1 //===--------- SCEVAffinator.cpp - Create Scops from LLVM IR -------------===//
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 // Create a polyhedral description for a SCEV value.
10 //
11 //===----------------------------------------------------------------------===//
12
13 #include "polly/Support/SCEVAffinator.h"
14 #include "polly/Options.h"
15 #include "polly/ScopInfo.h"
16 #include "polly/Support/GICHelper.h"
17 #include "polly/Support/SCEVValidator.h"
18 #include "isl/aff.h"
19 #include "isl/local_space.h"
20 #include "isl/set.h"
21 #include "isl/val.h"
22
23 using namespace llvm;
24 using namespace polly;
25
26 static cl::opt<bool> IgnoreIntegerWrapping(
27 "polly-ignore-integer-wrapping",
28 cl::desc("Do not build run-time checks to proof absence of integer "
29 "wrapping"),
30 cl::Hidden, cl::ZeroOrMore, cl::init(false), cl::cat(PollyCategory));
31
32 // The maximal number of basic sets we allow during the construction of a
33 // piecewise affine function. More complex ones will result in very high
34 // compile time.
35 static int const MaxDisjunctionsInPwAff = 100;
36
37 // The maximal number of bits for which a general expression is modeled
38 // precisely.
39 static unsigned const MaxSmallBitWidth = 7;
40
41 /// Add the number of basic sets in @p Domain to @p User
addNumBasicSets(__isl_take isl_set * Domain,__isl_take isl_aff * Aff,void * User)42 static isl_stat addNumBasicSets(__isl_take isl_set *Domain,
43 __isl_take isl_aff *Aff, void *User) {
44 auto *NumBasicSets = static_cast<unsigned *>(User);
45 *NumBasicSets += isl_set_n_basic_set(Domain);
46 isl_set_free(Domain);
47 isl_aff_free(Aff);
48 return isl_stat_ok;
49 }
50
51 /// Determine if @p PWAC is too complex to continue.
isTooComplex(PWACtx PWAC)52 static bool isTooComplex(PWACtx PWAC) {
53 unsigned NumBasicSets = 0;
54 isl_pw_aff_foreach_piece(PWAC.first.get(), addNumBasicSets, &NumBasicSets);
55 if (NumBasicSets <= MaxDisjunctionsInPwAff)
56 return false;
57 return true;
58 }
59
60 /// Return the flag describing the possible wrapping of @p Expr.
getNoWrapFlags(const SCEV * Expr)61 static SCEV::NoWrapFlags getNoWrapFlags(const SCEV *Expr) {
62 if (auto *NAry = dyn_cast<SCEVNAryExpr>(Expr))
63 return NAry->getNoWrapFlags();
64 return SCEV::NoWrapMask;
65 }
66
combine(PWACtx PWAC0,PWACtx PWAC1,__isl_give isl_pw_aff * (Fn)(__isl_take isl_pw_aff *,__isl_take isl_pw_aff *))67 static PWACtx combine(PWACtx PWAC0, PWACtx PWAC1,
68 __isl_give isl_pw_aff *(Fn)(__isl_take isl_pw_aff *,
69 __isl_take isl_pw_aff *)) {
70 PWAC0.first = isl::manage(Fn(PWAC0.first.release(), PWAC1.first.release()));
71 PWAC0.second = PWAC0.second.unite(PWAC1.second);
72 return PWAC0;
73 }
74
getWidthExpValOnDomain(unsigned Width,__isl_take isl_set * Dom)75 static __isl_give isl_pw_aff *getWidthExpValOnDomain(unsigned Width,
76 __isl_take isl_set *Dom) {
77 auto *Ctx = isl_set_get_ctx(Dom);
78 auto *WidthVal = isl_val_int_from_ui(Ctx, Width);
79 auto *ExpVal = isl_val_2exp(WidthVal);
80 return isl_pw_aff_val_on_domain(Dom, ExpVal);
81 }
82
SCEVAffinator(Scop * S,LoopInfo & LI)83 SCEVAffinator::SCEVAffinator(Scop *S, LoopInfo &LI)
