xref: /freebsd/contrib/llvm-project/llvm/lib/CodeGen/IntrinsicLowering.cpp (revision aae38d10b4eebf81c0942947e8b83a9bb8651d88) 
1 //===-- IntrinsicLowering.cpp - Intrinsic Lowering default implementation -===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file implements the IntrinsicLowering class.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "llvm/CodeGen/IntrinsicLowering.h"
14 #include "llvm/ADT/SmallVector.h"
15 #include "llvm/IR/CallSite.h"
16 #include "llvm/IR/Constants.h"
17 #include "llvm/IR/DataLayout.h"
18 #include "llvm/IR/DerivedTypes.h"
19 #include "llvm/IR/IRBuilder.h"
20 #include "llvm/IR/Module.h"
21 #include "llvm/IR/Type.h"
22 #include "llvm/Support/ErrorHandling.h"
23 #include "llvm/Support/raw_ostream.h"
24 using namespace llvm;
25 
26 /// This function is used when we want to lower an intrinsic call to a call of
27 /// an external function. This handles hard cases such as when there was already
28 /// a prototype for the external function, but that prototype doesn't match the
29 /// arguments we expect to pass in.
30 template <class ArgIt>
31 static CallInst *ReplaceCallWith(const char *NewFn, CallInst *CI,
32                                  ArgIt ArgBegin, ArgIt ArgEnd,
33                                  Type *RetTy) {
34   // If we haven't already looked up this function, check to see if the
35   // program already contains a function with this name.
36   Module *M = CI->getModule();
37   // Get or insert the definition now.
38   std::vector<Type *> ParamTys;
39   for (ArgIt I = ArgBegin; I != ArgEnd; ++I)
40     ParamTys.push_back((*I)->getType());
41   FunctionCallee FCache =
42       M->getOrInsertFunction(NewFn, FunctionType::get(RetTy, ParamTys, false));
43 
44   IRBuilder<> Builder(CI->getParent(), CI->getIterator());
45   SmallVector<Value *, 8> Args(ArgBegin, ArgEnd);
46   CallInst *NewCI = Builder.CreateCall(FCache, Args);
47   NewCI->setName(CI->getName());
48   if (!CI->use_empty())
49     CI->replaceAllUsesWith(NewCI);
50   return NewCI;
51 }
52 
53 // VisualStudio defines setjmp as _setjmp
54 #if defined(_MSC_VER) && defined(setjmp) && \
55                          !defined(setjmp_undefined_for_msvc)
56 #  pragma push_macro("setjmp")
57 #  undef setjmp
58 #  define setjmp_undefined_for_msvc
59 #endif
60 
61 /// Emit the code to lower bswap of V before the specified instruction IP.
62 static Value *LowerBSWAP(LLVMContext &Context, Value *V, Instruction *IP) {
63   assert(V->getType()->isIntOrIntVectorTy() && "Can't bswap a non-integer type!");
64 
65   unsigned BitSize = V->getType()->getScalarSizeInBits();
66 
67   IRBuilder<> Builder(IP);
68 
69   switch(BitSize) {
70   default: llvm_unreachable("Unhandled type size of value to byteswap!");
71   case 16: {
72     Value *Tmp1 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8),
73                                     "bswap.2");
74     Value *Tmp2 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8),
75                                      "bswap.1");
76     V = Builder.CreateOr(Tmp1, Tmp2, "bswap.i16");
77     break;
78   }
79   case 32: {
80     Value *Tmp4 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 24),
81                                     "bswap.4");
82     Value *Tmp3 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8),
83                                     "bswap.3");
84     Value *Tmp2 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8),
85                                      "bswap.2");
86     Value *Tmp1 = Builder.CreateLShr(V,ConstantInt::get(V->getType(), 24),
87                                      "bswap.1");
88     Tmp3 = Builder.CreateAnd(Tmp3,
89                          ConstantInt::get(V->getType(), 0xFF0000),
90                              "bswap.and3");
91     Tmp2 = Builder.CreateAnd(Tmp2,
92                            ConstantInt::get(V->getType(), 0xFF00),
93                              "bswap.and2");
94     Tmp4 = Builder.CreateOr(Tmp4, Tmp3, "bswap.or1");
95     Tmp2 = Builder.CreateOr(Tmp2, Tmp1, "bswap.or2");
96     V = Builder.CreateOr(Tmp4, Tmp2, "bswap.i32");
97     break;
98   }
99   case 64: {
100     Value *Tmp8 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 56),
101                                     "bswap.8");
102     Value *Tmp7 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 40),
