LLVM  8.0.0svn
SimplifyLibCalls.cpp
Go to the documentation of this file.
1 //===------ SimplifyLibCalls.cpp - Library calls simplifier ---------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements the library calls simplifier. It does not implement
11 // any pass, but can't be used by other passes to do simplifications.
12 //
13 //===----------------------------------------------------------------------===//
14 
16 #include "llvm/ADT/SmallString.h"
17 #include "llvm/ADT/StringMap.h"
18 #include "llvm/ADT/Triple.h"
25 #include "llvm/Analysis/Loads.h"
26 #include "llvm/IR/DataLayout.h"
27 #include "llvm/IR/Function.h"
28 #include "llvm/IR/IRBuilder.h"
29 #include "llvm/IR/IntrinsicInst.h"
30 #include "llvm/IR/Intrinsics.h"
31 #include "llvm/IR/LLVMContext.h"
32 #include "llvm/IR/Module.h"
33 #include "llvm/IR/PatternMatch.h"
35 #include "llvm/Support/KnownBits.h"
37 
38 using namespace llvm;
39 using namespace PatternMatch;
40 
41 static cl::opt<bool>
42  EnableUnsafeFPShrink("enable-double-float-shrink", cl::Hidden,
43  cl::init(false),
44  cl::desc("Enable unsafe double to float "
45  "shrinking for math lib calls"));
46 
47 
48 //===----------------------------------------------------------------------===//
49 // Helper Functions
50 //===----------------------------------------------------------------------===//
51 
52 static bool ignoreCallingConv(LibFunc Func) {
53  return Func == LibFunc_abs || Func == LibFunc_labs ||
54  Func == LibFunc_llabs || Func == LibFunc_strlen;
55 }
56 
58  switch(CI->getCallingConv()) {
59  default:
60  return false;
62  return true;
66 
67  // The iOS ABI diverges from the standard in some cases, so for now don't
68  // try to simplify those calls.
69  if (Triple(CI->getModule()->getTargetTriple()).isiOS())
70  return false;
71 
72  auto *FuncTy = CI->getFunctionType();
73 
74  if (!FuncTy->getReturnType()->isPointerTy() &&
75  !FuncTy->getReturnType()->isIntegerTy() &&
76  !FuncTy->getReturnType()->isVoidTy())
77  return false;
78 
79  for (auto Param : FuncTy->params()) {
80  if (!Param->isPointerTy() && !Param->isIntegerTy())
81  return false;
82  }
83  return true;
84  }
85  }
86  return false;
87 }
88 
89 /// Return true if it is only used in equality comparisons with With.
90 static bool isOnlyUsedInEqualityComparison(Value *V, Value *With) {
91  for (User *U : V->users()) {
92  if (ICmpInst *IC = dyn_cast<ICmpInst>(U))
93  if (IC->isEquality() && IC->getOperand(1) == With)
94  continue;
95  // Unknown instruction.
96  return false;
97  }
98  return true;
99 }
100 
101 static bool callHasFloatingPointArgument(const CallInst *CI) {
102  return any_of(CI->operands(), [](const Use &OI) {
103  return OI->getType()->isFloatingPointTy();
104  });
105 }
106 
107 static Value *convertStrToNumber(CallInst *CI, StringRef &Str, int64_t Base) {
108  if (Base < 2 || Base > 36)
109  // handle special zero base
110  if (Base != 0)
111  return nullptr;
112 
113  char *End;
114  std::string nptr = Str.str();
115  errno = 0;
116  long long int Result = strtoll(nptr.c_str(), &End, Base);
117  if (errno)
118  return nullptr;
119 
120  // if we assume all possible target locales are ASCII supersets,
121  // then if strtoll successfully parses a number on the host,
122  // it will also successfully parse the same way on the target
123  if (*End != '\0')
124  return nullptr;
125 
126  if (!isIntN(CI->getType()->getPrimitiveSizeInBits(), Result))
127  return nullptr;
128 
129  return ConstantInt::get(CI->getType(), Result);
130 }
131 
133  const TargetLibraryInfo *TLI) {
134  CallInst *FOpen = dyn_cast<CallInst>(File);
135  if (!FOpen)
136  return false;
137 
138  Function *InnerCallee = FOpen->getCalledFunction();
139  if (!InnerCallee)
140  return false;
141 
142  LibFunc Func;
143  if (!TLI->getLibFunc(*InnerCallee, Func) || !TLI->has(Func) ||
144  Func != LibFunc_fopen)
145  return false;
146 
148  if (PointerMayBeCaptured(File, true, true))
149  return false;
150 
151  return true;
152 }
153 
155  for (User *U : V->users()) {
156  if (ICmpInst *IC = dyn_cast<ICmpInst>(U))
157  if (Constant *C = dyn_cast<Constant>(IC->getOperand(1)))
158  if (C->isNullValue())
159  continue;
160  // Unknown instruction.
161  return false;
162  }
163  return true;
164 }
165 
166 static bool canTransformToMemCmp(CallInst *CI, Value *Str, uint64_t Len,
167  const DataLayout &DL) {
169  return false;
170 
171  if (!isDereferenceableAndAlignedPointer(Str, 1, APInt(64, Len), DL))
172  return false;
173 
174  return true;
175 }
176 
177 //===----------------------------------------------------------------------===//
178 // String and Memory Library Call Optimizations
179 //===----------------------------------------------------------------------===//
180 
181 Value *LibCallSimplifier::optimizeStrCat(CallInst *CI, IRBuilder<> &B) {
182  // Extract some information from the instruction
183  Value *Dst = CI->getArgOperand(0);
184  Value *Src = CI->getArgOperand(1);
185 
186  // See if we can get the length of the input string.
187  uint64_t Len = GetStringLength(Src);
188  if (Len == 0)
189  return nullptr;
190  --Len; // Unbias length.
191 
192  // Handle the simple, do-nothing case: strcat(x, "") -> x
193  if (Len == 0)
194  return Dst;
195 
196  return emitStrLenMemCpy(Src, Dst, Len, B);
197 }
198 
199 Value *LibCallSimplifier::emitStrLenMemCpy(Value *Src, Value *Dst, uint64_t Len,
200  IRBuilder<> &B) {
201  // We need to find the end of the destination string. That's where the
202  // memory is to be moved to. We just generate a call to strlen.
203  Value *DstLen = emitStrLen(Dst, B, DL, TLI);
204  if (!DstLen)
205  return nullptr;
206 
207  // Now that we have the destination's length, we must index into the
208  // destination's pointer to get the actual memcpy destination (end of
209  // the string .. we're concatenating).
210  Value *CpyDst = B.CreateGEP(B.getInt8Ty(), Dst, DstLen, "endptr");
211 
212  // We have enough information to now generate the memcpy call to do the
213  // concatenation for us. Make a memcpy to copy the nul byte with align = 1.
214  B.CreateMemCpy(CpyDst, 1, Src, 1,
215  ConstantInt::get(DL.getIntPtrType(Src->getContext()), Len + 1));
216  return Dst;
217 }
218 
219 Value *LibCallSimplifier::optimizeStrNCat(CallInst *CI, IRBuilder<> &B) {
220  // Extract some information from the instruction.
221  Value *Dst = CI->getArgOperand(0);
222  Value *Src = CI->getArgOperand(1);
223  uint64_t Len;
224 
225  // We don't do anything if length is not constant.
226  if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(CI->getArgOperand(2)))
227  Len = LengthArg->getZExtValue();
228  else
229  return nullptr;
230 
231  // See if we can get the length of the input string.
232  uint64_t SrcLen = GetStringLength(Src);
233  if (SrcLen == 0)
234  return nullptr;
235  --SrcLen; // Unbias length.
236 
237  // Handle the simple, do-nothing cases:
238  // strncat(x, "", c) -> x
239  // strncat(x, c, 0) -> x
240  if (SrcLen == 0 || Len == 0)
241  return Dst;
242 
243  // We don't optimize this case.
244  if (Len < SrcLen)
245  return nullptr;
246 
247  // strncat(x, s, c) -> strcat(x, s)
248  // s is constant so the strcat can be optimized further.
249  return emitStrLenMemCpy(Src, Dst, SrcLen, B);
250 }
251 
252 Value *LibCallSimplifier::optimizeStrChr(CallInst *CI, IRBuilder<> &B) {
254  FunctionType *FT = Callee->getFunctionType();
255  Value *SrcStr = CI->getArgOperand(0);
256 
257  // If the second operand is non-constant, see if we can compute the length
258  // of the input string and turn this into memchr.
259  ConstantInt *CharC = dyn_cast<ConstantInt>(CI->getArgOperand(1));
260  if (!CharC) {
261  uint64_t Len = GetStringLength(SrcStr);
262  if (Len == 0 || !FT->getParamType(1)->isIntegerTy(32)) // memchr needs i32.
263  return nullptr;
264 
265  return emitMemChr(SrcStr, CI->getArgOperand(1), // include nul.
266  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len),
267  B, DL, TLI);
268  }
269 
270  // Otherwise, the character is a constant, see if the first argument is
271  // a string literal. If so, we can constant fold.
272  StringRef Str;
273  if (!getConstantStringInfo(SrcStr, Str)) {
274  if (CharC->isZero()) // strchr(p, 0) -> p + strlen(p)
275  return B.CreateGEP(B.getInt8Ty(), SrcStr, emitStrLen(SrcStr, B, DL, TLI),
276  "strchr");
277  return nullptr;
278  }
279 
280  // Compute the offset, make sure to handle the case when we're searching for
281  // zero (a weird way to spell strlen).
282  size_t I = (0xFF & CharC->getSExtValue()) == 0
283  ? Str.size()
284  : Str.find(CharC->getSExtValue());
285  if (I == StringRef::npos) // Didn't find the char. strchr returns null.
286  return Constant::getNullValue(CI->getType());
287 
288  // strchr(s+n,c) -> gep(s+n+i,c)
289  return B.CreateGEP(B.getInt8Ty(), SrcStr, B.getInt64(I), "strchr");
290 }
291 
292 Value *LibCallSimplifier::optimizeStrRChr(CallInst *CI, IRBuilder<> &B) {
293  Value *SrcStr = CI->getArgOperand(0);
294  ConstantInt *CharC = dyn_cast<ConstantInt>(CI->getArgOperand(1));
295 
296  // Cannot fold anything if we're not looking for a constant.
297  if (!CharC)
298  return nullptr;
299 
300  StringRef Str;
301  if (!getConstantStringInfo(SrcStr, Str)) {
302  // strrchr(s, 0) -> strchr(s, 0)
303  if (CharC->isZero())
304  return emitStrChr(SrcStr, '\0', B, TLI);
305  return nullptr;
306  }
307 
308  // Compute the offset.
309  size_t I = (0xFF & CharC->getSExtValue()) == 0
310  ? Str.size()
311  : Str.rfind(CharC->getSExtValue());
312  if (I == StringRef::npos) // Didn't find the char. Return null.
313  return Constant::getNullValue(CI->getType());
314 
315  // strrchr(s+n,c) -> gep(s+n+i,c)
316  return B.CreateGEP(B.getInt8Ty(), SrcStr, B.getInt64(I), "strrchr");
317 }
318 
319 Value *LibCallSimplifier::optimizeStrCmp(CallInst *CI, IRBuilder<> &B) {
320  Value *Str1P = CI->getArgOperand(0), *Str2P = CI->getArgOperand(1);
321  if (Str1P == Str2P) // strcmp(x,x) -> 0
322  return ConstantInt::get(CI->getType(), 0);
323 
324  StringRef Str1, Str2;
325  bool HasStr1 = getConstantStringInfo(Str1P, Str1);
326  bool HasStr2 = getConstantStringInfo(Str2P, Str2);
327 
328  // strcmp(x, y) -> cnst (if both x and y are constant strings)
329  if (HasStr1 && HasStr2)
330  return ConstantInt::get(CI->getType(), Str1.compare(Str2));
331 
332  if (HasStr1 && Str1.empty()) // strcmp("", x) -> -*x
333  return B.CreateNeg(
334  B.CreateZExt(B.CreateLoad(Str2P, "strcmpload"), CI->getType()));
335 
336  if (HasStr2 && Str2.empty()) // strcmp(x,"") -> *x
337  return B.CreateZExt(B.CreateLoad(Str1P, "strcmpload"), CI->getType());
338 
339  // strcmp(P, "x") -> memcmp(P, "x", 2)
340  uint64_t Len1 = GetStringLength(Str1P);
341  uint64_t Len2 = GetStringLength(Str2P);
342  if (Len1 && Len2) {
343  return emitMemCmp(Str1P, Str2P,
344  ConstantInt::get(DL.getIntPtrType(CI->getContext()),
345  std::min(Len1, Len2)),
346  B, DL, TLI);
347  }
348 
349  // strcmp to memcmp
350  if (!HasStr1 && HasStr2) {
351  if (canTransformToMemCmp(CI, Str1P, Len2, DL))
352  return emitMemCmp(
353  Str1P, Str2P,
354  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len2), B, DL,
355  TLI);
356  } else if (HasStr1 && !HasStr2) {
357  if (canTransformToMemCmp(CI, Str2P, Len1, DL))
358  return emitMemCmp(
359  Str1P, Str2P,
360  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len1), B, DL,
361  TLI);
362  }
363 
364  return nullptr;
365 }
366 
367 Value *LibCallSimplifier::optimizeStrNCmp(CallInst *CI, IRBuilder<> &B) {
368  Value *Str1P = CI->getArgOperand(0), *Str2P = CI->getArgOperand(1);
369  if (Str1P == Str2P) // strncmp(x,x,n) -> 0
370  return ConstantInt::get(CI->getType(), 0);
371 
372  // Get the length argument if it is constant.
373  uint64_t Length;
374  if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(CI->getArgOperand(2)))
375  Length = LengthArg->getZExtValue();
376  else
377  return nullptr;
378 
379  if (Length == 0) // strncmp(x,y,0) -> 0
380  return ConstantInt::get(CI->getType(), 0);
381 
382  if (Length == 1) // strncmp(x,y,1) -> memcmp(x,y,1)
383  return emitMemCmp(Str1P, Str2P, CI->getArgOperand(2), B, DL, TLI);
384 
385  StringRef Str1, Str2;
386  bool HasStr1 = getConstantStringInfo(Str1P, Str1);
387  bool HasStr2 = getConstantStringInfo(Str2P, Str2);
388 
389  // strncmp(x, y) -> cnst (if both x and y are constant strings)
390  if (HasStr1 && HasStr2) {
391  StringRef SubStr1 = Str1.substr(0, Length);
392  StringRef SubStr2 = Str2.substr(0, Length);
393  return ConstantInt::get(CI->getType(), SubStr1.compare(SubStr2));
394  }
395 
396  if (HasStr1 && Str1.empty()) // strncmp("", x, n) -> -*x
397  return B.CreateNeg(
398  B.CreateZExt(B.CreateLoad(Str2P, "strcmpload"), CI->getType()));
399 
400  if (HasStr2 && Str2.empty()) // strncmp(x, "", n) -> *x
401  return B.CreateZExt(B.CreateLoad(Str1P, "strcmpload"), CI->getType());
402 
403  uint64_t Len1 = GetStringLength(Str1P);
404  uint64_t Len2 = GetStringLength(Str2P);
405 
406  // strncmp to memcmp
407  if (!HasStr1 && HasStr2) {
408  Len2 = std::min(Len2, Length);
409  if (canTransformToMemCmp(CI, Str1P, Len2, DL))
410  return emitMemCmp(
411  Str1P, Str2P,
412  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len2), B, DL,
413  TLI);
414  } else if (HasStr1 && !HasStr2) {
415  Len1 = std::min(Len1, Length);
416  if (canTransformToMemCmp(CI, Str2P, Len1, DL))
417  return emitMemCmp(
418  Str1P, Str2P,
419  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len1), B, DL,
420  TLI);
421  }
422 
423  return nullptr;
424 }
425 
426 Value *LibCallSimplifier::optimizeStrCpy(CallInst *CI, IRBuilder<> &B) {
427  Value *Dst = CI->getArgOperand(0), *Src = CI->getArgOperand(1);
428  if (Dst == Src) // strcpy(x,x) -> x
429  return Src;
430 
431  // See if we can get the length of the input string.
432  uint64_t Len = GetStringLength(Src);
433  if (Len == 0)
434  return nullptr;
435 
436  // We have enough information to now generate the memcpy call to do the
437  // copy for us. Make a memcpy to copy the nul byte with align = 1.
438  B.CreateMemCpy(Dst, 1, Src, 1,
439  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len));
440  return Dst;
441 }
442 
443 Value *LibCallSimplifier::optimizeStpCpy(CallInst *CI, IRBuilder<> &B) {
445  Value *Dst = CI->getArgOperand(0), *Src = CI->getArgOperand(1);
446  if (Dst == Src) { // stpcpy(x,x) -> x+strlen(x)
447  Value *StrLen = emitStrLen(Src, B, DL, TLI);
448  return StrLen ? B.CreateInBoundsGEP(B.getInt8Ty(), Dst, StrLen) : nullptr;
449  }
450 
451  // See if we can get the length of the input string.
452  uint64_t Len = GetStringLength(Src);
453  if (Len == 0)
454  return nullptr;
455 
456  Type *PT = Callee->getFunctionType()->getParamType(0);
457  Value *LenV = ConstantInt::get(DL.getIntPtrType(PT), Len);
458  Value *DstEnd = B.CreateGEP(B.getInt8Ty(), Dst,
459  ConstantInt::get(DL.getIntPtrType(PT), Len - 1));
460 
461  // We have enough information to now generate the memcpy call to do the
462  // copy for us. Make a memcpy to copy the nul byte with align = 1.
463  B.CreateMemCpy(Dst, 1, Src, 1, LenV);
464  return DstEnd;
465 }
466 
467 Value *LibCallSimplifier::optimizeStrNCpy(CallInst *CI, IRBuilder<> &B) {
469  Value *Dst = CI->getArgOperand(0);
470  Value *Src = CI->getArgOperand(1);
471  Value *LenOp = CI->getArgOperand(2);
472 
473  // See if we can get the length of the input string.
