1 | //===- InstCombineMulDivRem.cpp -------------------------------------------===// |
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 visit functions for mul, fmul, sdiv, udiv, fdiv, |
10 | // srem, urem, frem. |
11 | // |
12 | //===----------------------------------------------------------------------===// |
13 | |
14 | #include "InstCombineInternal.h" |
15 | #include "llvm/ADT/APInt.h" |
16 | #include "llvm/ADT/SmallVector.h" |
17 | #include "llvm/Analysis/InstructionSimplify.h" |
18 | #include "llvm/Analysis/ValueTracking.h" |
19 | #include "llvm/IR/BasicBlock.h" |
20 | #include "llvm/IR/Constant.h" |
21 | #include "llvm/IR/Constants.h" |
22 | #include "llvm/IR/InstrTypes.h" |
23 | #include "llvm/IR/Instruction.h" |
24 | #include "llvm/IR/Instructions.h" |
25 | #include "llvm/IR/IntrinsicInst.h" |
26 | #include "llvm/IR/Intrinsics.h" |
27 | #include "llvm/IR/Operator.h" |
28 | #include "llvm/IR/PatternMatch.h" |
29 | #include "llvm/IR/Type.h" |
30 | #include "llvm/IR/Value.h" |
31 | #include "llvm/Support/Casting.h" |
32 | #include "llvm/Support/ErrorHandling.h" |
33 | #include "llvm/Transforms/InstCombine/InstCombiner.h" |
34 | #include "llvm/Transforms/Utils/BuildLibCalls.h" |
35 | #include <cassert> |
36 | |
37 | #define DEBUG_TYPE "instcombine" |
38 | #include "llvm/Transforms/Utils/InstructionWorklist.h" |
39 | |
40 | using namespace llvm; |
41 | using namespace PatternMatch; |
42 | |
43 | /// The specific integer value is used in a context where it is known to be |
44 | /// non-zero. If this allows us to simplify the computation, do so and return |
45 | /// the new operand, otherwise return null. |
46 | static Value *simplifyValueKnownNonZero(Value *V, InstCombinerImpl &IC, |
47 | Instruction &CxtI) { |
48 | // If V has multiple uses, then we would have to do more analysis to determine |
49 | // if this is safe. For example, the use could be in dynamically unreached |
50 | // code. |
51 | if (!V->hasOneUse()) return nullptr; |
52 | |
53 | bool MadeChange = false; |
54 | |
55 | // ((1 << A) >>u B) --> (1 << (A-B)) |
56 | // Because V cannot be zero, we know that B is less than A. |
57 | Value *A = nullptr, *B = nullptr, *One = nullptr; |
58 | if (match(V, P: m_LShr(L: m_OneUse(SubPattern: m_Shl(L: m_Value(V&: One), R: m_Value(V&: A))), R: m_Value(V&: B))) && |
59 | match(V: One, P: m_One())) { |
60 | A = IC.Builder.CreateSub(LHS: A, RHS: B); |
61 | return IC.Builder.CreateShl(LHS: One, RHS: A); |
62 | } |
63 | |
64 | // (PowerOfTwo >>u B) --> isExact since shifting out the result would make it |
65 | // inexact. Similarly for <<. |
66 | BinaryOperator *I = dyn_cast<BinaryOperator>(Val: V); |
67 | if (I && I->isLogicalShift() && |
68 | IC.isKnownToBeAPowerOfTwo(V: I->getOperand(i_nocapture: 0), OrZero: false, Depth: 0, CxtI: &CxtI)) { |
69 | // We know that this is an exact/nuw shift and that the input is a |
70 | // non-zero context as well. |
71 | if (Value *V2 = simplifyValueKnownNonZero(V: I->getOperand(i_nocapture: 0), IC, CxtI)) { |
72 | IC.replaceOperand(I&: *I, OpNum: 0, V: V2); |
73 | MadeChange = true; |
74 | } |
75 | |
76 | if (I->getOpcode() == Instruction::LShr && !I->isExact()) { |
77 | I->setIsExact(); |
78 | MadeChange = true; |
79 | } |
80 | |
81 | if (I->getOpcode() == Instruction::Shl && !I->hasNoUnsignedWrap()) { |
82 | I->setHasNoUnsignedWrap(); |
83 | MadeChange = true; |
84 | } |
85 | } |
86 | |
87 | // TODO: Lots more we could do here: |
88 | // If V is a phi node, we can call this on each of its operands. |
89 | // "select cond, X, 0" can simplify to "X". |
90 | |
91 | return MadeChange ? V : nullptr; |
92 | } |
93 | |
94 | // TODO: This is a specific form of a much more general pattern. |
95 | // We could detect a select with any binop identity constant, or we |
96 | // could use SimplifyBinOp to see if either arm of the select reduces. |
97 | // But that needs to be done carefully and/or while removing potential |
98 | // reverse canonicalizations as in InstCombiner::foldSelectIntoOp(). |
99 | static Value *foldMulSelectToNegate(BinaryOperator &I, |
100 | InstCombiner::BuilderTy &Builder) { |
101 | Value *Cond, *OtherOp; |
102 | |
103 | // mul (select Cond, 1, -1), OtherOp --> select Cond, OtherOp, -OtherOp |
104 | // mul OtherOp, (select Cond, 1, -1) --> select Cond, OtherOp, -OtherOp |
105 | if (match(V: &I, P: m_c_Mul(L: m_OneUse(SubPattern: m_Select(C: m_Value(V&: Cond), L: m_One(), R: m_AllOnes())), |
106 | R: m_Value(V&: OtherOp)))) { |
107 | bool HasAnyNoWrap = I.hasNoSignedWrap() || I.hasNoUnsignedWrap(); |
108 | Value *Neg = Builder.CreateNeg(V: OtherOp, Name: "" , HasNSW: HasAnyNoWrap); |
109 | return Builder.CreateSelect(C: Cond, True: OtherOp, False: Neg); |
110 | } |
111 | // mul (select Cond, -1, 1), OtherOp --> select Cond, -OtherOp, OtherOp |
112 | // mul OtherOp, (select Cond, -1, 1) --> select Cond, -OtherOp, OtherOp |
113 | if (match(V: &I, P: m_c_Mul(L: m_OneUse(SubPattern: m_Select(C: m_Value(V&: Cond), L: m_AllOnes(), R: m_One())), |
114 | R: m_Value(V&: OtherOp)))) { |
115 | bool HasAnyNoWrap = I.hasNoSignedWrap() || I.hasNoUnsignedWrap(); |
116 | Value *Neg = Builder.CreateNeg(V: OtherOp, Name: "" , HasNSW: HasAnyNoWrap); |
117 | return Builder.CreateSelect(C: Cond, True: Neg, False: OtherOp); |
118 | } |
119 | |
120 | // fmul (select Cond, 1.0, -1.0), OtherOp --> select Cond, OtherOp, -OtherOp |
121 | // fmul OtherOp, (select Cond, 1.0, -1.0) --> select Cond, OtherOp, -OtherOp |
122 | if (match(V: &I, P: m_c_FMul(L: m_OneUse(SubPattern: m_Select(C: m_Value(V&: Cond), L: m_SpecificFP(V: 1.0), |
123 | R: m_SpecificFP(V: -1.0))), |
124 | R: m_Value(V&: OtherOp)))) { |
125 | IRBuilder<>::FastMathFlagGuard FMFGuard(Builder); |
126 | Builder.setFastMathFlags(I.getFastMathFlags()); |
127 | return Builder.CreateSelect(C: Cond, True: OtherOp, False: Builder.CreateFNeg(V: OtherOp)); |
128 | } |
129 | |
130 | // fmul (select Cond, -1.0, 1.0), OtherOp --> select Cond, -OtherOp, OtherOp |
131 | // fmul OtherOp, (select Cond, -1.0, 1.0) --> select Cond, -OtherOp, OtherOp |
132 | if (match(V: &I, P: m_c_FMul(L: m_OneUse(SubPattern: m_Select(C: m_Value(V&: Cond), L: m_SpecificFP(V: -1.0), |
133 | R: m_SpecificFP(V: 1.0))), |
134 | R: m_Value(V&: OtherOp)))) { |
135 | IRBuilder<>::FastMathFlagGuard FMFGuard(Builder); |
136 | Builder.setFastMathFlags(I.getFastMathFlags()); |
137 | return Builder.CreateSelect(C: Cond, True: Builder.CreateFNeg(V: OtherOp), False: OtherOp); |
138 | } |
139 | |
140 | return nullptr; |
141 | } |
142 | |
143 | /// Reduce integer multiplication patterns that contain a (+/-1 << Z) factor. |
144 | /// Callers are expected to call this twice to handle commuted patterns. |
145 | static Value *foldMulShl1(BinaryOperator &Mul, bool CommuteOperands, |
146 | InstCombiner::BuilderTy &Builder) { |
147 | Value *X = Mul.getOperand(i_nocapture: 0), *Y = Mul.getOperand(i_nocapture: 1); |
148 | if (CommuteOperands) |
149 | std::swap(a&: X, b&: Y); |
150 | |
151 | const bool HasNSW = Mul.hasNoSignedWrap(); |
152 | const bool HasNUW = Mul.hasNoUnsignedWrap(); |
153 | |
154 | // X * (1 << Z) --> X << Z |
155 | Value *Z; |
156 | if (match(V: Y, P: m_Shl(L: m_One(), R: m_Value(V&: Z)))) { |
157 | bool PropagateNSW = HasNSW && cast<ShlOperator>(Val: Y)->hasNoSignedWrap(); |
158 | return Builder.CreateShl(LHS: X, RHS: Z, Name: Mul.getName(), HasNUW, HasNSW: PropagateNSW); |
159 | } |
160 | |
161 | // Similar to above, but an increment of the shifted value becomes an add: |
162 | // X * ((1 << Z) + 1) --> (X * (1 << Z)) + X --> (X << Z) + X |
163 | // This increases uses of X, so it may require a freeze, but that is still |
164 | // expected to be an improvement because it removes the multiply. |
165 | BinaryOperator *Shift; |
166 | if (match(V: Y, P: m_OneUse(SubPattern: m_Add(L: m_BinOp(I&: Shift), R: m_One()))) && |
167 | match(V: Shift, P: m_OneUse(SubPattern: m_Shl(L: m_One(), R: m_Value(V&: Z))))) { |
168 | bool PropagateNSW = HasNSW && Shift->hasNoSignedWrap(); |
169 | Value *FrX = X; |
170 | if (!isGuaranteedNotToBeUndef(V: X)) |
171 | FrX = Builder.CreateFreeze(V: X, Name: X->getName() + ".fr" ); |
172 | Value *Shl = Builder.CreateShl(LHS: FrX, RHS: Z, Name: "mulshl" , HasNUW, HasNSW: PropagateNSW); |
173 | return Builder.CreateAdd(LHS: Shl, RHS: FrX, Name: Mul.getName(), HasNUW, HasNSW: PropagateNSW); |
174 | } |
175 | |
176 | // Similar to above, but a decrement of the shifted value is disguised as |
177 | // 'not' and becomes a sub: |
178 | // X * (~(-1 << Z)) --> X * ((1 << Z) - 1) --> (X << Z) - X |
179 | // This increases uses of X, so it may require a freeze, but that is still |
180 | // expected to be an improvement because it removes the multiply. |
181 | if (match(V: Y, P: m_OneUse(SubPattern: m_Not(V: m_OneUse(SubPattern: m_Shl(L: m_AllOnes(), R: m_Value(V&: Z))))))) { |
182 | Value *FrX = X; |
183 | if (!isGuaranteedNotToBeUndef(V: X)) |
184 | FrX = Builder.CreateFreeze(V: X, Name: X->getName() + ".fr" ); |
185 | Value *Shl = Builder.CreateShl(LHS: FrX, RHS: Z, Name: "mulshl" ); |
186 | return Builder.CreateSub(LHS: Shl, RHS: FrX, Name: Mul.getName()); |
187 | } |
188 | |
189 | return nullptr; |
190 | } |
191 | |
192 | static Value *takeLog2(IRBuilderBase &Builder, Value *Op, unsigned Depth, |
193 | bool AssumeNonZero, bool DoFold); |
194 | |
195 | Instruction *InstCombinerImpl::visitMul(BinaryOperator &I) { |
196 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
197 | if (Value *V = |
198 | simplifyMulInst(LHS: Op0, RHS: Op1, IsNSW: I.hasNoSignedWrap(), IsNUW: I.hasNoUnsignedWrap(), |
199 | Q: SQ.getWithInstruction(I: &I))) |
200 | return replaceInstUsesWith(I, V); |
201 | |
202 | if (SimplifyAssociativeOrCommutative(I)) |
203 | return &I; |
204 | |
205 | if (Instruction *X = foldVectorBinop(Inst&: I)) |
206 | return X; |
207 | |
208 | if (Instruction *Phi = foldBinopWithPhiOperands(BO&: I)) |
209 | return Phi; |
210 | |
211 | if (Value *V = foldUsingDistributiveLaws(I)) |
212 | return replaceInstUsesWith(I, V); |
213 | |
214 | Type *Ty = I.getType(); |
215 | const unsigned BitWidth = Ty->getScalarSizeInBits(); |
216 | const bool HasNSW = I.hasNoSignedWrap(); |
217 | const bool HasNUW = I.hasNoUnsignedWrap(); |
218 | |
219 | // X * -1 --> 0 - X |
220 | if (match(V: Op1, P: m_AllOnes())) { |
221 | return HasNSW ? BinaryOperator::CreateNSWNeg(Op: Op0) |
222 | : BinaryOperator::CreateNeg(Op: Op0); |
223 | } |
224 | |
225 | // Also allow combining multiply instructions on vectors. |
226 | { |
227 | Value *NewOp; |
228 | Constant *C1, *C2; |
229 | const APInt *IVal; |
230 | if (match(V: &I, P: m_Mul(L: m_Shl(L: m_Value(V&: NewOp), R: m_ImmConstant(C&: C2)), |
231 | R: m_ImmConstant(C&: C1))) && |
232 | match(V: C1, P: m_APInt(Res&: IVal))) { |
233 | // ((X << C2)*C1) == (X * (C1 << C2)) |
234 | Constant *Shl = |
235 | ConstantFoldBinaryOpOperands(Opcode: Instruction::Shl, LHS: C1, RHS: C2, DL); |
236 | assert(Shl && "Constant folding of immediate constants failed" ); |
237 | BinaryOperator *Mul = cast<BinaryOperator>(Val: I.getOperand(i_nocapture: 0)); |
238 | BinaryOperator *BO = BinaryOperator::CreateMul(V1: NewOp, V2: Shl); |
239 | if (HasNUW && Mul->hasNoUnsignedWrap()) |
240 | BO->setHasNoUnsignedWrap(); |
241 | if (HasNSW && Mul->hasNoSignedWrap() && Shl->isNotMinSignedValue()) |
242 | BO->setHasNoSignedWrap(); |
243 | return BO; |
244 | } |
245 | |
246 | if (match(V: &I, P: m_Mul(L: m_Value(V&: NewOp), R: m_Constant(C&: C1)))) { |
247 | // Replace X*(2^C) with X << C, where C is either a scalar or a vector. |
248 | if (Constant *NewCst = ConstantExpr::getExactLogBase2(C: C1)) { |
249 | BinaryOperator *Shl = BinaryOperator::CreateShl(V1: NewOp, V2: NewCst); |
250 | |
251 | if (HasNUW) |
252 | Shl->setHasNoUnsignedWrap(); |
253 | if (HasNSW) { |
254 | const APInt *V; |
255 | if (match(V: NewCst, P: m_APInt(Res&: V)) && *V != V->getBitWidth() - 1) |
256 | Shl->setHasNoSignedWrap(); |
257 | } |
258 | |
259 | return Shl; |
260 | } |
261 | } |
262 | } |
263 | |
264 | if (Op0->hasOneUse() && match(V: Op1, P: m_NegatedPower2())) { |
265 | // Interpret X * (-1<<C) as (-X) * (1<<C) and try to sink the negation. |
266 | // The "* (1<<C)" thus becomes a potential shifting opportunity. |
267 | if (Value *NegOp0 = |
268 | Negator::Negate(/*IsNegation*/ LHSIsZero: true, IsNSW: HasNSW, Root: Op0, IC&: *this)) { |
269 | auto *Op1C = cast<Constant>(Val: Op1); |
270 | return replaceInstUsesWith( |
271 | I, V: Builder.CreateMul(LHS: NegOp0, RHS: ConstantExpr::getNeg(C: Op1C), Name: "" , |
272 | /* HasNUW */ false, |
273 | HasNSW: HasNSW && Op1C->isNotMinSignedValue())); |
274 | } |
275 | |
276 | // Try to convert multiply of extended operand to narrow negate and shift |
277 | // for better analysis. |
278 | // This is valid if the shift amount (trailing zeros in the multiplier |
279 | // constant) clears more high bits than the bitwidth difference between |
280 | // source and destination types: |
281 | // ({z/s}ext X) * (-1<<C) --> (zext (-X)) << C |
282 | const APInt *NegPow2C; |
283 | Value *X; |
284 | if (match(V: Op0, P: m_ZExtOrSExt(Op: m_Value(V&: X))) && |
285 | match(V: Op1, P: m_APIntAllowPoison(Res&: NegPow2C))) { |
286 | unsigned SrcWidth = X->getType()->getScalarSizeInBits(); |
287 | unsigned ShiftAmt = NegPow2C->countr_zero(); |
288 | if (ShiftAmt >= BitWidth - SrcWidth) { |
289 | Value *N = Builder.CreateNeg(V: X, Name: X->getName() + ".neg" ); |
290 | Value *Z = Builder.CreateZExt(V: N, DestTy: Ty, Name: N->getName() + ".z" ); |
291 | return BinaryOperator::CreateShl(V1: Z, V2: ConstantInt::get(Ty, V: ShiftAmt)); |
292 | } |
293 | } |
294 | } |
295 | |
296 | if (Instruction *FoldedMul = foldBinOpIntoSelectOrPhi(I)) |
