xref: /freebsd/contrib/llvm-project/llvm/lib/Transforms/Scalar/MergedLoadStoreMotion.cpp (revision 67be1e195acfaec99ce4fffeb17111ce085755f7)
1 //===- MergedLoadStoreMotion.cpp - merge and hoist/sink load/stores -------===//
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 //! \file
10 //! This pass performs merges of loads and stores on both sides of a
11 //  diamond (hammock). It hoists the loads and sinks the stores.
12 //
13 // The algorithm iteratively hoists two loads to the same address out of a
14 // diamond (hammock) and merges them into a single load in the header. Similar
15 // it sinks and merges two stores to the tail block (footer). The algorithm
16 // iterates over the instructions of one side of the diamond and attempts to
17 // find a matching load/store on the other side. New tail/footer block may be
18 // insterted if the tail/footer block has more predecessors (not only the two
19 // predecessors that are forming the diamond). It hoists / sinks when it thinks
20 // it safe to do so.  This optimization helps with eg. hiding load latencies,
21 // triggering if-conversion, and reducing static code size.
22 //
23 // NOTE: This code no longer performs load hoisting, it is subsumed by GVNHoist.
24 //
25 //===----------------------------------------------------------------------===//
26 //
27 //
28 // Example:
29 // Diamond shaped code before merge:
30 //
31 //            header:
32 //                     br %cond, label %if.then, label %if.else
33 //                        +                    +
34 //                       +                      +
35 //                      +                        +
36 //            if.then:                         if.else:
37 //               %lt = load %addr_l               %le = load %addr_l
38 //               <use %lt>                        <use %le>
39 //               <...>                            <...>
40 //               store %st, %addr_s               store %se, %addr_s
41 //               br label %if.end                 br label %if.end
42 //                     +                         +
43 //                      +                       +
44 //                       +                     +
45 //            if.end ("footer"):
46 //                     <...>
47 //
48 // Diamond shaped code after merge:
49 //
50 //            header:
51 //                     %l = load %addr_l
52 //                     br %cond, label %if.then, label %if.else
53 //                        +                    +
54 //                       +                      +
55 //                      +                        +
56 //            if.then:                         if.else:
57 //               <use %l>                         <use %l>
58 //               <...>                            <...>
59 //               br label %if.end                 br label %if.end
60 //                      +                        +
61 //                       +                      +
62 //                        +                    +
63 //            if.end ("footer"):
64 //                     %s.sink = phi [%st, if.then], [%se, if.else]
65 //                     <...>
66 //                     store %s.sink, %addr_s
67 //                     <...>
68 //
69 //
70 //===----------------------- TODO -----------------------------------------===//
71 //
72 // 1) Generalize to regions other than diamonds
73 // 2) Be more aggressive merging memory operations
74 // Note that both changes require register pressure control
75 //
76 //===----------------------------------------------------------------------===//
77 
78 #include "llvm/Transforms/Scalar/MergedLoadStoreMotion.h"
79 #include "llvm/Analysis/AliasAnalysis.h"
80 #include "llvm/Analysis/GlobalsModRef.h"
81 #include "llvm/IR/IRBuilder.h"
82 #include "llvm/IR/Instructions.h"
83 #include "llvm/Support/Debug.h"
84 #include "llvm/Support/raw_ostream.h"
85 #include "llvm/Transforms/Scalar.h"
86 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
87 
88 using namespace llvm;
89 
90 #define DEBUG_TYPE "mldst-motion"
91 
92 namespace {
93 //===----------------------------------------------------------------------===//
94 //                         MergedLoadStoreMotion Pass
95 //===----------------------------------------------------------------------===//
96 class MergedLoadStoreMotion {
97   AliasAnalysis *AA = nullptr;
98 
99   // The mergeLoad/Store algorithms could have Size0 * Size1 complexity,
100   // where Size0 and Size1 are the #instructions on the two sides of
101   // the diamond. The constant chosen here is arbitrary. Compiler Time
102   // Control is enforced by the check Size0 * Size1 < MagicCompileTimeControl.