474  uint64_t SrcLen = GetStringLength(Src);
475  if (SrcLen == 0)
476  return nullptr;
477  --SrcLen;
478 
479  if (SrcLen == 0) {
480  // strncpy(x, "", y) -> memset(align 1 x, '\0', y)
481  B.CreateMemSet(Dst, B.getInt8('\0'), LenOp, 1);
482  return Dst;
483  }
484 
485  uint64_t Len;
486  if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(LenOp))
487  Len = LengthArg->getZExtValue();
488  else
489  return nullptr;
490 
491  if (Len == 0)
492  return Dst; // strncpy(x, y, 0) -> x
493 
494  // Let strncpy handle the zero padding
495  if (Len > SrcLen + 1)
496  return nullptr;
497 
498  Type *PT = Callee->getFunctionType()->getParamType(0);
499  // strncpy(x, s, c) -> memcpy(align 1 x, align 1 s, c) [s and c are constant]
500  B.CreateMemCpy(Dst, 1, Src, 1, ConstantInt::get(DL.getIntPtrType(PT), Len));
501 
502  return Dst;
503 }
504 
505 Value *LibCallSimplifier::optimizeStringLength(CallInst *CI, IRBuilder<> &B,
506  unsigned CharSize) {
507  Value *Src = CI->getArgOperand(0);
508 
509  // Constant folding: strlen("xyz") -> 3
510  if (uint64_t Len = GetStringLength(Src, CharSize))
511  return ConstantInt::get(CI->getType(), Len - 1);
512 
513  // If s is a constant pointer pointing to a string literal, we can fold
514  // strlen(s + x) to strlen(s) - x, when x is known to be in the range
515  // [0, strlen(s)] or the string has a single null terminator '\0' at the end.
516  // We only try to simplify strlen when the pointer s points to an array
517  // of i8. Otherwise, we would need to scale the offset x before doing the
518  // subtraction. This will make the optimization more complex, and it's not
519  // very useful because calling strlen for a pointer of other types is
520  // very uncommon.
521  if (GEPOperator *GEP = dyn_cast<GEPOperator>(Src)) {
522  if (!isGEPBasedOnPointerToString(GEP, CharSize))
523  return nullptr;
524 
526  if (getConstantDataArrayInfo(GEP->getOperand(0), Slice, CharSize)) {
527  uint64_t NullTermIdx;
528  if (Slice.Array == nullptr) {
529  NullTermIdx = 0;
530  } else {
531  NullTermIdx = ~((uint64_t)0);
532  for (uint64_t I = 0, E = Slice.Length; I < E; ++I) {
533  if (Slice.Array->getElementAsInteger(I + Slice.Offset) == 0) {
534  NullTermIdx = I;
535  break;
536  }
537  }
538  // If the string does not have '\0', leave it to strlen to compute
539  // its length.
540  if (NullTermIdx == ~((uint64_t)0))
541  return nullptr;
542  }
543 
544  Value *Offset = GEP->getOperand(2);
545  KnownBits Known = computeKnownBits(Offset, DL, 0, nullptr, CI, nullptr);
546  Known.Zero.flipAllBits();
547  uint64_t ArrSize =
548  cast<ArrayType>(GEP->getSourceElementType())->getNumElements();
549 
550  // KnownZero's bits are flipped, so zeros in KnownZero now represent
551  // bits known to be zeros in Offset, and ones in KnowZero represent
552  // bits unknown in Offset. Therefore, Offset is known to be in range
553  // [0, NullTermIdx] when the flipped KnownZero is non-negative and
554  // unsigned-less-than NullTermIdx.
555  //
556  // If Offset is not provably in the range [0, NullTermIdx], we can still
557  // optimize if we can prove that the program has undefined behavior when
558  // Offset is outside that range. That is the case when GEP->getOperand(0)
559  // is a pointer to an object whose memory extent is NullTermIdx+1.
560  if ((Known.Zero.isNonNegative() && Known.Zero.ule(NullTermIdx)) ||
561  (GEP->isInBounds() && isa<GlobalVariable>(GEP->getOperand(0)) &&
562  NullTermIdx == ArrSize - 1)) {
563  Offset = B.CreateSExtOrTrunc(Offset, CI->getType());
564  return B.CreateSub(ConstantInt::get(CI->getType(), NullTermIdx),
565  Offset);
566  }
567  }
568 
569  return nullptr;
570  }
571 
572  // strlen(x?"foo":"bars") --> x ? 3 : 4
573  if (SelectInst *SI = dyn_cast<SelectInst>(Src)) {
574  uint64_t LenTrue = GetStringLength(SI->getTrueValue(), CharSize);
575  uint64_t LenFalse = GetStringLength(SI->getFalseValue(), CharSize);
576  if (LenTrue && LenFalse) {
577  ORE.emit([&]() {
578  return OptimizationRemark("instcombine", "simplify-libcalls", CI)
579  << "folded strlen(select) to select of constants";
580  });
581  return B.CreateSelect(SI->getCondition(),
582  ConstantInt::get(CI->getType(), LenTrue - 1),
583  ConstantInt::get(CI->getType(), LenFalse - 1));
584  }
585  }
586 
587  // strlen(x) != 0 --> *x != 0
588  // strlen(x) == 0 --> *x == 0
590  return B.CreateZExt(B.CreateLoad(Src, "strlenfirst"), CI->getType());
591 
592  return nullptr;
593 }
594 
595 Value *LibCallSimplifier::optimizeStrLen(CallInst *CI, IRBuilder<> &B) {
596  return optimizeStringLength(CI, B, 8);
597 }
598 
599 Value *LibCallSimplifier::optimizeWcslen(CallInst *CI, IRBuilder<> &B) {
600  Module &M = *CI->getModule();
601  unsigned WCharSize = TLI->getWCharSize(M) * 8;
602  // We cannot perform this optimization without wchar_size metadata.
603  if (WCharSize == 0)
604  return nullptr;
605 
606  return optimizeStringLength(CI, B, WCharSize);
607 }
608 
609 Value *LibCallSimplifier::optimizeStrPBrk(CallInst *CI, IRBuilder<> &B) {
610  StringRef S1, S2;
611  bool HasS1 = getConstantStringInfo(CI->getArgOperand(0), S1);
612  bool HasS2 = getConstantStringInfo(CI->getArgOperand(1), S2);
613 
614  // strpbrk(s, "") -> nullptr
615  // strpbrk("", s) -> nullptr
616  if ((HasS1 && S1.empty()) || (HasS2 && S2.empty()))
617  return Constant::getNullValue(CI->getType());
618 
619  // Constant folding.
620  if (HasS1 && HasS2) {
621  size_t I = S1.find_first_of(S2);
622  if (I == StringRef::npos) // No match.
623  return Constant::getNullValue(CI->getType());
624 
625  return B.CreateGEP(B.getInt8Ty(), CI->getArgOperand(0), B.getInt64(I),
626  "strpbrk");
627  }
628 
629  // strpbrk(s, "a") -> strchr(s, 'a')
630  if (HasS2 && S2.size() == 1)
631  return emitStrChr(CI->getArgOperand(0), S2[0], B, TLI);
632 
633  return nullptr;
634 }
635 
636 Value *LibCallSimplifier::optimizeStrTo(CallInst *CI, IRBuilder<> &B) {
637  Value *EndPtr = CI->getArgOperand(1);
638  if (isa<ConstantPointerNull>(EndPtr)) {
639  // With a null EndPtr, this function won't capture the main argument.
640  // It would be readonly too, except that it still may write to errno.
641  CI->addParamAttr(0, Attribute::NoCapture);
642  }
643 
644  return nullptr;
645 }
646 
647 Value *LibCallSimplifier::optimizeStrSpn(CallInst *CI, IRBuilder<> &B) {
648  StringRef S1, S2;
649  bool HasS1 = getConstantStringInfo(CI->getArgOperand(0), S1);
650  bool HasS2 = getConstantStringInfo(CI->getArgOperand(1), S2);
651 
652  // strspn(s, "") -> 0
653  // strspn("", s) -> 0
654  if ((HasS1 && S1.empty()) || (HasS2 && S2.empty()))
655  return Constant::getNullValue(CI->getType());
656 
657  // Constant folding.
658  if (HasS1 && HasS2) {
659  size_t Pos = S1.find_first_not_of(S2);
660  if (Pos == StringRef::npos)
661  Pos = S1.size();
662  return ConstantInt::get(CI->getType(), Pos);
663  }
664 
665  return nullptr;
666 }
667 
668 Value *LibCallSimplifier::optimizeStrCSpn(CallInst *CI, IRBuilder<> &B) {
669  StringRef S1, S2;
670  bool HasS1 = getConstantStringInfo(CI->getArgOperand(0), S1);
671  bool HasS2 = getConstantStringInfo(CI->getArgOperand(1), S2);
672 
673  // strcspn("", s) -> 0
674  if (HasS1 && S1.empty())
675  return Constant::getNullValue(CI->getType());
676 
677  // Constant folding.
678  if (HasS1 && HasS2) {
679  size_t Pos = S1.find_first_of(S2);
680  if (Pos == StringRef::npos)
681  Pos = S1.size();
682  return ConstantInt::get(CI->getType(), Pos);
683  }
684 
685  // strcspn(s, "") -> strlen(s)
686  if (HasS2 && S2.empty())
687  return emitStrLen(CI->getArgOperand(0), B, DL, TLI);
688 
689  return nullptr;
690 }
691 
692 Value *LibCallSimplifier::optimizeStrStr(CallInst *CI, IRBuilder<> &B) {
693  // fold strstr(x, x) -> x.
694  if (CI->getArgOperand(0) == CI->getArgOperand(1))
695  return B.CreateBitCast(CI->getArgOperand(0), CI->getType());
696 
697  // fold strstr(a, b) == a -> strncmp(a, b, strlen(b)) == 0
699  Value *StrLen = emitStrLen(CI->getArgOperand(1), B, DL, TLI);
700  if (!StrLen)
701  return nullptr;
702  Value *StrNCmp = emitStrNCmp(CI->getArgOperand(0), CI->getArgOperand(1),
703  StrLen, B, DL, TLI);
704  if (!StrNCmp)
705  return nullptr;
706  for (auto UI = CI->user_begin(), UE = CI->user_end(); UI != UE;) {
707  ICmpInst *Old = cast<ICmpInst>(*UI++);
708  Value *Cmp =
709  B.CreateICmp(Old->getPredicate(), StrNCmp,
710  ConstantInt::getNullValue(StrNCmp->getType()), "cmp");
711  replaceAllUsesWith(Old, Cmp);
712  }
713  return CI;
714  }
715 
716  // See if either input string is a constant string.
717  StringRef SearchStr, ToFindStr;
718  bool HasStr1 = getConstantStringInfo(CI->getArgOperand(0), SearchStr);
719  bool HasStr2 = getConstantStringInfo(CI->getArgOperand(1), ToFindStr);
720 
721  // fold strstr(x, "") -> x.
722  if (HasStr2 && ToFindStr.empty())
723  return B.CreateBitCast(CI->getArgOperand(0), CI->getType());
724 
725  // If both strings are known, constant fold it.
726  if (HasStr1 && HasStr2) {
727  size_t Offset = SearchStr.find(ToFindStr);
728 
729  if (Offset == StringRef::npos) // strstr("foo", "bar") -> null
730  return Constant::getNullValue(CI->getType());
731 
732  // strstr("abcd", "bc") -> gep((char*)"abcd", 1)
733  Value *Result = castToCStr(CI->getArgOperand(0), B);
734  Result = B.CreateConstInBoundsGEP1_64(Result, Offset, "strstr");
735  return B.CreateBitCast(Result, CI->getType());
736  }
737 
738  // fold strstr(x, "y") -> strchr(x, 'y').
739  if (HasStr2 && ToFindStr.size() == 1) {
740  Value *StrChr = emitStrChr(CI->getArgOperand(0), ToFindStr[0], B, TLI);
741  return StrChr ? B.CreateBitCast(StrChr, CI->getType()) : nullptr;
742  }
743  return nullptr;
744 }
745 
746 Value *LibCallSimplifier::optimizeMemChr(CallInst *CI, IRBuilder<> &B) {
747  Value *SrcStr = CI->getArgOperand(0);
748  ConstantInt *CharC = dyn_cast<ConstantInt>(CI->getArgOperand(1));
750 
751  // memchr(x, y, 0) -> null
752  if (LenC && LenC->isZero())
753  return Constant::getNullValue(CI->getType());
754 
755  // From now on we need at least constant length and string.
756  StringRef Str;
757  if (!LenC || !getConstantStringInfo(SrcStr, Str, 0, /*TrimAtNul=*/false))
758  return nullptr;
759 
760  // Truncate the string to LenC. If Str is smaller than LenC we will still only
761  // scan the string, as reading past the end of it is undefined and we can just
762  // return null if we don't find the char.
763  Str = Str.substr(0, LenC->getZExtValue());
764 
765  // If the char is variable but the input str and length are not we can turn
766  // this memchr call into a simple bit field test. Of course this only works
767  // when the return value is only checked against null.
768  //
769  // It would be really nice to reuse switch lowering here but we can't change
770  // the CFG at this point.
771  //
772  // memchr("\r\n", C, 2) != nullptr -> (C & ((1 << '\r') | (1 << '\n'))) != 0
773  // after bounds check.
774  if (!CharC && !Str.empty() && isOnlyUsedInZeroEqualityComparison(CI)) {
775  unsigned char Max =
776  *std::max_element(reinterpret_cast<const unsigned char *>(Str.begin()),
777  reinterpret_cast<const unsigned char *>(Str.end()));
778 
779  // Make sure the bit field we're about to create fits in a register on the
780  // target.
781  // FIXME: On a 64 bit architecture this prevents us from using the
782  // interesting range of alpha ascii chars. We could do better by emitting
783  // two bitfields or shifting the range by 64 if no lower chars are used.
784  if (!DL.fitsInLegalInteger(Max + 1))
785  return nullptr;
786 
787  // For the bit field use a power-of-2 type with at least 8 bits to avoid
788  // creating unnecessary illegal types.
789  unsigned char Width = NextPowerOf2(std::max((unsigned char)7, Max));
790 
791  // Now build the bit field.
792  APInt Bitfield(Width, 0);
793  for (char C : Str)
794  Bitfield.setBit((unsigned char)C);
795  Value *BitfieldC = B.getInt(Bitfield);
796 
797  // First check that the bit field access is within bounds.
798  Value *C = B.CreateZExtOrTrunc(CI->getArgOperand(1), BitfieldC->getType());
799  Value *Bounds = B.CreateICmp(ICmpInst::ICMP_ULT, C, B.getIntN(Width, Width),
800  "memchr.bounds");
801 
802  // Create code that checks if the given bit is set in the field.
803  Value *Shl = B.CreateShl(B.getIntN(Width, 1ULL), C);
804  Value *Bits = B.CreateIsNotNull(B.CreateAnd(Shl, BitfieldC), "memchr.bits");
805 
806  // Finally merge both checks and cast to pointer type. The inttoptr
807  // implicitly zexts the i1 to intptr type.
808  return B.CreateIntToPtr(B.CreateAnd(Bounds, Bits, "memchr"), CI->getType());
809  }
810 
811  // Check if all arguments are constants. If so, we can constant fold.
812  if (!CharC)
813  return nullptr;
814 
815  // Compute the offset.
816  size_t I = Str.find(CharC->getSExtValue() & 0xFF);
817  if (I == StringRef::npos) // Didn't find the char. memchr returns null.
818  return Constant::getNullValue(CI->getType());
819 
820  // memchr(s+n,c,l) -> gep(s+n+i,c)
821  return B.CreateGEP(B.getInt8Ty(), SrcStr, B.getInt64(I), "memchr");
822 }
823 
824 Value *LibCallSimplifier::optimizeMemCmp(CallInst *CI, IRBuilder<> &B) {
825  Value *LHS = CI->getArgOperand(0), *RHS = CI->getArgOperand(1);
826 
827  if (LHS == RHS) // memcmp(s,s,x) -> 0
828  return Constant::getNullValue(CI->getType());
829 
830  // Make sure we have a constant length.
832  if (!LenC)
833  return nullptr;
834 
835  uint64_t Len = LenC->getZExtValue();
836  if (Len == 0) // memcmp(s1,s2,0) -> 0
837  return Constant::getNullValue(CI->getType());
838 
839  // memcmp(S1,S2,1) -> *(unsigned char*)LHS - *(unsigned char*)RHS
840  if (Len == 1) {
841  Value *LHSV = B.CreateZExt(B.CreateLoad(castToCStr(LHS, B), "lhsc"),
842  CI->getType(), "lhsv");
843  Value *RHSV = B.CreateZExt(B.CreateLoad(castToCStr(RHS, B), "rhsc"),
844  CI->getType(), "rhsv");
845  return B.CreateSub(LHSV, RHSV, "chardiff");
846  }
847 
848  // memcmp(S1,S2,N/8)==0 -> (*(intN_t*)S1 != *(intN_t*)S2)==0
849  // TODO: The case where both inputs are constants does not need to be limited
850  // to legal integers or equality comparison. See block below this.
851  if (DL.isLegalInteger(Len * 8) && isOnlyUsedInZeroEqualityComparison(CI)) {
852  IntegerType *IntType = IntegerType::get(CI->getContext(), Len * 8);
853  unsigned PrefAlignment = DL.getPrefTypeAlignment(IntType);
854 
855  // First, see if we can fold either argument to a constant.
856  Value *LHSV = nullptr;
857  if (auto *LHSC = dyn_cast<Constant>(LHS)) {
858  LHSC = ConstantExpr::getBitCast(LHSC, IntType->getPointerTo());
859  LHSV = ConstantFoldLoadFromConstPtr(LHSC, IntType, DL);
860  }
861  Value *RHSV = nullptr;
862  if (auto *RHSC = dyn_cast<Constant>(RHS)) {
863  RHSC = ConstantExpr::getBitCast(RHSC, IntType->getPointerTo());
864  RHSV = ConstantFoldLoadFromConstPtr(RHSC, IntType, DL);
865  }
866 
867  // Don't generate unaligned loads. If either source is constant data,
868  // alignment doesn't matter for that source because there is no load.