103                                     "bswap.7");
104     Value *Tmp6 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 24),
105                                     "bswap.6");
106     Value *Tmp5 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8),
107                                     "bswap.5");
108     Value* Tmp4 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8),
109                                      "bswap.4");
110     Value* Tmp3 = Builder.CreateLShr(V,
111                                      ConstantInt::get(V->getType(), 24),
112                                      "bswap.3");
113     Value* Tmp2 = Builder.CreateLShr(V,
114                                      ConstantInt::get(V->getType(), 40),
115                                      "bswap.2");
116     Value* Tmp1 = Builder.CreateLShr(V,
117                                      ConstantInt::get(V->getType(), 56),
118                                      "bswap.1");
119     Tmp7 = Builder.CreateAnd(Tmp7,
120                              ConstantInt::get(V->getType(),
121                                               0xFF000000000000ULL),
122                              "bswap.and7");
123     Tmp6 = Builder.CreateAnd(Tmp6,
124                              ConstantInt::get(V->getType(),
125                                               0xFF0000000000ULL),
126                              "bswap.and6");
127     Tmp5 = Builder.CreateAnd(Tmp5,
128                         ConstantInt::get(V->getType(),
129                              0xFF00000000ULL),
130                              "bswap.and5");
131     Tmp4 = Builder.CreateAnd(Tmp4,
132                         ConstantInt::get(V->getType(),
133                              0xFF000000ULL),
134                              "bswap.and4");
135     Tmp3 = Builder.CreateAnd(Tmp3,
136                              ConstantInt::get(V->getType(),
137                              0xFF0000ULL),
138                              "bswap.and3");
139     Tmp2 = Builder.CreateAnd(Tmp2,
140                              ConstantInt::get(V->getType(),
141                              0xFF00ULL),
142                              "bswap.and2");
143     Tmp8 = Builder.CreateOr(Tmp8, Tmp7, "bswap.or1");
144     Tmp6 = Builder.CreateOr(Tmp6, Tmp5, "bswap.or2");
145     Tmp4 = Builder.CreateOr(Tmp4, Tmp3, "bswap.or3");
146     Tmp2 = Builder.CreateOr(Tmp2, Tmp1, "bswap.or4");
147     Tmp8 = Builder.CreateOr(Tmp8, Tmp6, "bswap.or5");
148     Tmp4 = Builder.CreateOr(Tmp4, Tmp2, "bswap.or6");
149     V = Builder.CreateOr(Tmp8, Tmp4, "bswap.i64");
150     break;
151   }
152   }
153   return V;
154 }
155 
156 /// Emit the code to lower ctpop of V before the specified instruction IP.
157 static Value *LowerCTPOP(LLVMContext &Context, Value *V, Instruction *IP) {
158   assert(V->getType()->isIntegerTy() && "Can't ctpop a non-integer type!");
159 
160   static const uint64_t MaskValues[6] = {
161     0x5555555555555555ULL, 0x3333333333333333ULL,
162     0x0F0F0F0F0F0F0F0FULL, 0x00FF00FF00FF00FFULL,
163     0x0000FFFF0000FFFFULL, 0x00000000FFFFFFFFULL
164   };
165 
166   IRBuilder<> Builder(IP);
167 
168   unsigned BitSize = V->getType()->getPrimitiveSizeInBits();
169   unsigned WordSize = (BitSize + 63) / 64;
170   Value *Count = ConstantInt::get(V->getType(), 0);
171 
172   for (unsigned n = 0; n < WordSize; ++n) {
173     Value *PartValue = V;
174     for (unsigned i = 1, ct = 0; i < (BitSize>64 ? 64 : BitSize);
175          i <<= 1, ++ct) {
176       Value *MaskCst = ConstantInt::get(V->getType(), MaskValues[ct]);
177       Value *LHS = Builder.CreateAnd(PartValue, MaskCst, "cppop.and1");
178       Value *VShift = Builder.CreateLShr(PartValue,
179                                         ConstantInt::get(V->getType(), i),
180                                          "ctpop.sh");
181       Value *RHS = Builder.CreateAnd(VShift, MaskCst, "cppop.and2");
182       PartValue = Builder.CreateAdd(LHS, RHS, "ctpop.step");
183     }
184     Count = Builder.CreateAdd(PartValue, Count, "ctpop.part");
185     if (BitSize > 64) {
186       V = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 64),
187                              "ctpop.part.sh");
188       BitSize -= 64;
189     }
190   }
191 
192   return Count;
193 }
194 
195 /// Emit the code to lower ctlz of V before the specified instruction IP.
