AutoUpgrade.cpp revision 59d3ae6cdc4316ad338cd848251f33a236ccb36c 
1//===-- AutoUpgrade.cpp - Implement auto-upgrade helper functions ---------===//
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 auto-upgrade helper functions
11//
12//===----------------------------------------------------------------------===//
13
14#include "llvm/AutoUpgrade.h"
15#include "llvm/IR/Constants.h"
16#include "llvm/IR/Function.h"
17#include "llvm/IR/IRBuilder.h"
18#include "llvm/IR/Instruction.h"
19#include "llvm/IR/IntrinsicInst.h"
20#include "llvm/IR/LLVMContext.h"
21#include "llvm/IR/Module.h"
22#include "llvm/Support/CFG.h"
23#include "llvm/Support/CallSite.h"
24#include "llvm/Support/ErrorHandling.h"
25#include <cstring>
26using namespace llvm;
27
28// Upgrade the declarations of the SSE4.1 functions whose arguments have
29// changed their type from v4f32 to v2i64.
30static bool UpgradeSSE41Function(Function* F, Intrinsic::ID IID,
31                                 Function *&NewFn) {
32  // Check whether this is an old version of the function, which received
33  // v4f32 arguments.
34  Type *Arg0Type = F->getFunctionType()->getParamType(0);
35  if (Arg0Type != VectorType::get(Type::getFloatTy(F->getContext()), 4))
36    return false;
37
38  // Yes, it's old, replace it with new version.
39  F->setName(F->getName() + ".old");
40  NewFn = Intrinsic::getDeclaration(F->getParent(), IID);
41  return true;
42}
43
44static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
45  assert(F && "Illegal to upgrade a non-existent Function.");
46
47  // Quickly eliminate it, if it's not a candidate.
48  StringRef Name = F->getName();
49  if (Name.size() <= 8 || !Name.startswith("llvm."))
50    return false;
51  Name = Name.substr(5); // Strip off "llvm."
52
53  switch (Name[0]) {
54  default: break;
55  case 'a': {
56    if (Name.startswith("arm.neon.vclz")) {
57      Type* args[2] = {
58        F->arg_begin()->getType(),
59        Type::getInt1Ty(F->getContext())
60      };
61      // Can't use Intrinsic::getDeclaration here as it adds a ".i1" to
62      // the end of the name. Change name from llvm.arm.neon.vclz.* to
63      //  llvm.ctlz.*
64      FunctionType* fType = FunctionType::get(F->getReturnType(), args, false);
65      NewFn = Function::Create(fType, F->getLinkage(),
66                               "llvm.ctlz." + Name.substr(14), F->getParent());
67      return true;
68    }
69    if (Name.startswith("arm.neon.vcnt")) {
70      NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::ctpop,
71                                        F->arg_begin()->getType());
72      return true;
73    }
74    break;
75  }
76  case 'c': {
77    if (Name.startswith("ctlz.") && F->arg_size() == 1) {
78      F->setName(Name + ".old");
79      NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::ctlz,
80                                        F->arg_begin()->getType());
81      return true;
82    }
83    if (Name.startswith("cttz.") && F->arg_size() == 1) {
84      F->setName(Name + ".old");
85      NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::cttz,
86                                        F->arg_begin()->getType());
87      return true;
88    }
89    break;
90  }
91  case 'o':
92    // We only need to change the name to match the mangling including the
93    // address space.
94    if (F->arg_size() == 2 && Name.startswith("objectsize.")) {
95      Type *Tys[2] = { F->getReturnType(), F->arg_begin()->getType() };
96      if (F->getName() != Intrinsic::getName(Intrinsic::objectsize, Tys)) {
97        F->setName(Name + ".old");
98        NewFn = Intrinsic::getDeclaration(F->getParent(),
99                                          Intrinsic::objectsize, Tys);
100        return true;
101      }
102    }
103    break;
104
105  case 'x': {
106    if (Name.startswith("x86.sse2.pcmpeq.") ||
107        Name.startswith("x86.sse2.pcmpgt.") ||
108        Name.startswith("x86.avx2.pcmpeq.") ||
109        Name.startswith("x86.avx2.pcmpgt.") ||
110        Name.startswith("x86.avx.vpermil.") ||
111        Name == "x86.avx.movnt.dq.256" ||
112        Name == "x86.avx.movnt.pd.256" ||
113        Name == "x86.avx.movnt.ps.256" ||
114        Name == "x86.sse42.crc32.64.8" ||
115        (Name.startswith("x86.xop.vpcom") && F->arg_size() == 2)) {
116      NewFn = 0;
117      return true;
118    }
119    // SSE4.1 ptest functions may have an old signature.