869  if ((LHSV || getKnownAlignment(LHS, DL, CI) >= PrefAlignment) &&
870  (RHSV || getKnownAlignment(RHS, DL, CI) >= PrefAlignment)) {
871  if (!LHSV) {
872  Type *LHSPtrTy =
873  IntType->getPointerTo(LHS->getType()->getPointerAddressSpace());
874  LHSV = B.CreateLoad(B.CreateBitCast(LHS, LHSPtrTy), "lhsv");
875  }
876  if (!RHSV) {
877  Type *RHSPtrTy =
878  IntType->getPointerTo(RHS->getType()->getPointerAddressSpace());
879  RHSV = B.CreateLoad(B.CreateBitCast(RHS, RHSPtrTy), "rhsv");
880  }
881  return B.CreateZExt(B.CreateICmpNE(LHSV, RHSV), CI->getType(), "memcmp");
882  }
883  }
884 
885  // Constant folding: memcmp(x, y, Len) -> constant (all arguments are const).
886  // TODO: This is limited to i8 arrays.
887  StringRef LHSStr, RHSStr;
888  if (getConstantStringInfo(LHS, LHSStr) &&
889  getConstantStringInfo(RHS, RHSStr)) {
890  // Make sure we're not reading out-of-bounds memory.
891  if (Len > LHSStr.size() || Len > RHSStr.size())
892  return nullptr;
893  // Fold the memcmp and normalize the result. This way we get consistent
894  // results across multiple platforms.
895  uint64_t Ret = 0;
896  int Cmp = memcmp(LHSStr.data(), RHSStr.data(), Len);
897  if (Cmp < 0)
898  Ret = -1;
899  else if (Cmp > 0)
900  Ret = 1;
901  return ConstantInt::get(CI->getType(), Ret);
902  }
903 
904  return nullptr;
905 }
906 
907 Value *LibCallSimplifier::optimizeMemCpy(CallInst *CI, IRBuilder<> &B) {
908  // memcpy(x, y, n) -> llvm.memcpy(align 1 x, align 1 y, n)
909  B.CreateMemCpy(CI->getArgOperand(0), 1, CI->getArgOperand(1), 1,
910  CI->getArgOperand(2));
911  return CI->getArgOperand(0);
912 }
913 
914 Value *LibCallSimplifier::optimizeMemMove(CallInst *CI, IRBuilder<> &B) {
915  // memmove(x, y, n) -> llvm.memmove(align 1 x, align 1 y, n)
916  B.CreateMemMove(CI->getArgOperand(0), 1, CI->getArgOperand(1), 1,
917  CI->getArgOperand(2));
918  return CI->getArgOperand(0);
919 }
920 
921 /// Fold memset[_chk](malloc(n), 0, n) --> calloc(1, n).
923  const TargetLibraryInfo &TLI) {
924  // This has to be a memset of zeros (bzero).
925  auto *FillValue = dyn_cast<ConstantInt>(Memset->getArgOperand(1));
926  if (!FillValue || FillValue->getZExtValue() != 0)
927  return nullptr;
928 
929  // TODO: We should handle the case where the malloc has more than one use.
930  // This is necessary to optimize common patterns such as when the result of
931  // the malloc is checked against null or when a memset intrinsic is used in
932  // place of a memset library call.
933  auto *Malloc = dyn_cast<CallInst>(Memset->getArgOperand(0));
934  if (!Malloc || !Malloc->hasOneUse())
935  return nullptr;
936 
937  // Is the inner call really malloc()?
938  Function *InnerCallee = Malloc->getCalledFunction();
939  if (!InnerCallee)
940  return nullptr;
941 
942  LibFunc Func;
943  if (!TLI.getLibFunc(*InnerCallee, Func) || !TLI.has(Func) ||
944  Func != LibFunc_malloc)
945  return nullptr;
946 
947  // The memset must cover the same number of bytes that are malloc'd.
948  if (Memset->getArgOperand(2) != Malloc->getArgOperand(0))
949  return nullptr;
950 
951  // Replace the malloc with a calloc. We need the data layout to know what the
952  // actual size of a 'size_t' parameter is.
953  B.SetInsertPoint(Malloc->getParent(), ++Malloc->getIterator());
954  const DataLayout &DL = Malloc->getModule()->getDataLayout();
955  IntegerType *SizeType = DL.getIntPtrType(B.GetInsertBlock()->getContext());
956  Value *Calloc = emitCalloc(ConstantInt::get(SizeType, 1),
957  Malloc->getArgOperand(0), Malloc->getAttributes(),
958  B, TLI);
959  if (!Calloc)
960  return nullptr;
961 
962  Malloc->replaceAllUsesWith(Calloc);
963  Malloc->eraseFromParent();
964 
965  return Calloc;
966 }
967 
968 Value *LibCallSimplifier::optimizeMemSet(CallInst *CI, IRBuilder<> &B) {
969  if (auto *Calloc = foldMallocMemset(CI, B, *TLI))
970  return Calloc;
971 
972  // memset(p, v, n) -> llvm.memset(align 1 p, v, n)
973  Value *Val = B.CreateIntCast(CI->getArgOperand(1), B.getInt8Ty(), false);
974  B.CreateMemSet(CI->getArgOperand(0), Val, CI->getArgOperand(2), 1);
975  return CI->getArgOperand(0);
976 }
977 
978 Value *LibCallSimplifier::optimizeRealloc(CallInst *CI, IRBuilder<> &B) {
979  if (isa<ConstantPointerNull>(CI->getArgOperand(0)))
980  return emitMalloc(CI->getArgOperand(1), B, DL, TLI);
981 
982  return nullptr;
983 }
984 
985 //===----------------------------------------------------------------------===//
986 // Math Library Optimizations
987 //===----------------------------------------------------------------------===//
988 
989 // Replace a libcall \p CI with a call to intrinsic \p IID
991  // Propagate fast-math flags from the existing call to the new call.
994 
995  Module *M = CI->getModule();
996  Value *V = CI->getArgOperand(0);
997  Function *F = Intrinsic::getDeclaration(M, IID, CI->getType());
998  CallInst *NewCall = B.CreateCall(F, V);
999  NewCall->takeName(CI);
1000  return NewCall;
1001 }
1002 
1003 /// Return a variant of Val with float type.
1004 /// Currently this works in two cases: If Val is an FPExtension of a float
1005 /// value to something bigger, simply return the operand.
1006 /// If Val is a ConstantFP but can be converted to a float ConstantFP without
1007 /// loss of precision do so.
1009  if (FPExtInst *Cast = dyn_cast<FPExtInst>(Val)) {
1010  Value *Op = Cast->getOperand(0);
1011  if (Op->getType()->isFloatTy())
1012  return Op;
1013  }
1014  if (ConstantFP *Const = dyn_cast<ConstantFP>(Val)) {
1015  APFloat F = Const->getValueAPF();
1016  bool losesInfo;
1018  &losesInfo);
1019  if (!losesInfo)
1020  return ConstantFP::get(Const->getContext(), F);
1021  }
1022  return nullptr;
1023 }
1024 
1025 /// Shrink double -> float functions.
1027  bool isBinary, bool isPrecise = false) {
1028  if (!CI->getType()->isDoubleTy())
1029  return nullptr;
1030 
1031  // If not all the uses of the function are converted to float, then bail out.
1032  // This matters if the precision of the result is more important than the
1033  // precision of the arguments.
1034  if (isPrecise)
1035  for (User *U : CI->users()) {
1036  FPTruncInst *Cast = dyn_cast<FPTruncInst>(U);
1037  if (!Cast || !Cast->getType()->isFloatTy())
1038  return nullptr;
1039  }
1040 
1041  // If this is something like 'g((double) float)', convert to 'gf(float)'.
1042  Value *V[2];
1043  V[0] = valueHasFloatPrecision(CI->getArgOperand(0));
1044  V[1] = isBinary ? valueHasFloatPrecision(CI->getArgOperand(1)) : nullptr;
1045  if (!V[0] || (isBinary && !V[1]))
1046  return nullptr;
1047 
1048  // If call isn't an intrinsic, check that it isn't within a function with the
1049  // same name as the float version of this call, otherwise the result is an
1050  // infinite loop. For example, from MinGW-w64:
1051  //
1052  // float expf(float val) { return (float) exp((double) val); }
1053  Function *CalleeFn = CI->getCalledFunction();
1054  StringRef CalleeNm = CalleeFn->getName();
1055  AttributeList CalleeAt = CalleeFn->getAttributes();
1056  if (CalleeFn && !CalleeFn->isIntrinsic()) {
1057  const Function *Fn = CI->getFunction();
1058  StringRef FnName = Fn->getName();
1059  if (FnName.back() == 'f' &&
1060  FnName.size() == (CalleeNm.size() + 1) &&
1061  FnName.startswith(CalleeNm))
1062  return nullptr;
1063  }
1064 
1065  // Propagate the math semantics from the current function to the new function.
1068 
1069  // g((double) float) -> (double) gf(float)
1070  Value *R;
1071  if (CalleeFn->isIntrinsic()) {
1072  Module *M = CI->getModule();
1073  Intrinsic::ID IID = CalleeFn->getIntrinsicID();
1074  Function *Fn = Intrinsic::getDeclaration(M, IID, B.getFloatTy());
1075  R = isBinary ? B.CreateCall(Fn, V) : B.CreateCall(Fn, V[0]);
1076  }
1077  else
1078  R = isBinary ? emitBinaryFloatFnCall(V[0], V[1], CalleeNm, B, CalleeAt)
1079  : emitUnaryFloatFnCall(V[0], CalleeNm, B, CalleeAt);
1080 
1081  return B.CreateFPExt(R, B.getDoubleTy());
1082 }
1083 
1084 /// Shrink double -> float for unary functions.
1086  bool isPrecise = false) {
1087  return optimizeDoubleFP(CI, B, false, isPrecise);
1088 }
1089 
1090 /// Shrink double -> float for binary functions.
1092  bool isPrecise = false) {
1093  return optimizeDoubleFP(CI, B, true, isPrecise);
1094 }
1095 
1096 // cabs(z) -> sqrt((creal(z)*creal(z)) + (cimag(z)*cimag(z)))
1097 Value *LibCallSimplifier::optimizeCAbs(CallInst *CI, IRBuilder<> &B) {
1098  if (!CI->isFast())
1099  return nullptr;
1100 
1101  // Propagate fast-math flags from the existing call to new instructions.
1104 
1105  Value *Real, *Imag;
1106  if (CI->getNumArgOperands() == 1) {
1107  Value *Op = CI->getArgOperand(0);
1108  assert(Op->getType()->isArrayTy() && "Unexpected signature for cabs!");
1109  Real = B.CreateExtractValue(Op, 0, "real");
1110  Imag = B.CreateExtractValue(Op, 1, "imag");
1111  } else {
1112  assert(CI->getNumArgOperands() == 2 && "Unexpected signature for cabs!");
1113  Real = CI->getArgOperand(0);
1114  Imag = CI->getArgOperand(1);
1115  }
1116 
1117  Value *RealReal = B.CreateFMul(Real, Real);
1118  Value *ImagImag = B.CreateFMul(Imag, Imag);
1119 
1120  Function *FSqrt = Intrinsic::getDeclaration(CI->getModule(), Intrinsic::sqrt,
1121  CI->getType());
1122  return B.CreateCall(FSqrt, B.CreateFAdd(RealReal, ImagImag), "cabs");
1123 }
1124 
1126  IRBuilder<> &B) {
1127  if (!isa<FPMathOperator>(Call))
1128  return nullptr;
1129 
1132 
1133  // TODO: Add tan() and other calls.
1134  // TODO: Can this be shared to also handle LLVM intrinsics?
1135  Value *X;
1136  switch (Func) {
1137  case LibFunc_sin:
1138  case LibFunc_sinf:
1139  case LibFunc_sinl:
1140  // sin(-X) --> -sin(X)
1141  if (match(Call->getArgOperand(0), m_OneUse(m_FNeg(m_Value(X)))))
1142  return B.CreateFNeg(B.CreateCall(Call->getCalledFunction(), X, "sin"));
1143  break;
1144  case LibFunc_cos:
1145  case LibFunc_cosf:
1146  case LibFunc_cosl:
1147  // cos(-X) --> cos(X)
1148  if (match(Call->getArgOperand(0), m_FNeg(m_Value(X))))
1149  return B.CreateCall(Call->getCalledFunction(), X, "cos");
1150  break;
1151  default:
1152  break;
1153  }
1154  return nullptr;
1155 }
1156 
1157 static Value *getPow(Value *InnerChain[33], unsigned Exp, IRBuilder<> &B) {
1158  // Multiplications calculated using Addition Chains.
1159  // Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html
1160 
1161  assert(Exp != 0 && "Incorrect exponent 0 not handled");
1162 
1163  if (InnerChain[Exp])
1164  return InnerChain[Exp];
1165 
1166  static const unsigned AddChain[33][2] = {
1167  {0, 0}, // Unused.
1168  {0, 0}, // Unused (base case = pow1).
1169  {1, 1}, // Unused (pre-computed).
1170  {1, 2}, {2, 2}, {2, 3}, {3, 3}, {2, 5}, {4, 4},
1171  {1, 8}, {5, 5}, {1, 10}, {6, 6}, {4, 9}, {7, 7},
1172  {3, 12}, {8, 8}, {8, 9}, {2, 16}, {1, 18}, {10, 10},
1173  {6, 15}, {11, 11}, {3, 20}, {12, 12}, {8, 17}, {13, 13},
1174  {3, 24}, {14, 14}, {4, 25}, {15, 15}, {3, 28}, {16, 16},
1175  };
1176 
1177  InnerChain[Exp] = B.CreateFMul(getPow(InnerChain, AddChain[Exp][0], B),
1178  getPow(InnerChain, AddChain[Exp][1], B));
1179  return InnerChain[Exp];
1180 }
1181 
1182 /// Use square root in place of pow(x, +/-0.5).
1183 Value *LibCallSimplifier::replacePowWithSqrt(CallInst *Pow, IRBuilder<> &B) {
1184  // TODO: There is some subset of 'fast' under which these transforms should
1185  // be allowed.
1186  if (!Pow->isFast())
1187  return nullptr;
1188 
1189  Value *Sqrt, *Base = Pow->getArgOperand(0), *Expo = Pow->getArgOperand(1);
1190  Type *Ty = Pow->getType();
1191 
1192  const APFloat *ExpoF;
1193  if (!match(Expo, m_APFloat(ExpoF)) ||
1194  (!ExpoF->isExactlyValue(0.5) && !ExpoF->isExactlyValue(-0.5)))
1195  return nullptr;
1196 
1197  // If errno is never set, then use the intrinsic for sqrt().
1198  if (Pow->hasFnAttr(Attribute::ReadNone)) {
1199  Function *SqrtFn = Intrinsic::getDeclaration(Pow->getModule(),
1200  Intrinsic::sqrt, Ty);
1201  Sqrt = B.CreateCall(SqrtFn, Base);
1202  }
1203  // Otherwise, use the libcall for sqrt().
1204  else if (hasUnaryFloatFn(TLI, Ty, LibFunc_sqrt, LibFunc_sqrtf, LibFunc_sqrtl))
1205  // TODO: We also should check that the target can in fact lower the sqrt()
1206  // libcall. We currently have no way to ask this question, so we ask if
1207  // the target has a sqrt() libcall, which is not exactly the same.
1208  Sqrt = emitUnaryFloatFnCall(Base, TLI->getName(LibFunc_sqrt), B,
1209  Pow->getCalledFunction()->getAttributes());
1210  else
1211  return nullptr;
1212 
1213  // If the exponent is negative, then get the reciprocal.
1214  if (ExpoF->isNegative())
1215  Sqrt = B.CreateFDiv(ConstantFP::get(Ty, 1.0), Sqrt, "reciprocal");
1216 
1217  return Sqrt;
1218 }
1219 
1220 Value *LibCallSimplifier::optimizePow(CallInst *Pow, IRBuilder<> &B) {
1221  Value *Base = Pow->getArgOperand(0), *Expo = Pow->getArgOperand(1);
1222  Function *Callee = Pow->getCalledFunction();
1223  AttributeList Attrs = Callee->getAttributes();
1224  StringRef Name = Callee->getName();
1225  Module *Module = Pow->getModule();
1226  Type *Ty = Pow->getType();
1227  Value *Shrunk = nullptr;
1228  bool Ignored;
1229 
1230  // Bail out if simplifying libcalls to pow() is disabled.
1231  if (!hasUnaryFloatFn(TLI, Ty, LibFunc_pow, LibFunc_powf, LibFunc_powl))
1232  return nullptr;
1233 
1234  // Propagate the math semantics from the call to any created instructions.
1237 
1238  // Shrink pow() to powf() if the arguments are single precision,
1239  // unless the result is expected to be double precision.
1240  if (UnsafeFPShrink &&
1241  Name == TLI->getName(LibFunc_pow) && hasFloatVersion(Name))
1242  Shrunk = optimizeBinaryDoubleFP(Pow, B, true);
1243 
1244  // Evaluate special cases related to the base.
1245 
1246  // pow(1.0, x) -> 1.0
1247  if (match(Base, m_FPOne()))
1248  return Base;
1249 
1250  // pow(2.0, x) -> exp2(x)
1251  if (match(Base, m_SpecificFP(2.0))) {
1252  Value *Exp2 = Intrinsic::getDeclaration(Module, Intrinsic::exp2, Ty);
1253  return B.CreateCall(Exp2, Expo, "exp2");
1254  }
1255 
1256  // pow(10.0, x) -> exp10(x)
1257  if (ConstantFP *BaseC = dyn_cast<ConstantFP>(Base))
1258  // There's no exp10 intrinsic yet, but, maybe, some day there shall be one.
1259  if (BaseC->isExactlyValue(10.0) &&
1260  hasUnaryFloatFn(TLI, Ty, LibFunc_exp10, LibFunc_exp10f, LibFunc_exp10l))
1261  return emitUnaryFloatFnCall(Expo, TLI->getName(LibFunc_exp10), B, Attrs);
1262 
1263  // pow(exp(x), y) -> exp(x * y)
1264  // pow(exp2(x), y) -> exp2(x * y)
1265  // We enable these only with fast-math. Besides rounding differences, the
1266  // transformation changes overflow and underflow behavior quite dramatically.
1267  // Example: x = 1000, y = 0.001.
1268  // pow(exp(x), y) = pow(inf, 0.001) = inf, whereas exp(x*y) = exp(1).