297 | return FoldedMul; |
298 | |
299 | if (Value *FoldedMul = foldMulSelectToNegate(I, Builder)) |
300 | return replaceInstUsesWith(I, V: FoldedMul); |
301 | |
302 | // Simplify mul instructions with a constant RHS. |
303 | Constant *MulC; |
304 | if (match(V: Op1, P: m_ImmConstant(C&: MulC))) { |
305 | // Canonicalize (X+C1)*MulC -> X*MulC+C1*MulC. |
306 | // Canonicalize (X|C1)*MulC -> X*MulC+C1*MulC. |
307 | Value *X; |
308 | Constant *C1; |
309 | if (match(V: Op0, P: m_OneUse(SubPattern: m_AddLike(L: m_Value(V&: X), R: m_ImmConstant(C&: C1))))) { |
310 | // C1*MulC simplifies to a tidier constant. |
311 | Value *NewC = Builder.CreateMul(LHS: C1, RHS: MulC); |
312 | auto *BOp0 = cast<BinaryOperator>(Val: Op0); |
313 | bool Op0NUW = |
314 | (BOp0->getOpcode() == Instruction::Or || BOp0->hasNoUnsignedWrap()); |
315 | Value *NewMul = Builder.CreateMul(LHS: X, RHS: MulC); |
316 | auto *BO = BinaryOperator::CreateAdd(V1: NewMul, V2: NewC); |
317 | if (HasNUW && Op0NUW) { |
318 | // If NewMulBO is constant we also can set BO to nuw. |
319 | if (auto *NewMulBO = dyn_cast<BinaryOperator>(Val: NewMul)) |
320 | NewMulBO->setHasNoUnsignedWrap(); |
321 | BO->setHasNoUnsignedWrap(); |
322 | } |
323 | return BO; |
324 | } |
325 | } |
326 | |
327 | // abs(X) * abs(X) -> X * X |
328 | Value *X; |
329 | if (Op0 == Op1 && match(V: Op0, P: m_Intrinsic<Intrinsic::abs>(Op0: m_Value(V&: X)))) |
330 | return BinaryOperator::CreateMul(V1: X, V2: X); |
331 | |
332 | { |
333 | Value *Y; |
334 | // abs(X) * abs(Y) -> abs(X * Y) |
335 | if (I.hasNoSignedWrap() && |
336 | match(V: Op0, |
337 | P: m_OneUse(SubPattern: m_Intrinsic<Intrinsic::abs>(Op0: m_Value(V&: X), Op1: m_One()))) && |
338 | match(V: Op1, P: m_OneUse(SubPattern: m_Intrinsic<Intrinsic::abs>(Op0: m_Value(V&: Y), Op1: m_One())))) |
339 | return replaceInstUsesWith( |
340 | I, V: Builder.CreateBinaryIntrinsic(ID: Intrinsic::abs, |
341 | LHS: Builder.CreateNSWMul(LHS: X, RHS: Y), |
342 | RHS: Builder.getTrue())); |
343 | } |
344 | |
345 | // -X * C --> X * -C |
346 | Value *Y; |
347 | Constant *Op1C; |
348 | if (match(V: Op0, P: m_Neg(V: m_Value(V&: X))) && match(V: Op1, P: m_Constant(C&: Op1C))) |
349 | return BinaryOperator::CreateMul(V1: X, V2: ConstantExpr::getNeg(C: Op1C)); |
350 | |
351 | // -X * -Y --> X * Y |
352 | if (match(V: Op0, P: m_Neg(V: m_Value(V&: X))) && match(V: Op1, P: m_Neg(V: m_Value(V&: Y)))) { |
353 | auto *NewMul = BinaryOperator::CreateMul(V1: X, V2: Y); |
354 | if (HasNSW && cast<OverflowingBinaryOperator>(Val: Op0)->hasNoSignedWrap() && |
355 | cast<OverflowingBinaryOperator>(Val: Op1)->hasNoSignedWrap()) |
356 | NewMul->setHasNoSignedWrap(); |
357 | return NewMul; |
358 | } |
359 | |
360 | // -X * Y --> -(X * Y) |
361 | // X * -Y --> -(X * Y) |
362 | if (match(V: &I, P: m_c_Mul(L: m_OneUse(SubPattern: m_Neg(V: m_Value(V&: X))), R: m_Value(V&: Y)))) |
363 | return BinaryOperator::CreateNeg(Op: Builder.CreateMul(LHS: X, RHS: Y)); |
364 | |
365 | // (-X * Y) * -X --> (X * Y) * X |
366 | // (-X << Y) * -X --> (X << Y) * X |
367 | if (match(V: Op1, P: m_Neg(V: m_Value(V&: X)))) { |
368 | if (Value *NegOp0 = Negator::Negate(LHSIsZero: false, /*IsNSW*/ false, Root: Op0, IC&: *this)) |
369 | return BinaryOperator::CreateMul(V1: NegOp0, V2: X); |
370 | } |
371 | |
372 | if (Op0->hasOneUse()) { |
373 | // (mul (div exact X, C0), C1) |
374 | // -> (div exact X, C0 / C1) |
375 | // iff C0 % C1 == 0 and X / (C0 / C1) doesn't create UB. |
376 | const APInt *C1; |
377 | auto UDivCheck = [&C1](const APInt &C) { return C.urem(RHS: *C1).isZero(); }; |
378 | auto SDivCheck = [&C1](const APInt &C) { |
379 | APInt Quot, Rem; |
380 | APInt::sdivrem(LHS: C, RHS: *C1, Quotient&: Quot, Remainder&: Rem); |
381 | return Rem.isZero() && !Quot.isAllOnes(); |
382 | }; |
383 | if (match(V: Op1, P: m_APInt(Res&: C1)) && |
384 | (match(V: Op0, P: m_Exact(SubPattern: m_UDiv(L: m_Value(V&: X), R: m_CheckedInt(CheckFn: UDivCheck)))) || |
385 | match(V: Op0, P: m_Exact(SubPattern: m_SDiv(L: m_Value(V&: X), R: m_CheckedInt(CheckFn: SDivCheck)))))) { |
386 | auto BOpc = cast<BinaryOperator>(Val: Op0)->getOpcode(); |
387 | return BinaryOperator::CreateExact( |
388 | Opc: BOpc, V1: X, |
389 | V2: Builder.CreateBinOp(Opc: BOpc, LHS: cast<BinaryOperator>(Val: Op0)->getOperand(i_nocapture: 1), |
390 | RHS: Op1)); |
391 | } |
392 | } |
393 | |
394 | // (X / Y) * Y = X - (X % Y) |
395 | // (X / Y) * -Y = (X % Y) - X |
396 | { |
397 | Value *Y = Op1; |
398 | BinaryOperator *Div = dyn_cast<BinaryOperator>(Val: Op0); |
399 | if (!Div || (Div->getOpcode() != Instruction::UDiv && |
400 | Div->getOpcode() != Instruction::SDiv)) { |
401 | Y = Op0; |
402 | Div = dyn_cast<BinaryOperator>(Val: Op1); |
403 | } |
404 | Value *Neg = dyn_castNegVal(V: Y); |
405 | if (Div && Div->hasOneUse() && |
406 | (Div->getOperand(i_nocapture: 1) == Y || Div->getOperand(i_nocapture: 1) == Neg) && |
407 | (Div->getOpcode() == Instruction::UDiv || |
408 | Div->getOpcode() == Instruction::SDiv)) { |
409 | Value *X = Div->getOperand(i_nocapture: 0), *DivOp1 = Div->getOperand(i_nocapture: 1); |
410 | |
411 | // If the division is exact, X % Y is zero, so we end up with X or -X. |
412 | if (Div->isExact()) { |
413 | if (DivOp1 == Y) |
414 | return replaceInstUsesWith(I, V: X); |
415 | return BinaryOperator::CreateNeg(Op: X); |
416 | } |
417 | |
418 | auto RemOpc = Div->getOpcode() == Instruction::UDiv ? Instruction::URem |
419 | : Instruction::SRem; |
420 | // X must be frozen because we are increasing its number of uses. |
421 | Value *XFreeze = X; |
422 | if (!isGuaranteedNotToBeUndef(V: X)) |
423 | XFreeze = Builder.CreateFreeze(V: X, Name: X->getName() + ".fr" ); |
424 | Value *Rem = Builder.CreateBinOp(Opc: RemOpc, LHS: XFreeze, RHS: DivOp1); |
425 | if (DivOp1 == Y) |
426 | return BinaryOperator::CreateSub(V1: XFreeze, V2: Rem); |
427 | return BinaryOperator::CreateSub(V1: Rem, V2: XFreeze); |
428 | } |
429 | } |
430 | |
431 | // Fold the following two scenarios: |
432 | // 1) i1 mul -> i1 and. |
433 | // 2) X * Y --> X & Y, iff X, Y can be only {0,1}. |
434 | // Note: We could use known bits to generalize this and related patterns with |
435 | // shifts/truncs |
436 | if (Ty->isIntOrIntVectorTy(BitWidth: 1) || |
437 | (match(V: Op0, P: m_And(L: m_Value(), R: m_One())) && |
438 | match(V: Op1, P: m_And(L: m_Value(), R: m_One())))) |
439 | return BinaryOperator::CreateAnd(V1: Op0, V2: Op1); |
440 | |
441 | if (Value *R = foldMulShl1(Mul&: I, /* CommuteOperands */ false, Builder)) |
442 | return replaceInstUsesWith(I, V: R); |
443 | if (Value *R = foldMulShl1(Mul&: I, /* CommuteOperands */ true, Builder)) |
444 | return replaceInstUsesWith(I, V: R); |
445 | |
446 | // (zext bool X) * (zext bool Y) --> zext (and X, Y) |
447 | // (sext bool X) * (sext bool Y) --> zext (and X, Y) |
448 | // Note: -1 * -1 == 1 * 1 == 1 (if the extends match, the result is the same) |
449 | if (((match(V: Op0, P: m_ZExt(Op: m_Value(V&: X))) && match(V: Op1, P: m_ZExt(Op: m_Value(V&: Y)))) || |
450 | (match(V: Op0, P: m_SExt(Op: m_Value(V&: X))) && match(V: Op1, P: m_SExt(Op: m_Value(V&: Y))))) && |
451 | X->getType()->isIntOrIntVectorTy(BitWidth: 1) && X->getType() == Y->getType() && |
452 | (Op0->hasOneUse() || Op1->hasOneUse() || X == Y)) { |
453 | Value *And = Builder.CreateAnd(LHS: X, RHS: Y, Name: "mulbool" ); |
454 | return CastInst::Create(Instruction::ZExt, S: And, Ty); |
455 | } |
456 | // (sext bool X) * (zext bool Y) --> sext (and X, Y) |
457 | // (zext bool X) * (sext bool Y) --> sext (and X, Y) |
458 | // Note: -1 * 1 == 1 * -1 == -1 |
459 | if (((match(V: Op0, P: m_SExt(Op: m_Value(V&: X))) && match(V: Op1, P: m_ZExt(Op: m_Value(V&: Y)))) || |
460 | (match(V: Op0, P: m_ZExt(Op: m_Value(V&: X))) && match(V: Op1, P: m_SExt(Op: m_Value(V&: Y))))) && |
461 | X->getType()->isIntOrIntVectorTy(BitWidth: 1) && X->getType() == Y->getType() && |
462 | (Op0->hasOneUse() || Op1->hasOneUse())) { |
463 | Value *And = Builder.CreateAnd(LHS: X, RHS: Y, Name: "mulbool" ); |
464 | return CastInst::Create(Instruction::SExt, S: And, Ty); |
465 | } |
466 | |
467 | // (zext bool X) * Y --> X ? Y : 0 |
468 | // Y * (zext bool X) --> X ? Y : 0 |
469 | if (match(V: Op0, P: m_ZExt(Op: m_Value(V&: X))) && X->getType()->isIntOrIntVectorTy(BitWidth: 1)) |
470 | return SelectInst::Create(C: X, S1: Op1, S2: ConstantInt::getNullValue(Ty)); |
471 | if (match(V: Op1, P: m_ZExt(Op: m_Value(V&: X))) && X->getType()->isIntOrIntVectorTy(BitWidth: 1)) |
472 | return SelectInst::Create(C: X, S1: Op0, S2: ConstantInt::getNullValue(Ty)); |
473 | |
474 | // mul (sext X), Y -> select X, -Y, 0 |
475 | // mul Y, (sext X) -> select X, -Y, 0 |
476 | if (match(V: &I, P: m_c_Mul(L: m_OneUse(SubPattern: m_SExt(Op: m_Value(V&: X))), R: m_Value(V&: Y))) && |
477 | X->getType()->isIntOrIntVectorTy(BitWidth: 1)) |
478 | return SelectInst::Create(C: X, S1: Builder.CreateNeg(V: Y, Name: "" , HasNSW: I.hasNoSignedWrap()), |
479 | S2: ConstantInt::getNullValue(Ty: Op0->getType())); |
480 | |
481 | Constant *ImmC; |
482 | if (match(V: Op1, P: m_ImmConstant(C&: ImmC))) { |
483 | // (sext bool X) * C --> X ? -C : 0 |
484 | if (match(V: Op0, P: m_SExt(Op: m_Value(V&: X))) && X->getType()->isIntOrIntVectorTy(BitWidth: 1)) { |
485 | Constant *NegC = ConstantExpr::getNeg(C: ImmC); |
486 | return SelectInst::Create(C: X, S1: NegC, S2: ConstantInt::getNullValue(Ty)); |
487 | } |
488 | |
489 | // (ashr i32 X, 31) * C --> (X < 0) ? -C : 0 |
490 | const APInt *C; |
491 | if (match(V: Op0, P: m_OneUse(SubPattern: m_AShr(L: m_Value(V&: X), R: m_APInt(Res&: C)))) && |
492 | *C == C->getBitWidth() - 1) { |
493 | Constant *NegC = ConstantExpr::getNeg(C: ImmC); |
494 | Value *IsNeg = Builder.CreateIsNeg(Arg: X, Name: "isneg" ); |
495 | return SelectInst::Create(C: IsNeg, S1: NegC, S2: ConstantInt::getNullValue(Ty)); |
496 | } |
497 | } |
498 | |
499 | // (lshr X, 31) * Y --> (X < 0) ? Y : 0 |
500 | // TODO: We are not checking one-use because the elimination of the multiply |
501 | // is better for analysis? |
502 | const APInt *C; |
503 | if (match(V: &I, P: m_c_BinOp(L: m_LShr(L: m_Value(V&: X), R: m_APInt(Res&: C)), R: m_Value(V&: Y))) && |
504 | *C == C->getBitWidth() - 1) { |
505 | Value *IsNeg = Builder.CreateIsNeg(Arg: X, Name: "isneg" ); |
506 | return SelectInst::Create(C: IsNeg, S1: Y, S2: ConstantInt::getNullValue(Ty)); |
507 | } |
508 | |
509 | // (and X, 1) * Y --> (trunc X) ? Y : 0 |
510 | if (match(V: &I, P: m_c_BinOp(L: m_OneUse(SubPattern: m_And(L: m_Value(V&: X), R: m_One())), R: m_Value(V&: Y)))) { |
511 | Value *Tr = Builder.CreateTrunc(V: X, DestTy: CmpInst::makeCmpResultType(opnd_type: Ty)); |
512 | return SelectInst::Create(C: Tr, S1: Y, S2: ConstantInt::getNullValue(Ty)); |
513 | } |
514 | |
515 | // ((ashr X, 31) | 1) * X --> abs(X) |
516 | // X * ((ashr X, 31) | 1) --> abs(X) |
517 | if (match(V: &I, P: m_c_BinOp(L: m_Or(L: m_AShr(L: m_Value(V&: X), |
518 | R: m_SpecificIntAllowPoison(V: BitWidth - 1)), |
519 | R: m_One()), |
520 | R: m_Deferred(V: X)))) { |
521 | Value *Abs = Builder.CreateBinaryIntrinsic( |
522 | ID: Intrinsic::abs, LHS: X, RHS: ConstantInt::getBool(Context&: I.getContext(), V: HasNSW)); |
523 | Abs->takeName(V: &I); |
524 | return replaceInstUsesWith(I, V: Abs); |
525 | } |
526 | |
527 | if (Instruction *Ext = narrowMathIfNoOverflow(I)) |
528 | return Ext; |
529 | |
530 | if (Instruction *Res = foldBinOpOfSelectAndCastOfSelectCondition(I)) |
531 | return Res; |
532 | |
533 | // (mul Op0 Op1): |
534 | // if Log2(Op0) folds away -> |
535 | // (shl Op1, Log2(Op0)) |
536 | // if Log2(Op1) folds away -> |
537 | // (shl Op0, Log2(Op1)) |
538 | if (takeLog2(Builder, Op: Op0, /*Depth*/ 0, /*AssumeNonZero*/ false, |
539 | /*DoFold*/ false)) { |
540 | Value *Res = takeLog2(Builder, Op: Op0, /*Depth*/ 0, /*AssumeNonZero*/ false, |
541 | /*DoFold*/ true); |
542 | BinaryOperator *Shl = BinaryOperator::CreateShl(V1: Op1, V2: Res); |
543 | // We can only propegate nuw flag. |
544 | Shl->setHasNoUnsignedWrap(HasNUW); |
545 | return Shl; |
546 | } |
547 | if (takeLog2(Builder, Op: Op1, /*Depth*/ 0, /*AssumeNonZero*/ false, |
548 | /*DoFold*/ false)) { |
549 | Value *Res = takeLog2(Builder, Op: Op1, /*Depth*/ 0, /*AssumeNonZero*/ false, |
550 | /*DoFold*/ true); |
551 | BinaryOperator *Shl = BinaryOperator::CreateShl(V1: Op0, V2: Res); |
552 | // We can only propegate nuw flag. |
553 | Shl->setHasNoUnsignedWrap(HasNUW); |
554 | return Shl; |
555 | } |
556 | |
557 | bool Changed = false; |
558 | if (!HasNSW && willNotOverflowSignedMul(LHS: Op0, RHS: Op1, CxtI: I)) { |
559 | Changed = true; |
560 | I.setHasNoSignedWrap(true); |
561 | } |
562 | |
563 | if (!HasNUW && willNotOverflowUnsignedMul(LHS: Op0, RHS: Op1, CxtI: I, IsNSW: I.hasNoSignedWrap())) { |
564 | Changed = true; |
565 | I.setHasNoUnsignedWrap(true); |
566 | } |
567 | |
568 | return Changed ? &I : nullptr; |
569 | } |
570 | |
571 | Instruction *InstCombinerImpl::foldFPSignBitOps(BinaryOperator &I) { |
572 | BinaryOperator::BinaryOps Opcode = I.getOpcode(); |
573 | assert((Opcode == Instruction::FMul || Opcode == Instruction::FDiv) && |
574 | "Expected fmul or fdiv" ); |
575 | |
576 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
577 | Value *X, *Y; |
578 | |
579 | // -X * -Y --> X * Y |
580 | // -X / -Y --> X / Y |
581 | if (match(V: Op0, P: m_FNeg(X: m_Value(V&: X))) && match(V: Op1, P: m_FNeg(X: m_Value(V&: Y)))) |
582 | return BinaryOperator::CreateWithCopiedFlags(Opc: Opcode, V1: X, V2: Y, CopyO: &I); |
583 | |
584 | // fabs(X) * fabs(X) -> X * X |
585 | // fabs(X) / fabs(X) -> X / X |
586 | if (Op0 == Op1 && match(V: Op0, P: m_FAbs(Op0: m_Value(V&: X)))) |
