1 //===-- LibCallsShrinkWrap.cpp ----------------------------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This pass shrink-wraps a call to function if the result is not used.
11 // The call can set errno but is otherwise side effect free. For example:
16 // Even if the result of library call is not being used, the compiler cannot
17 // safely delete the call because the function can set errno on error
19 // Note in many functions, the error condition solely depends on the incoming
20 // parameter. In this optimization, we can generate the condition can lead to
21 // the errno to shrink-wrap the call. Since the chances of hitting the error
22 // condition is low, the runtime call is effectively eliminated.
24 // These partially dead calls are usually results of C++ abstraction penalty
25 // exposed by inlining.
27 //===----------------------------------------------------------------------===//
29 #include "llvm/Transforms/Utils/LibCallsShrinkWrap.h"
30 #include "llvm/ADT/SmallVector.h"
31 #include "llvm/ADT/Statistic.h"
32 #include "llvm/Analysis/GlobalsModRef.h"
33 #include "llvm/Analysis/TargetLibraryInfo.h"
34 #include "llvm/IR/CFG.h"
35 #include "llvm/IR/Constants.h"
36 #include "llvm/IR/Dominators.h"
37 #include "llvm/IR/Function.h"
38 #include "llvm/IR/IRBuilder.h"
39 #include "llvm/IR/InstVisitor.h"
40 #include "llvm/IR/Instructions.h"
41 #include "llvm/IR/LLVMContext.h"
42 #include "llvm/IR/MDBuilder.h"
43 #include "llvm/Pass.h"
44 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
47 #define DEBUG_TYPE "libcalls-shrinkwrap"
49 STATISTIC(NumWrappedOneCond
, "Number of One-Condition Wrappers Inserted");
50 STATISTIC(NumWrappedTwoCond
, "Number of Two-Condition Wrappers Inserted");
53 class LibCallsShrinkWrapLegacyPass
: public FunctionPass
{
55 static char ID
; // Pass identification, replacement for typeid
56 explicit LibCallsShrinkWrapLegacyPass() : FunctionPass(ID
) {
57 initializeLibCallsShrinkWrapLegacyPassPass(
58 *PassRegistry::getPassRegistry());
60 void getAnalysisUsage(AnalysisUsage
&AU
) const override
;
61 bool runOnFunction(Function
&F
) override
;
65 char LibCallsShrinkWrapLegacyPass::ID
= 0;
66 INITIALIZE_PASS_BEGIN(LibCallsShrinkWrapLegacyPass
, "libcalls-shrinkwrap",
67 "Conditionally eliminate dead library calls", false,
69 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass
)
70 INITIALIZE_PASS_END(LibCallsShrinkWrapLegacyPass
, "libcalls-shrinkwrap",
71 "Conditionally eliminate dead library calls", false, false)
74 class LibCallsShrinkWrap
: public InstVisitor
<LibCallsShrinkWrap
> {
76 LibCallsShrinkWrap(const TargetLibraryInfo
&TLI
, DominatorTree
*DT
)
78 void visitCallInst(CallInst
&CI
) { checkCandidate(CI
); }
81 for (auto &CI
: WorkList
) {
82 LLVM_DEBUG(dbgs() << "CDCE calls: " << CI
->getCalledFunction()->getName()
86 LLVM_DEBUG(dbgs() << "Transformed\n");
93 bool perform(CallInst
*CI
);
94 void checkCandidate(CallInst
&CI
);
95 void shrinkWrapCI(CallInst
*CI
, Value
*Cond
);
96 bool performCallDomainErrorOnly(CallInst
*CI
, const LibFunc
&Func
);
97 bool performCallErrors(CallInst
*CI
, const LibFunc
&Func
);
98 bool performCallRangeErrorOnly(CallInst
*CI
, const LibFunc
&Func
);
99 Value
*generateOneRangeCond(CallInst
*CI
, const LibFunc
&Func
);
100 Value
*generateTwoRangeCond(CallInst
*CI
, const LibFunc
&Func
);
101 Value
*generateCondForPow(CallInst
*CI
, const LibFunc
&Func
);
103 // Create an OR of two conditions.
