AutoUpgrade.cpp revision 06cb8ed00696eb14d1b831921452e50ec0568ea2 
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/Constants.h"
16#include "llvm/Function.h"
17#include "llvm/IRBuilder.h"
18#include "llvm/Instruction.h"
19#include "llvm/IntrinsicInst.h"
20#include "llvm/LLVMContext.h"
21#include "llvm/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 'c': {
56    if (Name.startswith("ctlz.") && F->arg_size() == 1) {
57      F->setName(Name + ".old");
58      NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::ctlz,
59                                        F->arg_begin()->getType());
60      return true;
61    }
62    if (Name.startswith("cttz.") && F->arg_size() == 1) {
63      F->setName(Name + ".old");
64      NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::cttz,
65                                        F->arg_begin()->getType());
66      return true;
67    }
68    break;
69  }
70  case 'x': {
71    if (Name.startswith("x86.sse2.pcmpeq.") ||
72        Name.startswith("x86.sse2.pcmpgt.") ||
73        Name.startswith("x86.avx2.pcmpeq.") ||
74        Name.startswith("x86.avx2.pcmpgt.") ||
75        Name.startswith("x86.avx.vpermil.") ||
76        Name == "x86.avx.movnt.dq.256" ||
77        Name == "x86.avx.movnt.pd.256" ||
78        Name == "x86.avx.movnt.ps.256" ||
79        (Name.startswith("x86.xop.vpcom") && F->arg_size() == 2)) {
80      NewFn = 0;
81      return true;
82    }
83    // SSE4.1 ptest functions may have an old signature.
84    if (Name.startswith("x86.sse41.ptest")) {
85      if (Name == "x86.sse41.ptestc")
86        return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestc, NewFn);
87      if (Name == "x86.sse41.ptestz")
88        return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestz, NewFn);
89      if (Name == "x86.sse41.ptestnzc")
90        return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestnzc, NewFn);
91    }
92    // frcz.ss/sd may need to have an argument dropped
93    if (Name.startswith("x86.xop.vfrcz.ss") && F->arg_size() == 2) {
94      F->setName(Name + ".old");
95      NewFn = Intrinsic::getDeclaration(F->getParent(),
96                                        Intrinsic::x86_xop_vfrcz_ss);
97      return true;
98    }
99    if (Name.startswith("x86.xop.vfrcz.sd") && F->arg_size() == 2) {
100      F->setName(Name + ".old");
101      NewFn = Intrinsic::getDeclaration(F->getParent(),
102                                        Intrinsic::x86_xop_vfrcz_sd);
103      return true;
104    }
105    // Fix the FMA4 intrinsics to remove the 4
106    if (Name.startswith("x86.fma4.")) {
107      F->setName("llvm.x86.fma" + Name.substr(8));
108      NewFn = F;
109      return true;
110    }
111    break;
112  }
113  }
114
115  //  This may not belong here. This function is effectively being overloaded
116  //  to both detect an intrinsic which needs upgrading, and to provide the
117  //  upgraded form of the intrinsic. We should perhaps have two separate
118  //  functions for this.
119  return false;
120}
121
122bool llvm::UpgradeIntrinsicFunction(Function *F, Function *&NewFn) {
123  NewFn = 0;
124  bool Upgraded = UpgradeIntrinsicFunction1(F, NewFn);
125
126  // Upgrade intrinsic attributes.  This does not change the function.
127  if (NewFn)
128    F = NewFn;
129  if (unsigned id = F->getIntrinsicID())
130    F->setAttributes(Intrinsic::getAttributes((Intrinsic::ID)id));
131  return Upgraded;
132}
133
134bool llvm::UpgradeGlobalVariable(GlobalVariable *GV) {
135  // Nothing to do yet.
136  return false;
137}
138
139// UpgradeIntrinsicCall - Upgrade a call to an old intrinsic to be a call the
140// upgraded intrinsic. All argument and return casting must be provided in
141// order to seamlessly integrate with existing context.
142void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
143  Function *F = CI->getCalledFunction();
144  LLVMContext &C = CI->getContext();
145  IRBuilder<> Builder(C);
146  Builder.SetInsertPoint(CI->getParent(), CI);
147
148  assert(F && "Intrinsic call is not direct?");
149
150  if (!NewFn) {
151    // Get the Function's name.
