibFunc_exp2l:
2778  case LibFunc_exp2:
2779  case LibFunc_exp2f:
2780  return optimizeExp2(CI, Builder);
2781  case LibFunc_fabsf:
2782  case LibFunc_fabs:
2783  case LibFunc_fabsl:
2784  return replaceUnaryCall(CI, Builder, Intrinsic::fabs);
2785  case LibFunc_sqrtf:
2786  case LibFunc_sqrt:
2787  case LibFunc_sqrtl:
2788  return optimizeSqrt(CI, Builder);
2789  case LibFunc_log:
2790  case LibFunc_log10:
2791  case LibFunc_log1p:
2792  case LibFunc_log2:
2793  case LibFunc_logb:
2794  return optimizeLog(CI, Builder);
2795  case LibFunc_tan:
2796  case LibFunc_tanf:
2797  case LibFunc_tanl:
2798  return optimizeTan(CI, Builder);
2799  case LibFunc_ceil:
2800  return replaceUnaryCall(CI, Builder, Intrinsic::ceil);
2801  case LibFunc_floor:
2802  return replaceUnaryCall(CI, Builder, Intrinsic::floor);
2803  case LibFunc_round:
2804  return replaceUnaryCall(CI, Builder, Intrinsic::round);
2805  case LibFunc_nearbyint:
2806  return replaceUnaryCall(CI, Builder, Intrinsic::nearbyint);
2807  case LibFunc_rint:
2808  return replaceUnaryCall(CI, Builder, Intrinsic::rint);
2809  case LibFunc_trunc:
2810  return replaceUnaryCall(CI, Builder, Intrinsic::trunc);
2811  case LibFunc_acos:
2812  case LibFunc_acosh:
2813  case LibFunc_asin:
2814  case LibFunc_asinh:
2815  case LibFunc_atan:
2816  case LibFunc_atanh:
2817  case LibFunc_cbrt:
2818  case LibFunc_cosh:
2819  case LibFunc_exp:
2820  case LibFunc_exp10:
2821  case LibFunc_expm1:
2822  case LibFunc_cos:
2823  case LibFunc_sin:
2824  case LibFunc_sinh:
2825  case LibFunc_tanh:
2826  if (UnsafeFPShrink && hasFloatVersion(CI->getCalledFunction()->getName()))
2827  return optimizeUnaryDoubleFP(CI, Builder, true);
2828  return nullptr;
2829  case LibFunc_copysign:
2830  if (hasFloatVersion(CI->getCalledFunction()->getName()))
2831  return optimizeBinaryDoubleFP(CI, Builder);
2832  return nullptr;
2833  case LibFunc_fminf:
2834  case LibFunc_fmin:
2835  case LibFunc_fminl:
2836  case LibFunc_fmaxf:
2837  case LibFunc_fmax:
2838  case LibFunc_fmaxl:
2839  return optimizeFMinFMax(CI, Builder);
2840  case LibFunc_cabs:
2841  case LibFunc_cabsf:
2842  case LibFunc_cabsl:
2843  return optimizeCAbs(CI, Builder);
2844  default:
2845  return nullptr;
2846  }
2847 }
2848 
2850  // TODO: Split out the code below that operates on FP calls so that
2851  // we can all non-FP calls with the StrictFP attribute to be
2852  // optimized.
2853  if (CI->isNoBuiltin())
2854  return nullptr;
2855 
2856  LibFunc Func;
2857  Function *Callee = CI->getCalledFunction();
2858 
2860  CI->getOperandBundlesAsDefs(OpBundles);
2861  IRBuilder<> Builder(CI, /*FPMathTag=*/nullptr, OpBundles);
2862  bool isCallingConvC = isCallingConvCCompatible(CI);
2863 
2864  // Command-line parameter overrides instruction attribute.
2865  // This can't be moved to optimizeFloatingPointLibCall() because it may be
2866  // used by the intrinsic optimizations.
2867  if (EnableUnsafeFPShrink.getNumOccurrences() > 0)
2868  UnsafeFPShrink = EnableUnsafeFPShrink;
2869  else if (isa<FPMathOperator>(CI) && CI->isFast())
2870  UnsafeFPShrink = true;
2871 
2872  // First, check for intrinsics.
2873  if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI)) {
2874  if (!isCallingConvC)
2875  return nullptr;
2876  // The FP intrinsics have corresponding constrained versions so we don't
2877  // need to check for the StrictFP attribute here.
2878  switch (II->getIntrinsicID()) {
2879  case Intrinsic::pow:
2880  return optimizePow(CI, Builder);
2881  case Intrinsic::exp2:
2882  return optimizeExp2(CI, Builder);
2883  case Intrinsic::log:
2884  return optimizeLog(CI, Builder);
2885  case Intrinsic::sqrt:
2886  return optimizeSqrt(CI, Builder);
2887  // TODO: Use foldMallocMemset() with memset intrinsic.
2888  case Intrinsic::memset:
2889  return optimizeMemSet(CI, Builder);
2890  case Intrinsic::memcpy:
2891  return optimizeMemCpy(CI, Builder);
2892  case Intrinsic::memmove:
2893  return optimizeMemMove(CI, Builder);
2894  default:
2895  return nullptr;
2896  }
2897  }
2898 
2899  // Also try to simplify calls to fortified library functions.
2900  if (Value *SimplifiedFortifiedCI = FortifiedSimplifier.optimizeCall(CI)) {
2901  // Try to further simplify the result.
2902  CallInst *SimplifiedCI = dyn_cast<CallInst>(SimplifiedFortifiedCI);
2903  if (SimplifiedCI && SimplifiedCI->getCalledFunction()) {
2904  // Use an IR Builder from SimplifiedCI if available instead of CI
2905  // to guarantee we reach all uses we might replace later on.
2906  IRBuilder<> TmpBuilder(SimplifiedCI);
2907  if (Value *V = optimizeStringMemoryLibCall(SimplifiedCI, TmpBuilder)) {
2908  // If we were able to further simplify, remove the now redundant call.
2909  substituteInParent(SimplifiedCI, V);
2910  return V;
2911  }
2912  }
2913  return SimplifiedFortifiedCI;
2914  }
2915 
2916  // Then check for known library functions.
2917  if (TLI->getLibFunc(*Callee, Func) && TLI->has(Func)) {
2918  // We never change the calling convention.
2919  if (!ignoreCallingConv(Func) && !isCallingConvC)
2920  return nullptr;
2921  if (Value *V = optimizeStringMemoryLibCall(CI, Builder))
2922  return V;
2923  if (Value *V = optimizeFloatingPointLibCall(CI, Func, Builder))
2924  return V;
2925  switch (Func) {
2926  case LibFunc_ffs:
2927  case LibFunc_ffsl:
2928  case LibFunc_ffsll:
2929  return optimizeFFS(CI, Builder);
2930  case LibFunc_fls:
2931  case LibFunc_flsl:
2932  case LibFunc_flsll:
2933  return optimizeFls(CI, Builder);
2934  case LibFunc_abs:
2935  case LibFunc_labs:
2936  case LibFunc_llabs:
2937  return optimizeAbs(CI, Builder);
2938  case LibFunc_isdigit:
2939  return optimizeIsDigit(CI, Builder);
2940  case LibFunc_isascii:
2941  return optimizeIsAscii(CI, Builder);
2942  case LibFunc_toascii:
2943  return optimizeToAscii(CI, Builder);
2944  case LibFunc_atoi:
2945  case LibFunc_atol:
2946  case LibFunc_atoll:
2947  return optimizeAtoi(CI, Builder);
2948  case LibFunc_strtol:
2949  case LibFunc_strtoll:
2950  return optimizeStrtol(CI, Builder);
2951  case LibFunc_printf:
2952  return optimizePrintF(CI, Builder);
2953  case LibFunc_sprintf:
2954  return optimizeSPrintF(CI, Builder);
2955  case LibFunc_snprintf:
2956  return optimizeSnPrintF(CI, Builder);
2957  case LibFunc_fprintf:
2958  return optimizeFPrintF(CI, Builder);
2959  case LibFunc_fwrite:
2960  return optimizeFWrite(CI, Builder);
2961  case LibFunc_fread:
2962  return optimizeFRead(CI, Builder);
2963  case LibFunc_fputs:
2964  return optimizeFPuts(CI, Builder);
2965  case LibFunc_fgets:
2966  return optimizeFGets(CI, Builder);
2967  case LibFunc_fputc:
2968  return optimizeFPutc(CI, Builder);
2969  case LibFunc_fgetc:
2970  return optimizeFGetc(CI, Builder);
2971  case LibFunc_puts:
2972  return optimizePuts(CI, Builder);
2973  case LibFunc_perror:
2974  return optimizeErrorReporting(CI, Builder);
2975  case LibFunc_vfprintf:
2976  case LibFunc_fiprintf:
2977  return optimizeErrorReporting(CI, Builder, 0);
2978  default:
2979  return nullptr;
2980  }
2981  }
2982  return nullptr;
2983 }
2984 
2986  const DataLayout &DL, const TargetLibraryInfo *TLI,
2989  function_ref<void(Instruction *, Value *)> Replacer,
2990  function_ref<void(Instruction *)> Eraser)
2991  : FortifiedSimplifier(TLI), DL(DL), TLI(TLI), ORE(ORE), BFI(BFI), PSI(PSI),
2992  UnsafeFPShrink(false), Replacer(Replacer), Eraser(Eraser) {}
2993 
2994 void LibCallSimplifier::replaceAllUsesWith(Instruction *I, Value *With) {
2995  // Indirect through the replacer used in this instance.