196 static Value *LowerCTLZ(LLVMContext &Context, Value *V, Instruction *IP) {
197 
198   IRBuilder<> Builder(IP);
199 
200   unsigned BitSize = V->getType()->getPrimitiveSizeInBits();
201   for (unsigned i = 1; i < BitSize; i <<= 1) {
202     Value *ShVal = ConstantInt::get(V->getType(), i);
203     ShVal = Builder.CreateLShr(V, ShVal, "ctlz.sh");
204     V = Builder.CreateOr(V, ShVal, "ctlz.step");
205   }
206 
207   V = Builder.CreateNot(V);
208   return LowerCTPOP(Context, V, IP);
209 }
210 
211 static void ReplaceFPIntrinsicWithCall(CallInst *CI, const char *Fname,
212                                        const char *Dname,
213                                        const char *LDname) {
214   CallSite CS(CI);
215   switch (CI->getArgOperand(0)->getType()->getTypeID()) {
216   default: llvm_unreachable("Invalid type in intrinsic");
217   case Type::FloatTyID:
218     ReplaceCallWith(Fname, CI, CS.arg_begin(), CS.arg_end(),
219                   Type::getFloatTy(CI->getContext()));
220     break;
221   case Type::DoubleTyID:
222     ReplaceCallWith(Dname, CI, CS.arg_begin(), CS.arg_end(),
223                   Type::getDoubleTy(CI->getContext()));
224     break;
225   case Type::X86_FP80TyID:
226   case Type::FP128TyID:
227   case Type::PPC_FP128TyID:
228     ReplaceCallWith(LDname, CI, CS.arg_begin(), CS.arg_end(),
229                   CI->getArgOperand(0)->getType());
230     break;
231   }
232 }
233 
234 void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) {
235   IRBuilder<> Builder(CI);
236   LLVMContext &Context = CI->getContext();
237 
238   const Function *Callee = CI->getCalledFunction();
239   assert(Callee && "Cannot lower an indirect call!");
240 
241   CallSite CS(CI);
242   switch (Callee->getIntrinsicID()) {
243   case Intrinsic::not_intrinsic:
244     report_fatal_error("Cannot lower a call to a non-intrinsic function '"+
245                       Callee->getName() + "'!");
246   default:
247     report_fatal_error("Code generator does not support intrinsic function '"+
248                       Callee->getName()+"'!");
249 
250   case Intrinsic::expect: {
251     // Just replace __builtin_expect(exp, c) with EXP.
252     Value *V = CI->getArgOperand(0);
253     CI->replaceAllUsesWith(V);
254     break;
255   }
256 
257     // The setjmp/longjmp intrinsics should only exist in the code if it was
258     // never optimized (ie, right out of the CFE), or if it has been hacked on
259     // by the lowerinvoke pass.  In both cases, the right thing to do is to
260     // convert the call to an explicit setjmp or longjmp call.
261   case Intrinsic::setjmp: {
262     Value *V = ReplaceCallWith("setjmp", CI, CS.arg_begin(), CS.arg_end(),
263                                Type::getInt32Ty(Context));
264     if (!CI->getType()->isVoidTy())
265       CI->replaceAllUsesWith(V);
266     break;
267   }
268   case Intrinsic::sigsetjmp:
269      if (!CI->getType()->isVoidTy())
270        CI->replaceAllUsesWith(Constant::getNullValue(CI->getType()));
271      break;
272 
273   case Intrinsic::longjmp: {
274     ReplaceCallWith("longjmp", CI, CS.arg_begin(), CS.arg_end(),
275                     Type::getVoidTy(Context));
276     break;
277   }
278 
279   case Intrinsic::siglongjmp: {
280     // Insert the call to abort
281     ReplaceCallWith("abort", CI, CS.arg_end(), CS.arg_end(),
282                     Type::getVoidTy(Context));
283     break;
284   }
285   case Intrinsic::ctpop:
286     CI->replaceAllUsesWith(LowerCTPOP(Context, CI->getArgOperand(0), CI));
287     break;
288 
289   case Intrinsic::bswap:
290     CI->replaceAllUsesWith(LowerBSWAP(Context, CI->getArgOperand(0), CI));
291     break;
292 
293   case Intrinsic::ctlz:
294     CI->replaceAllUsesWith(LowerCTLZ(Context, CI->getArgOperand(0), CI));
295     break;
296 
297   case Intrinsic::cttz: {
298     // cttz(x) -> ctpop(~X & (X-1))
299     Value *Src = CI->getArgOperand(0);
300     Value *NotSrc = Builder.CreateNot(Src);
301     NotSrc->setName(Src->getName() + ".not");
302     Value *SrcM1 = ConstantInt::get(Src->getType(), 1);
303     SrcM1 = Builder.CreateSub(Src, SrcM1);
304     Src = LowerCTPOP(Context, Builder.CreateAnd(NotSrc, SrcM1), CI);
305     CI->replaceAllUsesWith(Src);
306     break;
307   }
308 
309   case Intrinsic::stacksave:
310   case Intrinsic::stackrestore: {
311     if (!Warned)
312       errs() << "WARNING: this target does not support the llvm.stack"
313              << (Callee->getIntrinsicID() == Intrinsic::stacksave ?
