AutoUpgrade.cpp revision c4a8c07f6489c0081207f722ce0a4502614aef69 
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.startswith("x86.xop.vpcom") && F->arg_size() == 2)) {
115      NewFn = 0;
116      return true;
117    }
118    // SSE4.1 ptest functions may have an old signature.
119    if (Name.startswith("x86.sse41.ptest")) {
120      if (Name == "x86.sse41.ptestc")
121        return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestc, NewFn);
122      if (Name == "x86.sse41.ptestz")
123        return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestz, NewFn);
124      if (Name == "x86.sse41.ptestnzc")
125        return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestnzc, NewFn);
126    }
127    // frcz.ss/sd may need to have an argument dropped
128    if (Name.startswith("x86.xop.vfrcz.ss") && F->arg_size() == 2) {
129      F->setName(Name + ".old");
130      NewFn = Intrinsic::getDeclaration(F->getParent(),
131                                        Intrinsic::x86_xop_vfrcz_ss);
132      return true;
133    }
134    if (Name.startswith("x86.xop.vfrcz.sd") && F->arg_size() == 2) {
135      F->setName(Name + ".old");
136      NewFn = Intrinsic::getDeclaration(F->getParent(),
137                                        Intrinsic::x86_xop_vfrcz_sd);
138      return true;
139    }
140    // Fix the FMA4 intrinsics to remove the 4
141    if (Name.startswith("x86.fma4.")) {
142      F->setName("llvm.x86.fma" + Name.substr(8));
143      NewFn = F;
144      return true;
145    }
146    break;
147  }
148  }
149
150  //  This may not belong here. This function is effectively being overloaded
151  //  to both detect an intrinsic which needs upgrading, and to provide the
152  //  upgraded form of the intrinsic. We should perhaps have two separate
153  //  functions for this.
154  return false;
155}
156
157bool llvm::UpgradeIntrinsicFunction(Function *F, Function *&NewFn) {
158  NewFn = 0;
159  bool Upgraded = UpgradeIntrinsicFunction1(F, NewFn);
160
161  // Upgrade intrinsic attributes.  This does not change the function.
162  if (NewFn)
163    F = NewFn;
164  if (unsigned id = F->getIntrinsicID())
165    F->setAttributes(Intrinsic::getAttributes(F->getContext(),
166                                              (Intrinsic::ID)id));
167  return Upgraded;
168}
169
170bool llvm::UpgradeGlobalVariable(GlobalVariable *GV) {
171  // Nothing to do yet.
172  return false;
173}
174
175// UpgradeIntrinsicCall - Upgrade a call to an old intrinsic to be a call the
176// upgraded intrinsic. All argument and return casting must be provided in
177// order to seamlessly integrate with existing context.
178void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
179  Function *F = CI->getCalledFunction();
180  LLVMContext &C = CI->getContext();
181  IRBuilder<> Builder(C);
182  Builder.SetInsertPoint(CI->getParent(), CI);
183
184  assert(F && "Intrinsic call is not direct?");
185
186  if (!NewFn) {
187    // Get the Function's name.
188    StringRef Name = F->getName();
189
190    Value *Rep;
191    // Upgrade packed integer vector compares intrinsics to compare instructions
192    if (Name.startswith("llvm.x86.sse2.pcmpeq.") ||
193        Name.startswith("llvm.x86.avx2.pcmpeq.")) {
194      Rep = Builder.CreateICmpEQ(CI->getArgOperand(0), CI->getArgOperand(1),
195                                 "pcmpeq");
196      // need to sign extend since icmp returns vector of i1
197      Rep = Builder.CreateSExt(Rep, CI->getType(), "");
198    } else if (Name.startswith("llvm.x86.sse2.pcmpgt.") ||
199               Name.startswith("llvm.x86.avx2.pcmpgt.")) {
200      Rep = Builder.CreateICmpSGT(CI->getArgOperand(0), CI->getArgOperand(1),
201                                  "pcmpgt");
202      // need to sign extend since icmp returns vector of i1
203      Rep = Builder.CreateSExt(Rep, CI->getType(), "");
204    } else if (Name == "llvm.x86.avx.movnt.dq.256" ||
205               Name == "llvm.x86.avx.movnt.ps.256" ||
206               Name == "llvm.x86.avx.movnt.pd.256") {
207      IRBuilder<> Builder(C);
208      Builder.SetInsertPoint(CI->getParent(), CI);
209
210      Module *M = F->getParent();
211      SmallVector<Value *, 1> Elts;
212      Elts.push_back(ConstantInt::get(Type::getInt32Ty(C), 1));
213      MDNode *Node = MDNode::get(C, Elts);
214
215      Value *Arg0 = CI->getArgOperand(0);
216      Value *Arg1 = CI->getArgOperand(1);
217
218      // Convert the type of the pointer to a pointer to the stored type.