152    StringRef Name = F->getName();
153
154    Value *Rep;
155    // Upgrade packed integer vector compares intrinsics to compare instructions
156    if (Name.startswith("llvm.x86.sse2.pcmpeq.") ||
157        Name.startswith("llvm.x86.avx2.pcmpeq.")) {
158      Rep = Builder.CreateICmpEQ(CI->getArgOperand(0), CI->getArgOperand(1),
159                                 "pcmpeq");
160      // need to sign extend since icmp returns vector of i1
161      Rep = Builder.CreateSExt(Rep, CI->getType(), "");
162    } else if (Name.startswith("llvm.x86.sse2.pcmpgt.") ||
163               Name.startswith("llvm.x86.avx2.pcmpgt.")) {
164      Rep = Builder.CreateICmpSGT(CI->getArgOperand(0), CI->getArgOperand(1),
165                                  "pcmpgt");
166      // need to sign extend since icmp returns vector of i1
167      Rep = Builder.CreateSExt(Rep, CI->getType(), "");
168    } else if (Name == "llvm.x86.avx.movnt.dq.256" ||
169               Name == "llvm.x86.avx.movnt.ps.256" ||
170               Name == "llvm.x86.avx.movnt.pd.256") {
171      IRBuilder<> Builder(C);
172      Builder.SetInsertPoint(CI->getParent(), CI);
173
174      Module *M = F->getParent();
175      SmallVector<Value *, 1> Elts;
176      Elts.push_back(ConstantInt::get(Type::getInt32Ty(C), 1));
177      MDNode *Node = MDNode::get(C, Elts);
178
179      Value *Arg0 = CI->getArgOperand(0);
180      Value *Arg1 = CI->getArgOperand(1);
181
182      // Convert the type of the pointer to a pointer to the stored type.
183      Value *BC = Builder.CreateBitCast(Arg0,
184                                        PointerType::getUnqual(Arg1->getType()),
185                                        "cast");
186      StoreInst *SI = Builder.CreateStore(Arg1, BC);
187      SI->setMetadata(M->getMDKindID("nontemporal"), Node);
188      SI->setAlignment(16);
189
190      // Remove intrinsic.
191      CI->eraseFromParent();
192      return;
193    } else if (Name.startswith("llvm.x86.xop.vpcom")) {
194      Intrinsic::ID intID;
195      if (Name.endswith("ub"))
196        intID = Intrinsic::x86_xop_vpcomub;
197      else if (Name.endswith("uw"))
198        intID = Intrinsic::x86_xop_vpcomuw;
199      else if (Name.endswith("ud"))
200        intID = Intrinsic::x86_xop_vpcomud;
201      else if (Name.endswith("uq"))
202        intID = Intrinsic::x86_xop_vpcomuq;
203      else if (Name.endswith("b"))
204        intID = Intrinsic::x86_xop_vpcomb;
205      else if (Name.endswith("w"))
206        intID = Intrinsic::x86_xop_vpcomw;
207      else if (Name.endswith("d"))
208        intID = Intrinsic::x86_xop_vpcomd;
209      else if (Name.endswith("q"))
210        intID = Intrinsic::x86_xop_vpcomq;
211      else
212        llvm_unreachable("Unknown suffix");
213
214      Name = Name.substr(18); // strip off "llvm.x86.xop.vpcom"
215      unsigned Imm;
216      if (Name.startswith("lt"))
217        Imm = 0;
218      else if (Name.startswith("le"))
219        Imm = 1;
220      else if (Name.startswith("gt"))
221        Imm = 2;
222      else if (Name.startswith("ge"))
223        Imm = 3;
224      else if (Name.startswith("eq"))
225        Imm = 4;
226      else if (Name.startswith("ne"))
227        Imm = 5;
228      else if (Name.startswith("true"))
229        Imm = 6;
230      else if (Name.startswith("false"))
231        Imm = 7;
232      else
233        llvm_unreachable("Unknown condition");
234
235      Function *VPCOM = Intrinsic::getDeclaration(F->getParent(), intID);
236      Rep = Builder.CreateCall3(VPCOM, CI->getArgOperand(0),
237                                CI->getArgOperand(1), Builder.getInt8(Imm));
238    } else {
239      bool PD128 = false, PD256 = false, PS128 = false, PS256 = false;
240      if (Name == "llvm.x86.avx.vpermil.pd.256")
241        PD256 = true;
242      else if (Name == "llvm.x86.avx.vpermil.pd")
243        PD128 = true;
244      else if (Name == "llvm.x86.avx.vpermil.ps.256")
245        PS256 = true;
246      else if (Name == "llvm.x86.avx.vpermil.ps")
247        PS128 = true;
248
249      if (PD256 || PD128 || PS256 || PS128) {