2996  Replacer(I, With);
2997 }
2998 
2999 void LibCallSimplifier::eraseFromParent(Instruction *I) {
3000  Eraser(I);
3001 }
3002 
3003 // TODO:
3004 // Additional cases that we need to add to this file:
3005 //
3006 // cbrt:
3007 // * cbrt(expN(X)) -> expN(x/3)
3008 // * cbrt(sqrt(x)) -> pow(x,1/6)
3009 // * cbrt(cbrt(x)) -> pow(x,1/9)
3010 //
3011 // exp, expf, expl:
3012 // * exp(log(x)) -> x
3013 //
3014 // log, logf, logl:
3015 // * log(exp(x)) -> x
3016 // * log(exp(y)) -> y*log(e)
3017 // * log(exp10(y)) -> y*log(10)
3018 // * log(sqrt(x)) -> 0.5*log(x)
3019 //
3020 // pow, powf, powl:
3021 // * pow(sqrt(x),y) -> pow(x,y*0.5)
3022 // * pow(pow(x,y),z)-> pow(x,y*z)
3023 //
3024 // signbit:
3025 // * signbit(cnst) -> cnst'
3026 // * signbit(nncst) -> 0 (if pstv is a non-negative constant)
3027 //
3028 // sqrt, sqrtf, sqrtl:
3029 // * sqrt(expN(x)) -> expN(x*0.5)
3030 // * sqrt(Nroot(x)) -> pow(x,1/(2*N))
3031 // * sqrt(pow(x,y)) -> pow(|x|,y*0.5)
3032 //
3033 
3034 //===----------------------------------------------------------------------===//
3035 // Fortified Library Call Optimizations
3036 //===----------------------------------------------------------------------===//
3037 
3038 bool
3039 FortifiedLibCallSimplifier::isFortifiedCallFoldable(CallInst *CI,
3040  unsigned ObjSizeOp,
3041  Optional<unsigned> SizeOp,
3042  Optional<unsigned> StrOp,
3043  Optional<unsigned> FlagOp) {
3044  // If this function takes a flag argument, the implementation may use it to
3045  // perform extra checks. Don't fold into the non-checking variant.
3046  if (FlagOp) {
3048  if (!Flag || !Flag->isZero())
3049  return false;
3050  }
3051 
3052  if (SizeOp && CI->getArgOperand(ObjSizeOp) == CI->getArgOperand(*SizeOp))
3053  return true;
3054 
3055  if (ConstantInt *ObjSizeCI =
3056  dyn_cast<ConstantInt>(CI->getArgOperand(ObjSizeOp))) {
3057  if (ObjSizeCI->isMinusOne())
3058  return true;
3059  // If the object size wasn't -1 (unknown), bail out if we were asked to.
3060  if (OnlyLowerUnknownSize)
3061  return false;
3062  if (StrOp) {
3063  uint64_t Len = GetStringLength(CI->getArgOperand(*StrOp));
3064  // If the length is 0 we don't know how long it is and so we can't
3065  // remove the check.
3066  if (Len)
3067  annotateDereferenceableBytes(CI, *StrOp, Len);
3068  else
3069  return false;
3070  return ObjSizeCI->getZExtValue() >= Len;
3071  }
3072 
3073  if (SizeOp) {
3074  if (ConstantInt *SizeCI =
3075  dyn_cast<ConstantInt>(CI->getArgOperand(*SizeOp)))
3076  return ObjSizeCI->getZExtValue() >= SizeCI->getZExtValue();
3077  }
3078  }
3079  return false;
3080 }
3081 
3082 Value *FortifiedLibCallSimplifier::optimizeMemCpyChk(CallInst *CI,
3083  IRBuilder<> &B) {
3084  if (isFortifiedCallFoldable(CI, 3, 2)) {
3085  CallInst *NewCI = B.CreateMemCpy(
3086  CI->getArgOperand(0), 1, CI->getArgOperand(1), 1, CI->getArgOperand(2));
3087  NewCI->setAttributes(CI->getAttributes());
3088  return CI->getArgOperand(0);
3089  }
3090  return nullptr;
3091 }
3092 
3093 Value *FortifiedLibCallSimplifier::optimizeMemMoveChk(CallInst *CI,
3094  IRBuilder<> &B) {
3095  if (isFortifiedCallFoldable(CI, 3, 2)) {
3096  CallInst *NewCI = B.CreateMemMove(
3097  CI->getArgOperand(0), 1, CI->getArgOperand(1), 1, CI->getArgOperand(2));
3098  NewCI->setAttributes(CI->getAttributes());
3099  return CI->getArgOperand(0);
3100  }
3101  return nullptr;
3102 }
3103 
3104 Value *FortifiedLibCallSimplifier::optimizeMemSetChk(CallInst *CI,
3105  IRBuilder<> &B) {
3106  // TODO: Try foldMallocMemset() here.
3107 
3108  if (isFortifiedCallFoldable(CI, 3, 2)) {
3109  Value *Val = B.CreateIntCast(CI->getArgOperand(1), B.getInt8Ty(), false);
3110  CallInst *NewCI =
3111  B.CreateMemSet(CI->getArgOperand(0), Val, CI->getArgOperand(2), 1);
3112  NewCI->setAttributes(CI->getAttributes());
3113  return CI->getArgOperand(0);
3114  }
3115  return nullptr;
3116 }
3117 
3118 Value *FortifiedLibCallSimplifier::optimizeStrpCpyChk(CallInst *CI,
3119  IRBuilder<> &B,
3120  LibFunc Func) {
3121  const DataLayout &DL = CI->getModule()->getDataLayout();
3122  Value *Dst = CI->getArgOperand(0), *Src = CI->getArgOperand(1),
3123  *ObjSize = CI->getArgOperand(2);
3124 
3125  // __stpcpy_chk(x,x,...) -> x+strlen(x)
3126  if (Func == LibFunc_stpcpy_chk && !OnlyLowerUnknownSize && Dst == Src) {
3127  Value *StrLen = emitStrLen(Src, B, DL, TLI);
3128  return StrLen ? B.CreateInBoundsGEP(B.getInt8Ty(), Dst, StrLen) : nullptr;
3129  }
3130 
3131  // If a) we don't have any length information, or b) we know this will
3132  // fit then just lower to a plain st[rp]cpy. Otherwise we'll keep our
3133  // st[rp]cpy_chk call which may fail at runtime if the size is too long.
3134  // TODO: It might be nice to get a maximum length out of the possible
3135  // string lengths for varying.
3136  if (isFortifiedCallFoldable(CI, 2, None, 1)) {
3137  if (Func == LibFunc_strcpy_chk)
3138  return emitStrCpy(Dst, Src, B, TLI);
3139  else
3140  return emitStpCpy(Dst, Src, B, TLI);
3141  }
3142 
3143  if (OnlyLowerUnknownSize)
3144  return nullptr;
3145 
3146  // Maybe we can stil fold __st[rp]cpy_chk to __memcpy_chk.