219      Value *BC = Builder.CreateBitCast(Arg0,
220                                        PointerType::getUnqual(Arg1->getType()),
221                                        "cast");
222      StoreInst *SI = Builder.CreateStore(Arg1, BC);
223      SI->setMetadata(M->getMDKindID("nontemporal"), Node);
224      SI->setAlignment(16);
225
226      // Remove intrinsic.
227      CI->eraseFromParent();
228      return;
229    } else if (Name.startswith("llvm.x86.xop.vpcom")) {
230      Intrinsic::ID intID;
231      if (Name.endswith("ub"))
232        intID = Intrinsic::x86_xop_vpcomub;
233      else if (Name.endswith("uw"))
234        intID = Intrinsic::x86_xop_vpcomuw;
235      else if (Name.endswith("ud"))
236        intID = Intrinsic::x86_xop_vpcomud;
237      else if (Name.endswith("uq"))
238        intID = Intrinsic::x86_xop_vpcomuq;
239      else if (Name.endswith("b"))
240        intID = Intrinsic::x86_xop_vpcomb;
241      else if (Name.endswith("w"))
242        intID = Intrinsic::x86_xop_vpcomw;
243      else if (Name.endswith("d"))
244        intID = Intrinsic::x86_xop_vpcomd;
245      else if (Name.endswith("q"))
246        intID = Intrinsic::x86_xop_vpcomq;
247      else
248        llvm_unreachable("Unknown suffix");
249
250      Name = Name.substr(18); // strip off "llvm.x86.xop.vpcom"
251      unsigned Imm;
252      if (Name.startswith("lt"))
253        Imm = 0;
254      else if (Name.startswith("le"))
255        Imm = 1;
256      else if (Name.startswith("gt"))
257        Imm = 2;
258      else if (Name.startswith("ge"))
259        Imm = 3;
260      else if (Name.startswith("eq"))
261        Imm = 4;
262      else if (Name.startswith("ne"))
263        Imm = 5;
264      else if (Name.startswith("true"))
265        Imm = 6;
266      else if (Name.startswith("false"))
267        Imm = 7;
268      else
269        llvm_unreachable("Unknown condition");
270
271      Function *VPCOM = Intrinsic::getDeclaration(F->getParent(), intID);
272      Rep = Builder.CreateCall3(VPCOM, CI->getArgOperand(0),
273                                CI->getArgOperand(1), Builder.getInt8(Imm));
274    } else {
275      bool PD128 = false, PD256 = false, PS128 = false, PS256 = false;
276      if (Name == "llvm.x86.avx.vpermil.pd.256")
277        PD256 = true;
278      else if (Name == "llvm.x86.avx.vpermil.pd")
279        PD128 = true;
280      else if (Name == "llvm.x86.avx.vpermil.ps.256")
281        PS256 = true;
282      else if (Name == "llvm.x86.avx.vpermil.ps")
283        PS128 = true;
284
285      if (PD256 || PD128 || PS256 || PS128) {
286        Value *Op0 = CI->getArgOperand(0);
287        unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
288        SmallVector<Constant*, 8> Idxs;
289
290        if (PD128)
291          for (unsigned i = 0; i != 2; ++i)
292            Idxs.push_back(Builder.getInt32((Imm >> i) & 0x1));
293        else if (PD256)
294          for (unsigned l = 0; l != 4; l+=2)
295            for (unsigned i = 0; i != 2; ++i)
296              Idxs.push_back(Builder.getInt32(((Imm >> (l+i)) & 0x1) + l));
297        else if (PS128)
298          for (unsigned i = 0; i != 4; ++i)
299            Idxs.push_back(Builder.getInt32((Imm >> (2 * i)) & 0x3));
300        else if (PS256)
301          for (unsigned l = 0; l != 8; l+=4)
302            for (unsigned i = 0; i != 4; ++i)
303              Idxs.push_back(Builder.getInt32(((Imm >> (2 * i)) & 0x3) + l));
304        else
305          llvm_unreachable("Unexpected function");
306
307        Rep = Builder.CreateShuffleVector(Op0, Op0, ConstantVector::get(Idxs));
308      } else {
309        llvm_unreachable("Unknown function for CallInst upgrade.");
310      }
311    }
312
313    CI->replaceAllUsesWith(Rep);
314    CI->eraseFromParent();
315    return;
316  }
317
318  std::string Name = CI->getName().str();
319  CI->setName(Name + ".old");
320
321  switch (NewFn->getIntrinsicID()) {
322  default:
323    llvm_unreachable("Unknown function for CallInst upgrade.");
324
325  case Intrinsic::ctlz:
326  case Intrinsic::cttz:
327    assert(CI->getNumArgOperands() == 1 &&
328           "Mismatch between function args and call args");
329    CI->replaceAllUsesWith(Builder.CreateCall2(NewFn, CI->getArgOperand(0),
330                                               Builder.getFalse(), Name));
