[InstCombine] Signed saturation patterns
[llvm-core.git] / lib / Transforms / Utils / IntegerDivision.cpp
blob9082049c82da67ce8a9d5df6219f33917f280070
1 //===-- IntegerDivision.cpp - Expand integer division ---------------------===//
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 contains an implementation of 32bit and 64bit scalar integer
10 // division for targets that don't have native support. It's largely derived
11 // from compiler-rt's implementations of __udivsi3 and __udivmoddi4,
12 // but hand-tuned for targets that prefer less control flow.
14 //===----------------------------------------------------------------------===//
16 #include "llvm/Transforms/Utils/IntegerDivision.h"
17 #include "llvm/IR/Function.h"
18 #include "llvm/IR/IRBuilder.h"
19 #include "llvm/IR/Instructions.h"
20 #include "llvm/IR/Intrinsics.h"
21 #include <utility>
23 using namespace llvm;
25 #define DEBUG_TYPE "integer-division"
27 /// Generate code to compute the remainder of two signed integers. Returns the
28 /// remainder, which will have the sign of the dividend. Builder's insert point
29 /// should be pointing where the caller wants code generated, e.g. at the srem
30 /// instruction. This will generate a urem in the process, and Builder's insert
31 /// point will be pointing at the uren (if present, i.e. not folded), ready to
32 /// be expanded if the user wishes
33 static Value *generateSignedRemainderCode(Value *Dividend, Value *Divisor,
34 IRBuilder<> &Builder) {
35 unsigned BitWidth = Dividend->getType()->getIntegerBitWidth();
36 ConstantInt *Shift;
38 if (BitWidth == 64) {
39 Shift = Builder.getInt64(63);
40 } else {
41 assert(BitWidth == 32 && "Unexpected bit width");
42 Shift = Builder.getInt32(31);
45 // Following instructions are generated for both i32 (shift 31) and
46 // i64 (shift 63).
48 // ; %dividend_sgn = ashr i32 %dividend, 31
49 // ; %divisor_sgn = ashr i32 %divisor, 31
50 // ; %dvd_xor = xor i32 %dividend, %dividend_sgn
51 // ; %dvs_xor = xor i32 %divisor, %divisor_sgn
52 // ; %u_dividend = sub i32 %dvd_xor, %dividend_sgn
53 // ; %u_divisor = sub i32 %dvs_xor, %divisor_sgn
54 // ; %urem = urem i32 %dividend, %divisor
55 // ; %xored = xor i32 %urem, %dividend_sgn
56 // ; %srem = sub i32 %xored, %dividend_sgn
57 Value *DividendSign = Builder.CreateAShr(Dividend, Shift);
58 Value *DivisorSign = Builder.CreateAShr(Divisor, Shift);
59 Value *DvdXor = Builder.CreateXor(Dividend, DividendSign);
60 Value *DvsXor = Builder.CreateXor(Divisor, DivisorSign);
61 Value *UDividend = Builder.CreateSub(DvdXor, DividendSign);
62 Value *UDivisor = Builder.CreateSub(DvsXor, DivisorSign);
63 Value *URem = Builder.CreateURem(UDividend, UDivisor);
64 Value *Xored = Builder.CreateXor(URem, DividendSign);
65 Value *SRem = Builder.CreateSub(Xored, DividendSign);
67 if (Instruction *URemInst = dyn_cast<Instruction>(URem))
68 Builder.SetInsertPoint(URemInst);
70 return SRem;
74 /// Generate code to compute the remainder of two unsigned integers. Returns the
75 /// remainder. Builder's insert point should be pointing where the caller wants
76 /// code generated, e.g. at the urem instruction. This will generate a udiv in
77 /// the process, and Builder's insert point will be pointing at the udiv (if
78 /// present, i.e. not folded), ready to be expanded if the user wishes
79 static Value *generatedUnsignedRemainderCode(Value *Dividend, Value *Divisor,
80 IRBuilder<> &Builder) {
81 // Remainder = Dividend - Quotient*Divisor
83 // Following instructions are generated for both i32 and i64
85 // ; %quotient = udiv i32 %dividend, %divisor
86 // ; %product = mul i32 %divisor, %quotient
87 // ; %remainder = sub i32 %dividend, %product
88 Value *Quotient = Builder.CreateUDiv(Dividend, Divisor);
89 Value *Product = Builder.CreateMul(Divisor, Quotient);
90 Value *Remainder = Builder.CreateSub(Dividend, Product);
92 if (Instruction *UDiv = dyn_cast<Instruction>(Quotient))
93 Builder.SetInsertPoint(UDiv);
95 return Remainder;
98 /// Generate code to divide two signed integers. Returns the quotient, rounded
99 /// towards 0. Builder's insert point should be pointing where the caller wants
100 /// code generated, e.g. at the sdiv instruction. This will generate a udiv in
101 /// the process, and Builder's insert point will be pointing at the udiv (if
102 /// present, i.e. not folded), ready to be expanded if the user wishes.
