1 //===-- IntrinsicLowering.cpp - Intrinsic Lowering default implementation -===//
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 file implements the IntrinsicLowering class.
12 //===----------------------------------------------------------------------===//
14 #include "llvm/Constants.h"
15 #include "llvm/DerivedTypes.h"
16 #include "llvm/Module.h"
17 #include "llvm/Type.h"
18 #include "llvm/CodeGen/IntrinsicLowering.h"
19 #include "llvm/Support/ErrorHandling.h"
20 #include "llvm/Support/IRBuilder.h"
21 #include "llvm/Support/raw_ostream.h"
22 #include "llvm/Target/TargetData.h"
23 #include "llvm/ADT/SmallVector.h"
26 template <class ArgIt
>
27 static void EnsureFunctionExists(Module
&M
, const char *Name
,
28 ArgIt ArgBegin
, ArgIt ArgEnd
,
30 // Insert a correctly-typed definition now.
31 std::vector
<const Type
*> ParamTys
;
32 for (ArgIt I
= ArgBegin
; I
!= ArgEnd
; ++I
)
33 ParamTys
.push_back(I
->getType());
34 M
.getOrInsertFunction(Name
, FunctionType::get(RetTy
, ParamTys
, false));
37 static void EnsureFPIntrinsicsExist(Module
&M
, Function
*Fn
,
39 const char *DName
, const char *LDName
) {
40 // Insert definitions for all the floating point types.
41 switch((int)Fn
->arg_begin()->getType()->getTypeID()) {
43 EnsureFunctionExists(M
, FName
, Fn
->arg_begin(), Fn
->arg_end(),
44 Type::getFloatTy(M
.getContext()));
46 case Type::DoubleTyID
:
47 EnsureFunctionExists(M
, DName
, Fn
->arg_begin(), Fn
->arg_end(),
48 Type::getDoubleTy(M
.getContext()));
50 case Type::X86_FP80TyID
:
52 case Type::PPC_FP128TyID
:
53 EnsureFunctionExists(M
, LDName
, Fn
->arg_begin(), Fn
->arg_end(),
54 Fn
->arg_begin()->getType());
59 /// ReplaceCallWith - This function is used when we want to lower an intrinsic
60 /// call to a call of an external function. This handles hard cases such as
61 /// when there was already a prototype for the external function, and if that
62 /// prototype doesn't match the arguments we expect to pass in.
63 template <class ArgIt
>
64 static CallInst
*ReplaceCallWith(const char *NewFn
, CallInst
*CI
,
65 ArgIt ArgBegin
, ArgIt ArgEnd
,
67 // If we haven't already looked up this function, check to see if the
68 // program already contains a function with this name.
69 Module
*M
= CI
->getParent()->getParent()->getParent();
70 // Get or insert the definition now.