120    if (Name.startswith("x86.sse41.ptest")) {
121      if (Name == "x86.sse41.ptestc")
122        return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestc, NewFn);
123      if (Name == "x86.sse41.ptestz")
124        return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestz, NewFn);
125      if (Name == "x86.sse41.ptestnzc")
126        return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestnzc, NewFn);
127    }
128    // frcz.ss/sd may need to have an argument dropped
129    if (Name.startswith("x86.xop.vfrcz.ss") && F->arg_size() == 2) {
130      F->setName(Name + ".old");
131      NewFn = Intrinsic::getDeclaration(F->getParent(),
132                                        Intrinsic::x86_xop_vfrcz_ss);
133      return true;
134    }
135    if (Name.startswith("x86.xop.vfrcz.sd") && F->arg_size() == 2) {
136      F->setName(Name + ".old");
137      NewFn = Intrinsic::getDeclaration(F->getParent(),
138                                        Intrinsic::x86_xop_vfrcz_sd);
139      return true;
140    }
141    // Fix the FMA4 intrinsics to remove the 4
142    if (Name.startswith("x86.fma4.")) {
143      F->setName("llvm.x86.fma" + Name.substr(8));
144      NewFn = F;
145      return true;
146    }
147    break;
148  }
149  }
150
151  //  This may not belong here. This function is effectively being overloaded
152  //  to both detect an intrinsic which needs upgrading, and to provide the
153  //  upgraded form of the intrinsic. We should perhaps have two separate
154  //  functions for this.
155  return false;
156}
157
158bool llvm::UpgradeIntrinsicFunction(Function *F, Function *&NewFn) {
159  NewFn = 0;
160  bool Upgraded = UpgradeIntrinsicFunction1(F, NewFn);
161
162  // Upgrade intrinsic attributes.  This does not change the function.
163  if (NewFn)
164    F = NewFn;
165  if (unsigned id = F->getIntrinsicID())
166    F->setAttributes(Intrinsic::getAttributes(F->getContext(),
167                                              (Intrinsic::ID)id));
168  return Upgraded;
169}
170
171bool llvm::UpgradeGlobalVariable(GlobalVariable *GV) {
172  // Nothing to do yet.
173  return false;
174}
175
176// UpgradeIntrinsicCall - Upgrade a call to an old intrinsic to be a call the
177// upgraded intrinsic. All argument and return casting must be provided in
178// order to seamlessly integrate with existing context.
179void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
180  Function *F = CI->getCalledFunction();
181  LLVMContext &C = CI->getContext();
182  IRBuilder<> Builder(C);
183  Builder.SetInsertPoint(CI->getParent(), CI);
184
185  assert(F && "Intrinsic call is not direct?");
186
187  if (!NewFn) {
188    // Get the Function's name.
189    StringRef Name = F->getName();
190
191    Value *Rep;
192    // Upgrade packed integer vector compares intrinsics to compare instructions
193    if (Name.startswith("llvm.x86.sse2.pcmpeq.") ||
194        Name.startswith("llvm.x86.avx2.pcmpeq.")) {
195      Rep = Builder.CreateICmpEQ(CI->getArgOperand(0), CI->getArgOperand(1),
196                                 "pcmpeq");
197      // need to sign extend since icmp returns vector of i1
198      Rep = Builder.CreateSExt(Rep, CI->getType(), "");
199    } else if (Name.startswith("llvm.x86.sse2.pcmpgt.") ||
200               Name.startswith("llvm.x86.avx2.pcmpgt.")) {
201      Rep = Builder.CreateICmpSGT(CI->getArgOperand(0), CI->getArgOperand(1),
202                                  "pcmpgt");
203      // need to sign extend since icmp returns vector of i1
204      Rep = Builder.CreateSExt(Rep, CI->getType(), "");
205    } else if (Name == "llvm.x86.avx.movnt.dq.256" ||
206               Name == "llvm.x86.avx.movnt.ps.256" ||
207               Name == "llvm.x86.avx.movnt.pd.256") {
208      IRBuilder<> Builder(C);
209      Builder.SetInsertPoint(CI->getParent(), CI);
210
211      Module *M = F->getParent();
212      SmallVector<Value *, 1> Elts;
213      Elts.push_back(ConstantInt::get(Type::getInt32Ty(C), 1));
214      MDNode *Node = MDNode::get(C, Elts);
215
216      Value *Arg0 = CI->getArgOperand(0);
217      Value *Arg1 = CI->getArgOperand(1);
218
219      // Convert the type of the pointer to a pointer to the stored type.