587 | return BinaryOperator::CreateWithCopiedFlags(Opc: Opcode, V1: X, V2: X, CopyO: &I); |
588 | |
589 | // fabs(X) * fabs(Y) --> fabs(X * Y) |
590 | // fabs(X) / fabs(Y) --> fabs(X / Y) |
591 | if (match(V: Op0, P: m_FAbs(Op0: m_Value(V&: X))) && match(V: Op1, P: m_FAbs(Op0: m_Value(V&: Y))) && |
592 | (Op0->hasOneUse() || Op1->hasOneUse())) { |
593 | IRBuilder<>::FastMathFlagGuard FMFGuard(Builder); |
594 | Builder.setFastMathFlags(I.getFastMathFlags()); |
595 | Value *XY = Builder.CreateBinOp(Opc: Opcode, LHS: X, RHS: Y); |
596 | Value *Fabs = Builder.CreateUnaryIntrinsic(ID: Intrinsic::fabs, V: XY); |
597 | Fabs->takeName(V: &I); |
598 | return replaceInstUsesWith(I, V: Fabs); |
599 | } |
600 | |
601 | return nullptr; |
602 | } |
603 | |
604 | Instruction *InstCombinerImpl::foldPowiReassoc(BinaryOperator &I) { |
605 | auto createPowiExpr = [](BinaryOperator &I, InstCombinerImpl &IC, Value *X, |
606 | Value *Y, Value *Z) { |
607 | InstCombiner::BuilderTy &Builder = IC.Builder; |
608 | Value *YZ = Builder.CreateAdd(LHS: Y, RHS: Z); |
609 | Instruction *NewPow = Builder.CreateIntrinsic( |
610 | ID: Intrinsic::powi, Types: {X->getType(), YZ->getType()}, Args: {X, YZ}, FMFSource: &I); |
611 | |
612 | return NewPow; |
613 | }; |
614 | |
615 | Value *X, *Y, *Z; |
616 | unsigned Opcode = I.getOpcode(); |
617 | assert((Opcode == Instruction::FMul || Opcode == Instruction::FDiv) && |
618 | "Unexpected opcode" ); |
619 | |
620 | // powi(X, Y) * X --> powi(X, Y+1) |
621 | // X * powi(X, Y) --> powi(X, Y+1) |
622 | if (match(V: &I, P: m_c_FMul(L: m_OneUse(SubPattern: m_AllowReassoc(SubPattern: m_Intrinsic<Intrinsic::powi>( |
623 | Op0: m_Value(V&: X), Op1: m_Value(V&: Y)))), |
624 | R: m_Deferred(V: X)))) { |
625 | Constant *One = ConstantInt::get(Ty: Y->getType(), V: 1); |
626 | if (willNotOverflowSignedAdd(LHS: Y, RHS: One, CxtI: I)) { |
627 | Instruction *NewPow = createPowiExpr(I, *this, X, Y, One); |
628 | return replaceInstUsesWith(I, V: NewPow); |
629 | } |
630 | } |
631 | |
632 | // powi(x, y) * powi(x, z) -> powi(x, y + z) |
633 | Value *Op0 = I.getOperand(i_nocapture: 0); |
634 | Value *Op1 = I.getOperand(i_nocapture: 1); |
635 | if (Opcode == Instruction::FMul && I.isOnlyUserOfAnyOperand() && |
636 | match(V: Op0, P: m_AllowReassoc( |
637 | SubPattern: m_Intrinsic<Intrinsic::powi>(Op0: m_Value(V&: X), Op1: m_Value(V&: Y)))) && |
638 | match(V: Op1, P: m_AllowReassoc(SubPattern: m_Intrinsic<Intrinsic::powi>(Op0: m_Specific(V: X), |
639 | Op1: m_Value(V&: Z)))) && |
640 | Y->getType() == Z->getType()) { |
641 | Instruction *NewPow = createPowiExpr(I, *this, X, Y, Z); |
642 | return replaceInstUsesWith(I, V: NewPow); |
643 | } |
644 | |
645 | if (Opcode == Instruction::FDiv && I.hasAllowReassoc() && I.hasNoNaNs()) { |
646 | // powi(X, Y) / X --> powi(X, Y-1) |
647 | // This is legal when (Y - 1) can't wraparound, in which case reassoc and |
648 | // nnan are required. |
649 | // TODO: Multi-use may be also better off creating Powi(x,y-1) |
650 | if (match(V: Op0, P: m_OneUse(SubPattern: m_AllowReassoc(SubPattern: m_Intrinsic<Intrinsic::powi>( |
651 | Op0: m_Specific(V: Op1), Op1: m_Value(V&: Y))))) && |
652 | willNotOverflowSignedSub(LHS: Y, RHS: ConstantInt::get(Ty: Y->getType(), V: 1), CxtI: I)) { |
653 | Constant *NegOne = ConstantInt::getAllOnesValue(Ty: Y->getType()); |
654 | Instruction *NewPow = createPowiExpr(I, *this, Op1, Y, NegOne); |
655 | return replaceInstUsesWith(I, V: NewPow); |
656 | } |
657 | |
658 | // powi(X, Y) / (X * Z) --> powi(X, Y-1) / Z |
659 | // This is legal when (Y - 1) can't wraparound, in which case reassoc and |
660 | // nnan are required. |
661 | // TODO: Multi-use may be also better off creating Powi(x,y-1) |
662 | if (match(V: Op0, P: m_OneUse(SubPattern: m_AllowReassoc(SubPattern: m_Intrinsic<Intrinsic::powi>( |
663 | Op0: m_Value(V&: X), Op1: m_Value(V&: Y))))) && |
664 | match(V: Op1, P: m_AllowReassoc(SubPattern: m_c_FMul(L: m_Specific(V: X), R: m_Value(V&: Z)))) && |
665 | willNotOverflowSignedSub(LHS: Y, RHS: ConstantInt::get(Ty: Y->getType(), V: 1), CxtI: I)) { |
666 | Constant *NegOne = ConstantInt::getAllOnesValue(Ty: Y->getType()); |
667 | auto *NewPow = createPowiExpr(I, *this, X, Y, NegOne); |
668 | return BinaryOperator::CreateFDivFMF(V1: NewPow, V2: Z, FMFSource: &I); |
669 | } |
670 | } |
671 | |
672 | return nullptr; |
673 | } |
674 | |
675 | Instruction *InstCombinerImpl::foldFMulReassoc(BinaryOperator &I) { |
676 | Value *Op0 = I.getOperand(i_nocapture: 0); |
677 | Value *Op1 = I.getOperand(i_nocapture: 1); |
678 | Value *X, *Y; |
679 | Constant *C; |
680 | BinaryOperator *Op0BinOp; |
681 | |
682 | // Reassociate constant RHS with another constant to form constant |
683 | // expression. |
684 | if (match(V: Op1, P: m_Constant(C)) && C->isFiniteNonZeroFP() && |
685 | match(V: Op0, P: m_AllowReassoc(SubPattern: m_BinOp(I&: Op0BinOp)))) { |
686 | // Everything in this scope folds I with Op0, intersecting their FMF. |
687 | FastMathFlags FMF = I.getFastMathFlags() & Op0BinOp->getFastMathFlags(); |
688 | IRBuilder<>::FastMathFlagGuard FMFGuard(Builder); |
689 | Builder.setFastMathFlags(FMF); |
690 | Constant *C1; |
691 | if (match(V: Op0, P: m_OneUse(SubPattern: m_FDiv(L: m_Constant(C&: C1), R: m_Value(V&: X))))) { |
692 | // (C1 / X) * C --> (C * C1) / X |
693 | Constant *CC1 = |
694 | ConstantFoldBinaryOpOperands(Opcode: Instruction::FMul, LHS: C, RHS: C1, DL); |
695 | if (CC1 && CC1->isNormalFP()) |
696 | return BinaryOperator::CreateFDivFMF(V1: CC1, V2: X, FMF); |
697 | } |
698 | if (match(V: Op0, P: m_FDiv(L: m_Value(V&: X), R: m_Constant(C&: C1)))) { |
699 | // FIXME: This seems like it should also be checking for arcp |
700 | // (X / C1) * C --> X * (C / C1) |
701 | Constant *CDivC1 = |
702 | ConstantFoldBinaryOpOperands(Opcode: Instruction::FDiv, LHS: C, RHS: C1, DL); |
703 | if (CDivC1 && CDivC1->isNormalFP()) |
704 | return BinaryOperator::CreateFMulFMF(V1: X, V2: CDivC1, FMF); |
705 | |
706 | // If the constant was a denormal, try reassociating differently. |
707 | // (X / C1) * C --> X / (C1 / C) |
708 | Constant *C1DivC = |
709 | ConstantFoldBinaryOpOperands(Opcode: Instruction::FDiv, LHS: C1, RHS: C, DL); |
710 | if (C1DivC && Op0->hasOneUse() && C1DivC->isNormalFP()) |
711 | return BinaryOperator::CreateFDivFMF(V1: X, V2: C1DivC, FMF); |
712 | } |
713 | |
714 | // We do not need to match 'fadd C, X' and 'fsub X, C' because they are |
715 | // canonicalized to 'fadd X, C'. Distributing the multiply may allow |
716 | // further folds and (X * C) + C2 is 'fma'. |
717 | if (match(V: Op0, P: m_OneUse(SubPattern: m_FAdd(L: m_Value(V&: X), R: m_Constant(C&: C1))))) { |
718 | // (X + C1) * C --> (X * C) + (C * C1) |
719 | if (Constant *CC1 = |
720 | ConstantFoldBinaryOpOperands(Opcode: Instruction::FMul, LHS: C, RHS: C1, DL)) { |
721 | Value *XC = Builder.CreateFMul(L: X, R: C); |
722 | return BinaryOperator::CreateFAddFMF(V1: XC, V2: CC1, FMF); |
723 | } |
724 | } |
725 | if (match(V: Op0, P: m_OneUse(SubPattern: m_FSub(L: m_Constant(C&: C1), R: m_Value(V&: X))))) { |
726 | // (C1 - X) * C --> (C * C1) - (X * C) |
727 | if (Constant *CC1 = |
728 | ConstantFoldBinaryOpOperands(Opcode: Instruction::FMul, LHS: C, RHS: C1, DL)) { |
729 | Value *XC = Builder.CreateFMul(L: X, R: C); |
730 | return BinaryOperator::CreateFSubFMF(V1: CC1, V2: XC, FMF); |
731 | } |
732 | } |
733 | } |
734 | |
735 | Value *Z; |
736 | if (match(V: &I, |
737 | P: m_c_FMul(L: m_AllowReassoc(SubPattern: m_OneUse(SubPattern: m_FDiv(L: m_Value(V&: X), R: m_Value(V&: Y)))), |
738 | R: m_Value(V&: Z)))) { |
739 | BinaryOperator *DivOp = cast<BinaryOperator>(Val: ((Z == Op0) ? Op1 : Op0)); |
740 | FastMathFlags FMF = I.getFastMathFlags() & DivOp->getFastMathFlags(); |
741 | if (FMF.allowReassoc()) { |
742 | // Sink division: (X / Y) * Z --> (X * Z) / Y |
743 | IRBuilder<>::FastMathFlagGuard FMFGuard(Builder); |
744 | Builder.setFastMathFlags(FMF); |
745 | auto *NewFMul = Builder.CreateFMul(L: X, R: Z); |
746 | return BinaryOperator::CreateFDivFMF(V1: NewFMul, V2: Y, FMF); |
747 | } |
748 | } |
749 | |
750 | // sqrt(X) * sqrt(Y) -> sqrt(X * Y) |
751 | // nnan disallows the possibility of returning a number if both operands are |
752 | // negative (in that case, we should return NaN). |
753 | if (I.hasNoNaNs() && match(V: Op0, P: m_OneUse(SubPattern: m_Sqrt(Op0: m_Value(V&: X)))) && |
754 | match(V: Op1, P: m_OneUse(SubPattern: m_Sqrt(Op0: m_Value(V&: Y))))) { |
755 | Value *XY = Builder.CreateFMulFMF(L: X, R: Y, FMFSource: &I); |
756 | Value *Sqrt = Builder.CreateUnaryIntrinsic(ID: Intrinsic::sqrt, V: XY, FMFSource: &I); |
757 | return replaceInstUsesWith(I, V: Sqrt); |
758 | } |
759 | |
760 | // The following transforms are done irrespective of the number of uses |
761 | // for the expression "1.0/sqrt(X)". |
762 | // 1) 1.0/sqrt(X) * X -> X/sqrt(X) |
763 | // 2) X * 1.0/sqrt(X) -> X/sqrt(X) |
764 | // We always expect the backend to reduce X/sqrt(X) to sqrt(X), if it |
765 | // has the necessary (reassoc) fast-math-flags. |
766 | if (I.hasNoSignedZeros() && |
767 | match(V: Op0, P: (m_FDiv(L: m_SpecificFP(V: 1.0), R: m_Value(V&: Y)))) && |
768 | match(V: Y, P: m_Sqrt(Op0: m_Value(V&: X))) && Op1 == X) |
769 | return BinaryOperator::CreateFDivFMF(V1: X, V2: Y, FMFSource: &I); |
770 | if (I.hasNoSignedZeros() && |
771 | match(V: Op1, P: (m_FDiv(L: m_SpecificFP(V: 1.0), R: m_Value(V&: Y)))) && |
772 | match(V: Y, P: m_Sqrt(Op0: m_Value(V&: X))) && Op0 == X) |
773 | return BinaryOperator::CreateFDivFMF(V1: X, V2: Y, FMFSource: &I); |
774 | |
775 | // Like the similar transform in instsimplify, this requires 'nsz' because |
776 | // sqrt(-0.0) = -0.0, and -0.0 * -0.0 does not simplify to -0.0. |
777 | if (I.hasNoNaNs() && I.hasNoSignedZeros() && Op0 == Op1 && Op0->hasNUses(N: 2)) { |
778 | // Peek through fdiv to find squaring of square root: |
779 | // (X / sqrt(Y)) * (X / sqrt(Y)) --> (X * X) / Y |
780 | if (match(V: Op0, P: m_FDiv(L: m_Value(V&: X), R: m_Sqrt(Op0: m_Value(V&: Y))))) { |
781 | Value *XX = Builder.CreateFMulFMF(L: X, R: X, FMFSource: &I); |
782 | return BinaryOperator::CreateFDivFMF(V1: XX, V2: Y, FMFSource: &I); |
783 | } |
784 | // (sqrt(Y) / X) * (sqrt(Y) / X) --> Y / (X * X) |
785 | if (match(V: Op0, P: m_FDiv(L: m_Sqrt(Op0: m_Value(V&: Y)), R: m_Value(V&: X)))) { |
786 | Value *XX = Builder.CreateFMulFMF(L: X, R: X, FMFSource: &I); |
787 | return BinaryOperator::CreateFDivFMF(V1: Y, V2: XX, FMFSource: &I); |
788 | } |
789 | } |
790 | |
791 | // pow(X, Y) * X --> pow(X, Y+1) |
792 | // X * pow(X, Y) --> pow(X, Y+1) |
793 | if (match(V: &I, P: m_c_FMul(L: m_OneUse(SubPattern: m_Intrinsic<Intrinsic::pow>(Op0: m_Value(V&: X), |
794 | Op1: m_Value(V&: Y))), |
795 | R: m_Deferred(V: X)))) { |
796 | Value *Y1 = Builder.CreateFAddFMF(L: Y, R: ConstantFP::get(Ty: I.getType(), V: 1.0), FMFSource: &I); |
797 | Value *Pow = Builder.CreateBinaryIntrinsic(ID: Intrinsic::pow, LHS: X, RHS: Y1, FMFSource: &I); |
798 | return replaceInstUsesWith(I, V: Pow); |
799 | } |
800 | |
801 | if (Instruction *FoldedPowi = foldPowiReassoc(I)) |
802 | return FoldedPowi; |
803 | |
804 | if (I.isOnlyUserOfAnyOperand()) { |
805 | // pow(X, Y) * pow(X, Z) -> pow(X, Y + Z) |
806 | if (match(V: Op0, P: m_Intrinsic<Intrinsic::pow>(Op0: m_Value(V&: X), Op1: m_Value(V&: Y))) && |
807 | match(V: Op1, P: m_Intrinsic<Intrinsic::pow>(Op0: m_Specific(V: X), Op1: m_Value(V&: Z)))) { |
808 | auto *YZ = Builder.CreateFAddFMF(L: Y, R: Z, FMFSource: &I); |
809 | auto *NewPow = Builder.CreateBinaryIntrinsic(ID: Intrinsic::pow, LHS: X, RHS: YZ, FMFSource: &I); |
810 | return replaceInstUsesWith(I, V: NewPow); |
811 | } |
812 | // pow(X, Y) * pow(Z, Y) -> pow(X * Z, Y) |
813 | if (match(V: Op0, P: m_Intrinsic<Intrinsic::pow>(Op0: m_Value(V&: X), Op1: m_Value(V&: Y))) && |
814 | match(V: Op1, P: m_Intrinsic<Intrinsic::pow>(Op0: m_Value(V&: Z), Op1: m_Specific(V: Y)))) { |
815 | auto *XZ = Builder.CreateFMulFMF(L: X, R: Z, FMFSource: &I); |
816 | auto *NewPow = Builder.CreateBinaryIntrinsic(ID: Intrinsic::pow, LHS: XZ, RHS: Y, FMFSource: &I); |
817 | return replaceInstUsesWith(I, V: NewPow); |
818 | } |
819 | |
820 | // exp(X) * exp(Y) -> exp(X + Y) |
821 | if (match(V: Op0, P: m_Intrinsic<Intrinsic::exp>(Op0: m_Value(V&: X))) && |
822 | match(V: Op1, P: m_Intrinsic<Intrinsic::exp>(Op0: m_Value(V&: Y)))) { |
823 | Value *XY = Builder.CreateFAddFMF(L: X, R: Y, FMFSource: &I); |
824 | Value *Exp = Builder.CreateUnaryIntrinsic(ID: Intrinsic::exp, V: XY, FMFSource: &I); |
825 | return replaceInstUsesWith(I, V: Exp); |
826 | } |
827 | |
828 | // exp2(X) * exp2(Y) -> exp2(X + Y) |
829 | if (match(V: Op0, P: m_Intrinsic<Intrinsic::exp2>(Op0: m_Value(V&: X))) && |
830 | match(V: Op1, P: m_Intrinsic<Intrinsic::exp2>(Op0: m_Value(V&: Y)))) { |
831 | Value *XY = Builder.CreateFAddFMF(L: X, R: Y, FMFSource: &I); |
832 | Value *Exp2 = Builder.CreateUnaryIntrinsic(ID: Intrinsic::exp2, V: XY, FMFSource: &I); |
833 | return replaceInstUsesWith(I, V: Exp2); |
834 | } |
835 | } |
836 | |
837 | // (X*Y) * X => (X*X) * Y where Y != X |
838 | // The purpose is two-fold: |
839 | // 1) to form a power expression (of X). |
840 | // 2) potentially shorten the critical path: After transformation, the |
841 | // latency of the instruction Y is amortized by the expression of X*X, |
842 | // and therefore Y is in a "less critical" position compared to what it |
843 | // was before the transformation. |
844 | if (match(V: Op0, P: m_OneUse(SubPattern: m_c_FMul(L: m_Specific(V: Op1), R: m_Value(V&: Y)))) && Op1 != Y) { |
845 | Value *XX = Builder.CreateFMulFMF(L: Op1, R: Op1, FMFSource: &I); |
