1 //=- AArch64LoadStoreOptimizer.cpp - AArch64 load/store opt. pass -*- C++ -*-=//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file contains a pass that performs load / store related peephole
11 // optimizations. This pass should be run after register allocation.
13 //===----------------------------------------------------------------------===//
15 #include "AArch64InstrInfo.h"
16 #include "AArch64Subtarget.h"
17 #include "MCTargetDesc/AArch64AddressingModes.h"
18 #include "llvm/ADT/BitVector.h"
19 #include "llvm/ADT/SmallVector.h"
20 #include "llvm/ADT/Statistic.h"
21 #include "llvm/CodeGen/MachineBasicBlock.h"
22 #include "llvm/CodeGen/MachineFunctionPass.h"
23 #include "llvm/CodeGen/MachineInstr.h"
24 #include "llvm/CodeGen/MachineInstrBuilder.h"
25 #include "llvm/Support/CommandLine.h"
26 #include "llvm/Support/Debug.h"
27 #include "llvm/Support/ErrorHandling.h"
28 #include "llvm/Support/raw_ostream.h"
29 #include "llvm/Target/TargetInstrInfo.h"
30 #include "llvm/Target/TargetMachine.h"
31 #include "llvm/Target/TargetRegisterInfo.h"
34 #define DEBUG_TYPE "aarch64-ldst-opt"
36 /// AArch64AllocLoadStoreOpt - Post-register allocation pass to combine
37 /// load / store instructions to form ldp / stp instructions.
39 STATISTIC(NumPairCreated, "Number of load/store pair instructions generated");
40 STATISTIC(NumPostFolded, "Number of post-index updates folded");
41 STATISTIC(NumPreFolded, "Number of pre-index updates folded");
42 STATISTIC(NumUnscaledPairCreated,
43 "Number of load/store from unscaled generated");
44 STATISTIC(NumSmallTypeMerged, "Number of small type loads merged");
46 static cl::opt<unsigned> ScanLimit("aarch64-load-store-scan-limit",
47 cl::init(20), cl::Hidden);
50 void initializeAArch64LoadStoreOptPass(PassRegistry &);
53 #define AARCH64_LOAD_STORE_OPT_NAME "AArch64 load / store optimization pass"
57 typedef struct LdStPairFlags {
58 // If a matching instruction is found, MergeForward is set to true if the
59 // merge is to remove the first instruction and replace the second with
60 // a pair-wise insn, and false if the reverse is true.
63 // SExtIdx gives the index of the result of the load pair that must be
64 // extended. The value of SExtIdx assumes that the paired load produces the
65 // value in this order: (I, returned iterator), i.e., -1 means no value has
66 // to be extended, 0 means I, and 1 means the returned iterator.
69 LdStPairFlags() : MergeForward(false), SExtIdx(-1) {}
71 void setMergeForward(bool V = true) { MergeForward = V; }
72 bool getMergeForward() const { return MergeForward; }
74 void setSExtIdx(int V) { SExtIdx = V; }
75 int getSExtIdx() const { return SExtIdx; }
79 struct AArch64LoadStoreOpt : public MachineFunctionPass {
81 AArch64LoadStoreOpt() : MachineFunctionPass(ID), IsStrictAlign(false) {
82 initializeAArch64LoadStoreOptPass(*PassRegistry::getPassRegistry());
85 const AArch64InstrInfo *TII;
86 const TargetRegisterInfo *TRI;
89 // Scan the instructions looking for a load/store that can be combined
90 // with the current instruction into a load/store pair.
91 // Return the matching instruction if one is found, else MBB->end().
92 MachineBasicBlock::iterator findMatchingInsn(MachineBasicBlock::iterator I,
95 // Merge the two instructions indicated into a single pair-wise instruction.
96 // If MergeForward is true, erase the first instruction and fold its
97 // operation into the second. If false, the reverse. Return the instruction
98 // following the first instruction (which may change during processing).
99 MachineBasicBlock::iterator
100 mergePairedInsns(MachineBasicBlock::iterator I,
101 MachineBasicBlock::iterator Paired,
102 const LdStPairFlags &Flags);
104 // Scan the instruction list to find a base register update that can
105 // be combined with the current instruction (a load or store) using
106 // pre or post indexed addressing with writeback. Scan forwards.
107 MachineBasicBlock::iterator
108 findMatchingUpdateInsnForward(MachineBasicBlock::iterator I, unsigned Limit,
111 // Scan the instruction list to find a base register update that can
112 // be combined with the current instruction (a load or store) using
113 // pre or post indexed addressing with writeback. Scan backwards.
114 MachineBasicBlock::iterator
115 findMatchingUpdateInsnBackward(MachineBasicBlock::iterator I, unsigned Limit);
117 // Find an instruction that updates the base register of the ld/st
119 bool isMatchingUpdateInsn(MachineInstr *MemMI, MachineInstr *MI,
120 unsigned BaseReg, int Offset);
122 // Merge a pre- or post-index base register update into a ld/st instruction.
123 MachineBasicBlock::iterator
124 mergeUpdateInsn(MachineBasicBlock::iterator I,
125 MachineBasicBlock::iterator Update, bool IsPreIdx);
127 // Find and merge foldable ldr/str instructions.
128 bool tryToMergeLdStInst(MachineBasicBlock::iterator &MBBI);
130 bool optimizeBlock(MachineBasicBlock &MBB);
132 bool runOnMachineFunction(MachineFunction &Fn) override;
134 const char *getPassName() const override {
135 return AARCH64_LOAD_STORE_OPT_NAME;
138 char AArch64LoadStoreOpt::ID = 0;
141 INITIALIZE_PASS(AArch64LoadStoreOpt, "aarch64-ldst-opt",
142 AARCH64_LOAD_STORE_OPT_NAME, false, false)
144 static bool isUnscaledLdSt(unsigned Opc) {
148 case AArch64::STURSi:
149 case AArch64::STURDi:
150 case AArch64::STURQi:
151 case AArch64::STURWi:
152 case AArch64::STURXi:
153 case AArch64::LDURSi:
154 case AArch64::LDURDi:
155 case AArch64::LDURQi:
156 case AArch64::LDURWi:
157 case AArch64::LDURXi:
158 case AArch64::LDURSWi:
159 case AArch64::LDURHHi:
164 static bool isUnscaledLdSt(MachineInstr *MI) {
165 return isUnscaledLdSt(MI->getOpcode());
168 static bool isSmallTypeLdMerge(unsigned Opc) {
172 case AArch64::LDRHHui:
173 case AArch64::LDURHHi:
175 // FIXME: Add other instructions (e.g, LDRBBui, LDURSHWi, LDRSHWui, etc.).