220      Value *BC = Builder.CreateBitCast(Arg0,
221                                        PointerType::getUnqual(Arg1->getType()),
222                                        "cast");
223      StoreInst *SI = Builder.CreateStore(Arg1, BC);
224      SI->setMetadata(M->getMDKindID("nontemporal"), Node);
225      SI->setAlignment(16);
226
227      // Remove intrinsic.
228      CI->eraseFromParent();
229      return;
230    } else if (Name.startswith("llvm.x86.xop.vpcom")) {
231      Intrinsic::ID intID;
232      if (Name.endswith("ub"))
233        intID = Intrinsic::x86_xop_vpcomub;
234      else if (Name.endswith("uw"))
235        intID = Intrinsic::x86_xop_vpcomuw;
236      else if (Name.endswith("ud"))
237        intID = Intrinsic::x86_xop_vpcomud;
238      else if (Name.endswith("uq"))
239        intID = Intrinsic::x86_xop_vpcomuq;
240      else if (Name.endswith("b"))
241        intID = Intrinsic::x86_xop_vpcomb;
242      else if (Name.endswith("w"))
243        intID = Intrinsic::x86_xop_vpcomw;
244      else if (Name.endswith("d"))
245        intID = Intrinsic::x86_xop_vpcomd;
246      else if (Name.endswith("q"))
247        intID = Intrinsic::x86_xop_vpcomq;
248      else
249        llvm_unreachable("Unknown suffix");
250
251      Name = Name.substr(18); // strip off "llvm.x86.xop.vpcom"
252      unsigned Imm;
253      if (Name.startswith("lt"))
254        Imm = 0;
255      else if (Name.startswith("le"))
256        Imm = 1;
257      else if (Name.startswith("gt"))
258        Imm = 2;
259      else if (Name.startswith("ge"))
260        Imm = 3;
261      else if (Name.startswith("eq"))
262        Imm = 4;
263      else if (Name.startswith("ne"))
264        Imm = 5;
265      else if (Name.startswith("true"))
266        Imm = 6;
267      else if (Name.startswith("false"))
268        Imm = 7;
269      else
270        llvm_unreachable("Unknown condition");
271
272      Function *VPCOM = Intrinsic::getDeclaration(F->getParent(), intID);
273      Rep = Builder.CreateCall3(VPCOM, CI->getArgOperand(0),
274                                CI->getArgOperand(1), Builder.getInt8(Imm));
275    } else if (Name == "llvm.x86.sse42.crc32.64.8") {
276      Function *CRC32 = Intrinsic::getDeclaration(F->getParent(),
277                                               Intrinsic::x86_sse42_crc32_32_8);
278      Value *Trunc0 = Builder.CreateTrunc(CI->getArgOperand(0), Type::getInt32Ty(C));
279      Rep = Builder.CreateCall2(CRC32, Trunc0, CI->getArgOperand(1));
280      Rep = Builder.CreateZExt(Rep, CI->getType(), "");
281    } else {
282      bool PD128 = false, PD256 = false, PS128 = false, PS256 = false;
283      if (Name == "llvm.x86.avx.vpermil.pd.256")
284        PD256 = true;
285      else if (Name == "llvm.x86.avx.vpermil.pd")
286        PD128 = true;
287      else if (Name == "llvm.x86.avx.vpermil.ps.256")
288        PS256 = true;
289      else if (Name == "llvm.x86.avx.vpermil.ps")
290        PS128 = true;
291
292      if (PD256 || PD128 || PS256 || PS128) {
293        Value *Op0 = CI->getArgOperand(0);
294        unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
295        SmallVector<Constant*, 8> Idxs;
296
297        if (PD128)
298          for (unsigned i = 0; i != 2; ++i)
299            Idxs.push_back(Builder.getInt32((Imm >> i) & 0x1));
300        else if (PD256)
301          for (unsigned l = 0; l != 4; l+=2)
302            for (unsigned i = 0; i != 2; ++i)
303              Idxs.push_back(Builder.getInt32(((Imm >> (l+i)) & 0x1) + l));
304        else if (PS128)
305          for (unsigned i = 0; i != 4; ++i)
306            Idxs.push_back(Builder.getInt32((Imm >> (2 * i)) & 0x3));