103 static Value *generateSignedDivisionCode(Value *Dividend, Value *Divisor,
104 IRBuilder<> &Builder) {
105 // Implementation taken from compiler-rt's __divsi3 and __divdi3
107 unsigned BitWidth = Dividend->getType()->getIntegerBitWidth();
108 ConstantInt *Shift;
110 if (BitWidth == 64) {
111 Shift = Builder.getInt64(63);
112 } else {
113 assert(BitWidth == 32 && "Unexpected bit width");
114 Shift = Builder.getInt32(31);
117 // Following instructions are generated for both i32 (shift 31) and
118 // i64 (shift 63).
120 // ; %tmp = ashr i32 %dividend, 31
121 // ; %tmp1 = ashr i32 %divisor, 31
122 // ; %tmp2 = xor i32 %tmp, %dividend
123 // ; %u_dvnd = sub nsw i32 %tmp2, %tmp
124 // ; %tmp3 = xor i32 %tmp1, %divisor
125 // ; %u_dvsr = sub nsw i32 %tmp3, %tmp1
126 // ; %q_sgn = xor i32 %tmp1, %tmp
127 // ; %q_mag = udiv i32 %u_dvnd, %u_dvsr
128 // ; %tmp4 = xor i32 %q_mag, %q_sgn
129 // ; %q = sub i32 %tmp4, %q_sgn
130 Value *Tmp = Builder.CreateAShr(Dividend, Shift);
131 Value *Tmp1 = Builder.CreateAShr(Divisor, Shift);
132 Value *Tmp2 = Builder.CreateXor(Tmp, Dividend);
133 Value *U_Dvnd = Builder.CreateSub(Tmp2, Tmp);
134 Value *Tmp3 = Builder.CreateXor(Tmp1, Divisor);
135 Value *U_Dvsr = Builder.CreateSub(Tmp3, Tmp1);
136 Value *Q_Sgn = Builder.CreateXor(Tmp1, Tmp);
137 Value *Q_Mag = Builder.CreateUDiv(U_Dvnd, U_Dvsr);
138 Value *Tmp4 = Builder.CreateXor(Q_Mag, Q_Sgn);
139 Value *Q = Builder.CreateSub(Tmp4, Q_Sgn);
141 if (Instruction *UDiv = dyn_cast<Instruction>(Q_Mag))
142 Builder.SetInsertPoint(UDiv);
144 return Q;
147 /// Generates code to divide two unsigned scalar 32-bit or 64-bit integers.
148 /// Returns the quotient, rounded towards 0. Builder's insert point should
149 /// point where the caller wants code generated, e.g. at the udiv instruction.
150 static Value *generateUnsignedDivisionCode(Value *Dividend, Value *Divisor,
151 IRBuilder<> &Builder) {
152 // The basic algorithm can be found in the compiler-rt project's
153 // implementation of __udivsi3.c. Here, we do a lower-level IR based approach
154 // that's been hand-tuned to lessen the amount of control flow involved.