71 std::vector
<const Type
*> ParamTys
;
72 for (ArgIt I
= ArgBegin
; I
!= ArgEnd
; ++I
)
73 ParamTys
.push_back((*I
)->getType());
74 Constant
* FCache
= M
->getOrInsertFunction(NewFn
,
75 FunctionType::get(RetTy
, ParamTys
, false));
77 IRBuilder
<> Builder(CI
->getParent(), CI
);
78 SmallVector
<Value
*, 8> Args(ArgBegin
, ArgEnd
);
79 CallInst
*NewCI
= Builder
.CreateCall(FCache
, Args
.begin(), Args
.end());
80 NewCI
->setName(CI
->getName());
82 CI
->replaceAllUsesWith(NewCI
);
86 void IntrinsicLowering::AddPrototypes(Module
&M
) {
87 LLVMContext
&Context
= M
.getContext();
88 for (Module::iterator I
= M
.begin(), E
= M
.end(); I
!= E
; ++I
)
89 if (I
->isDeclaration() && !I
->use_empty())
90 switch (I
->getIntrinsicID()) {
92 case Intrinsic::setjmp
:
93 EnsureFunctionExists(M
, "setjmp", I
->arg_begin(), I
->arg_end(),
94 Type::getInt32Ty(M
.getContext()));
96 case Intrinsic::longjmp
:
97 EnsureFunctionExists(M
, "longjmp", I
->arg_begin(), I
->arg_end(),
98 Type::getVoidTy(M
.getContext()));
100 case Intrinsic::siglongjmp
:
101 EnsureFunctionExists(M
, "abort", I
->arg_end(), I
->arg_end(),
102 Type::getVoidTy(M
.getContext()));
104 case Intrinsic::memcpy
:
105 M
.getOrInsertFunction("memcpy",
106 PointerType::getUnqual(Type::getInt8Ty(Context
)),
107 PointerType::getUnqual(Type::getInt8Ty(Context
)),
108 PointerType::getUnqual(Type::getInt8Ty(Context
)),
109 TD
.getIntPtrType(Context
), (Type
*)0);
111 case Intrinsic::memmove
:
112 M
.getOrInsertFunction("memmove",
113 PointerType::getUnqual(Type::getInt8Ty(Context
)),
114 PointerType::getUnqual(Type::getInt8Ty(Context
)),
115 PointerType::getUnqual(Type::getInt8Ty(Context
)),
116 TD
.getIntPtrType(Context
), (Type
*)0);
118 case Intrinsic::memset
:
119 M
.getOrInsertFunction("memset",
120 PointerType::getUnqual(Type::getInt8Ty(Context
)),
121 PointerType::getUnqual(Type::getInt8Ty(Context
)),
122 Type::getInt32Ty(M
.getContext()),
123 TD
.getIntPtrType(Context
), (Type
*)0);
125 case Intrinsic::sqrt
:
126 EnsureFPIntrinsicsExist(M
, I
, "sqrtf", "sqrt", "sqrtl");
129 EnsureFPIntrinsicsExist(M
, I
, "sinf", "sin", "sinl");
132 EnsureFPIntrinsicsExist(M
, I
, "cosf", "cos", "cosl");
135 EnsureFPIntrinsicsExist(M
, I
, "powf", "pow", "powl");
138 EnsureFPIntrinsicsExist(M
, I
, "logf", "log", "logl");
140 case Intrinsic::log2
:
141 EnsureFPIntrinsicsExist(M
, I
, "log2f", "log2", "log2l");
143 case Intrinsic::log10
:
144 EnsureFPIntrinsicsExist(M
, I
, "log10f", "log10", "log10l");
147 EnsureFPIntrinsicsExist(M
, I
, "expf", "exp", "expl");
149 case Intrinsic::exp2
:
150 EnsureFPIntrinsicsExist(M
, I
, "exp2f", "exp2", "exp2l");
155 /// LowerBSWAP - Emit the code to lower bswap of V before the specified
157 static Value
*LowerBSWAP(LLVMContext
&Context
, Value
*V
, Instruction
*IP
) {
158 assert(V
->getType()->isInteger() && "Can't bswap a non-integer type!");
160 unsigned BitSize
= V
->getType()->getPrimitiveSizeInBits();
162 IRBuilder
<> Builder(IP
->getParent(), IP