84 : S(S), Ctx(S->getIslCtx().get()), SE(*S->getSE()), LI(LI),
85 TD(S->getFunction().getParent()->getDataLayout()) {}
86
getScope()87 Loop *SCEVAffinator::getScope() { return BB ? LI.getLoopFor(BB) : nullptr; }
88
interpretAsUnsigned(PWACtx & PWAC,unsigned Width)89 void SCEVAffinator::interpretAsUnsigned(PWACtx &PWAC, unsigned Width) {
90 auto *NonNegDom = isl_pw_aff_nonneg_set(PWAC.first.copy());
91 auto *NonNegPWA =
92 isl_pw_aff_intersect_domain(PWAC.first.copy(), isl_set_copy(NonNegDom));
93 auto *ExpPWA = getWidthExpValOnDomain(Width, isl_set_complement(NonNegDom));
94 PWAC.first = isl::manage(isl_pw_aff_union_add(
95 NonNegPWA, isl_pw_aff_add(PWAC.first.release(), ExpPWA)));
96 }
97
takeNonNegativeAssumption(PWACtx & PWAC,RecordedAssumptionsTy * RecordedAssumptions)98 void SCEVAffinator::takeNonNegativeAssumption(
99 PWACtx &PWAC, RecordedAssumptionsTy *RecordedAssumptions) {
100 this->RecordedAssumptions = RecordedAssumptions;
101
102 auto *NegPWA = isl_pw_aff_neg(PWAC.first.copy());
103 auto *NegDom = isl_pw_aff_pos_set(NegPWA);
104 PWAC.second =
105 isl::manage(isl_set_union(PWAC.second.release(), isl_set_copy(NegDom)));
106 auto *Restriction = BB ? NegDom : isl_set_params(NegDom);
107 auto DL = BB ? BB->getTerminator()->getDebugLoc() : DebugLoc();
108 recordAssumption(RecordedAssumptions, UNSIGNED, isl::manage(Restriction), DL,
109 AS_RESTRICTION, BB);
110 }
111
getPWACtxFromPWA(isl::pw_aff PWA)112 PWACtx SCEVAffinator::getPWACtxFromPWA(isl::pw_aff PWA) {
113 return std::make_pair(PWA, isl::set::empty(isl::space(Ctx, 0, NumIterators)));
114 }
115
getPwAff(const SCEV * Expr,BasicBlock * BB,RecordedAssumptionsTy * RecordedAssumptions)116 PWACtx SCEVAffinator::getPwAff(const SCEV *Expr, BasicBlock *BB,
117 RecordedAssumptionsTy *RecordedAssumptions) {
118 this->BB = BB;
119 this->RecordedAssumptions = RecordedAssumptions;
120
121 if (BB) {
122 auto *DC = S->getDomainConditions(BB).release();
123 NumIterators = isl_set_n_dim(DC);
124 isl_set_free(DC);
125 } else
126 NumIterators = 0;
127
128 return visit(Expr);
129 }
130
checkForWrapping(const SCEV * Expr,PWACtx PWAC) const131 PWACtx SCEVAffinator::checkForWrapping(const SCEV *Expr, PWACtx PWAC) const {
132 // If the SCEV flags do contain NSW (no signed wrap) then PWA already
133 // represents Expr in modulo semantic (it is not allowed to overflow), thus we
134 // are done. Otherwise, we will compute:
135 // PWA = ((PWA + 2^(n-1)) mod (2 ^ n)) - 2^(n-1)
136 // whereas n is the number of bits of the Expr, hence:
137 // n = bitwidth(ExprType)
138
139 if (IgnoreIntegerWrapping || (getNoWrapFlags(Expr) & SCEV::FlagNSW))
140 return PWAC;
141
142 isl::pw_aff PWAMod = addModuloSemantic(PWAC.first, Expr->getType());
143
144 isl::set NotEqualSet = PWAC.first.ne_set(PWAMod);
145 PWAC.second = PWAC.second.unite(NotEqualSet).coalesce();
146
147 const DebugLoc &Loc = BB ? BB->getTerminator()->getDebugLoc() : DebugLoc();
148 if (!BB)
149 NotEqualSet = NotEqualSet.params();
150 NotEqualSet = NotEqualSet.coalesce();
151
152 if (!NotEqualSet.is_empty())
153 recordAssumption(RecordedAssumptions, WRAPPING, NotEqualSet, Loc,
154 AS_RESTRICTION, BB);