156 // Some helper values
157 IntegerType *DivTy = cast<IntegerType>(Dividend->getType());
158 unsigned BitWidth = DivTy->getBitWidth();
160 ConstantInt *Zero;
161 ConstantInt *One;
162 ConstantInt *NegOne;
163 ConstantInt *MSB;
165 if (BitWidth == 64) {
166 Zero = Builder.getInt64(0);
167 One = Builder.getInt64(1);
168 NegOne = ConstantInt::getSigned(DivTy, -1);
169 MSB = Builder.getInt64(63);
170 } else {
171 assert(BitWidth == 32 && "Unexpected bit width");
172 Zero = Builder.getInt32(0);
173 One = Builder.getInt32(1);
174 NegOne = ConstantInt::getSigned(DivTy, -1);
175 MSB = Builder.getInt32(31);
178 ConstantInt *True = Builder.getTrue();
180 BasicBlock *IBB = Builder.GetInsertBlock();
181 Function *F = IBB->getParent();
182 Function *CTLZ = Intrinsic::getDeclaration(F->getParent(), Intrinsic::ctlz,
183 DivTy);
185 // Our CFG is going to look like:
186 // +---------------------+
187 // | special-cases |
188 // | ... |
189 // +---------------------+
190 // | |
191 // | +----------+
192 // | | bb1 |
193 // | | ... |
194 // | +----------+
195 // | | |
196 // | | +------------+
197 // | | | preheader |
198 // | | | ... |
199 // | | +------------+
200 // | | |
201 // | | | +---+
202 // | | | | |
203 // | | +------------+ |
204 // | | | do-while | |
205 // | | | ... | |
206 // | | +------------+ |
207 // | | | | |
208 // | +-----------+ +---+
209 // | | loop-exit |
210 // | | ... |
211 // | +-----------+
212 // | |
213 // +-------+
214 // | ... |
215 // | end |
216 // +-------+
217 BasicBlock *SpecialCases = Builder.GetInsertBlock();
218 SpecialCases->setName(Twine(SpecialCases->getName(), "_udiv-special-cases"));
219 BasicBlock *End = SpecialCases->splitBasicBlock(Builder.GetInsertPoint(),
220 "udiv-end");
221 BasicBlock *LoopExit = BasicBlock::Create(Builder.getContext(),
222 "udiv-loop-exit", F, End);
223 BasicBlock *DoWhile = BasicBlock::Create(Builder.getContext(),
224 "udiv-do-while", F, End);
225 BasicBlock *Preheader = BasicBlock::Create(Builder.getContext(),
226 "udiv-preheader", F, End);
227 BasicBlock *BB1 = BasicBlock::Create(Builder.getContext(),
228 "udiv-bb1", F, End);
230 // We'll be overwriting the terminator to insert our extra blocks
231 SpecialCases->getTerminator()->eraseFromParent();
233 // Same instructions are generated for both i32 (msb 31) and i64 (msb 63).
235 // First off, check for special cases: dividend or divisor is zero, divisor
236 // is greater than dividend, and divisor is 1.
237 // ; special-cases:
238 // ; %ret0_1 = icmp eq i32 %divisor, 0
239 // ; %ret0_2 = icmp eq i32 %dividend, 0
240 // ; %ret0_3 = or i1 %ret0_1, %ret0_2
241 // ; %tmp0 = tail call i32 @llvm.ctlz.i32(i32 %divisor, i1 true)
242 // ; %tmp1 = tail call i32 @llvm.ctlz.i32(i32 %dividend, i1 true)
243 // ; %sr = sub nsw i32 %tmp0, %tmp1
244 // ; %ret0_4 = icmp ugt i32 %sr, 31
245 // ; %ret0 = or i1 %ret0_3, %ret0_4
246 // ; %retDividend = icmp eq i32 %sr, 31
247 // ; %retVal = select i1 %ret0, i32 0, i32 %dividend
248 // ; %earlyRet = or i1 %ret0, %retDividend
249 // ; br i1 %earlyRet, label %end, label %bb1
250 Builder.SetInsertPoint(SpecialCases);
