Bug Summary

File:build/source/llvm/lib/Transforms/Scalar/Reassociate.cpp
Warning:line 179, column 8
Dereference of null pointer (loaded from variable '__begin1')

Annotated Source Code

Press '?' to see keyboard shortcuts

clang -cc1 -cc1 -triple x86_64-pc-linux-gnu -analyze -disable-free -clear-ast-before-backend -disable-llvm-verifier -discard-value-names -main-file-name Reassociate.cpp -analyzer-checker=core -analyzer-checker=apiModeling -analyzer-checker=unix -analyzer-checker=deadcode -analyzer-checker=cplusplus -analyzer-checker=security.insecureAPI.UncheckedReturn -analyzer-checker=security.insecureAPI.getpw -analyzer-checker=security.insecureAPI.gets -analyzer-checker=security.insecureAPI.mktemp -analyzer-checker=security.insecureAPI.mkstemp -analyzer-checker=security.insecureAPI.vfork -analyzer-checker=nullability.NullPassedToNonnull -analyzer-checker=nullability.NullReturnedFromNonnull -analyzer-output plist -w -setup-static-analyzer -analyzer-config-compatibility-mode=true -mrelocation-model pic -pic-level 2 -mframe-pointer=none -fmath-errno -ffp-contract=on -fno-rounding-math -mconstructor-aliases -funwind-tables=2 -target-cpu x86-64 -tune-cpu generic -debugger-tuning=gdb -ffunction-sections -fdata-sections -fcoverage-compilation-dir=/build/source/build-llvm/tools/clang/stage2-bins -resource-dir /usr/lib/llvm-17/lib/clang/17 -D _DEBUG -D _GLIBCXX_ASSERTIONS -D _GNU_SOURCE -D _LIBCPP_ENABLE_ASSERTIONS -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I lib/Transforms/Scalar -I /build/source/llvm/lib/Transforms/Scalar -I include -I /build/source/llvm/include -D _FORTIFY_SOURCE=2 -D NDEBUG -U NDEBUG -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/c++/10 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/x86_64-linux-gnu/c++/10 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/c++/10/backward -internal-isystem /usr/lib/llvm-17/lib/clang/17/include -internal-isystem /usr/local/include -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../x86_64-linux-gnu/include -internal-externc-isystem /usr/include/x86_64-linux-gnu -internal-externc-isystem /include -internal-externc-isystem /usr/include -fmacro-prefix-map=/build/source/build-llvm/tools/clang/stage2-bins=build-llvm/tools/clang/stage2-bins -fmacro-prefix-map=/build/source/= -fcoverage-prefix-map=/build/source/build-llvm/tools/clang/stage2-bins=build-llvm/tools/clang/stage2-bins -fcoverage-prefix-map=/build/source/= -source-date-epoch 1683717183 -O2 -Wno-unused-command-line-argument -Wno-unused-parameter -Wwrite-strings -Wno-missing-field-initializers -Wno-long-long -Wno-maybe-uninitialized -Wno-class-memaccess -Wno-redundant-move -Wno-pessimizing-move -Wno-noexcept-type -Wno-comment -Wno-misleading-indentation -std=c++17 -fdeprecated-macro -fdebug-compilation-dir=/build/source/build-llvm/tools/clang/stage2-bins -fdebug-prefix-map=/build/source/build-llvm/tools/clang/stage2-bins=build-llvm/tools/clang/stage2-bins -fdebug-prefix-map=/build/source/= -ferror-limit 19 -fvisibility-inlines-hidden -stack-protector 2 -fgnuc-version=4.2.1 -fcolor-diagnostics -vectorize-loops -vectorize-slp -analyzer-output=html -analyzer-config stable-report-filename=true -faddrsig -D__GCC_HAVE_DWARF2_CFI_ASM=1 -o /tmp/scan-build-2023-05-10-133810-16478-1 -x c++ /build/source/llvm/lib/Transforms/Scalar/Reassociate.cpp
1//===- Reassociate.cpp - Reassociate binary expressions -------------------===//
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 pass reassociates commutative expressions in an order that is designed
10// to promote better constant propagation, GCSE, LICM, PRE, etc.
11//
12// For example: 4 + (x + 5) -> x + (4 + 5)
13//
14// In the implementation of this algorithm, constants are assigned rank = 0,
15// function arguments are rank = 1, and other values are assigned ranks
16// corresponding to the reverse post order traversal of current function
17// (starting at 2), which effectively gives values in deep loops higher rank
18// than values not in loops.
19//
20//===----------------------------------------------------------------------===//
21
22#include "llvm/Transforms/Scalar/Reassociate.h"
23#include "llvm/ADT/APFloat.h"
24#include "llvm/ADT/APInt.h"
25#include "llvm/ADT/DenseMap.h"
26#include "llvm/ADT/PostOrderIterator.h"
27#include "llvm/ADT/SmallPtrSet.h"
28#include "llvm/ADT/SmallSet.h"
29#include "llvm/ADT/SmallVector.h"
30#include "llvm/ADT/Statistic.h"
31#include "llvm/Analysis/BasicAliasAnalysis.h"
32#include "llvm/Analysis/ConstantFolding.h"
33#include "llvm/Analysis/GlobalsModRef.h"
34#include "llvm/Analysis/ValueTracking.h"
35#include "llvm/IR/Argument.h"
36#include "llvm/IR/BasicBlock.h"
37#include "llvm/IR/CFG.h"
38#include "llvm/IR/Constant.h"
39#include "llvm/IR/Constants.h"
40#include "llvm/IR/Function.h"
41#include "llvm/IR/IRBuilder.h"
42#include "llvm/IR/InstrTypes.h"
43#include "llvm/IR/Instruction.h"
44#include "llvm/IR/Instructions.h"
45#include "llvm/IR/Operator.h"
46#include "llvm/IR/PassManager.h"
47#include "llvm/IR/PatternMatch.h"
48#include "llvm/IR/Type.h"
49#include "llvm/IR/User.h"
50#include "llvm/IR/Value.h"
51#include "llvm/IR/ValueHandle.h"
52#include "llvm/InitializePasses.h"
53#include "llvm/Pass.h"
54#include "llvm/Support/Casting.h"
55#include "llvm/Support/Debug.h"
56#include "llvm/Support/raw_ostream.h"
57#include "llvm/Transforms/Scalar.h"
58#include "llvm/Transforms/Utils/Local.h"
59#include <algorithm>
60#include <cassert>
61#include <utility>
62
63using namespace llvm;
64using namespace reassociate;
65using namespace PatternMatch;
66
67#define DEBUG_TYPE"reassociate" "reassociate"
68
69STATISTIC(NumChanged, "Number of insts reassociated")static llvm::Statistic NumChanged = {"reassociate", "NumChanged"
, "Number of insts reassociated"}
;
70STATISTIC(NumAnnihil, "Number of expr tree annihilated")static llvm::Statistic NumAnnihil = {"reassociate", "NumAnnihil"
, "Number of expr tree annihilated"}
;
71STATISTIC(NumFactor , "Number of multiplies factored")static llvm::Statistic NumFactor = {"reassociate", "NumFactor"
, "Number of multiplies factored"}
;
72
73#ifndef NDEBUG
74/// Print out the expression identified in the Ops list.
75static void PrintOps(Instruction *I, const SmallVectorImpl<ValueEntry> &Ops) {
76 Module *M = I->getModule();
77 dbgs() << Instruction::getOpcodeName(I->getOpcode()) << " "
78 << *Ops[0].Op->getType() << '\t';
79 for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
80 dbgs() << "[ ";
81 Ops[i].Op->printAsOperand(dbgs(), false, M);
82 dbgs() << ", #" << Ops[i].Rank << "] ";
83 }
84}
85#endif
86
87/// Utility class representing a non-constant Xor-operand. We classify
88/// non-constant Xor-Operands into two categories:
89/// C1) The operand is in the form "X & C", where C is a constant and C != ~0
90/// C2)
91/// C2.1) The operand is in the form of "X | C", where C is a non-zero
92/// constant.
93/// C2.2) Any operand E which doesn't fall into C1 and C2.1, we view this
94/// operand as "E | 0"
95class llvm::reassociate::XorOpnd {
96public:
97 XorOpnd(Value *V);
98
99 bool isInvalid() const { return SymbolicPart == nullptr; }
100 bool isOrExpr() const { return isOr; }
101 Value *getValue() const { return OrigVal; }
102 Value *getSymbolicPart() const { return SymbolicPart; }
103 unsigned getSymbolicRank() const { return SymbolicRank; }
104 const APInt &getConstPart() const { return ConstPart; }
105
106 void Invalidate() { SymbolicPart = OrigVal = nullptr; }
107 void setSymbolicRank(unsigned R) { SymbolicRank = R; }
108
109private:
110 Value *OrigVal;
111 Value *SymbolicPart;
112 APInt ConstPart;
113 unsigned SymbolicRank;
114 bool isOr;
115};
116
117XorOpnd::XorOpnd(Value *V) {
118 assert(!isa<ConstantInt>(V) && "No ConstantInt")(static_cast <bool> (!isa<ConstantInt>(V) &&
"No ConstantInt") ? void (0) : __assert_fail ("!isa<ConstantInt>(V) && \"No ConstantInt\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 118, __extension__
__PRETTY_FUNCTION__))
;
119 OrigVal = V;
120 Instruction *I = dyn_cast<Instruction>(V);
121 SymbolicRank = 0;
122
123 if (I && (I->getOpcode() == Instruction::Or ||
124 I->getOpcode() == Instruction::And)) {
125 Value *V0 = I->getOperand(0);
126 Value *V1 = I->getOperand(1);
127 const APInt *C;
128 if (match(V0, m_APInt(C)))
129 std::swap(V0, V1);
130
131 if (match(V1, m_APInt(C))) {
132 ConstPart = *C;
133 SymbolicPart = V0;
134 isOr = (I->getOpcode() == Instruction::Or);
135 return;
136 }
137 }
138
139 // view the operand as "V | 0"
140 SymbolicPart = V;
141 ConstPart = APInt::getZero(V->getType()->getScalarSizeInBits());
142 isOr = true;
143}
144
145/// Return true if I is an instruction with the FastMathFlags that are needed
146/// for general reassociation set. This is not the same as testing
147/// Instruction::isAssociative() because it includes operations like fsub.
148/// (This routine is only intended to be called for floating-point operations.)
149static bool hasFPAssociativeFlags(Instruction *I) {
150 assert(I && isa<FPMathOperator>(I) && "Should only check FP ops")(static_cast <bool> (I && isa<FPMathOperator
>(I) && "Should only check FP ops") ? void (0) : __assert_fail
("I && isa<FPMathOperator>(I) && \"Should only check FP ops\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 150, __extension__
__PRETTY_FUNCTION__))
;
151 return I->hasAllowReassoc() && I->hasNoSignedZeros();
152}
153
154/// Return true if V is an instruction of the specified opcode and if it
155/// only has one use.
156static BinaryOperator *isReassociableOp(Value *V, unsigned Opcode) {
157 auto *BO = dyn_cast<BinaryOperator>(V);
158 if (BO && BO->hasOneUse() && BO->getOpcode() == Opcode)
159 if (!isa<FPMathOperator>(BO) || hasFPAssociativeFlags(BO))
160 return BO;
161 return nullptr;
162}
163
164static BinaryOperator *isReassociableOp(Value *V, unsigned Opcode1,
165 unsigned Opcode2) {
166 auto *BO = dyn_cast<BinaryOperator>(V);
167 if (BO && BO->hasOneUse() &&
168 (BO->getOpcode() == Opcode1 || BO->getOpcode() == Opcode2))
169 if (!isa<FPMathOperator>(BO) || hasFPAssociativeFlags(BO))
170 return BO;
171 return nullptr;
172}
173
174void ReassociatePass::BuildRankMap(Function &F,
175 ReversePostOrderTraversal<Function*> &RPOT) {
176 unsigned Rank = 2;
177
178 // Assign distinct ranks to function arguments.
179 for (auto &Arg :
8.1
'__begin1' is not equal to '__end1'
F.args()) {
5
Assuming '__begin1' is not equal to '__end1'
8
Null pointer value stored to '__begin1'
9
Dereference of null pointer (loaded from variable '__begin1')
180 ValueRankMap[&Arg] = ++Rank;
181 LLVM_DEBUG(dbgs() << "Calculated Rank[" << Arg.getName() << "] = " << Rankdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "Calculated Rank[" <<
Arg.getName() << "] = " << Rank << "\n"; }
} while (false)
6
Assuming 'DebugFlag' is false
7
Loop condition is false. Exiting loop
182 << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "Calculated Rank[" <<
Arg.getName() << "] = " << Rank << "\n"; }
} while (false)
;
183 }
184
185 // Traverse basic blocks in ReversePostOrder.
186 for (BasicBlock *BB : RPOT) {
187 unsigned BBRank = RankMap[BB] = ++Rank << 16;
188
189 // Walk the basic block, adding precomputed ranks for any instructions that
190 // we cannot move. This ensures that the ranks for these instructions are
191 // all different in the block.
192 for (Instruction &I : *BB)
193 if (mayHaveNonDefUseDependency(I))
194 ValueRankMap[&I] = ++BBRank;
195 }
196}
197
198unsigned ReassociatePass::getRank(Value *V) {
199 Instruction *I = dyn_cast<Instruction>(V);
200 if (!I) {
201 if (isa<Argument>(V)) return ValueRankMap[V]; // Function argument.
202 return 0; // Otherwise it's a global or constant, rank 0.
203 }
204
205 if (unsigned Rank = ValueRankMap[I])
206 return Rank; // Rank already known?
207
208 // If this is an expression, return the 1+MAX(rank(LHS), rank(RHS)) so that
209 // we can reassociate expressions for code motion! Since we do not recurse
210 // for PHI nodes, we cannot have infinite recursion here, because there
211 // cannot be loops in the value graph that do not go through PHI nodes.
212 unsigned Rank = 0, MaxRank = RankMap[I->getParent()];
213 for (unsigned i = 0, e = I->getNumOperands(); i != e && Rank != MaxRank; ++i)
214 Rank = std::max(Rank, getRank(I->getOperand(i)));
215
216 // If this is a 'not' or 'neg' instruction, do not count it for rank. This
217 // assures us that X and ~X will have the same rank.
218 if (!match(I, m_Not(m_Value())) && !match(I, m_Neg(m_Value())) &&
219 !match(I, m_FNeg(m_Value())))
220 ++Rank;
221
222 LLVM_DEBUG(dbgs() << "Calculated Rank[" << V->getName() << "] = " << Rankdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "Calculated Rank[" <<
V->getName() << "] = " << Rank << "\n";
} } while (false)
223 << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "Calculated Rank[" <<
V->getName() << "] = " << Rank << "\n";
} } while (false)
;
224
225 return ValueRankMap[I] = Rank;
226}
227
228// Canonicalize constants to RHS. Otherwise, sort the operands by rank.
229void ReassociatePass::canonicalizeOperands(Instruction *I) {
230 assert(isa<BinaryOperator>(I) && "Expected binary operator.")(static_cast <bool> (isa<BinaryOperator>(I) &&
"Expected binary operator.") ? void (0) : __assert_fail ("isa<BinaryOperator>(I) && \"Expected binary operator.\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 230, __extension__
__PRETTY_FUNCTION__))
;
231 assert(I->isCommutative() && "Expected commutative operator.")(static_cast <bool> (I->isCommutative() && "Expected commutative operator."
) ? void (0) : __assert_fail ("I->isCommutative() && \"Expected commutative operator.\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 231, __extension__
__PRETTY_FUNCTION__))
;
232
233 Value *LHS = I->getOperand(0);
234 Value *RHS = I->getOperand(1);
235 if (LHS == RHS || isa<Constant>(RHS))
236 return;
237 if (isa<Constant>(LHS) || getRank(RHS) < getRank(LHS))
238 cast<BinaryOperator>(I)->swapOperands();
239}
240
241static BinaryOperator *CreateAdd(Value *S1, Value *S2, const Twine &Name,
242 Instruction *InsertBefore, Value *FlagsOp) {
243 if (S1->getType()->isIntOrIntVectorTy())
244 return BinaryOperator::CreateAdd(S1, S2, Name, InsertBefore);
245 else {
246 BinaryOperator *Res =
247 BinaryOperator::CreateFAdd(S1, S2, Name, InsertBefore);
248 Res->setFastMathFlags(cast<FPMathOperator>(FlagsOp)->getFastMathFlags());
249 return Res;
250 }
251}
252
253static BinaryOperator *CreateMul(Value *S1, Value *S2, const Twine &Name,
254 Instruction *InsertBefore, Value *FlagsOp) {
255 if (S1->getType()->isIntOrIntVectorTy())
256 return BinaryOperator::CreateMul(S1, S2, Name, InsertBefore);
257 else {
258 BinaryOperator *Res =
259 BinaryOperator::CreateFMul(S1, S2, Name, InsertBefore);
260 Res->setFastMathFlags(cast<FPMathOperator>(FlagsOp)->getFastMathFlags());
261 return Res;
262 }
263}
264
265static Instruction *CreateNeg(Value *S1, const Twine &Name,
266 Instruction *InsertBefore, Value *FlagsOp) {
267 if (S1->getType()->isIntOrIntVectorTy())
268 return BinaryOperator::CreateNeg(S1, Name, InsertBefore);
269
270 if (auto *FMFSource = dyn_cast<Instruction>(FlagsOp))
271 return UnaryOperator::CreateFNegFMF(S1, FMFSource, Name, InsertBefore);
272
273 return UnaryOperator::CreateFNeg(S1, Name, InsertBefore);
274}
275
276/// Replace 0-X with X*-1.
277static BinaryOperator *LowerNegateToMultiply(Instruction *Neg) {
278 assert((isa<UnaryOperator>(Neg) || isa<BinaryOperator>(Neg)) &&(static_cast <bool> ((isa<UnaryOperator>(Neg) || isa
<BinaryOperator>(Neg)) && "Expected a Negate!")