3147  uint64_t Len = GetStringLength(Src);
3148  if (Len)
3149  annotateDereferenceableBytes(CI, 1, Len);
3150  else
3151  return nullptr;
3152 
3153  Type *SizeTTy = DL.getIntPtrType(CI->getContext());
3154  Value *LenV = ConstantInt::get(SizeTTy, Len);
3155  Value *Ret = emitMemCpyChk(Dst, Src, LenV, ObjSize, B, DL, TLI);
3156  // If the function was an __stpcpy_chk, and we were able to fold it into
3157  // a __memcpy_chk, we still need to return the correct end pointer.
3158  if (Ret && Func == LibFunc_stpcpy_chk)
3159  return B.CreateGEP(B.getInt8Ty(), Dst, ConstantInt::get(SizeTTy, Len - 1));
3160  return Ret;
3161 }
3162 
3163 Value *FortifiedLibCallSimplifier::optimizeStrpNCpyChk(CallInst *CI,
3164  IRBuilder<> &B,
3165  LibFunc Func) {
3166  if (isFortifiedCallFoldable(CI, 3, 2)) {
3167  if (Func == LibFunc_strncpy_chk)
3168  return emitStrNCpy(CI->getArgOperand(0), CI->getArgOperand(1),
3169  CI->getArgOperand(2), B, TLI);
3170  else
3171  return emitStpNCpy(CI->getArgOperand(0), CI->getArgOperand(1),
3172  CI->getArgOperand(2), B, TLI);
3173  }
3174 
3175  return nullptr;
3176 }
3177 
3178 Value *FortifiedLibCallSimplifier::optimizeMemCCpyChk(CallInst *CI,
3179  IRBuilder<> &B) {
3180  if (isFortifiedCallFoldable(CI, 4, 3))
3181  return emitMemCCpy(CI->getArgOperand(0), CI->getArgOperand(1),
3182  CI->getArgOperand(2), CI->getArgOperand(3), B, TLI);
3183 
3184  return nullptr;
3185 }
3186 
3187 Value *FortifiedLibCallSimplifier::optimizeSNPrintfChk(CallInst *CI,
3188  IRBuilder<> &B) {
3189  if (isFortifiedCallFoldable(CI, 3, 1, None, 2)) {
3190  SmallVector<Value *, 8> VariadicArgs(CI->arg_begin() + 5, CI->arg_end());
3191  return emitSNPrintf(CI->getArgOperand(0), CI->getArgOperand(1),
3192  CI->getArgOperand(4), VariadicArgs, B, TLI);
3193  }
3194 
3195  return nullptr;
3196 }
3197 
3198 Value *FortifiedLibCallSimplifier::optimizeSPrintfChk(CallInst *CI,
3199  IRBuilder<> &B) {
3200  if (isFortifiedCallFoldable(CI, 2, None, None, 1)) {
3201  SmallVector<Value *, 8> VariadicArgs(CI->arg_begin() + 4, CI->arg_end());
3202  return emitSPrintf(CI->getArgOperand(0), CI->getArgOperand(3), VariadicArgs,
3203  B, TLI);
3204  }
3205 
3206  return nullptr;
3207 }
3208 
3209 Value *FortifiedLibCallSimplifier::optimizeStrCatChk(CallInst *CI,
3210  IRBuilder<> &B) {
3211  if (isFortifiedCallFoldable(CI, 2))
3212  return emitStrCat(CI->getArgOperand(0), CI->getArgOperand(1), B, TLI);
3213 
3214  return nullptr;
3215 }
3216 
3217 Value *FortifiedLibCallSimplifier::optimizeStrLCat(CallInst *CI,
3218  IRBuilder<> &B) {
3219  if (isFortifiedCallFoldable(CI, 3))
3220  return emitStrLCat(CI->getArgOperand(0), CI->getArgOperand(1),
3221  CI->getArgOperand(2), B, TLI);
3222 
3223  return nullptr;
3224 }
3225 
3226 Value *FortifiedLibCallSimplifier::optimizeStrNCatChk(CallInst *CI,
3227  IRBuilder<> &B) {
3228  if (isFortifiedCallFoldable(CI, 3))
3229  return emitStrNCat(CI->getArgOperand(0), CI->getArgOperand(1),
3230  CI->getArgOperand(2), B, TLI);
3231 
3232  return nullptr;
3233 }
3234 
3235 Value *FortifiedLibCallSimplifier::optimizeStrLCpyChk(CallInst *CI,
3236  IRBuilder<> &B) {
3237  if (isFortifiedCallFoldable(CI, 3))
3238  return emitStrLCpy(CI->getArgOperand(0), CI->getArgOperand(1),
3239  CI->getArgOperand(2), B, TLI);
3240 
3241  return nullptr;
3242 }
3243 
3244 Value *FortifiedLibCallSimplifier::optimizeVSNPrintfChk(CallInst *CI,
3245  IRBuilder<> &B) {
3246  if (isFortifiedCallFoldable(CI, 3, 1, None, 2))
3247  return emitVSNPrintf(CI->getArgOperand(0), CI->getArgOperand(1),
3248  CI->getArgOperand(4), CI->getArgOperand(5), B, TLI);
3249 
3250  return nullptr;
3251 }
3252 
3253 Value *FortifiedLibCallSimplifier::optimizeVSPrintfChk(CallInst *CI,
3254  IRBuilder<> &B) {
3255  if (isFortifiedCallFoldable(CI, 2, None, None, 1))
3256  return emitVSPrintf(CI->getArgOperand(0), CI->getArgOperand(3),
3257  CI->getArgOperand(4), B, TLI);
3258 
3259  return nullptr;
3260 }
3261 
3263  // FIXME: We shouldn't be changing "nobuiltin" or TLI unavailable calls here.
3264  // Some clang users checked for _chk libcall availability using:
3265  // __has_builtin(__builtin___memcpy_chk)
3266  // When compiling with -fno-builtin, this is always true.
3267  // When passing -ffreestanding/-mkernel, which both imply -fno-builtin, we
3268  // end up with fortified libcalls, which isn't acceptable in a freestanding
3269  // environment which only provides their non-fortified counterparts.
3270  //
3271  // Until we change clang and/or teach external users to check for availability
3272  // differently, disregard the "nobuiltin" attribute and TLI::has.
3273  //
3274  // PR23093.
3275 
3276  LibFunc Func;
3277  Function *Callee = CI->getCalledFunction();
3278 
3280  CI->getOperandBundlesAsDefs(OpBundles);
3281  IRBuilder<> Builder(CI, /*FPMathTag=*/nullptr, OpBundles);
3282  bool isCallingConvC = isCallingConvCCompatible(CI);
3283 
3284  // First, check that this is a known library functions and that the prototype
3285  // is correct.
3286  if (!TLI->getLibFunc(*Callee, Func))
3287  return nullptr;
3288 
3289  // We never change the calling convention.