155
156 return PWAC;
157 }
158
addModuloSemantic(isl::pw_aff PWA,Type * ExprType) const159 isl::pw_aff SCEVAffinator::addModuloSemantic(isl::pw_aff PWA,
160 Type *ExprType) const {
161 unsigned Width = TD.getTypeSizeInBits(ExprType);
162
163 auto ModVal = isl::val::int_from_ui(Ctx, Width);
164 ModVal = ModVal.pow2();
165
166 isl::set Domain = PWA.domain();
167 isl::pw_aff AddPW =
168 isl::manage(getWidthExpValOnDomain(Width - 1, Domain.release()));
169
170 return PWA.add(AddPW).mod(ModVal).sub(AddPW);
171 }
172
hasNSWAddRecForLoop(Loop * L) const173 bool SCEVAffinator::hasNSWAddRecForLoop(Loop *L) const {
174 for (const auto &CachedPair : CachedExpressions) {
175 auto *AddRec = dyn_cast<SCEVAddRecExpr>(CachedPair.first.first);
176 if (!AddRec)
177 continue;
178 if (AddRec->getLoop() != L)
179 continue;
180 if (AddRec->getNoWrapFlags() & SCEV::FlagNSW)
181 return true;
182 }
183
184 return false;
185 }
186
computeModuloForExpr(const SCEV * Expr)187 bool SCEVAffinator::computeModuloForExpr(const SCEV *Expr) {
188 unsigned Width = TD.getTypeSizeInBits(Expr->getType());
189 // We assume nsw expressions never overflow.
190 if (auto *NAry = dyn_cast<SCEVNAryExpr>(Expr))
191 if (NAry->getNoWrapFlags() & SCEV::FlagNSW)
192 return false;
193 return Width <= MaxSmallBitWidth;
194 }
195
visit(const SCEV * Expr)196 PWACtx SCEVAffinator::visit(const SCEV *Expr) {
197
198 auto Key = std::make_pair(Expr, BB);
199 PWACtx PWAC = CachedExpressions[Key];
200 if (PWAC.first)
201 return PWAC;
202
203 auto ConstantAndLeftOverPair = extractConstantFactor(Expr, SE);
204 auto *Factor = ConstantAndLeftOverPair.first;
205 Expr = ConstantAndLeftOverPair.second;
206
207 auto *Scope = getScope();
208 S->addParams(getParamsInAffineExpr(&S->getRegion(), Scope, Expr, SE));
209
210 // In case the scev is a valid parameter, we do not further analyze this
211 // expression, but create a new parameter in the isl_pw_aff. This allows us
212 // to treat subexpressions that we cannot translate into an piecewise affine
213 // expression, as constant parameters of the piecewise affine expression.
214 if (isl_id *Id = S->getIdForParam(Expr).release()) {
215 isl_space *Space = isl_space_set_alloc(Ctx.get(), 1, NumIterators);
216 Space = isl_space_set_dim_id(Space, isl_dim_param, 0, Id);
217
218 isl_set *Domain = isl_set_universe(isl_space_copy(Space));
219 isl_aff *Affine = isl_aff_zero_on_domain(isl_local_space_from_space(Space));
220 Affine = isl_aff_add_coefficient_si(Affine, isl_dim_param, 0, 1);
221
222 PWAC = getPWACtxFromPWA(isl::manage(isl_pw_aff_alloc(Domain, Affine)));
223 } else {
224 PWAC = SCEVVisitor<SCEVAffinator, PWACtx>::visit(Expr);
225 if (computeModuloForExpr(Expr))
226 PWAC.first = addModuloSemantic(PWAC.first, Expr->getType());
227 else
228 PWAC = checkForWrapping(Expr, PWAC);
229 }
230
231 if (!Factor->getType()->isIntegerTy(1)) {
232 PWAC = combine(PWAC, visitConstant(Factor), isl_pw_aff_mul);
233 if (computeModuloForExpr(Key.first))
234 PWAC.first = addModuloSemantic(PWAC.first, Expr->getType());
235 }
236
237 // For compile time reasons we need to simplify the PWAC before we cache and
238 // return it.