251 Value *Ret0_1 = Builder.CreateICmpEQ(Divisor, Zero);
252 Value *Ret0_2 = Builder.CreateICmpEQ(Dividend, Zero);
253 Value *Ret0_3 = Builder.CreateOr(Ret0_1, Ret0_2);
254 Value *Tmp0 = Builder.CreateCall(CTLZ, {Divisor, True});
255 Value *Tmp1 = Builder.CreateCall(CTLZ, {Dividend, True});
256 Value *SR = Builder.CreateSub(Tmp0, Tmp1);
257 Value *Ret0_4 = Builder.CreateICmpUGT(SR, MSB);
258 Value *Ret0 = Builder.CreateOr(Ret0_3, Ret0_4);
259 Value *RetDividend = Builder.CreateICmpEQ(SR, MSB);
260 Value *RetVal = Builder.CreateSelect(Ret0, Zero, Dividend);
261 Value *EarlyRet = Builder.CreateOr(Ret0, RetDividend);
262 Builder.CreateCondBr(EarlyRet, End, BB1);
264 // ; bb1: ; preds = %special-cases
265 // ; %sr_1 = add i32 %sr, 1
266 // ; %tmp2 = sub i32 31, %sr
267 // ; %q = shl i32 %dividend, %tmp2
268 // ; %skipLoop = icmp eq i32 %sr_1, 0
269 // ; br i1 %skipLoop, label %loop-exit, label %preheader
270 Builder.SetInsertPoint(BB1);
271 Value *SR_1 = Builder.CreateAdd(SR, One);
272 Value *Tmp2 = Builder.CreateSub(MSB, SR);
273 Value *Q = Builder.CreateShl(Dividend, Tmp2);
274 Value *SkipLoop = Builder.CreateICmpEQ(SR_1, Zero);
275 Builder.CreateCondBr(SkipLoop, LoopExit, Preheader);
277 // ; preheader: ; preds = %bb1
278 // ; %tmp3 = lshr i32 %dividend, %sr_1
279 // ; %tmp4 = add i32 %divisor, -1
280 // ; br label %do-while
281 Builder.SetInsertPoint(Preheader);
282 Value *Tmp3 = Builder.CreateLShr(Dividend, SR_1);
283 Value *Tmp4 = Builder.CreateAdd(Divisor, NegOne);
284 Builder.CreateBr(DoWhile);
286 // ; do-while: ; preds = %do-while, %preheader
287 // ; %carry_1 = phi i32 [ 0, %preheader ], [ %carry, %do-while ]
288 // ; %sr_3 = phi i32 [ %sr_1, %preheader ], [ %sr_2, %do-while ]
289 // ; %r_1 = phi i32 [ %tmp3, %preheader ], [ %r, %do-while ]
290 // ; %q_2 = phi i32 [ %q, %preheader ], [ %q_1, %do-while ]
291 // ; %tmp5 = shl i32 %r_1, 1
292 // ; %tmp6 = lshr i32 %q_2, 31
293 // ; %tmp7 = or i32 %tmp5, %tmp6
294 // ; %tmp8 = shl i32 %q_2, 1
295 // ; %q_1 = or i32 %carry_1, %tmp8
296 // ; %tmp9 = sub i32 %tmp4, %tmp7
297 // ; %tmp10 = ashr i32 %tmp9, 31
298 // ; %carry = and i32 %tmp10, 1
299 // ; %tmp11 = and i32 %tmp10, %divisor
300 // ; %r = sub i32 %tmp7, %tmp11
301 // ; %sr_2 = add i32 %sr_3, -1
302 // ; %tmp12 = icmp eq i32 %sr_2, 0
303 // ; br i1 %tmp12, label %loop-exit, label %do-while
304 Builder.SetInsertPoint(DoWhile);
305 PHINode *Carry_1 = Builder.CreatePHI(DivTy, 2);
306 PHINode *SR_3 = Builder.CreatePHI(DivTy, 2);
307 PHINode *R_1 = Builder.CreatePHI(DivTy, 2);
308 PHINode *Q_2 = Builder.CreatePHI(DivTy, 2);
309 Value *Tmp5 = Builder.CreateShl(R_1, One);
310 Value *Tmp6 = Builder.CreateLShr(Q_2, MSB);
311 Value *Tmp7 = Builder.CreateOr(Tmp5, Tmp6);
312 Value *Tmp8 = Builder.CreateShl(Q_2, One);
313 Value *Q_1 = Builder.CreateOr(Carry_1, Tmp8);
314 Value *Tmp9 = Builder.CreateSub(Tmp4, Tmp7);
315 Value *Tmp10 = Builder.CreateAShr(Tmp9, MSB);
316 Value *Carry = Builder.CreateAnd(Tmp10, One);
317 Value *Tmp11 = Builder.CreateAnd(Tmp10, Divisor);
318 Value *R = Builder.CreateSub(Tmp7, Tmp11);
319 Value *SR_2 = Builder.CreateAdd(SR_3, NegOne);