3290  if (!ignoreCallingConv(Func) && !isCallingConvC)
3291  return nullptr;
3292 
3293  switch (Func) {
3294  case LibFunc_memcpy_chk:
3295  return optimizeMemCpyChk(CI, Builder);
3296  case LibFunc_memmove_chk:
3297  return optimizeMemMoveChk(CI, Builder);
3298  case LibFunc_memset_chk:
3299  return optimizeMemSetChk(CI, Builder);
3300  case LibFunc_stpcpy_chk:
3301  case LibFunc_strcpy_chk:
3302  return optimizeStrpCpyChk(CI, Builder, Func);
3303  case LibFunc_stpncpy_chk:
3304  case LibFunc_strncpy_chk:
3305  return optimizeStrpNCpyChk(CI, Builder, Func);
3306  case LibFunc_memccpy_chk:
3307  return optimizeMemCCpyChk(CI, Builder);
3308  case LibFunc_snprintf_chk:
3309  return optimizeSNPrintfChk(CI, Builder);
3310  case LibFunc_sprintf_chk:
3311  return optimizeSPrintfChk(CI, Builder);
3312  case LibFunc_strcat_chk:
3313  return optimizeStrCatChk(CI, Builder);
3314  case LibFunc_strlcat_chk:
3315  return optimizeStrLCat(CI, Builder);
3316  case LibFunc_strncat_chk:
3317  return optimizeStrNCatChk(CI, Builder);
3318  case LibFunc_strlcpy_chk:
3319  return optimizeStrLCpyChk(CI, Builder);
3320  case LibFunc_vsnprintf_chk:
3321  return optimizeVSNPrintfChk(CI, Builder);
3322  case LibFunc_vsprintf_chk:
3323  return optimizeVSPrintfChk(CI, Builder);
3324  default:
3325  break;
3326  }
3327  return nullptr;
3328 }
3329 
3331  const TargetLibraryInfo *TLI, bool OnlyLowerUnknownSize)
3332  : TLI(TLI), OnlyLowerUnknownSize(OnlyLowerUnknownSize) {}
Value * CreateNSWNeg(Value *V, const Twine &Name="")
Definition: IRBuilder.h:1495
Value * CreateInBoundsGEP(Value *Ptr, ArrayRef< Value *> IdxList, const Twine &Name="")
Definition: IRBuilder.h:1696
uint64_t CallInst * C
void computeKnownBits(const Value *V, KnownBits &Known, const DataLayout &DL, unsigned Depth=0, AssumptionCache *AC=nullptr, const Instruction *CxtI=nullptr, const DominatorTree *DT=nullptr, OptimizationRemarkEmitter *ORE=nullptr, bool UseInstrInfo=true)
Determine which bits of V are known to be either zero or one and return them in the KnownZero/KnownOn...
bool isIntrinsic() const
isIntrinsic - Returns true if the function&#39;s name starts with "llvm.".
Definition: Function.h:198
bool hasNoInfs() const
Determine whether the no-infs flag is set.
Value * emitStpCpy(Value *Dst, Value *Src, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the stpcpy function to the builder, for the specified pointer arguments.
A parsed version of the target data layout string in and methods for querying it. ...
Definition: DataLayout.h:111
const std::string & getTargetTriple() const
Get the target triple which is a string describing the target host.
Definition: Module.h:240
class_match< Value > m_Value()
Match an arbitrary value and ignore it.
Definition: PatternMatch.h:70
Value * CreateICmp(CmpInst::Predicate P, Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:2212
static GCMetadataPrinterRegistry::Add< ErlangGCPrinter > X("erlang", "erlang-compatible garbage collector")
bool hasNoSignedZeros() const
Determine whether the no-signed-zeros flag is set.
void flipAllBits()
Toggle every bit to its opposite value.
Definition: APInt.h:1483
uint64_t getZExtValue() const
Get zero extended value.
Definition: APInt.h:1571
ThreeOps_match< Cond, LHS, RHS, Instruction::Select > m_Select(const Cond &C, const LHS &L, const RHS &R)
Matches SelectInst.
bool isOnlyUsedInZeroEqualityComparison(const Instruction *CxtI)
GCNRegPressure max(const GCNRegPressure &P1, const GCNRegPressure &P2)
LLVM_NODISCARD std::string str() const
str - Get the contents as an std::string.
Definition: StringRef.h:232
bool doesNotAccessMemory(unsigned OpNo) const
Definition: InstrTypes.h:1551
void setFast(bool B=true)
Definition: Operator.h:236
Value * CreateICmpNE(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:2106
Value * CreateIsNotNull(Value *Arg, const Twine &Name="")
Return an i1 value testing if Arg is not null.
Definition: IRBuilder.h:2382
Value * CreateZExtOrTrunc(Value *V, Type *DestTy, const Twine &Name="")
Create a ZExt or Trunc from the integer value V to DestTy.
Definition: IRBuilder.h:1888
This class represents lattice values for constants.
Definition: AllocatorList.h:23
ARM_APCS - ARM Procedure Calling Standard calling convention (obsolete, but still used on some target...
Definition: CallingConv.h:95
static bool callHasFloatingPointArgument(const CallInst *CI)
Type * getParamType(unsigned i) const
Parameter type accessors.
Definition: DerivedTypes.h:135
Value * CreateICmpULT(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:2118
LoadInst * CreateLoad(Type *Ty, Value *Ptr, const char *Name)
Provided to resolve &#39;CreateLoad(Ty, Ptr, "...")&#39; correctly, instead of converting the string to &#39;bool...
Definition: IRBuilder.h:1576
Value * optimizeCall(CallInst *CI)
Take the given call instruction and return a more optimal value to replace the instruction with or 0 ...
A Module instance is used to store all the information related to an LLVM module. ...
Definition: Module.h:65
bool hasOptSize() const
Optimize this function for size (-Os) or minimum size (-Oz).
Definition: Function.h:622
static Constant * getInfinity(Type *Ty, bool Negative=false)
Definition: Constants.cpp:839
static Attribute getWithDereferenceableBytes(LLVMContext &Context, uint64_t Bytes)
Definition: Attributes.cpp:158
amdgpu Simplify well known AMD library false FunctionCallee Value const Twine & Name
Function * getCaller()
Helper to get the caller (the parent function).
A handy container for a FunctionType+Callee-pointer pair, which can be passed around as a single enti...
Definition: DerivedTypes.h:165
LLVM_NODISCARD bool startswith(StringRef Prefix) const
Check if this string starts with the given Prefix.
Definition: StringRef.h:270
LLVM_NODISCARD size_t rfind(char C, size_t From=npos) const
Search for the last character C in the string.
Definition: StringRef.h:359
void push_back(const T &Elt)
Definition: SmallVector.h:211
Analysis providing profile information.
LLVM_NODISCARD int compare(StringRef RHS) const
compare - Compare two strings; the result is -1, 0, or 1 if this string is lexicographically less tha...
Definition: StringRef.h:188
Value * CreateICmpSLT(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:2134
This class represents a function call, abstracting a target machine&#39;s calling convention.
static uint64_t round(uint64_t Acc, uint64_t Input)
Definition: xxhash.cpp:57
Value * emitSNPrintf(Value *Dest, Value *Size, Value *Fmt, ArrayRef< Value *> Args, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the snprintf function.
unsigned less than
Definition: InstrTypes.h:757
An efficient, type-erasing, non-owning reference to a callable.
Definition: STLExtras.h:104
float convertToFloat() const
Definition: APFloat.h:1113
LLVMContext & getContext() const
All values hold a context through their type.
Definition: Value.cpp:733
bool hasFnAttribute(Attribute::AttrKind Kind) const
Return true if the function has the attribute.
Definition: Function.h:323
Value * CreateSExt(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1882
F(f)
User::op_iterator arg_end()
Return the iterator pointing to the end of the argument list.
Definition: InstrTypes.h:1212
const fltSemantics & getSemantics() const
Definition: APFloat.h:1170
unsigned getPointerAddressSpace() const
Get the address space of this pointer or pointer vector type.
Definition: DerivedTypes.h:606
CallInst * CreateMemSet(Value *Ptr, Value *Val, uint64_t Size, unsigned Align, bool isVolatile=false, MDNode *TBAATag=nullptr, MDNode *ScopeTag=nullptr, MDNode *NoAliasTag=nullptr)
Create and insert a memset to the specified pointer and the specified value.
Definition: IRBuilder.h:440
An instruction for reading from memory.
Definition: Instructions.h:167
Value * emitMemCCpy(Value *Ptr1, Value *Ptr2, Value *Val, Value *Len, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the memccpy function.
Hexagon Common GEP
static bool isTrigLibCall(CallInst *CI)
bool isGEPBasedOnPointerToString(const GEPOperator *GEP, unsigned CharSize=8)
Returns true if the GEP is based on a pointer to a string (array of.
Value * emitFPutSUnlocked(Value *Str, Value *File, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the fputs_unlocked function.
uint64_t Offset
Slice starts at this Offset.
void addAttribute(unsigned i, Attribute::AttrKind Kind)
adds the attribute to the list of attributes.
Definition: InstrTypes.h:1383
bool shouldOptimizeForSize(Function *F, ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI)
Returns true if function F is suggested to be size-optimized base on the profile. ...
Definition: SizeOpts.cpp:23
LLVMContext & getContext() const
Get the context in which this basic block lives.
Definition: BasicBlock.cpp:32
Value * emitMalloc(Value *Num, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the malloc function.
bool isDereferenceableAndAlignedPointer(const Value *V, Type *Ty, unsigned Align, const DataLayout &DL, const Instruction *CtxI=nullptr, const DominatorTree *DT=nullptr)
Returns true if V is always a dereferenceable pointer with alignment greater or equal than requested...