);
165 default: llvm_unreachable("Unhandled type size of value to byteswap!");
167 Value
*Tmp1
= Builder
.CreateShl(V
, ConstantInt::get(V
->getType(), 8),
169 Value
*Tmp2
= Builder
.CreateLShr(V
, ConstantInt::get(V
->getType(), 8),
171 V
= Builder
.CreateOr(Tmp1
, Tmp2
, "bswap.i16");
175 Value
*Tmp4
= Builder
.CreateShl(V
, ConstantInt::get(V
->getType(), 24),
177 Value
*Tmp3
= Builder
.CreateShl(V
, ConstantInt::get(V
->getType(), 8),
179 Value
*Tmp2
= Builder
.CreateLShr(V
, ConstantInt::get(V
->getType(), 8),
181 Value
*Tmp1
= Builder
.CreateLShr(V
,ConstantInt::get(V
->getType(), 24),
183 Tmp3
= Builder
.CreateAnd(Tmp3
,
184 ConstantInt::get(Type::getInt32Ty(Context
), 0xFF0000),
186 Tmp2
= Builder
.CreateAnd(Tmp2
,
187 ConstantInt::get(Type::getInt32Ty(Context
), 0xFF00),
189 Tmp4
= Builder
.CreateOr(Tmp4
, Tmp3
, "bswap.or1");
190 Tmp2
= Builder
.CreateOr(Tmp2
, Tmp1
, "bswap.or2");
191 V
= Builder
.CreateOr(Tmp4
, Tmp2
, "bswap.i32");
195 Value
*Tmp8
= Builder
.CreateShl(V
, ConstantInt::get(V
->getType(), 56),
197 Value
*Tmp7
= Builder
.CreateShl(V
, ConstantInt::get(V
->getType(), 40),
199 Value
*Tmp6
= Builder
.CreateShl(V
, ConstantInt::get(V
->getType(), 24),
201 Value
*Tmp5
= Builder
.CreateShl(V
, ConstantInt::get(V
->getType(), 8),
203 Value
* Tmp4
= Builder
.CreateLShr(V
, ConstantInt::get(V
->getType(), 8),
205 Value
* Tmp3
= Builder
.CreateLShr(V
,
206 ConstantInt::get(V
->getType(), 24),
208 Value
* Tmp2
= Builder
.CreateLShr(V
,
209 ConstantInt::get(V
->getType(), 40),
211 Value
* Tmp1
= Builder
.CreateLShr(V
,
212 ConstantInt::get(V
->getType(), 56),
214 Tmp7
= Builder
.CreateAnd(Tmp7
,
215 ConstantInt::get(Type::getInt64Ty(Context
),
216 0xFF000000000000ULL
),
218 Tmp6
= Builder
.CreateAnd(Tmp6
,
219 ConstantInt::get(Type::getInt64Ty(Context
),
222 Tmp5
= Builder
.CreateAnd(Tmp5
,
223 ConstantInt::get(Type::getInt64Ty(Context
),
226 Tmp4
= Builder
.CreateAnd(Tmp4
,
227 ConstantInt::get(Type::getInt64Ty(Context
),
230 Tmp3
= Builder
.CreateAnd(Tmp3
,
231 ConstantInt::get(Type::getInt64Ty(Context
),
234 Tmp2
= Builder
.CreateAnd(Tmp2
,
235 ConstantInt::get(Type::getInt64Ty(Context
),
238 Tmp8
= Builder
.CreateOr(Tmp8
, Tmp7
, "bswap.or1");
239 Tmp6
= Builder
.CreateOr(Tmp6
, Tmp5
, "bswap.or2");
240 Tmp4
= Builder
.CreateOr(Tmp4
, Tmp3
, "bswap.or3");
241 Tmp2
= Builder
.CreateOr(Tmp2
, Tmp1
, "bswap.or4");
242 Tmp8
= Builder
.CreateOr(Tmp8
, Tmp6
, "bswap.or5");
243 Tmp4
= Builder
.CreateOr(Tmp4
, Tmp2
, "bswap.or6");
244 V
= Builder
.CreateOr(Tmp8
, Tmp4
, "bswap.i64");
251 /// LowerCTPOP - Emit the code to lower ctpop of V before the specified
253 static Value
*LowerCTPOP(LLVMContext
&Context
, Value
*V
, Instruction
*IP
) {
254 assert(V
->getType()->isInteger() && "Can't ctpop a non-integer type!");
256 static const uint64_t MaskValues
[6] = {
257 0x5555555555555555ULL
, 0x3333333333333333ULL
,
258 0x0F0F0F0F0F0F0F0FULL
, 0x00FF00FF00FF00FFULL
,
259 0x0000FFFF0000FFFFULL
, 0x00000000FFFFFFFFULL