178 static bool isSmallTypeLdMerge(MachineInstr *MI) {
179 return isSmallTypeLdMerge(MI->getOpcode());
182 // Scaling factor for unscaled load or store.
183 static int getMemScale(MachineInstr *MI) {
184 switch (MI->getOpcode()) {
186 llvm_unreachable("Opcode has unknown scale!");
187 case AArch64::LDRBBui:
188 case AArch64::STRBBui:
190 case AArch64::LDRHHui:
191 case AArch64::LDURHHi:
192 case AArch64::STRHHui:
194 case AArch64::LDRSui:
195 case AArch64::LDURSi:
196 case AArch64::LDRSWui:
197 case AArch64::LDURSWi:
198 case AArch64::LDRWui:
199 case AArch64::LDURWi:
200 case AArch64::STRSui:
201 case AArch64::STURSi:
202 case AArch64::STRWui:
203 case AArch64::STURWi:
205 case AArch64::LDPSWi:
210 case AArch64::LDRDui:
211 case AArch64::LDURDi:
212 case AArch64::LDRXui:
213 case AArch64::LDURXi:
214 case AArch64::STRDui:
215 case AArch64::STURDi:
216 case AArch64::STRXui:
217 case AArch64::STURXi:
223 case AArch64::LDRQui:
224 case AArch64::LDURQi:
225 case AArch64::STRQui:
226 case AArch64::STURQi:
233 static unsigned getMatchingNonSExtOpcode(unsigned Opc,
234 bool *IsValidLdStrOpc = nullptr) {
236 *IsValidLdStrOpc = true;
240 *IsValidLdStrOpc = false;
242 case AArch64::STRDui:
243 case AArch64::STURDi:
244 case AArch64::STRQui:
245 case AArch64::STURQi:
246 case AArch64::STRWui:
247 case AArch64::STURWi:
248 case AArch64::STRXui:
249 case AArch64::STURXi:
250 case AArch64::LDRDui:
251 case AArch64::LDURDi:
252 case AArch64::LDRQui:
253 case AArch64::LDURQi:
254 case AArch64::LDRWui:
255 case AArch64::LDURWi:
256 case AArch64::LDRXui:
257 case AArch64::LDURXi:
258 case AArch64::STRSui:
259 case AArch64::STURSi:
260 case AArch64::LDRSui:
261 case AArch64::LDURSi:
262 case AArch64::LDRHHui:
263 case AArch64::LDURHHi:
265 case AArch64::LDRSWui:
266 return AArch64::LDRWui;
267 case AArch64::LDURSWi:
268 return AArch64::LDURWi;
272 static unsigned getMatchingPairOpcode(unsigned Opc) {
275 llvm_unreachable("Opcode has no pairwise equivalent!");
276 case AArch64::STRSui:
277 case AArch64::STURSi:
278 return AArch64::STPSi;
279 case AArch64::STRDui:
280 case AArch64::STURDi:
281 return AArch64::STPDi;
282 case AArch64::STRQui:
283 case AArch64::STURQi:
284 return AArch64::STPQi;
285 case AArch64::STRWui:
286 case AArch64::STURWi:
287 return AArch64::STPWi;
288 case AArch64::STRXui:
289 case AArch64::STURXi:
290 return AArch64::STPXi;
291 case AArch64::LDRSui:
292 case AArch64::LDURSi:
293 return AArch64::LDPSi;
294 case AArch64::LDRDui:
295 case AArch64::LDURDi:
296 return AArch64::LDPDi;
297 case AArch64::LDRQui:
298 case AArch64::LDURQi:
299 return AArch64::LDPQi;
300 case AArch64::LDRWui:
301 case AArch64::LDURWi:
302 return AArch64::LDPWi;
303 case AArch64::LDRXui:
304 case AArch64::LDURXi:
305 return AArch64::LDPXi;
306 case AArch64::LDRSWui:
307 case AArch64::LDURSWi:
308 return AArch64::LDPSWi;
309 case AArch64::LDRHHui:
310 return AArch64::LDRWui;
311 case AArch64::LDURHHi:
312 return AArch64::LDURWi;
316 static unsigned getPreIndexedOpcode(unsigned Opc) {
319 llvm_unreachable("Opcode has no pre-indexed equivalent!");
320 case AArch64::STRSui:
321 return AArch64::STRSpre;
322 case AArch64::STRDui:
323 return AArch64::STRDpre;
324 case AArch64::STRQui:
325 return AArch64::STRQpre;
326 case AArch64::STRBBui:
327 return AArch64::STRBBpre;
328 case AArch64::STRHHui:
329 return AArch64::STRHHpre;
330 case AArch64::STRWui:
331 return AArch64::STRWpre;
332 case AArch64::STRXui:
333 return AArch64::STRXpre;
334 case AArch64::LDRSui:
335 return AArch64::LDRSpre;
336 case AArch64::LDRDui:
337 return AArch64::LDRDpre;
338 case AArch64::LDRQui:
339 return AArch64::LDRQpre;
340 case AArch64::LDRBBui:
341 return AArch64::LDRBBpre;
342 case AArch64::LDRHHui:
343 return AArch64::LDRHHpre;
344 case AArch64::LDRWui:
345 return AArch64::LDRWpre;
346 case AArch64::LDRXui:
347 return AArch64::LDRXpre;
348 case AArch64::LDRSWui:
349 return AArch64::LDRSWpre;
351 return AArch64::LDPSpre;
352 case AArch64::LDPSWi:
353 return AArch64::LDPSWpre;
355 return AArch64::LDPDpre;
357 return AArch64::LDPQpre;
359 return AArch64::LDPWpre;
361 return AArch64::LDPXpre;
363 return AArch64::STPSpre;
365 return AArch64::STPDpre;
367 return AArch64::STPQpre;
369 return AArch64::STPWpre;
371 return AArch64::STPXpre;
375 static unsigned getPostIndexedOpcode(unsigned Opc) {
378 llvm_unreachable("Opcode has no post-indexed wise equivalent!");
379 case AArch64::STRSui:
380 return AArch64::STRSpost;
381 case AArch64::STRDui:
382 return AArch64::STRDpost;
383 case AArch64::STRQui:
384 return AArch64::STRQpost;
385 case AArch64::STRBBui:
386 return AArch64::STRBBpost;
387 case AArch64::STRHHui:
388 return AArch64::STRHHpost;
389 case AArch64::STRWui:
390 return AArch64::STRWpost;
391 case AArch64::STRXui:
392 return AArch64::STRXpost;
393 case AArch64::LDRSui:
394 return AArch64::LDRSpost;
395 case AArch64::LDRDui:
396 return AArch64::LDRDpost;
397 case AArch64::LDRQui:
398 return AArch64::LDRQpost;
399 case AArch64::LDRBBui:
400 return AArch64::LDRBBpost;
401 case AArch64::LDRHHui:
402 return AArch64::LDRHHpost;
403 case AArch64::LDRWui:
404 return AArch64::LDRWpost;
405 case AArch64::LDRXui:
406 return AArch64::LDRXpost;
407 case AArch64::LDRSWui:
408 return AArch64::LDRSWpost;
410 return AArch64::LDPSpost;
411 case AArch64::LDPSWi:
412 return AArch64::LDPSWpost;
414 return AArch64::LDPDpost;
416 return AArch64::LDPQpost;
418 return AArch64::LDPWpost;
420 return AArch64::LDPXpost;
422 return AArch64::STPSpost;
424 return AArch64::STPDpost;
426 return AArch64::STPQpost;
428 return AArch64::STPWpost;
430 return AArch64::STPXpost;
434 static bool isPairedLdSt(const MachineInstr *MI) {
435 switch (MI->getOpcode()) {
439 case AArch64::LDPSWi:
453 static const MachineOperand &getLdStRegOp(const MachineInstr *MI,
454 unsigned PairedRegOp = 0) {
455 assert(PairedRegOp < 2 && "Unexpected register operand idx.");
456 unsigned Idx = isPairedLdSt(MI) ? PairedRegOp : 0;
457 return MI->getOperand(Idx);
460 static const MachineOperand &getLdStBaseOp(const MachineInstr *MI) {
461 unsigned Idx = isPairedLdSt(MI) ? 2 : 1;
462 return MI->getOperand(Idx);
465 static const MachineOperand &getLdStOffsetOp(const MachineInstr *MI) {
466 unsigned Idx = isPairedLdSt(MI) ? 3 : 2;
467 return MI->getOperand(Idx);
470 // Copy MachineMemOperands from Op0 and Op1 to a new array assigned to MI.
471 static void concatenateMemOperands(MachineInstr *MI, MachineInstr *Op0,
473 assert(MI->memoperands_empty() && "expected a new machineinstr");
474 size_t numMemRefs = (Op0->memoperands_end() - Op0->memoperands_begin()) +
475 (Op1->memoperands_end() - Op1->memoperands_begin());
477 MachineFunction *MF = MI->getParent()->getParent();
478 MachineSDNode::mmo_iterator MemBegin = MF->allocateMemRefsArray(numMemRefs);
479 MachineSDNode::mmo_iterator MemEnd =
480 std::copy(Op0->memoperands_begin(), Op0->memoperands_end(), MemBegin);
481 MemEnd = std::copy(Op1->memoperands_begin(), Op1->memoperands_end(), MemEnd);
482 MI->setMemRefs(MemBegin, MemEnd);
485 MachineBasicBlock::iterator
486 AArch64LoadStoreOpt::mergePairedInsns(MachineBasicBlock::iterator I,
487 MachineBasicBlock::iterator Paired,
488 const LdStPairFlags &Flags) {
489 MachineBasicBlock::iterator NextI = I;
491 // If NextI is the second of the two instructions to be merged, we need
492 // to skip one further. Either way we merge will invalidate the iterator,
493 // and we don't need to scan the new instruction, as it's a pairwise
494 // instruction, which we're not considering for further action anyway.
498 int SExtIdx = Flags.getSExtIdx();
500 SExtIdx == -1 ? I->getOpcode() : getMatchingNonSExtOpcode(I->getOpcode());
501 bool IsUnscaled = isUnscaledLdSt(Opc);
502 int OffsetStride = IsUnscaled ? getMemScale(I) : 1;
504 bool MergeForward = Flags.getMergeForward();
505 unsigned NewOpc = getMatchingPairOpcode(Opc);
506 // Insert our new paired instruction after whichever of the paired
507 // instructions MergeForward indicates.
508 MachineBasicBlock::iterator InsertionPoint = MergeForward ? Paired : I;
509 // Also based on MergeForward is from where we copy the base register operand
510 // so we get the flags compatible with the input code.
511 const MachineOperand &BaseRegOp =
512 MergeForward ? getLdStBaseOp(Paired) : getLdStBaseOp(I);
514 // Which register is Rt and which is Rt2 depends on the offset order.
515 MachineInstr *RtMI, *Rt2MI;
516 if (getLdStOffsetOp(I).getImm() ==
517 getLdStOffsetOp(Paired).getImm() + OffsetStride) {
520 // Here we swapped the assumption made for SExtIdx.
521 // I.e., we turn ldp I, Paired into ldp Paired, I.
522 // Update the index accordingly.
524 SExtIdx = (SExtIdx + 1) % 2;
530 int OffsetImm = getLdStOffsetOp(RtMI).getImm();
532 if (isSmallTypeLdMerge(Opc)) {
533 // Change the scaled offset from small to large type.
536 MachineInstr *RtNewDest = MergeForward ? I : Paired;
537 // Construct the new load instruction.
538 // FIXME: currently we support only halfword unsigned load. We need to
539 // handle byte type, signed, and store instructions as well.
540 MachineInstr *NewMemMI, *BitExtMI1, *BitExtMI2;
541 NewMemMI = BuildMI(*I->getParent(), InsertionPoint, I->getDebugLoc(),
543 .addOperand(getLdStRegOp(RtNewDest))
544 .addOperand(BaseRegOp)
547 // Copy MachineMemOperands from the original loads.