Definition: Loads.cpp:135
static bool isReportingError(Function *Callee, CallInst *CI, int StreamArg)
static void annotateDereferenceableBytes(CallInst *CI, ArrayRef< unsigned > ArgNos, uint64_t DereferenceableBytes)
static Constant * getNullValue(Type *Ty)
Constructor to create a &#39;0&#39; constant of arbitrary type.
Definition: Constants.cpp:289
iterator begin()
Instruction iterator methods.
Definition: BasicBlock.h:273
IntegerType * getInt32Ty()
Fetch the type representing a 32-bit integer.
Definition: IRBuilder.h:383
static Value * replaceUnaryCall(CallInst *CI, IRBuilder<> &B, Intrinsic::ID IID)
Value * getArgOperand(unsigned i) const
Definition: InstrTypes.h:1241
opStatus convertToInteger(MutableArrayRef< integerPart > Input, unsigned int Width, bool IsSigned, roundingMode RM, bool *IsExact) const
Definition: APFloat.h:1084
static GCMetadataPrinterRegistry::Add< OcamlGCMetadataPrinter > Y("ocaml", "ocaml 3.10-compatible collector")
specific_fpval m_SpecificFP(double V)
Match a specific floating point value or vector with all elements equal to the value.
Definition: PatternMatch.h:614
bool match(Val *V, const Pattern &P)
Definition: PatternMatch.h:47
static void annotateNonNullBasedOnAccess(CallInst *CI, ArrayRef< unsigned > ArgNos)
Value * CreateFPExt(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1947
void setBit(unsigned BitPosition)
Set a given bit to 1.
Definition: APInt.h:1402
This class represents the LLVM &#39;select&#39; instruction.
const DataLayout & getDataLayout() const
Get the data layout for the module&#39;s target platform.
Definition: Module.cpp:369
bool isNonNegative() const
Determine if this APInt Value is non-negative (>= 0)
Definition: APInt.h:368
FortifiedLibCallSimplifier(const TargetLibraryInfo *TLI, bool OnlyLowerUnknownSize=false)
A Use represents the edge between a Value definition and its users.
Definition: Use.h:55
PointerType * getPointerTo(unsigned AddrSpace=0) const
Return a pointer to the current type.
Definition: Type.cpp:654
Value * emitPutChar(Value *Char, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the putchar function. This assumes that Char is an integer.
bool isIntegerTy() const
True if this is an instance of IntegerType.
Definition: Type.h:196
This provides a uniform API for creating instructions and inserting them into a basic block: either a...
Definition: IRBuilder.h:779
bool hasApproxFunc() const
Determine whether the approximate-math-functions flag is set.
Value * emitStrLCpy(Value *Dest, Value *Src, Value *Size, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the strlcpy function.
Value * CreateAdd(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1118
CallInst * CreateMemMove(Value *Dst, unsigned DstAlign, Value *Src, unsigned SrcAlign, uint64_t Size, bool isVolatile=false, MDNode *TBAATag=nullptr, MDNode *ScopeTag=nullptr, MDNode *NoAliasTag=nullptr)
Create and insert a memmove between the specified pointers.
Definition: IRBuilder.h:530
Value * emitUnaryFloatFnCall(Value *Op, StringRef Name, IRBuilder<> &B, const AttributeList &Attrs)
Emit a call to the unary function named &#39;Name&#39; (e.g.
static cl::opt< bool > EnableUnsafeFPShrink("enable-double-float-shrink", cl::Hidden, cl::init(false), cl::desc("Enable unsafe double to float " "shrinking for math lib calls"))
Value * emitFPutCUnlocked(Value *Char, Value *File, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the fputc_unlocked function.
LLVM_NODISCARD StringRef substr(size_t Start, size_t N=npos) const
Return a reference to the substring from [Start, Start + N).
Definition: StringRef.h:592
LLVM_NODISCARD bool empty() const
empty - Check if the string is empty.
Definition: StringRef.h:140
static StructType * get(LLVMContext &Context, ArrayRef< Type *> Elements, bool isPacked=false)
This static method is the primary way to create a literal StructType.
Definition: Type.cpp:341
bool paramHasAttr(unsigned ArgNo, Attribute::AttrKind Kind) const
Determine whether the argument or parameter has the given attribute.
StoreInst * CreateStore(Value *Val, Value *Ptr, bool isVolatile=false)
Definition: IRBuilder.h:1605
static Value * convertStrToNumber(CallInst *CI, StringRef &Str, int64_t Base)
Value * CreateIntToPtr(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1959
Instruction * clone() const
Create a copy of &#39;this&#39; instruction that is identical in all ways except the following: ...
Class to represent function types.
Definition: DerivedTypes.h:103
Value * CreateBitCast(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1964
FastMathFlags getFastMathFlags() const
Convenience function for getting all the fast-math flags, which must be an operator which supports th...
Type * getType() const
All values are typed, get the type of this value.
Definition: Value.h:245
Value * CreateSExtOrTrunc(Value *V, Type *DestTy, const Twine &Name="")
Create a SExt or Trunc from the integer value V to DestTy.
Definition: IRBuilder.h:1903
apfloat_match m_APFloat(const APFloat *&Res)
Match a ConstantFP or splatted ConstantVector, binding the specified pointer to the contained APFloat...
Definition: PatternMatch.h:193
const ConstantDataArray * Array
ConstantDataArray pointer.
opStatus convert(const fltSemantics &ToSemantics, roundingMode RM, bool *losesInfo)
Definition: APFloat.cpp:4483
LibCallSimplifier(const DataLayout &DL, const TargetLibraryInfo *TLI, OptimizationRemarkEmitter &ORE, BlockFrequencyInfo *BFI, ProfileSummaryInfo *PSI, function_ref< void(Instruction *, Value *)> Replacer=&replaceAllUsesWithDefault, function_ref< void(Instruction *)> Eraser=&eraseFromParentDefault)
void setCalledFunction(Function *Fn)
Sets the function called, including updating the function type.
Definition: InstrTypes.h:1323
BasicBlock * GetInsertBlock() const
Definition: IRBuilder.h:126
Value * emitMemCpyChk(Value *Dst, Value *Src, Value *Len, Value *ObjSize, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the __memcpy_chk function to the builder.
AttributeSet getParamAttributes(unsigned ArgNo) const
The attributes for the argument or parameter at the given index are returned.
bool has(LibFunc F) const
Tests whether a library function is available.
Value * emitVSPrintf(Value *Dest, Value *Fmt, Value *VAList, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the vsprintf function.
LLVM_NODISCARD size_t size() const
size - Get the string size.
Definition: StringRef.h:144
SmallString - A SmallString is just a SmallVector with methods and accessors that make it work better...
Definition: SmallString.h:25
unsigned getOpcode() const
Returns a member of one of the enums like Instruction::Add.
Definition: Instruction.h:125
Value * CreateSub(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1135
AttributeList getAttributes() const
Return the attribute list for this Function.
Definition: Function.h:223
static const fltSemantics & IEEEdouble() LLVM_READNONE
Definition: APFloat.cpp:158
static Value * valueHasFloatPrecision(Value *Val)
Return a variant of Val with float type.
Value * CreateZExt(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1878
constexpr char Attrs[]
Key for Kernel::Metadata::mAttrs.
uint64_t GetStringLength(const Value *V, unsigned CharSize=8)
If we can compute the length of the string pointed to by the specified pointer, return &#39;len+1&#39;...
void takeName(Value *V)
Transfer the name from V to this value.
Definition: Value.cpp:291
Function * getDeclaration(Module *M, ID id, ArrayRef< Type *> Tys=None)
Create or insert an LLVM Function declaration for an intrinsic, and return it.
Definition: Function.cpp:1079
static Value * getIntToFPVal(Value *I2F, IRBuilder<> &B)
void SetInsertPoint(BasicBlock *TheBB)
This specifies that created instructions should be appended to the end of the specified block...
Definition: IRBuilder.h:132
Value * getOperand(unsigned i) const
Definition: User.h:169
ConstantInt * getIntN(unsigned N, uint64_t C)
Get a constant N-bit value, zero extended or truncated from a 64-bit value.