103   const int MagicCompileTimeControl = 250;
104 
105   const bool SplitFooterBB;
106 public:
107   MergedLoadStoreMotion(bool SplitFooterBB) : SplitFooterBB(SplitFooterBB) {}
108   bool run(Function &F, AliasAnalysis &AA);
109 
110 private:
111   BasicBlock *getDiamondTail(BasicBlock *BB);
112   bool isDiamondHead(BasicBlock *BB);
113   // Routines for sinking stores
114   StoreInst *canSinkFromBlock(BasicBlock *BB, StoreInst *SI);
115   PHINode *getPHIOperand(BasicBlock *BB, StoreInst *S0, StoreInst *S1);
116   bool isStoreSinkBarrierInRange(const Instruction &Start,
117                                  const Instruction &End, MemoryLocation Loc);
118   bool canSinkStoresAndGEPs(StoreInst *S0, StoreInst *S1) const;
119   void sinkStoresAndGEPs(BasicBlock *BB, StoreInst *SinkCand,
120                          StoreInst *ElseInst);
121   bool mergeStores(BasicBlock *BB);
122 };
123 } // end anonymous namespace
124 
125 ///
126 /// Return tail block of a diamond.
127 ///
128 BasicBlock *MergedLoadStoreMotion::getDiamondTail(BasicBlock *BB) {
129   assert(isDiamondHead(BB) && "Basic block is not head of a diamond");
130   return BB->getTerminator()->getSuccessor(0)->getSingleSuccessor();
131 }
132 
133 ///
134 /// True when BB is the head of a diamond (hammock)
135 ///
136 bool MergedLoadStoreMotion::isDiamondHead(BasicBlock *BB) {
137   if (!BB)
138     return false;
139   auto *BI = dyn_cast<BranchInst>(BB->getTerminator());
140   if (!BI || !BI->isConditional())
141     return false;
142 
143   BasicBlock *Succ0 = BI->getSuccessor(0);
144   BasicBlock *Succ1 = BI->getSuccessor(1);
145 
146   if (!Succ0->getSinglePredecessor())
147     return false;
148   if (!Succ1->getSinglePredecessor())
149     return false;
150 
151   BasicBlock *Succ0Succ = Succ0->getSingleSuccessor();
152   BasicBlock *Succ1Succ = Succ1->getSingleSuccessor();
153   // Ignore triangles.
154   if (!Succ0Succ || !Succ1Succ || Succ0Succ != Succ1Succ)
155     return false;
156   return true;
157 }
158 
159 
160 ///
161 /// True when instruction is a sink barrier for a store
162 /// located in Loc
163 ///
164 /// Whenever an instruction could possibly read or modify the
165 /// value being stored or protect against the store from
166 /// happening it is considered a sink barrier.
167 ///
168 bool MergedLoadStoreMotion::isStoreSinkBarrierInRange(const Instruction &Start,
169                                                       const Instruction &End,
170                                                       MemoryLocation Loc) {
171   for (const Instruction &Inst :
172        make_range(Start.getIterator(), End.getIterator()))
173     if (Inst.mayThrow())
174       return true;
175   return AA->canInstructionRangeModRef(Start, End, Loc, ModRefInfo::ModRef);
176 }
177 
178 ///
179 /// Check if \p BB contains a store to the same address as \p SI
180 ///
181 /// \return The store in \p  when it is safe to sink. Otherwise return Null.
182 ///
183 StoreInst *MergedLoadStoreMotion::canSinkFromBlock(BasicBlock *BB1,
184                                                    StoreInst *Store0) {
185   LLVM_DEBUG(dbgs() << "can Sink? : "; Store0->dump(); dbgs() << "\n");
186   BasicBlock *BB0 = Store0->getParent();
187   for (Instruction &Inst : reverse(*BB1)) {
188     auto *Store1 = dyn_cast<StoreInst>(&Inst);
189     if (!Store1)
190       continue;
191 
192     MemoryLocation Loc0 = MemoryLocation::get(Store0);
193     MemoryLocation Loc1 = MemoryLocation::get(Store1);
194 
195     if (AA->isMustAlias(Loc0, Loc1) &&
196         !isStoreSinkBarrierInRange(*Store1->getNextNode(), BB1->back(), Loc1) &&
197         !isStoreSinkBarrierInRange(*Store0->getNextNode(), BB0->back(), Loc0) &&
198         Store0->hasSameSpecialState(Store1) &&
199         CastInst::isBitOrNoopPointerCastable(
200             Store0->getValueOperand()->getType(),
201             Store1->getValueOperand()->getType(),
202             Store0->getModule()->getDataLayout()))
203       return Store1;
204   }
205   return nullptr;
206 }
207 
208 ///
209 /// Create a PHI node in BB for the operands of S0 and S1
210 ///
211 PHINode *MergedLoadStoreMotion::getPHIOperand(BasicBlock *BB, StoreInst *S0,
212                                               StoreInst *S1) {
213   // Create a phi if the values mismatch.