548 concatenateMemOperands(NewMemMI, I, Paired);
552 << "Creating the new load and extract. Replacing instructions:\n ");
553 DEBUG(I->print(dbgs()));
554 DEBUG(dbgs() << " ");
555 DEBUG(Paired->print(dbgs()));
556 DEBUG(dbgs() << " with instructions:\n ");
557 DEBUG((NewMemMI)->print(dbgs()));
559 MachineInstr *ExtDestMI = MergeForward ? Paired : I;
560 if (ExtDestMI == Rt2MI) {
561 // Create the bitfield extract for high half.
562 BitExtMI1 = BuildMI(*I->getParent(), InsertionPoint, I->getDebugLoc(),
563 TII->get(AArch64::UBFMWri))
564 .addOperand(getLdStRegOp(Rt2MI))
565 .addReg(getLdStRegOp(RtNewDest).getReg())
568 // Create the bitfield extract for low half.
569 BitExtMI2 = BuildMI(*I->getParent(), InsertionPoint, I->getDebugLoc(),
570 TII->get(AArch64::ANDWri))
571 .addOperand(getLdStRegOp(RtMI))
572 .addReg(getLdStRegOp(RtNewDest).getReg())
575 // Create the bitfield extract for low half.
576 BitExtMI1 = BuildMI(*I->getParent(), InsertionPoint, I->getDebugLoc(),
577 TII->get(AArch64::ANDWri))
578 .addOperand(getLdStRegOp(RtMI))
579 .addReg(getLdStRegOp(RtNewDest).getReg())
581 // Create the bitfield extract for high half.
582 BitExtMI2 = BuildMI(*I->getParent(), InsertionPoint, I->getDebugLoc(),
583 TII->get(AArch64::UBFMWri))
584 .addOperand(getLdStRegOp(Rt2MI))
585 .addReg(getLdStRegOp(RtNewDest).getReg())
589 DEBUG(dbgs() << " ");
590 DEBUG((BitExtMI1)->print(dbgs()));
591 DEBUG(dbgs() << " ");
592 DEBUG((BitExtMI2)->print(dbgs()));
593 DEBUG(dbgs() << "\n");
595 // Erase the old instructions.
596 I->eraseFromParent();
597 Paired->eraseFromParent();
603 OffsetImm /= OffsetStride;
605 // Construct the new instruction.
606 MachineInstrBuilder MIB = BuildMI(*I->getParent(), InsertionPoint,
607 I->getDebugLoc(), TII->get(NewOpc))
608 .addOperand(getLdStRegOp(RtMI))
609 .addOperand(getLdStRegOp(Rt2MI))
610 .addOperand(BaseRegOp)
614 // FIXME: Do we need/want to copy the mem operands from the source
615 // instructions? Probably. What uses them after this?
617 DEBUG(dbgs() << "Creating pair load/store. Replacing instructions:\n ");
618 DEBUG(I->print(dbgs()));
619 DEBUG(dbgs() << " ");
620 DEBUG(Paired->print(dbgs()));
621 DEBUG(dbgs() << " with instruction:\n ");
624 // Generate the sign extension for the proper result of the ldp.
625 // I.e., with X1, that would be:
626 // %W1<def> = KILL %W1, %X1<imp-def>
627 // %X1<def> = SBFMXri %X1<kill>, 0, 31
628 MachineOperand &DstMO = MIB->getOperand(SExtIdx);
629 // Right now, DstMO has the extended register, since it comes from an
631 unsigned DstRegX = DstMO.getReg();
632 // Get the W variant of that register.
633 unsigned DstRegW = TRI->getSubReg(DstRegX, AArch64::sub_32);
634 // Update the result of LDP to use the W instead of the X variant.
635 DstMO.setReg(DstRegW);
636 DEBUG(((MachineInstr *)MIB)->print(dbgs()));
637 DEBUG(dbgs() << "\n");
638 // Make the machine verifier happy by providing a definition for
640 // Insert this definition right after the generated LDP, i.e., before
642 MachineInstrBuilder MIBKill =
643 BuildMI(*I->getParent(), InsertionPoint, I->getDebugLoc(),
644 TII->get(TargetOpcode::KILL), DstRegW)
646 .addReg(DstRegX, RegState::Define);
647 MIBKill->getOperand(2).setImplicit();
648 // Create the sign extension.
649 MachineInstrBuilder MIBSXTW =
650 BuildMI(*I->getParent(), InsertionPoint, I->getDebugLoc(),
651 TII->get(AArch64::SBFMXri), DstRegX)
656 DEBUG(dbgs() << " Extend operand:\n ");
657 DEBUG(((MachineInstr *)MIBSXTW)->print(dbgs()));
658 DEBUG(dbgs() << "\n");
660 DEBUG(((MachineInstr *)MIB)->print(dbgs()));
661 DEBUG(dbgs() << "\n");
664 // Erase the old instructions.
665 I->eraseFromParent();
666 Paired->eraseFromParent();
671 /// trackRegDefsUses - Remember what registers the specified instruction uses
673 static void trackRegDefsUses(const MachineInstr *MI, BitVector &ModifiedRegs,
675 const TargetRegisterInfo *TRI) {
676 for (const MachineOperand &MO : MI->operands()) {
678 ModifiedRegs.setBitsNotInMask(MO.getRegMask());
682 unsigned Reg = MO.getReg();
684 for (MCRegAliasIterator AI(Reg, TRI, true); AI.isValid(); ++AI)
685 ModifiedRegs.set(*AI);
687 assert(MO.isUse() && "Reg operand not a def and not a use?!?");
688 for (MCRegAliasIterator AI(Reg, TRI, true); AI.isValid(); ++AI)
694 static bool inBoundsForPair(bool IsUnscaled, int Offset, int OffsetStride) {
695 // Convert the byte-offset used by unscaled into an "element" offset used
696 // by the scaled pair load/store instructions.
698 Offset /= OffsetStride;
700 return Offset <= 63 && Offset >= -64;
703 // Do alignment, specialized to power of 2 and for signed ints,
704 // avoiding having to do a C-style cast from uint_64t to int when
705 // using RoundUpToAlignment from include/llvm/Support/MathExtras.h.
706 // FIXME: Move this function to include/MathExtras.h?