846 | return BinaryOperator::CreateFMulFMF(V1: XX, V2: Y, FMFSource: &I); |
847 | } |
848 | if (match(V: Op1, P: m_OneUse(SubPattern: m_c_FMul(L: m_Specific(V: Op0), R: m_Value(V&: Y)))) && Op0 != Y) { |
849 | Value *XX = Builder.CreateFMulFMF(L: Op0, R: Op0, FMFSource: &I); |
850 | return BinaryOperator::CreateFMulFMF(V1: XX, V2: Y, FMFSource: &I); |
851 | } |
852 | |
853 | return nullptr; |
854 | } |
855 | |
856 | Instruction *InstCombinerImpl::visitFMul(BinaryOperator &I) { |
857 | if (Value *V = simplifyFMulInst(LHS: I.getOperand(i_nocapture: 0), RHS: I.getOperand(i_nocapture: 1), |
858 | FMF: I.getFastMathFlags(), |
859 | Q: SQ.getWithInstruction(I: &I))) |
860 | return replaceInstUsesWith(I, V); |
861 | |
862 | if (SimplifyAssociativeOrCommutative(I)) |
863 | return &I; |
864 | |
865 | if (Instruction *X = foldVectorBinop(Inst&: I)) |
866 | return X; |
867 | |
868 | if (Instruction *Phi = foldBinopWithPhiOperands(BO&: I)) |
869 | return Phi; |
870 | |
871 | if (Instruction *FoldedMul = foldBinOpIntoSelectOrPhi(I)) |
872 | return FoldedMul; |
873 | |
874 | if (Value *FoldedMul = foldMulSelectToNegate(I, Builder)) |
875 | return replaceInstUsesWith(I, V: FoldedMul); |
876 | |
877 | if (Instruction *R = foldFPSignBitOps(I)) |
878 | return R; |
879 | |
880 | if (Instruction *R = foldFBinOpOfIntCasts(I)) |
881 | return R; |
882 | |
883 | // X * -1.0 --> -X |
884 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
885 | if (match(V: Op1, P: m_SpecificFP(V: -1.0))) |
886 | return UnaryOperator::CreateFNegFMF(Op: Op0, FMFSource: &I); |
887 | |
888 | // With no-nans/no-infs: |
889 | // X * 0.0 --> copysign(0.0, X) |
890 | // X * -0.0 --> copysign(0.0, -X) |
891 | const APFloat *FPC; |
892 | if (match(V: Op1, P: m_APFloatAllowPoison(Res&: FPC)) && FPC->isZero() && |
893 | ((I.hasNoInfs() && |
894 | isKnownNeverNaN(V: Op0, /*Depth=*/0, SQ: SQ.getWithInstruction(I: &I))) || |
895 | isKnownNeverNaN(V: &I, /*Depth=*/0, SQ: SQ.getWithInstruction(I: &I)))) { |
896 | if (FPC->isNegative()) |
897 | Op0 = Builder.CreateFNegFMF(V: Op0, FMFSource: &I); |
898 | CallInst *CopySign = Builder.CreateIntrinsic(ID: Intrinsic::copysign, |
899 | Types: {I.getType()}, Args: {Op1, Op0}, FMFSource: &I); |
900 | return replaceInstUsesWith(I, V: CopySign); |
901 | } |
902 | |
903 | // -X * C --> X * -C |
904 | Value *X, *Y; |
905 | Constant *C; |
906 | if (match(V: Op0, P: m_FNeg(X: m_Value(V&: X))) && match(V: Op1, P: m_Constant(C))) |
907 | if (Constant *NegC = ConstantFoldUnaryOpOperand(Opcode: Instruction::FNeg, Op: C, DL)) |
908 | return BinaryOperator::CreateFMulFMF(V1: X, V2: NegC, FMFSource: &I); |
909 | |
910 | if (I.hasNoNaNs() && I.hasNoSignedZeros()) { |
911 | // (uitofp bool X) * Y --> X ? Y : 0 |
912 | // Y * (uitofp bool X) --> X ? Y : 0 |
913 | // Note INF * 0 is NaN. |
914 | if (match(V: Op0, P: m_UIToFP(Op: m_Value(V&: X))) && |
915 | X->getType()->isIntOrIntVectorTy(BitWidth: 1)) { |
916 | auto *SI = SelectInst::Create(C: X, S1: Op1, S2: ConstantFP::get(Ty: I.getType(), V: 0.0)); |
917 | SI->copyFastMathFlags(FMF: I.getFastMathFlags()); |
918 | return SI; |
919 | } |
920 | if (match(V: Op1, P: m_UIToFP(Op: m_Value(V&: X))) && |
921 | X->getType()->isIntOrIntVectorTy(BitWidth: 1)) { |
922 | auto *SI = SelectInst::Create(C: X, S1: Op0, S2: ConstantFP::get(Ty: I.getType(), V: 0.0)); |
923 | SI->copyFastMathFlags(FMF: I.getFastMathFlags()); |
924 | return SI; |
925 | } |
926 | } |
927 | |
928 | // (select A, B, C) * (select A, D, E) --> select A, (B*D), (C*E) |
929 | if (Value *V = SimplifySelectsFeedingBinaryOp(I, LHS: Op0, RHS: Op1)) |
930 | return replaceInstUsesWith(I, V); |
931 | |
932 | if (I.hasAllowReassoc()) |
933 | if (Instruction *FoldedMul = foldFMulReassoc(I)) |
934 | return FoldedMul; |
935 | |
936 | // log2(X * 0.5) * Y = log2(X) * Y - Y |
937 | if (I.isFast()) { |
938 | IntrinsicInst *Log2 = nullptr; |
939 | if (match(V: Op0, P: m_OneUse(SubPattern: m_Intrinsic<Intrinsic::log2>( |
940 | Op0: m_OneUse(SubPattern: m_FMul(L: m_Value(V&: X), R: m_SpecificFP(V: 0.5))))))) { |
941 | Log2 = cast<IntrinsicInst>(Val: Op0); |
942 | Y = Op1; |
943 | } |
944 | if (match(V: Op1, P: m_OneUse(SubPattern: m_Intrinsic<Intrinsic::log2>( |
945 | Op0: m_OneUse(SubPattern: m_FMul(L: m_Value(V&: X), R: m_SpecificFP(V: 0.5))))))) { |
946 | Log2 = cast<IntrinsicInst>(Val: Op1); |
947 | Y = Op0; |
948 | } |
949 | if (Log2) { |
950 | Value *Log2 = Builder.CreateUnaryIntrinsic(ID: Intrinsic::log2, V: X, FMFSource: &I); |
951 | Value *LogXTimesY = Builder.CreateFMulFMF(L: Log2, R: Y, FMFSource: &I); |
952 | return BinaryOperator::CreateFSubFMF(V1: LogXTimesY, V2: Y, FMFSource: &I); |
953 | } |
954 | } |
955 | |
956 | // Simplify FMUL recurrences starting with 0.0 to 0.0 if nnan and nsz are set. |
957 | // Given a phi node with entry value as 0 and it used in fmul operation, |
958 | // we can replace fmul with 0 safely and eleminate loop operation. |
959 | PHINode *PN = nullptr; |
960 | Value *Start = nullptr, *Step = nullptr; |
961 | if (matchSimpleRecurrence(I: &I, P&: PN, Start, Step) && I.hasNoNaNs() && |
962 | I.hasNoSignedZeros() && match(V: Start, P: m_Zero())) |
963 | return replaceInstUsesWith(I, V: Start); |
964 | |
965 | // minimum(X, Y) * maximum(X, Y) => X * Y. |
966 | if (match(V: &I, |
967 | P: m_c_FMul(L: m_Intrinsic<Intrinsic::maximum>(Op0: m_Value(V&: X), Op1: m_Value(V&: Y)), |
968 | R: m_c_Intrinsic<Intrinsic::minimum>(Op0: m_Deferred(V: X), |
969 | Op1: m_Deferred(V: Y))))) { |
970 | BinaryOperator *Result = BinaryOperator::CreateFMulFMF(V1: X, V2: Y, FMFSource: &I); |
971 | // We cannot preserve ninf if nnan flag is not set. |
972 | // If X is NaN and Y is Inf then in original program we had NaN * NaN, |
973 | // while in optimized version NaN * Inf and this is a poison with ninf flag. |
974 | if (!Result->hasNoNaNs()) |
975 | Result->setHasNoInfs(false); |
976 | return Result; |
977 | } |
978 | |
979 | return nullptr; |
980 | } |
981 | |
982 | /// Fold a divide or remainder with a select instruction divisor when one of the |
983 | /// select operands is zero. In that case, we can use the other select operand |
984 | /// because div/rem by zero is undefined. |
985 | bool InstCombinerImpl::simplifyDivRemOfSelectWithZeroOp(BinaryOperator &I) { |
986 | SelectInst *SI = dyn_cast<SelectInst>(Val: I.getOperand(i_nocapture: 1)); |
987 | if (!SI) |
988 | return false; |
989 | |
990 | int NonNullOperand; |
991 | if (match(V: SI->getTrueValue(), P: m_Zero())) |
992 | // div/rem X, (Cond ? 0 : Y) -> div/rem X, Y |
993 | NonNullOperand = 2; |
994 | else if (match(V: SI->getFalseValue(), P: m_Zero())) |
995 | // div/rem X, (Cond ? Y : 0) -> div/rem X, Y |
996 | NonNullOperand = 1; |
997 | else |
998 | return false; |
999 | |
1000 | // Change the div/rem to use 'Y' instead of the select. |
1001 | replaceOperand(I, OpNum: 1, V: SI->getOperand(i_nocapture: NonNullOperand)); |
1002 | |
1003 | // Okay, we know we replace the operand of the div/rem with 'Y' with no |
1004 | // problem. However, the select, or the condition of the select may have |
1005 | // multiple uses. Based on our knowledge that the operand must be non-zero, |
1006 | // propagate the known value for the select into other uses of it, and |
1007 | // propagate a known value of the condition into its other users. |
1008 | |
1009 | // If the select and condition only have a single use, don't bother with this, |
1010 | // early exit. |
1011 | Value *SelectCond = SI->getCondition(); |
1012 | if (SI->use_empty() && SelectCond->hasOneUse()) |
1013 | return true; |
1014 | |
1015 | // Scan the current block backward, looking for other uses of SI. |
1016 | BasicBlock::iterator BBI = I.getIterator(), BBFront = I.getParent()->begin(); |
1017 | Type *CondTy = SelectCond->getType(); |
1018 | while (BBI != BBFront) { |
1019 | --BBI; |
1020 | // If we found an instruction that we can't assume will return, so |
1021 | // information from below it cannot be propagated above it. |
1022 | if (!isGuaranteedToTransferExecutionToSuccessor(I: &*BBI)) |
1023 | break; |
1024 | |
1025 | // Replace uses of the select or its condition with the known values. |
1026 | for (Use &Op : BBI->operands()) { |
1027 | if (Op == SI) { |
1028 | replaceUse(U&: Op, NewValue: SI->getOperand(i_nocapture: NonNullOperand)); |
1029 | Worklist.push(I: &*BBI); |
1030 | } else if (Op == SelectCond) { |
1031 | replaceUse(U&: Op, NewValue: NonNullOperand == 1 ? ConstantInt::getTrue(Ty: CondTy) |
1032 | : ConstantInt::getFalse(Ty: CondTy)); |
1033 | Worklist.push(I: &*BBI); |
1034 | } |
1035 | } |
1036 | |
1037 | // If we past the instruction, quit looking for it. |
1038 | if (&*BBI == SI) |
1039 | SI = nullptr; |
1040 | if (&*BBI == SelectCond) |
1041 | SelectCond = nullptr; |
1042 | |
1043 | // If we ran out of things to eliminate, break out of the loop. |
1044 | if (!SelectCond && !SI) |
1045 | break; |
1046 | |
1047 | } |
1048 | return true; |
1049 | } |
1050 | |
1051 | /// True if the multiply can not be expressed in an int this size. |
1052 | static bool multiplyOverflows(const APInt &C1, const APInt &C2, APInt &Product, |
1053 | bool IsSigned) { |
1054 | bool Overflow; |
1055 | Product = IsSigned ? C1.smul_ov(RHS: C2, Overflow) : C1.umul_ov(RHS: C2, Overflow); |
1056 | return Overflow; |
1057 | } |
1058 | |
1059 | /// True if C1 is a multiple of C2. Quotient contains C1/C2. |
1060 | static bool isMultiple(const APInt &C1, const APInt &C2, APInt &Quotient, |
1061 | bool IsSigned) { |
1062 | assert(C1.getBitWidth() == C2.getBitWidth() && "Constant widths not equal" ); |
1063 | |
1064 | // Bail if we will divide by zero. |
1065 | if (C2.isZero()) |
1066 | return false; |
1067 | |
1068 | // Bail if we would divide INT_MIN by -1. |
1069 | if (IsSigned && C1.isMinSignedValue() && C2.isAllOnes()) |
1070 | return false; |
1071 | |
1072 | APInt Remainder(C1.getBitWidth(), /*val=*/0ULL, IsSigned); |
1073 | if (IsSigned) |
1074 | APInt::sdivrem(LHS: C1, RHS: C2, Quotient, Remainder); |
1075 | else |
1076 | APInt::udivrem(LHS: C1, RHS: C2, Quotient, Remainder); |
1077 | |
1078 | return Remainder.isMinValue(); |
1079 | } |
1080 | |
1081 | static Value *foldIDivShl(BinaryOperator &I, InstCombiner::BuilderTy &Builder) { |
1082 | assert((I.getOpcode() == Instruction::SDiv || |
1083 | I.getOpcode() == Instruction::UDiv) && |
1084 | "Expected integer divide" ); |
1085 | |
1086 | bool IsSigned = I.getOpcode() == Instruction::SDiv; |
1087 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
1088 | Type *Ty = I.getType(); |
1089 | |
1090 | Value *X, *Y, *Z; |
1091 | |
1092 | // With appropriate no-wrap constraints, remove a common factor in the |
1093 | // dividend and divisor that is disguised as a left-shifted value. |
1094 | if (match(V: Op1, P: m_Shl(L: m_Value(V&: X), R: m_Value(V&: Z))) && |
1095 | match(V: Op0, P: m_c_Mul(L: m_Specific(V: X), R: m_Value(V&: Y)))) { |
1096 | // Both operands must have the matching no-wrap for this kind of division. |
1097 | auto *Mul = cast<OverflowingBinaryOperator>(Val: Op0); |
1098 | auto *Shl = cast<OverflowingBinaryOperator>(Val: Op1); |
1099 | bool HasNUW = Mul->hasNoUnsignedWrap() && Shl->hasNoUnsignedWrap(); |
1100 | bool HasNSW = Mul->hasNoSignedWrap() && Shl->hasNoSignedWrap(); |
1101 | |
1102 | // (X * Y) u/ (X << Z) --> Y u>> Z |
1103 | if (!IsSigned && HasNUW) |
1104 | return Builder.CreateLShr(LHS: Y, RHS: Z, Name: "" , isExact: I.isExact()); |
1105 | |
1106 | // (X * Y) s/ (X << Z) --> Y s/ (1 << Z) |
1107 | if (IsSigned && HasNSW && (Op0->hasOneUse() || Op1->hasOneUse())) { |
1108 | Value *Shl = Builder.CreateShl(LHS: ConstantInt::get(Ty, V: 1), RHS: Z); |
1109 | return Builder.CreateSDiv(LHS: Y, RHS: Shl, Name: "" , isExact: I.isExact()); |
1110 | } |
1111 | } |
1112 | |
1113 | // With appropriate no-wrap constraints, remove a common factor in the |
1114 | // dividend and divisor that is disguised as a left-shift amount. |
1115 | if (match(V: Op0, P: m_Shl(L: m_Value(V&: X), R: m_Value(V&: Z))) && |
1116 | match(V: Op1, P: m_Shl(L: m_Value(V&: Y), R: m_Specific(V: Z)))) { |
1117 | auto *Shl0 = cast<OverflowingBinaryOperator>(Val: Op0); |
1118 | auto *Shl1 = cast<OverflowingBinaryOperator>(Val: Op1); |
1119 | |
1120 | // For unsigned div, we need 'nuw' on both shifts or |
1121 | // 'nsw' on both shifts + 'nuw' on the dividend. |
1122 | // (X << Z) / (Y << Z) --> X / Y |
1123 | if (!IsSigned && |
1124 | ((Shl0->hasNoUnsignedWrap() && Shl1->hasNoUnsignedWrap()) || |
1125 | (Shl0->hasNoUnsignedWrap() && Shl0->hasNoSignedWrap() && |
1126 | Shl1->hasNoSignedWrap()))) |
1127 | return Builder.CreateUDiv(LHS: X, RHS: Y, Name: "" , isExact: I.isExact()); |
1128 | |
1129 | // For signed div, we need 'nsw' on both shifts + 'nuw' on the divisor. |
1130 | // (X << Z) / (Y << Z) --> X / Y |
1131 | if (IsSigned && Shl0->hasNoSignedWrap() && Shl1->hasNoSignedWrap() && |
1132 | Shl1->hasNoUnsignedWrap()) |
1133 | return Builder.CreateSDiv(LHS: X, RHS: Y, Name: "" , isExact: I.isExact()); |
1134 | } |
1135 | |
1136 | // If X << Y and X << Z does not overflow, then: |
1137 | // (X << Y) / (X << Z) -> (1 << Y) / (1 << Z) -> 1 << Y >> Z |
1138 | if (match(V: Op0, P: m_Shl(L: m_Value(V&: X), R: m_Value(V&: Y))) && |
1139 | match(V: Op1, P: m_Shl(L: m_Specific(V: X), R: m_Value(V&: Z)))) { |
1140 | auto *Shl0 = cast<OverflowingBinaryOperator>(Val: Op0); |
1141 | auto *Shl1 = cast<OverflowingBinaryOperator>(Val: Op1); |
1142 | |
1143 | if (IsSigned ? (Shl0->hasNoSignedWrap() && Shl1->hasNoSignedWrap()) |
1144 | : (Shl0->hasNoUnsignedWrap() && Shl1->hasNoUnsignedWrap())) { |
1145 | Constant *One = ConstantInt::get(Ty: X->getType(), V: 1); |
1146 | // Only preserve the nsw flag if dividend has nsw |