214   Value *Opd1 = S0->getValueOperand();
215   Value *Opd2 = S1->getValueOperand();
216   if (Opd1 == Opd2)
217     return nullptr;
218 
219   auto *NewPN = PHINode::Create(Opd1->getType(), 2, Opd2->getName() + ".sink");
220   NewPN->insertBefore(BB->begin());
221   NewPN->applyMergedLocation(S0->getDebugLoc(), S1->getDebugLoc());
222   NewPN->addIncoming(Opd1, S0->getParent());
223   NewPN->addIncoming(Opd2, S1->getParent());
224   return NewPN;
225 }
226 
227 ///
228 /// Check if 2 stores can be sunk, optionally together with corresponding GEPs.
229 ///
230 bool MergedLoadStoreMotion::canSinkStoresAndGEPs(StoreInst *S0,
231                                                  StoreInst *S1) const {
232   if (S0->getPointerOperand() == S1->getPointerOperand())
233     return true;
234   auto *GEP0 = dyn_cast<GetElementPtrInst>(S0->getPointerOperand());
235   auto *GEP1 = dyn_cast<GetElementPtrInst>(S1->getPointerOperand());
236   return GEP0 && GEP1 && GEP0->isIdenticalTo(GEP1) && GEP0->hasOneUse() &&
237          (GEP0->getParent() == S0->getParent()) && GEP1->hasOneUse() &&
238          (GEP1->getParent() == S1->getParent());
239 }
240 
241 ///
242 /// Merge two stores to same address and sink into \p BB
243 ///
244 /// Optionally also sinks GEP instruction computing the store address
245 ///
246 void MergedLoadStoreMotion::sinkStoresAndGEPs(BasicBlock *BB, StoreInst *S0,
247                                               StoreInst *S1) {
248   Value *Ptr0 = S0->getPointerOperand();
249   Value *Ptr1 = S1->getPointerOperand();
250   // Only one definition?
251   LLVM_DEBUG(dbgs() << "Sink Instruction into BB \n"; BB->dump();
252              dbgs() << "Instruction Left\n"; S0->dump(); dbgs() << "\n";
253              dbgs() << "Instruction Right\n"; S1->dump(); dbgs() << "\n");
254   // Hoist the instruction.
255   BasicBlock::iterator InsertPt = BB->getFirstInsertionPt();
256   // Intersect optional metadata.
257   S0->andIRFlags(S1);
258   S0->dropUnknownNonDebugMetadata();
259   S0->applyMergedLocation(S0->getDebugLoc(), S1->getDebugLoc());
260   S0->mergeDIAssignID(S1);
261 
262   // Insert bitcast for conflicting typed stores (or just use original value if
263   // same type).
264   IRBuilder<> Builder(S0);
265   auto Cast = Builder.CreateBitOrPointerCast(S0->getValueOperand(),
266                                              S1->getValueOperand()->getType());
267   S0->setOperand(0, Cast);
268 
269   // Create the new store to be inserted at the join point.
270   StoreInst *SNew = cast<StoreInst>(S0->clone());
271   SNew->insertBefore(InsertPt);
272   // New PHI operand? Use it.
273   if (PHINode *NewPN = getPHIOperand(BB, S0, S1))
274     SNew->setOperand(0, NewPN);
275   S0->eraseFromParent();
276   S1->eraseFromParent();
277 
278   if (Ptr0 != Ptr1) {
279     auto *GEP0 = cast<GetElementPtrInst>(Ptr0);
280     auto *GEP1 = cast<GetElementPtrInst>(Ptr1);
281     Instruction *GEPNew = GEP0->clone();
282     GEPNew->insertBefore(SNew);
283     GEPNew->applyMergedLocation(GEP0->getDebugLoc(), GEP1->getDebugLoc());
284     SNew->setOperand(1, GEPNew);
285     GEP0->replaceAllUsesWith(GEPNew);
286     GEP0->eraseFromParent();
287     GEP1->replaceAllUsesWith(GEPNew);
288     GEP1->eraseFromParent();
289   }
290 }
291 
292 ///
293 /// True when two stores are equivalent and can sink into the footer
294 ///
295 /// Starting from a diamond head block, iterate over the instructions in one
296 /// successor block and try to match a store in the second successor.