314                "save" : "restore") << " intrinsic.\n";
315     Warned = true;
316     if (Callee->getIntrinsicID() == Intrinsic::stacksave)
317       CI->replaceAllUsesWith(Constant::getNullValue(CI->getType()));
318     break;
319   }
320 
321   case Intrinsic::get_dynamic_area_offset:
322     errs() << "WARNING: this target does not support the custom llvm.get."
323               "dynamic.area.offset.  It is being lowered to a constant 0\n";
324     // Just lower it to a constant 0 because for most targets
325     // @llvm.get.dynamic.area.offset is lowered to zero.
326     CI->replaceAllUsesWith(ConstantInt::get(CI->getType(), 0));
327     break;
328   case Intrinsic::returnaddress:
329   case Intrinsic::frameaddress:
330     errs() << "WARNING: this target does not support the llvm."
331            << (Callee->getIntrinsicID() == Intrinsic::returnaddress ?
332              "return" : "frame") << "address intrinsic.\n";
333     CI->replaceAllUsesWith(
334         ConstantPointerNull::get(cast<PointerType>(CI->getType())));
335     break;
336   case Intrinsic::addressofreturnaddress:
337     errs() << "WARNING: this target does not support the "
338               "llvm.addressofreturnaddress intrinsic.\n";
339     CI->replaceAllUsesWith(
340         ConstantPointerNull::get(cast<PointerType>(CI->getType())));
341     break;
342 
343   case Intrinsic::prefetch:
344     break;    // Simply strip out prefetches on unsupported architectures
345 
346   case Intrinsic::pcmarker:
347     break;    // Simply strip out pcmarker on unsupported architectures
348   case Intrinsic::readcyclecounter: {
349     errs() << "WARNING: this target does not support the llvm.readcyclecoun"
350            << "ter intrinsic.  It is being lowered to a constant 0\n";
351     CI->replaceAllUsesWith(ConstantInt::get(Type::getInt64Ty(Context), 0));
352     break;
353   }
354 
355   case Intrinsic::dbg_declare:
356   case Intrinsic::dbg_label:
357     break;    // Simply strip out debugging intrinsics
358 
359   case Intrinsic::eh_typeid_for:
360     // Return something different to eh_selector.
361     CI->replaceAllUsesWith(ConstantInt::get(CI->getType(), 1));
362     break;
363 
364   case Intrinsic::annotation:
365   case Intrinsic::ptr_annotation:
366     // Just drop the annotation, but forward the value
367     CI->replaceAllUsesWith(CI->getOperand(0));
368     break;
369 
370   case Intrinsic::assume:
371   case Intrinsic::var_annotation:
372     break;   // Strip out these intrinsics
373 
374   case Intrinsic::memcpy: {
375     Type *IntPtr = DL.getIntPtrType(Context);
376     Value *Size = Builder.CreateIntCast(CI->getArgOperand(2), IntPtr,
377                                         /* isSigned */ false);
378     Value *Ops[3];
379     Ops[0] = CI->getArgOperand(0);
380     Ops[1] = CI->getArgOperand(1);
381     Ops[2] = Size;
382     ReplaceCallWith("memcpy", CI, Ops, Ops+3, CI->getArgOperand(0)->getType());
383     break;
384   }
385   case Intrinsic::memmove: {
386     Type *IntPtr = DL.getIntPtrType(Context);
387     Value *Size = Builder.CreateIntCast(CI->getArgOperand(2), IntPtr,
388                                         /* isSigned */ false);
389     Value *Ops[3];
390     Ops[0] = CI->getArgOperand(0);
391     Ops[1] = CI->getArgOperand(1);
392     Ops[2] = Size;
393     ReplaceCallWith("memmove", CI, Ops, Ops+3, CI->getArgOperand(0)->getType());
394     break;
395   }
396   case Intrinsic::memset: {
397     Value *Op0 = CI->getArgOperand(0);
398     Type *IntPtr = DL.getIntPtrType(Op0->getType());
399     Value *Size = Builder.CreateIntCast(CI->getArgOperand(2), IntPtr,
400                                         /* isSigned */ false);
401     Value *Ops[3];
402     Ops[0] = Op0;
403     // Extend the amount to i32.