1147 | // or divisor has nsw and operator is sdiv. |
1148 | Value *Dividend = Builder.CreateShl( |
1149 | LHS: One, RHS: Y, Name: "shl.dividend" , |
1150 | /*HasNUW*/ true, |
1151 | /*HasNSW*/ |
1152 | IsSigned ? (Shl0->hasNoUnsignedWrap() || Shl1->hasNoUnsignedWrap()) |
1153 | : Shl0->hasNoSignedWrap()); |
1154 | return Builder.CreateLShr(LHS: Dividend, RHS: Z, Name: "" , isExact: I.isExact()); |
1155 | } |
1156 | } |
1157 | |
1158 | return nullptr; |
1159 | } |
1160 | |
1161 | /// This function implements the transforms common to both integer division |
1162 | /// instructions (udiv and sdiv). It is called by the visitors to those integer |
1163 | /// division instructions. |
1164 | /// Common integer divide transforms |
1165 | Instruction *InstCombinerImpl::commonIDivTransforms(BinaryOperator &I) { |
1166 | if (Instruction *Phi = foldBinopWithPhiOperands(BO&: I)) |
1167 | return Phi; |
1168 | |
1169 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
1170 | bool IsSigned = I.getOpcode() == Instruction::SDiv; |
1171 | Type *Ty = I.getType(); |
1172 | |
1173 | // The RHS is known non-zero. |
1174 | if (Value *V = simplifyValueKnownNonZero(V: I.getOperand(i_nocapture: 1), IC&: *this, CxtI&: I)) |
1175 | return replaceOperand(I, OpNum: 1, V); |
1176 | |
1177 | // Handle cases involving: [su]div X, (select Cond, Y, Z) |
1178 | // This does not apply for fdiv. |
1179 | if (simplifyDivRemOfSelectWithZeroOp(I)) |
1180 | return &I; |
1181 | |
1182 | // If the divisor is a select-of-constants, try to constant fold all div ops: |
1183 | // C / (select Cond, TrueC, FalseC) --> select Cond, (C / TrueC), (C / FalseC) |
1184 | // TODO: Adapt simplifyDivRemOfSelectWithZeroOp to allow this and other folds. |
1185 | if (match(V: Op0, P: m_ImmConstant()) && |
1186 | match(V: Op1, P: m_Select(C: m_Value(), L: m_ImmConstant(), R: m_ImmConstant()))) { |
1187 | if (Instruction *R = FoldOpIntoSelect(Op&: I, SI: cast<SelectInst>(Val: Op1), |
1188 | /*FoldWithMultiUse*/ true)) |
1189 | return R; |
1190 | } |
1191 | |
1192 | const APInt *C2; |
1193 | if (match(V: Op1, P: m_APInt(Res&: C2))) { |
1194 | Value *X; |
1195 | const APInt *C1; |
1196 | |
1197 | // (X / C1) / C2 -> X / (C1*C2) |
1198 | if ((IsSigned && match(V: Op0, P: m_SDiv(L: m_Value(V&: X), R: m_APInt(Res&: C1)))) || |
1199 | (!IsSigned && match(V: Op0, P: m_UDiv(L: m_Value(V&: X), R: m_APInt(Res&: C1))))) { |
1200 | APInt Product(C1->getBitWidth(), /*val=*/0ULL, IsSigned); |
1201 | if (!multiplyOverflows(C1: *C1, C2: *C2, Product, IsSigned)) |
1202 | return BinaryOperator::Create(Op: I.getOpcode(), S1: X, |
1203 | S2: ConstantInt::get(Ty, V: Product)); |
1204 | } |
1205 | |
1206 | APInt Quotient(C2->getBitWidth(), /*val=*/0ULL, IsSigned); |
1207 | if ((IsSigned && match(V: Op0, P: m_NSWMul(L: m_Value(V&: X), R: m_APInt(Res&: C1)))) || |
1208 | (!IsSigned && match(V: Op0, P: m_NUWMul(L: m_Value(V&: X), R: m_APInt(Res&: C1))))) { |
1209 | |
1210 | // (X * C1) / C2 -> X / (C2 / C1) if C2 is a multiple of C1. |
1211 | if (isMultiple(C1: *C2, C2: *C1, Quotient, IsSigned)) { |
1212 | auto *NewDiv = BinaryOperator::Create(Op: I.getOpcode(), S1: X, |
1213 | S2: ConstantInt::get(Ty, V: Quotient)); |
1214 | NewDiv->setIsExact(I.isExact()); |
1215 | return NewDiv; |
1216 | } |
1217 | |
1218 | // (X * C1) / C2 -> X * (C1 / C2) if C1 is a multiple of C2. |
1219 | if (isMultiple(C1: *C1, C2: *C2, Quotient, IsSigned)) { |
1220 | auto *Mul = BinaryOperator::Create(Op: Instruction::Mul, S1: X, |
1221 | S2: ConstantInt::get(Ty, V: Quotient)); |
1222 | auto *OBO = cast<OverflowingBinaryOperator>(Val: Op0); |
1223 | Mul->setHasNoUnsignedWrap(!IsSigned && OBO->hasNoUnsignedWrap()); |
1224 | Mul->setHasNoSignedWrap(OBO->hasNoSignedWrap()); |
1225 | return Mul; |
1226 | } |
1227 | } |
1228 | |
1229 | if ((IsSigned && match(V: Op0, P: m_NSWShl(L: m_Value(V&: X), R: m_APInt(Res&: C1))) && |
1230 | C1->ult(RHS: C1->getBitWidth() - 1)) || |
1231 | (!IsSigned && match(V: Op0, P: m_NUWShl(L: m_Value(V&: X), R: m_APInt(Res&: C1))) && |
1232 | C1->ult(RHS: C1->getBitWidth()))) { |
1233 | APInt C1Shifted = APInt::getOneBitSet( |
1234 | numBits: C1->getBitWidth(), BitNo: static_cast<unsigned>(C1->getZExtValue())); |
1235 | |
1236 | // (X << C1) / C2 -> X / (C2 >> C1) if C2 is a multiple of 1 << C1. |
1237 | if (isMultiple(C1: *C2, C2: C1Shifted, Quotient, IsSigned)) { |
1238 | auto *BO = BinaryOperator::Create(Op: I.getOpcode(), S1: X, |
1239 | S2: ConstantInt::get(Ty, V: Quotient)); |
1240 | BO->setIsExact(I.isExact()); |
1241 | return BO; |
1242 | } |
1243 | |
1244 | // (X << C1) / C2 -> X * ((1 << C1) / C2) if 1 << C1 is a multiple of C2. |
1245 | if (isMultiple(C1: C1Shifted, C2: *C2, Quotient, IsSigned)) { |
1246 | auto *Mul = BinaryOperator::Create(Op: Instruction::Mul, S1: X, |
1247 | S2: ConstantInt::get(Ty, V: Quotient)); |
1248 | auto *OBO = cast<OverflowingBinaryOperator>(Val: Op0); |
1249 | Mul->setHasNoUnsignedWrap(!IsSigned && OBO->hasNoUnsignedWrap()); |
1250 | Mul->setHasNoSignedWrap(OBO->hasNoSignedWrap()); |
1251 | return Mul; |
1252 | } |
1253 | } |
1254 | |
1255 | // Distribute div over add to eliminate a matching div/mul pair: |
1256 | // ((X * C2) + C1) / C2 --> X + C1/C2 |
1257 | // We need a multiple of the divisor for a signed add constant, but |
1258 | // unsigned is fine with any constant pair. |
1259 | if (IsSigned && |
1260 | match(V: Op0, P: m_NSWAddLike(L: m_NSWMul(L: m_Value(V&: X), R: m_SpecificInt(V: *C2)), |
1261 | R: m_APInt(Res&: C1))) && |
1262 | isMultiple(C1: *C1, C2: *C2, Quotient, IsSigned)) { |
1263 | return BinaryOperator::CreateNSWAdd(V1: X, V2: ConstantInt::get(Ty, V: Quotient)); |
1264 | } |
1265 | if (!IsSigned && |
1266 | match(V: Op0, P: m_NUWAddLike(L: m_NUWMul(L: m_Value(V&: X), R: m_SpecificInt(V: *C2)), |
1267 | R: m_APInt(Res&: C1)))) { |
1268 | return BinaryOperator::CreateNUWAdd(V1: X, |
1269 | V2: ConstantInt::get(Ty, V: C1->udiv(RHS: *C2))); |
1270 | } |
1271 | |
1272 | if (!C2->isZero()) // avoid X udiv 0 |
1273 | if (Instruction *FoldedDiv = foldBinOpIntoSelectOrPhi(I)) |
1274 | return FoldedDiv; |
1275 | } |
1276 | |
1277 | if (match(V: Op0, P: m_One())) { |
1278 | assert(!Ty->isIntOrIntVectorTy(1) && "i1 divide not removed?" ); |
1279 | if (IsSigned) { |
1280 | // 1 / 0 --> undef ; 1 / 1 --> 1 ; 1 / -1 --> -1 ; 1 / anything else --> 0 |
1281 | // (Op1 + 1) u< 3 ? Op1 : 0 |
1282 | // Op1 must be frozen because we are increasing its number of uses. |
1283 | Value *F1 = Op1; |
1284 | if (!isGuaranteedNotToBeUndef(V: Op1)) |
1285 | F1 = Builder.CreateFreeze(V: Op1, Name: Op1->getName() + ".fr" ); |
1286 | Value *Inc = Builder.CreateAdd(LHS: F1, RHS: Op0); |
1287 | Value *Cmp = Builder.CreateICmpULT(LHS: Inc, RHS: ConstantInt::get(Ty, V: 3)); |
1288 | return SelectInst::Create(C: Cmp, S1: F1, S2: ConstantInt::get(Ty, V: 0)); |
1289 | } else { |
1290 | // If Op1 is 0 then it's undefined behaviour. If Op1 is 1 then the |
1291 | // result is one, otherwise it's zero. |
1292 | return new ZExtInst(Builder.CreateICmpEQ(LHS: Op1, RHS: Op0), Ty); |
1293 | } |
1294 | } |
1295 | |
1296 | // See if we can fold away this div instruction. |
1297 | if (SimplifyDemandedInstructionBits(Inst&: I)) |
1298 | return &I; |
1299 | |
1300 | // (X - (X rem Y)) / Y -> X / Y; usually originates as ((X / Y) * Y) / Y |
1301 | Value *X, *Z; |
1302 | if (match(V: Op0, P: m_Sub(L: m_Value(V&: X), R: m_Value(V&: Z)))) // (X - Z) / Y; Y = Op1 |
1303 | if ((IsSigned && match(V: Z, P: m_SRem(L: m_Specific(V: X), R: m_Specific(V: Op1)))) || |
1304 | (!IsSigned && match(V: Z, P: m_URem(L: m_Specific(V: X), R: m_Specific(V: Op1))))) |
1305 | return BinaryOperator::Create(Op: I.getOpcode(), S1: X, S2: Op1); |
1306 | |
1307 | // (X << Y) / X -> 1 << Y |
1308 | Value *Y; |
1309 | if (IsSigned && match(V: Op0, P: m_NSWShl(L: m_Specific(V: Op1), R: m_Value(V&: Y)))) |
1310 | return BinaryOperator::CreateNSWShl(V1: ConstantInt::get(Ty, V: 1), V2: Y); |
1311 | if (!IsSigned && match(V: Op0, P: m_NUWShl(L: m_Specific(V: Op1), R: m_Value(V&: Y)))) |
1312 | return BinaryOperator::CreateNUWShl(V1: ConstantInt::get(Ty, V: 1), V2: Y); |
1313 | |
1314 | // X / (X * Y) -> 1 / Y if the multiplication does not overflow. |
1315 | if (match(V: Op1, P: m_c_Mul(L: m_Specific(V: Op0), R: m_Value(V&: Y)))) { |
1316 | bool HasNSW = cast<OverflowingBinaryOperator>(Val: Op1)->hasNoSignedWrap(); |
1317 | bool HasNUW = cast<OverflowingBinaryOperator>(Val: Op1)->hasNoUnsignedWrap(); |
1318 | if ((IsSigned && HasNSW) || (!IsSigned && HasNUW)) { |
1319 | replaceOperand(I, OpNum: 0, V: ConstantInt::get(Ty, V: 1)); |
1320 | replaceOperand(I, OpNum: 1, V: Y); |
1321 | return &I; |
1322 | } |
1323 | } |
1324 | |
1325 | // (X << Z) / (X * Y) -> (1 << Z) / Y |
1326 | // TODO: Handle sdiv. |
1327 | if (!IsSigned && Op1->hasOneUse() && |
1328 | match(V: Op0, P: m_NUWShl(L: m_Value(V&: X), R: m_Value(V&: Z))) && |
1329 | match(V: Op1, P: m_c_Mul(L: m_Specific(V: X), R: m_Value(V&: Y)))) |
1330 | if (cast<OverflowingBinaryOperator>(Val: Op1)->hasNoUnsignedWrap()) { |
1331 | Instruction *NewDiv = BinaryOperator::CreateUDiv( |
1332 | V1: Builder.CreateShl(LHS: ConstantInt::get(Ty, V: 1), RHS: Z, Name: "" , /*NUW*/ HasNUW: true), V2: Y); |
1333 | NewDiv->setIsExact(I.isExact()); |
1334 | return NewDiv; |
1335 | } |
1336 | |
1337 | if (Value *R = foldIDivShl(I, Builder)) |
1338 | return replaceInstUsesWith(I, V: R); |
1339 | |
1340 | // With the appropriate no-wrap constraint, remove a multiply by the divisor |
1341 | // after peeking through another divide: |
1342 | // ((Op1 * X) / Y) / Op1 --> X / Y |
1343 | if (match(V: Op0, P: m_BinOp(Opcode: I.getOpcode(), L: m_c_Mul(L: m_Specific(V: Op1), R: m_Value(V&: X)), |
1344 | R: m_Value(V&: Y)))) { |
1345 | auto *InnerDiv = cast<PossiblyExactOperator>(Val: Op0); |
1346 | auto *Mul = cast<OverflowingBinaryOperator>(Val: InnerDiv->getOperand(i_nocapture: 0)); |
1347 | Instruction *NewDiv = nullptr; |
1348 | if (!IsSigned && Mul->hasNoUnsignedWrap()) |
1349 | NewDiv = BinaryOperator::CreateUDiv(V1: X, V2: Y); |
1350 | else if (IsSigned && Mul->hasNoSignedWrap()) |
1351 | NewDiv = BinaryOperator::CreateSDiv(V1: X, V2: Y); |
1352 | |
1353 | // Exact propagates only if both of the original divides are exact. |
1354 | if (NewDiv) { |
1355 | NewDiv->setIsExact(I.isExact() && InnerDiv->isExact()); |
1356 | return NewDiv; |
1357 | } |
1358 | } |
1359 | |
1360 | // (X * Y) / (X * Z) --> Y / Z (and commuted variants) |
1361 | if (match(V: Op0, P: m_Mul(L: m_Value(V&: X), R: m_Value(V&: Y)))) { |
1362 | auto OB0HasNSW = cast<OverflowingBinaryOperator>(Val: Op0)->hasNoSignedWrap(); |
1363 | auto OB0HasNUW = cast<OverflowingBinaryOperator>(Val: Op0)->hasNoUnsignedWrap(); |
1364 | |
1365 | auto CreateDivOrNull = [&](Value *A, Value *B) -> Instruction * { |
1366 | auto OB1HasNSW = cast<OverflowingBinaryOperator>(Val: Op1)->hasNoSignedWrap(); |
1367 | auto OB1HasNUW = |
1368 | cast<OverflowingBinaryOperator>(Val: Op1)->hasNoUnsignedWrap(); |
1369 | const APInt *C1, *C2; |
1370 | if (IsSigned && OB0HasNSW) { |
1371 | if (OB1HasNSW && match(V: B, P: m_APInt(Res&: C1)) && !C1->isAllOnes()) |
1372 | return BinaryOperator::CreateSDiv(V1: A, V2: B); |
1373 | } |
1374 | if (!IsSigned && OB0HasNUW) { |
1375 | if (OB1HasNUW) |
1376 | return BinaryOperator::CreateUDiv(V1: A, V2: B); |
1377 | if (match(V: A, P: m_APInt(Res&: C1)) && match(V: B, P: m_APInt(Res&: C2)) && C2->ule(RHS: *C1)) |
1378 | return BinaryOperator::CreateUDiv(V1: A, V2: B); |
1379 | } |
1380 | return nullptr; |
1381 | }; |
1382 | |
1383 | if (match(V: Op1, P: m_c_Mul(L: m_Specific(V: X), R: m_Value(V&: Z)))) { |
1384 | if (auto *Val = CreateDivOrNull(Y, Z)) |
1385 | return Val; |
1386 | } |
1387 | if (match(V: Op1, P: m_c_Mul(L: m_Specific(V: Y), R: m_Value(V&: Z)))) { |
1388 | if (auto *Val = CreateDivOrNull(X, Z)) |
1389 | return Val; |
1390 | } |
1391 | } |
1392 | return nullptr; |
1393 | } |
1394 | |
1395 | static const unsigned MaxDepth = 6; |
1396 | |
1397 | // Take the exact integer log2 of the value. If DoFold is true, create the |
1398 | // actual instructions, otherwise return a non-null dummy value. Return nullptr |
1399 | // on failure. |
1400 | static Value *takeLog2(IRBuilderBase &Builder, Value *Op, unsigned Depth, |
1401 | bool AssumeNonZero, bool DoFold) { |
1402 | auto IfFold = [DoFold](function_ref<Value *()> Fn) { |
1403 | if (!DoFold) |
1404 | return reinterpret_cast<Value *>(-1); |
1405 | return Fn(); |
1406 | }; |
1407 | |
1408 | // FIXME: assert that Op1 isn't/doesn't contain undef. |
1409 | |
1410 | // log2(2^C) -> C |
1411 | if (match(V: Op, P: m_Power2())) |
1412 | return IfFold([&]() { |
1413 | Constant *C = ConstantExpr::getExactLogBase2(C: cast<Constant>(Val: Op)); |
1414 | if (!C) |
1415 | llvm_unreachable("Failed to constant fold udiv -> logbase2" ); |
1416 | return C; |
1417 | }); |
1418 | |
1419 | // The remaining tests are all recursive, so bail out if we hit the limit. |
1420 | if (Depth++ == MaxDepth) |
1421 | return nullptr; |
1422 | |
1423 | // log2(zext X) -> zext log2(X) |
1424 | // FIXME: Require one use? |
1425 | Value *X, *Y; |
1426 | if (match(V: Op, P: m_ZExt(Op: m_Value(V&: X)))) |
1427 | if (Value *LogX = takeLog2(Builder, Op: X, Depth, AssumeNonZero, DoFold)) |
1428 | return IfFold([&]() { return Builder.CreateZExt(V: LogX, DestTy: Op->getType()); }); |
1429 | |
1430 | // log2(X << Y) -> log2(X) + Y |
1431 | // FIXME: Require one use unless X is 1? |
1432 | if (match(V: Op, P: m_Shl(L: m_Value(V&: X), R: m_Value(V&: Y)))) { |
1433 | auto *BO = cast<OverflowingBinaryOperator>(Val: Op); |
1434 | // nuw will be set if the `shl` is trivially non-zero. |