707 static int alignTo(int Num, int PowOf2) {
708 return (Num + PowOf2 - 1) & ~(PowOf2 - 1);
711 static bool mayAlias(MachineInstr *MIa, MachineInstr *MIb,
712 const AArch64InstrInfo *TII) {
713 // One of the instructions must modify memory.
714 if (!MIa->mayStore() && !MIb->mayStore())
717 // Both instructions must be memory operations.
718 if (!MIa->mayLoadOrStore() && !MIb->mayLoadOrStore())
721 return !TII->areMemAccessesTriviallyDisjoint(MIa, MIb);
724 static bool mayAlias(MachineInstr *MIa,
725 SmallVectorImpl<MachineInstr *> &MemInsns,
726 const AArch64InstrInfo *TII) {
727 for (auto &MIb : MemInsns)
728 if (mayAlias(MIa, MIb, TII))
734 /// findMatchingInsn - Scan the instructions looking for a load/store that can
735 /// be combined with the current instruction into a load/store pair.
736 MachineBasicBlock::iterator
737 AArch64LoadStoreOpt::findMatchingInsn(MachineBasicBlock::iterator I,
738 LdStPairFlags &Flags, unsigned Limit) {
739 MachineBasicBlock::iterator E = I->getParent()->end();
740 MachineBasicBlock::iterator MBBI = I;
741 MachineInstr *FirstMI = I;
744 unsigned Opc = FirstMI->getOpcode();
745 bool MayLoad = FirstMI->mayLoad();
746 bool IsUnscaled = isUnscaledLdSt(FirstMI);
747 unsigned Reg = getLdStRegOp(FirstMI).getReg();
748 unsigned BaseReg = getLdStBaseOp(FirstMI).getReg();
749 int Offset = getLdStOffsetOp(FirstMI).getImm();
751 // Early exit if the first instruction modifies the base register.
752 // e.g., ldr x0, [x0]
753 if (FirstMI->modifiesRegister(BaseReg, TRI))
756 // Early exit if the offset if not possible to match. (6 bits of positive
757 // range, plus allow an extra one in case we find a later insn that matches
759 int OffsetStride = IsUnscaled ? getMemScale(FirstMI) : 1;
760 if (!isSmallTypeLdMerge(Opc) &&
761 !inBoundsForPair(IsUnscaled, Offset, OffsetStride))
764 // Track which registers have been modified and used between the first insn
765 // (inclusive) and the second insn.
766 BitVector ModifiedRegs, UsedRegs;
767 ModifiedRegs.resize(TRI->getNumRegs());
768 UsedRegs.resize(TRI->getNumRegs());
770 // Remember any instructions that read/write memory between FirstMI and MI.
771 SmallVector<MachineInstr *, 4> MemInsns;
773 for (unsigned Count = 0; MBBI != E && Count < Limit; ++MBBI) {
774 MachineInstr *MI = MBBI;
775 // Skip DBG_VALUE instructions. Otherwise debug info can affect the
776 // optimization by changing how far we scan.
777 if (MI->isDebugValue())
780 // Now that we know this is a real instruction, count it.
783 bool CanMergeOpc = Opc == MI->getOpcode();
784 Flags.setSExtIdx(-1);
786 bool IsValidLdStrOpc;
787 unsigned NonSExtOpc = getMatchingNonSExtOpcode(Opc, &IsValidLdStrOpc);
788 assert(IsValidLdStrOpc &&
789 "Given Opc should be a Load or Store with an immediate");
790 // Opc will be the first instruction in the pair.
791 Flags.setSExtIdx(NonSExtOpc == (unsigned)Opc ? 1 : 0);
792 CanMergeOpc = NonSExtOpc == getMatchingNonSExtOpcode(MI->getOpcode());
795 if (CanMergeOpc && getLdStOffsetOp(MI).isImm()) {
796 assert(MI->mayLoadOrStore() && "Expected memory operation.");
797 // If we've found another instruction with the same opcode, check to see
798 // if the base and offset are compatible with our starting instruction.
799 // These instructions all have scaled immediate operands, so we just
800 // check for +1/-1. Make sure to check the new instruction offset is
801 // actually an immediate and not a symbolic reference destined for
804 // Pairwise instructions have a 7-bit signed offset field. Single insns
805 // have a 12-bit unsigned offset field. To be a valid combine, the
806 // final offset must be in range.
807 unsigned MIBaseReg = getLdStBaseOp(MI).getReg();
808 int MIOffset = getLdStOffsetOp(MI).getImm();
809 if (BaseReg == MIBaseReg && ((Offset == MIOffset + OffsetStride) ||
810 (Offset + OffsetStride == MIOffset))) {
811 int MinOffset = Offset < MIOffset ? Offset : MIOffset;
812 // If this is a volatile load/store that otherwise matched, stop looking
813 // as something is going on that we don't have enough information to
814 // safely transform. Similarly, stop if we see a hint to avoid pairs.
815 if (MI->hasOrderedMemoryRef() || TII->isLdStPairSuppressed(MI))
817 // If the resultant immediate offset of merging these instructions
818 // is out of range for a pairwise instruction, bail and keep looking.
819 bool MIIsUnscaled = isUnscaledLdSt(MI);
820 bool IsSmallTypeLd = isSmallTypeLdMerge(MI->getOpcode());
821 if (!IsSmallTypeLd &&
822 !inBoundsForPair(MIIsUnscaled, MinOffset, OffsetStride)) {
823 trackRegDefsUses(MI, ModifiedRegs, UsedRegs, TRI);
824 MemInsns.push_back(MI);
829 // If the alignment requirements of the larger type scaled load
830 // instruction can't express the scaled offset of the smaller type
831 // input, bail and keep looking.
832 if (!IsUnscaled && alignTo(MinOffset, 2) != MinOffset) {
833 trackRegDefsUses(MI, ModifiedRegs, UsedRegs, TRI);
834 MemInsns.push_back(MI);
838 // If the alignment requirements of the paired (scaled) instruction
839 // can't express the offset of the unscaled input, bail and keep
841 if (IsUnscaled && (alignTo(MinOffset, OffsetStride) != MinOffset)) {
842 trackRegDefsUses(MI, ModifiedRegs, UsedRegs, TRI);
843 MemInsns.push_back(MI);
847 // If the destination register of the loads is the same register, bail
848 // and keep looking. A load-pair instruction with both destination
849 // registers the same is UNPREDICTABLE and will result in an exception.