239 PWAC.first = PWAC.first.coalesce();
240 if (!computeModuloForExpr(Key.first))
241 PWAC = checkForWrapping(Key.first, PWAC);
242
243 CachedExpressions[Key] = PWAC;
244 return PWAC;
245 }
246
visitConstant(const SCEVConstant * Expr)247 PWACtx SCEVAffinator::visitConstant(const SCEVConstant *Expr) {
248 ConstantInt *Value = Expr->getValue();
249 isl_val *v;
250
251 // LLVM does not define if an integer value is interpreted as a signed or
252 // unsigned value. Hence, without further information, it is unknown how
253 // this value needs to be converted to GMP. At the moment, we only support
254 // signed operations. So we just interpret it as signed. Later, there are
255 // two options:
256 //
257 // 1. We always interpret any value as signed and convert the values on
258 // demand.
259 // 2. We pass down the signedness of the calculation and use it to interpret
260 // this constant correctly.
261 v = isl_valFromAPInt(Ctx.get(), Value->getValue(), /* isSigned */ true);
262
263 isl_space *Space = isl_space_set_alloc(Ctx.get(), 0, NumIterators);
264 isl_local_space *ls = isl_local_space_from_space(Space);
265 return getPWACtxFromPWA(
266 isl::manage(isl_pw_aff_from_aff(isl_aff_val_on_domain(ls, v))));
267 }
268
visitPtrToIntExpr(const SCEVPtrToIntExpr * Expr)269 PWACtx SCEVAffinator::visitPtrToIntExpr(const SCEVPtrToIntExpr *Expr) {
270 return visit(Expr->getOperand(0));
271 }
272
visitTruncateExpr(const SCEVTruncateExpr * Expr)273 PWACtx SCEVAffinator::visitTruncateExpr(const SCEVTruncateExpr *Expr) {
274 // Truncate operations are basically modulo operations, thus we can
275 // model them that way. However, for large types we assume the operand
276 // to fit in the new type size instead of introducing a modulo with a very
277 // large constant.
278
279 auto *Op = Expr->getOperand();
280 auto OpPWAC = visit(Op);
281
282 unsigned Width = TD.getTypeSizeInBits(Expr->getType());
283
284 if (computeModuloForExpr(Expr))
285 return OpPWAC;
286
287 auto *Dom = OpPWAC.first.domain().release();
288 auto *ExpPWA = getWidthExpValOnDomain(Width - 1, Dom);
289 auto *GreaterDom =
290 isl_pw_aff_ge_set(OpPWAC.first.copy(), isl_pw_aff_copy(ExpPWA));
291 auto *SmallerDom =
292 isl_pw_aff_lt_set(OpPWAC.first.copy(), isl_pw_aff_neg(ExpPWA));
293 auto *OutOfBoundsDom = isl_set_union(SmallerDom, GreaterDom);
294 OpPWAC.second = OpPWAC.second.unite(isl::manage_copy(OutOfBoundsDom));
295
296 if (!BB) {
297 assert(isl_set_dim(OutOfBoundsDom, isl_dim_set) == 0 &&
298 "Expected a zero dimensional set for non-basic-block domains");
299 OutOfBoundsDom = isl_set_params(OutOfBoundsDom);
300 }
301
302 recordAssumption(RecordedAssumptions, UNSIGNED, isl::manage(OutOfBoundsDom),
303 DebugLoc(), AS_RESTRICTION, BB);
304
305 return OpPWAC;
306 }
307
visitZeroExtendExpr(const SCEVZeroExtendExpr * Expr)308 PWACtx SCEVAffinator::visitZeroExtendExpr(const SCEVZeroExtendExpr *Expr) {
309 // A zero-extended value can be interpreted as a piecewise defined signed
310 // value. If the value was non-negative it stays the same, otherwise it
311 // is the sum of the original value and 2^n where n is the bit-width of
312 // the original (or operand) type. Examples:
313 // zext i8 127 to i32 -> { [127] }
314 // zext i8 -1 to i32 -> { [256 + (-1)] } = { [255] }
315 // zext i8 %v to i32 -> [v] -> { [v] | v >= 0; [256 + v] | v < 0 }
316 //
317 // However, LLVM/Scalar Evolution uses zero-extend (potentially lead by a
318 // truncate) to represent some forms of modulo computation. The left-hand side
319 // of the condition in the code below would result in the SCEV
320 // "zext i1 <false, +, true>for.body" which is just another description
321 // of the C expression "i & 1 != 0" or, equivalently, "i % 2 != 0".