? void (0) : __assert_fail ("(isa<UnaryOperator>(Neg) || isa<BinaryOperator>(Neg)) && \"Expected a Negate!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 279, __extension__
__PRETTY_FUNCTION__))
279 "Expected a Negate!")(static_cast <bool> ((isa<UnaryOperator>(Neg) || isa
<BinaryOperator>(Neg)) && "Expected a Negate!")
? void (0) : __assert_fail ("(isa<UnaryOperator>(Neg) || isa<BinaryOperator>(Neg)) && \"Expected a Negate!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 279, __extension__
__PRETTY_FUNCTION__))
;
280 // FIXME: It's not safe to lower a unary FNeg into a FMul by -1.0.
281 unsigned OpNo = isa<BinaryOperator>(Neg) ? 1 : 0;
282 Type *Ty = Neg->getType();
283 Constant *NegOne = Ty->isIntOrIntVectorTy() ?
284 ConstantInt::getAllOnesValue(Ty) : ConstantFP::get(Ty, -1.0);
285
286 BinaryOperator *Res = CreateMul(Neg->getOperand(OpNo), NegOne, "", Neg, Neg);
287 Neg->setOperand(OpNo, Constant::getNullValue(Ty)); // Drop use of op.
288 Res->takeName(Neg);
289 Neg->replaceAllUsesWith(Res);
290 Res->setDebugLoc(Neg->getDebugLoc());
291 return Res;
292}
293
294/// Returns k such that lambda(2^Bitwidth) = 2^k, where lambda is the Carmichael
295/// function. This means that x^(2^k) === 1 mod 2^Bitwidth for
296/// every odd x, i.e. x^(2^k) = 1 for every odd x in Bitwidth-bit arithmetic.
297/// Note that 0 <= k < Bitwidth, and if Bitwidth > 3 then x^(2^k) = 0 for every
298/// even x in Bitwidth-bit arithmetic.
299static unsigned CarmichaelShift(unsigned Bitwidth) {
300 if (Bitwidth < 3)
301 return Bitwidth - 1;
302 return Bitwidth - 2;
303}
304
305/// Add the extra weight 'RHS' to the existing weight 'LHS',
306/// reducing the combined weight using any special properties of the operation.
307/// The existing weight LHS represents the computation X op X op ... op X where
308/// X occurs LHS times. The combined weight represents X op X op ... op X with
309/// X occurring LHS + RHS times. If op is "Xor" for example then the combined
310/// operation is equivalent to X if LHS + RHS is odd, or 0 if LHS + RHS is even;
311/// the routine returns 1 in LHS in the first case, and 0 in LHS in the second.
312static void IncorporateWeight(APInt &LHS, const APInt &RHS, unsigned Opcode) {
313 // If we were working with infinite precision arithmetic then the combined
314 // weight would be LHS + RHS. But we are using finite precision arithmetic,
315 // and the APInt sum LHS + RHS may not be correct if it wraps (it is correct
316 // for nilpotent operations and addition, but not for idempotent operations
317 // and multiplication), so it is important to correctly reduce the combined
318 // weight back into range if wrapping would be wrong.
319
320 // If RHS is zero then the weight didn't change.
321 if (RHS.isMinValue())
322 return;
323 // If LHS is zero then the combined weight is RHS.
324 if (LHS.isMinValue()) {
325 LHS = RHS;
326 return;
327 }
328 // From this point on we know that neither LHS nor RHS is zero.
329
330 if (Instruction::isIdempotent(Opcode)) {
331 // Idempotent means X op X === X, so any non-zero weight is equivalent to a
332 // weight of 1. Keeping weights at zero or one also means that wrapping is
333 // not a problem.
334 assert(LHS == 1 && RHS == 1 && "Weights not reduced!")(static_cast <bool> (LHS == 1 && RHS == 1 &&
"Weights not reduced!") ? void (0) : __assert_fail ("LHS == 1 && RHS == 1 && \"Weights not reduced!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 334, __extension__
__PRETTY_FUNCTION__))
;
335 return; // Return a weight of 1.
336 }
337 if (Instruction::isNilpotent(Opcode)) {
338 // Nilpotent means X op X === 0, so reduce weights modulo 2.
339 assert(LHS == 1 && RHS == 1 && "Weights not reduced!")(static_cast <bool> (LHS == 1 && RHS == 1 &&
"Weights not reduced!") ? void (0) : __assert_fail ("LHS == 1 && RHS == 1 && \"Weights not reduced!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 339, __extension__
__PRETTY_FUNCTION__))
;
340 LHS = 0; // 1 + 1 === 0 modulo 2.
341 return;
342 }
343 if (Opcode == Instruction::Add || Opcode == Instruction::FAdd) {
344 // TODO: Reduce the weight by exploiting nsw/nuw?
345 LHS += RHS;
346 return;
347 }
348
349 assert((Opcode == Instruction::Mul || Opcode == Instruction::FMul) &&(static_cast <bool> ((Opcode == Instruction::Mul || Opcode
== Instruction::FMul) && "Unknown associative operation!"
) ? void (0) : __assert_fail ("(Opcode == Instruction::Mul || Opcode == Instruction::FMul) && \"Unknown associative operation!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 350, __extension__
__PRETTY_FUNCTION__))
350 "Unknown associative operation!")(static_cast <bool> ((Opcode == Instruction::Mul || Opcode
== Instruction::FMul) && "Unknown associative operation!"
) ? void (0) : __assert_fail ("(Opcode == Instruction::Mul || Opcode == Instruction::FMul) && \"Unknown associative operation!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 350, __extension__
__PRETTY_FUNCTION__))
;
351 unsigned Bitwidth = LHS.getBitWidth();
352 // If CM is the Carmichael number then a weight W satisfying W >= CM+Bitwidth
353 // can be replaced with W-CM. That's because x^W=x^(W-CM) for every Bitwidth
354 // bit number x, since either x is odd in which case x^CM = 1, or x is even in
355 // which case both x^W and x^(W - CM) are zero. By subtracting off multiples
356 // of CM like this weights can always be reduced to the range [0, CM+Bitwidth)
357 // which by a happy accident means that they can always be represented using
358 // Bitwidth bits.
359 // TODO: Reduce the weight by exploiting nsw/nuw? (Could do much better than
360 // the Carmichael number).
361 if (Bitwidth > 3) {
362 /// CM - The value of Carmichael's lambda function.
363 APInt CM = APInt::getOneBitSet(Bitwidth, CarmichaelShift(Bitwidth));
364 // Any weight W >= Threshold can be replaced with W - CM.
365 APInt Threshold = CM + Bitwidth;
366 assert(LHS.ult(Threshold) && RHS.ult(Threshold) && "Weights not reduced!")(static_cast <bool> (LHS.ult(Threshold) && RHS.
ult(Threshold) && "Weights not reduced!") ? void (0) :
__assert_fail ("LHS.ult(Threshold) && RHS.ult(Threshold) && \"Weights not reduced!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 366, __extension__
__PRETTY_FUNCTION__))
;
367 // For Bitwidth 4 or more the following sum does not overflow.
368 LHS += RHS;
369 while (LHS.uge(Threshold))
370 LHS -= CM;
371 } else {
372 // To avoid problems with overflow do everything the same as above but using
373 // a larger type.
374 unsigned CM = 1U << CarmichaelShift(Bitwidth);
375 unsigned Threshold = CM + Bitwidth;
376 assert(LHS.getZExtValue() < Threshold && RHS.getZExtValue() < Threshold &&(static_cast <bool> (LHS.getZExtValue() < Threshold &&
RHS.getZExtValue() < Threshold && "Weights not reduced!"
) ? void (0) : __assert_fail ("LHS.getZExtValue() < Threshold && RHS.getZExtValue() < Threshold && \"Weights not reduced!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 377, __extension__
__PRETTY_FUNCTION__))
377 "Weights not reduced!")(static_cast <bool> (LHS.getZExtValue() < Threshold &&
RHS.getZExtValue() < Threshold && "Weights not reduced!"
) ? void (0) : __assert_fail ("LHS.getZExtValue() < Threshold && RHS.getZExtValue() < Threshold && \"Weights not reduced!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 377, __extension__
__PRETTY_FUNCTION__))
;
378 unsigned Total = LHS.getZExtValue() + RHS.getZExtValue();
379 while (Total >= Threshold)
380 Total -= CM;
381 LHS = Total;
382 }
383}
384
385using RepeatedValue = std::pair<Value*, APInt>;
386
387/// Given an associative binary expression, return the leaf
388/// nodes in Ops along with their weights (how many times the leaf occurs). The
389/// original expression is the same as
390/// (Ops[0].first op Ops[0].first op ... Ops[0].first) <- Ops[0].second times
391/// op
392/// (Ops[1].first op Ops[1].first op ... Ops[1].first) <- Ops[1].second times
393/// op
394/// ...
395/// op
396/// (Ops[N].first op Ops[N].first op ... Ops[N].first) <- Ops[N].second times
397///
398/// Note that the values Ops[0].first, ..., Ops[N].first are all distinct.
399///
400/// This routine may modify the function, in which case it returns 'true'. The
401/// changes it makes may well be destructive, changing the value computed by 'I'
402/// to something completely different. Thus if the routine returns 'true' then
403/// you MUST either replace I with a new expression computed from the Ops array,
404/// or use RewriteExprTree to put the values back in.
405///
406/// A leaf node is either not a binary operation of the same kind as the root
407/// node 'I' (i.e. is not a binary operator at all, or is, but with a different
408/// opcode), or is the same kind of binary operator but has a use which either
409/// does not belong to the expression, or does belong to the expression but is
410/// a leaf node. Every leaf node has at least one use that is a non-leaf node
411/// of the expression, while for non-leaf nodes (except for the root 'I') every
412/// use is a non-leaf node of the expression.
413///
414/// For example:
415/// expression graph node names
416///
417/// + | I
418/// / \ |
419/// + + | A, B
420/// / \ / \ |
421/// * + * | C, D, E
422/// / \ / \ / \ |
423/// + * | F, G
424///
425/// The leaf nodes are C, E, F and G. The Ops array will contain (maybe not in
426/// that order) (C, 1), (E, 1), (F, 2), (G, 2).
427///
428/// The expression is maximal: if some instruction is a binary operator of the
429/// same kind as 'I', and all of its uses are non-leaf nodes of the expression,
430/// then the instruction also belongs to the expression, is not a leaf node of
431/// it, and its operands also belong to the expression (but may be leaf nodes).
432///
433/// NOTE: This routine will set operands of non-leaf non-root nodes to undef in
434/// order to ensure that every non-root node in the expression has *exactly one*
435/// use by a non-leaf node of the expression. This destruction means that the
436/// caller MUST either replace 'I' with a new expression or use something like
437/// RewriteExprTree to put the values back in if the routine indicates that it
438/// made a change by returning 'true'.
439///
440/// In the above example either the right operand of A or the left operand of B
441/// will be replaced by undef. If it is B's operand then this gives:
442///
443/// + | I
444/// / \ |
445/// + + | A, B - operand of B replaced with undef
446/// / \ \ |
447/// * + * | C, D, E
448/// / \ / \ / \ |
449/// + * | F, G
450///
451/// Note that such undef operands can only be reached by passing through 'I'.
452/// For example, if you visit operands recursively starting from a leaf node
453/// then you will never see such an undef operand unless you get back to 'I',
454/// which requires passing through a phi node.
455///
456/// Note that this routine may also mutate binary operators of the wrong type
457/// that have all uses inside the expression (i.e. only used by non-leaf nodes
458/// of the expression) if it can turn them into binary operators of the right
459/// type and thus make the expression bigger.
460static bool LinearizeExprTree(Instruction *I,
461 SmallVectorImpl<RepeatedValue> &Ops,
462 ReassociatePass::OrderedSet &ToRedo) {
463 assert((isa<UnaryOperator>(I) || isa<BinaryOperator>(I)) &&(static_cast <bool> ((isa<UnaryOperator>(I) || isa
<BinaryOperator>(I)) && "Expected a UnaryOperator or BinaryOperator!"
) ? void (0) : __assert_fail ("(isa<UnaryOperator>(I) || isa<BinaryOperator>(I)) && \"Expected a UnaryOperator or BinaryOperator!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 464, __extension__
__PRETTY_FUNCTION__))
464 "Expected a UnaryOperator or BinaryOperator!")(static_cast <bool> ((isa<UnaryOperator>(I) || isa
<BinaryOperator>(I)) && "Expected a UnaryOperator or BinaryOperator!"
) ? void (0) : __assert_fail ("(isa<UnaryOperator>(I) || isa<BinaryOperator>(I)) && \"Expected a UnaryOperator or BinaryOperator!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 464, __extension__
__PRETTY_FUNCTION__))
;
465 LLVM_DEBUG(dbgs() << "LINEARIZE: " << *I << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "LINEARIZE: " << *I <<
'\n'; } } while (false)
;
466 unsigned Bitwidth = I->getType()->getScalarType()->getPrimitiveSizeInBits();
467 unsigned Opcode = I->getOpcode();
468 assert(I->isAssociative() && I->isCommutative() &&(static_cast <bool> (I->isAssociative() && I
->isCommutative() && "Expected an associative and commutative operation!"
) ? void (0) : __assert_fail ("I->isAssociative() && I->isCommutative() && \"Expected an associative and commutative operation!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 469, __extension__
__PRETTY_FUNCTION__))
469 "Expected an associative and commutative operation!")(static_cast <bool> (I->isAssociative() && I
->isCommutative() && "Expected an associative and commutative operation!"
) ? void (0) : __assert_fail ("I->isAssociative() && I->isCommutative() && \"Expected an associative and commutative operation!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 469, __extension__
__PRETTY_FUNCTION__))
;
470
471 // Visit all operands of the expression, keeping track of their weight (the
472 // number of paths from the expression root to the operand, or if you like
473 // the number of times that operand occurs in the linearized expression).
474 // For example, if I = X + A, where X = A + B, then I, X and B have weight 1
475 // while A has weight two.
476
477 // Worklist of non-leaf nodes (their operands are in the expression too) along
478 // with their weights, representing a certain number of paths to the operator.
479 // If an operator occurs in the worklist multiple times then we found multiple
480 // ways to get to it.
481 SmallVector<std::pair<Instruction*, APInt>, 8> Worklist; // (Op, Weight)
482 Worklist.push_back(std::make_pair(I, APInt(Bitwidth, 1)));
483 bool Changed = false;
484
485 // Leaves of the expression are values that either aren't the right kind of
486 // operation (eg: a constant, or a multiply in an add tree), or are, but have
487 // some uses that are not inside the expression. For example, in I = X + X,
488 // X = A + B, the value X has two uses (by I) that are in the expression. If
489 // X has any other uses, for example in a return instruction, then we consider
490 // X to be a leaf, and won't analyze it further. When we first visit a value,
491 // if it has more than one use then at first we conservatively consider it to
492 // be a leaf. Later, as the expression is explored, we may discover some more
493 // uses of the value from inside the expression. If all uses turn out to be
494 // from within the expression (and the value is a binary operator of the right
495 // kind) then the value is no longer considered to be a leaf, and its operands
496 // are explored.
497
498 // Leaves - Keeps track of the set of putative leaves as well as the number of
499 // paths to each leaf seen so far.
500 using LeafMap = DenseMap<Value *, APInt>;
501 LeafMap Leaves; // Leaf -> Total weight so far.
502 SmallVector<Value *, 8> LeafOrder; // Ensure deterministic leaf output order.
503
504#ifndef NDEBUG
505 SmallPtrSet<Value *, 8> Visited; // For checking the iteration scheme.
506#endif
507 while (!Worklist.empty()) {
508 std::pair<Instruction*, APInt> P = Worklist.pop_back_val();
509 I = P.first; // We examine the operands of this binary operator.
510
511 for (unsigned OpIdx = 0; OpIdx < I->getNumOperands(); ++OpIdx) { // Visit operands.
512 Value *Op = I->getOperand(OpIdx);
513 APInt Weight = P.second; // Number of paths to this operand.
514 LLVM_DEBUG(dbgs() << "OPERAND: " << *Op << " (" << Weight << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "OPERAND: " << *Op <<
" (" << Weight << ")\n"; } } while (false)
;
515 assert(!Op->use_empty() && "No uses, so how did we get to it?!")(static_cast <bool> (!Op->use_empty() && "No uses, so how did we get to it?!"
) ? void (0) : __assert_fail ("!Op->use_empty() && \"No uses, so how did we get to it?!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 515, __extension__
__PRETTY_FUNCTION__))
;
516
517 // If this is a binary operation of the right kind with only one use then
518 // add its operands to the expression.
519 if (BinaryOperator *BO = isReassociableOp(Op, Opcode)) {
520 assert(Visited.insert(Op).second && "Not first visit!")(static_cast <bool> (Visited.insert(Op).second &&
"Not first visit!") ? void (0) : __assert_fail ("Visited.insert(Op).second && \"Not first visit!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 520, __extension__
__PRETTY_FUNCTION__))
;
521 LLVM_DEBUG(dbgs() << "DIRECT ADD: " << *Op << " (" << Weight << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "DIRECT ADD: " << *Op
<< " (" << Weight << ")\n"; } } while (false
)
;
522 Worklist.push_back(std::make_pair(BO, Weight));
523 continue;
524 }
525
526 // Appears to be a leaf. Is the operand already in the set of leaves?
527 LeafMap::iterator It = Leaves.find(Op);
528 if (It == Leaves.end()) {
529 // Not in the leaf map. Must be the first time we saw this operand.
530 assert(Visited.insert(Op).second && "Not first visit!")(static_cast <bool> (Visited.insert(Op).second &&
"Not first visit!") ? void (0) : __assert_fail ("Visited.insert(Op).second && \"Not first visit!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 530, __extension__
__PRETTY_FUNCTION__))
;
531 if (!Op->hasOneUse()) {
532 // This value has uses not accounted for by the expression, so it is
533 // not safe to modify. Mark it as being a leaf.