250        Value *Op0 = CI->getArgOperand(0);
251        unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
252        SmallVector<Constant*, 8> Idxs;
253
254        if (PD128)
255          for (unsigned i = 0; i != 2; ++i)
256            Idxs.push_back(Builder.getInt32((Imm >> i) & 0x1));
257        else if (PD256)
258          for (unsigned l = 0; l != 4; l+=2)
259            for (unsigned i = 0; i != 2; ++i)
260              Idxs.push_back(Builder.getInt32(((Imm >> (l+i)) & 0x1) + l));
261        else if (PS128)
262          for (unsigned i = 0; i != 4; ++i)
263            Idxs.push_back(Builder.getInt32((Imm >> (2 * i)) & 0x3));
264        else if (PS256)
265          for (unsigned l = 0; l != 8; l+=4)
266            for (unsigned i = 0; i != 4; ++i)
267              Idxs.push_back(Builder.getInt32(((Imm >> (2 * i)) & 0x3) + l));
268        else
269          llvm_unreachable("Unexpected function");
270
271        Rep = Builder.CreateShuffleVector(Op0, Op0, ConstantVector::get(Idxs));
272      } else {
273        llvm_unreachable("Unknown function for CallInst upgrade.");
274      }
275    }
276
277    CI->replaceAllUsesWith(Rep);
278    CI->eraseFromParent();
279    return;
280  }
281
282  StringRef Name = CI->getName();
283
284  switch (NewFn->getIntrinsicID()) {
285  default:
286    llvm_unreachable("Unknown function for CallInst upgrade.");
287
288  case Intrinsic::ctlz:
289  case Intrinsic::cttz:
290    assert(CI->getNumArgOperands() == 1 &&
291           "Mismatch between function args and call args");
292    CI->setName(Name + ".old");
293    CI->replaceAllUsesWith(Builder.CreateCall2(NewFn, CI->getArgOperand(0),
294                                               Builder.getFalse(), Name));
295    CI->eraseFromParent();
296    return;
297
298  case Intrinsic::x86_xop_vfrcz_ss:
299  case Intrinsic::x86_xop_vfrcz_sd:
300    CI->replaceAllUsesWith(Builder.CreateCall(NewFn, CI->getArgOperand(1),
301                                              Name));
302    CI->eraseFromParent();
303    return;
304
305  case Intrinsic::x86_sse41_ptestc:
306  case Intrinsic::x86_sse41_ptestz:
307  case Intrinsic::x86_sse41_ptestnzc: {
308    // The arguments for these intrinsics used to be v4f32, and changed
309    // to v2i64. This is purely a nop, since those are bitwise intrinsics.
310    // So, the only thing required is a bitcast for both arguments.
311    // First, check the arguments have the old type.
312    Value *Arg0 = CI->getArgOperand(0);
313    if (Arg0->getType() != VectorType::get(Type::getFloatTy(C), 4))
314      return;
315
316    // Old intrinsic, add bitcasts
317    Value *Arg1 = CI->getArgOperand(1);
318
319    Value *BC0 =
320      Builder.CreateBitCast(Arg0,
321                            VectorType::get(Type::getInt64Ty(C), 2),
322                            "cast");
323    Value *BC1 =
324      Builder.CreateBitCast(Arg1,
325                            VectorType::get(Type::getInt64Ty(C), 2),
326                            "cast");
327
328    CallInst* NewCall = Builder.CreateCall2(NewFn, BC0, BC1, Name);
329    CI->replaceAllUsesWith(NewCall);
330    CI->eraseFromParent();
331    return;
332  }
333  }
334}
335
336// This tests each Function to determine if it needs upgrading. When we find
337// one we are interested in, we then upgrade all calls to reflect the new
338// function.
339void llvm::UpgradeCallsToIntrinsic(Function* F) {
340  assert(F && "Illegal attempt to upgrade a non-existent intrinsic.");
341
342  // Upgrade the function and check if it is a totaly new function.
343  Function *NewFn;
344  if (UpgradeIntrinsicFunction(F, NewFn)) {
345    if (NewFn != F) {
346      // Replace all uses to the old function with the new one if necessary.
347      for (Value::use_iterator UI = F->use_begin(), UE = F->use_end();
348           UI != UE; ) {
349        if (CallInst *CI = dyn_cast<CallInst>(*UI++))
350          UpgradeIntrinsicCall(CI, NewFn);
351      }
352      // Remove old function, no longer used, from the module.
353      F->eraseFromParent();
354    }
355  }
356}
357
358