322 //
323 // for (i = 0; i < N; i++)
324 // if (i & 1 != 0 /* == i % 2 */)
325 // /* do something */
326 //
327 // If we do not make the modulo explicit but only use the mechanism described
328 // above we will get the very restrictive assumption "N < 3", because for all
329 // values of N >= 3 the SCEVAddRecExpr operand of the zero-extend would wrap.
330 // Alternatively, we can make the modulo in the operand explicit in the
331 // resulting piecewise function and thereby avoid the assumption on N. For the
332 // example this would result in the following piecewise affine function:
333 // { [i0] -> [(1)] : 2*floor((-1 + i0)/2) = -1 + i0;
334 // [i0] -> [(0)] : 2*floor((i0)/2) = i0 }
335 // To this end we can first determine if the (immediate) operand of the
336 // zero-extend can wrap and, in case it might, we will use explicit modulo
337 // semantic to compute the result instead of emitting non-wrapping
338 // assumptions.
339 //
340 // Note that operands with large bit-widths are less likely to be negative
341 // because it would result in a very large access offset or loop bound after
342 // the zero-extend. To this end one can optimistically assume the operand to
343 // be positive and avoid the piecewise definition if the bit-width is bigger
344 // than some threshold (here MaxZextSmallBitWidth).
345 //
346 // We choose to go with a hybrid solution of all modeling techniques described
347 // above. For small bit-widths (up to MaxZextSmallBitWidth) we will model the
348 // wrapping explicitly and use a piecewise defined function. However, if the
349 // bit-width is bigger than MaxZextSmallBitWidth we will employ overflow
350 // assumptions and assume the "former negative" piece will not exist.
351
352 auto *Op = Expr->getOperand();
353 auto OpPWAC = visit(Op);
354
355 // If the width is to big we assume the negative part does not occur.
356 if (!computeModuloForExpr(Op)) {
357 takeNonNegativeAssumption(OpPWAC, RecordedAssumptions);
358 return OpPWAC;
359 }
360
361 // If the width is small build the piece for the non-negative part and
362 // the one for the negative part and unify them.
363 unsigned Width = TD.getTypeSizeInBits(Op->getType());
364 interpretAsUnsigned(OpPWAC, Width);
365 return OpPWAC;
366 }
367
visitSignExtendExpr(const SCEVSignExtendExpr * Expr)368 PWACtx SCEVAffinator::visitSignExtendExpr(const SCEVSignExtendExpr *Expr) {
369 // As all values are represented as signed, a sign extension is a noop.
370 return visit(Expr->getOperand());
371 }
372
visitAddExpr(const SCEVAddExpr * Expr)373 PWACtx SCEVAffinator::visitAddExpr(const SCEVAddExpr *Expr) {
374 PWACtx Sum = visit(Expr->getOperand(0));
375
376 for (int i = 1, e = Expr->getNumOperands(); i < e; ++i) {
377 Sum = combine(Sum, visit(Expr->getOperand(i)), isl_pw_aff_add);
378 if (isTooComplex(Sum))
379 return complexityBailout();
380 }
381
382 return Sum;
383 }
384
visitMulExpr(const SCEVMulExpr * Expr)385 PWACtx SCEVAffinator::visitMulExpr(const SCEVMulExpr *Expr) {
386 PWACtx Prod = visit(Expr->getOperand(0));
387
388 for (int i = 1, e = Expr->getNumOperands(); i < e; ++i) {
389 Prod = combine(Prod, visit(Expr->getOperand(i)), isl_pw_aff_mul);
390 if (isTooComplex(Prod))
391 return complexityBailout();
392 }
393
394 return Prod;
395 }
396
visitAddRecExpr(const SCEVAddRecExpr * Expr)397 PWACtx SCEVAffinator::visitAddRecExpr(const SCEVAddRecExpr *Expr) {
398 assert(Expr->isAffine() && "Only affine AddRecurrences allowed");
399
400 auto Flags = Expr->getNoWrapFlags();
401
402 // Directly generate isl_pw_aff for Expr if 'start' is zero.