534 LLVM_DEBUG(dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "ADD USES LEAF: " <<
*Op << " (" << Weight << ")\n"; } } while (
false)
535 << "ADD USES LEAF: " << *Op << " (" << Weight << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "ADD USES LEAF: " <<
*Op << " (" << Weight << ")\n"; } } while (
false)
;
536 LeafOrder.push_back(Op);
537 Leaves[Op] = Weight;
538 continue;
539 }
540 // No uses outside the expression, try morphing it.
541 } else {
542 // Already in the leaf map.
543 assert(It != Leaves.end() && Visited.count(Op) &&(static_cast <bool> (It != Leaves.end() && Visited
.count(Op) && "In leaf map but not visited!") ? void (
0) : __assert_fail ("It != Leaves.end() && Visited.count(Op) && \"In leaf map but not visited!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 544, __extension__
__PRETTY_FUNCTION__))
544 "In leaf map but not visited!")(static_cast <bool> (It != Leaves.end() && Visited
.count(Op) && "In leaf map but not visited!") ? void (
0) : __assert_fail ("It != Leaves.end() && Visited.count(Op) && \"In leaf map but not visited!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 544, __extension__
__PRETTY_FUNCTION__))
;
545
546 // Update the number of paths to the leaf.
547 IncorporateWeight(It->second, Weight, Opcode);
548
549#if 0 // TODO: Re-enable once PR13021 is fixed.
550 // The leaf already has one use from inside the expression. As we want
551 // exactly one such use, drop this new use of the leaf.
552 assert(!Op->hasOneUse() && "Only one use, but we got here twice!")(static_cast <bool> (!Op->hasOneUse() && "Only one use, but we got here twice!"
) ? void (0) : __assert_fail ("!Op->hasOneUse() && \"Only one use, but we got here twice!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 552, __extension__
__PRETTY_FUNCTION__))
;
553 I->setOperand(OpIdx, UndefValue::get(I->getType()));
554 Changed = true;
555
556 // If the leaf is a binary operation of the right kind and we now see
557 // that its multiple original uses were in fact all by nodes belonging
558 // to the expression, then no longer consider it to be a leaf and add
559 // its operands to the expression.
560 if (BinaryOperator *BO = isReassociableOp(Op, Opcode)) {
561 LLVM_DEBUG(dbgs() << "UNLEAF: " << *Op << " (" << It->second << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "UNLEAF: " << *Op <<
" (" << It->second << ")\n"; } } while (false
)
;
562 Worklist.push_back(std::make_pair(BO, It->second));
563 Leaves.erase(It);
564 continue;
565 }
566#endif
567
568 // If we still have uses that are not accounted for by the expression
569 // then it is not safe to modify the value.
570 if (!Op->hasOneUse())
571 continue;
572
573 // No uses outside the expression, try morphing it.
574 Weight = It->second;
575 Leaves.erase(It); // Since the value may be morphed below.
576 }
577
578 // At this point we have a value which, first of all, is not a binary
579 // expression of the right kind, and secondly, is only used inside the
580 // expression. This means that it can safely be modified. See if we
581 // can usefully morph it into an expression of the right kind.
582 assert((!isa<Instruction>(Op) ||(static_cast <bool> ((!isa<Instruction>(Op) || cast
<Instruction>(Op)->getOpcode() != Opcode || (isa<
FPMathOperator>(Op) && !hasFPAssociativeFlags(cast
<Instruction>(Op)))) && "Should have been handled above!"
) ? void (0) : __assert_fail ("(!isa<Instruction>(Op) || cast<Instruction>(Op)->getOpcode() != Opcode || (isa<FPMathOperator>(Op) && !hasFPAssociativeFlags(cast<Instruction>(Op)))) && \"Should have been handled above!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 586, __extension__
__PRETTY_FUNCTION__))
583 cast<Instruction>(Op)->getOpcode() != Opcode(static_cast <bool> ((!isa<Instruction>(Op) || cast
<Instruction>(Op)->getOpcode() != Opcode || (isa<
FPMathOperator>(Op) && !hasFPAssociativeFlags(cast
<Instruction>(Op)))) && "Should have been handled above!"
) ? void (0) : __assert_fail ("(!isa<Instruction>(Op) || cast<Instruction>(Op)->getOpcode() != Opcode || (isa<FPMathOperator>(Op) && !hasFPAssociativeFlags(cast<Instruction>(Op)))) && \"Should have been handled above!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 586, __extension__
__PRETTY_FUNCTION__))
584 || (isa<FPMathOperator>(Op) &&(static_cast <bool> ((!isa<Instruction>(Op) || cast
<Instruction>(Op)->getOpcode() != Opcode || (isa<
FPMathOperator>(Op) && !hasFPAssociativeFlags(cast
<Instruction>(Op)))) && "Should have been handled above!"
) ? void (0) : __assert_fail ("(!isa<Instruction>(Op) || cast<Instruction>(Op)->getOpcode() != Opcode || (isa<FPMathOperator>(Op) && !hasFPAssociativeFlags(cast<Instruction>(Op)))) && \"Should have been handled above!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 586, __extension__
__PRETTY_FUNCTION__))
585 !hasFPAssociativeFlags(cast<Instruction>(Op)))) &&(static_cast <bool> ((!isa<Instruction>(Op) || cast
<Instruction>(Op)->getOpcode() != Opcode || (isa<
FPMathOperator>(Op) && !hasFPAssociativeFlags(cast
<Instruction>(Op)))) && "Should have been handled above!"
) ? void (0) : __assert_fail ("(!isa<Instruction>(Op) || cast<Instruction>(Op)->getOpcode() != Opcode || (isa<FPMathOperator>(Op) && !hasFPAssociativeFlags(cast<Instruction>(Op)))) && \"Should have been handled above!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 586, __extension__
__PRETTY_FUNCTION__))
586 "Should have been handled above!")(static_cast <bool> ((!isa<Instruction>(Op) || cast
<Instruction>(Op)->getOpcode() != Opcode || (isa<
FPMathOperator>(Op) && !hasFPAssociativeFlags(cast
<Instruction>(Op)))) && "Should have been handled above!"
) ? void (0) : __assert_fail ("(!isa<Instruction>(Op) || cast<Instruction>(Op)->getOpcode() != Opcode || (isa<FPMathOperator>(Op) && !hasFPAssociativeFlags(cast<Instruction>(Op)))) && \"Should have been handled above!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 586, __extension__
__PRETTY_FUNCTION__))
;
587 assert(Op->hasOneUse() && "Has uses outside the expression tree!")(static_cast <bool> (Op->hasOneUse() && "Has uses outside the expression tree!"
) ? void (0) : __assert_fail ("Op->hasOneUse() && \"Has uses outside the expression tree!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 587, __extension__
__PRETTY_FUNCTION__))
;
588
589 // If this is a multiply expression, turn any internal negations into
590 // multiplies by -1 so they can be reassociated. Add any users of the
591 // newly created multiplication by -1 to the redo list, so any
592 // reassociation opportunities that are exposed will be reassociated
593 // further.
594 Instruction *Neg;
595 if (((Opcode == Instruction::Mul && match(Op, m_Neg(m_Value()))) ||
596 (Opcode == Instruction::FMul && match(Op, m_FNeg(m_Value())))) &&
597 match(Op, m_Instruction(Neg))) {
598 LLVM_DEBUG(dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "MORPH LEAF: " << *Op
<< " (" << Weight << ") TO "; } } while (false
)
599 << "MORPH LEAF: " << *Op << " (" << Weight << ") TO ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "MORPH LEAF: " << *Op
<< " (" << Weight << ") TO "; } } while (false
)
;
600 Instruction *Mul = LowerNegateToMultiply(Neg);
601 LLVM_DEBUG(dbgs() << *Mul << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << *Mul << '\n'; } } while
(false)
;
602 Worklist.push_back(std::make_pair(Mul, Weight));
603 for (User *U : Mul->users()) {
604 if (BinaryOperator *UserBO = dyn_cast<BinaryOperator>(U))
605 ToRedo.insert(UserBO);
606 }
607 ToRedo.insert(Neg);
608 Changed = true;
609 continue;
610 }
611
612 // Failed to morph into an expression of the right type. This really is
613 // a leaf.
614 LLVM_DEBUG(dbgs() << "ADD LEAF: " << *Op << " (" << Weight << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "ADD LEAF: " << *Op <<
" (" << Weight << ")\n"; } } while (false)
;
615 assert(!isReassociableOp(Op, Opcode) && "Value was morphed?")(static_cast <bool> (!isReassociableOp(Op, Opcode) &&
"Value was morphed?") ? void (0) : __assert_fail ("!isReassociableOp(Op, Opcode) && \"Value was morphed?\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 615, __extension__
__PRETTY_FUNCTION__))
;
616 LeafOrder.push_back(Op);
617 Leaves[Op] = Weight;
618 }
619 }
620
621 // The leaves, repeated according to their weights, represent the linearized
622 // form of the expression.
623 for (Value *V : LeafOrder) {
624 LeafMap::iterator It = Leaves.find(V);
625 if (It == Leaves.end())
626 // Node initially thought to be a leaf wasn't.
627 continue;
628 assert(!isReassociableOp(V, Opcode) && "Shouldn't be a leaf!")(static_cast <bool> (!isReassociableOp(V, Opcode) &&
"Shouldn't be a leaf!") ? void (0) : __assert_fail ("!isReassociableOp(V, Opcode) && \"Shouldn't be a leaf!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 628, __extension__
__PRETTY_FUNCTION__))
;
629 APInt Weight = It->second;
630 if (Weight.isMinValue())
631 // Leaf already output or weight reduction eliminated it.
632 continue;
633 // Ensure the leaf is only output once.
634 It->second = 0;
635 Ops.push_back(std::make_pair(V, Weight));
636 }
637
638 // For nilpotent operations or addition there may be no operands, for example
639 // because the expression was "X xor X" or consisted of 2^Bitwidth additions:
640 // in both cases the weight reduces to 0 causing the value to be skipped.
641 if (Ops.empty()) {
642 Constant *Identity = ConstantExpr::getBinOpIdentity(Opcode, I->getType());
643 assert(Identity && "Associative operation without identity!")(static_cast <bool> (Identity && "Associative operation without identity!"
) ? void (0) : __assert_fail ("Identity && \"Associative operation without identity!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 643, __extension__
__PRETTY_FUNCTION__))
;
644 Ops.emplace_back(Identity, APInt(Bitwidth, 1));
645 }
646
647 return Changed;
648}
649
650/// Now that the operands for this expression tree are
651/// linearized and optimized, emit them in-order.
652void ReassociatePass::RewriteExprTree(BinaryOperator *I,
653 SmallVectorImpl<ValueEntry> &Ops) {
654 assert(Ops.size() > 1 && "Single values should be used directly!")(static_cast <bool> (Ops.size() > 1 && "Single values should be used directly!"
) ? void (0) : __assert_fail ("Ops.size() > 1 && \"Single values should be used directly!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 654, __extension__
__PRETTY_FUNCTION__))
;
655
656 // Since our optimizations should never increase the number of operations, the
657 // new expression can usually be written reusing the existing binary operators
658 // from the original expression tree, without creating any new instructions,
659 // though the rewritten expression may have a completely different topology.
660 // We take care to not change anything if the new expression will be the same
661 // as the original. If more than trivial changes (like commuting operands)
662 // were made then we are obliged to clear out any optional subclass data like
663 // nsw flags.
664
665 /// NodesToRewrite - Nodes from the original expression available for writing
666 /// the new expression into.
667 SmallVector<BinaryOperator*, 8> NodesToRewrite;
668 unsigned Opcode = I->getOpcode();
669 BinaryOperator *Op = I;
670
671 /// NotRewritable - The operands being written will be the leaves of the new
672 /// expression and must not be used as inner nodes (via NodesToRewrite) by
673 /// mistake. Inner nodes are always reassociable, and usually leaves are not
674 /// (if they were they would have been incorporated into the expression and so
675 /// would not be leaves), so most of the time there is no danger of this. But
676 /// in rare cases a leaf may become reassociable if an optimization kills uses
677 /// of it, or it may momentarily become reassociable during rewriting (below)
678 /// due it being removed as an operand of one of its uses. Ensure that misuse
679 /// of leaf nodes as inner nodes cannot occur by remembering all of the future
680 /// leaves and refusing to reuse any of them as inner nodes.
681 SmallPtrSet<Value*, 8> NotRewritable;
682 for (unsigned i = 0, e = Ops.size(); i != e; ++i)
683 NotRewritable.insert(Ops[i].Op);
684
685 // ExpressionChanged - Non-null if the rewritten expression differs from the
686 // original in some non-trivial way, requiring the clearing of optional flags.
687 // Flags are cleared from the operator in ExpressionChanged up to I inclusive.
688 BinaryOperator *ExpressionChanged = nullptr;
689 for (unsigned i = 0; ; ++i) {
690 // The last operation (which comes earliest in the IR) is special as both
691 // operands will come from Ops, rather than just one with the other being
692 // a subexpression.
693 if (i+2 == Ops.size()) {
694 Value *NewLHS = Ops[i].Op;
695 Value *NewRHS = Ops[i+1].Op;
696 Value *OldLHS = Op->getOperand(0);
697 Value *OldRHS = Op->getOperand(1);
698
699 if (NewLHS == OldLHS && NewRHS == OldRHS)
700 // Nothing changed, leave it alone.
701 break;
702
703 if (NewLHS == OldRHS && NewRHS == OldLHS) {
704 // The order of the operands was reversed. Swap them.
705 LLVM_DEBUG(dbgs() << "RA: " << *Op << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "RA: " << *Op <<
'\n'; } } while (false)
;
706 Op->swapOperands();
707 LLVM_DEBUG(dbgs() << "TO: " << *Op << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "TO: " << *Op <<
'\n'; } } while (false)
;
708 MadeChange = true;
709 ++NumChanged;
710 break;
711 }
712
713 // The new operation differs non-trivially from the original. Overwrite
714 // the old operands with the new ones.
715 LLVM_DEBUG(dbgs() << "RA: " << *Op << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "RA: " << *Op <<
'\n'; } } while (false)
;
716 if (NewLHS != OldLHS) {
717 BinaryOperator *BO = isReassociableOp(OldLHS, Opcode);
718 if (BO && !NotRewritable.count(BO))
719 NodesToRewrite.push_back(BO);
720 Op->setOperand(0, NewLHS);
721 }
722 if (NewRHS != OldRHS) {
723 BinaryOperator *BO = isReassociableOp(OldRHS, Opcode);
724 if (BO && !NotRewritable.count(BO))
725 NodesToRewrite.push_back(BO);
726 Op->setOperand(1, NewRHS);
727 }
728 LLVM_DEBUG(dbgs() << "TO: " << *Op << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "TO: " << *Op <<
'\n'; } } while (false)
;
729
730 ExpressionChanged = Op;
731 MadeChange = true;
732 ++NumChanged;
733
734 break;
735 }
736
737 // Not the last operation. The left-hand side will be a sub-expression
738 // while the right-hand side will be the current element of Ops.
739 Value *NewRHS = Ops[i].Op;
740 if (NewRHS != Op->getOperand(1)) {
741 LLVM_DEBUG(dbgs() << "RA: " << *Op << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "RA: " << *Op <<
'\n'; } } while (false)
;
742 if (NewRHS == Op->getOperand(0)) {
743 // The new right-hand side was already present as the left operand. If
744 // we are lucky then swapping the operands will sort out both of them.
745 Op->swapOperands();
746 } else {
747 // Overwrite with the new right-hand side.
748 BinaryOperator *BO = isReassociableOp(Op->getOperand(1), Opcode);
749 if (BO && !NotRewritable.count(BO))
750 NodesToRewrite.push_back(BO);
751 Op->setOperand(1, NewRHS);
752 ExpressionChanged = Op;
753 }
754 LLVM_DEBUG(dbgs() << "TO: " << *Op << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "TO: " << *Op <<
'\n'; } } while (false)
;
755 MadeChange = true;
756 ++NumChanged;
757 }
758
759 // Now deal with the left-hand side. If this is already an operation node
760 // from the original expression then just rewrite the rest of the expression
761 // into it.
762 BinaryOperator *BO = isReassociableOp(Op->getOperand(0), Opcode);
763 if (BO && !NotRewritable.count(BO)) {
764 Op = BO;
765 continue;
766 }
767
768 // Otherwise, grab a spare node from the original expression and use that as
769 // the left-hand side. If there are no nodes left then the optimizers made
770 // an expression with more nodes than the original! This usually means that
771 // they did something stupid but it might mean that the problem was just too
772 // hard (finding the mimimal number of multiplications needed to realize a
773 // multiplication expression is NP-complete). Whatever the reason, smart or
774 // stupid, create a new node if there are none left.
775 BinaryOperator *NewOp;
776 if (NodesToRewrite.empty()) {
777 Constant *Undef = UndefValue::get(I->getType());
778 NewOp = BinaryOperator::Create(Instruction::BinaryOps(Opcode),
779 Undef, Undef, "", I);
780 if (isa<FPMathOperator>(NewOp))
781 NewOp->setFastMathFlags(I->getFastMathFlags());
782 } else {
783 NewOp = NodesToRewrite.pop_back_val();
784 }
785
786 LLVM_DEBUG(dbgs() << "RA: " << *Op << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "RA: " << *Op <<
'\n'; } } while (false)
;
787 Op->setOperand(0, NewOp);
788 LLVM_DEBUG(dbgs() << "TO: " << *Op << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "TO: " << *Op <<
'\n'; } } while (false)
;
789 ExpressionChanged = Op;
790 MadeChange = true;
791 ++NumChanged;
792 Op = NewOp;
793 }
794
795 // If the expression changed non-trivially then clear out all subclass data
796 // starting from the operator specified in ExpressionChanged, and compactify
797 // the operators to just before the expression root to guarantee that the
798 // expression tree is dominated by all of Ops.
799 if (ExpressionChanged)
800 do {
801 // Preserve FastMathFlags.
802 if (isa<FPMathOperator>(I)) {
803 FastMathFlags Flags = I->getFastMathFlags();
804 ExpressionChanged->clearSubclassOptionalData();
805 ExpressionChanged->setFastMathFlags(Flags);
806 } else
807 ExpressionChanged->clearSubclassOptionalData();
808
809 if (ExpressionChanged == I)
810 break;
811
812 // Discard any debug info related to the expressions that has changed (we
813 // can leave debug infor related to the root, since the result of the
814 // expression tree should be the same even after reassociation).