104 Value
*createOrCond(CallInst
*CI
, CmpInst::Predicate Cmp
, float Val
,
105 CmpInst::Predicate Cmp2
, float Val2
) {
106 IRBuilder
<> BBBuilder(CI
);
107 Value
*Arg
= CI
->getArgOperand(0);
108 auto Cond2
= createCond(BBBuilder
, Arg
, Cmp2
, Val2
);
109 auto Cond1
= createCond(BBBuilder
, Arg
, Cmp
, Val
);
110 return BBBuilder
.CreateOr(Cond1
, Cond2
);
113 // Create a single condition using IRBuilder.
114 Value
*createCond(IRBuilder
<> &BBBuilder
, Value
*Arg
, CmpInst::Predicate Cmp
,
116 Constant
*V
= ConstantFP::get(BBBuilder
.getContext(), APFloat(Val
));
117 if (!Arg
->getType()->isFloatTy())
118 V
= ConstantExpr::getFPExtend(V
, Arg
->getType());
119 return BBBuilder
.CreateFCmp(Cmp
, Arg
, V
);
122 // Create a single condition.
123 Value
*createCond(CallInst
*CI
, CmpInst::Predicate Cmp
, float Val
) {
124 IRBuilder
<> BBBuilder(CI
);
125 Value
*Arg
= CI
->getArgOperand(0);
126 return createCond(BBBuilder
, Arg
, Cmp
, Val
);
129 const TargetLibraryInfo
&TLI
;
131 SmallVector
<CallInst
*, 16> WorkList
;
133 } // end anonymous namespace
135 // Perform the transformation to calls with errno set by domain error.
136 bool LibCallsShrinkWrap::performCallDomainErrorOnly(CallInst
*CI
,
137 const LibFunc
&Func
) {
138 Value
*Cond
= nullptr;
141 case LibFunc_acos
: // DomainError: (x < -1 || x > 1)
142 case LibFunc_acosf
: // Same as acos
143 case LibFunc_acosl
: // Same as acos
144 case LibFunc_asin
: // DomainError: (x < -1 || x > 1)
145 case LibFunc_asinf
: // Same as asin
146 case LibFunc_asinl
: // Same as asin
149 Cond
= createOrCond(CI
, CmpInst::FCMP_OLT
, -1.0f
, CmpInst::FCMP_OGT
, 1.0f
);
152 case LibFunc_cos
: // DomainError: (x == +inf || x == -inf)
153 case LibFunc_cosf
: // Same as cos
154 case LibFunc_cosl
: // Same as cos
155 case LibFunc_sin
: // DomainError: (x == +inf || x == -inf)
156 case LibFunc_sinf
: // Same as sin
157 case LibFunc_sinl
: // Same as sin
160 Cond
= createOrCond(CI
, CmpInst::FCMP_OEQ
, INFINITY
, CmpInst::FCMP_OEQ
,
164 case LibFunc_acosh
: // DomainError: (x < 1)
165 case LibFunc_acoshf
: // Same as acosh
166 case LibFunc_acoshl
: // Same as acosh
169 Cond
= createCond(CI
, CmpInst::FCMP_OLT
, 1.0f
);
172 case LibFunc_sqrt
: // DomainError: (x < 0)
173 case LibFunc_sqrtf
: // Same as sqrt
174 case LibFunc_sqrtl
: // Same as sqrt
177 Cond
= createCond(CI
, CmpInst::FCMP_OLT
, 0.0f
);
183 shrinkWrapCI(CI
, Cond
);
187 // Perform the transformation to calls with errno set by range error.
188 bool LibCallsShrinkWrap::performCallRangeErrorOnly(CallInst
*CI
,
189 const LibFunc
&Func
) {
190 Value
*Cond
= nullptr;
207 case LibFunc_sinhl
: {
208 Cond
= generateTwoRangeCond(CI
, Func
);
211 case LibFunc_expm1
: // RangeError: (709, inf)
212 case LibFunc_expm1f
: // RangeError: (88, inf)
213 case LibFunc_expm1l
: // RangeError: (11356, inf)
215 Cond
= generateOneRangeCond(CI
, Func
);
221 shrinkWrapCI(CI
, Cond
);
225 // Perform the transformation to calls with errno set by combination of errors.