1269  auto *BaseFn = dyn_cast<CallInst>(Base);
1270  if (BaseFn && BaseFn->isFast() && Pow->isFast()) {
1271  LibFunc LibFn;
1272  Function *CalleeFn = BaseFn->getCalledFunction();
1273  if (CalleeFn && TLI->getLibFunc(CalleeFn->getName(), LibFn) &&
1274  (LibFn == LibFunc_exp || LibFn == LibFunc_exp2) && TLI->has(LibFn)) {
1275  Value *FMul = B.CreateFMul(BaseFn->getArgOperand(0), Expo, "mul");
1276  return emitUnaryFloatFnCall(FMul, CalleeFn->getName(), B,
1277  CalleeFn->getAttributes());
1278  }
1279  }
1280 
1281  // Evaluate special cases related to the exponent.
1282 
1283  // pow(x, -1.0) -> 1.0 / x
1284  if (match(Expo, m_SpecificFP(-1.0)))
1285  return B.CreateFDiv(ConstantFP::get(Ty, 1.0), Base, "reciprocal");
1286 
1287  // pow(x, 0.0) -> 1.0
1288  if (match(Expo, m_SpecificFP(0.0)))
1289  return ConstantFP::get(Ty, 1.0);
1290 
1291  // pow(x, 1.0) -> x
1292  if (match(Expo, m_FPOne()))
1293  return Base;
1294 
1295  // pow(x, 2.0) -> x * x
1296  if (match(Expo, m_SpecificFP(2.0)))
1297  return B.CreateFMul(Base, Base, "square");
1298 
1299  if (Value *Sqrt = replacePowWithSqrt(Pow, B))
1300  return Sqrt;
1301 
1302  // FIXME: Correct the transforms and pull this into replacePowWithSqrt().
1303  ConstantFP *ExpoC = dyn_cast<ConstantFP>(Expo);
1304  if (ExpoC && ExpoC->isExactlyValue(0.5) &&
1305  hasUnaryFloatFn(TLI, Ty, LibFunc_sqrt, LibFunc_sqrtf, LibFunc_sqrtl)) {
1306  // Expand pow(x, 0.5) to (x == -infinity ? +infinity : fabs(sqrt(x))).
1307  // This is faster than calling pow(), and still handles -0.0 and
1308  // negative infinity correctly.
1309  // TODO: In finite-only mode, this could be just fabs(sqrt(x)).
1310  Value *PosInf = ConstantFP::getInfinity(Ty);
1311  Value *NegInf = ConstantFP::getInfinity(Ty, true);
1312 
1313  // TODO: As above, we should lower to the sqrt() intrinsic if the pow() is
1314  // an intrinsic, to match errno semantics.
1315  Value *Sqrt = emitUnaryFloatFnCall(Base, TLI->getName(LibFunc_sqrt),
1316  B, Attrs);
1317  Function *FAbsFn = Intrinsic::getDeclaration(Module, Intrinsic::fabs, Ty);
1318  Value *FAbs = B.CreateCall(FAbsFn, Sqrt, "abs");
1319  Value *FCmp = B.CreateFCmpOEQ(Base, NegInf, "isinf");
1320  Sqrt = B.CreateSelect(FCmp, PosInf, FAbs);
1321  return Sqrt;
1322  }
1323 
1324  // pow(x, n) -> x * x * x * ...
1325  const APFloat *ExpoF;
1326  if (Pow->isFast() && match(Expo, m_APFloat(ExpoF))) {
1327  // We limit to a max of 7 multiplications, thus the maximum exponent is 32.
1328  APFloat LimF(ExpoF->getSemantics(), 33.0),
1329  ExpoA(abs(*ExpoF));
1330  if (ExpoA.isInteger() && ExpoA.compare(LimF) == APFloat::cmpLessThan) {
1331  // We will memoize intermediate products of the Addition Chain.
1332  Value *InnerChain[33] = {nullptr};
1333  InnerChain[1] = Base;
1334  InnerChain[2] = B.CreateFMul(Base, Base, "square");
1335 
1336  // We cannot readily convert a non-double type (like float) to a double.
1337  // So we first convert it to something which could be converted to double.
1339  Value *FMul = getPow(InnerChain, ExpoA.convertToDouble(), B);
1340 
1341  // If the exponent is negative, then get the reciprocal.
1342  if (ExpoF->isNegative())
1343  FMul = B.CreateFDiv(ConstantFP::get(Ty, 1.0), FMul, "reciprocal");
1344 
1345  return FMul;
1346  }
1347  }
1348 
1349  return Shrunk;
1350 }
1351 
1352 Value *LibCallSimplifier::optimizeExp2(CallInst *CI, IRBuilder<> &B) {
1354  Value *Ret = nullptr;
1355  StringRef Name = Callee->getName();
1356  if (UnsafeFPShrink && Name == "exp2" && hasFloatVersion(Name))
1357  Ret = optimizeUnaryDoubleFP(CI, B, true);
1358 
1359  Value *Op = CI->getArgOperand(0);
1360  // Turn exp2(sitofp(x)) -> ldexp(1.0, sext(x)) if sizeof(x) <= 32
1361  // Turn exp2(uitofp(x)) -> ldexp(1.0, zext(x)) if sizeof(x) < 32
1362  LibFunc LdExp = LibFunc_ldexpl;
1363  if (Op->getType()->isFloatTy())
1364  LdExp = LibFunc_ldexpf;
1365  else if (Op->getType()->isDoubleTy())
1366  LdExp = LibFunc_ldexp;
1367 
1368  if (TLI->has(LdExp)) {
1369  Value *LdExpArg = nullptr;
1370  if (SIToFPInst *OpC = dyn_cast<SIToFPInst>(Op)) {
1371  if (OpC->getOperand(0)->getType()->getPrimitiveSizeInBits() <= 32)
1372  LdExpArg = B.CreateSExt(OpC->getOperand(0), B.getInt32Ty());
1373  } else if (UIToFPInst *OpC = dyn_cast<UIToFPInst>(Op)) {
1374  if (OpC->getOperand(0)->getType()->getPrimitiveSizeInBits() < 32)
1375  LdExpArg = B.CreateZExt(OpC->getOperand(0), B.getInt32Ty());
1376  }
1377 
1378  if (LdExpArg) {
1379  Constant *One = ConstantFP::get(CI->getContext(), APFloat(1.0f));
1380  if (!Op->getType()->isFloatTy())
1381  One = ConstantExpr::getFPExtend(One, Op->getType());
1382 
1383  Module *M = CI->getModule();
1384  Value *NewCallee =
1385  M->getOrInsertFunction(TLI->getName(LdExp), Op->getType(),
1386  Op->getType(), B.getInt32Ty());
1387  CallInst *CI = B.CreateCall(NewCallee, {One, LdExpArg});
1388  if (const Function *F = dyn_cast<Function>(Callee->stripPointerCasts()))
1389  CI->setCallingConv(F->getCallingConv());
1390 
1391  return CI;
1392  }
1393  }
1394  return Ret;
1395 }
1396 
1397 Value *LibCallSimplifier::optimizeFMinFMax(CallInst *CI, IRBuilder<> &B) {
1399  // If we can shrink the call to a float function rather than a double
1400  // function, do that first.
1401  StringRef Name = Callee->getName();
1402  if ((Name == "fmin" || Name == "fmax") && hasFloatVersion(Name))
1403  if (Value *Ret = optimizeBinaryDoubleFP(CI, B))
1404  return Ret;
1405 
1407  FastMathFlags FMF;
1408  if (CI->isFast()) {
1409  // If the call is 'fast', then anything we create here will also be 'fast'.
1410  FMF.setFast();
1411  } else {
1412  // At a minimum, no-nans-fp-math must be true.
1413  if (!CI->hasNoNaNs())
1414  return nullptr;
1415  // No-signed-zeros is implied by the definitions of fmax/fmin themselves:
1416  // "Ideally, fmax would be sensitive to the sign of zero, for example
1417  // fmax(-0. 0, +0. 0) would return +0; however, implementation in software
1418  // might be impractical."
1419  FMF.setNoSignedZeros();
1420  FMF.setNoNaNs();
1421  }
1422  B.setFastMathFlags(FMF);
1423 
1424  // We have a relaxed floating-point environment. We can ignore NaN-handling
1425  // and transform to a compare and select. We do not have to consider errno or
1426  // exceptions, because fmin/fmax do not have those.
1427  Value *Op0 = CI->getArgOperand(0);
1428  Value *Op1 = CI->getArgOperand(1);
1429  Value *Cmp = Callee->getName().startswith("fmin") ?
1430  B.CreateFCmpOLT(Op0, Op1) : B.CreateFCmpOGT(Op0, Op1);
1431  return B.CreateSelect(Cmp, Op0, Op1);
1432 }
1433 
1434 Value *LibCallSimplifier::optimizeLog(CallInst *CI, IRBuilder<> &B) {
1436  Value *Ret = nullptr;
1437  StringRef Name = Callee->getName();
1438  if (UnsafeFPShrink && hasFloatVersion(Name))
1439  Ret = optimizeUnaryDoubleFP(CI, B, true);
1440 
1441  if (!CI->isFast())
1442  return Ret;
1443  Value *Op1 = CI->getArgOperand(0);
1444  auto *OpC = dyn_cast<CallInst>(Op1);
1445 
1446  // The earlier call must also be 'fast' in order to do these transforms.
1447  if (!OpC || !OpC->isFast())
1448  return Ret;
1449 
1450  // log(pow(x,y)) -> y*log(x)
1451  // This is only applicable to log, log2, log10.
1452  if (Name != "log" && Name != "log2" && Name != "log10")
1453  return Ret;
1454 
1456  FastMathFlags FMF;
1457  FMF.setFast();
1458  B.setFastMathFlags(FMF);
1459 
1460  LibFunc Func;
1461  Function *F = OpC->getCalledFunction();
1462  if (F && ((TLI->getLibFunc(F->getName(), Func) && TLI->has(Func) &&
1463  Func == LibFunc_pow) || F->getIntrinsicID() == Intrinsic::pow))
1464  return B.CreateFMul(OpC->getArgOperand(1),
1465  emitUnaryFloatFnCall(OpC->getOperand(0), Callee->getName(), B,
1466  Callee->getAttributes()), "mul");
1467 
1468  // log(exp2(y)) -> y*log(2)
1469  if (F && Name == "log" && TLI->getLibFunc(F->getName(), Func) &&
1470  TLI->has(Func) && Func == LibFunc_exp2)
1471  return B.CreateFMul(
1472  OpC->getArgOperand(0),
1474  Callee->getName(), B, Callee->getAttributes()),
1475  "logmul");
1476  return Ret;
1477 }
1478 
1479 Value *LibCallSimplifier::optimizeSqrt(CallInst *CI, IRBuilder<> &B) {
1481  Value *Ret = nullptr;
1482  // TODO: Once we have a way (other than checking for the existince of the
1483  // libcall) to tell whether our target can lower @llvm.sqrt, relax the
1484  // condition below.
1485  if (TLI->has(LibFunc_sqrtf) && (Callee->getName() == "sqrt" ||
1486  Callee->getIntrinsicID() == Intrinsic::sqrt))
1487  Ret = optimizeUnaryDoubleFP(CI, B, true);
1488 
1489  if (!CI->isFast())
1490  return Ret;
1491 
1493  if (!I || I->getOpcode() != Instruction::FMul || !I->isFast())
1494  return Ret;
1495 
1496  // We're looking for a repeated factor in a multiplication tree,
1497  // so we can do this fold: sqrt(x * x) -> fabs(x);
1498  // or this fold: sqrt((x * x) * y) -> fabs(x) * sqrt(y).
1499  Value *Op0 = I->getOperand(0);
1500  Value *Op1 = I->getOperand(1);
1501  Value *RepeatOp = nullptr;
1502  Value *OtherOp = nullptr;
1503  if (Op0 == Op1) {
1504  // Simple match: the operands of the multiply are identical.
1505  RepeatOp = Op0;
1506  } else {
1507  // Look for a more complicated pattern: one of the operands is itself
1508  // a multiply, so search for a common factor in that multiply.
1509  // Note: We don't bother looking any deeper than this first level or for
1510  // variations of this pattern because instcombine's visitFMUL and/or the
1511  // reassociation pass should give us this form.
1512  Value *OtherMul0, *OtherMul1;
1513  if (match(Op0, m_FMul(m_Value(OtherMul0), m_Value(OtherMul1)))) {
1514  // Pattern: sqrt((x * y) * z)
1515  if (OtherMul0 == OtherMul1 && cast<Instruction>(Op0)->isFast()) {
1516  // Matched: sqrt((x * x) * z)
1517  RepeatOp = OtherMul0;
1518  OtherOp = Op1;
1519  }
1520  }
1521  }
1522  if (!RepeatOp)
1523  return Ret;
1524 
1525  // Fast math flags for any created instructions should match the sqrt
1526  // and multiply.
1529 
1530  // If we found a repeated factor, hoist it out of the square root and
1531  // replace it with the fabs of that factor.
1532  Module *M = Callee->getParent();
1533  Type *ArgType = I->getType();
1534  Value *Fabs = Intrinsic::getDeclaration(M, Intrinsic::fabs, ArgType);
1535  Value *FabsCall = B.CreateCall(Fabs, RepeatOp, "fabs");
1536  if (OtherOp) {
1537  // If we found a non-repeated factor, we still need to get its square
1538  // root. We then multiply that by the value that was simplified out
1539  // of the square root calculation.
1540  Value *Sqrt = Intrinsic::getDeclaration(M, Intrinsic::sqrt, ArgType);
1541  Value *SqrtCall = B.CreateCall(Sqrt, OtherOp, "sqrt");
1542  return B.CreateFMul(FabsCall, SqrtCall);
1543  }
1544  return FabsCall;
1545 }
1546 
1547 // TODO: Generalize to handle any trig function and its inverse.
1548 Value *LibCallSimplifier::optimizeTan(CallInst *CI, IRBuilder<> &B) {
1550  Value *Ret = nullptr;
1551  StringRef Name = Callee->getName();
1552  if (UnsafeFPShrink && Name == "tan" && hasFloatVersion(Name))
1553  Ret = optimizeUnaryDoubleFP(CI, B, true);
1554 
1555  Value *Op1 = CI->getArgOperand(0);
1556  auto *OpC = dyn_cast<CallInst>(Op1);
1557  if (!OpC)
1558  return Ret;
1559 
1560  // Both calls must be 'fast' in order to remove them.
1561  if (!CI->isFast() || !OpC->isFast())
1562  return Ret;
1563 
1564  // tan(atan(x)) -> x
1565  // tanf(atanf(x)) -> x
1566  // tanl(atanl(x)) -> x
1567  LibFunc Func;
1568  Function *F = OpC->getCalledFunction();
1569  if (F && TLI->getLibFunc(F->getName(), Func) && TLI->has(Func) &&
1570  ((Func == LibFunc_atan && Callee->getName() == "tan") ||
1571  (Func == LibFunc_atanf && Callee->getName() == "tanf") ||
1572  (Func == LibFunc_atanl && Callee->getName() == "tanl")))
1573  Ret = OpC->getArgOperand(0);
1574  return Ret;
1575 }
1576 
1577 static bool isTrigLibCall(CallInst *CI) {
1578  // We can only hope to do anything useful if we can ignore things like errno
1579  // and floating-point exceptions.
1580  // We already checked the prototype.
1581  return CI->hasFnAttr(Attribute::NoUnwind) &&
1582  CI->hasFnAttr(Attribute::ReadNone);
1583 }
1584 
1585 static void insertSinCosCall(IRBuilder<> &B, Function *OrigCallee, Value *Arg,
1586  bool UseFloat, Value *&Sin, Value *&Cos,
1587  Value *&SinCos) {
1588  Type *ArgTy = Arg->getType();
1589  Type *ResTy;
1590  StringRef Name;
1591 
1592  Triple T(OrigCallee->getParent()->getTargetTriple());
1593  if (UseFloat) {
1594  Name = "__sincospif_stret";
1595 
1596  assert(T.getArch() != Triple::x86 && "x86 messy and unsupported for now");
1597  // x86_64 can't use {float, float} since that would be returned in both
1598  // xmm0 and xmm1, which isn't what a real struct would do.
1599  ResTy = T.getArch() == Triple::x86_64
1600  ? static_cast<Type *>(VectorType::get(ArgTy, 2))
1601  : static_cast<Type *>(StructType::get(ArgTy, ArgTy));
1602  } else {
1603  Name = "__sincospi_stret";
1604  ResTy = StructType::get(ArgTy, ArgTy);
1605  }
1606 
1607  Module *M = OrigCallee->getParent();
1608  Value *Callee = M->getOrInsertFunction(Name, OrigCallee->getAttributes(),
1609  ResTy, ArgTy);
1610 
1611  if (Instruction *ArgInst = dyn_cast<Instruction>(Arg)) {
1612  // If the argument is an instruction, it must dominate all uses so put our
1613  // sincos call there.
1614  B.SetInsertPoint(ArgInst->getParent(), ++ArgInst->getIterator());
1615  } else {
1616  // Otherwise (e.g. for a constant) the beginning of the function is as
1617  // good a place as any.
1618  BasicBlock &EntryBB = B.GetInsertBlock()->getParent()->getEntryBlock();
1619  B.SetInsertPoint(&EntryBB, EntryBB.begin());
1620  }
1621 
1622  SinCos = B.CreateCall(Callee, Arg, "sincospi");
1623 
1624  if (SinCos->getType()->isStructTy()) {
1625  Sin = B.CreateExtractValue(SinCos, 0, "sinpi");
1626  Cos = B.CreateExtractValue(SinCos, 1, "cospi");
1627  } else {
1628  Sin = B.CreateExtractElement(SinCos, ConstantInt::get(B.getInt32Ty(), 0),
1629  "sinpi");
1630  Cos = B.CreateExtractElement(SinCos, ConstantInt::get(B.getInt32Ty(), 1),
1631  "cospi");
1632  }
1633 }
1634 
1635 Value *LibCallSimplifier::optimizeSinCosPi(CallInst *CI, IRBuilder<> &B) {
1636  // Make sure the prototype is as expected, otherwise the rest of the
1637  // function is probably invalid and likely to abort.