815 replaceDbgUsesWithUndef(ExpressionChanged);
816
817 ExpressionChanged->moveBefore(I);
818 ExpressionChanged = cast<BinaryOperator>(*ExpressionChanged->user_begin());
819 } while (true);
820
821 // Throw away any left over nodes from the original expression.
822 for (unsigned i = 0, e = NodesToRewrite.size(); i != e; ++i)
823 RedoInsts.insert(NodesToRewrite[i]);
824}
825
826/// Insert instructions before the instruction pointed to by BI,
827/// that computes the negative version of the value specified. The negative
828/// version of the value is returned, and BI is left pointing at the instruction
829/// that should be processed next by the reassociation pass.
830/// Also add intermediate instructions to the redo list that are modified while
831/// pushing the negates through adds. These will be revisited to see if
832/// additional opportunities have been exposed.
833static Value *NegateValue(Value *V, Instruction *BI,
834 ReassociatePass::OrderedSet &ToRedo) {
835 if (auto *C = dyn_cast<Constant>(V)) {
836 const DataLayout &DL = BI->getModule()->getDataLayout();
837 Constant *Res = C->getType()->isFPOrFPVectorTy()
838 ? ConstantFoldUnaryOpOperand(Instruction::FNeg, C, DL)
839 : ConstantExpr::getNeg(C);
840 if (Res)
841 return Res;
842 }
843
844 // We are trying to expose opportunity for reassociation. One of the things
845 // that we want to do to achieve this is to push a negation as deep into an
846 // expression chain as possible, to expose the add instructions. In practice,
847 // this means that we turn this:
848 // X = -(A+12+C+D) into X = -A + -12 + -C + -D = -12 + -A + -C + -D
849 // so that later, a: Y = 12+X could get reassociated with the -12 to eliminate
850 // the constants. We assume that instcombine will clean up the mess later if
851 // we introduce tons of unnecessary negation instructions.
852 //
853 if (BinaryOperator *I =
854 isReassociableOp(V, Instruction::Add, Instruction::FAdd)) {
855 // Push the negates through the add.
856 I->setOperand(0, NegateValue(I->getOperand(0), BI, ToRedo));
857 I->setOperand(1, NegateValue(I->getOperand(1), BI, ToRedo));
858 if (I->getOpcode() == Instruction::Add) {
859 I->setHasNoUnsignedWrap(false);
860 I->setHasNoSignedWrap(false);
861 }
862
863 // We must move the add instruction here, because the neg instructions do
864 // not dominate the old add instruction in general. By moving it, we are
865 // assured that the neg instructions we just inserted dominate the
866 // instruction we are about to insert after them.
867 //
868 I->moveBefore(BI);
869 I->setName(I->getName()+".neg");
870
871 // Add the intermediate negates to the redo list as processing them later
872 // could expose more reassociating opportunities.
873 ToRedo.insert(I);
874 return I;
875 }
876
877 // Okay, we need to materialize a negated version of V with an instruction.
878 // Scan the use lists of V to see if we have one already.
879 for (User *U : V->users()) {
880 if (!match(U, m_Neg(m_Value())) && !match(U, m_FNeg(m_Value())))
881 continue;
882
883 // We found one! Now we have to make sure that the definition dominates
884 // this use. We do this by moving it to the entry block (if it is a
885 // non-instruction value) or right after the definition. These negates will
886 // be zapped by reassociate later, so we don't need much finesse here.
887 Instruction *TheNeg = dyn_cast<Instruction>(U);
888
889 // We can't safely propagate a vector zero constant with poison/undef lanes.
890 Constant *C;
891 if (match(TheNeg, m_BinOp(m_Constant(C), m_Value())) &&
892 C->containsUndefOrPoisonElement())
893 continue;
894
895 // Verify that the negate is in this function, V might be a constant expr.
896 if (!TheNeg ||
897 TheNeg->getParent()->getParent() != BI->getParent()->getParent())
898 continue;
899
900 Instruction *InsertPt;
901 if (Instruction *InstInput = dyn_cast<Instruction>(V)) {
902 InsertPt = InstInput->getInsertionPointAfterDef();
903 if (!InsertPt)
904 continue;
905 } else {
906 InsertPt = &*TheNeg->getFunction()->getEntryBlock().begin();
907 }
908
909 TheNeg->moveBefore(InsertPt);
910 if (TheNeg->getOpcode() == Instruction::Sub) {
911 TheNeg->setHasNoUnsignedWrap(false);
912 TheNeg->setHasNoSignedWrap(false);
913 } else {
914 TheNeg->andIRFlags(BI);
915 }
916 ToRedo.insert(TheNeg);
917 return TheNeg;
918 }
919
920 // Insert a 'neg' instruction that subtracts the value from zero to get the
921 // negation.
922 Instruction *NewNeg = CreateNeg(V, V->getName() + ".neg", BI, BI);
923 ToRedo.insert(NewNeg);
924 return NewNeg;
925}
926
927// See if this `or` looks like an load widening reduction, i.e. that it
928// consists of an `or`/`shl`/`zext`/`load` nodes only. Note that we don't
929// ensure that the pattern is *really* a load widening reduction,
930// we do not ensure that it can really be replaced with a widened load,
931// only that it mostly looks like one.
932static bool isLoadCombineCandidate(Instruction *Or) {
933 SmallVector<Instruction *, 8> Worklist;
934 SmallSet<Instruction *, 8> Visited;
935
936 auto Enqueue = [&](Value *V) {
937 auto *I = dyn_cast<Instruction>(V);
938 // Each node of an `or` reduction must be an instruction,
939 if (!I)
940 return false; // Node is certainly not part of an `or` load reduction.
941 // Only process instructions we have never processed before.
942 if (Visited.insert(I).second)
943 Worklist.emplace_back(I);
944 return true; // Will need to look at parent nodes.
945 };
946
947 if (!Enqueue(Or))
948 return false; // Not an `or` reduction pattern.
949
950 while (!Worklist.empty()) {
951 auto *I = Worklist.pop_back_val();
952
953 // Okay, which instruction is this node?
954 switch (I->getOpcode()) {
955 case Instruction::Or:
956 // Got an `or` node. That's fine, just recurse into it's operands.
957 for (Value *Op : I->operands())
958 if (!Enqueue(Op))
959 return false; // Not an `or` reduction pattern.
960 continue;
961
962 case Instruction::Shl:
963 case Instruction::ZExt:
964 // `shl`/`zext` nodes are fine, just recurse into their base operand.
965 if (!Enqueue(I->getOperand(0)))
966 return false; // Not an `or` reduction pattern.
967 continue;
968
969 case Instruction::Load:
970 // Perfect, `load` node means we've reached an edge of the graph.
971 continue;
972
973 default: // Unknown node.
974 return false; // Not an `or` reduction pattern.
975 }
976 }
977
978 return true;
979}
980
981/// Return true if it may be profitable to convert this (X|Y) into (X+Y).
982static bool shouldConvertOrWithNoCommonBitsToAdd(Instruction *Or) {
983 // Don't bother to convert this up unless either the LHS is an associable add
984 // or subtract or mul or if this is only used by one of the above.
985 // This is only a compile-time improvement, it is not needed for correctness!
986 auto isInteresting = [](Value *V) {
987 for (auto Op : {Instruction::Add, Instruction::Sub, Instruction::Mul,
988 Instruction::Shl})
989 if (isReassociableOp(V, Op))
990 return true;
991 return false;
992 };
993
994 if (any_of(Or->operands(), isInteresting))
995 return true;
996
997 Value *VB = Or->user_back();
998 if (Or->hasOneUse() && isInteresting(VB))
999 return true;
1000
1001 return false;
1002}
1003
1004/// If we have (X|Y), and iff X and Y have no common bits set,
1005/// transform this into (X+Y) to allow arithmetics reassociation.
1006static BinaryOperator *convertOrWithNoCommonBitsToAdd(Instruction *Or) {
1007 // Convert an or into an add.
1008 BinaryOperator *New =
1009 CreateAdd(Or->getOperand(0), Or->getOperand(1), "", Or, Or);
1010 New->setHasNoSignedWrap();
1011 New->setHasNoUnsignedWrap();
1012 New->takeName(Or);
1013
1014 // Everyone now refers to the add instruction.
1015 Or->replaceAllUsesWith(New);
1016 New->setDebugLoc(Or->getDebugLoc());
1017
1018 LLVM_DEBUG(dbgs() << "Converted or into an add: " << *New << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "Converted or into an add: "
<< *New << '\n'; } } while (false)
;
1019 return New;
1020}
1021
1022/// Return true if we should break up this subtract of X-Y into (X + -Y).
1023static bool ShouldBreakUpSubtract(Instruction *Sub) {
1024 // If this is a negation, we can't split it up!
1025 if (match(Sub, m_Neg(m_Value())) || match(Sub, m_FNeg(m_Value())))
1026 return false;
1027
1028 // Don't breakup X - undef.
1029 if (isa<UndefValue>(Sub->getOperand(1)))
1030 return false;
1031
1032 // Don't bother to break this up unless either the LHS is an associable add or
1033 // subtract or if this is only used by one.
1034 Value *V0 = Sub->getOperand(0);
1035 if (isReassociableOp(V0, Instruction::Add, Instruction::FAdd) ||
1036 isReassociableOp(V0, Instruction::Sub, Instruction::FSub))
1037 return true;
1038 Value *V1 = Sub->getOperand(1);
1039 if (isReassociableOp(V1, Instruction::Add, Instruction::FAdd) ||
1040 isReassociableOp(V1, Instruction::Sub, Instruction::FSub))
1041 return true;
1042 Value *VB = Sub->user_back();
1043 if (Sub->hasOneUse() &&
1044 (isReassociableOp(VB, Instruction::Add, Instruction::FAdd) ||
1045 isReassociableOp(VB, Instruction::Sub, Instruction::FSub)))
1046 return true;
1047
1048 return false;
1049}
1050
1051/// If we have (X-Y), and if either X is an add, or if this is only used by an
1052/// add, transform this into (X+(0-Y)) to promote better reassociation.
1053static BinaryOperator *BreakUpSubtract(Instruction *Sub,
1054 ReassociatePass::OrderedSet &ToRedo) {
1055 // Convert a subtract into an add and a neg instruction. This allows sub
1056 // instructions to be commuted with other add instructions.
1057 //
1058 // Calculate the negative value of Operand 1 of the sub instruction,
1059 // and set it as the RHS of the add instruction we just made.
1060 Value *NegVal = NegateValue(Sub->getOperand(1), Sub, ToRedo);
1061 BinaryOperator *New = CreateAdd(Sub->getOperand(0), NegVal, "", Sub, Sub);
1062 Sub->setOperand(0, Constant::getNullValue(Sub->getType())); // Drop use of op.
1063 Sub->setOperand(1, Constant::getNullValue(Sub->getType())); // Drop use of op.
1064 New->takeName(Sub);
1065
1066 // Everyone now refers to the add instruction.
1067 Sub->replaceAllUsesWith(New);
1068 New->setDebugLoc(Sub->getDebugLoc());
1069
1070 LLVM_DEBUG(dbgs() << "Negated: " << *New << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "Negated: " << *New <<
'\n'; } } while (false)
;
1071 return New;
1072}
1073
1074/// If this is a shift of a reassociable multiply or is used by one, change
1075/// this into a multiply by a constant to assist with further reassociation.
1076static BinaryOperator *ConvertShiftToMul(Instruction *Shl) {
1077 Constant *MulCst = ConstantInt::get(Shl->getType(), 1);
1078 auto *SA = cast<ConstantInt>(Shl->getOperand(1));
1079 MulCst = ConstantExpr::getShl(MulCst, SA);
1080
1081 BinaryOperator *Mul =
1082 BinaryOperator::CreateMul(Shl->getOperand(0), MulCst, "", Shl);
1083 Shl->setOperand(0, PoisonValue::get(Shl->getType())); // Drop use of op.
1084 Mul->takeName(Shl);
1085
1086 // Everyone now refers to the mul instruction.
1087 Shl->replaceAllUsesWith(Mul);
1088 Mul->setDebugLoc(Shl->getDebugLoc());
1089
1090 // We can safely preserve the nuw flag in all cases. It's also safe to turn a
1091 // nuw nsw shl into a nuw nsw mul. However, nsw in isolation requires special
1092 // handling. It can be preserved as long as we're not left shifting by
1093 // bitwidth - 1.
1094 bool NSW = cast<BinaryOperator>(Shl)->hasNoSignedWrap();
1095 bool NUW = cast<BinaryOperator>(Shl)->hasNoUnsignedWrap();
1096 unsigned BitWidth = Shl->getType()->getIntegerBitWidth();
1097 if (NSW && (NUW || SA->getValue().ult(BitWidth - 1)))
1098 Mul->setHasNoSignedWrap(true);
1099 Mul->setHasNoUnsignedWrap(NUW);
1100 return Mul;
1101}
1102
1103/// Scan backwards and forwards among values with the same rank as element i
1104/// to see if X exists. If X does not exist, return i. This is useful when
1105/// scanning for 'x' when we see '-x' because they both get the same rank.
1106static unsigned FindInOperandList(const SmallVectorImpl<ValueEntry> &Ops,
1107 unsigned i, Value *X) {
1108 unsigned XRank = Ops[i].Rank;
1109 unsigned e = Ops.size();
1110 for (unsigned j = i+1; j != e && Ops[j].Rank == XRank; ++j) {
1111 if (Ops[j].Op == X)
1112 return j;
1113 if (Instruction *I1 = dyn_cast<Instruction>(Ops[j].Op))
1114 if (Instruction *I2 = dyn_cast<Instruction>(X))
1115 if (I1->isIdenticalTo(I2))
1116 return j;
1117 }
1118 // Scan backwards.
1119 for (unsigned j = i-1; j != ~0U && Ops[j].Rank == XRank; --j) {
1120 if (Ops[j].Op == X)
1121 return j;
1122 if (Instruction *I1 = dyn_cast<Instruction>(Ops[j].Op))
1123 if (Instruction *I2 = dyn_cast<Instruction>(X))
1124 if (I1->isIdenticalTo(I2))
1125 return j;
1126 }
1127 return i;
1128}
1129
1130/// Emit a tree of add instructions, summing Ops together
1131/// and returning the result. Insert the tree before I.
1132static Value *EmitAddTreeOfValues(Instruction *I,
1133 SmallVectorImpl<WeakTrackingVH> &Ops) {
1134 if (Ops.size() == 1) return Ops.back();
1135
1136 Value *V1 = Ops.pop_back_val();
1137 Value *V2 = EmitAddTreeOfValues(I, Ops);
1138 return CreateAdd(V2, V1, "reass.add", I, I);
1139}
1140
1141/// If V is an expression tree that is a multiplication sequence,
1142/// and if this sequence contains a multiply by Factor,
1143/// remove Factor from the tree and return the new tree.
1144Value *ReassociatePass::RemoveFactorFromExpression(Value *V, Value *Factor) {
1145 BinaryOperator *BO = isReassociableOp(V, Instruction::Mul, Instruction::FMul);
1146 if (!BO)
1147 return nullptr;
1148
1149 SmallVector<RepeatedValue, 8> Tree;
1150 MadeChange |= LinearizeExprTree(BO, Tree, RedoInsts);
1151 SmallVector<ValueEntry, 8> Factors;
1152 Factors.reserve(Tree.size());
1153 for (unsigned i = 0, e = Tree.size(); i != e; ++i) {
1154 RepeatedValue E = Tree[i];
1155 Factors.append(E.second.getZExtValue(),
1156 ValueEntry(getRank(E.first), E.first));
1157 }
1158
1159 bool FoundFactor = false;
1160 bool NeedsNegate = false;
1161 for (unsigned i = 0, e = Factors.size(); i != e; ++i) {
1162 if (Factors[i].Op == Factor) {
1163 FoundFactor = true;
1164 Factors.erase(Factors.begin()+i);
1165 break;
1166 }
1167
1168 // If this is a negative version of this factor, remove it.
1169 if (ConstantInt *FC1 = dyn_cast<ConstantInt>(Factor)) {
1170 if (ConstantInt *FC2 = dyn_cast<ConstantInt>(Factors[i].Op))
1171 if (FC1->getValue() == -FC2->getValue()) {
1172 FoundFactor = NeedsNegate = true;
1173 Factors.erase(Factors.begin()+i);
1174 break;
1175 }
1176 } else if (ConstantFP *FC1 = dyn_cast<ConstantFP>(Factor)) {
1177 if (ConstantFP *FC2 = dyn_cast<ConstantFP>(Factors[i].Op)) {
1178 const APFloat &F1 = FC1->getValueAPF();
1179 APFloat F2(FC2->getValueAPF());
1180 F2.changeSign();
1181 if (F1 == F2) {
1182 FoundFactor = NeedsNegate = true;
1183 Factors.erase(Factors.begin() + i);
1184 break;
1185 }
1186 }
1187 }
1188 }
1189
1190 if (!FoundFactor) {
1191 // Make sure to restore the operands to the expression tree.
1192 RewriteExprTree(BO, Factors);
1193 return nullptr;
1194 }
1195
1196 BasicBlock::iterator InsertPt = ++BO->getIterator();
1197
1198 // If this was just a single multiply, remove the multiply and return the only
1199 // remaining operand.
1200 if (Factors.size() == 1) {
1201 RedoInsts.insert(BO);
1202 V = Factors[0].Op;
1203 } else {
1204 RewriteExprTree(BO, Factors);
1205 V = BO;
1206 }
1207
1208 if (NeedsNegate)
1209 V = CreateNeg(V, "neg", &*InsertPt, BO);
1210
1211 return V;
1212}
1213
1214/// If V is a single-use multiply, recursively add its operands as factors,
1215/// otherwise add V to the list of factors.
1216///
1217/// Ops is the top-level list of add operands we're trying to factor.