226 bool LibCallsShrinkWrap::performCallErrors(CallInst
*CI
,
227 const LibFunc
&Func
) {
228 Value
*Cond
= nullptr;
231 case LibFunc_atanh
: // DomainError: (x < -1 || x > 1)
232 // PoleError: (x == -1 || x == 1)
233 // Overall Cond: (x <= -1 || x >= 1)
234 case LibFunc_atanhf
: // Same as atanh
235 case LibFunc_atanhl
: // Same as atanh
238 Cond
= createOrCond(CI
, CmpInst::FCMP_OLE
, -1.0f
, CmpInst::FCMP_OGE
, 1.0f
);
241 case LibFunc_log
: // DomainError: (x < 0)
242 // PoleError: (x == 0)
243 // Overall Cond: (x <= 0)
244 case LibFunc_logf
: // Same as log
245 case LibFunc_logl
: // Same as log
246 case LibFunc_log10
: // Same as log
247 case LibFunc_log10f
: // Same as log
248 case LibFunc_log10l
: // Same as log
249 case LibFunc_log2
: // Same as log
250 case LibFunc_log2f
: // Same as log
251 case LibFunc_log2l
: // Same as log
252 case LibFunc_logb
: // Same as log
253 case LibFunc_logbf
: // Same as log
254 case LibFunc_logbl
: // Same as log
257 Cond
= createCond(CI
, CmpInst::FCMP_OLE
, 0.0f
);
260 case LibFunc_log1p
: // DomainError: (x < -1)
261 // PoleError: (x == -1)
262 // Overall Cond: (x <= -1)
263 case LibFunc_log1pf
: // Same as log1p
264 case LibFunc_log1pl
: // Same as log1p
267 Cond
= createCond(CI
, CmpInst::FCMP_OLE
, -1.0f
);
270 case LibFunc_pow
: // DomainError: x < 0 and y is noninteger
271 // PoleError: x == 0 and y < 0
272 // RangeError: overflow or underflow
275 Cond
= generateCondForPow(CI
, Func
);
283 assert(Cond
&& "performCallErrors should not see an empty condition");
284 shrinkWrapCI(CI
, Cond
);
288 // Checks if CI is a candidate for shrinkwrapping and put it into work list if
290 void LibCallsShrinkWrap::checkCandidate(CallInst
&CI
) {
291 if (CI
.isNoBuiltin())
293 // A possible improvement is to handle the calls with the return value being
294 // used. If there is API for fast libcall implementation without setting
295 // errno, we can use the same framework to direct/wrap the call to the fast
296 // API in the error free path, and leave the original call in the slow path.
301 Function
*Callee
= CI
.getCalledFunction();
304 if (!TLI
.getLibFunc(*Callee
, Func
) || !TLI
.has(Func
))
307 if (CI
.getNumArgOperands() == 0)
309 // TODO: Handle long double in other formats.
310 Type
*ArgType
= CI
.getArgOperand(0)->getType();
311 if (!(ArgType
->isFloatTy() || ArgType
->isDoubleTy() ||
312 ArgType
->isX86_FP80Ty()))
315 WorkList
.push_back(&CI
);
318 // Generate the upper bound condition for RangeError.
319 Value
*LibCallsShrinkWrap::generateOneRangeCond(CallInst
*CI
,
320 const LibFunc
&Func
) {
323 case LibFunc_expm1
: // RangeError: (709, inf)
326 case LibFunc_expm1f
: // RangeError: (88, inf)
329 case LibFunc_expm1l
: // RangeError: (11356, inf)
330 UpperBound
= 11356.0f
;
333 llvm_unreachable("Unhandled library call!");
337 return createCond(CI
, CmpInst::FCMP_OGT
, UpperBound
);
340 // Generate the lower and upper bound condition for RangeError.