Definition: IRBuilder.h:354
Flag
These should be considered private to the implementation of the MCInstrDesc class.
Definition: MCInstrDesc.h:131
static bool isOnlyUsedInComparisonWithZero(Value *V)
bool getConstantDataArrayInfo(const Value *V, ConstantDataArraySlice &Slice, unsigned ElementSize, uint64_t Offset=0)
Returns true if the value V is a pointer into a ConstantDataArray.
static Constant * getBitCast(Constant *C, Type *Ty, bool OnlyIfReduced=false)
Definition: Constants.cpp:1804
Value * emitFPutS(Value *Str, Value *File, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the fputs function.
bool isFloatTy() const
Return true if this is &#39;float&#39;, a 32-bit IEEE fp type.
Definition: Type.h:146
const BasicBlock & getEntryBlock() const
Definition: Function.h:664
IntegerType * getIntPtrType(LLVMContext &C, unsigned AddressSpace=0) const
Returns an integer type with size at least as big as that of a pointer in the given address space...
Definition: DataLayout.cpp:774
bool inferLibFuncAttributes(Function &F, const TargetLibraryInfo &TLI)
Analyze the name and prototype of the given function and set any applicable attributes.
Type * getDoubleTy()
Fetch the type representing a 64-bit floating point value.
Definition: IRBuilder.h:411
BlockFrequencyInfo pass uses BlockFrequencyInfoImpl implementation to estimate IR basic block frequen...
C - The default llvm calling convention, compatible with C.
Definition: CallingConv.h:34
OneUse_match< T > m_OneUse(const T &SubPattern)
Definition: PatternMatch.h:61
initializer< Ty > init(const Ty &Val)
Definition: CommandLine.h:432
uint64_t getDereferenceableOrNullBytes(unsigned i) const
Extract the number of dereferenceable_or_null bytes for a call or parameter (0=unknown).
Definition: InstrTypes.h:1591
static bool isLocallyOpenedFile(Value *File, CallInst *CI, IRBuilder<> &B, const TargetLibraryInfo *TLI)
uint64_t getZExtValue() const
Return the constant as a 64-bit unsigned integer value after it has been zero extended as appropriate...
Definition: Constants.h:148
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
Value * CreateFMul(Value *L, Value *R, const Twine &Name="", MDNode *FPMD=nullptr)
Definition: IRBuilder.h:1383
void setNoSignedZeros(bool B=true)
Definition: Operator.h:224
apint_match m_APInt(const APInt *&Res)
Match a ConstantInt or splatted ConstantVector, binding the specified pointer to the contained APInt...
Definition: PatternMatch.h:189
bool isNegative() const
Definition: APFloat.h:1162
LLVM Basic Block Representation.
Definition: BasicBlock.h:57
The instances of the Type class are immutable: once they are created, they are never changed...
Definition: Type.h:45
Value * CreateConstInBoundsGEP1_64(Type *Ty, Value *Ptr, uint64_t Idx0, const Twine &Name="")
Definition: IRBuilder.h:1799
ConstantInt * getTrue()
Get the constant value for i1 true.
Definition: IRBuilder.h:323
bool isNoBuiltin() const
Return true if the call should not be treated as a call to a builtin.
Definition: InstrTypes.h:1611
Instrumentation for Order File
ARM_AAPCS - ARM Architecture Procedure Calling Standard calling convention (aka EABI).
Definition: CallingConv.h:99
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
This is an important base class in LLVM.
Definition: Constant.h:41
Value * CreateSelect(Value *C, Value *True, Value *False, const Twine &Name="", Instruction *MDFrom=nullptr)
Definition: IRBuilder.h:2285
bool isPointerTy() const
True if this is an instance of PointerType.
Definition: Type.h:223
LLVM_NODISCARD size_t find_first_not_of(char C, size_t From=0) const
Find the first character in the string that is not C or npos if not found.
Definition: StringRef.cpp:249
FunctionType * getFunctionType() const
Definition: InstrTypes.h:1144
ConstantFP - Floating Point Values [float, double].
Definition: Constants.h:263
double convertToDouble() const
Definition: APFloat.h:1112
Diagnostic information for applied optimization remarks.
LLVM_NODISCARD size_t find(char C, size_t From=0) const
Search for the first character C in the string.
Definition: StringRef.h:299
LLVM_NODISCARD AttributeList addParamAttributes(LLVMContext &C, unsigned ArgNo, const AttrBuilder &B) const
Add an argument attribute to the list.
Definition: Attributes.h:434
unsigned getPrefTypeAlignment(Type *Ty) const
Returns the preferred stack/global alignment for the specified type.
Definition: DataLayout.cpp:770
bool isExactlyValue(double V) const
We don&#39;t rely on operator== working on double values, as it returns true for things that are clearly ...
Definition: APFloat.h:1145
bool isFast() const
Determine whether all fast-math-flags are set.
bool any_of(R &&range, UnaryPredicate P)
Provide wrappers to std::any_of which take ranges instead of having to pass begin/end explicitly...
Definition: STLExtras.h:1172
Value * CreateNeg(Value *V, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1485
This instruction compares its operands according to the predicate given to the constructor.
amdgpu Simplify well known AMD library false FunctionCallee Value * Arg
op_range operands()
Definition: User.h:237
Value * getPointerOperand()
Definition: Instructions.h:284
static bool isCallingConvCCompatible(CallInst *CI)
Class to represent integer types.
Definition: DerivedTypes.h:40
ConstantInt * getInt64(uint64_t C)
Get a constant 64-bit value.
Definition: IRBuilder.h:348
static Value * createPowWithIntegerExponent(Value *Base, Value *Expo, Module *M, IRBuilder<> &B)
Value * castToCStr(Value *V, IRBuilder<> &B)
Return V if it is an i8*, otherwise cast it to i8*.
IntegerType * getIntNTy(unsigned N)
Fetch the type representing an N-bit integer.
Definition: IRBuilder.h:396
Represents offset+length into a ConstantDataArray.
Value * CreateExtractElement(Value *Vec, Value *Idx, const Twine &Name="")
Definition: IRBuilder.h:2307
const Function * getFunction() const
Return the function this instruction belongs to.
Definition: Instruction.cpp:59
uint64_t getDereferenceableBytes(unsigned i) const
Extract the number of dereferenceable bytes for a call or parameter (0=unknown).
Definition: InstrTypes.h:1585
Constant * ConstantFoldLoadFromConstPtr(Constant *C, Type *Ty, const DataLayout &DL)
ConstantFoldLoadFromConstPtr - Return the value that a load from C would produce if it is constant an...
static double log2(double V)
static Value * optimizeMemCmpConstantSize(CallInst *CI, Value *LHS, Value *RHS, uint64_t Len, IRBuilder<> &B, const DataLayout &DL)
bool isIntN(unsigned N, int64_t x)
Checks if an signed integer fits into the given (dynamic) bit width.
Definition: MathExtras.h:397
uint64_t NextPowerOf2(uint64_t A)
Returns the next power of two (in 64-bits) that is strictly greater than A.
Definition: MathExtras.h:644
Value * emitStrChr(Value *Ptr, char C, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the strchr function to the builder, for the specified pointer and character.
Value * CreateExtractValue(Value *Agg, ArrayRef< unsigned > Idxs, const Twine &Name="")
Definition: IRBuilder.h:2349
LLVM_NODISCARD char back() const
back - Get the last character in the string.
Definition: StringRef.h:155
bool hasFnAttr(Attribute::AttrKind Kind) const
Determine whether this call has the given attribute.
Definition: InstrTypes.h:1373
static bool isBinary(MachineInstr &MI)
LLVM_READONLY APFloat maxnum(const APFloat &A, const APFloat &B)
Implements IEEE maxNum semantics.
Definition: APFloat.h:1253
Value * CreateTrunc(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1874
Triple - Helper class for working with autoconf configuration names.
Definition: Triple.h:43
Value * emitFWriteUnlocked(Value *Ptr, Value *Size, Value *N, Value *File, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the fwrite_unlocked function.