1218static void FindSingleUseMultiplyFactors(Value *V,
1219 SmallVectorImpl<Value*> &Factors) {
1220 BinaryOperator *BO = isReassociableOp(V, Instruction::Mul, Instruction::FMul);
1221 if (!BO) {
1222 Factors.push_back(V);
1223 return;
1224 }
1225
1226 // Otherwise, add the LHS and RHS to the list of factors.
1227 FindSingleUseMultiplyFactors(BO->getOperand(1), Factors);
1228 FindSingleUseMultiplyFactors(BO->getOperand(0), Factors);
1229}
1230
1231/// Optimize a series of operands to an 'and', 'or', or 'xor' instruction.
1232/// This optimizes based on identities. If it can be reduced to a single Value,
1233/// it is returned, otherwise the Ops list is mutated as necessary.
1234static Value *OptimizeAndOrXor(unsigned Opcode,
1235 SmallVectorImpl<ValueEntry> &Ops) {
1236 // Scan the operand lists looking for X and ~X pairs, along with X,X pairs.
1237 // If we find any, we can simplify the expression. X&~X == 0, X|~X == -1.
1238 for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
1239 // First, check for X and ~X in the operand list.
1240 assert(i < Ops.size())(static_cast <bool> (i < Ops.size()) ? void (0) : __assert_fail
("i < Ops.size()", "llvm/lib/Transforms/Scalar/Reassociate.cpp"
, 1240, __extension__ __PRETTY_FUNCTION__))
;
1241 Value *X;
1242 if (match(Ops[i].Op, m_Not(m_Value(X)))) { // Cannot occur for ^.
1243 unsigned FoundX = FindInOperandList(Ops, i, X);
1244 if (FoundX != i) {
1245 if (Opcode == Instruction::And) // ...&X&~X = 0
1246 return Constant::getNullValue(X->getType());
1247
1248 if (Opcode == Instruction::Or) // ...|X|~X = -1
1249 return Constant::getAllOnesValue(X->getType());
1250 }
1251 }
1252
1253 // Next, check for duplicate pairs of values, which we assume are next to
1254 // each other, due to our sorting criteria.
1255 assert(i < Ops.size())(static_cast <bool> (i < Ops.size()) ? void (0) : __assert_fail
("i < Ops.size()", "llvm/lib/Transforms/Scalar/Reassociate.cpp"
, 1255, __extension__ __PRETTY_FUNCTION__))
;
1256 if (i+1 != Ops.size() && Ops[i+1].Op == Ops[i].Op) {
1257 if (Opcode == Instruction::And || Opcode == Instruction::Or) {
1258 // Drop duplicate values for And and Or.
1259 Ops.erase(Ops.begin()+i);
1260 --i; --e;
1261 ++NumAnnihil;
1262 continue;
1263 }
1264
1265 // Drop pairs of values for Xor.
1266 assert(Opcode == Instruction::Xor)(static_cast <bool> (Opcode == Instruction::Xor) ? void
(0) : __assert_fail ("Opcode == Instruction::Xor", "llvm/lib/Transforms/Scalar/Reassociate.cpp"
, 1266, __extension__ __PRETTY_FUNCTION__))
;
1267 if (e == 2)
1268 return Constant::getNullValue(Ops[0].Op->getType());
1269
1270 // Y ^ X^X -> Y
1271 Ops.erase(Ops.begin()+i, Ops.begin()+i+2);
1272 i -= 1; e -= 2;
1273 ++NumAnnihil;
1274 }
1275 }
1276 return nullptr;
1277}
1278
1279/// Helper function of CombineXorOpnd(). It creates a bitwise-and
1280/// instruction with the given two operands, and return the resulting
1281/// instruction. There are two special cases: 1) if the constant operand is 0,
1282/// it will return NULL. 2) if the constant is ~0, the symbolic operand will
1283/// be returned.
1284static Value *createAndInstr(Instruction *InsertBefore, Value *Opnd,
1285 const APInt &ConstOpnd) {
1286 if (ConstOpnd.isZero())
1287 return nullptr;
1288
1289 if (ConstOpnd.isAllOnes())
1290 return Opnd;
1291
1292 Instruction *I = BinaryOperator::CreateAnd(
1293 Opnd, ConstantInt::get(Opnd->getType(), ConstOpnd), "and.ra",
1294 InsertBefore);
1295 I->setDebugLoc(InsertBefore->getDebugLoc());
1296 return I;
1297}
1298
1299// Helper function of OptimizeXor(). It tries to simplify "Opnd1 ^ ConstOpnd"
1300// into "R ^ C", where C would be 0, and R is a symbolic value.
1301//
1302// If it was successful, true is returned, and the "R" and "C" is returned
1303// via "Res" and "ConstOpnd", respectively; otherwise, false is returned,
1304// and both "Res" and "ConstOpnd" remain unchanged.
1305bool ReassociatePass::CombineXorOpnd(Instruction *I, XorOpnd *Opnd1,
1306 APInt &ConstOpnd, Value *&Res) {
1307 // Xor-Rule 1: (x | c1) ^ c2 = (x | c1) ^ (c1 ^ c1) ^ c2
1308 // = ((x | c1) ^ c1) ^ (c1 ^ c2)
1309 // = (x & ~c1) ^ (c1 ^ c2)
1310 // It is useful only when c1 == c2.
1311 if (!Opnd1->isOrExpr() || Opnd1->getConstPart().isZero())
1312 return false;
1313
1314 if (!Opnd1->getValue()->hasOneUse())
1315 return false;
1316
1317 const APInt &C1 = Opnd1->getConstPart();
1318 if (C1 != ConstOpnd)
1319 return false;
1320
1321 Value *X = Opnd1->getSymbolicPart();
1322 Res = createAndInstr(I, X, ~C1);
1323 // ConstOpnd was C2, now C1 ^ C2.
1324 ConstOpnd ^= C1;
1325
1326 if (Instruction *T = dyn_cast<Instruction>(Opnd1->getValue()))
1327 RedoInsts.insert(T);
1328 return true;
1329}
1330
1331// Helper function of OptimizeXor(). It tries to simplify
1332// "Opnd1 ^ Opnd2 ^ ConstOpnd" into "R ^ C", where C would be 0, and R is a
1333// symbolic value.
1334//
1335// If it was successful, true is returned, and the "R" and "C" is returned
1336// via "Res" and "ConstOpnd", respectively (If the entire expression is
1337// evaluated to a constant, the Res is set to NULL); otherwise, false is
1338// returned, and both "Res" and "ConstOpnd" remain unchanged.
1339bool ReassociatePass::CombineXorOpnd(Instruction *I, XorOpnd *Opnd1,
1340 XorOpnd *Opnd2, APInt &ConstOpnd,
1341 Value *&Res) {
1342 Value *X = Opnd1->getSymbolicPart();
1343 if (X != Opnd2->getSymbolicPart())
1344 return false;
1345
1346 // This many instruction become dead.(At least "Opnd1 ^ Opnd2" will die.)
1347 int DeadInstNum = 1;
1348 if (Opnd1->getValue()->hasOneUse())
1349 DeadInstNum++;
1350 if (Opnd2->getValue()->hasOneUse())
1351 DeadInstNum++;
1352
1353 // Xor-Rule 2:
1354 // (x | c1) ^ (x & c2)
1355 // = (x|c1) ^ (x&c2) ^ (c1 ^ c1) = ((x|c1) ^ c1) ^ (x & c2) ^ c1
1356 // = (x & ~c1) ^ (x & c2) ^ c1 // Xor-Rule 1
1357 // = (x & c3) ^ c1, where c3 = ~c1 ^ c2 // Xor-rule 3
1358 //
1359 if (Opnd1->isOrExpr() != Opnd2->isOrExpr()) {
1360 if (Opnd2->isOrExpr())
1361 std::swap(Opnd1, Opnd2);
1362
1363 const APInt &C1 = Opnd1->getConstPart();
1364 const APInt &C2 = Opnd2->getConstPart();
1365 APInt C3((~C1) ^ C2);
1366
1367 // Do not increase code size!
1368 if (!C3.isZero() && !C3.isAllOnes()) {
1369 int NewInstNum = ConstOpnd.getBoolValue() ? 1 : 2;
1370 if (NewInstNum > DeadInstNum)
1371 return false;
1372 }
1373
1374 Res = createAndInstr(I, X, C3);
1375 ConstOpnd ^= C1;
1376 } else if (Opnd1->isOrExpr()) {
1377 // Xor-Rule 3: (x | c1) ^ (x | c2) = (x & c3) ^ c3 where c3 = c1 ^ c2
1378 //
1379 const APInt &C1 = Opnd1->getConstPart();
1380 const APInt &C2 = Opnd2->getConstPart();
1381 APInt C3 = C1 ^ C2;
1382
1383 // Do not increase code size
1384 if (!C3.isZero() && !C3.isAllOnes()) {
1385 int NewInstNum = ConstOpnd.getBoolValue() ? 1 : 2;
1386 if (NewInstNum > DeadInstNum)
1387 return false;
1388 }
1389
1390 Res = createAndInstr(I, X, C3);
1391 ConstOpnd ^= C3;
1392 } else {
1393 // Xor-Rule 4: (x & c1) ^ (x & c2) = (x & (c1^c2))
1394 //
1395 const APInt &C1 = Opnd1->getConstPart();
1396 const APInt &C2 = Opnd2->getConstPart();
1397 APInt C3 = C1 ^ C2;
1398 Res = createAndInstr(I, X, C3);
1399 }
1400
1401 // Put the original operands in the Redo list; hope they will be deleted
1402 // as dead code.
1403 if (Instruction *T = dyn_cast<Instruction>(Opnd1->getValue()))
1404 RedoInsts.insert(T);
1405 if (Instruction *T = dyn_cast<Instruction>(Opnd2->getValue()))
1406 RedoInsts.insert(T);
1407
1408 return true;
1409}
1410
1411/// Optimize a series of operands to an 'xor' instruction. If it can be reduced
1412/// to a single Value, it is returned, otherwise the Ops list is mutated as
1413/// necessary.
1414Value *ReassociatePass::OptimizeXor(Instruction *I,
1415 SmallVectorImpl<ValueEntry> &Ops) {
1416 if (Value *V = OptimizeAndOrXor(Instruction::Xor, Ops))
1417 return V;
1418
1419 if (Ops.size() == 1)
1420 return nullptr;
1421
1422 SmallVector<XorOpnd, 8> Opnds;
1423 SmallVector<XorOpnd*, 8> OpndPtrs;
1424 Type *Ty = Ops[0].Op->getType();
1425 APInt ConstOpnd(Ty->getScalarSizeInBits(), 0);
1426
1427 // Step 1: Convert ValueEntry to XorOpnd
1428 for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
1429 Value *V = Ops[i].Op;
1430 const APInt *C;
1431 // TODO: Support non-splat vectors.
1432 if (match(V, m_APInt(C))) {
1433 ConstOpnd ^= *C;
1434 } else {
1435 XorOpnd O(V);
1436 O.setSymbolicRank(getRank(O.getSymbolicPart()));
1437 Opnds.push_back(O);
1438 }
1439 }
1440
1441 // NOTE: From this point on, do *NOT* add/delete element to/from "Opnds".
1442 // It would otherwise invalidate the "Opnds"'s iterator, and hence invalidate
1443 // the "OpndPtrs" as well. For the similar reason, do not fuse this loop
1444 // with the previous loop --- the iterator of the "Opnds" may be invalidated
1445 // when new elements are added to the vector.
1446 for (unsigned i = 0, e = Opnds.size(); i != e; ++i)
1447 OpndPtrs.push_back(&Opnds[i]);
1448
1449 // Step 2: Sort the Xor-Operands in a way such that the operands containing
1450 // the same symbolic value cluster together. For instance, the input operand
1451 // sequence ("x | 123", "y & 456", "x & 789") will be sorted into:
1452 // ("x | 123", "x & 789", "y & 456").
1453 //
1454 // The purpose is twofold:
1455 // 1) Cluster together the operands sharing the same symbolic-value.
1456 // 2) Operand having smaller symbolic-value-rank is permuted earlier, which
1457 // could potentially shorten crital path, and expose more loop-invariants.
1458 // Note that values' rank are basically defined in RPO order (FIXME).
1459 // So, if Rank(X) < Rank(Y) < Rank(Z), it means X is defined earlier
1460 // than Y which is defined earlier than Z. Permute "x | 1", "Y & 2",
1461 // "z" in the order of X-Y-Z is better than any other orders.
1462 llvm::stable_sort(OpndPtrs, [](XorOpnd *LHS, XorOpnd *RHS) {
1463 return LHS->getSymbolicRank() < RHS->getSymbolicRank();
1464 });
1465
1466 // Step 3: Combine adjacent operands
1467 XorOpnd *PrevOpnd = nullptr;
1468 bool Changed = false;
1469 for (unsigned i = 0, e = Opnds.size(); i < e; i++) {
1470 XorOpnd *CurrOpnd = OpndPtrs[i];
1471 // The combined value
1472 Value *CV;
1473
1474 // Step 3.1: Try simplifying "CurrOpnd ^ ConstOpnd"
1475 if (!ConstOpnd.isZero() && CombineXorOpnd(I, CurrOpnd, ConstOpnd, CV)) {
1476 Changed = true;
1477 if (CV)
1478 *CurrOpnd = XorOpnd(CV);
1479 else {
1480 CurrOpnd->Invalidate();
1481 continue;
1482 }
1483 }
1484
1485 if (!PrevOpnd || CurrOpnd->getSymbolicPart() != PrevOpnd->getSymbolicPart()) {
1486 PrevOpnd = CurrOpnd;
1487 continue;
1488 }
1489
1490 // step 3.2: When previous and current operands share the same symbolic
1491 // value, try to simplify "PrevOpnd ^ CurrOpnd ^ ConstOpnd"
1492 if (CombineXorOpnd(I, CurrOpnd, PrevOpnd, ConstOpnd, CV)) {
1493 // Remove previous operand
1494 PrevOpnd->Invalidate();
1495 if (CV) {
1496 *CurrOpnd = XorOpnd(CV);
1497 PrevOpnd = CurrOpnd;
1498 } else {
1499 CurrOpnd->Invalidate();
1500 PrevOpnd = nullptr;
1501 }
1502 Changed = true;
1503 }
1504 }
1505
1506 // Step 4: Reassemble the Ops
1507 if (Changed) {
1508 Ops.clear();
1509 for (const XorOpnd &O : Opnds) {
1510 if (O.isInvalid())
1511 continue;
1512 ValueEntry VE(getRank(O.getValue()), O.getValue());
1513 Ops.push_back(VE);
1514 }
1515 if (!ConstOpnd.isZero()) {
1516 Value *C = ConstantInt::get(Ty, ConstOpnd);
1517 ValueEntry VE(getRank(C), C);
1518 Ops.push_back(VE);
1519 }
1520 unsigned Sz = Ops.size();
1521 if (Sz == 1)
1522 return Ops.back().Op;
1523 if (Sz == 0) {
1524 assert(ConstOpnd.isZero())(static_cast <bool> (ConstOpnd.isZero()) ? void (0) : __assert_fail
("ConstOpnd.isZero()", "llvm/lib/Transforms/Scalar/Reassociate.cpp"
, 1524, __extension__ __PRETTY_FUNCTION__))
;
1525 return ConstantInt::get(Ty, ConstOpnd);
1526 }
1527 }
1528
1529 return nullptr;
1530}
1531
1532/// Optimize a series of operands to an 'add' instruction. This
1533/// optimizes based on identities. If it can be reduced to a single Value, it
1534/// is returned, otherwise the Ops list is mutated as necessary.
1535Value *ReassociatePass::OptimizeAdd(Instruction *I,
1536 SmallVectorImpl<ValueEntry> &Ops) {
1537 // Scan the operand lists looking for X and -X pairs. If we find any, we
1538 // can simplify expressions like X+-X == 0 and X+~X ==-1. While we're at it,
1539 // scan for any
1540 // duplicates. We want to canonicalize Y+Y+Y+Z -> 3*Y+Z.
1541
1542 for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
1543 Value *TheOp = Ops[i].Op;
1544 // Check to see if we've seen this operand before. If so, we factor all
1545 // instances of the operand together. Due to our sorting criteria, we know
1546 // that these need to be next to each other in the vector.
1547 if (i+1 != Ops.size() && Ops[i+1].Op == TheOp) {
1548 // Rescan the list, remove all instances of this operand from the expr.
1549 unsigned NumFound = 0;
1550 do {
1551 Ops.erase(Ops.begin()+i);
1552 ++NumFound;
1553 } while (i != Ops.size() && Ops[i].Op == TheOp);
1554
1555 LLVM_DEBUG(dbgs() << "\nFACTORING [" << NumFound << "]: " << *TheOpdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "\nFACTORING [" << NumFound
<< "]: " << *TheOp << '\n'; } } while (false
)
1556 << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "\nFACTORING [" << NumFound
<< "]: " << *TheOp << '\n'; } } while (false
)
;
1557 ++NumFactor;
1558
1559 // Insert a new multiply.
1560 Type *Ty = TheOp->getType();
1561 Constant *C = Ty->isIntOrIntVectorTy() ?
1562 ConstantInt::get(Ty, NumFound) : ConstantFP::get(Ty, NumFound);
1563 Instruction *Mul = CreateMul(TheOp, C, "factor", I, I);
1564
1565 // Now that we have inserted a multiply, optimize it. This allows us to
1566 // handle cases that require multiple factoring steps, such as this:
1567 // (X*2) + (X*2) + (X*2) -> (X*2)*3 -> X*6
1568 RedoInsts.insert(Mul);
1569
1570 // If every add operand was a duplicate, return the multiply.
1571 if (Ops.empty())
1572 return Mul;
1573
1574 // Otherwise, we had some input that didn't have the dupe, such as
1575 // "A + A + B" -> "A*2 + B". Add the new multiply to the list of
1576 // things being added by this operation.
1577 Ops.insert(Ops.begin(), ValueEntry(getRank(Mul), Mul));
1578
1579 --i;
1580 e = Ops.size();
1581 continue;
1582 }
1583
1584 // Check for X and -X or X and ~X in the operand list.