297 ///
298 bool MergedLoadStoreMotion::mergeStores(BasicBlock *HeadBB) {
299 
300   bool MergedStores = false;
301   BasicBlock *TailBB = getDiamondTail(HeadBB);
302   BasicBlock *SinkBB = TailBB;
303   assert(SinkBB && "Footer of a diamond cannot be empty");
304 
305   succ_iterator SI = succ_begin(HeadBB);
306   assert(SI != succ_end(HeadBB) && "Diamond head cannot have zero successors");
307   BasicBlock *Pred0 = *SI;
308   ++SI;
309   assert(SI != succ_end(HeadBB) && "Diamond head cannot have single successor");
310   BasicBlock *Pred1 = *SI;
311   // tail block  of a diamond/hammock?
312   if (Pred0 == Pred1)
313     return false; // No.
314   // bail out early if we can not merge into the footer BB
315   if (!SplitFooterBB && TailBB->hasNPredecessorsOrMore(3))
316     return false;
317   // #Instructions in Pred1 for Compile Time Control
318   auto InstsNoDbg = Pred1->instructionsWithoutDebug();
319   int Size1 = std::distance(InstsNoDbg.begin(), InstsNoDbg.end());
320   int NStores = 0;
321 
322   for (BasicBlock::reverse_iterator RBI = Pred0->rbegin(), RBE = Pred0->rend();
323        RBI != RBE;) {
324 
325     Instruction *I = &*RBI;
326     ++RBI;
327 
328     // Don't sink non-simple (atomic, volatile) stores.
329     auto *S0 = dyn_cast<StoreInst>(I);
330     if (!S0 || !S0->isSimple())
331       continue;
332 
333     ++NStores;
334     if (NStores * Size1 >= MagicCompileTimeControl)
335       break;
336     if (StoreInst *S1 = canSinkFromBlock(Pred1, S0)) {
337       if (!canSinkStoresAndGEPs(S0, S1))
338         // Don't attempt to sink below stores that had to stick around
339         // But after removal of a store and some of its feeding
340         // instruction search again from the beginning since the iterator
341         // is likely stale at this point.
342         break;
343 
344       if (SinkBB == TailBB && TailBB->hasNPredecessorsOrMore(3)) {
345         // We have more than 2 predecessors. Insert a new block
346         // postdominating 2 predecessors we're going to sink from.
347         SinkBB = SplitBlockPredecessors(TailBB, {Pred0, Pred1}, ".sink.split");
348         if (!SinkBB)
349           break;
350       }
351 
352       MergedStores = true;
353       sinkStoresAndGEPs(SinkBB, S0, S1);
354       RBI = Pred0->rbegin();
355       RBE = Pred0->rend();
356       LLVM_DEBUG(dbgs() << "Search again\n"; Instruction *I = &*RBI; I->dump());
357     }
358   }
359   return MergedStores;
360 }
361 
362 bool MergedLoadStoreMotion::run(Function &F, AliasAnalysis &AA) {
363   this->AA = &AA;
364 
365   bool Changed = false;
366   LLVM_DEBUG(dbgs() << "Instruction Merger\n");
367 
368   // Merge unconditional branches, allowing PRE to catch more
369   // optimization opportunities.
370   // This loop doesn't care about newly inserted/split blocks
371   // since they never will be diamond heads.
372   for (BasicBlock &BB : make_early_inc_range(F))
373     // Hoist equivalent loads and sink stores
374     // outside diamonds when possible
375     if (isDiamondHead(&BB))
376       Changed |= mergeStores(&BB);
377   return Changed;
378 }
379 
380 PreservedAnalyses
381 MergedLoadStoreMotionPass::run(Function &F, FunctionAnalysisManager &AM) {
382   MergedLoadStoreMotion Impl(Options.SplitFooterBB);
383   auto &AA = AM.getResult<AAManager>(F);
384   if (!Impl.run(F, AA))
385     return PreservedAnalyses::all();
386 
387   PreservedAnalyses PA;
388   if (!Options.SplitFooterBB)
389     PA.preserveSet<CFGAnalyses>();
390   return PA;
391 }
392 
393 void MergedLoadStoreMotionPass::printPipeline(
394     raw_ostream &OS, function_ref<StringRef(StringRef)> MapClassName2PassName) {
395   static_cast<PassInfoMixin<MergedLoadStoreMotionPass> *>(this)->printPipeline(
396       OS, MapClassName2PassName);
397   OS << '<';
398   OS << (Options.SplitFooterBB ? "" : "no-") << "split-footer-bb";
399   OS << '>';
400 }
401