1638  if (!isTrigLibCall(CI))
1639  return nullptr;
1640 
1641  Value *Arg = CI->getArgOperand(0);
1642  SmallVector<CallInst *, 1> SinCalls;
1643  SmallVector<CallInst *, 1> CosCalls;
1644  SmallVector<CallInst *, 1> SinCosCalls;
1645 
1646  bool IsFloat = Arg->getType()->isFloatTy();
1647 
1648  // Look for all compatible sinpi, cospi and sincospi calls with the same
1649  // argument. If there are enough (in some sense) we can make the
1650  // substitution.
1651  Function *F = CI->getFunction();
1652  for (User *U : Arg->users())
1653  classifyArgUse(U, F, IsFloat, SinCalls, CosCalls, SinCosCalls);
1654 
1655  // It's only worthwhile if both sinpi and cospi are actually used.
1656  if (SinCosCalls.empty() && (SinCalls.empty() || CosCalls.empty()))
1657  return nullptr;
1658 
1659  Value *Sin, *Cos, *SinCos;
1660  insertSinCosCall(B, CI->getCalledFunction(), Arg, IsFloat, Sin, Cos, SinCos);
1661 
1662  auto replaceTrigInsts = [this](SmallVectorImpl<CallInst *> &Calls,
1663  Value *Res) {
1664  for (CallInst *C : Calls)
1665  replaceAllUsesWith(C, Res);
1666  };
1667 
1668  replaceTrigInsts(SinCalls, Sin);
1669  replaceTrigInsts(CosCalls, Cos);
1670  replaceTrigInsts(SinCosCalls, SinCos);
1671 
1672  return nullptr;
1673 }
1674 
1675 void LibCallSimplifier::classifyArgUse(
1676  Value *Val, Function *F, bool IsFloat,
1677  SmallVectorImpl<CallInst *> &SinCalls,
1678  SmallVectorImpl<CallInst *> &CosCalls,
1679  SmallVectorImpl<CallInst *> &SinCosCalls) {
1680  CallInst *CI = dyn_cast<CallInst>(Val);
1681 
1682  if (!CI)
1683  return;
1684 
1685  // Don't consider calls in other functions.
1686  if (CI->getFunction() != F)
1687  return;
1688 
1690  LibFunc Func;
1691  if (!Callee || !TLI->getLibFunc(*Callee, Func) || !TLI->has(Func) ||
1692  !isTrigLibCall(CI))
1693  return;
1694 
1695  if (IsFloat) {
1696  if (Func == LibFunc_sinpif)
1697  SinCalls.push_back(CI);
1698  else if (Func == LibFunc_cospif)
1699  CosCalls.push_back(CI);
1700  else if (Func == LibFunc_sincospif_stret)
1701  SinCosCalls.push_back(CI);
1702  } else {
1703  if (Func == LibFunc_sinpi)
1704  SinCalls.push_back(CI);
1705  else if (Func == LibFunc_cospi)
1706  CosCalls.push_back(CI);
1707  else if (Func == LibFunc_sincospi_stret)
1708  SinCosCalls.push_back(CI);
1709  }
1710 }
1711 
1712 //===----------------------------------------------------------------------===//
1713 // Integer Library Call Optimizations
1714 //===----------------------------------------------------------------------===//
1715 
1716 Value *LibCallSimplifier::optimizeFFS(CallInst *CI, IRBuilder<> &B) {
1717  // ffs(x) -> x != 0 ? (i32)llvm.cttz(x)+1 : 0
1718  Value *Op = CI->getArgOperand(0);
1719  Type *ArgType = Op->getType();
1721  Intrinsic::cttz, ArgType);
1722  Value *V = B.CreateCall(F, {Op, B.getTrue()}, "cttz");
1723  V = B.CreateAdd(V, ConstantInt::get(V->getType(), 1));
1724  V = B.CreateIntCast(V, B.getInt32Ty(), false);
1725 
1726  Value *Cond = B.CreateICmpNE(Op, Constant::getNullValue(ArgType));
1727  return B.CreateSelect(Cond, V, B.getInt32(0));
1728 }
1729 
1730 Value *LibCallSimplifier::optimizeFls(CallInst *CI, IRBuilder<> &B) {
1731  // fls(x) -> (i32)(sizeInBits(x) - llvm.ctlz(x, false))
1732  Value *Op = CI->getArgOperand(0);
1733  Type *ArgType = Op->getType();
1735  Intrinsic::ctlz, ArgType);
1736  Value *V = B.CreateCall(F, {Op, B.getFalse()}, "ctlz");
1737  V = B.CreateSub(ConstantInt::get(V->getType(), ArgType->getIntegerBitWidth()),
1738  V);
1739  return B.CreateIntCast(V, CI->getType(), false);
1740 }
1741 
1742 Value *LibCallSimplifier::optimizeAbs(CallInst *CI, IRBuilder<> &B) {
1743  // abs(x) -> x <s 0 ? -x : x
1744  // The negation has 'nsw' because abs of INT_MIN is undefined.
1745  Value *X = CI->getArgOperand(0);
1746  Value *IsNeg = B.CreateICmpSLT(X, Constant::getNullValue(X->getType()));
1747  Value *NegX = B.CreateNSWNeg(X, "neg");
1748  return B.CreateSelect(IsNeg, NegX, X);
1749 }
1750 
1751 Value *LibCallSimplifier::optimizeIsDigit(CallInst *CI, IRBuilder<> &B) {
1752  // isdigit(c) -> (c-'0') <u 10
1753  Value *Op = CI->getArgOperand(0);
1754  Op = B.CreateSub(Op, B.getInt32('0'), "isdigittmp");
1755  Op = B.CreateICmpULT(Op, B.getInt32(10), "isdigit");
1756  return B.CreateZExt(Op, CI->getType());
1757 }
1758 
1759 Value *LibCallSimplifier::optimizeIsAscii(CallInst *CI, IRBuilder<> &B) {
1760  // isascii(c) -> c <u 128
1761  Value *Op = CI->getArgOperand(0);
1762  Op = B.CreateICmpULT(Op, B.getInt32(128), "isascii");
1763  return B.CreateZExt(Op, CI->getType());
1764 }
1765 
1766 Value *LibCallSimplifier::optimizeToAscii(CallInst *CI, IRBuilder<> &B) {
1767  // toascii(c) -> c & 0x7f
1768  return B.CreateAnd(CI->getArgOperand(0),
1769  ConstantInt::get(CI->getType(), 0x7F));
1770 }
1771 
1772 Value *LibCallSimplifier::optimizeAtoi(CallInst *CI, IRBuilder<> &B) {
1773  StringRef Str;
1774  if (!getConstantStringInfo(CI->getArgOperand(0), Str))
1775  return nullptr;
1776 
1777  return convertStrToNumber(CI, Str, 10);
1778 }
1779 
1780 Value *LibCallSimplifier::optimizeStrtol(CallInst *CI, IRBuilder<> &B) {
1781  StringRef Str;
1782  if (!getConstantStringInfo(CI->getArgOperand(0), Str))
1783  return nullptr;
1784 
1785  if (!isa<ConstantPointerNull>(CI->getArgOperand(1)))
1786  return nullptr;
1787 
1788  if (ConstantInt *CInt = dyn_cast<ConstantInt>(CI->getArgOperand(2))) {
1789  return convertStrToNumber(CI, Str, CInt->getSExtValue());
1790  }
1791 
1792  return nullptr;
1793 }
1794 
1795 //===----------------------------------------------------------------------===//
1796 // Formatting and IO Library Call Optimizations
1797 //===----------------------------------------------------------------------===//
1798 
1799 static bool isReportingError(Function *Callee, CallInst *CI, int StreamArg);
1800 
1801 Value *LibCallSimplifier::optimizeErrorReporting(CallInst *CI, IRBuilder<> &B,
1802  int StreamArg) {
1803  Function *Callee = CI->getCalledFunction();
1804  // Error reporting calls should be cold, mark them as such.
1805  // This applies even to non-builtin calls: it is only a hint and applies to
1806  // functions that the frontend might not understand as builtins.
1807 
1808  // This heuristic was suggested in:
1809  // Improving Static Branch Prediction in a Compiler
1810  // Brian L. Deitrich, Ben-Chung Cheng, Wen-mei W. Hwu
1811  // Proceedings of PACT'98, Oct. 1998, IEEE
1812  if (!CI->hasFnAttr(Attribute::Cold) &&
1813  isReportingError(Callee, CI, StreamArg)) {
1815  }
1816 
1817  return nullptr;
1818 }
1819 
1820 static bool isReportingError(Function *Callee, CallInst *CI, int StreamArg) {
1821  if (!Callee || !Callee->isDeclaration())
1822  return false;
1823 
1824  if (StreamArg < 0)
1825  return true;
1826 
1827  // These functions might be considered cold, but only if their stream
1828  // argument is stderr.
1829 
1830  if (StreamArg >= (int)CI->getNumArgOperands())
1831  return false;
1832  LoadInst *LI = dyn_cast<LoadInst>(CI->getArgOperand(StreamArg));
1833  if (!LI)
1834  return false;
1836  if (!GV || !GV->isDeclaration())
1837  return false;
1838  return GV->getName() == "stderr";
1839 }
1840 
1841 Value *LibCallSimplifier::optimizePrintFString(CallInst *CI, IRBuilder<> &B) {
1842  // Check for a fixed format string.
1843  StringRef FormatStr;
1844  if (!getConstantStringInfo(CI->getArgOperand(0), FormatStr))
1845  return nullptr;
1846 
1847  // Empty format string -> noop.
1848  if (FormatStr.empty()) // Tolerate printf's declared void.
1849  return CI->use_empty() ? (Value *)CI : ConstantInt::get(CI->getType(), 0);
1850 
1851  // Do not do any of the following transformations if the printf return value
1852  // is used, in general the printf return value is not compatible with either
1853  // putchar() or puts().
1854  if (!CI->use_empty())
1855  return nullptr;
1856 
1857  // printf("x") -> putchar('x'), even for "%" and "%%".
1858  if (FormatStr.size() == 1 || FormatStr == "%%")
1859  return emitPutChar(B.getInt32(FormatStr[0]), B, TLI);
1860 
1861  // printf("%s", "a") --> putchar('a')
1862  if (FormatStr == "%s" && CI->getNumArgOperands() > 1) {
1863  StringRef ChrStr;
1864  if (!getConstantStringInfo(CI->getOperand(1), ChrStr))
1865  return nullptr;
1866  if (ChrStr.size() != 1)
1867  return nullptr;
1868  return emitPutChar(B.getInt32(ChrStr[0]), B, TLI);
1869  }
1870 
1871  // printf("foo\n") --> puts("foo")
1872  if (FormatStr[FormatStr.size() - 1] == '\n' &&
1873  FormatStr.find('%') == StringRef::npos) { // No format characters.
1874  // Create a string literal with no \n on it. We expect the constant merge
1875  // pass to be run after this pass, to merge duplicate strings.
1876  FormatStr = FormatStr.drop_back();
1877  Value *GV = B.CreateGlobalString(FormatStr, "str");
1878  return emitPutS(GV, B, TLI);
1879  }
1880 
1881  // Optimize specific format strings.
1882  // printf("%c", chr) --> putchar(chr)
1883  if (FormatStr == "%c" && CI->getNumArgOperands() > 1 &&
1884  CI->getArgOperand(1)->getType()->isIntegerTy())
1885  return emitPutChar(CI->getArgOperand(1), B, TLI);
1886 
1887  // printf("%s\n", str) --> puts(str)
1888  if (FormatStr == "%s\n" && CI->getNumArgOperands() > 1 &&
1889  CI->getArgOperand(1)->getType()->isPointerTy())
1890  return emitPutS(CI->getArgOperand(1), B, TLI);
1891  return nullptr;
1892 }
1893 
1894 Value *LibCallSimplifier::optimizePrintF(CallInst *CI, IRBuilder<> &B) {
1895 
1896  Function *Callee = CI->getCalledFunction();
1897  FunctionType *FT = Callee->getFunctionType();
1898  if (Value *V = optimizePrintFString(CI, B)) {
1899  return V;
1900  }
1901 
1902  // printf(format, ...) -> iprintf(format, ...) if no floating point
1903  // arguments.
1904  if (TLI->has(LibFunc_iprintf) && !callHasFloatingPointArgument(CI)) {
1905  Module *M = B.GetInsertBlock()->getParent()->getParent();
1906  Constant *IPrintFFn =
1907  M->getOrInsertFunction("iprintf", FT, Callee->getAttributes());
1908  CallInst *New = cast<CallInst>(CI->clone());
1909  New->setCalledFunction(IPrintFFn);
1910  B.Insert(New);
1911  return New;
1912  }
1913  return nullptr;
1914 }
1915 
1916 Value *LibCallSimplifier::optimizeSPrintFString(CallInst *CI, IRBuilder<> &B) {
1917  // Check for a fixed format string.
1918  StringRef FormatStr;
1919  if (!getConstantStringInfo(CI->getArgOperand(1), FormatStr))
1920  return nullptr;
1921 
1922  // If we just have a format string (nothing else crazy) transform it.
1923  if (CI->getNumArgOperands() == 2) {
1924  // Make sure there's no % in the constant array. We could try to handle
1925  // %% -> % in the future if we cared.
1926  if (FormatStr.find('%') != StringRef::npos)
1927  return nullptr; // we found a format specifier, bail out.
1928 
1929  // sprintf(str, fmt) -> llvm.memcpy(align 1 str, align 1 fmt, strlen(fmt)+1)
1930  B.CreateMemCpy(CI->getArgOperand(0), 1, CI->getArgOperand(1), 1,
1931  ConstantInt::get(DL.getIntPtrType(CI->getContext()),
1932  FormatStr.size() + 1)); // Copy the null byte.
1933  return ConstantInt::get(CI->getType(), FormatStr.size());
1934  }
1935 
1936  // The remaining optimizations require the format string to be "%s" or "%c"
1937  // and have an extra operand.
1938  if (FormatStr.size() != 2 || FormatStr[0] != '%' ||
1939  CI->getNumArgOperands() < 3)
1940  return nullptr;
1941 
1942  // Decode the second character of the format string.
1943  if (FormatStr[1] == 'c') {
1944  // sprintf(dst, "%c", chr) --> *(i8*)dst = chr; *((i8*)dst+1) = 0
1945  if (!CI->getArgOperand(2)->getType()->isIntegerTy())
1946  return nullptr;
1947  Value *V = B.CreateTrunc(CI->getArgOperand(2), B.getInt8Ty(), "char");
1948  Value *Ptr = castToCStr(CI->getArgOperand(0), B);
1949  B.CreateStore(V, Ptr);
1950  Ptr = B.CreateGEP(B.getInt8Ty(), Ptr, B.getInt32(1), "nul");
1951  B.CreateStore(B.getInt8(0), Ptr);
1952 
1953  return ConstantInt::get(CI->getType(), 1);
1954  }
1955 
1956  if (FormatStr[1] == 's') {
1957  // sprintf(dest, "%s", str) -> llvm.memcpy(dest, str, strlen(str)+1, 1)
1958  if (!CI->getArgOperand(2)->getType()->isPointerTy())
1959  return nullptr;
1960 
1961  Value *Len = emitStrLen(CI->getArgOperand(2), B, DL, TLI);
1962  if (!Len)
1963  return nullptr;
1964  Value *IncLen =
1965  B.CreateAdd(Len, ConstantInt::get(Len->getType(), 1), "leninc");
1966  B.CreateMemCpy(CI->getArgOperand(0), 1, CI->getArgOperand(2), 1, IncLen);
1967 
1968  // The sprintf result is the unincremented number of bytes in the string.
1969  return B.CreateIntCast(Len, CI->getType(), false);
1970  }
1971  return nullptr;
1972 }
1973 
1974 Value *LibCallSimplifier::optimizeSPrintF(CallInst *CI, IRBuilder<> &B) {
1975  Function *Callee = CI->getCalledFunction();
1976  FunctionType *FT = Callee->getFunctionType();
1977  if (Value *V = optimizeSPrintFString(CI, B)) {
1978  return V;
1979  }
1980 
1981  // sprintf(str, format, ...) -> siprintf(str, format, ...) if no floating
1982  // point arguments.
1983  if (TLI->has(LibFunc_siprintf) && !callHasFloatingPointArgument(CI)) {
1984  Module *M = B.GetInsertBlock()->getParent()->getParent();
1985  Constant *SIPrintFFn =
1986  M->getOrInsertFunction("siprintf", FT, Callee->getAttributes());
1987  CallInst *New = cast<CallInst>(CI->clone());
1988  New->setCalledFunction(SIPrintFFn);
1989  B.Insert(New);
1990  return New;
1991  }
1992  return nullptr;
1993 }
1994 
1995 Value *LibCallSimplifier::optimizeSnPrintFString(CallInst *CI, IRBuilder<> &B) {
1996  // Check for a fixed format string.
1997  StringRef FormatStr;
1998  if (!getConstantStringInfo(CI->getArgOperand(2), FormatStr))
1999  return nullptr;
2000 
2001  // Check for size
2003  if (!Size)
2004  return nullptr;
2005 
2006  uint64_t N = Size->getZExtValue();
2007 
2008  // If we just have a format string (nothing else crazy) transform it.
2009  if (CI->getNumArgOperands() == 3) {
2010  // Make sure there's no % in the constant array. We could try to handle
2011  // %% -> % in the future if we cared.
2012  if (FormatStr.find('%') != StringRef::npos)
2013  return nullptr; // we found a format specifier, bail out.
2014 
2015  if (N == 0)
2016  return ConstantInt::get(CI->getType(), FormatStr.size());
2017  else if (N < FormatStr.size() + 1)
2018  return nullptr;
2019 
2020  // sprintf(str, size, fmt) -> llvm.memcpy(align 1 str, align 1 fmt,
2021  // strlen(fmt)+1)
2022  B.CreateMemCpy(
2023  CI->getArgOperand(0), 1, CI->getArgOperand(2), 1,
2024  ConstantInt::get(DL.getIntPtrType(CI->getContext()),
2025  FormatStr.size() + 1)); // Copy the null byte.
2026  return ConstantInt::get(CI->getType(), FormatStr.size());
2027  }
2028 
2029  // The remaining optimizations require the format string to be "%s" or "%c"
2030  // and have an extra operand.
2031  if (FormatStr.size() == 2 && FormatStr[0] == '%' &&
2032  CI->getNumArgOperands() == 4) {
2033 
2034  // Decode the second character of the format string.