320 Value *Tmp12 = Builder.CreateICmpEQ(SR_2, Zero);
321 Builder.CreateCondBr(Tmp12, LoopExit, DoWhile);
323 // ; loop-exit: ; preds = %do-while, %bb1
324 // ; %carry_2 = phi i32 [ 0, %bb1 ], [ %carry, %do-while ]
325 // ; %q_3 = phi i32 [ %q, %bb1 ], [ %q_1, %do-while ]
326 // ; %tmp13 = shl i32 %q_3, 1
327 // ; %q_4 = or i32 %carry_2, %tmp13
328 // ; br label %end
329 Builder.SetInsertPoint(LoopExit);
330 PHINode *Carry_2 = Builder.CreatePHI(DivTy, 2);
331 PHINode *Q_3 = Builder.CreatePHI(DivTy, 2);
332 Value *Tmp13 = Builder.CreateShl(Q_3, One);
333 Value *Q_4 = Builder.CreateOr(Carry_2, Tmp13);
334 Builder.CreateBr(End);
336 // ; end: ; preds = %loop-exit, %special-cases
337 // ; %q_5 = phi i32 [ %q_4, %loop-exit ], [ %retVal, %special-cases ]
338 // ; ret i32 %q_5
339 Builder.SetInsertPoint(End, End->begin());
340 PHINode *Q_5 = Builder.CreatePHI(DivTy, 2);
342 // Populate the Phis, since all values have now been created. Our Phis were:
343 // ; %carry_1 = phi i32 [ 0, %preheader ], [ %carry, %do-while ]
344 Carry_1->addIncoming(Zero, Preheader);
345 Carry_1->addIncoming(Carry, DoWhile);
346 // ; %sr_3 = phi i32 [ %sr_1, %preheader ], [ %sr_2, %do-while ]
347 SR_3->addIncoming(SR_1, Preheader);
348 SR_3->addIncoming(SR_2, DoWhile);
349 // ; %r_1 = phi i32 [ %tmp3, %preheader ], [ %r, %do-while ]
350 R_1->addIncoming(Tmp3, Preheader);
351 R_1->addIncoming(R, DoWhile);
352 // ; %q_2 = phi i32 [ %q, %preheader ], [ %q_1, %do-while ]
353 Q_2->addIncoming(Q, Preheader);
354 Q_2->addIncoming(Q_1, DoWhile);
355 // ; %carry_2 = phi i32 [ 0, %bb1 ], [ %carry, %do-while ]
356 Carry_2->addIncoming(Zero, BB1);
357 Carry_2->addIncoming(Carry, DoWhile);
358 // ; %q_3 = phi i32 [ %q, %bb1 ], [ %q_1, %do-while ]
359 Q_3->addIncoming(Q, BB1);
360 Q_3->addIncoming(Q_1, DoWhile);
361 // ; %q_5 = phi i32 [ %q_4, %loop-exit ], [ %retVal, %special-cases ]
362 Q_5->addIncoming(Q_4, LoopExit);
363 Q_5->addIncoming(RetVal, SpecialCases);
365 return Q_5;
368 /// Generate code to calculate the remainder of two integers, replacing Rem with
369 /// the generated code. This currently generates code using the udiv expansion,
370 /// but future work includes generating more specialized code, e.g. when more
371 /// information about the operands are known. Implements both 32bit and 64bit
372 /// scalar division.
374 /// Replace Rem with generated code.
375 bool llvm::expandRemainder(BinaryOperator *Rem) {
376 assert((Rem->getOpcode() == Instruction::SRem ||
377 Rem->getOpcode() == Instruction::URem) &&
378 "Trying to expand remainder from a non-remainder function");
380 IRBuilder<> Builder(Rem);
382 assert(!Rem->getType()->isVectorTy() && "Div over vectors not supported");
383 assert((Rem->getType()->getIntegerBitWidth() == 32 ||
384 Rem->getType()->getIntegerBitWidth() == 64) &&
385 "Div of bitwidth other than 32 or 64 not supported");
387 // First prepare the sign if it's a signed remainder
388 if (Rem->getOpcode() == Instruction::SRem) {
389 Value *Remainder = generateSignedRemainderCode(Rem->getOperand(0),
390 Rem->getOperand(1), Builder);
392 // Check whether this is the insert point while Rem is still valid.