1585 Value *X;
1586 if (!match(TheOp, m_Neg(m_Value(X))) && !match(TheOp, m_Not(m_Value(X))) &&
1587 !match(TheOp, m_FNeg(m_Value(X))))
1588 continue;
1589
1590 unsigned FoundX = FindInOperandList(Ops, i, X);
1591 if (FoundX == i)
1592 continue;
1593
1594 // Remove X and -X from the operand list.
1595 if (Ops.size() == 2 &&
1596 (match(TheOp, m_Neg(m_Value())) || match(TheOp, m_FNeg(m_Value()))))
1597 return Constant::getNullValue(X->getType());
1598
1599 // Remove X and ~X from the operand list.
1600 if (Ops.size() == 2 && match(TheOp, m_Not(m_Value())))
1601 return Constant::getAllOnesValue(X->getType());
1602
1603 Ops.erase(Ops.begin()+i);
1604 if (i < FoundX)
1605 --FoundX;
1606 else
1607 --i; // Need to back up an extra one.
1608 Ops.erase(Ops.begin()+FoundX);
1609 ++NumAnnihil;
1610 --i; // Revisit element.
1611 e -= 2; // Removed two elements.
1612
1613 // if X and ~X we append -1 to the operand list.
1614 if (match(TheOp, m_Not(m_Value()))) {
1615 Value *V = Constant::getAllOnesValue(X->getType());
1616 Ops.insert(Ops.end(), ValueEntry(getRank(V), V));
1617 e += 1;
1618 }
1619 }
1620
1621 // Scan the operand list, checking to see if there are any common factors
1622 // between operands. Consider something like A*A+A*B*C+D. We would like to
1623 // reassociate this to A*(A+B*C)+D, which reduces the number of multiplies.
1624 // To efficiently find this, we count the number of times a factor occurs
1625 // for any ADD operands that are MULs.
1626 DenseMap<Value*, unsigned> FactorOccurrences;
1627
1628 // Keep track of each multiply we see, to avoid triggering on (X*4)+(X*4)
1629 // where they are actually the same multiply.
1630 unsigned MaxOcc = 0;
1631 Value *MaxOccVal = nullptr;
1632 for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
1633 BinaryOperator *BOp =
1634 isReassociableOp(Ops[i].Op, Instruction::Mul, Instruction::FMul);
1635 if (!BOp)
1636 continue;
1637
1638 // Compute all of the factors of this added value.
1639 SmallVector<Value*, 8> Factors;
1640 FindSingleUseMultiplyFactors(BOp, Factors);
1641 assert(Factors.size() > 1 && "Bad linearize!")(static_cast <bool> (Factors.size() > 1 && "Bad linearize!"
) ? void (0) : __assert_fail ("Factors.size() > 1 && \"Bad linearize!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 1641, __extension__
__PRETTY_FUNCTION__))
;
1642
1643 // Add one to FactorOccurrences for each unique factor in this op.
1644 SmallPtrSet<Value*, 8> Duplicates;
1645 for (Value *Factor : Factors) {
1646 if (!Duplicates.insert(Factor).second)
1647 continue;
1648
1649 unsigned Occ = ++FactorOccurrences[Factor];
1650 if (Occ > MaxOcc) {
1651 MaxOcc = Occ;
1652 MaxOccVal = Factor;
1653 }
1654
1655 // If Factor is a negative constant, add the negated value as a factor
1656 // because we can percolate the negate out. Watch for minint, which
1657 // cannot be positivified.
1658 if (ConstantInt *CI = dyn_cast<ConstantInt>(Factor)) {
1659 if (CI->isNegative() && !CI->isMinValue(true)) {
1660 Factor = ConstantInt::get(CI->getContext(), -CI->getValue());
1661 if (!Duplicates.insert(Factor).second)
1662 continue;
1663 unsigned Occ = ++FactorOccurrences[Factor];
1664 if (Occ > MaxOcc) {
1665 MaxOcc = Occ;
1666 MaxOccVal = Factor;
1667 }
1668 }
1669 } else if (ConstantFP *CF = dyn_cast<ConstantFP>(Factor)) {
1670 if (CF->isNegative()) {
1671 APFloat F(CF->getValueAPF());
1672 F.changeSign();
1673 Factor = ConstantFP::get(CF->getContext(), F);
1674 if (!Duplicates.insert(Factor).second)
1675 continue;
1676 unsigned Occ = ++FactorOccurrences[Factor];
1677 if (Occ > MaxOcc) {
1678 MaxOcc = Occ;
1679 MaxOccVal = Factor;
1680 }
1681 }
1682 }
1683 }
1684 }
1685
1686 // If any factor occurred more than one time, we can pull it out.
1687 if (MaxOcc > 1) {
1688 LLVM_DEBUG(dbgs() << "\nFACTORING [" << MaxOcc << "]: " << *MaxOccValdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "\nFACTORING [" << MaxOcc
<< "]: " << *MaxOccVal << '\n'; } } while (
false)
1689 << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "\nFACTORING [" << MaxOcc
<< "]: " << *MaxOccVal << '\n'; } } while (
false)
;
1690 ++NumFactor;
1691
1692 // Create a new instruction that uses the MaxOccVal twice. If we don't do
1693 // this, we could otherwise run into situations where removing a factor
1694 // from an expression will drop a use of maxocc, and this can cause
1695 // RemoveFactorFromExpression on successive values to behave differently.
1696 Instruction *DummyInst =
1697 I->getType()->isIntOrIntVectorTy()
1698 ? BinaryOperator::CreateAdd(MaxOccVal, MaxOccVal)
1699 : BinaryOperator::CreateFAdd(MaxOccVal, MaxOccVal);
1700
1701 SmallVector<WeakTrackingVH, 4> NewMulOps;
1702 for (unsigned i = 0; i != Ops.size(); ++i) {
1703 // Only try to remove factors from expressions we're allowed to.
1704 BinaryOperator *BOp =
1705 isReassociableOp(Ops[i].Op, Instruction::Mul, Instruction::FMul);
1706 if (!BOp)
1707 continue;
1708
1709 if (Value *V = RemoveFactorFromExpression(Ops[i].Op, MaxOccVal)) {
1710 // The factorized operand may occur several times. Convert them all in
1711 // one fell swoop.
1712 for (unsigned j = Ops.size(); j != i;) {
1713 --j;
1714 if (Ops[j].Op == Ops[i].Op) {
1715 NewMulOps.push_back(V);
1716 Ops.erase(Ops.begin()+j);
1717 }
1718 }
1719 --i;
1720 }
1721 }
1722
1723 // No need for extra uses anymore.
1724 DummyInst->deleteValue();
1725
1726 unsigned NumAddedValues = NewMulOps.size();
1727 Value *V = EmitAddTreeOfValues(I, NewMulOps);
1728
1729 // Now that we have inserted the add tree, optimize it. This allows us to
1730 // handle cases that require multiple factoring steps, such as this:
1731 // A*A*B + A*A*C --> A*(A*B+A*C) --> A*(A*(B+C))
1732 assert(NumAddedValues > 1 && "Each occurrence should contribute a value")(static_cast <bool> (NumAddedValues > 1 && "Each occurrence should contribute a value"
) ? void (0) : __assert_fail ("NumAddedValues > 1 && \"Each occurrence should contribute a value\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 1732, __extension__
__PRETTY_FUNCTION__))
;
1733 (void)NumAddedValues;
1734 if (Instruction *VI = dyn_cast<Instruction>(V))
1735 RedoInsts.insert(VI);
1736
1737 // Create the multiply.
1738 Instruction *V2 = CreateMul(V, MaxOccVal, "reass.mul", I, I);
1739
1740 // Rerun associate on the multiply in case the inner expression turned into
1741 // a multiply. We want to make sure that we keep things in canonical form.
1742 RedoInsts.insert(V2);
1743
1744 // If every add operand included the factor (e.g. "A*B + A*C"), then the
1745 // entire result expression is just the multiply "A*(B+C)".
1746 if (Ops.empty())
1747 return V2;
1748
1749 // Otherwise, we had some input that didn't have the factor, such as
1750 // "A*B + A*C + D" -> "A*(B+C) + D". Add the new multiply to the list of
1751 // things being added by this operation.
1752 Ops.insert(Ops.begin(), ValueEntry(getRank(V2), V2));
1753 }
1754
1755 return nullptr;
1756}
1757
1758/// Build up a vector of value/power pairs factoring a product.
1759///
1760/// Given a series of multiplication operands, build a vector of factors and
1761/// the powers each is raised to when forming the final product. Sort them in
1762/// the order of descending power.
1763///
1764/// (x*x) -> [(x, 2)]
1765/// ((x*x)*x) -> [(x, 3)]
1766/// ((((x*y)*x)*y)*x) -> [(x, 3), (y, 2)]
1767///
1768/// \returns Whether any factors have a power greater than one.
1769static bool collectMultiplyFactors(SmallVectorImpl<ValueEntry> &Ops,
1770 SmallVectorImpl<Factor> &Factors) {
1771 // FIXME: Have Ops be (ValueEntry, Multiplicity) pairs, simplifying this.
1772 // Compute the sum of powers of simplifiable factors.
1773 unsigned FactorPowerSum = 0;
1774 for (unsigned Idx = 1, Size = Ops.size(); Idx < Size; ++Idx) {
1775 Value *Op = Ops[Idx-1].Op;
1776
1777 // Count the number of occurrences of this value.
1778 unsigned Count = 1;
1779 for (; Idx < Size && Ops[Idx].Op == Op; ++Idx)
1780 ++Count;
1781 // Track for simplification all factors which occur 2 or more times.
1782 if (Count > 1)
1783 FactorPowerSum += Count;
1784 }
1785
1786 // We can only simplify factors if the sum of the powers of our simplifiable
1787 // factors is 4 or higher. When that is the case, we will *always* have
1788 // a simplification. This is an important invariant to prevent cyclicly
1789 // trying to simplify already minimal formations.
1790 if (FactorPowerSum < 4)
1791 return false;
1792
1793 // Now gather the simplifiable factors, removing them from Ops.
1794 FactorPowerSum = 0;
1795 for (unsigned Idx = 1; Idx < Ops.size(); ++Idx) {
1796 Value *Op = Ops[Idx-1].Op;
1797
1798 // Count the number of occurrences of this value.
1799 unsigned Count = 1;
1800 for (; Idx < Ops.size() && Ops[Idx].Op == Op; ++Idx)
1801 ++Count;
1802 if (Count == 1)
1803 continue;
1804 // Move an even number of occurrences to Factors.
1805 Count &= ~1U;
1806 Idx -= Count;
1807 FactorPowerSum += Count;
1808 Factors.push_back(Factor(Op, Count));
1809 Ops.erase(Ops.begin()+Idx, Ops.begin()+Idx+Count);
1810 }
1811
1812 // None of the adjustments above should have reduced the sum of factor powers
1813 // below our mininum of '4'.
1814 assert(FactorPowerSum >= 4)(static_cast <bool> (FactorPowerSum >= 4) ? void (0)
: __assert_fail ("FactorPowerSum >= 4", "llvm/lib/Transforms/Scalar/Reassociate.cpp"
, 1814, __extension__ __PRETTY_FUNCTION__))
;
1815
1816 llvm::stable_sort(Factors, [](const Factor &LHS, const Factor &RHS) {
1817 return LHS.Power > RHS.Power;
1818 });
1819 return true;
1820}
1821
1822/// Build a tree of multiplies, computing the product of Ops.
1823static Value *buildMultiplyTree(IRBuilderBase &Builder,
1824 SmallVectorImpl<Value*> &Ops) {
1825 if (Ops.size() == 1)
1826 return Ops.back();
1827
1828 Value *LHS = Ops.pop_back_val();
1829 do {
1830 if (LHS->getType()->isIntOrIntVectorTy())
1831 LHS = Builder.CreateMul(LHS, Ops.pop_back_val());
1832 else
1833 LHS = Builder.CreateFMul(LHS, Ops.pop_back_val());
1834 } while (!Ops.empty());
1835
1836 return LHS;
1837}
1838
1839/// Build a minimal multiplication DAG for (a^x)*(b^y)*(c^z)*...
1840///
1841/// Given a vector of values raised to various powers, where no two values are
1842/// equal and the powers are sorted in decreasing order, compute the minimal
1843/// DAG of multiplies to compute the final product, and return that product
1844/// value.
1845Value *
1846ReassociatePass::buildMinimalMultiplyDAG(IRBuilderBase &Builder,
1847 SmallVectorImpl<Factor> &Factors) {
1848 assert(Factors[0].Power)(static_cast <bool> (Factors[0].Power) ? void (0) : __assert_fail
("Factors[0].Power", "llvm/lib/Transforms/Scalar/Reassociate.cpp"
, 1848, __extension__ __PRETTY_FUNCTION__))
;
1849 SmallVector<Value *, 4> OuterProduct;
1850 for (unsigned LastIdx = 0, Idx = 1, Size = Factors.size();
1851 Idx < Size && Factors[Idx].Power > 0; ++Idx) {
1852 if (Factors[Idx].Power != Factors[LastIdx].Power) {
1853 LastIdx = Idx;
1854 continue;
1855 }
1856
1857 // We want to multiply across all the factors with the same power so that
1858 // we can raise them to that power as a single entity. Build a mini tree
1859 // for that.
1860 SmallVector<Value *, 4> InnerProduct;
1861 InnerProduct.push_back(Factors[LastIdx].Base);
1862 do {
1863 InnerProduct.push_back(Factors[Idx].Base);
1864 ++Idx;
1865 } while (Idx < Size && Factors[Idx].Power == Factors[LastIdx].Power);
1866
1867 // Reset the base value of the first factor to the new expression tree.
1868 // We'll remove all the factors with the same power in a second pass.
1869 Value *M = Factors[LastIdx].Base = buildMultiplyTree(Builder, InnerProduct);
1870 if (Instruction *MI = dyn_cast<Instruction>(M))
1871 RedoInsts.insert(MI);
1872
1873 LastIdx = Idx;
1874 }
1875 // Unique factors with equal powers -- we've folded them into the first one's
1876 // base.
1877 Factors.erase(std::unique(Factors.begin(), Factors.end(),
1878 [](const Factor &LHS, const Factor &RHS) {
1879 return LHS.Power == RHS.Power;
1880 }),
1881 Factors.end());
1882
1883 // Iteratively collect the base of each factor with an add power into the
1884 // outer product, and halve each power in preparation for squaring the
1885 // expression.
1886 for (Factor &F : Factors) {
1887 if (F.Power & 1)
1888 OuterProduct.push_back(F.Base);
1889 F.Power >>= 1;
1890 }
1891 if (Factors[0].Power) {
1892 Value *SquareRoot = buildMinimalMultiplyDAG(Builder, Factors);
1893 OuterProduct.push_back(SquareRoot);
1894 OuterProduct.push_back(SquareRoot);
1895 }
1896 if (OuterProduct.size() == 1)
1897 return OuterProduct.front();
1898
1899 Value *V = buildMultiplyTree(Builder, OuterProduct);
1900 return V;
1901}
1902
1903Value *ReassociatePass::OptimizeMul(BinaryOperator *I,
1904 SmallVectorImpl<ValueEntry> &Ops) {
1905 // We can only optimize the multiplies when there is a chain of more than
1906 // three, such that a balanced tree might require fewer total multiplies.
1907 if (Ops.size() < 4)
1908 return nullptr;
1909
1910 // Try to turn linear trees of multiplies without other uses of the
1911 // intermediate stages into minimal multiply DAGs with perfect sub-expression
1912 // re-use.
1913 SmallVector<Factor, 4> Factors;
1914 if (!collectMultiplyFactors(Ops, Factors))
1915 return nullptr; // All distinct factors, so nothing left for us to do.
1916
1917 IRBuilder<> Builder(I);
1918 // The reassociate transformation for FP operations is performed only
1919 // if unsafe algebra is permitted by FastMathFlags. Propagate those flags
1920 // to the newly generated operations.
1921 if (auto FPI = dyn_cast<FPMathOperator>(I))
1922 Builder.setFastMathFlags(FPI->getFastMathFlags());
1923
1924 Value *V = buildMinimalMultiplyDAG(Builder, Factors);
1925 if (Ops.empty())
1926 return V;
1927
1928 ValueEntry NewEntry = ValueEntry(getRank(V), V);
1929 Ops.insert(llvm::lower_bound(Ops, NewEntry), NewEntry);
1930 return nullptr;
1931}
1932
1933Value *ReassociatePass::OptimizeExpression(BinaryOperator *I,
1934 SmallVectorImpl<ValueEntry> &Ops) {
1935 // Now that we have the linearized expression tree, try to optimize it.
1936 // Start by folding any constants that we found.
1937 const DataLayout &DL = I->getModule()->getDataLayout();
1938 Constant *Cst = nullptr;
1939 unsigned Opcode = I->getOpcode();
1940 while (!Ops.empty()) {
1941 if (auto *C = dyn_cast<Constant>(Ops.back().Op)) {
1942 if (!Cst) {
1943 Ops.pop_back();
1944 Cst = C;
1945 continue;
1946 }
1947 if (Constant *Res = ConstantFoldBinaryOpOperands(Opcode, C, Cst, DL)) {
1948 Ops.pop_back();
1949 Cst = Res;
1950 continue;
1951 }
1952 }
1953 break;
1954 }
1955 // If there was nothing but constants then we are done.
1956 if (Ops.empty())
1957 return Cst;
1958
1959 // Put the combined constant back at the end of the operand list, except if
1960 // there is no point. For example, an add of 0 gets dropped here, while a
1961 // multiplication by zero turns the whole expression into zero.
1962 if (Cst && Cst != ConstantExpr::getBinOpIdentity(Opcode, I->getType())) {
1963 if (Cst == ConstantExpr::getBinOpAbsorber(Opcode, I->getType()))
1964 return Cst;
1965 Ops.push_back(ValueEntry(0, Cst));
1966 }
1967
1968 if (Ops.size() == 1) return Ops[0].Op;
1969
1970 // Handle destructive annihilation due to identities between elements in the
1971 // argument list here.