2035  if (FormatStr[1] == 'c') {
2036  if (N == 0)
2037  return ConstantInt::get(CI->getType(), 1);
2038  else if (N == 1)
2039  return nullptr;
2040 
2041  // snprintf(dst, size, "%c", chr) --> *(i8*)dst = chr; *((i8*)dst+1) = 0
2042  if (!CI->getArgOperand(3)->getType()->isIntegerTy())
2043  return nullptr;
2044  Value *V = B.CreateTrunc(CI->getArgOperand(3), B.getInt8Ty(), "char");
2045  Value *Ptr = castToCStr(CI->getArgOperand(0), B);
2046  B.CreateStore(V, Ptr);
2047  Ptr = B.CreateGEP(B.getInt8Ty(), Ptr, B.getInt32(1), "nul");
2048  B.CreateStore(B.getInt8(0), Ptr);
2049 
2050  return ConstantInt::get(CI->getType(), 1);
2051  }
2052 
2053  if (FormatStr[1] == 's') {
2054  // snprintf(dest, size, "%s", str) to llvm.memcpy(dest, str, len+1, 1)
2055  StringRef Str;
2056  if (!getConstantStringInfo(CI->getArgOperand(3), Str))
2057  return nullptr;
2058 
2059  if (N == 0)
2060  return ConstantInt::get(CI->getType(), Str.size());
2061  else if (N < Str.size() + 1)
2062  return nullptr;
2063 
2064  B.CreateMemCpy(CI->getArgOperand(0), 1, CI->getArgOperand(3), 1,
2065  ConstantInt::get(CI->getType(), Str.size() + 1));
2066 
2067  // The snprintf result is the unincremented number of bytes in the string.
2068  return ConstantInt::get(CI->getType(), Str.size());
2069  }
2070  }
2071  return nullptr;
2072 }
2073 
2074 Value *LibCallSimplifier::optimizeSnPrintF(CallInst *CI, IRBuilder<> &B) {
2075  if (Value *V = optimizeSnPrintFString(CI, B)) {
2076  return V;
2077  }
2078 
2079  return nullptr;
2080 }
2081 
2082 Value *LibCallSimplifier::optimizeFPrintFString(CallInst *CI, IRBuilder<> &B) {
2083  optimizeErrorReporting(CI, B, 0);
2084 
2085  // All the optimizations depend on the format string.
2086  StringRef FormatStr;
2087  if (!getConstantStringInfo(CI->getArgOperand(1), FormatStr))
2088  return nullptr;
2089 
2090  // Do not do any of the following transformations if the fprintf return
2091  // value is used, in general the fprintf return value is not compatible
2092  // with fwrite(), fputc() or fputs().
2093  if (!CI->use_empty())
2094  return nullptr;
2095 
2096  // fprintf(F, "foo") --> fwrite("foo", 3, 1, F)
2097  if (CI->getNumArgOperands() == 2) {
2098  // Could handle %% -> % if we cared.
2099  if (FormatStr.find('%') != StringRef::npos)
2100  return nullptr; // We found a format specifier.
2101 
2102  return emitFWrite(
2103  CI->getArgOperand(1),
2104  ConstantInt::get(DL.getIntPtrType(CI->getContext()), FormatStr.size()),
2105  CI->getArgOperand(0), B, DL, TLI);
2106  }
2107 
2108  // The remaining optimizations require the format string to be "%s" or "%c"
2109  // and have an extra operand.
2110  if (FormatStr.size() != 2 || FormatStr[0] != '%' ||
2111  CI->getNumArgOperands() < 3)
2112  return nullptr;
2113 
2114  // Decode the second character of the format string.
2115  if (FormatStr[1] == 'c') {
2116  // fprintf(F, "%c", chr) --> fputc(chr, F)
2117  if (!CI->getArgOperand(2)->getType()->isIntegerTy())
2118  return nullptr;
2119  return emitFPutC(CI->getArgOperand(2), CI->getArgOperand(0), B, TLI);
2120  }
2121 
2122  if (FormatStr[1] == 's') {
2123  // fprintf(F, "%s", str) --> fputs(str, F)
2124  if (!CI->getArgOperand(2)->getType()->isPointerTy())
2125  return nullptr;
2126  return emitFPutS(CI->getArgOperand(2), CI->getArgOperand(0), B, TLI);
2127  }
2128  return nullptr;
2129 }
2130 
2131 Value *LibCallSimplifier::optimizeFPrintF(CallInst *CI, IRBuilder<> &B) {
2132  Function *Callee = CI->getCalledFunction();
2133  FunctionType *FT = Callee->getFunctionType();
2134  if (Value *V = optimizeFPrintFString(CI, B)) {
2135  return V;
2136  }
2137 
2138  // fprintf(stream, format, ...) -> fiprintf(stream, format, ...) if no
2139  // floating point arguments.
2140  if (TLI->has(LibFunc_fiprintf) && !callHasFloatingPointArgument(CI)) {
2141  Module *M = B.GetInsertBlock()->getParent()->getParent();
2142  Constant *FIPrintFFn =
2143  M->getOrInsertFunction("fiprintf", FT, Callee->getAttributes());
2144  CallInst *New = cast<CallInst>(CI->clone());
2145  New->setCalledFunction(FIPrintFFn);
2146  B.Insert(New);
2147  return New;
2148  }
2149  return nullptr;
2150 }
2151 
2152 Value *LibCallSimplifier::optimizeFWrite(CallInst *CI, IRBuilder<> &B) {
2153  optimizeErrorReporting(CI, B, 3);
2154 
2155  // Get the element size and count.
2156  ConstantInt *SizeC = dyn_cast<ConstantInt>(CI->getArgOperand(1));
2157  ConstantInt *CountC = dyn_cast<ConstantInt>(CI->getArgOperand(2));
2158  if (SizeC && CountC) {
2159  uint64_t Bytes = SizeC->getZExtValue() * CountC->getZExtValue();
2160 
2161  // If this is writing zero records, remove the call (it's a noop).
2162  if (Bytes == 0)
2163  return ConstantInt::get(CI->getType(), 0);
2164 
2165  // If this is writing one byte, turn it into fputc.
2166  // This optimisation is only valid, if the return value is unused.
2167  if (Bytes == 1 && CI->use_empty()) { // fwrite(S,1,1,F) -> fputc(S[0],F)
2168  Value *Char = B.CreateLoad(castToCStr(CI->getArgOperand(0), B), "char");
2169  Value *NewCI = emitFPutC(Char, CI->getArgOperand(3), B, TLI);
2170  return NewCI ? ConstantInt::get(CI->getType(), 1) : nullptr;
2171  }
2172  }
2173 
2174  if (isLocallyOpenedFile(CI->getArgOperand(3), CI, B, TLI))
2175  return emitFWriteUnlocked(CI->getArgOperand(0), CI->getArgOperand(1),
2176  CI->getArgOperand(2), CI->getArgOperand(3), B, DL,
2177  TLI);
2178 
2179  return nullptr;
2180 }
2181 
2182 Value *LibCallSimplifier::optimizeFPuts(CallInst *CI, IRBuilder<> &B) {
2183  optimizeErrorReporting(CI, B, 1);
2184 
2185  // Don't rewrite fputs to fwrite when optimising for size because fwrite
2186  // requires more arguments and thus extra MOVs are required.
2187  if (CI->getFunction()->optForSize())
2188  return nullptr;
2189 
2190  // Check if has any use
2191  if (!CI->use_empty()) {
2192  if (isLocallyOpenedFile(CI->getArgOperand(1), CI, B, TLI))
2193  return emitFPutSUnlocked(CI->getArgOperand(0), CI->getArgOperand(1), B,
2194  TLI);
2195  else
2196  // We can't optimize if return value is used.
2197  return nullptr;
2198  }
2199 
2200  // fputs(s,F) --> fwrite(s,1,strlen(s),F)
2201  uint64_t Len = GetStringLength(CI->getArgOperand(0));
2202  if (!Len)
2203  return nullptr;
2204 
2205  // Known to have no uses (see above).
2206  return emitFWrite(
2207  CI->getArgOperand(0),
2208  ConstantInt::get(DL.getIntPtrType(CI->getContext()), Len - 1),
2209  CI->getArgOperand(1), B, DL, TLI);
2210 }
2211 
2212 Value *LibCallSimplifier::optimizeFPutc(CallInst *CI, IRBuilder<> &B) {
2213  optimizeErrorReporting(CI, B, 1);
2214 
2215  if (isLocallyOpenedFile(CI->getArgOperand(1), CI, B, TLI))
2216  return emitFPutCUnlocked(CI->getArgOperand(0), CI->getArgOperand(1), B,
2217  TLI);
2218 
2219  return nullptr;
2220 }
2221 
2222 Value *LibCallSimplifier::optimizeFGetc(CallInst *CI, IRBuilder<> &B) {
2223  if (isLocallyOpenedFile(CI->getArgOperand(0), CI, B, TLI))
2224  return emitFGetCUnlocked(CI->getArgOperand(0), B, TLI);
2225 
2226  return nullptr;
2227 }
2228 
2229 Value *LibCallSimplifier::optimizeFGets(CallInst *CI, IRBuilder<> &B) {
2230  if (isLocallyOpenedFile(CI->getArgOperand(2), CI, B, TLI))
2231  return emitFGetSUnlocked(CI->getArgOperand(0), CI->getArgOperand(1),
2232  CI->getArgOperand(2), B, TLI);
2233 
2234  return nullptr;
2235 }
2236 
2237 Value *LibCallSimplifier::optimizeFRead(CallInst *CI, IRBuilder<> &B) {
2238  if (isLocallyOpenedFile(CI->getArgOperand(3), CI, B, TLI))
2239  return emitFReadUnlocked(CI->getArgOperand(0), CI->getArgOperand(1),
2240  CI->getArgOperand(2), CI->getArgOperand(3), B, DL,
2241  TLI);
2242 
2243  return nullptr;
2244 }
2245 
2246 Value *LibCallSimplifier::optimizePuts(CallInst *CI, IRBuilder<> &B) {
2247  // Check for a constant string.
2248  StringRef Str;
2249  if (!getConstantStringInfo(CI->getArgOperand(0), Str))
2250  return nullptr;
2251 
2252  if (Str.empty() && CI->use_empty()) {
2253  // puts("") -> putchar('\n')
2254  Value *Res = emitPutChar(B.getInt32('\n'), B, TLI);
2255  if (CI->use_empty() || !Res)
2256  return Res;
2257  return B.CreateIntCast(Res, CI->getType(), true);
2258  }
2259 
2260  return nullptr;
2261 }
2262 
2263 bool LibCallSimplifier::hasFloatVersion(StringRef FuncName) {
2264  LibFunc Func;
2265  SmallString<20> FloatFuncName = FuncName;
2266  FloatFuncName += 'f';
2267  if (TLI->getLibFunc(FloatFuncName, Func))
2268  return TLI->has(Func);
2269  return false;
2270 }
2271 
2272 Value *LibCallSimplifier::optimizeStringMemoryLibCall(CallInst *CI,
2273  IRBuilder<> &Builder) {
2274  LibFunc Func;
2275  Function *Callee = CI->getCalledFunction();
2276  // Check for string/memory library functions.
2277  if (TLI->getLibFunc(*Callee, Func) && TLI->has(Func)) {
2278  // Make sure we never change the calling convention.
2279  assert((ignoreCallingConv(Func) ||
2280  isCallingConvCCompatible(CI)) &&
2281  "Optimizing string/memory libcall would change the calling convention");
2282  switch (Func) {
2283  case LibFunc_strcat:
2284  return optimizeStrCat(CI, Builder);
2285  case LibFunc_strncat:
2286  return optimizeStrNCat(CI, Builder);
2287  case LibFunc_strchr:
2288  return optimizeStrChr(CI, Builder);
2289  case LibFunc_strrchr:
2290  return optimizeStrRChr(CI, Builder);
2291  case LibFunc_strcmp:
2292  return optimizeStrCmp(CI, Builder);
2293  case LibFunc_strncmp:
2294  return optimizeStrNCmp(CI, Builder);
2295  case LibFunc_strcpy:
2296  return optimizeStrCpy(CI, Builder);
2297  case LibFunc_stpcpy:
2298  return optimizeStpCpy(CI, Builder);
2299  case LibFunc_strncpy:
2300  return optimizeStrNCpy(CI, Builder);
2301  case LibFunc_strlen:
2302  return optimizeStrLen(CI, Builder);
2303  case LibFunc_strpbrk:
2304  return optimizeStrPBrk(CI, Builder);
2305  case LibFunc_strtol:
2306  case LibFunc_strtod:
2307  case LibFunc_strtof:
2308  case LibFunc_strtoul:
2309  case LibFunc_strtoll:
2310  case LibFunc_strtold:
2311  case LibFunc_strtoull:
2312  return optimizeStrTo(CI, Builder);
2313  case LibFunc_strspn:
2314  return optimizeStrSpn(CI, Builder);
2315  case LibFunc_strcspn:
2316  return optimizeStrCSpn(CI, Builder);
2317  case LibFunc_strstr:
2318  return optimizeStrStr(CI, Builder);
2319  case LibFunc_memchr:
2320  return optimizeMemChr(CI, Builder);
2321  case LibFunc_memcmp:
2322  return optimizeMemCmp(CI, Builder);
2323  case LibFunc_memcpy:
2324  return optimizeMemCpy(CI, Builder);
2325  case LibFunc_memmove:
2326  return optimizeMemMove(CI, Builder);
2327  case LibFunc_memset:
2328  return optimizeMemSet(CI, Builder);
2329  case LibFunc_realloc:
2330  return optimizeRealloc(CI, Builder);
2331  case LibFunc_wcslen:
2332  return optimizeWcslen(CI, Builder);
2333  default:
2334  break;
2335  }
2336  }
2337  return nullptr;
2338 }
2339 
2340 Value *LibCallSimplifier::optimizeFloatingPointLibCall(CallInst *CI,
2341  LibFunc Func,
2342  IRBuilder<> &Builder) {
2343  // Don't optimize calls that require strict floating point semantics.
2344  if (CI->isStrictFP())
2345  return nullptr;
2346 
2347  if (Value *V = optimizeTrigReflections(CI, Func, Builder))
2348  return V;
2349 
2350  switch (Func) {
2351  case LibFunc_sinpif:
2352  case LibFunc_sinpi:
2353  case LibFunc_cospif:
2354  case LibFunc_cospi:
2355  return optimizeSinCosPi(CI, Builder);
2356  case LibFunc_powf:
2357  case LibFunc_pow:
2358  case LibFunc_powl:
2359  return optimizePow(CI, Builder);
2360  case LibFunc_exp2l:
2361  case LibFunc_exp2:
2362  case LibFunc_exp2f:
2363  return optimizeExp2(CI, Builder);
2364  case LibFunc_fabsf:
2365  case LibFunc_fabs:
2366  case LibFunc_fabsl:
2367  return replaceUnaryCall(CI, Builder, Intrinsic::fabs);
2368  case LibFunc_sqrtf:
2369  case LibFunc_sqrt:
2370  case LibFunc_sqrtl:
2371  return optimizeSqrt(CI, Builder);
2372  case LibFunc_log:
2373  case LibFunc_log10:
2374  case LibFunc_log1p:
2375  case LibFunc_log2:
2376  case LibFunc_logb:
2377  return optimizeLog(CI, Builder);
2378  case LibFunc_tan:
2379  case LibFunc_tanf:
2380  case LibFunc_tanl:
2381  return optimizeTan(CI, Builder);
2382  case LibFunc_ceil:
2383  return replaceUnaryCall(CI, Builder, Intrinsic::ceil);
2384  case LibFunc_floor:
2385  return replaceUnaryCall(CI, Builder, Intrinsic::floor);
2386  case LibFunc_round:
2387  return replaceUnaryCall(CI, Builder, Intrinsic::round);
2388  case LibFunc_nearbyint:
2389  return replaceUnaryCall(CI, Builder, Intrinsic::nearbyint);
2390  case LibFunc_rint:
2391  return replaceUnaryCall(CI, Builder, Intrinsic::rint);
2392  case LibFunc_trunc:
2393  return replaceUnaryCall(CI, Builder, Intrinsic::trunc);
2394  case LibFunc_acos:
2395  case LibFunc_acosh:
2396  case LibFunc_asin:
2397  case LibFunc_asinh:
2398  case LibFunc_atan:
2399  case LibFunc_atanh:
2400  case LibFunc_cbrt:
2401  case LibFunc_cosh:
2402  case LibFunc_exp:
2403  case LibFunc_exp10:
2404  case LibFunc_expm1:
2405  case LibFunc_cos:
2406  case LibFunc_sin:
2407  case LibFunc_sinh:
2408  case LibFunc_tanh:
2409  if (UnsafeFPShrink && hasFloatVersion(CI->getCalledFunction()->getName()))
2410  return optimizeUnaryDoubleFP(CI, Builder, true);
2411  return nullptr;
2412  case LibFunc_copysign:
2413  if (hasFloatVersion(CI->getCalledFunction()->getName()))
2414  return optimizeBinaryDoubleFP(CI, Builder);
2415  return nullptr;
2416  case LibFunc_fminf:
2417  case LibFunc_fmin:
2418  case LibFunc_fminl:
2419  case LibFunc_fmaxf:
2420  case LibFunc_fmax:
2421  case LibFunc_fmaxl:
2422  return optimizeFMinFMax(CI, Builder);
2423  case LibFunc_cabs:
2424  case LibFunc_cabsf:
2425  case LibFunc_cabsl:
2426  return optimizeCAbs(CI, Builder);
2427  default:
2428  return nullptr;
2429  }
2430 }
2431 
2433  // TODO: Split out the code below that operates on FP calls so that
2434  // we can all non-FP calls with the StrictFP attribute to be
2435  // optimized.
2436  if (CI->isNoBuiltin())
2437  return nullptr;
2438 
2439  LibFunc Func;
2440  Function *Callee = CI->getCalledFunction();
2441 
2443  CI->getOperandBundlesAsDefs(OpBundles);
2444  IRBuilder<> Builder(CI, /*FPMathTag=*/nullptr, OpBundles);
2445  bool isCallingConvC = isCallingConvCCompatible(CI);
2446 
2447  // Command-line parameter overrides instruction attribute.