393 bool IsInsertPoint = Rem->getIterator() == Builder.GetInsertPoint();
394 Rem->replaceAllUsesWith(Remainder);
395 Rem->dropAllReferences();
396 Rem->eraseFromParent();
398 // If we didn't actually generate an urem instruction, we're done
399 // This happens for example if the input were constant. In this case the
400 // Builder insertion point was unchanged
401 if (IsInsertPoint)
402 return true;
404 BinaryOperator *BO = dyn_cast<BinaryOperator>(Builder.GetInsertPoint());
405 Rem = BO;
408 Value *Remainder = generatedUnsignedRemainderCode(Rem->getOperand(0),
409 Rem->getOperand(1),
410 Builder);
412 Rem->replaceAllUsesWith(Remainder);
413 Rem->dropAllReferences();
414 Rem->eraseFromParent();
416 // Expand the udiv
417 if (BinaryOperator *UDiv = dyn_cast<BinaryOperator>(Builder.GetInsertPoint())) {
418 assert(UDiv->getOpcode() == Instruction::UDiv && "Non-udiv in expansion?");
419 expandDivision(UDiv);
422 return true;
426 /// Generate code to divide two integers, replacing Div with the generated
427 /// code. This currently generates code similarly to compiler-rt's
428 /// implementations, but future work includes generating more specialized code
429 /// when more information about the operands are known. Implements both
430 /// 32bit and 64bit scalar division.
432 /// Replace Div with generated code.
433 bool llvm::expandDivision(BinaryOperator *Div) {
434 assert((Div->getOpcode() == Instruction::SDiv ||
435 Div->getOpcode() == Instruction::UDiv) &&
436 "Trying to expand division from a non-division function");
438 IRBuilder<> Builder(Div);
440 assert(!Div->getType()->isVectorTy() && "Div over vectors not supported");
441 assert((Div->getType()->getIntegerBitWidth() == 32 ||
442 Div->getType()->getIntegerBitWidth() == 64) &&
443 "Div of bitwidth other than 32 or 64 not supported");
445 // First prepare the sign if it's a signed division
446 if (Div->getOpcode() == Instruction::SDiv) {
447 // Lower the code to unsigned division, and reset Div to point to the udiv.
448 Value *Quotient = generateSignedDivisionCode(Div->getOperand(0),
449 Div->getOperand(1), Builder);
451 // Check whether this is the insert point while Div is still valid.
452 bool IsInsertPoint = Div->getIterator() == Builder.GetInsertPoint();
453 Div->replaceAllUsesWith(Quotient);
454 Div->dropAllReferences();
455 Div->eraseFromParent();
457 // If we didn't actually generate an udiv instruction, we're done
458 // This happens for example if the input were constant. In this case the
459 // Builder insertion point was unchanged
460 if (IsInsertPoint)
461 return true;
463 BinaryOperator *BO = dyn_cast<BinaryOperator>(Builder.GetInsertPoint());
464 Div = BO;
467 // Insert the unsigned division code
468 Value *Quotient = generateUnsignedDivisionCode(Div->getOperand(0),
469 Div->getOperand(1),
470 Builder);
471 Div->replaceAllUsesWith(Quotient);
472 Div->dropAllReferences();
473 Div->eraseFromParent();
475 return true;
478 /// Generate code to compute the remainder of two integers of bitwidth up to
479 /// 32 bits. Uses the above routines and extends the inputs/truncates the
480 /// outputs to operate in 32 bits; that is, these routines are good for targets
481 /// that have no or very little suppport for smaller than 32 bit integer
482 /// arithmetic.
484 /// Replace Rem with emulation code.