262 IRBuilder
<> Builder(IP
->getParent(), IP
);
264 unsigned BitSize
= V
->getType()->getPrimitiveSizeInBits();
265 unsigned WordSize
= (BitSize
+ 63) / 64;
266 Value
*Count
= ConstantInt::get(V
->getType(), 0);
268 for (unsigned n
= 0; n
< WordSize
; ++n
) {
269 Value
*PartValue
= V
;
270 for (unsigned i
= 1, ct
= 0; i
< (BitSize
>64 ? 64 : BitSize
);
272 Value
*MaskCst
= ConstantInt::get(V
->getType(), MaskValues
[ct
]);
273 Value
*LHS
= Builder
.CreateAnd(PartValue
, MaskCst
, "cppop.and1");
274 Value
*VShift
= Builder
.CreateLShr(PartValue
,
275 ConstantInt::get(V
->getType(), i
),
277 Value
*RHS
= Builder
.CreateAnd(VShift
, MaskCst
, "cppop.and2");
278 PartValue
= Builder
.CreateAdd(LHS
, RHS
, "ctpop.step");
280 Count
= Builder
.CreateAdd(PartValue
, Count
, "ctpop.part");
282 V
= Builder
.CreateLShr(V
, ConstantInt::get(V
->getType(), 64),
291 /// LowerCTLZ - Emit the code to lower ctlz of V before the specified
293 static Value
*LowerCTLZ(LLVMContext
&Context
, Value
*V
, Instruction
*IP
) {
295 IRBuilder
<> Builder(IP
->getParent(), IP
);
297 unsigned BitSize
= V
->getType()->getPrimitiveSizeInBits();
298 for (unsigned i
= 1; i
< BitSize
; i
<<= 1) {
299 Value
*ShVal
= ConstantInt::get(V
->getType(), i
);
300 ShVal
= Builder
.CreateLShr(V
, ShVal
, "ctlz.sh");
301 V
= Builder
.CreateOr(V
, ShVal
, "ctlz.step");
304 V
= Builder
.CreateNot(V
);
305 return LowerCTPOP(Context
, V
, IP
);
308 static void ReplaceFPIntrinsicWithCall(CallInst
*CI
, const char *Fname
,
310 const char *LDname
) {
311 switch (CI
->getOperand(1)->getType()->getTypeID()) {
312 default: llvm_unreachable("Invalid type in intrinsic");
313 case Type::FloatTyID
:
314 ReplaceCallWith(Fname
, CI
, CI
->op_begin() + 1, CI
->op_end(),
315 Type::getFloatTy(CI
->getContext()));
317 case Type::DoubleTyID
:
318 ReplaceCallWith(Dname
, CI
, CI
->op_begin() + 1, CI
->op_end(),
319 Type::getDoubleTy(CI
->getContext()));
321 case Type::X86_FP80TyID
:
322 case Type::FP128TyID
:
323 case Type::PPC_FP128TyID
:
324 ReplaceCallWith(LDname
, CI
, CI
->op_begin() + 1, CI
->op_end(),
325 CI
->getOperand(1)->getType());
330 void IntrinsicLowering::LowerIntrinsicCall(CallInst
*CI
) {
331 IRBuilder
<> Builder(CI
->getParent(), CI
);
332 LLVMContext
&Context
= CI
->getContext();
334 Function
*Callee
= CI
->getCalledFunction();
335 assert(Callee
&& "Cannot lower an indirect call!");
337 switch (Callee
->getIntrinsicID()) {
338 case Intrinsic::not_intrinsic
:
339 llvm_report_error("Cannot lower a call to a non-intrinsic function '"+
340 Callee
->getName() + "'!");
342 llvm_report_error("Code generator does not support intrinsic function '"+
343 Callee
->getName()+"'!");
345 // The setjmp/longjmp intrinsics should only exist in the code if it was
346 // never optimized (ie, right out of the CFE), or if it has been hacked on
347 // by the lowerinvoke pass. In both cases, the right thing to do is to
348 // convert the call to an explicit setjmp or longjmp call.