331    CI->eraseFromParent();
332    return;
333
334  case Intrinsic::objectsize:
335    CI->replaceAllUsesWith(Builder.CreateCall2(NewFn,
336                                               CI->getArgOperand(0),
337                                               CI->getArgOperand(1),
338                                               Name));
339    CI->eraseFromParent();
340    return;
341
342  case Intrinsic::arm_neon_vclz: {
343    // Change name from llvm.arm.neon.vclz.* to llvm.ctlz.*
344    CI->replaceAllUsesWith(Builder.CreateCall2(NewFn, CI->getArgOperand(0),
345                                               Builder.getFalse(),
346                                               "llvm.ctlz." + Name.substr(14)));
347    CI->eraseFromParent();
348    return;
349  }
350  case Intrinsic::ctpop: {
351    CI->replaceAllUsesWith(Builder.CreateCall(NewFn, CI->getArgOperand(0)));
352    CI->eraseFromParent();
353    return;
354  }
355
356  case Intrinsic::x86_xop_vfrcz_ss:
357  case Intrinsic::x86_xop_vfrcz_sd:
358    CI->replaceAllUsesWith(Builder.CreateCall(NewFn, CI->getArgOperand(1),
359                                              Name));
360    CI->eraseFromParent();
361    return;
362
363  case Intrinsic::x86_sse41_ptestc:
364  case Intrinsic::x86_sse41_ptestz:
365  case Intrinsic::x86_sse41_ptestnzc: {
366    // The arguments for these intrinsics used to be v4f32, and changed
367    // to v2i64. This is purely a nop, since those are bitwise intrinsics.
368    // So, the only thing required is a bitcast for both arguments.
369    // First, check the arguments have the old type.
370    Value *Arg0 = CI->getArgOperand(0);
371    if (Arg0->getType() != VectorType::get(Type::getFloatTy(C), 4))
372      return;
373
374    // Old intrinsic, add bitcasts
375    Value *Arg1 = CI->getArgOperand(1);
376
377    Value *BC0 =
378      Builder.CreateBitCast(Arg0,
379                            VectorType::get(Type::getInt64Ty(C), 2),
380                            "cast");
381    Value *BC1 =
382      Builder.CreateBitCast(Arg1,
383                            VectorType::get(Type::getInt64Ty(C), 2),
384                            "cast");
385
386    CallInst* NewCall = Builder.CreateCall2(NewFn, BC0, BC1, Name);
387    CI->replaceAllUsesWith(NewCall);
388    CI->eraseFromParent();
389    return;
390  }
391  }
392}
393
394// This tests each Function to determine if it needs upgrading. When we find
395// one we are interested in, we then upgrade all calls to reflect the new
396// function.
397void llvm::UpgradeCallsToIntrinsic(Function* F) {
398  assert(F && "Illegal attempt to upgrade a non-existent intrinsic.");
399
400  // Upgrade the function and check if it is a totaly new function.
401  Function *NewFn;
402  if (UpgradeIntrinsicFunction(F, NewFn)) {
403    if (NewFn != F) {
404      // Replace all uses to the old function with the new one if necessary.
405      for (Value::use_iterator UI = F->use_begin(), UE = F->use_end();
406           UI != UE; ) {
407        if (CallInst *CI = dyn_cast<CallInst>(*UI++))
408          UpgradeIntrinsicCall(CI, NewFn);
409      }
410      // Remove old function, no longer used, from the module.
411      F->eraseFromParent();
412    }
413  }
414}
415
416void llvm::UpgradeInstWithTBAATag(Instruction *I) {
417  MDNode *MD = I->getMetadata(LLVMContext::MD_tbaa);
418  assert(MD && "UpgradeInstWithTBAATag should have a TBAA tag");
419  // Check if the tag uses struct-path aware TBAA format.
420  if (isa<MDNode>(MD->getOperand(0)) && MD->getNumOperands() >= 3)
421    return;
422
423  if (MD->getNumOperands() == 3) {
424    Value *Elts[] = {
425      MD->getOperand(0),
426      MD->getOperand(1)
427    };
428    MDNode *ScalarType = MDNode::get(I->getContext(), Elts);
429    // Create a MDNode <ScalarType, ScalarType, offset 0, const>
430    Value *Elts2[] = {
431      ScalarType, ScalarType,
432      Constant::getNullValue(Type::getInt64Ty(I->getContext())),
433      MD->getOperand(2)
434    };
435    I->setMetadata(LLVMContext::MD_tbaa, MDNode::get(I->getContext(), Elts2));
436  } else {
437    // Create a MDNode <MD, MD, offset 0>
438    Value *Elts[] = {MD, MD,
439      Constant::getNullValue(Type::getInt64Ty(I->getContext()))};
440    I->setMetadata(LLVMContext::MD_tbaa, MDNode::get(I->getContext(), Elts));
441  }
442}
443