307        else if (PS256)
308          for (unsigned l = 0; l != 8; l+=4)
309            for (unsigned i = 0; i != 4; ++i)
310              Idxs.push_back(Builder.getInt32(((Imm >> (2 * i)) & 0x3) + l));
311        else
312          llvm_unreachable("Unexpected function");
313
314        Rep = Builder.CreateShuffleVector(Op0, Op0, ConstantVector::get(Idxs));
315      } else {
316        llvm_unreachable("Unknown function for CallInst upgrade.");
317      }
318    }
319
320    CI->replaceAllUsesWith(Rep);
321    CI->eraseFromParent();
322    return;
323  }
324
325  std::string Name = CI->getName().str();
326  CI->setName(Name + ".old");
327
328  switch (NewFn->getIntrinsicID()) {
329  default:
330    llvm_unreachable("Unknown function for CallInst upgrade.");
331
332  case Intrinsic::ctlz:
333  case Intrinsic::cttz:
334    assert(CI->getNumArgOperands() == 1 &&
335           "Mismatch between function args and call args");
336    CI->replaceAllUsesWith(Builder.CreateCall2(NewFn, CI->getArgOperand(0),
337                                               Builder.getFalse(), Name));
338    CI->eraseFromParent();
339    return;
340
341  case Intrinsic::objectsize:
342    CI->replaceAllUsesWith(Builder.CreateCall2(NewFn,
343                                               CI->getArgOperand(0),
344                                               CI->getArgOperand(1),
345                                               Name));
346    CI->eraseFromParent();
347    return;
348
349  case Intrinsic::arm_neon_vclz: {
350    // Change name from llvm.arm.neon.vclz.* to llvm.ctlz.*
351    CI->replaceAllUsesWith(Builder.CreateCall2(NewFn, CI->getArgOperand(0),
352                                               Builder.getFalse(),
353                                               "llvm.ctlz." + Name.substr(14)));
354    CI->eraseFromParent();
355    return;
356  }
357  case Intrinsic::ctpop: {
358    CI->replaceAllUsesWith(Builder.CreateCall(NewFn, CI->getArgOperand(0)));
359    CI->eraseFromParent();
360    return;
361  }
362
363  case Intrinsic::x86_xop_vfrcz_ss:
364  case Intrinsic::x86_xop_vfrcz_sd:
365    CI->replaceAllUsesWith(Builder.CreateCall(NewFn, CI->getArgOperand(1),
366                                              Name));
367    CI->eraseFromParent();
368    return;
369
370  case Intrinsic::x86_sse41_ptestc:
371  case Intrinsic::x86_sse41_ptestz:
372  case Intrinsic::x86_sse41_ptestnzc: {
373    // The arguments for these intrinsics used to be v4f32, and changed
374    // to v2i64. This is purely a nop, since those are bitwise intrinsics.
375    // So, the only thing required is a bitcast for both arguments.
376    // First, check the arguments have the old type.
377    Value *Arg0 = CI->getArgOperand(0);
378    if (Arg0->getType() != VectorType::get(Type::getFloatTy(C), 4))
379      return;
380
381    // Old intrinsic, add bitcasts
382    Value *Arg1 = CI->getArgOperand(1);
383
384    Value *BC0 =
385      Builder.CreateBitCast(Arg0,
386                            VectorType::get(Type::getInt64Ty(C), 2),
387                            "cast");
388    Value *BC1 =
389      Builder.CreateBitCast(Arg1,
390                            VectorType::get(Type::getInt64Ty(C), 2),
391                            "cast");
392
393    CallInst* NewCall = Builder.CreateCall2(NewFn, BC0, BC1, Name);
394    CI->replaceAllUsesWith(NewCall);
395    CI->eraseFromParent();
396    return;
397  }
398  }
399}
400
401// This tests each Function to determine if it needs upgrading. When we find
402// one we are interested in, we then upgrade all calls to reflect the new
403// function.