1435 | if (AssumeNonZero || BO->hasNoUnsignedWrap() || BO->hasNoSignedWrap()) |
1436 | if (Value *LogX = takeLog2(Builder, Op: X, Depth, AssumeNonZero, DoFold)) |
1437 | return IfFold([&]() { return Builder.CreateAdd(LHS: LogX, RHS: Y); }); |
1438 | } |
1439 | |
1440 | // log2(Cond ? X : Y) -> Cond ? log2(X) : log2(Y) |
1441 | // FIXME: Require one use? |
1442 | if (SelectInst *SI = dyn_cast<SelectInst>(Val: Op)) |
1443 | if (Value *LogX = takeLog2(Builder, Op: SI->getOperand(i_nocapture: 1), Depth, |
1444 | AssumeNonZero, DoFold)) |
1445 | if (Value *LogY = takeLog2(Builder, Op: SI->getOperand(i_nocapture: 2), Depth, |
1446 | AssumeNonZero, DoFold)) |
1447 | return IfFold([&]() { |
1448 | return Builder.CreateSelect(C: SI->getOperand(i_nocapture: 0), True: LogX, False: LogY); |
1449 | }); |
1450 | |
1451 | // log2(umin(X, Y)) -> umin(log2(X), log2(Y)) |
1452 | // log2(umax(X, Y)) -> umax(log2(X), log2(Y)) |
1453 | auto *MinMax = dyn_cast<MinMaxIntrinsic>(Val: Op); |
1454 | if (MinMax && MinMax->hasOneUse() && !MinMax->isSigned()) { |
1455 | // Use AssumeNonZero as false here. Otherwise we can hit case where |
1456 | // log2(umax(X, Y)) != umax(log2(X), log2(Y)) (because overflow). |
1457 | if (Value *LogX = takeLog2(Builder, Op: MinMax->getLHS(), Depth, |
1458 | /*AssumeNonZero*/ false, DoFold)) |
1459 | if (Value *LogY = takeLog2(Builder, Op: MinMax->getRHS(), Depth, |
1460 | /*AssumeNonZero*/ false, DoFold)) |
1461 | return IfFold([&]() { |
1462 | return Builder.CreateBinaryIntrinsic(ID: MinMax->getIntrinsicID(), LHS: LogX, |
1463 | RHS: LogY); |
1464 | }); |
1465 | } |
1466 | |
1467 | return nullptr; |
1468 | } |
1469 | |
1470 | /// If we have zero-extended operands of an unsigned div or rem, we may be able |
1471 | /// to narrow the operation (sink the zext below the math). |
1472 | static Instruction *narrowUDivURem(BinaryOperator &I, |
1473 | InstCombinerImpl &IC) { |
1474 | Instruction::BinaryOps Opcode = I.getOpcode(); |
1475 | Value *N = I.getOperand(i_nocapture: 0); |
1476 | Value *D = I.getOperand(i_nocapture: 1); |
1477 | Type *Ty = I.getType(); |
1478 | Value *X, *Y; |
1479 | if (match(V: N, P: m_ZExt(Op: m_Value(V&: X))) && match(V: D, P: m_ZExt(Op: m_Value(V&: Y))) && |
1480 | X->getType() == Y->getType() && (N->hasOneUse() || D->hasOneUse())) { |
1481 | // udiv (zext X), (zext Y) --> zext (udiv X, Y) |
1482 | // urem (zext X), (zext Y) --> zext (urem X, Y) |
1483 | Value *NarrowOp = IC.Builder.CreateBinOp(Opc: Opcode, LHS: X, RHS: Y); |
1484 | return new ZExtInst(NarrowOp, Ty); |
1485 | } |
1486 | |
1487 | Constant *C; |
1488 | if (isa<Instruction>(Val: N) && match(V: N, P: m_OneUse(SubPattern: m_ZExt(Op: m_Value(V&: X)))) && |
1489 | match(V: D, P: m_Constant(C))) { |
1490 | // If the constant is the same in the smaller type, use the narrow version. |
1491 | Constant *TruncC = IC.getLosslessUnsignedTrunc(C, TruncTy: X->getType()); |
1492 | if (!TruncC) |
1493 | return nullptr; |
1494 | |
1495 | // udiv (zext X), C --> zext (udiv X, C') |
1496 | // urem (zext X), C --> zext (urem X, C') |
1497 | return new ZExtInst(IC.Builder.CreateBinOp(Opc: Opcode, LHS: X, RHS: TruncC), Ty); |
1498 | } |
1499 | if (isa<Instruction>(Val: D) && match(V: D, P: m_OneUse(SubPattern: m_ZExt(Op: m_Value(V&: X)))) && |
1500 | match(V: N, P: m_Constant(C))) { |
1501 | // If the constant is the same in the smaller type, use the narrow version. |
1502 | Constant *TruncC = IC.getLosslessUnsignedTrunc(C, TruncTy: X->getType()); |
1503 | if (!TruncC) |
1504 | return nullptr; |
1505 | |
1506 | // udiv C, (zext X) --> zext (udiv C', X) |
1507 | // urem C, (zext X) --> zext (urem C', X) |
1508 | return new ZExtInst(IC.Builder.CreateBinOp(Opc: Opcode, LHS: TruncC, RHS: X), Ty); |
1509 | } |
1510 | |
1511 | return nullptr; |
1512 | } |
1513 | |
1514 | Instruction *InstCombinerImpl::visitUDiv(BinaryOperator &I) { |
1515 | if (Value *V = simplifyUDivInst(LHS: I.getOperand(i_nocapture: 0), RHS: I.getOperand(i_nocapture: 1), IsExact: I.isExact(), |
1516 | Q: SQ.getWithInstruction(I: &I))) |
1517 | return replaceInstUsesWith(I, V); |
1518 | |
1519 | if (Instruction *X = foldVectorBinop(Inst&: I)) |
1520 | return X; |
1521 | |
1522 | // Handle the integer div common cases |
1523 | if (Instruction *Common = commonIDivTransforms(I)) |
1524 | return Common; |
1525 | |
1526 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
1527 | Value *X; |
1528 | const APInt *C1, *C2; |
1529 | if (match(V: Op0, P: m_LShr(L: m_Value(V&: X), R: m_APInt(Res&: C1))) && match(V: Op1, P: m_APInt(Res&: C2))) { |
1530 | // (X lshr C1) udiv C2 --> X udiv (C2 << C1) |
1531 | bool Overflow; |
1532 | APInt C2ShlC1 = C2->ushl_ov(Amt: *C1, Overflow); |
1533 | if (!Overflow) { |
1534 | bool IsExact = I.isExact() && match(V: Op0, P: m_Exact(SubPattern: m_Value())); |
1535 | BinaryOperator *BO = BinaryOperator::CreateUDiv( |
1536 | V1: X, V2: ConstantInt::get(Ty: X->getType(), V: C2ShlC1)); |
1537 | if (IsExact) |
1538 | BO->setIsExact(); |
1539 | return BO; |
1540 | } |
1541 | } |
1542 | |
1543 | // Op0 / C where C is large (negative) --> zext (Op0 >= C) |
1544 | // TODO: Could use isKnownNegative() to handle non-constant values. |
1545 | Type *Ty = I.getType(); |
1546 | if (match(V: Op1, P: m_Negative())) { |
1547 | Value *Cmp = Builder.CreateICmpUGE(LHS: Op0, RHS: Op1); |
1548 | return CastInst::CreateZExtOrBitCast(S: Cmp, Ty); |
1549 | } |
1550 | // Op0 / (sext i1 X) --> zext (Op0 == -1) (if X is 0, the div is undefined) |
1551 | if (match(V: Op1, P: m_SExt(Op: m_Value(V&: X))) && X->getType()->isIntOrIntVectorTy(BitWidth: 1)) { |
1552 | Value *Cmp = Builder.CreateICmpEQ(LHS: Op0, RHS: ConstantInt::getAllOnesValue(Ty)); |
1553 | return CastInst::CreateZExtOrBitCast(S: Cmp, Ty); |
1554 | } |
1555 | |
1556 | if (Instruction *NarrowDiv = narrowUDivURem(I, IC&: *this)) |
1557 | return NarrowDiv; |
1558 | |
1559 | Value *A, *B; |
1560 | |
1561 | // Look through a right-shift to find the common factor: |
1562 | // ((Op1 *nuw A) >> B) / Op1 --> A >> B |
1563 | if (match(V: Op0, P: m_LShr(L: m_NUWMul(L: m_Specific(V: Op1), R: m_Value(V&: A)), R: m_Value(V&: B))) || |
1564 | match(V: Op0, P: m_LShr(L: m_NUWMul(L: m_Value(V&: A), R: m_Specific(V: Op1)), R: m_Value(V&: B)))) { |
1565 | Instruction *Lshr = BinaryOperator::CreateLShr(V1: A, V2: B); |
1566 | if (I.isExact() && cast<PossiblyExactOperator>(Val: Op0)->isExact()) |
1567 | Lshr->setIsExact(); |
1568 | return Lshr; |
1569 | } |
1570 | |
1571 | // Op1 udiv Op2 -> Op1 lshr log2(Op2), if log2() folds away. |
1572 | if (takeLog2(Builder, Op: Op1, /*Depth*/ 0, /*AssumeNonZero*/ true, |
1573 | /*DoFold*/ false)) { |
1574 | Value *Res = takeLog2(Builder, Op: Op1, /*Depth*/ 0, |
1575 | /*AssumeNonZero*/ true, /*DoFold*/ true); |
1576 | return replaceInstUsesWith( |
1577 | I, V: Builder.CreateLShr(LHS: Op0, RHS: Res, Name: I.getName(), isExact: I.isExact())); |
1578 | } |
1579 | |
1580 | return nullptr; |
1581 | } |
1582 | |
1583 | Instruction *InstCombinerImpl::visitSDiv(BinaryOperator &I) { |
1584 | if (Value *V = simplifySDivInst(LHS: I.getOperand(i_nocapture: 0), RHS: I.getOperand(i_nocapture: 1), IsExact: I.isExact(), |
1585 | Q: SQ.getWithInstruction(I: &I))) |
1586 | return replaceInstUsesWith(I, V); |
1587 | |
1588 | if (Instruction *X = foldVectorBinop(Inst&: I)) |
1589 | return X; |
1590 | |
1591 | // Handle the integer div common cases |
1592 | if (Instruction *Common = commonIDivTransforms(I)) |
1593 | return Common; |
1594 | |
1595 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
1596 | Type *Ty = I.getType(); |
1597 | Value *X; |
1598 | // sdiv Op0, -1 --> -Op0 |
1599 | // sdiv Op0, (sext i1 X) --> -Op0 (because if X is 0, the op is undefined) |
1600 | if (match(V: Op1, P: m_AllOnes()) || |
1601 | (match(V: Op1, P: m_SExt(Op: m_Value(V&: X))) && X->getType()->isIntOrIntVectorTy(BitWidth: 1))) |
1602 | return BinaryOperator::CreateNSWNeg(Op: Op0); |
1603 | |
1604 | // X / INT_MIN --> X == INT_MIN |
1605 | if (match(V: Op1, P: m_SignMask())) |
1606 | return new ZExtInst(Builder.CreateICmpEQ(LHS: Op0, RHS: Op1), Ty); |
1607 | |
1608 | if (I.isExact()) { |
1609 | // sdiv exact X, 1<<C --> ashr exact X, C iff 1<<C is non-negative |
1610 | if (match(V: Op1, P: m_Power2()) && match(V: Op1, P: m_NonNegative())) { |
1611 | Constant *C = ConstantExpr::getExactLogBase2(C: cast<Constant>(Val: Op1)); |
1612 | return BinaryOperator::CreateExactAShr(V1: Op0, V2: C); |
1613 | } |
1614 | |
1615 | // sdiv exact X, (1<<ShAmt) --> ashr exact X, ShAmt (if shl is non-negative) |
1616 | Value *ShAmt; |
1617 | if (match(V: Op1, P: m_NSWShl(L: m_One(), R: m_Value(V&: ShAmt)))) |
1618 | return BinaryOperator::CreateExactAShr(V1: Op0, V2: ShAmt); |
1619 | |
1620 | // sdiv exact X, -1<<C --> -(ashr exact X, C) |
1621 | if (match(V: Op1, P: m_NegatedPower2())) { |
1622 | Constant *NegPow2C = ConstantExpr::getNeg(C: cast<Constant>(Val: Op1)); |
1623 | Constant *C = ConstantExpr::getExactLogBase2(C: NegPow2C); |
1624 | Value *Ashr = Builder.CreateAShr(LHS: Op0, RHS: C, Name: I.getName() + ".neg" , isExact: true); |
1625 | return BinaryOperator::CreateNSWNeg(Op: Ashr); |
1626 | } |
1627 | } |
1628 | |
1629 | const APInt *Op1C; |
1630 | if (match(V: Op1, P: m_APInt(Res&: Op1C))) { |
1631 | // If the dividend is sign-extended and the constant divisor is small enough |
1632 | // to fit in the source type, shrink the division to the narrower type: |
1633 | // (sext X) sdiv C --> sext (X sdiv C) |
1634 | Value *Op0Src; |
1635 | if (match(V: Op0, P: m_OneUse(SubPattern: m_SExt(Op: m_Value(V&: Op0Src)))) && |
1636 | Op0Src->getType()->getScalarSizeInBits() >= |
1637 | Op1C->getSignificantBits()) { |
1638 | |
1639 | // In the general case, we need to make sure that the dividend is not the |
1640 | // minimum signed value because dividing that by -1 is UB. But here, we |
1641 | // know that the -1 divisor case is already handled above. |
1642 | |
1643 | Constant *NarrowDivisor = |
1644 | ConstantExpr::getTrunc(C: cast<Constant>(Val: Op1), Ty: Op0Src->getType()); |
1645 | Value *NarrowOp = Builder.CreateSDiv(LHS: Op0Src, RHS: NarrowDivisor); |
1646 | return new SExtInst(NarrowOp, Ty); |
1647 | } |
1648 | |
1649 | // -X / C --> X / -C (if the negation doesn't overflow). |
1650 | // TODO: This could be enhanced to handle arbitrary vector constants by |
1651 | // checking if all elements are not the min-signed-val. |
1652 | if (!Op1C->isMinSignedValue() && match(V: Op0, P: m_NSWNeg(V: m_Value(V&: X)))) { |
1653 | Constant *NegC = ConstantInt::get(Ty, V: -(*Op1C)); |
1654 | Instruction *BO = BinaryOperator::CreateSDiv(V1: X, V2: NegC); |
1655 | BO->setIsExact(I.isExact()); |
1656 | return BO; |
1657 | } |
1658 | } |
1659 | |
1660 | // -X / Y --> -(X / Y) |
1661 | Value *Y; |
1662 | if (match(V: &I, P: m_SDiv(L: m_OneUse(SubPattern: m_NSWNeg(V: m_Value(V&: X))), R: m_Value(V&: Y)))) |
1663 | return BinaryOperator::CreateNSWNeg( |
1664 | Op: Builder.CreateSDiv(LHS: X, RHS: Y, Name: I.getName(), isExact: I.isExact())); |
1665 | |
1666 | // abs(X) / X --> X > -1 ? 1 : -1 |
1667 | // X / abs(X) --> X > -1 ? 1 : -1 |
1668 | if (match(V: &I, P: m_c_BinOp( |
1669 | L: m_OneUse(SubPattern: m_Intrinsic<Intrinsic::abs>(Op0: m_Value(V&: X), Op1: m_One())), |
1670 | R: m_Deferred(V: X)))) { |
1671 | Value *Cond = Builder.CreateIsNotNeg(Arg: X); |
1672 | return SelectInst::Create(C: Cond, S1: ConstantInt::get(Ty, V: 1), |
1673 | S2: ConstantInt::getAllOnesValue(Ty)); |
1674 | } |
1675 | |
1676 | KnownBits KnownDividend = computeKnownBits(V: Op0, Depth: 0, CxtI: &I); |
1677 | if (!I.isExact() && |
1678 | (match(V: Op1, P: m_Power2(V&: Op1C)) || match(V: Op1, P: m_NegatedPower2(V&: Op1C))) && |
1679 | KnownDividend.countMinTrailingZeros() >= Op1C->countr_zero()) { |
1680 | I.setIsExact(); |
1681 | return &I; |
1682 | } |
1683 | |
1684 | if (KnownDividend.isNonNegative()) { |
1685 | // If both operands are unsigned, turn this into a udiv. |
1686 | if (isKnownNonNegative(V: Op1, SQ: SQ.getWithInstruction(I: &I))) { |
1687 | auto *BO = BinaryOperator::CreateUDiv(V1: Op0, V2: Op1, Name: I.getName()); |
1688 | BO->setIsExact(I.isExact()); |
1689 | return BO; |
1690 | } |
1691 | |
1692 | if (match(V: Op1, P: m_NegatedPower2())) { |
1693 | // X sdiv (-(1 << C)) -> -(X sdiv (1 << C)) -> |
1694 | // -> -(X udiv (1 << C)) -> -(X u>> C) |
1695 | Constant *CNegLog2 = ConstantExpr::getExactLogBase2( |
1696 | C: ConstantExpr::getNeg(C: cast<Constant>(Val: Op1))); |
1697 | Value *Shr = Builder.CreateLShr(LHS: Op0, RHS: CNegLog2, Name: I.getName(), isExact: I.isExact()); |
1698 | return BinaryOperator::CreateNeg(Op: Shr); |
1699 | } |
1700 | |
1701 | if (isKnownToBeAPowerOfTwo(V: Op1, /*OrZero*/ true, Depth: 0, CxtI: &I)) { |
1702 | // X sdiv (1 << Y) -> X udiv (1 << Y) ( -> X u>> Y) |
1703 | // Safe because the only negative value (1 << Y) can take on is |
1704 | // INT_MIN, and X sdiv INT_MIN == X udiv INT_MIN == 0 if X doesn't have |
1705 | // the sign bit set. |
1706 | auto *BO = BinaryOperator::CreateUDiv(V1: Op0, V2: Op1, Name: I.getName()); |
1707 | BO->setIsExact(I.isExact()); |
1708 | return BO; |
1709 | } |
1710 | } |
1711 | |
1712 | // -X / X --> X == INT_MIN ? 1 : -1 |
1713 | if (isKnownNegation(X: Op0, Y: Op1)) { |
1714 | APInt MinVal = APInt::getSignedMinValue(numBits: Ty->getScalarSizeInBits()); |
1715 | Value *Cond = Builder.CreateICmpEQ(LHS: Op0, RHS: ConstantInt::get(Ty, V: MinVal)); |
1716 | return SelectInst::Create(C: Cond, S1: ConstantInt::get(Ty, V: 1), |
1717 | S2: ConstantInt::getAllOnesValue(Ty)); |
1718 | } |
1719 | return nullptr; |
1720 | } |
1721 | |
1722 | /// Remove negation and try to convert division into multiplication. |