341 Value
*LibCallsShrinkWrap::generateTwoRangeCond(CallInst
*CI
,
342 const LibFunc
&Func
) {
343 float UpperBound
, LowerBound
;
345 case LibFunc_cosh
: // RangeError: (x < -710 || x > 710)
346 case LibFunc_sinh
: // Same as cosh
347 LowerBound
= -710.0f
;
350 case LibFunc_coshf
: // RangeError: (x < -89 || x > 89)
351 case LibFunc_sinhf
: // Same as coshf
355 case LibFunc_coshl
: // RangeError: (x < -11357 || x > 11357)
356 case LibFunc_sinhl
: // Same as coshl
357 LowerBound
= -11357.0f
;
358 UpperBound
= 11357.0f
;
360 case LibFunc_exp
: // RangeError: (x < -745 || x > 709)
361 LowerBound
= -745.0f
;
364 case LibFunc_expf
: // RangeError: (x < -103 || x > 88)
365 LowerBound
= -103.0f
;
368 case LibFunc_expl
: // RangeError: (x < -11399 || x > 11356)
369 LowerBound
= -11399.0f
;
370 UpperBound
= 11356.0f
;
372 case LibFunc_exp10
: // RangeError: (x < -323 || x > 308)
373 LowerBound
= -323.0f
;
376 case LibFunc_exp10f
: // RangeError: (x < -45 || x > 38)
380 case LibFunc_exp10l
: // RangeError: (x < -4950 || x > 4932)
381 LowerBound
= -4950.0f
;
382 UpperBound
= 4932.0f
;
384 case LibFunc_exp2
: // RangeError: (x < -1074 || x > 1023)
385 LowerBound
= -1074.0f
;
386 UpperBound
= 1023.0f
;
388 case LibFunc_exp2f
: // RangeError: (x < -149 || x > 127)
389 LowerBound
= -149.0f
;
392 case LibFunc_exp2l
: // RangeError: (x < -16445 || x > 11383)
393 LowerBound
= -16445.0f
;
394 UpperBound
= 11383.0f
;
397 llvm_unreachable("Unhandled library call!");
401 return createOrCond(CI
, CmpInst::FCMP_OGT
, UpperBound
, CmpInst::FCMP_OLT
,
405 // For pow(x,y), We only handle the following cases:
406 // (1) x is a constant && (x >= 1) && (x < MaxUInt8)
407 // Cond is: (y > 127)
408 // (2) x is a value coming from an integer type.
409 // (2.1) if x's bit_size == 8
410 // Cond: (x <= 0 || y > 128)
411 // (2.2) if x's bit_size is 16
412 // Cond: (x <= 0 || y > 64)
413 // (2.3) if x's bit_size is 32
414 // Cond: (x <= 0 || y > 32)
415 // Support for powl(x,y) and powf(x,y) are TBD.
417 // Note that condition can be more conservative than the actual condition
418 // (i.e. we might invoke the calls that will not set the errno.).
420 Value
*LibCallsShrinkWrap::generateCondForPow(CallInst
*CI
,
421 const LibFunc
&Func
) {
422 // FIXME: LibFunc_powf and powl TBD.
423 if (Func
!= LibFunc_pow
) {
424 LLVM_DEBUG(dbgs() << "Not handled powf() and powl()\n");
428 Value
*Base
= CI
->getArgOperand(0);
429 Value
*Exp
= CI
->getArgOperand(1);
430 IRBuilder
<> BBBuilder(CI
);
432 // Constant Base case.
433 if (ConstantFP
*CF
= dyn_cast
<ConstantFP
>(Base
)) {
434 double D
= CF
->getValueAPF().convertToDouble();
435 if (D
< 1.0f
|| D
> APInt::getMaxValue(8).getZExtValue()) {
436 LLVM_DEBUG(dbgs() << "Not handled pow(): constant base out of range\n");
441 Constant
*V
= ConstantFP::get(CI
->getContext(), APFloat(127.0f
));
442 if (!Exp
->getType()->isFloatTy())
443 V
= ConstantExpr::getFPExtend(V
, Exp
->getType());
444 return BBBuilder
.CreateFCmp(CmpInst::FCMP_OGT
, Exp
, V
);
447 // If the Base value coming from an integer type.
448 Instruction
*I
= dyn_cast
<Instruction
>(Base
);
450 LLVM_DEBUG(dbgs() << "Not handled pow(): FP type base\n");
453 unsigned Opcode
= I
->getOpcode();
454 if (Opcode
== Instruction::UIToFP
|| Opcode
== Instruction::SIToFP
) {
455 unsigned BW
= I
->getOperand(0)->getType()->getPrimitiveSizeInBits();
464 LLVM_DEBUG(dbgs() << "Not handled pow(): type too wide\n");
469 Constant
*V
= ConstantFP::get(CI
->getContext(), APFloat(UpperV
));
470 Constant
*V0
= ConstantFP::get(CI
->getContext(), APFloat(0.0f
));
471 if (!Exp
->getType()->isFloatTy())
472 V
= ConstantExpr::getFPExtend(V
, Exp
->getType());
473 if (!Base
->getType()->isFloatTy())
474 V0
= ConstantExpr::getFPExtend(V0
, Exp
->getType());
476 Value
*Cond
= BBBuilder
.CreateFCmp(CmpInst::FCMP_OGT
, Exp
, V
);
477 Value
*Cond0
= BBBuilder
.CreateFCmp(CmpInst::FCMP_OLE
, Base
, V0
);
478 return BBBuilder
.CreateOr(Cond0
, Cond
);
480 LLVM_DEBUG(dbgs() << "Not handled pow(): base not from integer convert\n");
484 // Wrap conditions that can potentially generate errno to the library call.