404     Ops[1] = Builder.CreateIntCast(CI->getArgOperand(1),
405                                    Type::getInt32Ty(Context),
406                                    /* isSigned */ false);
407     Ops[2] = Size;
408     ReplaceCallWith("memset", CI, Ops, Ops+3, CI->getArgOperand(0)->getType());
409     break;
410   }
411   case Intrinsic::sqrt: {
412     ReplaceFPIntrinsicWithCall(CI, "sqrtf", "sqrt", "sqrtl");
413     break;
414   }
415   case Intrinsic::log: {
416     ReplaceFPIntrinsicWithCall(CI, "logf", "log", "logl");
417     break;
418   }
419   case Intrinsic::log2: {
420     ReplaceFPIntrinsicWithCall(CI, "log2f", "log2", "log2l");
421     break;
422   }
423   case Intrinsic::log10: {
424     ReplaceFPIntrinsicWithCall(CI, "log10f", "log10", "log10l");
425     break;
426   }
427   case Intrinsic::exp: {
428     ReplaceFPIntrinsicWithCall(CI, "expf", "exp", "expl");
429     break;
430   }
431   case Intrinsic::exp2: {
432     ReplaceFPIntrinsicWithCall(CI, "exp2f", "exp2", "exp2l");
433     break;
434   }
435   case Intrinsic::pow: {
436     ReplaceFPIntrinsicWithCall(CI, "powf", "pow", "powl");
437     break;
438   }
439   case Intrinsic::sin: {
440     ReplaceFPIntrinsicWithCall(CI, "sinf", "sin", "sinl");
441     break;
442   }
443   case Intrinsic::cos: {
444     ReplaceFPIntrinsicWithCall(CI, "cosf", "cos", "cosl");
445     break;
446   }
447   case Intrinsic::floor: {
448     ReplaceFPIntrinsicWithCall(CI, "floorf", "floor", "floorl");
449     break;
450   }
451   case Intrinsic::ceil: {
452     ReplaceFPIntrinsicWithCall(CI, "ceilf", "ceil", "ceill");
453     break;
454   }
455   case Intrinsic::trunc: {
456     ReplaceFPIntrinsicWithCall(CI, "truncf", "trunc", "truncl");
457     break;
458   }
459   case Intrinsic::round: {
460     ReplaceFPIntrinsicWithCall(CI, "roundf", "round", "roundl");
461     break;
462   }
463   case Intrinsic::copysign: {
464     ReplaceFPIntrinsicWithCall(CI, "copysignf", "copysign", "copysignl");
465     break;
466   }
467   case Intrinsic::flt_rounds:
468      // Lower to "round to the nearest"
469      if (!CI->getType()->isVoidTy())
470        CI->replaceAllUsesWith(ConstantInt::get(CI->getType(), 1));
471      break;
472   case Intrinsic::invariant_start:
473   case Intrinsic::lifetime_start:
474     // Discard region information.
475     CI->replaceAllUsesWith(UndefValue::get(CI->getType()));
476     break;
477   case Intrinsic::invariant_end:
478   case Intrinsic::lifetime_end:
479     // Discard region information.
480     break;
481   }
482 
483   assert(CI->use_empty() &&
484          "Lowering should have eliminated any uses of the intrinsic call!");
485   CI->eraseFromParent();
486 }
487 
488 bool IntrinsicLowering::LowerToByteSwap(CallInst *CI) {
489   // Verify this is a simple bswap.
490   if (CI->getNumArgOperands() != 1 ||
491       CI->getType() != CI->getArgOperand(0)->getType() ||
492       !CI->getType()->isIntegerTy())
493     return false;
494 
495   IntegerType *Ty = dyn_cast<IntegerType>(CI->getType());
496   if (!Ty)
497     return false;
498 
499   // Okay, we can do this xform, do so now.
500   Module *M = CI->getModule();
501   Function *Int = Intrinsic::getDeclaration(M, Intrinsic::bswap, Ty);
502 
503   Value *Op = CI->getArgOperand(0);
504   Op = CallInst::Create(Int, Op, CI->getName(), CI);
505 
506   CI->replaceAllUsesWith(Op);
507   CI->eraseFromParent();
508   return true;
509 }
510