349 case Intrinsic::setjmp
: {
350 Value
*V
= ReplaceCallWith("setjmp", CI
, CI
->op_begin() + 1, CI
->op_end(),
351 Type::getInt32Ty(Context
));
352 if (CI
->getType() != Type::getVoidTy(Context
))
353 CI
->replaceAllUsesWith(V
);
356 case Intrinsic::sigsetjmp
:
357 if (CI
->getType() != Type::getVoidTy(Context
))
358 CI
->replaceAllUsesWith(Constant::getNullValue(CI
->getType()));
361 case Intrinsic::longjmp
: {
362 ReplaceCallWith("longjmp", CI
, CI
->op_begin() + 1, CI
->op_end(),
363 Type::getVoidTy(Context
));
367 case Intrinsic::siglongjmp
: {
368 // Insert the call to abort
369 ReplaceCallWith("abort", CI
, CI
->op_end(), CI
->op_end(),
370 Type::getVoidTy(Context
));
373 case Intrinsic::ctpop
:
374 CI
->replaceAllUsesWith(LowerCTPOP(Context
, CI
->getOperand(1), CI
));
377 case Intrinsic::bswap
:
378 CI
->replaceAllUsesWith(LowerBSWAP(Context
, CI
->getOperand(1), CI
));
381 case Intrinsic::ctlz
:
382 CI
->replaceAllUsesWith(LowerCTLZ(Context
, CI
->getOperand(1), CI
));
385 case Intrinsic::cttz
: {
386 // cttz(x) -> ctpop(~X & (X-1))
387 Value
*Src
= CI
->getOperand(1);
388 Value
*NotSrc
= Builder
.CreateNot(Src
);
389 NotSrc
->setName(Src
->getName() + ".not");
390 Value
*SrcM1
= ConstantInt::get(Src
->getType(), 1);
391 SrcM1
= Builder
.CreateSub(Src
, SrcM1
);
392 Src
= LowerCTPOP(Context
, Builder
.CreateAnd(NotSrc
, SrcM1
), CI
);
393 CI
->replaceAllUsesWith(Src
);
397 case Intrinsic::stacksave
:
398 case Intrinsic::stackrestore
: {
400 errs() << "WARNING: this target does not support the llvm.stack"
401 << (Callee
->getIntrinsicID() == Intrinsic::stacksave
?
402 "save" : "restore") << " intrinsic.\n";
404 if (Callee
->getIntrinsicID() == Intrinsic::stacksave
)
405 CI
->replaceAllUsesWith(Constant::getNullValue(CI
->getType()));
409 case Intrinsic::returnaddress
:
410 case Intrinsic::frameaddress
:
411 errs() << "WARNING: this target does not support the llvm."
412 << (Callee
->getIntrinsicID() == Intrinsic::returnaddress
?
413 "return" : "frame") << "address intrinsic.\n";
414 CI
->replaceAllUsesWith(ConstantPointerNull::get(
415 cast
<PointerType
>(CI
->getType())));
418 case Intrinsic::prefetch
:
419 break; // Simply strip out prefetches on unsupported architectures
421 case Intrinsic::pcmarker
:
422 break; // Simply strip out pcmarker on unsupported architectures
423 case Intrinsic::readcyclecounter
: {
424 errs() << "WARNING: this target does not support the llvm.readcyclecoun"
425 << "ter intrinsic. It is being lowered to a constant 0\n";
426 CI
->replaceAllUsesWith(ConstantInt::get(Type::getInt64Ty(Context
), 0));
430 case Intrinsic::dbg_stoppoint
:
431 case Intrinsic::dbg_region_start
:
432 case Intrinsic::dbg_region_end
:
433 case Intrinsic::dbg_func_start
:
434 case Intrinsic::dbg_declare
:
435 break; // Simply strip out debugging intrinsics
437 case Intrinsic::eh_exception
:
438 case Intrinsic::eh_selector_i32
:
439 case Intrinsic::eh_selector_i64
:
440 CI
->replaceAllUsesWith(Constant::getNullValue(CI
->getType()));
443 case Intrinsic::eh_typeid_for_i32
:
444 case Intrinsic::eh_typeid_for_i64
:
445 // Return something different to eh_selector.