485 void LibCallsShrinkWrap::shrinkWrapCI(CallInst
*CI
, Value
*Cond
) {
486 assert(Cond
!= nullptr && "ShrinkWrapCI is not expecting an empty call inst");
487 MDNode
*BranchWeights
=
488 MDBuilder(CI
->getContext()).createBranchWeights(1, 2000);
490 TerminatorInst
*NewInst
=
491 SplitBlockAndInsertIfThen(Cond
, CI
, false, BranchWeights
, DT
);
492 BasicBlock
*CallBB
= NewInst
->getParent();
493 CallBB
->setName("cdce.call");
494 BasicBlock
*SuccBB
= CallBB
->getSingleSuccessor();
495 assert(SuccBB
&& "The split block should have a single successor");
496 SuccBB
->setName("cdce.end");
497 CI
->removeFromParent();
498 CallBB
->getInstList().insert(CallBB
->getFirstInsertionPt(), CI
);
499 LLVM_DEBUG(dbgs() << "== Basic Block After ==");
500 LLVM_DEBUG(dbgs() << *CallBB
->getSinglePredecessor() << *CallBB
501 << *CallBB
->getSingleSuccessor() << "\n");
504 // Perform the transformation to a single candidate.
505 bool LibCallsShrinkWrap::perform(CallInst
*CI
) {
507 Function
*Callee
= CI
->getCalledFunction();
508 assert(Callee
&& "perform() should apply to a non-empty callee");
509 TLI
.getLibFunc(*Callee
, Func
);
510 assert(Func
&& "perform() is not expecting an empty function");
512 if (performCallDomainErrorOnly(CI
, Func
) || performCallRangeErrorOnly(CI
, Func
))
514 return performCallErrors(CI
, Func
);
517 void LibCallsShrinkWrapLegacyPass::getAnalysisUsage(AnalysisUsage
&AU
) const {
518 AU
.addPreserved
<DominatorTreeWrapperPass
>();
519 AU
.addPreserved
<GlobalsAAWrapperPass
>();
520 AU
.addRequired
<TargetLibraryInfoWrapperPass
>();
523 static bool runImpl(Function
&F
, const TargetLibraryInfo
&TLI
,
525 if (F
.hasFnAttribute(Attribute::OptimizeForSize
))
527 LibCallsShrinkWrap
CCDCE(TLI
, DT
);
529 bool Changed
= CCDCE
.perform();
531 // Verify the dominator after we've updated it locally.
532 assert(!DT
|| DT
->verify(DominatorTree::VerificationLevel::Fast
));
536 bool LibCallsShrinkWrapLegacyPass::runOnFunction(Function
&F
) {
537 auto &TLI
= getAnalysis
<TargetLibraryInfoWrapperPass
>().getTLI();
538 auto *DTWP
= getAnalysisIfAvailable
<DominatorTreeWrapperPass
>();
539 auto *DT
= DTWP
? &DTWP
->getDomTree() : nullptr;
540 return runImpl(F
, TLI
, DT
);
544 char &LibCallsShrinkWrapPassID
= LibCallsShrinkWrapLegacyPass::ID
;
546 // Public interface to LibCallsShrinkWrap pass.
547 FunctionPass
*createLibCallsShrinkWrapPass() {
548 return new LibCallsShrinkWrapLegacyPass();
551 PreservedAnalyses
LibCallsShrinkWrapPass::run(Function
&F
,
552 FunctionAnalysisManager
&FAM
) {
553 auto &TLI
= FAM
.getResult
<TargetLibraryAnalysis
>(F
);
554 auto *DT
= FAM
.getCachedResult
<DominatorTreeAnalysis
>(F
);
555 if (!runImpl(F
, TLI
, DT
))
556 return PreservedAnalyses::all();
557 auto PA
= PreservedAnalyses();
558 PA
.preserve
<GlobalsAA
>();
559 PA
.preserve
<DominatorTreeAnalysis
>();