850 if (MayLoad && Reg == getLdStRegOp(MI).getReg()) {
851 trackRegDefsUses(MI, ModifiedRegs, UsedRegs, TRI);
852 MemInsns.push_back(MI);
856 // If the Rt of the second instruction was not modified or used between
857 // the two instructions and none of the instructions between the second
858 // and first alias with the second, we can combine the second into the
860 if (!ModifiedRegs[getLdStRegOp(MI).getReg()] &&
861 !(MI->mayLoad() && UsedRegs[getLdStRegOp(MI).getReg()]) &&
862 !mayAlias(MI, MemInsns, TII)) {
863 Flags.setMergeForward(false);
867 // Likewise, if the Rt of the first instruction is not modified or used
868 // between the two instructions and none of the instructions between the
869 // first and the second alias with the first, we can combine the first
871 if (!ModifiedRegs[getLdStRegOp(FirstMI).getReg()] &&
872 !(MayLoad && UsedRegs[getLdStRegOp(FirstMI).getReg()]) &&
873 !mayAlias(FirstMI, MemInsns, TII)) {
874 Flags.setMergeForward(true);
877 // Unable to combine these instructions due to interference in between.
882 // If the instruction wasn't a matching load or store. Stop searching if we
883 // encounter a call instruction that might modify memory.
887 // Update modified / uses register lists.
888 trackRegDefsUses(MI, ModifiedRegs, UsedRegs, TRI);
890 // Otherwise, if the base register is modified, we have no match, so
892 if (ModifiedRegs[BaseReg])
895 // Update list of instructions that read/write memory.
896 if (MI->mayLoadOrStore())
897 MemInsns.push_back(MI);
902 MachineBasicBlock::iterator
903 AArch64LoadStoreOpt::mergeUpdateInsn(MachineBasicBlock::iterator I,
904 MachineBasicBlock::iterator Update,
906 assert((Update->getOpcode() == AArch64::ADDXri ||
907 Update->getOpcode() == AArch64::SUBXri) &&
908 "Unexpected base register update instruction to merge!");
909 MachineBasicBlock::iterator NextI = I;
910 // Return the instruction following the merged instruction, which is
911 // the instruction following our unmerged load. Unless that's the add/sub
912 // instruction we're merging, in which case it's the one after that.
913 if (++NextI == Update)
916 int Value = Update->getOperand(2).getImm();
917 assert(AArch64_AM::getShiftValue(Update->getOperand(3).getImm()) == 0 &&
918 "Can't merge 1 << 12 offset into pre-/post-indexed load / store");
919 if (Update->getOpcode() == AArch64::SUBXri)
922 unsigned NewOpc = IsPreIdx ? getPreIndexedOpcode(I->getOpcode())
923 : getPostIndexedOpcode(I->getOpcode());
924 MachineInstrBuilder MIB;
925 if (!isPairedLdSt(I)) {
926 // Non-paired instruction.
927 MIB = BuildMI(*I->getParent(), I, I->getDebugLoc(), TII->get(NewOpc))
928 .addOperand(getLdStRegOp(Update))
929 .addOperand(getLdStRegOp(I))
930 .addOperand(getLdStBaseOp(I))
933 // Paired instruction.
934 int Scale = getMemScale(I);
935 MIB = BuildMI(*I->getParent(), I, I->getDebugLoc(), TII->get(NewOpc))
936 .addOperand(getLdStRegOp(Update))
937 .addOperand(getLdStRegOp(I, 0))
938 .addOperand(getLdStRegOp(I, 1))
939 .addOperand(getLdStBaseOp(I))
940 .addImm(Value / Scale);
945 DEBUG(dbgs() << "Creating pre-indexed load/store.");
947 DEBUG(dbgs() << "Creating post-indexed load/store.");
948 DEBUG(dbgs() << " Replacing instructions:\n ");
949 DEBUG(I->print(dbgs()));
950 DEBUG(dbgs() << " ");
951 DEBUG(Update->print(dbgs()));
952 DEBUG(dbgs() << " with instruction:\n ");
953 DEBUG(((MachineInstr *)MIB)->print(dbgs()));
954 DEBUG(dbgs() << "\n");
956 // Erase the old instructions for the block.
957 I->eraseFromParent();
958 Update->eraseFromParent();
963 bool AArch64LoadStoreOpt::isMatchingUpdateInsn(MachineInstr *MemMI,
965 unsigned BaseReg, int Offset) {
966 switch (MI->getOpcode()) {
969 case AArch64::SUBXri:
970 // Negate the offset for a SUB instruction.
973 case AArch64::ADDXri:
974 // Make sure it's a vanilla immediate operand, not a relocation or
975 // anything else we can't handle.
976 if (!MI->getOperand(2).isImm())
978 // Watch out for 1 << 12 shifted value.
979 if (AArch64_AM::getShiftValue(MI->getOperand(3).getImm()))
982 // The update instruction source and destination register must be the
983 // same as the load/store base register.
984 if (MI->getOperand(0).getReg() != BaseReg ||
985 MI->getOperand(1).getReg() != BaseReg)
988 bool IsPairedInsn = isPairedLdSt(MemMI);
989 int UpdateOffset = MI->getOperand(2).getImm();
990 // For non-paired load/store instructions, the immediate must fit in a
991 // signed 9-bit integer.
992 if (!IsPairedInsn && (UpdateOffset > 255 || UpdateOffset < -256))
995 // For paired load/store instructions, the immediate must be a multiple of
996 // the scaling factor. The scaled offset must also fit into a signed 7-bit
999 int Scale = getMemScale(MemMI);
1000 if (UpdateOffset % Scale != 0)
1003 int ScaledOffset = UpdateOffset / Scale;
1004 if (ScaledOffset > 64 || ScaledOffset < -64)
1008 // If we have a non-zero Offset, we check that it matches the amount
1009 // we're adding to the register.