1723 | Instruction *InstCombinerImpl::foldFDivConstantDivisor(BinaryOperator &I) { |
1724 | Constant *C; |
1725 | if (!match(V: I.getOperand(i_nocapture: 1), P: m_Constant(C))) |
1726 | return nullptr; |
1727 | |
1728 | // -X / C --> X / -C |
1729 | Value *X; |
1730 | const DataLayout &DL = I.getDataLayout(); |
1731 | if (match(V: I.getOperand(i_nocapture: 0), P: m_FNeg(X: m_Value(V&: X)))) |
1732 | if (Constant *NegC = ConstantFoldUnaryOpOperand(Opcode: Instruction::FNeg, Op: C, DL)) |
1733 | return BinaryOperator::CreateFDivFMF(V1: X, V2: NegC, FMFSource: &I); |
1734 | |
1735 | // nnan X / +0.0 -> copysign(inf, X) |
1736 | // nnan nsz X / -0.0 -> copysign(inf, X) |
1737 | if (I.hasNoNaNs() && |
1738 | (match(V: I.getOperand(i_nocapture: 1), P: m_PosZeroFP()) || |
1739 | (I.hasNoSignedZeros() && match(V: I.getOperand(i_nocapture: 1), P: m_AnyZeroFP())))) { |
1740 | IRBuilder<> B(&I); |
1741 | CallInst *CopySign = B.CreateIntrinsic( |
1742 | ID: Intrinsic::copysign, Types: {C->getType()}, |
1743 | Args: {ConstantFP::getInfinity(Ty: I.getType()), I.getOperand(i_nocapture: 0)}, FMFSource: &I); |
1744 | CopySign->takeName(V: &I); |
1745 | return replaceInstUsesWith(I, V: CopySign); |
1746 | } |
1747 | |
1748 | // If the constant divisor has an exact inverse, this is always safe. If not, |
1749 | // then we can still create a reciprocal if fast-math-flags allow it and the |
1750 | // constant is a regular number (not zero, infinite, or denormal). |
1751 | if (!(C->hasExactInverseFP() || (I.hasAllowReciprocal() && C->isNormalFP()))) |
1752 | return nullptr; |
1753 | |
1754 | // Disallow denormal constants because we don't know what would happen |
1755 | // on all targets. |
1756 | // TODO: Use Intrinsic::canonicalize or let function attributes tell us that |
1757 | // denorms are flushed? |
1758 | auto *RecipC = ConstantFoldBinaryOpOperands( |
1759 | Opcode: Instruction::FDiv, LHS: ConstantFP::get(Ty: I.getType(), V: 1.0), RHS: C, DL); |
1760 | if (!RecipC || !RecipC->isNormalFP()) |
1761 | return nullptr; |
1762 | |
1763 | // X / C --> X * (1 / C) |
1764 | return BinaryOperator::CreateFMulFMF(V1: I.getOperand(i_nocapture: 0), V2: RecipC, FMFSource: &I); |
1765 | } |
1766 | |
1767 | /// Remove negation and try to reassociate constant math. |
1768 | static Instruction *foldFDivConstantDividend(BinaryOperator &I) { |
1769 | Constant *C; |
1770 | if (!match(V: I.getOperand(i_nocapture: 0), P: m_Constant(C))) |
1771 | return nullptr; |
1772 | |
1773 | // C / -X --> -C / X |
1774 | Value *X; |
1775 | const DataLayout &DL = I.getDataLayout(); |
1776 | if (match(V: I.getOperand(i_nocapture: 1), P: m_FNeg(X: m_Value(V&: X)))) |
1777 | if (Constant *NegC = ConstantFoldUnaryOpOperand(Opcode: Instruction::FNeg, Op: C, DL)) |
1778 | return BinaryOperator::CreateFDivFMF(V1: NegC, V2: X, FMFSource: &I); |
1779 | |
1780 | if (!I.hasAllowReassoc() || !I.hasAllowReciprocal()) |
1781 | return nullptr; |
1782 | |
1783 | // Try to reassociate C / X expressions where X includes another constant. |
1784 | Constant *C2, *NewC = nullptr; |
1785 | if (match(V: I.getOperand(i_nocapture: 1), P: m_FMul(L: m_Value(V&: X), R: m_Constant(C&: C2)))) { |
1786 | // C / (X * C2) --> (C / C2) / X |
1787 | NewC = ConstantFoldBinaryOpOperands(Opcode: Instruction::FDiv, LHS: C, RHS: C2, DL); |
1788 | } else if (match(V: I.getOperand(i_nocapture: 1), P: m_FDiv(L: m_Value(V&: X), R: m_Constant(C&: C2)))) { |
1789 | // C / (X / C2) --> (C * C2) / X |
1790 | NewC = ConstantFoldBinaryOpOperands(Opcode: Instruction::FMul, LHS: C, RHS: C2, DL); |
1791 | } |
1792 | // Disallow denormal constants because we don't know what would happen |
1793 | // on all targets. |
1794 | // TODO: Use Intrinsic::canonicalize or let function attributes tell us that |
1795 | // denorms are flushed? |
1796 | if (!NewC || !NewC->isNormalFP()) |
1797 | return nullptr; |
1798 | |
1799 | return BinaryOperator::CreateFDivFMF(V1: NewC, V2: X, FMFSource: &I); |
1800 | } |
1801 | |
1802 | /// Negate the exponent of pow/exp to fold division-by-pow() into multiply. |
1803 | static Instruction *foldFDivPowDivisor(BinaryOperator &I, |
1804 | InstCombiner::BuilderTy &Builder) { |
1805 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
1806 | auto *II = dyn_cast<IntrinsicInst>(Val: Op1); |
1807 | if (!II || !II->hasOneUse() || !I.hasAllowReassoc() || |
1808 | !I.hasAllowReciprocal()) |
1809 | return nullptr; |
1810 | |
1811 | // Z / pow(X, Y) --> Z * pow(X, -Y) |
1812 | // Z / exp{2}(Y) --> Z * exp{2}(-Y) |
1813 | // In the general case, this creates an extra instruction, but fmul allows |
1814 | // for better canonicalization and optimization than fdiv. |
1815 | Intrinsic::ID IID = II->getIntrinsicID(); |
1816 | SmallVector<Value *> Args; |
1817 | switch (IID) { |
1818 | case Intrinsic::pow: |
1819 | Args.push_back(Elt: II->getArgOperand(i: 0)); |
1820 | Args.push_back(Elt: Builder.CreateFNegFMF(V: II->getArgOperand(i: 1), FMFSource: &I)); |
1821 | break; |
1822 | case Intrinsic::powi: { |
1823 | // Require 'ninf' assuming that makes powi(X, -INT_MIN) acceptable. |
1824 | // That is, X ** (huge negative number) is 0.0, ~1.0, or INF and so |
1825 | // dividing by that is INF, ~1.0, or 0.0. Code that uses powi allows |
1826 | // non-standard results, so this corner case should be acceptable if the |
1827 | // code rules out INF values. |
1828 | if (!I.hasNoInfs()) |
1829 | return nullptr; |
1830 | Args.push_back(Elt: II->getArgOperand(i: 0)); |
1831 | Args.push_back(Elt: Builder.CreateNeg(V: II->getArgOperand(i: 1))); |
1832 | Type *Tys[] = {I.getType(), II->getArgOperand(i: 1)->getType()}; |
1833 | Value *Pow = Builder.CreateIntrinsic(ID: IID, Types: Tys, Args, FMFSource: &I); |
1834 | return BinaryOperator::CreateFMulFMF(V1: Op0, V2: Pow, FMFSource: &I); |
1835 | } |
1836 | case Intrinsic::exp: |
1837 | case Intrinsic::exp2: |
1838 | Args.push_back(Elt: Builder.CreateFNegFMF(V: II->getArgOperand(i: 0), FMFSource: &I)); |
1839 | break; |
1840 | default: |
1841 | return nullptr; |
1842 | } |
1843 | Value *Pow = Builder.CreateIntrinsic(ID: IID, Types: I.getType(), Args, FMFSource: &I); |
1844 | return BinaryOperator::CreateFMulFMF(V1: Op0, V2: Pow, FMFSource: &I); |
1845 | } |
1846 | |
1847 | /// Convert div to mul if we have an sqrt divisor iff sqrt's operand is a fdiv |
1848 | /// instruction. |
1849 | static Instruction *foldFDivSqrtDivisor(BinaryOperator &I, |
1850 | InstCombiner::BuilderTy &Builder) { |
1851 | // X / sqrt(Y / Z) --> X * sqrt(Z / Y) |
1852 | if (!I.hasAllowReassoc() || !I.hasAllowReciprocal()) |
1853 | return nullptr; |
1854 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
1855 | auto *II = dyn_cast<IntrinsicInst>(Val: Op1); |
1856 | if (!II || II->getIntrinsicID() != Intrinsic::sqrt || !II->hasOneUse() || |
1857 | !II->hasAllowReassoc() || !II->hasAllowReciprocal()) |
1858 | return nullptr; |
1859 | |
1860 | Value *Y, *Z; |
1861 | auto *DivOp = dyn_cast<Instruction>(Val: II->getOperand(i_nocapture: 0)); |
1862 | if (!DivOp) |
1863 | return nullptr; |
1864 | if (!match(V: DivOp, P: m_FDiv(L: m_Value(V&: Y), R: m_Value(V&: Z)))) |
1865 | return nullptr; |
1866 | if (!DivOp->hasAllowReassoc() || !I.hasAllowReciprocal() || |
1867 | !DivOp->hasOneUse()) |
1868 | return nullptr; |
1869 | Value *SwapDiv = Builder.CreateFDivFMF(L: Z, R: Y, FMFSource: DivOp); |
1870 | Value *NewSqrt = |
1871 | Builder.CreateUnaryIntrinsic(ID: II->getIntrinsicID(), V: SwapDiv, FMFSource: II); |
1872 | return BinaryOperator::CreateFMulFMF(V1: Op0, V2: NewSqrt, FMFSource: &I); |
1873 | } |
1874 | |
1875 | Instruction *InstCombinerImpl::visitFDiv(BinaryOperator &I) { |
1876 | Module *M = I.getModule(); |
1877 | |
1878 | if (Value *V = simplifyFDivInst(LHS: I.getOperand(i_nocapture: 0), RHS: I.getOperand(i_nocapture: 1), |
1879 | FMF: I.getFastMathFlags(), |
1880 | Q: SQ.getWithInstruction(I: &I))) |
1881 | return replaceInstUsesWith(I, V); |
1882 | |
1883 | if (Instruction *X = foldVectorBinop(Inst&: I)) |
1884 | return X; |
1885 | |
1886 | if (Instruction *Phi = foldBinopWithPhiOperands(BO&: I)) |
1887 | return Phi; |
1888 | |
1889 | if (Instruction *R = foldFDivConstantDivisor(I)) |
1890 | return R; |
1891 | |
1892 | if (Instruction *R = foldFDivConstantDividend(I)) |
1893 | return R; |
1894 | |
1895 | if (Instruction *R = foldFPSignBitOps(I)) |
1896 | return R; |
1897 | |
1898 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
1899 | if (isa<Constant>(Val: Op0)) |
1900 | if (SelectInst *SI = dyn_cast<SelectInst>(Val: Op1)) |
1901 | if (Instruction *R = FoldOpIntoSelect(Op&: I, SI)) |
1902 | return R; |
1903 | |
1904 | if (isa<Constant>(Val: Op1)) |
1905 | if (SelectInst *SI = dyn_cast<SelectInst>(Val: Op0)) |
1906 | if (Instruction *R = FoldOpIntoSelect(Op&: I, SI)) |
1907 | return R; |
1908 | |
1909 | if (I.hasAllowReassoc() && I.hasAllowReciprocal()) { |
1910 | Value *X, *Y; |
1911 | if (match(V: Op0, P: m_OneUse(SubPattern: m_FDiv(L: m_Value(V&: X), R: m_Value(V&: Y)))) && |
1912 | (!isa<Constant>(Val: Y) || !isa<Constant>(Val: Op1))) { |
1913 | // (X / Y) / Z => X / (Y * Z) |
1914 | Value *YZ = Builder.CreateFMulFMF(L: Y, R: Op1, FMFSource: &I); |
1915 | return BinaryOperator::CreateFDivFMF(V1: X, V2: YZ, FMFSource: &I); |
1916 | } |
1917 | if (match(V: Op1, P: m_OneUse(SubPattern: m_FDiv(L: m_Value(V&: X), R: m_Value(V&: Y)))) && |
1918 | (!isa<Constant>(Val: Y) || !isa<Constant>(Val: Op0))) { |
1919 | // Z / (X / Y) => (Y * Z) / X |
1920 | Value *YZ = Builder.CreateFMulFMF(L: Y, R: Op0, FMFSource: &I); |
1921 | return BinaryOperator::CreateFDivFMF(V1: YZ, V2: X, FMFSource: &I); |
1922 | } |
1923 | // Z / (1.0 / Y) => (Y * Z) |
1924 | // |
1925 | // This is a special case of Z / (X / Y) => (Y * Z) / X, with X = 1.0. The |
1926 | // m_OneUse check is avoided because even in the case of the multiple uses |
1927 | // for 1.0/Y, the number of instructions remain the same and a division is |
1928 | // replaced by a multiplication. |
1929 | if (match(V: Op1, P: m_FDiv(L: m_SpecificFP(V: 1.0), R: m_Value(V&: Y)))) |
1930 | return BinaryOperator::CreateFMulFMF(V1: Y, V2: Op0, FMFSource: &I); |
1931 | } |
1932 | |
1933 | if (I.hasAllowReassoc() && Op0->hasOneUse() && Op1->hasOneUse()) { |
1934 | // sin(X) / cos(X) -> tan(X) |
1935 | // cos(X) / sin(X) -> 1/tan(X) (cotangent) |
1936 | Value *X; |
1937 | bool IsTan = match(V: Op0, P: m_Intrinsic<Intrinsic::sin>(Op0: m_Value(V&: X))) && |
1938 | match(V: Op1, P: m_Intrinsic<Intrinsic::cos>(Op0: m_Specific(V: X))); |
1939 | bool IsCot = |
1940 | !IsTan && match(V: Op0, P: m_Intrinsic<Intrinsic::cos>(Op0: m_Value(V&: X))) && |
1941 | match(V: Op1, P: m_Intrinsic<Intrinsic::sin>(Op0: m_Specific(V: X))); |
1942 | |
1943 | if ((IsTan || IsCot) && hasFloatFn(M, TLI: &TLI, Ty: I.getType(), DoubleFn: LibFunc_tan, |
1944 | FloatFn: LibFunc_tanf, LongDoubleFn: LibFunc_tanl)) { |
1945 | IRBuilder<> B(&I); |
1946 | IRBuilder<>::FastMathFlagGuard FMFGuard(B); |
1947 | B.setFastMathFlags(I.getFastMathFlags()); |
1948 | AttributeList Attrs = |
1949 | cast<CallBase>(Val: Op0)->getCalledFunction()->getAttributes(); |
1950 | Value *Res = emitUnaryFloatFnCall(Op: X, TLI: &TLI, DoubleFn: LibFunc_tan, FloatFn: LibFunc_tanf, |
1951 | LongDoubleFn: LibFunc_tanl, B, Attrs); |
1952 | if (IsCot) |
1953 | Res = B.CreateFDiv(L: ConstantFP::get(Ty: I.getType(), V: 1.0), R: Res); |
1954 | return replaceInstUsesWith(I, V: Res); |
1955 | } |
1956 | } |
1957 | |
1958 | // X / (X * Y) --> 1.0 / Y |
1959 | // Reassociate to (X / X -> 1.0) is legal when NaNs are not allowed. |
1960 | // We can ignore the possibility that X is infinity because INF/INF is NaN. |
1961 | Value *X, *Y; |
1962 | if (I.hasNoNaNs() && I.hasAllowReassoc() && |
1963 | match(V: Op1, P: m_c_FMul(L: m_Specific(V: Op0), R: m_Value(V&: Y)))) { |
1964 | replaceOperand(I, OpNum: 0, V: ConstantFP::get(Ty: I.getType(), V: 1.0)); |
1965 | replaceOperand(I, OpNum: 1, V: Y); |
1966 | return &I; |
1967 | } |
1968 | |
1969 | // X / fabs(X) -> copysign(1.0, X) |
1970 | // fabs(X) / X -> copysign(1.0, X) |
1971 | if (I.hasNoNaNs() && I.hasNoInfs() && |
1972 | (match(V: &I, P: m_FDiv(L: m_Value(V&: X), R: m_FAbs(Op0: m_Deferred(V: X)))) || |
1973 | match(V: &I, P: m_FDiv(L: m_FAbs(Op0: m_Value(V&: X)), R: m_Deferred(V: X))))) { |
1974 | Value *V = Builder.CreateBinaryIntrinsic( |
1975 | ID: Intrinsic::copysign, LHS: ConstantFP::get(Ty: I.getType(), V: 1.0), RHS: X, FMFSource: &I); |
1976 | return replaceInstUsesWith(I, V); |
1977 | } |
1978 | |
1979 | if (Instruction *Mul = foldFDivPowDivisor(I, Builder)) |
1980 | return Mul; |
1981 | |
1982 | if (Instruction *Mul = foldFDivSqrtDivisor(I, Builder)) |
1983 | return Mul; |
1984 | |
1985 | // pow(X, Y) / X --> pow(X, Y-1) |
1986 | if (I.hasAllowReassoc() && |
1987 | match(V: Op0, P: m_OneUse(SubPattern: m_Intrinsic<Intrinsic::pow>(Op0: m_Specific(V: Op1), |
1988 | Op1: m_Value(V&: Y))))) { |
1989 | Value *Y1 = |
1990 | Builder.CreateFAddFMF(L: Y, R: ConstantFP::get(Ty: I.getType(), V: -1.0), FMFSource: &I); |
1991 | Value *Pow = Builder.CreateBinaryIntrinsic(ID: Intrinsic::pow, LHS: Op1, RHS: Y1, FMFSource: &I); |
1992 | return replaceInstUsesWith(I, V: Pow); |
1993 | } |
1994 | |
1995 | if (Instruction *FoldedPowi = foldPowiReassoc(I)) |
1996 | return FoldedPowi; |
1997 | |
1998 | return nullptr; |
1999 | } |
2000 | |
2001 | // Variety of transform for: |
2002 | // (urem/srem (mul X, Y), (mul X, Z)) |
2003 | // (urem/srem (shl X, Y), (shl X, Z)) |
2004 | // (urem/srem (shl Y, X), (shl Z, X)) |
2005 | // NB: The shift cases are really just extensions of the mul case. We treat |
2006 | // shift as Val * (1 << Amt). |
2007 | static Instruction *simplifyIRemMulShl(BinaryOperator &I, |