2448  // This can't be moved to optimizeFloatingPointLibCall() because it may be
2449  // used by the intrinsic optimizations.
2450  if (EnableUnsafeFPShrink.getNumOccurrences() > 0)
2451  UnsafeFPShrink = EnableUnsafeFPShrink;
2452  else if (isa<FPMathOperator>(CI) && CI->isFast())
2453  UnsafeFPShrink = true;
2454 
2455  // First, check for intrinsics.
2456  if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI)) {
2457  if (!isCallingConvC)
2458  return nullptr;
2459  // The FP intrinsics have corresponding constrained versions so we don't
2460  // need to check for the StrictFP attribute here.
2461  switch (II->getIntrinsicID()) {
2462  case Intrinsic::pow:
2463  return optimizePow(CI, Builder);
2464  case Intrinsic::exp2:
2465  return optimizeExp2(CI, Builder);
2466  case Intrinsic::log:
2467  return optimizeLog(CI, Builder);
2468  case Intrinsic::sqrt:
2469  return optimizeSqrt(CI, Builder);
2470  // TODO: Use foldMallocMemset() with memset intrinsic.
2471  default:
2472  return nullptr;
2473  }
2474  }
2475 
2476  // Also try to simplify calls to fortified library functions.
2477  if (Value *SimplifiedFortifiedCI = FortifiedSimplifier.optimizeCall(CI)) {
2478  // Try to further simplify the result.
2479  CallInst *SimplifiedCI = dyn_cast<CallInst>(SimplifiedFortifiedCI);
2480  if (SimplifiedCI && SimplifiedCI->getCalledFunction()) {
2481  // Use an IR Builder from SimplifiedCI if available instead of CI
2482  // to guarantee we reach all uses we might replace later on.
2483  IRBuilder<> TmpBuilder(SimplifiedCI);
2484  if (Value *V = optimizeStringMemoryLibCall(SimplifiedCI, TmpBuilder)) {
2485  // If we were able to further simplify, remove the now redundant call.
2486  SimplifiedCI->replaceAllUsesWith(V);
2487  SimplifiedCI->eraseFromParent();
2488  return V;
2489  }
2490  }
2491  return SimplifiedFortifiedCI;
2492  }
2493 
2494  // Then check for known library functions.
2495  if (TLI->getLibFunc(*Callee, Func) && TLI->has(Func)) {
2496  // We never change the calling convention.
2497  if (!ignoreCallingConv(Func) && !isCallingConvC)
2498  return nullptr;
2499  if (Value *V = optimizeStringMemoryLibCall(CI, Builder))
2500  return V;
2501  if (Value *V = optimizeFloatingPointLibCall(CI, Func, Builder))
2502  return V;
2503  switch (Func) {
2504  case LibFunc_ffs:
2505  case LibFunc_ffsl:
2506  case LibFunc_ffsll:
2507  return optimizeFFS(CI, Builder);
2508  case LibFunc_fls:
2509  case LibFunc_flsl:
2510  case LibFunc_flsll:
2511  return optimizeFls(CI, Builder);
2512  case LibFunc_abs:
2513  case LibFunc_labs:
2514  case LibFunc_llabs:
2515  return optimizeAbs(CI, Builder);
2516  case LibFunc_isdigit:
2517  return optimizeIsDigit(CI, Builder);
2518  case LibFunc_isascii:
2519  return optimizeIsAscii(CI, Builder);
2520  case LibFunc_toascii:
2521  return optimizeToAscii(CI, Builder);
2522  case LibFunc_atoi:
2523  case LibFunc_atol:
2524  case LibFunc_atoll:
2525  return optimizeAtoi(CI, Builder);
2526  case LibFunc_strtol:
2527  case LibFunc_strtoll:
2528  return optimizeStrtol(CI, Builder);
2529  case LibFunc_printf:
2530  return optimizePrintF(CI, Builder);
2531  case LibFunc_sprintf:
2532  return optimizeSPrintF(CI, Builder);
2533  case LibFunc_snprintf:
2534  return optimizeSnPrintF(CI, Builder);
2535  case LibFunc_fprintf:
2536  return optimizeFPrintF(CI, Builder);
2537  case LibFunc_fwrite:
2538  return optimizeFWrite(CI, Builder);
2539  case LibFunc_fread:
2540  return optimizeFRead(CI, Builder);
2541  case LibFunc_fputs:
2542  return optimizeFPuts(CI, Builder);
2543  case LibFunc_fgets:
2544  return optimizeFGets(CI, Builder);
2545  case LibFunc_fputc:
2546  return optimizeFPutc(CI, Builder);
2547  case LibFunc_fgetc:
2548  return optimizeFGetc(CI, Builder);
2549  case LibFunc_puts:
2550  return optimizePuts(CI, Builder);
2551  case LibFunc_perror:
2552  return optimizeErrorReporting(CI, Builder);
2553  case LibFunc_vfprintf:
2554  case LibFunc_fiprintf:
2555  return optimizeErrorReporting(CI, Builder, 0);
2556  default:
2557  return nullptr;
2558  }
2559  }
2560  return nullptr;
2561 }
2562 
2564  const DataLayout &DL, const TargetLibraryInfo *TLI,
2566  function_ref<void(Instruction *, Value *)> Replacer)
2567  : FortifiedSimplifier(TLI), DL(DL), TLI(TLI), ORE(ORE),
2568  UnsafeFPShrink(false), Replacer(Replacer) {}
2569 
2570 void LibCallSimplifier::replaceAllUsesWith(Instruction *I, Value *With) {
2571  // Indirect through the replacer used in this instance.
2572  Replacer(I, With);
2573 }
2574 
2575 // TODO:
2576 // Additional cases that we need to add to this file:
2577 //
2578 // cbrt:
2579 // * cbrt(expN(X)) -> expN(x/3)
2580 // * cbrt(sqrt(x)) -> pow(x,1/6)
2581 // * cbrt(cbrt(x)) -> pow(x,1/9)
2582 //
2583 // exp, expf, expl:
2584 // * exp(log(x)) -> x
2585 //
2586 // log, logf, logl:
2587 // * log(exp(x)) -> x
2588 // * log(exp(y)) -> y*log(e)
2589 // * log(exp10(y)) -> y*log(10)
2590 // * log(sqrt(x)) -> 0.5*log(x)
2591 //
2592 // pow, powf, powl:
2593 // * pow(sqrt(x),y) -> pow(x,y*0.5)
2594 // * pow(pow(x,y),z)-> pow(x,y*z)
2595 //
2596 // signbit:
2597 // * signbit(cnst) -> cnst'
2598 // * signbit(nncst) -> 0 (if pstv is a non-negative constant)
2599 //
2600 // sqrt, sqrtf, sqrtl:
2601 // * sqrt(expN(x)) -> expN(x*0.5)
2602 // * sqrt(Nroot(x)) -> pow(x,1/(2*N))
2603 // * sqrt(pow(x,y)) -> pow(|x|,y*0.5)
2604 //
2605 
2606 //===----------------------------------------------------------------------===//
2607 // Fortified Library Call Optimizations
2608 //===----------------------------------------------------------------------===//
2609 
2610 bool FortifiedLibCallSimplifier::isFortifiedCallFoldable(CallInst *CI,
2611  unsigned ObjSizeOp,
2612  unsigned SizeOp,
2613  bool isString) {
2614  if (CI->getArgOperand(ObjSizeOp) == CI->getArgOperand(SizeOp))
2615  return true;
2616  if (ConstantInt *ObjSizeCI =
2617  dyn_cast<ConstantInt>(CI->getArgOperand(ObjSizeOp))) {
2618  if (ObjSizeCI->isMinusOne())
2619  return true;
2620  // If the object size wasn't -1 (unknown), bail out if we were asked to.
2621  if (OnlyLowerUnknownSize)
2622  return false;
2623  if (isString) {
2624  uint64_t Len = GetStringLength(CI->getArgOperand(SizeOp));
2625  // If the length is 0 we don't know how long it is and so we can't
2626  // remove the check.
2627  if (Len == 0)
2628  return false;
2629  return ObjSizeCI->getZExtValue() >= Len;
2630  }
2631  if (ConstantInt *SizeCI = dyn_cast<ConstantInt>(CI->getArgOperand(SizeOp)))
2632  return ObjSizeCI->getZExtValue() >= SizeCI->getZExtValue();
2633  }
2634  return false;
2635 }
2636 
2637 Value *FortifiedLibCallSimplifier::optimizeMemCpyChk(CallInst *CI,
2638  IRBuilder<> &B) {
2639  if (isFortifiedCallFoldable(CI, 3, 2, false)) {
2640  B.CreateMemCpy(CI->getArgOperand(0), 1, CI->getArgOperand(1), 1,
2641  CI->getArgOperand(2));
2642  return CI->getArgOperand(0);
2643  }
2644  return nullptr;
2645 }
2646 
2647 Value *FortifiedLibCallSimplifier::optimizeMemMoveChk(CallInst *CI,
2648  IRBuilder<> &B) {
2649  if (isFortifiedCallFoldable(CI, 3, 2, false)) {
2650  B.CreateMemMove(CI->getArgOperand(0), 1, CI->getArgOperand(1), 1,
2651  CI->getArgOperand(2));
2652  return CI->getArgOperand(0);
2653  }
2654  return nullptr;
2655 }
2656 
2657 Value *FortifiedLibCallSimplifier::optimizeMemSetChk(CallInst *CI,
2658  IRBuilder<> &B) {
2659  // TODO: Try foldMallocMemset() here.
2660 
2661  if (isFortifiedCallFoldable(CI, 3, 2, false)) {
2662  Value *Val = B.CreateIntCast(CI->getArgOperand(1), B.getInt8Ty(), false);
2663  B.CreateMemSet(CI->getArgOperand(0), Val, CI->getArgOperand(2), 1);
2664  return CI->getArgOperand(0);
2665  }
2666  return nullptr;
2667 }
2668 
2669 Value *FortifiedLibCallSimplifier::optimizeStrpCpyChk(CallInst *CI,
2670  IRBuilder<> &B,
2671  LibFunc Func) {
2672  Function *Callee = CI->getCalledFunction();
2673  StringRef Name = Callee->getName();
2674  const DataLayout &DL = CI->getModule()->getDataLayout();
2675  Value *Dst = CI->getArgOperand(0), *Src = CI->getArgOperand(1),
2676  *ObjSize = CI->getArgOperand(2);
2677 
2678  // __stpcpy_chk(x,x,...) -> x+strlen(x)
2679  if (Func == LibFunc_stpcpy_chk && !OnlyLowerUnknownSize && Dst == Src) {
2680  Value *StrLen = emitStrLen(Src, B, DL, TLI);
2681  return StrLen ? B.CreateInBoundsGEP(B.getInt8Ty(), Dst, StrLen) : nullptr;
2682  }
2683 
2684  // If a) we don't have any length information, or b) we know this will
2685  // fit then just lower to a plain st[rp]cpy. Otherwise we'll keep our
2686  // st[rp]cpy_chk call which may fail at runtime if the size is too long.
2687  // TODO: It might be nice to get a maximum length out of the possible
2688  // string lengths for varying.
2689  if (isFortifiedCallFoldable(CI, 2, 1, true))
2690  return emitStrCpy(Dst, Src, B, TLI, Name.substr(2, 6));
2691 
2692  if (OnlyLowerUnknownSize)
2693  return nullptr;
2694 
2695  // Maybe we can stil fold __st[rp]cpy_chk to __memcpy_chk.
2696  uint64_t Len = GetStringLength(Src);
2697  if (Len == 0)
2698  return nullptr;
2699 
2700  Type *SizeTTy = DL.getIntPtrType(CI->getContext());
2701  Value *LenV = ConstantInt::get(SizeTTy, Len);
2702  Value *Ret = emitMemCpyChk(Dst, Src, LenV, ObjSize, B, DL, TLI);
2703  // If the function was an __stpcpy_chk, and we were able to fold it into
2704  // a __memcpy_chk, we still need to return the correct end pointer.
2705  if (Ret && Func == LibFunc_stpcpy_chk)
2706  return B.CreateGEP(B.getInt8Ty(), Dst, ConstantInt::get(SizeTTy, Len - 1));
2707  return Ret;
2708 }
2709 
2710 Value *FortifiedLibCallSimplifier::optimizeStrpNCpyChk(CallInst *CI,
2711  IRBuilder<> &B,
2712  LibFunc Func) {
2713  Function *Callee = CI->getCalledFunction();
2714  StringRef Name = Callee->getName();
2715  if (isFortifiedCallFoldable(CI, 3, 2, false)) {
2716  Value *Ret = emitStrNCpy(CI->getArgOperand(0), CI->getArgOperand(1),
2717  CI->getArgOperand(2), B, TLI, Name.substr(2, 7));
2718  return Ret;
2719  }
2720  return nullptr;
2721 }
2722 
2724  // FIXME: We shouldn't be changing "nobuiltin" or TLI unavailable calls here.
2725  // Some clang users checked for _chk libcall availability using:
2726  // __has_builtin(__builtin___memcpy_chk)
2727  // When compiling with -fno-builtin, this is always true.
2728  // When passing -ffreestanding/-mkernel, which both imply -fno-builtin, we
2729  // end up with fortified libcalls, which isn't acceptable in a freestanding
2730  // environment which only provides their non-fortified counterparts.
2731  //
2732  // Until we change clang and/or teach external users to check for availability
2733  // differently, disregard the "nobuiltin" attribute and TLI::has.
2734  //
2735  // PR23093.
2736 
2737  LibFunc Func;
2738  Function *Callee = CI->getCalledFunction();
2739 
2741  CI->getOperandBundlesAsDefs(OpBundles);
2742  IRBuilder<> Builder(CI, /*FPMathTag=*/nullptr, OpBundles);
2743  bool isCallingConvC = isCallingConvCCompatible(CI);
2744 
2745  // First, check that this is a known library functions and that the prototype
2746  // is correct.
2747  if (!TLI->getLibFunc(*Callee, Func))
2748  return nullptr;
2749 
2750  // We never change the calling convention.
2751  if (!ignoreCallingConv(Func) && !isCallingConvC)
2752  return nullptr;
2753 
2754  switch (Func) {
2755  case LibFunc_memcpy_chk:
2756  return optimizeMemCpyChk(CI, Builder);
2757  case LibFunc_memmove_chk:
2758  return optimizeMemMoveChk(CI, Builder);
2759  case LibFunc_memset_chk:
2760  return optimizeMemSetChk(CI, Builder);
2761  case LibFunc_stpcpy_chk:
2762  case LibFunc_strcpy_chk:
2763  return optimizeStrpCpyChk(CI, Builder, Func);
2764  case LibFunc_stpncpy_chk:
2765  case LibFunc_strncpy_chk:
2766  return optimizeStrpNCpyChk(CI, Builder, Func);
2767  default:
2768  break;
2769  }
2770  return nullptr;
2771 }
2772 
2774  const TargetLibraryInfo *TLI, bool OnlyLowerUnknownSize)
2775  : TLI(TLI), OnlyLowerUnknownSize(OnlyLowerUnknownSize) {}
Value * CreateNSWNeg(Value *V, const Twine &Name="")
Definition: IRBuilder.h:1266
Value * CreateInBoundsGEP(Value *Ptr, ArrayRef< Value *> IdxList, const Twine &Name="")
Definition: IRBuilder.h:1393
uint64_t CallInst * C
bool isIntrinsic() const
isIntrinsic - Returns true if the function&#39;s name starts with "llvm.".
Definition: Function.h:185
SymbolTableList< Instruction >::iterator eraseFromParent()
This method unlinks &#39;this&#39; from the containing basic block and deletes it.
Definition: Instruction.cpp:68
BinaryOp_match< cstfp_pred_ty< is_neg_zero_fp >, RHS, Instruction::FSub > m_FNeg(const RHS &X)
Match &#39;fneg X&#39; as &#39;fsub -0.0, X&#39;.
Definition: PatternMatch.h:665
void push_back(const T &Elt)
Definition: SmallVector.h:218
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:238
Function * getCalledFunction() const
Return the function called, or null if this is an indirect function invocation.
class_match< Value > m_Value()
Match an arbitrary value and ignore it.
Definition: PatternMatch.h:72
Value * CreateICmp(CmpInst::Predicate P, Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1846
static GCMetadataPrinterRegistry::Add< ErlangGCPrinter > X("erlang", "erlang-compatible garbage collector")
static Value * foldMallocMemset(CallInst *Memset, IRBuilder<> &B, const TargetLibraryInfo &TLI)
Fold memset[_chk](malloc(n), 0, n) –> calloc(1, n).
void flipAllBits()
Toggle every bit to its opposite value.
Definition: APInt.h:1471
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:228
void setFast(bool B=true)
Definition: Operator.h:231
Value * CreateICmpNE(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1740
Value * CreateIsNotNull(Value *Arg, const Twine &Name="")
Return an i1 value testing if Arg is not null.
Definition: IRBuilder.h:1996
Value * CreateZExtOrTrunc(Value *V, Type *DestTy, const Twine &Name="")
Create a ZExt or Trunc from the integer value V to DestTy.
Definition: IRBuilder.h:1566
void addAttribute(unsigned i, Attribute::AttrKind Kind)
adds the attribute to the list of attributes.
Compute iterated dominance frontiers using a linear time algorithm.
Definition: AllocatorList.h:24
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:1752
Constant * getOrInsertFunction(StringRef Name, FunctionType *T, AttributeList AttributeList)
Look up the specified function in the module symbol table.
Definition: Module.cpp:142
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:63
static Constant * getInfinity(Type *Ty, bool Negative=false)
Definition: Constants.cpp:785
LLVM_NODISCARD size_t rfind(char C, size_t From=npos) const
Search for the last character C in the string.
Definition: StringRef.h:360
LLVM_NODISCARD LLVM_ATTRIBUTE_ALWAYS_INLINE size_t size() const
size - Get the string size.
Definition: StringRef.h:138
Value * CreateICmpSLT(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1768
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
unsigned less than
Definition: InstrTypes.h:910
An efficient, type-erasing, non-owning reference to a callable.
Definition: STLExtras.h:107
LLVMContext & getContext() const
All values hold a context through their type.
Definition: Value.cpp:714
Value * CreateSExt(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1560
F(f)
const fltSemantics & getSemantics() const
Definition: APFloat.h:1155
unsigned getPointerAddressSpace() const
Get the address space of this pointer or pointer vector type.