485 bool llvm::expandRemainderUpTo32Bits(BinaryOperator *Rem) {
486 assert((Rem->getOpcode() == Instruction::SRem ||
487 Rem->getOpcode() == Instruction::URem) &&
488 "Trying to expand remainder from a non-remainder function");
490 Type *RemTy = Rem->getType();
491 assert(!RemTy->isVectorTy() && "Div over vectors not supported");
493 unsigned RemTyBitWidth = RemTy->getIntegerBitWidth();
495 assert(RemTyBitWidth <= 32 &&
496 "Div of bitwidth greater than 32 not supported");
498 if (RemTyBitWidth == 32)
499 return expandRemainder(Rem);
501 // If bitwidth smaller than 32 extend inputs, extend output and proceed
502 // with 32 bit division.
503 IRBuilder<> Builder(Rem);
505 Value *ExtDividend;
506 Value *ExtDivisor;
507 Value *ExtRem;
508 Value *Trunc;
509 Type *Int32Ty = Builder.getInt32Ty();
511 if (Rem->getOpcode() == Instruction::SRem) {
512 ExtDividend = Builder.CreateSExt(Rem->getOperand(0), Int32Ty);
513 ExtDivisor = Builder.CreateSExt(Rem->getOperand(1), Int32Ty);
514 ExtRem = Builder.CreateSRem(ExtDividend, ExtDivisor);
515 } else {
516 ExtDividend = Builder.CreateZExt(Rem->getOperand(0), Int32Ty);
517 ExtDivisor = Builder.CreateZExt(Rem->getOperand(1), Int32Ty);
518 ExtRem = Builder.CreateURem(ExtDividend, ExtDivisor);
520 Trunc = Builder.CreateTrunc(ExtRem, RemTy);
522 Rem->replaceAllUsesWith(Trunc);
523 Rem->dropAllReferences();
524 Rem->eraseFromParent();
526 return expandRemainder(cast<BinaryOperator>(ExtRem));
529 /// Generate code to compute the remainder of two integers of bitwidth up to
530 /// 64 bits. Uses the above routines and extends the inputs/truncates the
531 /// outputs to operate in 64 bits.
533 /// Replace Rem with emulation code.
534 bool llvm::expandRemainderUpTo64Bits(BinaryOperator *Rem) {
535 assert((Rem->getOpcode() == Instruction::SRem ||
536 Rem->getOpcode() == Instruction::URem) &&
537 "Trying to expand remainder from a non-remainder function");
539 Type *RemTy = Rem->getType();
540 assert(!RemTy->isVectorTy() && "Div over vectors not supported");
542 unsigned RemTyBitWidth = RemTy->getIntegerBitWidth();
544 assert(RemTyBitWidth <= 64 && "Div of bitwidth greater than 64 not supported");
546 if (RemTyBitWidth == 64)
547 return expandRemainder(Rem);
549 // If bitwidth smaller than 64 extend inputs, extend output and proceed
550 // with 64 bit division.
551 IRBuilder<> Builder(Rem);
553 Value *ExtDividend;
554 Value *ExtDivisor;
555 Value *ExtRem;
556 Value *Trunc;
557 Type *Int64Ty = Builder.getInt64Ty();
559 if (Rem->getOpcode() == Instruction::SRem) {
560 ExtDividend = Builder.CreateSExt(Rem->getOperand(0), Int64Ty);
561 ExtDivisor = Builder.CreateSExt(Rem->getOperand(1), Int64Ty);
562 ExtRem = Builder.CreateSRem(ExtDividend, ExtDivisor);
563 } else {
564 ExtDividend = Builder.CreateZExt(Rem->getOperand(0), Int64Ty);
565 ExtDivisor = Builder.CreateZExt(Rem->getOperand(1), Int64Ty);
566 ExtRem = Builder.CreateURem(ExtDividend, ExtDivisor);
568 Trunc = Builder.CreateTrunc(ExtRem, RemTy);
570 Rem->replaceAllUsesWith(Trunc);
571 Rem->dropAllReferences();
572 Rem->eraseFromParent();
574 return expandRemainder(cast<BinaryOperator>(ExtRem));
577 /// Generate code to divide two integers of bitwidth up to 32 bits. Uses the
578 /// above routines and extends the inputs/truncates the outputs to operate
579 /// in 32 bits; that is, these routines are good for targets that have no
580 /// or very little support for smaller than 32 bit integer arithmetic.