403 if (Expr->getStart()->isZero()) {
404 assert(S->contains(Expr->getLoop()) &&
405 "Scop does not contain the loop referenced in this AddRec");
406
407 PWACtx Step = visit(Expr->getOperand(1));
408 isl_space *Space = isl_space_set_alloc(Ctx.get(), 0, NumIterators);
409 isl_local_space *LocalSpace = isl_local_space_from_space(Space);
410
411 unsigned loopDimension = S->getRelativeLoopDepth(Expr->getLoop());
412
413 isl_aff *LAff = isl_aff_set_coefficient_si(
414 isl_aff_zero_on_domain(LocalSpace), isl_dim_in, loopDimension, 1);
415 isl_pw_aff *LPwAff = isl_pw_aff_from_aff(LAff);
416
417 Step.first = Step.first.mul(isl::manage(LPwAff));
418 return Step;
419 }
420
421 // Translate AddRecExpr from '{start, +, inc}' into 'start + {0, +, inc}'
422 // if 'start' is not zero.
423 // TODO: Using the original SCEV no-wrap flags is not always safe, however
424 // as our code generation is reordering the expression anyway it doesn't
425 // really matter.
426 const SCEV *ZeroStartExpr =
427 SE.getAddRecExpr(SE.getConstant(Expr->getStart()->getType(), 0),
428 Expr->getStepRecurrence(SE), Expr->getLoop(), Flags);
429
430 PWACtx Result = visit(ZeroStartExpr);
431 PWACtx Start = visit(Expr->getStart());
432 Result = combine(Result, Start, isl_pw_aff_add);
433 return Result;
434 }
435
visitSMaxExpr(const SCEVSMaxExpr * Expr)436 PWACtx SCEVAffinator::visitSMaxExpr(const SCEVSMaxExpr *Expr) {
437 PWACtx Max = visit(Expr->getOperand(0));
438
439 for (int i = 1, e = Expr->getNumOperands(); i < e; ++i) {
440 Max = combine(Max, visit(Expr->getOperand(i)), isl_pw_aff_max);
441 if (isTooComplex(Max))
442 return complexityBailout();
443 }
444
445 return Max;
446 }
447
visitSMinExpr(const SCEVSMinExpr * Expr)448 PWACtx SCEVAffinator::visitSMinExpr(const SCEVSMinExpr *Expr) {
449 PWACtx Min = visit(Expr->getOperand(0));
450
451 for (int i = 1, e = Expr->getNumOperands(); i < e; ++i) {
452 Min = combine(Min, visit(Expr->getOperand(i)), isl_pw_aff_min);
453 if (isTooComplex(Min))
454 return complexityBailout();
455 }
456
457 return Min;
458 }
459
visitUMaxExpr(const SCEVUMaxExpr * Expr)460 PWACtx SCEVAffinator::visitUMaxExpr(const SCEVUMaxExpr *Expr) {
461 llvm_unreachable("SCEVUMaxExpr not yet supported");
462 }
463
visitUMinExpr(const SCEVUMinExpr * Expr)464 PWACtx SCEVAffinator::visitUMinExpr(const SCEVUMinExpr *Expr) {
465 llvm_unreachable("SCEVUMinExpr not yet supported");
466 }
467
visitUDivExpr(const SCEVUDivExpr * Expr)468 PWACtx SCEVAffinator::visitUDivExpr(const SCEVUDivExpr *Expr) {
469 // The handling of unsigned division is basically the same as for signed
470 // division, except the interpretation of the operands. As the divisor
471 // has to be constant in both cases we can simply interpret it as an
472 // unsigned value without additional complexity in the representation.
473 // For the dividend we could choose from the different representation
474 // schemes introduced for zero-extend operations but for now we will
475 // simply use an assumption.
476 auto *Dividend = Expr->getLHS();
477 auto *Divisor = Expr->getRHS();
478 assert(isa<SCEVConstant>(Divisor) &&
479 "UDiv is no parameter but has a non-constant RHS.");
480
481 auto DividendPWAC = visit(Dividend);
482 auto DivisorPWAC = visit(Divisor);
483
484 if (SE.isKnownNegative(Divisor)) {
485 // Interpret negative divisors unsigned. This is a special case of the
486 // piece-wise defined value described for zero-extends as we already know
487 // the actual value of the constant divisor.