404void llvm::UpgradeCallsToIntrinsic(Function* F) {
405  assert(F && "Illegal attempt to upgrade a non-existent intrinsic.");
406
407  // Upgrade the function and check if it is a totaly new function.
408  Function *NewFn;
409  if (UpgradeIntrinsicFunction(F, NewFn)) {
410    if (NewFn != F) {
411      // Replace all uses to the old function with the new one if necessary.
412      for (Value::use_iterator UI = F->use_begin(), UE = F->use_end();
413           UI != UE; ) {
414        if (CallInst *CI = dyn_cast<CallInst>(*UI++))
415          UpgradeIntrinsicCall(CI, NewFn);
416      }
417      // Remove old function, no longer used, from the module.
418      F->eraseFromParent();
419    }
420  }
421}
422
423void llvm::UpgradeInstWithTBAATag(Instruction *I) {
424  MDNode *MD = I->getMetadata(LLVMContext::MD_tbaa);
425  assert(MD && "UpgradeInstWithTBAATag should have a TBAA tag");
426  // Check if the tag uses struct-path aware TBAA format.
427  if (isa<MDNode>(MD->getOperand(0)) && MD->getNumOperands() >= 3)
428    return;
429
430  if (MD->getNumOperands() == 3) {
431    Value *Elts[] = {
432      MD->getOperand(0),
433      MD->getOperand(1)
434    };
435    MDNode *ScalarType = MDNode::get(I->getContext(), Elts);
436    // Create a MDNode <ScalarType, ScalarType, offset 0, const>
437    Value *Elts2[] = {
438      ScalarType, ScalarType,
439      Constant::getNullValue(Type::getInt64Ty(I->getContext())),
440      MD->getOperand(2)
441    };
442    I->setMetadata(LLVMContext::MD_tbaa, MDNode::get(I->getContext(), Elts2));
443  } else {
444    // Create a MDNode <MD, MD, offset 0>
445    Value *Elts[] = {MD, MD,
446      Constant::getNullValue(Type::getInt64Ty(I->getContext()))};
447    I->setMetadata(LLVMContext::MD_tbaa, MDNode::get(I->getContext(), Elts));
448  }
449}
450
451Instruction *llvm::UpgradeBitCastInst(unsigned Opc, Value *V, Type *DestTy,
452                                      Instruction *&Temp) {
453  if (Opc != Instruction::BitCast)
454    return 0;
455
456  Temp = 0;
457  Type *SrcTy = V->getType();
458  if (SrcTy->isPtrOrPtrVectorTy() && DestTy->isPtrOrPtrVectorTy() &&
459      SrcTy->getPointerAddressSpace() != DestTy->getPointerAddressSpace()) {
460    LLVMContext &Context = V->getContext();
461
462    // We have no information about target data layout, so we assume that
463    // the maximum pointer size is 64bit.
464    Type *MidTy = Type::getInt64Ty(Context);
465    Temp = CastInst::Create(Instruction::PtrToInt, V, MidTy);
466
467    return CastInst::Create(Instruction::IntToPtr, Temp, DestTy);
468  }
469
470  return 0;
471}
472
473Value *llvm::UpgradeBitCastExpr(unsigned Opc, Constant *C, Type *DestTy) {
474  if (Opc != Instruction::BitCast)
475    return 0;
476
477  Type *SrcTy = C->getType();
478  if (SrcTy->isPtrOrPtrVectorTy() && DestTy->isPtrOrPtrVectorTy() &&
479      SrcTy->getPointerAddressSpace() != DestTy->getPointerAddressSpace()) {
480    LLVMContext &Context = C->getContext();
481
482    // We have no information about target data layout, so we assume that
483    // the maximum pointer size is 64bit.
484    Type *MidTy = Type::getInt64Ty(Context);
485
486    return ConstantExpr::getIntToPtr(ConstantExpr::getPtrToInt(C, MidTy),
487                                     DestTy);
488  }
489
490  return 0;
491}
492