Definition: DerivedTypes.h:503
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:404
An instruction for reading from memory.
Definition: Instructions.h:168
Hexagon Common GEP
static bool isTrigLibCall(CallInst *CI)
CallingConv::ID getCallingConv() const
getCallingConv/setCallingConv - Get or set the calling convention of this function call...
Value * emitStrNCpy(Value *Dst, Value *Src, Value *Len, IRBuilder<> &B, const TargetLibraryInfo *TLI, StringRef Name="strncpy")
Emit a call to the strncpy function to the builder, for the specified pointer arguments and length...
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 addParamAttr(unsigned ArgNo, Attribute::AttrKind Kind)
Adds the attribute to the indicated argument.
LLVMContext & getContext() const
Get the context in which this basic block lives.
Definition: BasicBlock.cpp:33
Value * emitMalloc(Value *Num, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the malloc function.
LLVM_NODISCARD LLVM_ATTRIBUTE_ALWAYS_INLINE const char * data() const
data - Get a pointer to the start of the string (which may not be null terminated).
Definition: StringRef.h:128
static bool isReportingError(Function *Callee, CallInst *CI, int StreamArg)
static Constant * getNullValue(Type *Ty)
Constructor to create a &#39;0&#39; constant of arbitrary type.
Definition: Constants.cpp:268
iterator begin()
Instruction iterator methods.
Definition: BasicBlock.h:264
ARM_AAPCS_VFP - Same as ARM_AAPCS, but uses hard floating point ABI.
Definition: CallingConv.h:103
IntegerType * getInt32Ty()
Fetch the type representing a 32-bit integer.
Definition: IRBuilder.h:347
static Value * replaceUnaryCall(CallInst *CI, IRBuilder<> &B, Intrinsic::ID IID)
specific_fpval m_SpecificFP(double V)
Match a specific floating point value or vector with all elements equal to the value.
Definition: PatternMatch.h:537
bool match(Val *V, const Pattern &P)
Definition: PatternMatch.h:49
Value * CreateFPExt(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1615
void setBit(unsigned BitPosition)
Set a given bit to 1.
Definition: APInt.h:1397
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:361
FunctionType * getFunctionType() const
bool isNonNegative() const
Determine if this APInt Value is non-negative (>= 0)
Definition: APInt.h:369
FortifiedLibCallSimplifier(const TargetLibraryInfo *TLI, bool OnlyLowerUnknownSize=false)
A Use represents the edge between a Value definition and its users.
Definition: Use.h:56
PointerType * getPointerTo(unsigned AddrSpace=0) const
Return a pointer to the current type.
Definition: Type.cpp:639
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:197
This provides a uniform API for creating instructions and inserting them into a basic block: either a...
Definition: IRBuilder.h:731
LibCallSimplifier(const DataLayout &DL, const TargetLibraryInfo *TLI, OptimizationRemarkEmitter &ORE, function_ref< void(Instruction *, Value *)> Replacer=&replaceAllUsesWithDefault)
Value * CreateAdd(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:962
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:494
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.
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:336
StoreInst * CreateStore(Value *Val, Value *Ptr, bool isVolatile=false)
Definition: IRBuilder.h:1321
static Value * convertStrToNumber(CallInst *CI, StringRef &Str, int64_t Base)
Value * CreateIntToPtr(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1624
Instruction * clone() const
Create a copy of &#39;this&#39; instruction that is identical in all ways except the following: ...
LLVM_NODISCARD LLVM_ATTRIBUTE_ALWAYS_INLINE bool startswith(StringRef Prefix) const
Check if this string starts with the given Prefix.
Definition: StringRef.h:267
Class to represent function types.
Definition: DerivedTypes.h:103
Value * CreateBitCast(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1629
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:1581
apfloat_match m_APFloat(const APFloat *&Res)
Match a ConstantFP or splatted ConstantVector, binding the specified pointer to the contained APFloat...
Definition: PatternMatch.h:181
const ConstantDataArray * Array
ConstantDataArray pointer.
opStatus convert(const fltSemantics &ToSemantics, roundingMode RM, bool *losesInfo)
Definition: APFloat.cpp:4444
ARM_AAPCS - ARM Architecture Procedure Calling Standard calling convention (aka EABI).
Definition: CallingConv.h:100
#define T
BasicBlock * GetInsertBlock() const
Definition: IRBuilder.h:121
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.
bool isNoBuiltin() const
Return true if the call should not be treated as a call to a builtin.
bool has(LibFunc F) const
Tests whether a library function is available.
LLVM_NODISCARD LLVM_ATTRIBUTE_ALWAYS_INLINE bool empty() const
empty - Check if the string is empty.
Definition: StringRef.h:133
SmallString - A SmallString is just a SmallVector with methods and accessors that make it work better...
Definition: SmallString.h:26
unsigned getOpcode() const
Returns a member of one of the enums like Instruction::Add.
Definition: Instruction.h:126
Value * emitStrCpy(Value *Dst, Value *Src, IRBuilder<> &B, const TargetLibraryInfo *TLI, StringRef Name="strcpy")
Emit a call to the strcpy function to the builder, for the specified pointer arguments.
Value * CreateSub(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:979
AttributeList getAttributes() const
Return the attribute list for this Function.
Definition: Function.h:210
void replaceAllUsesWith(Value *V)
Change all uses of this to point to a new Value.
Definition: Value.cpp:439
static const fltSemantics & IEEEdouble() LLVM_READNONE
Definition: APFloat.cpp:123
LLVM_NODISCARD LLVM_ATTRIBUTE_ALWAYS_INLINE StringRef substr(size_t Start, size_t N=npos) const
Return a reference to the substring from [Start, Start + N).
Definition: StringRef.h:598
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:1556
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:301
Value * CreateFCmpOEQ(Value *LHS, Value *RHS, const Twine &Name="", MDNode *FPMathTag=nullptr)
Definition: IRBuilder.h:1776
amdgpu Simplify well known AMD library false Value * Callee
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:1007
void SetInsertPoint(BasicBlock *TheBB)
This specifies that created instructions should be appended to the end of the specified block...
Definition: IRBuilder.h:127
Value * getOperand(unsigned i) const
Definition: User.h:170
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:318
Value * getOperand(unsigned i_nocapture) const
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:1750
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:147
const BasicBlock & getEntryBlock() const
Definition: Function.h:626
LoadInst * CreateLoad(Value *Ptr, const char *Name)
Provided to resolve &#39;CreateLoad(Ptr, "...")&#39; correctly, instead of converting the string to &#39;bool&#39; fo...
Definition: IRBuilder.h:1305
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:742
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:375
OneUse_match< T > m_OneUse(const T &SubPattern)
Definition: PatternMatch.h:63
initializer< Ty > init(const Ty &Val)
Definition: CommandLine.h:410
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:149
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:1195
bool hasFnAttr(Attribute::AttrKind Kind) const
Determine whether this call has the given attribute.
void setNoSignedZeros(bool B=true)
Definition: Operator.h:219
bool isNegative() const
Definition: APFloat.h:1147
LLVM Basic Block Representation.
Definition: BasicBlock.h:59
The instances of the Type class are immutable: once they are created, they are never changed...
Definition: Type.h:46
ConstantInt * getTrue()
Get the constant value for i1 true.
Definition: IRBuilder.h:287
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
This is an important base class in LLVM.
Definition: Constant.h:42
Value * CreateSelect(Value *C, Value *True, Value *False, const Twine &Name="", Instruction *MDFrom=nullptr)
Definition: IRBuilder.h:1901
bool isPointerTy() const
True if this is an instance of PointerType.
Definition: Type.h:224
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:250
ConstantFP - Floating Point Values [float, double].
Definition: Constants.h:264
Value * CreateFCmpOGT(Value *LHS, Value *RHS, const Twine &Name="", MDNode *FPMathTag=nullptr)
Definition: IRBuilder.h:1781
double convertToDouble() const
Definition: APFloat.h:1097
Diagnostic information for applied optimization remarks.
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:1130
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:1025
Value * CreateNeg(Value *V, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1256
void setCalledFunction(Value *Fn)
Set the function called.
bool optForSize() const
Optimize this function for size (-Os) or minimum size (-Oz).
Definition: Function.h:584
This instruction compares its operands according to the predicate given to the constructor.
op_range operands()
Definition: User.h:238
Value * getPointerOperand()
Definition: Instructions.h:274
LLVM_NODISCARD LLVM_ATTRIBUTE_ALWAYS_INLINE 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:184
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:312
Value * castToCStr(Value *V, IRBuilder<> &B)
Return V if it is an i8*, otherwise cast it to i8*.
Represents offset+length into a ConstantDataArray.
Value * CreateExtractElement(Value *Vec, Value *Idx, const Twine &Name="")
Definition: IRBuilder.h:1921
const Function * getFunction() const
Return the function this instruction belongs to.
Definition: Instruction.cpp:60
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...
bool isIntN(unsigned N, int64_t x)
Checks if an signed integer fits into the given (dynamic) bit width.
Definition: MathExtras.h:398
uint64_t NextPowerOf2(uint64_t A)
Returns the next power of two (in 64-bits) that is strictly greater than A.
Definition: MathExtras.h:640
const Constant * stripPointerCasts() const
Definition: Constant.h:173
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.
bool PointerMayBeCaptured(const Value *V, bool ReturnCaptures, bool StoreCaptures)
PointerMayBeCaptured - Return true if this pointer value may be captured by the enclosing function (w...
Value * CreateExtractValue(Value *Agg, ArrayRef< unsigned > Idxs, const Twine &Name="")
Definition: IRBuilder.h:1963
LLVM_NODISCARD char back() const
back - Get the last character in the string.
Definition: StringRef.h:149
static bool isBinary(MachineInstr &MI)
Value * CreateTrunc(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1552
Triple - Helper class for working with autoconf configuration names.
Definition: Triple.h:44
void setCallingConv(CallingConv::ID CC)
LLVM_NODISCARD LLVM_ATTRIBUTE_ALWAYS_INLINE 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:654
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:1374
Value * CreateFCmpOLT(Value *LHS, Value *RHS, const Twine &Name="", MDNode *FPMathTag=nullptr)
Definition: IRBuilder.h:1791
C - The default llvm calling convention, compatible with C.
Definition: CallingConv.h:35
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:240
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:120
static Value * optimizeDoubleFP(CallInst *CI, IRBuilder<> &B, bool isBinary, bool isPrecise=false)
Shrink double -> float functions.
bool isInteger() const
Definition: APFloat.h:1162
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:84
GlobalVariable * CreateGlobalString(StringRef Str, const Twine &Name="", unsigned AddressSpace=0)
Make a new global variable with initializer type i8*.
Definition: IRBuilder.cpp:43
Value * CreateIntCast(Value *V, Type *DestTy, bool isSigned, const Twine &Name="")
Definition: IRBuilder.h:1698
This is a &#39;vector&#39; (really, a variable-sized array), optimized for the case when the array is small...
Definition: SmallVector.h:847
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:2628
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.
ConstantInt * getInt32(uint32_t C)
Get a constant 32-bit value.
Definition: IRBuilder.h:307
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:446
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:621
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:684
Value * CreateConstInBoundsGEP1_64(Value *Ptr, uint64_t Idx0, const Twine &Name="")
Definition: IRBuilder.h:1491
BinaryOp_match< LHS, RHS, Instruction::FMul > m_FMul(const LHS &L, const RHS &R)
Definition: PatternMatch.h:676
Intrinsic::ID getIntrinsicID() const LLVM_READONLY
getIntrinsicID - This method returns the ID number of the specified function, or Intrinsic::not_intri...
Definition: Function.h:180
void setNoNaNs(bool B=true)
Definition: Operator.h:213
FunctionType * getFunctionType() const
Returns the FunctionType for me.
Definition: Function.h:150
const Module * getModule() const
Return the module owning the function this instruction belongs to or nullptr it the function does not...
Definition: Instruction.cpp:56
Class for arbitrary precision integers.
Definition: APInt.h:70
bool ule(const APInt &RHS) const
Unsigned less or equal comparison.
Definition: APInt.h:1217
iterator_range< user_iterator > users()
Definition: Value.h:400
IntegerType * getInt8Ty()
Fetch the type representing an 8-bit integer.
Definition: IRBuilder.h:337
Value * CreateShl(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1051
bool getLibFunc(StringRef funcName, LibFunc &F) const
Searches for a particular function name.
iterator begin() const
Definition: StringRef.h:106
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:292
amdgpu Simplify well known AMD library false Value Value * Arg
bool isStrictFP() const
Determine if the call requires strict floating point semantics.
Value * emitFGetSUnlocked(Value *Str, Value *Size, Value *File, IRBuilder<> &B, const TargetLibraryInfo *TLI)
Emit a call to the fgets_unlocked function.
specific_fpval m_FPOne()
Match a float 1.0 or vector with all elements equal to 1.0.
Definition: PatternMatch.h:540
Predicate getPredicate() const
Return the predicate for this instruction.
Definition: InstrTypes.h:959
Merge contiguous icmps into a memcmp
Definition: MergeICmps.cpp:794
static const size_t npos
Definition: StringRef.h:51
unsigned getIntegerBitWidth() const
Definition: DerivedTypes.h:97
LLVM_NODISCARD bool empty() const
Definition: SmallVector.h:56
unsigned getNumArgOperands() const
Return the number of call arguments.
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.
StringRef getName() const
Return a constant reference to the value&#39;s name.
Definition: Value.cpp:224
const Function * getParent() const
Return the enclosing method, or null if none.
Definition: BasicBlock.h:108
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:395
#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:1213
This class represents a cast unsigned integer to floating point.
bool isZero() const
This is just a convenience method to make client code smaller for a common code.
Definition: Constants.h:193
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...
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:260
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:323
static void insertSinCosCall(IRBuilder<> &B, Function *OrigCallee, Value *Arg, bool UseFloat, Value *&Sin, Value *&Cos, Value *&SinCos)
Type * getFloatTy()
Fetch the type representing a 32-bit floating point value.
Definition: IRBuilder.h:370
Value * CreateFAdd(Value *L, Value *R, const Twine &Name="", MDNode *FPMD=nullptr)
Definition: IRBuilder.h:1161
Value * CreateAnd(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1112
void getOperandBundlesAsDefs(SmallVectorImpl< OperandBundleDef > &Defs) const
Return the list of operand bundles attached to this instruction as a vector of OperandBundleDefs.
Definition: InstrTypes.h:1458
Value * CreateFDiv(Value *L, Value *R, const Twine &Name="", MDNode *FPMD=nullptr)
Definition: IRBuilder.h:1212
bool isDeclaration() const
Return true if the primary definition of this global value is outside of the current translation unit...
Definition: Globals.cpp:201
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:782
Value * emitMemCmp(Value *Ptr1, Value *Ptr2, Value *Len, IRBuilder<> &B, const DataLayout &DL, const TargetLibraryInfo *TLI)
Emit a call to the memcmp function.
This class represents a cast from signed integer to floating point.
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())
ARM_APCS - ARM Procedure Calling Standard calling convention (obsolete, but still used on some target...
Definition: CallingConv.h:96
user_iterator user_begin()
Definition: Value.h:376
Value * getArgOperand(unsigned i) const
getArgOperand/setArgOperand - Return/set the i-th call argument.
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:115
Module * getParent()
Get the module that this global value is contained inside of...
Definition: GlobalValue.h:565
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.
constexpr char Size[]
Key for Kernel::Arg::Metadata::mSize.
static VectorType * get(Type *ElementType, unsigned NumElements)
This static method is the primary way to construct an VectorType.
Definition: Type.cpp:593
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:220
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.
bool isExactlyValue(const APFloat &V) const
We don&#39;t rely on operator== working on double values, as it returns true for things that are clearly ...
Definition: Constants.cpp:801
ConstantInt * getInt8(uint8_t C)
Get a constant 8-bit value.
Definition: IRBuilder.h:297
static Value * optimizeTrigReflections(CallInst *Call, LibFunc Func, IRBuilder<> &B)
bool hasNoNaNs() const
Determine whether the no-NaNs flag is set.
Convenience struct for specifying and reasoning about fast-math flags.
Definition: Operator.h:160
StringRef - Represent a constant reference to a string, i.e.
Definition: StringRef.h:49
bool isDereferenceableAndAlignedPointer(const Value *V, 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:129
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:157
ConstantInt * getInt(const APInt &AI)
Get a constant integer value.
Definition: IRBuilder.h:323
LLVM_NODISCARD LLVM_ATTRIBUTE_ALWAYS_INLINE size_t find(char C, size_t From=0) const
Search for the first character C in the string.
Definition: StringRef.h:298
static bool ignoreCallingConv(LibFunc Func)
This class represents an extension of floating point types.
iterator end() const
Definition: StringRef.h:108
static Constant * getFPExtend(Constant *C, Type *Ty, bool OnlyIfReduced=false)
Definition: Constants.cpp:1668
bool isDoubleTy() const
Return true if this is &#39;double&#39;, a 64-bit IEEE fp type.
Definition: Type.h:150
bool hasUnaryFloatFn(const TargetLibraryInfo *TLI, Type *Ty, LibFunc DoubleFn, LibFunc FloatFn, LibFunc LongDoubleFn)
Check whether the overloaded unary floating point function corresponding to Ty is available...
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)
Determine which bits of V are known to be either zero or one and return them in the KnownZero/KnownOn...
The optimization diagnostic interface.
bool use_empty() const
Definition: Value.h:323
static Value * getPow(Value *InnerChain[33], unsigned Exp, IRBuilder<> &B)
Value * CreateFNeg(Value *V, const Twine &Name="", MDNode *FPMathTag=nullptr)
Definition: IRBuilder.h:1274
bool isStructTy() const
True if this is an instance of StructType.
Definition: Type.h:218
cmpResult compare(const APFloat &RHS) const
Definition: APFloat.h:1102
bool isArrayTy() const
True if this is an instance of ArrayType.
Definition: Type.h:221
A wrapper class for inspecting calls to intrinsic functions.
Definition: IntrinsicInst.h:44
CallInst * CreateCall(Value *Callee, ArrayRef< Value *> Args=None, const Twine &Name="", MDNode *FPMathTag=nullptr)
Definition: IRBuilder.h:1871
user_iterator user_end()
Definition: Value.h:384