Value * CreateGEP(Value *Ptr, ArrayRef< Value *> IdxList, const Twine &Name="")
Definition: IRBuilder.h:1677
Value * emitStrLen(Value *Ptr, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the strlen function to the builder, for the specified pointer.
static IntegerType * get(LLVMContext &C, unsigned NumBits)
This static method is the primary way of constructing an IntegerType.
Definition: Type.cpp:239
Value * emitBinaryFloatFnCall(Value *Op1, Value *Op2, StringRef Name, IRBuilder<> &B, const AttributeList &Attrs)
Emit a call to the binary function named &#39;Name&#39; (e.g.
static const fltSemantics & IEEEsingle() LLVM_READNONE
Definition: APFloat.cpp:155
void addParamAttr(unsigned ArgNo, Attribute::AttrKind Kind)
Adds the attribute to the indicated argument.
Definition: InstrTypes.h:1397
static Value * optimizeDoubleFP(CallInst *CI, IRBuilder<> &B, bool isBinary, bool isPrecise=false)
Shrink double -> float functions.
bool isInteger() const
Definition: APFloat.h:1177
Value * emitStrNCpy(Value *Dst, Value *Src, Value *Len, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the strncpy function to the builder, for the specified pointer arguments and length...
Value * emitFGetCUnlocked(Value *File, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the fgetc_unlocked function. File is a pointer to FILE.
This is the shared class of boolean and integer constants.
Definition: Constants.h:83
GlobalVariable * CreateGlobalString(StringRef Str, const Twine &Name="", unsigned AddressSpace=0)
Make a new global variable with initializer type i8*.
Definition: IRBuilder.cpp:42
Value * emitStrCat(Value *Dest, Value *Src, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the strcat function.
Value * emitStrCpy(Value *Dst, Value *Src, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the strcpy function to the builder, for the specified pointer arguments.
Value * CreateIntCast(Value *V, Type *DestTy, bool isSigned, const Twine &Name="")
Definition: IRBuilder.h:2033
This is a &#39;vector&#39; (really, a variable-sized array), optimized for the case when the array is small...
Definition: SmallVector.h:837
Value * emitVSNPrintf(Value *Dest, Value *Size, Value *Fmt, Value *VAList, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the vsnprintf function.
uint64_t getElementAsInteger(unsigned i) const
If this is a sequential container of integers (of any size), return the specified element in the low ...
Definition: Constants.cpp:2692
Module.h This file contains the declarations for the Module class.
uint64_t Length
Length of the slice.
Provides information about what library functions are available for the current target.
bool isLegalInteger(uint64_t Width) const
Returns true if the specified type is known to be a native integer type supported by the CPU...
Definition: DataLayout.h:255
ConstantInt * getInt32(uint32_t C)
Get a constant 32-bit value.
Definition: IRBuilder.h:343
void getOperandBundlesAsDefs(SmallVectorImpl< OperandBundleDef > &Defs) const
Return the list of operand bundles attached to this instruction as a vector of OperandBundleDefs.
Definition: InstrTypes.h:1843
CallInst * CreateMemCpy(Value *Dst, unsigned DstAlign, Value *Src, unsigned SrcAlign, uint64_t Size, bool isVolatile=false, MDNode *TBAATag=nullptr, MDNode *TBAAStructTag=nullptr, MDNode *ScopeTag=nullptr, MDNode *NoAliasTag=nullptr)
Create and insert a memcpy between the specified pointers.
Definition: IRBuilder.h:482
static Constant * get(Type *Ty, uint64_t V, bool isSigned=false)
If Ty is a vector type, return a Constant with a splat of the given value.
Definition: Constants.cpp:653
FunctionCallee getOrInsertFunction(StringRef Name, FunctionType *T, AttributeList AttributeList)
Look up the specified function in the module symbol table.
Definition: Module.cpp:143
static Constant * get(Type *Ty, double V)
This returns a ConstantFP, or a vector containing a splat of a ConstantFP, for the specified value in...
Definition: Constants.cpp:716
AttributeList getAttributes() const
Return the parameter attributes for this call.
Definition: InstrTypes.h:1366
BinaryOp_match< LHS, RHS, Instruction::FMul > m_FMul(const LHS &L, const RHS &R)
Definition: PatternMatch.h:789
bool NullPointerIsDefined(const Function *F, unsigned AS=0)
Check whether null pointer dereferencing is considered undefined behavior for a given function or an ...
Definition: Function.cpp:1580
Intrinsic::ID getIntrinsicID() const LLVM_READONLY
getIntrinsicID - This method returns the ID number of the specified function, or Intrinsic::not_intri...
Definition: Function.h:193
unsigned logBase2() const
Definition: APInt.h:1756
The access may modify the value stored in memory.
FunctionType * getFunctionType() const
Returns the FunctionType for me.
Definition: Function.h:163
const Module * getModule() const
Return the module owning the function this instruction belongs to or nullptr it the function does not...
Definition: Instruction.cpp:55
Class for arbitrary precision integers.
Definition: APInt.h:69
bool ule(const APInt &RHS) const
Unsigned less or equal comparison.
Definition: APInt.h:1222
amdgpu Simplify well known AMD library false FunctionCallee Callee
bool isKnownNonZero(const Value *V, const DataLayout &DL, unsigned Depth=0, AssumptionCache *AC=nullptr, const Instruction *CxtI=nullptr, const DominatorTree *DT=nullptr, bool UseInstrInfo=true)
Return true if the given value is known to be non-zero when defined.
bool isPowerOf2() const
Check if this APInt&#39;s value is a power of two greater than zero.
Definition: APInt.h:463
iterator_range< user_iterator > users()
Definition: Value.h:419
IntegerType * getInt8Ty()
Fetch the type representing an 8-bit integer.
Definition: IRBuilder.h:373
Value * CreateShl(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1207
bool getLibFunc(StringRef funcName, LibFunc &F) const
Searches for a particular function name.
iterator begin() const
Definition: StringRef.h:115
Value * emitFPutC(Value *Char, Value *File, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the fputc function.
ConstantInt * getFalse()
Get the constant value for i1 false.
Definition: IRBuilder.h:328
Value * emitFGetSUnlocked(Value *Str, Value *Size, Value *File, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the fgets_unlocked function.
User::op_iterator arg_begin()
Return the iterator pointing to the beginning of the argument list.
Definition: InstrTypes.h:1206
specific_fpval m_FPOne()
Match a float 1.0 or vector with all elements equal to 1.0.
Definition: PatternMatch.h:617
opStatus add(const APFloat &RHS, roundingMode RM)
Definition: APFloat.h:956
Predicate getPredicate() const
Return the predicate for this instruction.
Definition: InstrTypes.h:807
FNeg_match< OpTy > m_FNeg(const OpTy &X)
Match &#39;fneg X&#39; as &#39;fsub -0.0, X&#39;.
Definition: PatternMatch.h:771
unsigned getNumArgOperands() const
Definition: InstrTypes.h:1239
Merge contiguous icmps into a memcmp
Definition: MergeICmps.cpp:927
static const size_t npos
Definition: StringRef.h:50
unsigned getIntegerBitWidth() const
Definition: DerivedTypes.h:97
LLVM_NODISCARD bool empty() const
Definition: SmallVector.h:55
static Value * optimizeUnaryDoubleFP(CallInst *CI, IRBuilder<> &B, bool isPrecise=false)
Shrink double -> float for unary functions.
Value * emitStrNCmp(Value *Ptr1, Value *Ptr2, Value *Len, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the strncmp function to the builder.
CallingConv::ID getCallingConv() const
Definition: InstrTypes.h:1344
static VectorType * get(Type *ElementType, ElementCount EC)
This static method is the primary way to construct an VectorType.
Definition: Type.cpp:609
StringRef getName() const
Return a constant reference to the value&#39;s name.
Definition: Value.cpp:214
Function * getCalledFunction() const
Returns the function called, or null if this is an indirect function invocation.
Definition: InstrTypes.h:1287
const Function * getParent() const
Return the enclosing method, or null if none.
Definition: BasicBlock.h:106
LLVM_NODISCARD size_t find_first_of(char C, size_t From=0) const
Find the first character in the string that is C, or npos if not found.
Definition: StringRef.h:394
#define I(x, y, z)
Definition: MD5.cpp:58
#define N
APFloat abs(APFloat X)
Returns the absolute value of the argument.