2008 | InstCombinerImpl &IC) { |
2009 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1), *X = nullptr; |
2010 | APInt Y, Z; |
2011 | bool ShiftByX = false; |
2012 | |
2013 | // If V is not nullptr, it will be matched using m_Specific. |
2014 | auto MatchShiftOrMulXC = [](Value *Op, Value *&V, APInt &C) -> bool { |
2015 | const APInt *Tmp = nullptr; |
2016 | if ((!V && match(V: Op, P: m_Mul(L: m_Value(V), R: m_APInt(Res&: Tmp)))) || |
2017 | (V && match(V: Op, P: m_Mul(L: m_Specific(V), R: m_APInt(Res&: Tmp))))) |
2018 | C = *Tmp; |
2019 | else if ((!V && match(V: Op, P: m_Shl(L: m_Value(V), R: m_APInt(Res&: Tmp)))) || |
2020 | (V && match(V: Op, P: m_Shl(L: m_Specific(V), R: m_APInt(Res&: Tmp))))) |
2021 | C = APInt(Tmp->getBitWidth(), 1) << *Tmp; |
2022 | if (Tmp != nullptr) |
2023 | return true; |
2024 | |
2025 | // Reset `V` so we don't start with specific value on next match attempt. |
2026 | V = nullptr; |
2027 | return false; |
2028 | }; |
2029 | |
2030 | auto MatchShiftCX = [](Value *Op, APInt &C, Value *&V) -> bool { |
2031 | const APInt *Tmp = nullptr; |
2032 | if ((!V && match(V: Op, P: m_Shl(L: m_APInt(Res&: Tmp), R: m_Value(V)))) || |
2033 | (V && match(V: Op, P: m_Shl(L: m_APInt(Res&: Tmp), R: m_Specific(V))))) { |
2034 | C = *Tmp; |
2035 | return true; |
2036 | } |
2037 | |
2038 | // Reset `V` so we don't start with specific value on next match attempt. |
2039 | V = nullptr; |
2040 | return false; |
2041 | }; |
2042 | |
2043 | if (MatchShiftOrMulXC(Op0, X, Y) && MatchShiftOrMulXC(Op1, X, Z)) { |
2044 | // pass |
2045 | } else if (MatchShiftCX(Op0, Y, X) && MatchShiftCX(Op1, Z, X)) { |
2046 | ShiftByX = true; |
2047 | } else { |
2048 | return nullptr; |
2049 | } |
2050 | |
2051 | bool IsSRem = I.getOpcode() == Instruction::SRem; |
2052 | |
2053 | OverflowingBinaryOperator *BO0 = cast<OverflowingBinaryOperator>(Val: Op0); |
2054 | // TODO: We may be able to deduce more about nsw/nuw of BO0/BO1 based on Y >= |
2055 | // Z or Z >= Y. |
2056 | bool BO0HasNSW = BO0->hasNoSignedWrap(); |
2057 | bool BO0HasNUW = BO0->hasNoUnsignedWrap(); |
2058 | bool BO0NoWrap = IsSRem ? BO0HasNSW : BO0HasNUW; |
2059 | |
2060 | APInt RemYZ = IsSRem ? Y.srem(RHS: Z) : Y.urem(RHS: Z); |
2061 | // (rem (mul nuw/nsw X, Y), (mul X, Z)) |
2062 | // if (rem Y, Z) == 0 |
2063 | // -> 0 |
2064 | if (RemYZ.isZero() && BO0NoWrap) |
2065 | return IC.replaceInstUsesWith(I, V: ConstantInt::getNullValue(Ty: I.getType())); |
2066 | |
2067 | // Helper function to emit either (RemSimplificationC << X) or |
2068 | // (RemSimplificationC * X) depending on whether we matched Op0/Op1 as |
2069 | // (shl V, X) or (mul V, X) respectively. |
2070 | auto CreateMulOrShift = |
2071 | [&](const APInt &RemSimplificationC) -> BinaryOperator * { |
2072 | Value *RemSimplification = |
2073 | ConstantInt::get(Ty: I.getType(), V: RemSimplificationC); |
2074 | return ShiftByX ? BinaryOperator::CreateShl(V1: RemSimplification, V2: X) |
2075 | : BinaryOperator::CreateMul(V1: X, V2: RemSimplification); |
2076 | }; |
2077 | |
2078 | OverflowingBinaryOperator *BO1 = cast<OverflowingBinaryOperator>(Val: Op1); |
2079 | bool BO1HasNSW = BO1->hasNoSignedWrap(); |
2080 | bool BO1HasNUW = BO1->hasNoUnsignedWrap(); |
2081 | bool BO1NoWrap = IsSRem ? BO1HasNSW : BO1HasNUW; |
2082 | // (rem (mul X, Y), (mul nuw/nsw X, Z)) |
2083 | // if (rem Y, Z) == Y |
2084 | // -> (mul nuw/nsw X, Y) |
2085 | if (RemYZ == Y && BO1NoWrap) { |
2086 | BinaryOperator *BO = CreateMulOrShift(Y); |
2087 | // Copy any overflow flags from Op0. |
2088 | BO->setHasNoSignedWrap(IsSRem || BO0HasNSW); |
2089 | BO->setHasNoUnsignedWrap(!IsSRem || BO0HasNUW); |
2090 | return BO; |
2091 | } |
2092 | |
2093 | // (rem (mul nuw/nsw X, Y), (mul {nsw} X, Z)) |
2094 | // if Y >= Z |
2095 | // -> (mul {nuw} nsw X, (rem Y, Z)) |
2096 | if (Y.uge(RHS: Z) && (IsSRem ? (BO0HasNSW && BO1HasNSW) : BO0HasNUW)) { |
2097 | BinaryOperator *BO = CreateMulOrShift(RemYZ); |
2098 | BO->setHasNoSignedWrap(); |
2099 | BO->setHasNoUnsignedWrap(BO0HasNUW); |
2100 | return BO; |
2101 | } |
2102 | |
2103 | return nullptr; |
2104 | } |
2105 | |
2106 | /// This function implements the transforms common to both integer remainder |
2107 | /// instructions (urem and srem). It is called by the visitors to those integer |
2108 | /// remainder instructions. |
2109 | /// Common integer remainder transforms |
2110 | Instruction *InstCombinerImpl::commonIRemTransforms(BinaryOperator &I) { |
2111 | if (Instruction *Phi = foldBinopWithPhiOperands(BO&: I)) |
2112 | return Phi; |
2113 | |
2114 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
2115 | |
2116 | // The RHS is known non-zero. |
2117 | if (Value *V = simplifyValueKnownNonZero(V: I.getOperand(i_nocapture: 1), IC&: *this, CxtI&: I)) |
2118 | return replaceOperand(I, OpNum: 1, V); |
2119 | |
2120 | // Handle cases involving: rem X, (select Cond, Y, Z) |
2121 | if (simplifyDivRemOfSelectWithZeroOp(I)) |
2122 | return &I; |
2123 | |
2124 | // If the divisor is a select-of-constants, try to constant fold all rem ops: |
2125 | // C % (select Cond, TrueC, FalseC) --> select Cond, (C % TrueC), (C % FalseC) |
2126 | // TODO: Adapt simplifyDivRemOfSelectWithZeroOp to allow this and other folds. |
2127 | if (match(V: Op0, P: m_ImmConstant()) && |
2128 | match(V: Op1, P: m_Select(C: m_Value(), L: m_ImmConstant(), R: m_ImmConstant()))) { |
2129 | if (Instruction *R = FoldOpIntoSelect(Op&: I, SI: cast<SelectInst>(Val: Op1), |
2130 | /*FoldWithMultiUse*/ true)) |
2131 | return R; |
2132 | } |
2133 | |
2134 | if (isa<Constant>(Val: Op1)) { |
2135 | if (Instruction *Op0I = dyn_cast<Instruction>(Val: Op0)) { |
2136 | if (SelectInst *SI = dyn_cast<SelectInst>(Val: Op0I)) { |
2137 | if (Instruction *R = FoldOpIntoSelect(Op&: I, SI)) |
2138 | return R; |
2139 | } else if (auto *PN = dyn_cast<PHINode>(Val: Op0I)) { |
2140 | const APInt *Op1Int; |
2141 | if (match(V: Op1, P: m_APInt(Res&: Op1Int)) && !Op1Int->isMinValue() && |
2142 | (I.getOpcode() == Instruction::URem || |
2143 | !Op1Int->isMinSignedValue())) { |
2144 | // foldOpIntoPhi will speculate instructions to the end of the PHI's |
2145 | // predecessor blocks, so do this only if we know the srem or urem |
2146 | // will not fault. |
2147 | if (Instruction *NV = foldOpIntoPhi(I, PN)) |
2148 | return NV; |
2149 | } |
2150 | } |
2151 | |
2152 | // See if we can fold away this rem instruction. |
2153 | if (SimplifyDemandedInstructionBits(Inst&: I)) |
2154 | return &I; |
2155 | } |
2156 | } |
2157 | |
2158 | if (Instruction *R = simplifyIRemMulShl(I, IC&: *this)) |
2159 | return R; |
2160 | |
2161 | return nullptr; |
2162 | } |
2163 | |
2164 | Instruction *InstCombinerImpl::visitURem(BinaryOperator &I) { |
2165 | if (Value *V = simplifyURemInst(LHS: I.getOperand(i_nocapture: 0), RHS: I.getOperand(i_nocapture: 1), |
2166 | Q: SQ.getWithInstruction(I: &I))) |
2167 | return replaceInstUsesWith(I, V); |
2168 | |
2169 | if (Instruction *X = foldVectorBinop(Inst&: I)) |
2170 | return X; |
2171 | |
2172 | if (Instruction *common = commonIRemTransforms(I)) |
2173 | return common; |
2174 | |
2175 | if (Instruction *NarrowRem = narrowUDivURem(I, IC&: *this)) |
2176 | return NarrowRem; |
2177 | |
2178 | // X urem Y -> X and Y-1, where Y is a power of 2, |
2179 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
2180 | Type *Ty = I.getType(); |
2181 | if (isKnownToBeAPowerOfTwo(V: Op1, /*OrZero*/ true, Depth: 0, CxtI: &I)) { |
2182 | // This may increase instruction count, we don't enforce that Y is a |
2183 | // constant. |
2184 | Constant *N1 = Constant::getAllOnesValue(Ty); |
2185 | Value *Add = Builder.CreateAdd(LHS: Op1, RHS: N1); |
2186 | return BinaryOperator::CreateAnd(V1: Op0, V2: Add); |
2187 | } |
2188 | |
2189 | // 1 urem X -> zext(X != 1) |
2190 | if (match(V: Op0, P: m_One())) { |
2191 | Value *Cmp = Builder.CreateICmpNE(LHS: Op1, RHS: ConstantInt::get(Ty, V: 1)); |
2192 | return CastInst::CreateZExtOrBitCast(S: Cmp, Ty); |
2193 | } |
2194 | |
2195 | // Op0 urem C -> Op0 < C ? Op0 : Op0 - C, where C >= signbit. |
2196 | // Op0 must be frozen because we are increasing its number of uses. |
2197 | if (match(V: Op1, P: m_Negative())) { |
2198 | Value *F0 = Op0; |
2199 | if (!isGuaranteedNotToBeUndef(V: Op0)) |
2200 | F0 = Builder.CreateFreeze(V: Op0, Name: Op0->getName() + ".fr" ); |
2201 | Value *Cmp = Builder.CreateICmpULT(LHS: F0, RHS: Op1); |
2202 | Value *Sub = Builder.CreateSub(LHS: F0, RHS: Op1); |
2203 | return SelectInst::Create(C: Cmp, S1: F0, S2: Sub); |
2204 | } |
2205 | |
2206 | // If the divisor is a sext of a boolean, then the divisor must be max |
2207 | // unsigned value (-1). Therefore, the remainder is Op0 unless Op0 is also |
2208 | // max unsigned value. In that case, the remainder is 0: |
2209 | // urem Op0, (sext i1 X) --> (Op0 == -1) ? 0 : Op0 |
2210 | Value *X; |
2211 | if (match(V: Op1, P: m_SExt(Op: m_Value(V&: X))) && X->getType()->isIntOrIntVectorTy(BitWidth: 1)) { |
2212 | Value *FrozenOp0 = Op0; |
2213 | if (!isGuaranteedNotToBeUndef(V: Op0)) |
2214 | FrozenOp0 = Builder.CreateFreeze(V: Op0, Name: Op0->getName() + ".frozen" ); |
2215 | Value *Cmp = |
2216 | Builder.CreateICmpEQ(LHS: FrozenOp0, RHS: ConstantInt::getAllOnesValue(Ty)); |
2217 | return SelectInst::Create(C: Cmp, S1: ConstantInt::getNullValue(Ty), S2: FrozenOp0); |
2218 | } |
2219 | |
2220 | // For "(X + 1) % Op1" and if (X u< Op1) => (X + 1) == Op1 ? 0 : X + 1 . |
2221 | if (match(V: Op0, P: m_Add(L: m_Value(V&: X), R: m_One()))) { |
2222 | Value *Val = |
2223 | simplifyICmpInst(Predicate: ICmpInst::ICMP_ULT, LHS: X, RHS: Op1, Q: SQ.getWithInstruction(I: &I)); |
2224 | if (Val && match(V: Val, P: m_One())) { |
2225 | Value *FrozenOp0 = Op0; |
2226 | if (!isGuaranteedNotToBeUndef(V: Op0)) |
2227 | FrozenOp0 = Builder.CreateFreeze(V: Op0, Name: Op0->getName() + ".frozen" ); |
2228 | Value *Cmp = Builder.CreateICmpEQ(LHS: FrozenOp0, RHS: Op1); |
2229 | return SelectInst::Create(C: Cmp, S1: ConstantInt::getNullValue(Ty), S2: FrozenOp0); |
2230 | } |
2231 | } |
2232 | |
2233 | return nullptr; |
2234 | } |
2235 | |
2236 | Instruction *InstCombinerImpl::visitSRem(BinaryOperator &I) { |
2237 | if (Value *V = simplifySRemInst(LHS: I.getOperand(i_nocapture: 0), RHS: I.getOperand(i_nocapture: 1), |
2238 | Q: SQ.getWithInstruction(I: &I))) |
2239 | return replaceInstUsesWith(I, V); |
2240 | |
2241 | if (Instruction *X = foldVectorBinop(Inst&: I)) |
2242 | return X; |
2243 | |
2244 | // Handle the integer rem common cases |
2245 | if (Instruction *Common = commonIRemTransforms(I)) |
2246 | return Common; |
2247 | |
2248 | Value *Op0 = I.getOperand(i_nocapture: 0), *Op1 = I.getOperand(i_nocapture: 1); |
2249 | { |
2250 | const APInt *Y; |
2251 | // X % -Y -> X % Y |
2252 | if (match(V: Op1, P: m_Negative(V&: Y)) && !Y->isMinSignedValue()) |
2253 | return replaceOperand(I, OpNum: 1, V: ConstantInt::get(Ty: I.getType(), V: -*Y)); |
2254 | } |
2255 | |
2256 | // -X srem Y --> -(X srem Y) |
2257 | Value *X, *Y; |
2258 | if (match(V: &I, P: m_SRem(L: m_OneUse(SubPattern: m_NSWNeg(V: m_Value(V&: X))), R: m_Value(V&: Y)))) |
2259 | return BinaryOperator::CreateNSWNeg(Op: Builder.CreateSRem(LHS: X, RHS: Y)); |
2260 | |
2261 | // If the sign bits of both operands are zero (i.e. we can prove they are |
2262 | // unsigned inputs), turn this into a urem. |
2263 | APInt Mask(APInt::getSignMask(BitWidth: I.getType()->getScalarSizeInBits())); |
2264 | if (MaskedValueIsZero(V: Op1, Mask, Depth: 0, CxtI: &I) && |
2265 | MaskedValueIsZero(V: Op0, Mask, Depth: 0, CxtI: &I)) { |
2266 | // X srem Y -> X urem Y, iff X and Y don't have sign bit set |
2267 | return BinaryOperator::CreateURem(V1: Op0, V2: Op1, Name: I.getName()); |
2268 | } |
2269 | |
2270 | // If it's a constant vector, flip any negative values positive. |
2271 | if (isa<ConstantVector>(Val: Op1) || isa<ConstantDataVector>(Val: Op1)) { |
2272 | Constant *C = cast<Constant>(Val: Op1); |
2273 | unsigned VWidth = cast<FixedVectorType>(Val: C->getType())->getNumElements(); |
2274 | |
2275 | bool hasNegative = false; |
2276 | bool hasMissing = false; |
2277 | for (unsigned i = 0; i != VWidth; ++i) { |
2278 | Constant *Elt = C->getAggregateElement(Elt: i); |
2279 | if (!Elt) { |
2280 | hasMissing = true; |
2281 | break; |
2282 | } |
2283 | |
2284 | if (ConstantInt *RHS = dyn_cast<ConstantInt>(Val: Elt)) |
2285 | if (RHS->isNegative()) |
2286 | hasNegative = true; |
2287 | } |
2288 | |
2289 | if (hasNegative && !hasMissing) { |
2290 | SmallVector<Constant *, 16> Elts(VWidth); |
2291 | for (unsigned i = 0; i != VWidth; ++i) { |
2292 | Elts[i] = C->getAggregateElement(Elt: i); // Handle undef, etc. |
2293 | if (ConstantInt *RHS = dyn_cast<ConstantInt>(Val: Elts[i])) { |
2294 | if (RHS->isNegative()) |
2295 | Elts[i] = cast<ConstantInt>(Val: ConstantExpr::getNeg(C: RHS)); |
2296 | } |
2297 | } |
2298 | |
2299 | Constant *NewRHSV = ConstantVector::get(V: Elts); |
2300 | if (NewRHSV != C) // Don't loop on -MININT |
2301 | return replaceOperand(I, OpNum: 1, V: NewRHSV); |
2302 | } |
2303 | } |
2304 | |
2305 | return nullptr; |
2306 | } |
2307 | |
2308 | Instruction *InstCombinerImpl::visitFRem(BinaryOperator &I) { |
2309 | if (Value *V = simplifyFRemInst(LHS: I.getOperand(i_nocapture: 0), RHS: I.getOperand(i_nocapture: 1), |
2310 | FMF: I.getFastMathFlags(), |
2311 | Q: SQ.getWithInstruction(I: &I))) |
2312 | return replaceInstUsesWith(I, V); |
2313 | |
2314 | if (Instruction *X = foldVectorBinop(Inst&: I)) |
2315 | return X; |
2316 | |
2317 | if (Instruction *Phi = foldBinopWithPhiOperands(BO&: I)) |
2318 | return Phi; |
2319 | |
2320 | return nullptr; |
2321 | } |
2322 | |