1010 if (!Offset || Offset == MI->getOperand(2).getImm())
1017 MachineBasicBlock::iterator AArch64LoadStoreOpt::findMatchingUpdateInsnForward(
1018 MachineBasicBlock::iterator I, unsigned Limit, int UnscaledOffset) {
1019 MachineBasicBlock::iterator E = I->getParent()->end();
1020 MachineInstr *MemMI = I;
1021 MachineBasicBlock::iterator MBBI = I;
1023 unsigned BaseReg = getLdStBaseOp(MemMI).getReg();
1024 int MIUnscaledOffset = getLdStOffsetOp(MemMI).getImm() * getMemScale(MemMI);
1026 // Scan forward looking for post-index opportunities. Updating instructions
1027 // can't be formed if the memory instruction doesn't have the offset we're
1029 if (MIUnscaledOffset != UnscaledOffset)
1032 // If the base register overlaps a destination register, we can't
1033 // merge the update.
1034 bool IsPairedInsn = isPairedLdSt(MemMI);
1035 for (unsigned i = 0, e = IsPairedInsn ? 2 : 1; i != e; ++i) {
1036 unsigned DestReg = getLdStRegOp(MemMI, i).getReg();
1037 if (DestReg == BaseReg || TRI->isSubRegister(BaseReg, DestReg))
1041 // Track which registers have been modified and used between the first insn
1042 // (inclusive) and the second insn.
1043 BitVector ModifiedRegs, UsedRegs;
1044 ModifiedRegs.resize(TRI->getNumRegs());
1045 UsedRegs.resize(TRI->getNumRegs());
1047 for (unsigned Count = 0; MBBI != E; ++MBBI) {
1048 MachineInstr *MI = MBBI;
1049 // Skip DBG_VALUE instructions. Otherwise debug info can affect the
1050 // optimization by changing how far we scan.
1051 if (MI->isDebugValue())
1054 // Now that we know this is a real instruction, count it.
1057 // If we found a match, return it.
1058 if (isMatchingUpdateInsn(I, MI, BaseReg, UnscaledOffset))
1061 // Update the status of what the instruction clobbered and used.
1062 trackRegDefsUses(MI, ModifiedRegs, UsedRegs, TRI);
1064 // Otherwise, if the base register is used or modified, we have no match, so
1066 if (ModifiedRegs[BaseReg] || UsedRegs[BaseReg])
1072 MachineBasicBlock::iterator AArch64LoadStoreOpt::findMatchingUpdateInsnBackward(
1073 MachineBasicBlock::iterator I, unsigned Limit) {
1074 MachineBasicBlock::iterator B = I->getParent()->begin();
1075 MachineBasicBlock::iterator E = I->getParent()->end();
1076 MachineInstr *MemMI = I;
1077 MachineBasicBlock::iterator MBBI = I;
1079 unsigned BaseReg = getLdStBaseOp(MemMI).getReg();
1080 int Offset = getLdStOffsetOp(MemMI).getImm();
1082 // If the load/store is the first instruction in the block, there's obviously
1083 // not any matching update. Ditto if the memory offset isn't zero.
1084 if (MBBI == B || Offset != 0)
1086 // If the base register overlaps a destination register, we can't
1087 // merge the update.
1088 bool IsPairedInsn = isPairedLdSt(MemMI);
1089 for (unsigned i = 0, e = IsPairedInsn ? 2 : 1; i != e; ++i) {
1090 unsigned DestReg = getLdStRegOp(MemMI, i).getReg();
1091 if (DestReg == BaseReg || TRI->isSubRegister(BaseReg, DestReg))
1095 // Track which registers have been modified and used between the first insn
1096 // (inclusive) and the second insn.
1097 BitVector ModifiedRegs, UsedRegs;
1098 ModifiedRegs.resize(TRI->getNumRegs());
1099 UsedRegs.resize(TRI->getNumRegs());
1101 for (unsigned Count = 0; MBBI != B; --MBBI) {
1102 MachineInstr *MI = MBBI;
1103 // Skip DBG_VALUE instructions. Otherwise debug info can affect the
1104 // optimization by changing how far we scan.
1105 if (MI->isDebugValue())
1108 // Now that we know this is a real instruction, count it.
1111 // If we found a match, return it.
1112 if (isMatchingUpdateInsn(I, MI, BaseReg, Offset))
1115 // Update the status of what the instruction clobbered and used.
1116 trackRegDefsUses(MI, ModifiedRegs, UsedRegs, TRI);
1118 // Otherwise, if the base register is used or modified, we have no match, so
1120 if (ModifiedRegs[BaseReg] || UsedRegs[BaseReg])
1126 bool AArch64LoadStoreOpt::tryToMergeLdStInst(
1127 MachineBasicBlock::iterator &MBBI) {
1128 MachineInstr *MI = MBBI;
1129 MachineBasicBlock::iterator E = MI->getParent()->end();
1130 // If this is a volatile load/store, don't mess with it.
1131 if (MI->hasOrderedMemoryRef())
1134 // Make sure this is a reg+imm (as opposed to an address reloc).
1135 if (!getLdStOffsetOp(MI).isImm())
1138 // Check if this load/store has a hint to avoid pair formation.
1139 // MachineMemOperands hints are set by the AArch64StorePairSuppress pass.
1140 if (TII->isLdStPairSuppressed(MI))
1143 // Look ahead up to ScanLimit instructions for a pairable instruction.
1144 LdStPairFlags Flags;
1145 MachineBasicBlock::iterator Paired = findMatchingInsn(MBBI, Flags, ScanLimit);
1147 if (isSmallTypeLdMerge(MI)) {
1148 ++NumSmallTypeMerged;
1151 if (isUnscaledLdSt(MI))
1152 ++NumUnscaledPairCreated;
1155 // Merge the loads into a pair. Keeping the iterator straight is a
1156 // pain, so we let the merge routine tell us what the next instruction
1157 // is after it's done mucking about.
1158 MBBI = mergePairedInsns(MBBI, Paired, Flags);
1164 bool AArch64LoadStoreOpt::optimizeBlock(MachineBasicBlock &MBB) {
1165 bool Modified = false;
1166 // Three tranformations to do here:
1167 // 1) Find halfword loads that can be merged into a single 32-bit word load
1168 // with bitfield extract instructions.
1171 // ldrh w1, [x2, #2]
1174 // ubfx w1, w0, #16, #16
1175 // and w0, w0, #ffff
1176 // 2) Find loads and stores that can be merged into a single load or store
1177 // pair instruction.
1183 // 3) Find base register updates that can be merged into the load or store
1184 // as a base-reg writeback.