1972 unsigned NumOps = Ops.size();
1973 switch (Opcode) {
1974 default: break;
1975 case Instruction::And:
1976 case Instruction::Or:
1977 if (Value *Result = OptimizeAndOrXor(Opcode, Ops))
1978 return Result;
1979 break;
1980
1981 case Instruction::Xor:
1982 if (Value *Result = OptimizeXor(I, Ops))
1983 return Result;
1984 break;
1985
1986 case Instruction::Add:
1987 case Instruction::FAdd:
1988 if (Value *Result = OptimizeAdd(I, Ops))
1989 return Result;
1990 break;
1991
1992 case Instruction::Mul:
1993 case Instruction::FMul:
1994 if (Value *Result = OptimizeMul(I, Ops))
1995 return Result;
1996 break;
1997 }
1998
1999 if (Ops.size() != NumOps)
2000 return OptimizeExpression(I, Ops);
2001 return nullptr;
2002}
2003
2004// Remove dead instructions and if any operands are trivially dead add them to
2005// Insts so they will be removed as well.
2006void ReassociatePass::RecursivelyEraseDeadInsts(Instruction *I,
2007 OrderedSet &Insts) {
2008 assert(isInstructionTriviallyDead(I) && "Trivially dead instructions only!")(static_cast <bool> (isInstructionTriviallyDead(I) &&
"Trivially dead instructions only!") ? void (0) : __assert_fail
("isInstructionTriviallyDead(I) && \"Trivially dead instructions only!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2008, __extension__
__PRETTY_FUNCTION__))
;
2009 SmallVector<Value *, 4> Ops(I->operands());
2010 ValueRankMap.erase(I);
2011 Insts.remove(I);
2012 RedoInsts.remove(I);
2013 llvm::salvageDebugInfo(*I);
2014 I->eraseFromParent();
2015 for (auto *Op : Ops)
2016 if (Instruction *OpInst = dyn_cast<Instruction>(Op))
2017 if (OpInst->use_empty())
2018 Insts.insert(OpInst);
2019}
2020
2021/// Zap the given instruction, adding interesting operands to the work list.
2022void ReassociatePass::EraseInst(Instruction *I) {
2023 assert(isInstructionTriviallyDead(I) && "Trivially dead instructions only!")(static_cast <bool> (isInstructionTriviallyDead(I) &&
"Trivially dead instructions only!") ? void (0) : __assert_fail
("isInstructionTriviallyDead(I) && \"Trivially dead instructions only!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2023, __extension__
__PRETTY_FUNCTION__))
;
2024 LLVM_DEBUG(dbgs() << "Erasing dead inst: "; I->dump())do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "Erasing dead inst: "; I->
dump(); } } while (false)
;
2025
2026 SmallVector<Value *, 8> Ops(I->operands());
2027 // Erase the dead instruction.
2028 ValueRankMap.erase(I);
2029 RedoInsts.remove(I);
2030 llvm::salvageDebugInfo(*I);
2031 I->eraseFromParent();
2032 // Optimize its operands.
2033 SmallPtrSet<Instruction *, 8> Visited; // Detect self-referential nodes.
2034 for (unsigned i = 0, e = Ops.size(); i != e; ++i)
2035 if (Instruction *Op = dyn_cast<Instruction>(Ops[i])) {
2036 // If this is a node in an expression tree, climb to the expression root
2037 // and add that since that's where optimization actually happens.
2038 unsigned Opcode = Op->getOpcode();
2039 while (Op->hasOneUse() && Op->user_back()->getOpcode() == Opcode &&
2040 Visited.insert(Op).second)
2041 Op = Op->user_back();
2042
2043 // The instruction we're going to push may be coming from a
2044 // dead block, and Reassociate skips the processing of unreachable
2045 // blocks because it's a waste of time and also because it can
2046 // lead to infinite loop due to LLVM's non-standard definition
2047 // of dominance.
2048 if (ValueRankMap.contains(Op))
2049 RedoInsts.insert(Op);
2050 }
2051
2052 MadeChange = true;
2053}
2054
2055/// Recursively analyze an expression to build a list of instructions that have
2056/// negative floating-point constant operands. The caller can then transform
2057/// the list to create positive constants for better reassociation and CSE.
2058static void getNegatibleInsts(Value *V,
2059 SmallVectorImpl<Instruction *> &Candidates) {
2060 // Handle only one-use instructions. Combining negations does not justify
2061 // replicating instructions.
2062 Instruction *I;
2063 if (!match(V, m_OneUse(m_Instruction(I))))
2064 return;
2065
2066 // Handle expressions of multiplications and divisions.
2067 // TODO: This could look through floating-point casts.
2068 const APFloat *C;
2069 switch (I->getOpcode()) {
2070 case Instruction::FMul:
2071 // Not expecting non-canonical code here. Bail out and wait.
2072 if (match(I->getOperand(0), m_Constant()))
2073 break;
2074
2075 if (match(I->getOperand(1), m_APFloat(C)) && C->isNegative()) {
2076 Candidates.push_back(I);
2077 LLVM_DEBUG(dbgs() << "FMul with negative constant: " << *I << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "FMul with negative constant: "
<< *I << '\n'; } } while (false)
;
2078 }
2079 getNegatibleInsts(I->getOperand(0), Candidates);
2080 getNegatibleInsts(I->getOperand(1), Candidates);
2081 break;
2082 case Instruction::FDiv:
2083 // Not expecting non-canonical code here. Bail out and wait.
2084 if (match(I->getOperand(0), m_Constant()) &&
2085 match(I->getOperand(1), m_Constant()))
2086 break;
2087
2088 if ((match(I->getOperand(0), m_APFloat(C)) && C->isNegative()) ||
2089 (match(I->getOperand(1), m_APFloat(C)) && C->isNegative())) {
2090 Candidates.push_back(I);
2091 LLVM_DEBUG(dbgs() << "FDiv with negative constant: " << *I << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "FDiv with negative constant: "
<< *I << '\n'; } } while (false)
;
2092 }
2093 getNegatibleInsts(I->getOperand(0), Candidates);
2094 getNegatibleInsts(I->getOperand(1), Candidates);
2095 break;
2096 default:
2097 break;
2098 }
2099}
2100
2101/// Given an fadd/fsub with an operand that is a one-use instruction
2102/// (the fadd/fsub), try to change negative floating-point constants into
2103/// positive constants to increase potential for reassociation and CSE.
2104Instruction *ReassociatePass::canonicalizeNegFPConstantsForOp(Instruction *I,
2105 Instruction *Op,
2106 Value *OtherOp) {
2107 assert((I->getOpcode() == Instruction::FAdd ||(static_cast <bool> ((I->getOpcode() == Instruction::
FAdd || I->getOpcode() == Instruction::FSub) && "Expected fadd/fsub"
) ? void (0) : __assert_fail ("(I->getOpcode() == Instruction::FAdd || I->getOpcode() == Instruction::FSub) && \"Expected fadd/fsub\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2108, __extension__
__PRETTY_FUNCTION__))
2108 I->getOpcode() == Instruction::FSub) && "Expected fadd/fsub")(static_cast <bool> ((I->getOpcode() == Instruction::
FAdd || I->getOpcode() == Instruction::FSub) && "Expected fadd/fsub"
) ? void (0) : __assert_fail ("(I->getOpcode() == Instruction::FAdd || I->getOpcode() == Instruction::FSub) && \"Expected fadd/fsub\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2108, __extension__
__PRETTY_FUNCTION__))
;
2109
2110 // Collect instructions with negative FP constants from the subtree that ends
2111 // in Op.
2112 SmallVector<Instruction *, 4> Candidates;
2113 getNegatibleInsts(Op, Candidates);
2114 if (Candidates.empty())
2115 return nullptr;
2116
2117 // Don't canonicalize x + (-Constant * y) -> x - (Constant * y), if the
2118 // resulting subtract will be broken up later. This can get us into an
2119 // infinite loop during reassociation.
2120 bool IsFSub = I->getOpcode() == Instruction::FSub;
2121 bool NeedsSubtract = !IsFSub && Candidates.size() % 2 == 1;
2122 if (NeedsSubtract && ShouldBreakUpSubtract(I))
2123 return nullptr;
2124
2125 for (Instruction *Negatible : Candidates) {
2126 const APFloat *C;
2127 if (match(Negatible->getOperand(0), m_APFloat(C))) {
2128 assert(!match(Negatible->getOperand(1), m_Constant()) &&(static_cast <bool> (!match(Negatible->getOperand(1)
, m_Constant()) && "Expecting only 1 constant operand"
) ? void (0) : __assert_fail ("!match(Negatible->getOperand(1), m_Constant()) && \"Expecting only 1 constant operand\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2129, __extension__
__PRETTY_FUNCTION__))
2129 "Expecting only 1 constant operand")(static_cast <bool> (!match(Negatible->getOperand(1)
, m_Constant()) && "Expecting only 1 constant operand"
) ? void (0) : __assert_fail ("!match(Negatible->getOperand(1), m_Constant()) && \"Expecting only 1 constant operand\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2129, __extension__
__PRETTY_FUNCTION__))
;
2130 assert(C->isNegative() && "Expected negative FP constant")(static_cast <bool> (C->isNegative() && "Expected negative FP constant"
) ? void (0) : __assert_fail ("C->isNegative() && \"Expected negative FP constant\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2130, __extension__
__PRETTY_FUNCTION__))
;
2131 Negatible->setOperand(0, ConstantFP::get(Negatible->getType(), abs(*C)));
2132 MadeChange = true;
2133 }
2134 if (match(Negatible->getOperand(1), m_APFloat(C))) {
2135 assert(!match(Negatible->getOperand(0), m_Constant()) &&(static_cast <bool> (!match(Negatible->getOperand(0)
, m_Constant()) && "Expecting only 1 constant operand"
) ? void (0) : __assert_fail ("!match(Negatible->getOperand(0), m_Constant()) && \"Expecting only 1 constant operand\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2136, __extension__
__PRETTY_FUNCTION__))
2136 "Expecting only 1 constant operand")(static_cast <bool> (!match(Negatible->getOperand(0)
, m_Constant()) && "Expecting only 1 constant operand"
) ? void (0) : __assert_fail ("!match(Negatible->getOperand(0), m_Constant()) && \"Expecting only 1 constant operand\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2136, __extension__
__PRETTY_FUNCTION__))
;
2137 assert(C->isNegative() && "Expected negative FP constant")(static_cast <bool> (C->isNegative() && "Expected negative FP constant"
) ? void (0) : __assert_fail ("C->isNegative() && \"Expected negative FP constant\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2137, __extension__
__PRETTY_FUNCTION__))
;
2138 Negatible->setOperand(1, ConstantFP::get(Negatible->getType(), abs(*C)));
2139 MadeChange = true;
2140 }
2141 }
2142 assert(MadeChange == true && "Negative constant candidate was not changed")(static_cast <bool> (MadeChange == true && "Negative constant candidate was not changed"
) ? void (0) : __assert_fail ("MadeChange == true && \"Negative constant candidate was not changed\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2142, __extension__
__PRETTY_FUNCTION__))
;
2143
2144 // Negations cancelled out.
2145 if (Candidates.size() % 2 == 0)
2146 return I;
2147
2148 // Negate the final operand in the expression by flipping the opcode of this
2149 // fadd/fsub.
2150 assert(Candidates.size() % 2 == 1 && "Expected odd number")(static_cast <bool> (Candidates.size() % 2 == 1 &&
"Expected odd number") ? void (0) : __assert_fail ("Candidates.size() % 2 == 1 && \"Expected odd number\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2150, __extension__
__PRETTY_FUNCTION__))
;
2151 IRBuilder<> Builder(I);
2152 Value *NewInst = IsFSub ? Builder.CreateFAddFMF(OtherOp, Op, I)
2153 : Builder.CreateFSubFMF(OtherOp, Op, I);
2154 I->replaceAllUsesWith(NewInst);
2155 RedoInsts.insert(I);
2156 return dyn_cast<Instruction>(NewInst);
2157}
2158
2159/// Canonicalize expressions that contain a negative floating-point constant
2160/// of the following form:
2161/// OtherOp + (subtree) -> OtherOp {+/-} (canonical subtree)
2162/// (subtree) + OtherOp -> OtherOp {+/-} (canonical subtree)
2163/// OtherOp - (subtree) -> OtherOp {+/-} (canonical subtree)
2164///
2165/// The fadd/fsub opcode may be switched to allow folding a negation into the
2166/// input instruction.
2167Instruction *ReassociatePass::canonicalizeNegFPConstants(Instruction *I) {
2168 LLVM_DEBUG(dbgs() << "Combine negations for: " << *I << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "Combine negations for: " <<
*I << '\n'; } } while (false)
;
2169 Value *X;
2170 Instruction *Op;
2171 if (match(I, m_FAdd(m_Value(X), m_OneUse(m_Instruction(Op)))))
2172 if (Instruction *R = canonicalizeNegFPConstantsForOp(I, Op, X))
2173 I = R;
2174 if (match(I, m_FAdd(m_OneUse(m_Instruction(Op)), m_Value(X))))
2175 if (Instruction *R = canonicalizeNegFPConstantsForOp(I, Op, X))
2176 I = R;
2177 if (match(I, m_FSub(m_Value(X), m_OneUse(m_Instruction(Op)))))
2178 if (Instruction *R = canonicalizeNegFPConstantsForOp(I, Op, X))
2179 I = R;
2180 return I;
2181}
2182
2183/// Inspect and optimize the given instruction. Note that erasing
2184/// instructions is not allowed.
2185void ReassociatePass::OptimizeInst(Instruction *I) {
2186 // Only consider operations that we understand.
2187 if (!isa<UnaryOperator>(I) && !isa<BinaryOperator>(I))
2188 return;
2189
2190 if (I->getOpcode() == Instruction::Shl && isa<ConstantInt>(I->getOperand(1)))
2191 // If an operand of this shift is a reassociable multiply, or if the shift
2192 // is used by a reassociable multiply or add, turn into a multiply.
2193 if (isReassociableOp(I->getOperand(0), Instruction::Mul) ||
2194 (I->hasOneUse() &&
2195 (isReassociableOp(I->user_back(), Instruction::Mul) ||
2196 isReassociableOp(I->user_back(), Instruction::Add)))) {
2197 Instruction *NI = ConvertShiftToMul(I);
2198 RedoInsts.insert(I);
2199 MadeChange = true;
2200 I = NI;
2201 }
2202
2203 // Commute binary operators, to canonicalize the order of their operands.
2204 // This can potentially expose more CSE opportunities, and makes writing other
2205 // transformations simpler.
2206 if (I->isCommutative())
2207 canonicalizeOperands(I);
2208
2209 // Canonicalize negative constants out of expressions.
2210 if (Instruction *Res = canonicalizeNegFPConstants(I))
2211 I = Res;
2212
2213 // Don't optimize floating-point instructions unless they have the
2214 // appropriate FastMathFlags for reassociation enabled.
2215 if (isa<FPMathOperator>(I) && !hasFPAssociativeFlags(I))
2216 return;
2217
2218 // Do not reassociate boolean (i1) expressions. We want to preserve the
2219 // original order of evaluation for short-circuited comparisons that
2220 // SimplifyCFG has folded to AND/OR expressions. If the expression
2221 // is not further optimized, it is likely to be transformed back to a
2222 // short-circuited form for code gen, and the source order may have been
2223 // optimized for the most likely conditions.
2224 if (I->getType()->isIntegerTy(1))
2225 return;
2226
2227 // If this is a bitwise or instruction of operands
2228 // with no common bits set, convert it to X+Y.
2229 if (I->getOpcode() == Instruction::Or &&
2230 shouldConvertOrWithNoCommonBitsToAdd(I) && !isLoadCombineCandidate(I) &&
2231 haveNoCommonBitsSet(I->getOperand(0), I->getOperand(1),
2232 I->getModule()->getDataLayout(), /*AC=*/nullptr, I,
2233 /*DT=*/nullptr)) {
2234 Instruction *NI = convertOrWithNoCommonBitsToAdd(I);
2235 RedoInsts.insert(I);
2236 MadeChange = true;
2237 I = NI;
2238 }
2239
2240 // If this is a subtract instruction which is not already in negate form,
2241 // see if we can convert it to X+-Y.
2242 if (I->getOpcode() == Instruction::Sub) {
2243 if (ShouldBreakUpSubtract(I)) {
2244 Instruction *NI = BreakUpSubtract(I, RedoInsts);
2245 RedoInsts.insert(I);
2246 MadeChange = true;
2247 I = NI;
2248 } else if (match(I, m_Neg(m_Value()))) {
2249 // Otherwise, this is a negation. See if the operand is a multiply tree
2250 // and if this is not an inner node of a multiply tree.
2251 if (isReassociableOp(I->getOperand(1), Instruction::Mul) &&
2252 (!I->hasOneUse() ||
2253 !isReassociableOp(I->user_back(), Instruction::Mul))) {
2254 Instruction *NI = LowerNegateToMultiply(I);
2255 // If the negate was simplified, revisit the users to see if we can
2256 // reassociate further.
2257 for (User *U : NI->users()) {
2258 if (BinaryOperator *Tmp = dyn_cast<BinaryOperator>(U))
2259 RedoInsts.insert(Tmp);
2260 }
2261 RedoInsts.insert(I);
2262 MadeChange = true;
2263 I = NI;
2264 }
2265 }
2266 } else if (I->getOpcode() == Instruction::FNeg ||
2267 I->getOpcode() == Instruction::FSub) {
2268 if (ShouldBreakUpSubtract(I)) {
2269 Instruction *NI = BreakUpSubtract(I, RedoInsts);
2270 RedoInsts.insert(I);
2271 MadeChange = true;
2272 I = NI;
2273 } else if (match(I, m_FNeg(m_Value()))) {
2274 // Otherwise, this is a negation. See if the operand is a multiply tree
2275 // and if this is not an inner node of a multiply tree.