582 /// Replace Div with emulation code.
583 bool llvm::expandDivisionUpTo32Bits(BinaryOperator *Div) {
584 assert((Div->getOpcode() == Instruction::SDiv ||
585 Div->getOpcode() == Instruction::UDiv) &&
586 "Trying to expand division from a non-division function");
588 Type *DivTy = Div->getType();
589 assert(!DivTy->isVectorTy() && "Div over vectors not supported");
591 unsigned DivTyBitWidth = DivTy->getIntegerBitWidth();
593 assert(DivTyBitWidth <= 32 && "Div of bitwidth greater than 32 not supported");
595 if (DivTyBitWidth == 32)
596 return expandDivision(Div);
598 // If bitwidth smaller than 32 extend inputs, extend output and proceed
599 // with 32 bit division.
600 IRBuilder<> Builder(Div);
602 Value *ExtDividend;
603 Value *ExtDivisor;
604 Value *ExtDiv;
605 Value *Trunc;
606 Type *Int32Ty = Builder.getInt32Ty();
608 if (Div->getOpcode() == Instruction::SDiv) {
609 ExtDividend = Builder.CreateSExt(Div->getOperand(0), Int32Ty);
610 ExtDivisor = Builder.CreateSExt(Div->getOperand(1), Int32Ty);
611 ExtDiv = Builder.CreateSDiv(ExtDividend, ExtDivisor);
612 } else {
613 ExtDividend = Builder.CreateZExt(Div->getOperand(0), Int32Ty);
614 ExtDivisor = Builder.CreateZExt(Div->getOperand(1), Int32Ty);
615 ExtDiv = Builder.CreateUDiv(ExtDividend, ExtDivisor);
617 Trunc = Builder.CreateTrunc(ExtDiv, DivTy);
619 Div->replaceAllUsesWith(Trunc);
620 Div->dropAllReferences();
621 Div->eraseFromParent();
623 return expandDivision(cast<BinaryOperator>(ExtDiv));
626 /// Generate code to divide two integers of bitwidth up to 64 bits. Uses the
627 /// above routines and extends the inputs/truncates the outputs to operate
628 /// in 64 bits.
630 /// Replace Div with emulation code.
631 bool llvm::expandDivisionUpTo64Bits(BinaryOperator *Div) {
632 assert((Div->getOpcode() == Instruction::SDiv ||
633 Div->getOpcode() == Instruction::UDiv) &&
634 "Trying to expand division from a non-division function");
636 Type *DivTy = Div->getType();
637 assert(!DivTy->isVectorTy() && "Div over vectors not supported");
639 unsigned DivTyBitWidth = DivTy->getIntegerBitWidth();
641 assert(DivTyBitWidth <= 64 &&
642 "Div of bitwidth greater than 64 not supported");
644 if (DivTyBitWidth == 64)
645 return expandDivision(Div);
647 // If bitwidth smaller than 64 extend inputs, extend output and proceed
648 // with 64 bit division.
649 IRBuilder<> Builder(Div);
651 Value *ExtDividend;
652 Value *ExtDivisor;
653 Value *ExtDiv;
654 Value *Trunc;
655 Type *Int64Ty = Builder.getInt64Ty();
657 if (Div->getOpcode() == Instruction::SDiv) {
658 ExtDividend = Builder.CreateSExt(Div->getOperand(0), Int64Ty);
659 ExtDivisor = Builder.CreateSExt(Div->getOperand(1), Int64Ty);
660 ExtDiv = Builder.CreateSDiv(ExtDividend, ExtDivisor);
661 } else {
662 ExtDividend = Builder.CreateZExt(Div->getOperand(0), Int64Ty);
663 ExtDivisor = Builder.CreateZExt(Div->getOperand(1), Int64Ty);
664 ExtDiv = Builder.CreateUDiv(ExtDividend, ExtDivisor);
666 Trunc = Builder.CreateTrunc(ExtDiv, DivTy);
668 Div->replaceAllUsesWith(Trunc);
669 Div->dropAllReferences();
670 Div->eraseFromParent();
672 return expandDivision(cast<BinaryOperator>(ExtDiv));