446 CI
->replaceAllUsesWith(ConstantInt::get(CI
->getType(), 1));
449 case Intrinsic::var_annotation
:
450 break; // Strip out annotate intrinsic
452 case Intrinsic::memcpy
: {
453 const IntegerType
*IntPtr
= TD
.getIntPtrType(Context
);
454 Value
*Size
= Builder
.CreateIntCast(CI
->getOperand(3), IntPtr
,
455 /* isSigned */ false);
457 Ops
[0] = CI
->getOperand(1);
458 Ops
[1] = CI
->getOperand(2);
460 ReplaceCallWith("memcpy", CI
, Ops
, Ops
+3, CI
->getOperand(1)->getType());
463 case Intrinsic::memmove
: {
464 const IntegerType
*IntPtr
= TD
.getIntPtrType(Context
);
465 Value
*Size
= Builder
.CreateIntCast(CI
->getOperand(3), IntPtr
,
466 /* isSigned */ false);
468 Ops
[0] = CI
->getOperand(1);
469 Ops
[1] = CI
->getOperand(2);
471 ReplaceCallWith("memmove", CI
, Ops
, Ops
+3, CI
->getOperand(1)->getType());
474 case Intrinsic::memset
: {
475 const IntegerType
*IntPtr
= TD
.getIntPtrType(Context
);
476 Value
*Size
= Builder
.CreateIntCast(CI
->getOperand(3), IntPtr
,
477 /* isSigned */ false);
479 Ops
[0] = CI
->getOperand(1);
480 // Extend the amount to i32.
481 Ops
[1] = Builder
.CreateIntCast(CI
->getOperand(2), Type::getInt32Ty(Context
),
482 /* isSigned */ false);
484 ReplaceCallWith("memset", CI
, Ops
, Ops
+3, CI
->getOperand(1)->getType());
487 case Intrinsic::sqrt
: {
488 ReplaceFPIntrinsicWithCall(CI
, "sqrtf", "sqrt", "sqrtl");
491 case Intrinsic::log
: {
492 ReplaceFPIntrinsicWithCall(CI
, "logf", "log", "logl");
495 case Intrinsic::log2
: {
496 ReplaceFPIntrinsicWithCall(CI
, "log2f", "log2", "log2l");
499 case Intrinsic::log10
: {
500 ReplaceFPIntrinsicWithCall(CI
, "log10f", "log10", "log10l");
503 case Intrinsic::exp
: {
504 ReplaceFPIntrinsicWithCall(CI
, "expf", "exp", "expl");
507 case Intrinsic::exp2
: {
508 ReplaceFPIntrinsicWithCall(CI
, "exp2f", "exp2", "exp2l");
511 case Intrinsic::pow
: {
512 ReplaceFPIntrinsicWithCall(CI
, "powf", "pow", "powl");
515 case Intrinsic::flt_rounds
:
516 // Lower to "round to the nearest"
517 if (CI
->getType() != Type::getVoidTy(Context
))
518 CI
->replaceAllUsesWith(ConstantInt::get(CI
->getType(), 1));
522 assert(CI
->use_empty() &&
523 "Lowering should have eliminated any uses of the intrinsic call!");
524 CI
->eraseFromParent();