2276 Value *Op = isa<BinaryOperator>(I) ? I->getOperand(1) :
2277 I->getOperand(0);
2278 if (isReassociableOp(Op, Instruction::FMul) &&
2279 (!I->hasOneUse() ||
2280 !isReassociableOp(I->user_back(), Instruction::FMul))) {
2281 // If the negate was simplified, revisit the users to see if we can
2282 // reassociate further.
2283 Instruction *NI = LowerNegateToMultiply(I);
2284 for (User *U : NI->users()) {
2285 if (BinaryOperator *Tmp = dyn_cast<BinaryOperator>(U))
2286 RedoInsts.insert(Tmp);
2287 }
2288 RedoInsts.insert(I);
2289 MadeChange = true;
2290 I = NI;
2291 }
2292 }
2293 }
2294
2295 // If this instruction is an associative binary operator, process it.
2296 if (!I->isAssociative()) return;
2297 BinaryOperator *BO = cast<BinaryOperator>(I);
2298
2299 // If this is an interior node of a reassociable tree, ignore it until we
2300 // get to the root of the tree, to avoid N^2 analysis.
2301 unsigned Opcode = BO->getOpcode();
2302 if (BO->hasOneUse() && BO->user_back()->getOpcode() == Opcode) {
2303 // During the initial run we will get to the root of the tree.
2304 // But if we get here while we are redoing instructions, there is no
2305 // guarantee that the root will be visited. So Redo later
2306 if (BO->user_back() != BO &&
2307 BO->getParent() == BO->user_back()->getParent())
2308 RedoInsts.insert(BO->user_back());
2309 return;
2310 }
2311
2312 // If this is an add tree that is used by a sub instruction, ignore it
2313 // until we process the subtract.
2314 if (BO->hasOneUse() && BO->getOpcode() == Instruction::Add &&
2315 cast<Instruction>(BO->user_back())->getOpcode() == Instruction::Sub)
2316 return;
2317 if (BO->hasOneUse() && BO->getOpcode() == Instruction::FAdd &&
2318 cast<Instruction>(BO->user_back())->getOpcode() == Instruction::FSub)
2319 return;
2320
2321 ReassociateExpression(BO);
2322}
2323
2324void ReassociatePass::ReassociateExpression(BinaryOperator *I) {
2325 // First, walk the expression tree, linearizing the tree, collecting the
2326 // operand information.
2327 SmallVector<RepeatedValue, 8> Tree;
2328 MadeChange |= LinearizeExprTree(I, Tree, RedoInsts);
2329 SmallVector<ValueEntry, 8> Ops;
2330 Ops.reserve(Tree.size());
2331 for (const RepeatedValue &E : Tree)
2332 Ops.append(E.second.getZExtValue(), ValueEntry(getRank(E.first), E.first));
2333
2334 LLVM_DEBUG(dbgs() << "RAIn:\t"; PrintOps(I, Ops); dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "RAIn:\t"; PrintOps(I, Ops
); dbgs() << '\n'; } } while (false)
;
2335
2336 // Now that we have linearized the tree to a list and have gathered all of
2337 // the operands and their ranks, sort the operands by their rank. Use a
2338 // stable_sort so that values with equal ranks will have their relative
2339 // positions maintained (and so the compiler is deterministic). Note that
2340 // this sorts so that the highest ranking values end up at the beginning of
2341 // the vector.
2342 llvm::stable_sort(Ops);
2343
2344 // Now that we have the expression tree in a convenient
2345 // sorted form, optimize it globally if possible.
2346 if (Value *V = OptimizeExpression(I, Ops)) {
2347 if (V == I)
2348 // Self-referential expression in unreachable code.
2349 return;
2350 // This expression tree simplified to something that isn't a tree,
2351 // eliminate it.
2352 LLVM_DEBUG(dbgs() << "Reassoc to scalar: " << *V << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "Reassoc to scalar: " <<
*V << '\n'; } } while (false)
;
2353 I->replaceAllUsesWith(V);
2354 if (Instruction *VI = dyn_cast<Instruction>(V))
2355 if (I->getDebugLoc())
2356 VI->setDebugLoc(I->getDebugLoc());
2357 RedoInsts.insert(I);
2358 ++NumAnnihil;
2359 return;
2360 }
2361
2362 // We want to sink immediates as deeply as possible except in the case where
2363 // this is a multiply tree used only by an add, and the immediate is a -1.
2364 // In this case we reassociate to put the negation on the outside so that we
2365 // can fold the negation into the add: (-X)*Y + Z -> Z-X*Y
2366 if (I->hasOneUse()) {
2367 if (I->getOpcode() == Instruction::Mul &&
2368 cast<Instruction>(I->user_back())->getOpcode() == Instruction::Add &&
2369 isa<ConstantInt>(Ops.back().Op) &&
2370 cast<ConstantInt>(Ops.back().Op)->isMinusOne()) {
2371 ValueEntry Tmp = Ops.pop_back_val();
2372 Ops.insert(Ops.begin(), Tmp);
2373 } else if (I->getOpcode() == Instruction::FMul &&
2374 cast<Instruction>(I->user_back())->getOpcode() ==
2375 Instruction::FAdd &&
2376 isa<ConstantFP>(Ops.back().Op) &&
2377 cast<ConstantFP>(Ops.back().Op)->isExactlyValue(-1.0)) {
2378 ValueEntry Tmp = Ops.pop_back_val();
2379 Ops.insert(Ops.begin(), Tmp);
2380 }
2381 }
2382
2383 LLVM_DEBUG(dbgs() << "RAOut:\t"; PrintOps(I, Ops); dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("reassociate")) { dbgs() << "RAOut:\t"; PrintOps(I, Ops
); dbgs() << '\n'; } } while (false)
;
2384
2385 if (Ops.size() == 1) {
2386 if (Ops[0].Op == I)
2387 // Self-referential expression in unreachable code.
2388 return;
2389
2390 // This expression tree simplified to something that isn't a tree,
2391 // eliminate it.
2392 I->replaceAllUsesWith(Ops[0].Op);
2393 if (Instruction *OI = dyn_cast<Instruction>(Ops[0].Op))
2394 OI->setDebugLoc(I->getDebugLoc());
2395 RedoInsts.insert(I);
2396 return;
2397 }
2398
2399 if (Ops.size() > 2 && Ops.size() <= GlobalReassociateLimit) {
2400 // Find the pair with the highest count in the pairmap and move it to the
2401 // back of the list so that it can later be CSE'd.
2402 // example:
2403 // a*b*c*d*e
2404 // if c*e is the most "popular" pair, we can express this as
2405 // (((c*e)*d)*b)*a
2406 unsigned Max = 1;
2407 unsigned BestRank = 0;
2408 std::pair<unsigned, unsigned> BestPair;
2409 unsigned Idx = I->getOpcode() - Instruction::BinaryOpsBegin;
2410 for (unsigned i = 0; i < Ops.size() - 1; ++i)
2411 for (unsigned j = i + 1; j < Ops.size(); ++j) {
2412 unsigned Score = 0;
2413 Value *Op0 = Ops[i].Op;
2414 Value *Op1 = Ops[j].Op;
2415 if (std::less<Value *>()(Op1, Op0))
2416 std::swap(Op0, Op1);
2417 auto it = PairMap[Idx].find({Op0, Op1});
2418 if (it != PairMap[Idx].end()) {
2419 // Functions like BreakUpSubtract() can erase the Values we're using
2420 // as keys and create new Values after we built the PairMap. There's a
2421 // small chance that the new nodes can have the same address as
2422 // something already in the table. We shouldn't accumulate the stored
2423 // score in that case as it refers to the wrong Value.
2424 if (it->second.isValid())
2425 Score += it->second.Score;
2426 }
2427
2428 unsigned MaxRank = std::max(Ops[i].Rank, Ops[j].Rank);
2429 if (Score > Max || (Score == Max && MaxRank < BestRank)) {
2430 BestPair = {i, j};
2431 Max = Score;
2432 BestRank = MaxRank;
2433 }
2434 }
2435 if (Max > 1) {
2436 auto Op0 = Ops[BestPair.first];
2437 auto Op1 = Ops[BestPair.second];
2438 Ops.erase(&Ops[BestPair.second]);
2439 Ops.erase(&Ops[BestPair.first]);
2440 Ops.push_back(Op0);
2441 Ops.push_back(Op1);
2442 }
2443 }
2444 // Now that we ordered and optimized the expressions, splat them back into
2445 // the expression tree, removing any unneeded nodes.
2446 RewriteExprTree(I, Ops);
2447}
2448
2449void
2450ReassociatePass::BuildPairMap(ReversePostOrderTraversal<Function *> &RPOT) {
2451 // Make a "pairmap" of how often each operand pair occurs.
2452 for (BasicBlock *BI : RPOT) {
2453 for (Instruction &I : *BI) {
2454 if (!I.isAssociative())
2455 continue;
2456
2457 // Ignore nodes that aren't at the root of trees.
2458 if (I.hasOneUse() && I.user_back()->getOpcode() == I.getOpcode())
2459 continue;
2460
2461 // Collect all operands in a single reassociable expression.
2462 // Since Reassociate has already been run once, we can assume things
2463 // are already canonical according to Reassociation's regime.
2464 SmallVector<Value *, 8> Worklist = { I.getOperand(0), I.getOperand(1) };
2465 SmallVector<Value *, 8> Ops;
2466 while (!Worklist.empty() && Ops.size() <= GlobalReassociateLimit) {
2467 Value *Op = Worklist.pop_back_val();
2468 Instruction *OpI = dyn_cast<Instruction>(Op);
2469 if (!OpI || OpI->getOpcode() != I.getOpcode() || !OpI->hasOneUse()) {
2470 Ops.push_back(Op);
2471 continue;
2472 }
2473 // Be paranoid about self-referencing expressions in unreachable code.
2474 if (OpI->getOperand(0) != OpI)
2475 Worklist.push_back(OpI->getOperand(0));
2476 if (OpI->getOperand(1) != OpI)
2477 Worklist.push_back(OpI->getOperand(1));
2478 }
2479 // Skip extremely long expressions.
2480 if (Ops.size() > GlobalReassociateLimit)
2481 continue;
2482
2483 // Add all pairwise combinations of operands to the pair map.
2484 unsigned BinaryIdx = I.getOpcode() - Instruction::BinaryOpsBegin;
2485 SmallSet<std::pair<Value *, Value*>, 32> Visited;
2486 for (unsigned i = 0; i < Ops.size() - 1; ++i) {
2487 for (unsigned j = i + 1; j < Ops.size(); ++j) {
2488 // Canonicalize operand orderings.
2489 Value *Op0 = Ops[i];
2490 Value *Op1 = Ops[j];
2491 if (std::less<Value *>()(Op1, Op0))
2492 std::swap(Op0, Op1);
2493 if (!Visited.insert({Op0, Op1}).second)
2494 continue;
2495 auto res = PairMap[BinaryIdx].insert({{Op0, Op1}, {Op0, Op1, 1}});
2496 if (!res.second) {
2497 // If either key value has been erased then we've got the same
2498 // address by coincidence. That can't happen here because nothing is
2499 // erasing values but it can happen by the time we're querying the
2500 // map.
2501 assert(res.first->second.isValid() && "WeakVH invalidated")(static_cast <bool> (res.first->second.isValid() &&
"WeakVH invalidated") ? void (0) : __assert_fail ("res.first->second.isValid() && \"WeakVH invalidated\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2501, __extension__
__PRETTY_FUNCTION__))
;
2502 ++res.first->second.Score;
2503 }
2504 }
2505 }
2506 }
2507 }
2508}
2509
2510PreservedAnalyses ReassociatePass::run(Function &F, FunctionAnalysisManager &) {
2511 // Get the functions basic blocks in Reverse Post Order. This order is used by
2512 // BuildRankMap to pre calculate ranks correctly. It also excludes dead basic
2513 // blocks (it has been seen that the analysis in this pass could hang when
2514 // analysing dead basic blocks).
2515 ReversePostOrderTraversal<Function *> RPOT(&F);
2516
2517 // Calculate the rank map for F.
2518 BuildRankMap(F, RPOT);
4
Calling 'ReassociatePass::BuildRankMap'
2519
2520 // Build the pair map before running reassociate.
2521 // Technically this would be more accurate if we did it after one round
2522 // of reassociation, but in practice it doesn't seem to help much on
2523 // real-world code, so don't waste the compile time running reassociate
2524 // twice.
2525 // If a user wants, they could expicitly run reassociate twice in their
2526 // pass pipeline for further potential gains.
2527 // It might also be possible to update the pair map during runtime, but the
2528 // overhead of that may be large if there's many reassociable chains.
2529 BuildPairMap(RPOT);
2530
2531 MadeChange = false;
2532
2533 // Traverse the same blocks that were analysed by BuildRankMap.
2534 for (BasicBlock *BI : RPOT) {
2535 assert(RankMap.count(&*BI) && "BB should be ranked.")(static_cast <bool> (RankMap.count(&*BI) &&
"BB should be ranked.") ? void (0) : __assert_fail ("RankMap.count(&*BI) && \"BB should be ranked.\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2535, __extension__
__PRETTY_FUNCTION__))
;
2536 // Optimize every instruction in the basic block.
2537 for (BasicBlock::iterator II = BI->begin(), IE = BI->end(); II != IE;)
2538 if (isInstructionTriviallyDead(&*II)) {
2539 EraseInst(&*II++);
2540 } else {
2541 OptimizeInst(&*II);
2542 assert(II->getParent() == &*BI && "Moved to a different block!")(static_cast <bool> (II->getParent() == &*BI &&
"Moved to a different block!") ? void (0) : __assert_fail ("II->getParent() == &*BI && \"Moved to a different block!\""
, "llvm/lib/Transforms/Scalar/Reassociate.cpp", 2542, __extension__
__PRETTY_FUNCTION__))
;
2543 ++II;
2544 }
2545
2546 // Make a copy of all the instructions to be redone so we can remove dead
2547 // instructions.
2548 OrderedSet ToRedo(RedoInsts);
2549 // Iterate over all instructions to be reevaluated and remove trivially dead
2550 // instructions. If any operand of the trivially dead instruction becomes
2551 // dead mark it for deletion as well. Continue this process until all
2552 // trivially dead instructions have been removed.
2553 while (!ToRedo.empty()) {
2554 Instruction *I = ToRedo.pop_back_val();
2555 if (isInstructionTriviallyDead(I)) {
2556 RecursivelyEraseDeadInsts(I, ToRedo);
2557 MadeChange = true;
2558 }
2559 }
2560
2561 // Now that we have removed dead instructions, we can reoptimize the
2562 // remaining instructions.
2563 while (!RedoInsts.empty()) {
2564 Instruction *I = RedoInsts.front();
2565 RedoInsts.erase(RedoInsts.begin());
2566 if (isInstructionTriviallyDead(I))
2567 EraseInst(I);
2568 else
2569 OptimizeInst(I);
2570 }
2571 }
2572
2573 // We are done with the rank map and pair map.
2574 RankMap.clear();
2575 ValueRankMap.clear();
2576 for (auto &Entry : PairMap)
2577 Entry.clear();
2578
2579 if (MadeChange) {
2580 PreservedAnalyses PA;
2581 PA.preserveSet<CFGAnalyses>();
2582 return PA;
2583 }
2584
2585 return PreservedAnalyses::all();
2586}
2587
2588namespace {
2589
2590 class ReassociateLegacyPass : public FunctionPass {
2591 ReassociatePass Impl;
2592
2593 public:
2594 static char ID; // Pass identification, replacement for typeid
2595
2596 ReassociateLegacyPass() : FunctionPass(ID) {
2597 initializeReassociateLegacyPassPass(*PassRegistry::getPassRegistry());
2598 }
2599
2600 bool runOnFunction(Function &F) override {
2601 if (skipFunction(F))
1
Assuming the condition is false
2
Taking false branch
2602 return false;
2603
2604 FunctionAnalysisManager DummyFAM;
2605 auto PA = Impl.run(F, DummyFAM);
3
Calling 'ReassociatePass::run'
2606 return !PA.areAllPreserved();
2607 }
2608
2609 void getAnalysisUsage(AnalysisUsage &AU) const override {
2610 AU.setPreservesCFG();
2611 AU.addPreserved<AAResultsWrapperPass>();
2612 AU.addPreserved<BasicAAWrapperPass>();
2613 AU.addPreserved<GlobalsAAWrapperPass>();
2614 }
2615 };
2616
2617} // end anonymous namespace
2618
2619char ReassociateLegacyPass::ID = 0;
2620
2621INITIALIZE_PASS(ReassociateLegacyPass, "reassociate",static void *initializeReassociateLegacyPassPassOnce(PassRegistry
&Registry) { PassInfo *PI = new PassInfo( "Reassociate expressions"
, "reassociate", &ReassociateLegacyPass::ID, PassInfo::NormalCtor_t
(callDefaultCtor<ReassociateLegacyPass>), false, false)
; Registry.registerPass(*PI, true); return PI; } static llvm::
once_flag InitializeReassociateLegacyPassPassFlag; void llvm::
initializeReassociateLegacyPassPass(PassRegistry &Registry
) { llvm::call_once(InitializeReassociateLegacyPassPassFlag, initializeReassociateLegacyPassPassOnce
, std::ref(Registry)); }
2622 "Reassociate expressions", false, false)static void *initializeReassociateLegacyPassPassOnce(PassRegistry
&Registry) { PassInfo *PI = new PassInfo( "Reassociate expressions"
, "reassociate", &ReassociateLegacyPass::ID, PassInfo::NormalCtor_t
(callDefaultCtor<ReassociateLegacyPass>), false, false)
; Registry.registerPass(*PI, true); return PI; } static llvm::
once_flag InitializeReassociateLegacyPassPassFlag; void llvm::
initializeReassociateLegacyPassPass(PassRegistry &Registry
) { llvm::call_once(InitializeReassociateLegacyPassPassFlag, initializeReassociateLegacyPassPassOnce
, std::ref(Registry)); }
2623
2624// Public interface to the Reassociate pass
2625FunctionPass *llvm::createReassociatePass() {
2626 return new ReassociateLegacyPass();
2627}