488 unsigned Width = TD.getTypeSizeInBits(Expr->getType());
489 auto *DivisorDom = DivisorPWAC.first.domain().release();
490 auto *WidthExpPWA = getWidthExpValOnDomain(Width, DivisorDom);
491 DivisorPWAC.first = DivisorPWAC.first.add(isl::manage(WidthExpPWA));
492 }
493
494 // TODO: One can represent the dividend as piece-wise function to be more
495 // precise but therefor a heuristic is needed.
496
497 // Assume a non-negative dividend.
498 takeNonNegativeAssumption(DividendPWAC, RecordedAssumptions);
499
500 DividendPWAC = combine(DividendPWAC, DivisorPWAC, isl_pw_aff_div);
501 DividendPWAC.first = DividendPWAC.first.floor();
502
503 return DividendPWAC;
504 }
505
visitSDivInstruction(Instruction * SDiv)506 PWACtx SCEVAffinator::visitSDivInstruction(Instruction *SDiv) {
507 assert(SDiv->getOpcode() == Instruction::SDiv && "Assumed SDiv instruction!");
508
509 auto *Scope = getScope();
510 auto *Divisor = SDiv->getOperand(1);
511 auto *DivisorSCEV = SE.getSCEVAtScope(Divisor, Scope);
512 auto DivisorPWAC = visit(DivisorSCEV);
513 assert(isa<SCEVConstant>(DivisorSCEV) &&
514 "SDiv is no parameter but has a non-constant RHS.");
515
516 auto *Dividend = SDiv->getOperand(0);
517 auto *DividendSCEV = SE.getSCEVAtScope(Dividend, Scope);
518 auto DividendPWAC = visit(DividendSCEV);
519 DividendPWAC = combine(DividendPWAC, DivisorPWAC, isl_pw_aff_tdiv_q);
520 return DividendPWAC;
521 }
522
visitSRemInstruction(Instruction * SRem)523 PWACtx SCEVAffinator::visitSRemInstruction(Instruction *SRem) {
524 assert(SRem->getOpcode() == Instruction::SRem && "Assumed SRem instruction!");
525
526 auto *Scope = getScope();
527 auto *Divisor = SRem->getOperand(1);
528 auto *DivisorSCEV = SE.getSCEVAtScope(Divisor, Scope);
529 auto DivisorPWAC = visit(DivisorSCEV);
530 assert(isa<ConstantInt>(Divisor) &&
531 "SRem is no parameter but has a non-constant RHS.");
532
533 auto *Dividend = SRem->getOperand(0);
534 auto *DividendSCEV = SE.getSCEVAtScope(Dividend, Scope);
535 auto DividendPWAC = visit(DividendSCEV);
536 DividendPWAC = combine(DividendPWAC, DivisorPWAC, isl_pw_aff_tdiv_r);
537 return DividendPWAC;
538 }
539
visitUnknown(const SCEVUnknown * Expr)540 PWACtx SCEVAffinator::visitUnknown(const SCEVUnknown *Expr) {
541 if (Instruction *I = dyn_cast<Instruction>(Expr->getValue())) {
542 switch (I->getOpcode()) {
543 case Instruction::IntToPtr:
544 return visit(SE.getSCEVAtScope(I->getOperand(0), getScope()));
545 case Instruction::SDiv:
546 return visitSDivInstruction(I);
547 case Instruction::SRem:
548 return visitSRemInstruction(I);
549 default:
550 break; // Fall through.
551 }
552 }
553
554 llvm_unreachable(
555 "Unknowns SCEV was neither parameter nor a valid instruction.");
556 }
557
complexityBailout()558 PWACtx SCEVAffinator::complexityBailout() {
559 // We hit the complexity limit for affine expressions; invalidate the scop
560 // and return a constant zero.
561 const DebugLoc &Loc = BB ? BB->getTerminator()->getDebugLoc() : DebugLoc();
562 S->invalidate(COMPLEXITY, Loc);
563 return visit(SE.getZero(Type::getInt32Ty(S->getFunction().getContext())));
564 }
565