1191 for (MachineBasicBlock::iterator MBBI = MBB.begin(), E = MBB.end();
1192 !IsStrictAlign && MBBI != E;) {
1193 MachineInstr *MI = MBBI;
1194 switch (MI->getOpcode()) {
1196 // Just move on to the next instruction.
1199 // Scaled instructions.
1200 case AArch64::LDRHHui:
1201 // Unscaled instructions.
1202 case AArch64::LDURHHi: {
1203 if (tryToMergeLdStInst(MBBI)) {
1210 // FIXME: Do the other instructions.
1214 for (MachineBasicBlock::iterator MBBI = MBB.begin(), E = MBB.end();
1216 MachineInstr *MI = MBBI;
1217 switch (MI->getOpcode()) {
1219 // Just move on to the next instruction.
1222 // Scaled instructions.
1223 case AArch64::STRSui:
1224 case AArch64::STRDui:
1225 case AArch64::STRQui:
1226 case AArch64::STRXui:
1227 case AArch64::STRWui:
1228 case AArch64::LDRSui:
1229 case AArch64::LDRDui:
1230 case AArch64::LDRQui:
1231 case AArch64::LDRXui:
1232 case AArch64::LDRWui:
1233 case AArch64::LDRSWui:
1234 // Unscaled instructions.
1235 case AArch64::STURSi:
1236 case AArch64::STURDi:
1237 case AArch64::STURQi:
1238 case AArch64::STURWi:
1239 case AArch64::STURXi:
1240 case AArch64::LDURSi:
1241 case AArch64::LDURDi:
1242 case AArch64::LDURQi:
1243 case AArch64::LDURWi:
1244 case AArch64::LDURXi:
1245 case AArch64::LDURSWi: {
1246 if (tryToMergeLdStInst(MBBI)) {
1253 // FIXME: Do the other instructions.
1257 for (MachineBasicBlock::iterator MBBI = MBB.begin(), E = MBB.end();
1259 MachineInstr *MI = MBBI;
1260 // Do update merging. It's simpler to keep this separate from the above
1261 // switch, though not strictly necessary.
1262 unsigned Opc = MI->getOpcode();
1265 // Just move on to the next instruction.
1268 // Scaled instructions.
1269 case AArch64::STRSui:
1270 case AArch64::STRDui:
1271 case AArch64::STRQui:
1272 case AArch64::STRXui:
1273 case AArch64::STRWui:
1274 case AArch64::STRHHui:
1275 case AArch64::STRBBui:
1276 case AArch64::LDRSui:
1277 case AArch64::LDRDui:
1278 case AArch64::LDRQui:
1279 case AArch64::LDRXui:
1280 case AArch64::LDRWui:
1281 case AArch64::LDRHHui:
1282 case AArch64::LDRBBui:
1283 // Unscaled instructions.
1284 case AArch64::STURSi:
1285 case AArch64::STURDi:
1286 case AArch64::STURQi:
1287 case AArch64::STURWi:
1288 case AArch64::STURXi:
1289 case AArch64::LDURSi:
1290 case AArch64::LDURDi:
1291 case AArch64::LDURQi:
1292 case AArch64::LDURWi:
1293 case AArch64::LDURXi:
1294 // Paired instructions.
1295 case AArch64::LDPSi:
1296 case AArch64::LDPSWi:
1297 case AArch64::LDPDi:
1298 case AArch64::LDPQi:
1299 case AArch64::LDPWi:
1300 case AArch64::LDPXi:
1301 case AArch64::STPSi:
1302 case AArch64::STPDi:
1303 case AArch64::STPQi:
1304 case AArch64::STPWi:
1305 case AArch64::STPXi: {
1306 // Make sure this is a reg+imm (as opposed to an address reloc).
1307 if (!getLdStOffsetOp(MI).isImm()) {
1311 // Look forward to try to form a post-index instruction. For example,
1313 // add x20, x20, #32
1315 // ldr x0, [x20], #32
1316 MachineBasicBlock::iterator Update =
1317 findMatchingUpdateInsnForward(MBBI, ScanLimit, 0);
1319 // Merge the update into the ld/st.
1320 MBBI = mergeUpdateInsn(MBBI, Update, /*IsPreIdx=*/false);
1325 // Don't know how to handle pre/post-index versions, so move to the next
1327 if (isUnscaledLdSt(Opc)) {
1332 // Look back to try to find a pre-index instruction. For example,
1336 // ldr x1, [x0, #8]!
1337 Update = findMatchingUpdateInsnBackward(MBBI, ScanLimit);
1339 // Merge the update into the ld/st.
1340 MBBI = mergeUpdateInsn(MBBI, Update, /*IsPreIdx=*/true);
1345 // The immediate in the load/store is scaled by the size of the memory
1346 // operation. The immediate in the add we're looking for,
1347 // however, is not, so adjust here.
1348 int UnscaledOffset = getLdStOffsetOp(MI).getImm() * getMemScale(MI);
1350 // Look forward to try to find a post-index instruction. For example,
1351 // ldr x1, [x0, #64]
1354 // ldr x1, [x0, #64]!
1355 Update = findMatchingUpdateInsnForward(MBBI, ScanLimit, UnscaledOffset);
1357 // Merge the update into the ld/st.
1358 MBBI = mergeUpdateInsn(MBBI, Update, /*IsPreIdx=*/true);
1364 // Nothing found. Just move to the next instruction.
1368 // FIXME: Do the other instructions.
1375 bool AArch64LoadStoreOpt::runOnMachineFunction(MachineFunction &Fn) {
1376 TII = static_cast<const AArch64InstrInfo *>(Fn.getSubtarget().getInstrInfo());
1377 TRI = Fn.getSubtarget().getRegisterInfo();
1378 IsStrictAlign = (static_cast<const AArch64Subtarget &>(Fn.getSubtarget()))
1379 .requiresStrictAlign();
1381 bool Modified = false;
1382 for (auto &MBB : Fn)
1383 Modified |= optimizeBlock(MBB);
1388 // FIXME: Do we need/want a pre-alloc pass like ARM has to try to keep
1389 // loads and stores near one another?
1391 /// createAArch64LoadStoreOptimizationPass - returns an instance of the
1392 /// load / store optimization pass.
1393 FunctionPass *llvm::createAArch64LoadStoreOptimizationPass() {
1394 return new AArch64LoadStoreOpt();