Definition: APFloat.h:1228
bool isNormal() const
Definition: APFloat.h:1166
bool isZero() const
This is just a convenience method to make client code smaller for a common code.
Definition: Constants.h:192
static Value * optimizeBinaryDoubleFP(CallInst *CI, IRBuilder<> &B, bool isPrecise=false)
Shrink double -> float for binary functions.
Value * optimizeCall(CallInst *CI)
optimizeCall - Take the given call instruction and return a more optimal value to replace the instruc...
Value * emitStrLCat(Value *Dest, Value *Src, Value *Size, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the strlcat function.
void setAttributes(AttributeList A)
Set the parameter attributes for this call.
Definition: InstrTypes.h:1370
void removeParamAttr(unsigned ArgNo, Attribute::AttrKind Kind)
Removes the attribute from the given argument.
Definition: InstrTypes.h:1427
unsigned getKnownAlignment(Value *V, const DataLayout &DL, const Instruction *CxtI=nullptr, AssumptionCache *AC=nullptr, const DominatorTree *DT=nullptr)
Try to infer an alignment for the specified pointer.
Definition: Local.h:267
LLVM_NODISCARD std::enable_if<!is_simple_type< Y >::value, typename cast_retty< X, const Y >::ret_type >::type dyn_cast(const Y &Val)
Definition: Casting.h:332
uint32_t Size
Definition: Profile.cpp:46
CallInst * CreateCall(FunctionType *FTy, Value *Callee, ArrayRef< Value *> Args=None, const Twine &Name="", MDNode *FPMathTag=nullptr)
Definition: IRBuilder.h:2237
static void insertSinCosCall(IRBuilder<> &B, Function *OrigCallee, Value *Arg, bool UseFloat, Value *&Sin, Value *&Cos, Value *&SinCos)
Value * emitBCmp(Value *Ptr1, Value *Ptr2, Value *Len, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the bcmp function.
Type * getFloatTy()
Fetch the type representing a 32-bit floating point value.
Definition: IRBuilder.h:406
Value * emitSPrintf(Value *Dest, Value *Fmt, ArrayRef< Value *> VariadicArgs, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the sprintf function.
Value * CreateFAdd(Value *L, Value *R, const Twine &Name="", MDNode *FPMD=nullptr)
Definition: IRBuilder.h:1333
bool isStrictFP() const
Determine if the call requires strict floating point semantics.
Definition: InstrTypes.h:1617
Value * CreateAnd(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1268
Value * CreateFDiv(Value *L, Value *R, const Twine &Name="", MDNode *FPMD=nullptr)
Definition: IRBuilder.h:1408
bool isDeclaration() const
Return true if the primary definition of this global value is outside of the current translation unit...
Definition: Globals.cpp:231
bool getConstantStringInfo(const Value *V, StringRef &Str, uint64_t Offset=0, bool TrimAtNul=true)
This function computes the length of a null-terminated C string pointed to by V.
Value * emitFWrite(Value *Ptr, Value *Size, Value *File, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the fwrite function.
InstTy * Insert(InstTy *I, const Twine &Name="") const
Insert and return the specified instruction.
Definition: IRBuilder.h:830
Value * emitMemCmp(Value *Ptr1, Value *Ptr2, Value *Len, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the memcmp function.
LLVM_NODISCARD const char * data() const
data - Get a pointer to the start of the string (which may not be null terminated).
Definition: StringRef.h:136
static bool isOnlyUsedInEqualityComparison(Value *V, Value *With)
Return true if it is only used in equality comparisons with With.
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
user_iterator user_begin()
Definition: Value.h:395
This class represents a truncation of floating point types.
Value * emitFReadUnlocked(Value *Ptr, Value *Size, Value *N, Value *File, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the fread_unlocked function.
unsigned getPrimitiveSizeInBits() const LLVM_READONLY
Return the basic size of this type if it is a primitive type.
Definition: Type.cpp:114
Module * getParent()
Get the module that this global value is contained inside of...
Definition: GlobalValue.h:575
LLVM Value Representation.
Definition: Value.h:73
Value * emitPutS(Value *Str, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the puts function. This assumes that Str is some pointer.
ARM_AAPCS_VFP - Same as ARM_AAPCS, but uses hard floating point ABI.
Definition: CallingConv.h:102
LLVM_NODISCARD StringRef drop_back(size_t N=1) const
Return a StringRef equal to &#39;this&#39; but with the last N elements dropped.
Definition: StringRef.h:642
Value * emitStpNCpy(Value *Dst, Value *Src, Value *Len, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the stpncpy function to the builder, for the specified pointer arguments and length...
Value * emitMemChr(Value *Ptr, Value *Val, Value *Len, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the memchr function.
void setFastMathFlags(FastMathFlags NewFMF)
Set the fast-math flags to be used with generated fp-math operators.
Definition: IRBuilder.h:225
static bool canTransformToMemCmp(CallInst *CI, Value *Str, uint64_t Len, const DataLayout &DL)
Value * emitCalloc(Value *Num, Value *Size, const AttributeList &Attrs, IRBuilder<> &B, const TargetLibraryInfo &TLI)
Emit a call to the calloc function.
static bool callHasFP128Argument(const CallInst *CI)
ConstantInt * getInt8(uint8_t C)
Get a constant 8-bit value.
Definition: IRBuilder.h:333
static Value * optimizeTrigReflections(CallInst *Call, LibFunc Func, IRBuilder<> &B)
bool hasNoNaNs() const
Determine whether the no-NaNs flag is set.
bool hasOneUse() const
Return true if there is exactly one user of this value.
Definition: Value.h:432
static Value * getSqrtCall(Value *V, AttributeList Attrs, bool NoErrno, Module *M, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Convenience struct for specifying and reasoning about fast-math flags.
Definition: Operator.h:159
StringRef - Represent a constant reference to a string, i.e.
Definition: StringRef.h:48
static void annotateNonNullAndDereferenceable(CallInst *CI, ArrayRef< unsigned > ArgNos, Value *Size, const DataLayout &DL)
int64_t getSExtValue() const
Return the constant as a 64-bit integer value after it has been sign extended as appropriate for the ...
Definition: Constants.h:156
cstfp_pred_ty< is_any_zero_fp > m_AnyZeroFP()
Match a floating-point negative zero or positive zero.
Definition: PatternMatch.h:510
ConstantInt * getInt(const APInt &AI)
Get a constant integer value.
Definition: IRBuilder.h:359
static bool ignoreCallingConv(LibFunc Func)
This class represents an extension of floating point types.
iterator end() const
Definition: StringRef.h:117
bool isDoubleTy() const
Return true if this is &#39;double&#39;, a 64-bit IEEE fp type.
Definition: Type.h:149
Value * emitStrNCat(Value *Dest, Value *Src, Value *Size, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the strncat function.
bool hasFloatFn(const TargetLibraryInfo *TLI, Type *Ty, LibFunc DoubleFn, LibFunc FloatFn, LibFunc LongDoubleFn)
Check whether the overloaded floating point function corresponding to Ty is available.
The optimization diagnostic interface.
bool use_empty() const
Definition: Value.h:342
static Value * getPow(Value *InnerChain[33], unsigned Exp, IRBuilder<> &B)
constexpr char Args[]
Key for Kernel::Metadata::mArgs.
Value * CreateFNeg(Value *V, const Twine &Name="", MDNode *FPMathTag=nullptr)
Definition: IRBuilder.h:1503
LLVM_READONLY APFloat minnum(const APFloat &A, const APFloat &B)
Implements IEEE minNum semantics.
Definition: APFloat.h:1242
bool isStructTy() const
True if this is an instance of StructType.
Definition: Type.h:217
cmpResult compare(const APFloat &RHS) const
Definition: APFloat.h:1117
bool isArrayTy() const
True if this is an instance of ArrayType.
Definition: Type.h:220
A wrapper class for inspecting calls to intrinsic functions.
Definition: IntrinsicInst.h:43
const BasicBlock * getParent() const
Definition: Instruction.h:66
bool PointerMayBeCaptured(const Value *V, bool ReturnCaptures, bool StoreCaptures, unsigned MaxUsesToExplore=DefaultMaxUsesToExplore)
PointerMayBeCaptured - Return true if this pointer value may be captured by the enclosing function (w...
user_iterator user_end()
Definition: Value.h:403