[Alignment][NFC] Use Align with TargetLowering::setMinFunctionAlignment
[llvm-core.git] / include / llvm / CodeGen / GlobalISel / RegBankSelect.h
blobd9d076ba312c0d15b4e6e567ce5fee15f8352b27
1 //=- llvm/CodeGen/GlobalISel/RegBankSelect.h - Reg Bank Selector --*- C++ -*-=//
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 This file describes the interface of the MachineFunctionPass
10 /// responsible for assigning the generic virtual registers to register bank.
12 /// By default, the reg bank selector relies on local decisions to
13 /// assign the register bank. In other words, it looks at one instruction
14 /// at a time to decide where the operand of that instruction should live.
15 ///
16 /// At higher optimization level, we could imagine that the reg bank selector
17 /// would use more global analysis and do crazier thing like duplicating
18 /// instructions and so on. This is future work.
19 ///
20 /// For now, the pass uses a greedy algorithm to decide where the operand
21 /// of an instruction should live. It asks the target which banks may be
22 /// used for each operand of the instruction and what is the cost. Then,
23 /// it chooses the solution which minimize the cost of the instruction plus
24 /// the cost of any move that may be needed to the values into the right
25 /// register bank.
26 /// In other words, the cost for an instruction on a register bank RegBank
27 /// is: Cost of I on RegBank plus the sum of the cost for bringing the
28 /// input operands from their current register bank to RegBank.
29 /// Thus, the following formula:
30 /// cost(I, RegBank) = cost(I.Opcode, RegBank) +
31 /// sum(for each arg in I.arguments: costCrossCopy(arg.RegBank, RegBank))
32 ///
33 /// E.g., Let say we are assigning the register bank for the instruction
34 /// defining v2.
35 /// v0(A_REGBANK) = ...
36 /// v1(A_REGBANK) = ...
37 /// v2 = G_ADD i32 v0, v1 <-- MI
38 ///
39 /// The target may say it can generate G_ADD i32 on register bank A and B
40 /// with a cost of respectively 5 and 1.
41 /// Then, let say the cost of a cross register bank copies from A to B is 1.
42 /// The reg bank selector would compare the following two costs:
43 /// cost(MI, A_REGBANK) = cost(G_ADD, A_REGBANK) + cost(v0.RegBank, A_REGBANK) +
44 /// cost(v1.RegBank, A_REGBANK)
45 /// = 5 + cost(A_REGBANK, A_REGBANK) + cost(A_REGBANK,
46 /// A_REGBANK)
47 /// = 5 + 0 + 0 = 5
48 /// cost(MI, B_REGBANK) = cost(G_ADD, B_REGBANK) + cost(v0.RegBank, B_REGBANK) +
49 /// cost(v1.RegBank, B_REGBANK)
50 /// = 1 + cost(A_REGBANK, B_REGBANK) + cost(A_REGBANK,
51 /// B_REGBANK)
52 /// = 1 + 1 + 1 = 3
53 /// Therefore, in this specific example, the reg bank selector would choose
54 /// bank B for MI.
55 /// v0(A_REGBANK) = ...
56 /// v1(A_REGBANK) = ...
57 /// tmp0(B_REGBANK) = COPY v0
58 /// tmp1(B_REGBANK) = COPY v1
59 /// v2(B_REGBANK) = G_ADD i32 tmp0, tmp1
61 //===----------------------------------------------------------------------===//
63 #ifndef LLVM_CODEGEN_GLOBALISEL_REGBANKSELECT_H
64 #define LLVM_CODEGEN_GLOBALISEL_REGBANKSELECT_H
66 #include "llvm/ADT/SmallVector.h"
67 #include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
68 #include "llvm/CodeGen/GlobalISel/RegisterBankInfo.h"
69 #include "llvm/CodeGen/MachineBasicBlock.h"
70 #include "llvm/CodeGen/MachineFunctionPass.h"
71 #include "llvm/CodeGen/MachineOptimizationRemarkEmitter.h"
72 #include <cassert>
73 #include <cstdint>
74 #include <memory>
76 namespace llvm {
78 class BlockFrequency;
79 class MachineBlockFrequencyInfo;
80 class MachineBranchProbabilityInfo;
81 class MachineOperand;
82 class MachineRegisterInfo;
83 class Pass;
84 class raw_ostream;
85 class TargetPassConfig;
86 class TargetRegisterInfo;
88 /// This pass implements the reg bank selector pass used in the GlobalISel
89 /// pipeline. At the end of this pass, all register operands have been assigned
90 class RegBankSelect : public MachineFunctionPass {
91 public:
92 static char ID;
94 /// List of the modes supported by the RegBankSelect pass.
95 enum Mode {
96 /// Assign the register banks as fast as possible (default).
97 Fast,
98 /// Greedily minimize the cost of assigning register banks.
99 /// This should produce code of greater quality, but will
100 /// require more compile time.
101 Greedy
104 /// Abstract class used to represent an insertion point in a CFG.
105 /// This class records an insertion point and materializes it on
106 /// demand.
107 /// It allows to reason about the frequency of this insertion point,
108 /// without having to logically materialize it (e.g., on an edge),
109 /// before we actually need to insert something.
110 class InsertPoint {
111 protected:
112 /// Tell if the insert point has already been materialized.
113 bool WasMaterialized = false;
115 /// Materialize the insertion point.
117 /// If isSplit() is true, this involves actually splitting
118 /// the block or edge.
120 /// \post getPointImpl() returns a valid iterator.
121 /// \post getInsertMBBImpl() returns a valid basic block.
122 /// \post isSplit() == false ; no more splitting should be required.
123 virtual void materialize() = 0;
125 /// Return the materialized insertion basic block.
126 /// Code will be inserted into that basic block.
128 /// \pre ::materialize has been called.
129 virtual MachineBasicBlock &getInsertMBBImpl() = 0;
131 /// Return the materialized insertion point.
132 /// Code will be inserted before that point.
134 /// \pre ::materialize has been called.
135 virtual MachineBasicBlock::iterator getPointImpl() = 0;
137 public:
138 virtual ~InsertPoint() = default;
140 /// The first call to this method will cause the splitting to
141 /// happen if need be, then sub sequent calls just return
142 /// the iterator to that point. I.e., no more splitting will
143 /// occur.
145 /// \return The iterator that should be used with
146 /// MachineBasicBlock::insert. I.e., additional code happens
147 /// before that point.
148 MachineBasicBlock::iterator getPoint() {
149 if (!WasMaterialized) {
150 WasMaterialized = true;
151 assert(canMaterialize() && "Impossible to materialize this point");
152 materialize();
154 // When we materialized the point we should have done the splitting.
155 assert(!isSplit() && "Wrong pre-condition");
156 return getPointImpl();
159 /// The first call to this method will cause the splitting to
160 /// happen if need be, then sub sequent calls just return
161 /// the basic block that contains the insertion point.
162 /// I.e., no more splitting will occur.
164 /// \return The basic block should be used with
165 /// MachineBasicBlock::insert and ::getPoint. The new code should
166 /// happen before that point.
167 MachineBasicBlock &getInsertMBB() {
168 if (!WasMaterialized) {
169 WasMaterialized = true;
170 assert(canMaterialize() && "Impossible to materialize this point");
171 materialize();
173 // When we materialized the point we should have done the splitting.
174 assert(!isSplit() && "Wrong pre-condition");
175 return getInsertMBBImpl();
178 /// Insert \p MI in the just before ::getPoint()
179 MachineBasicBlock::iterator insert(MachineInstr &MI) {
180 return getInsertMBB().insert(getPoint(), &MI);
183 /// Does this point involve splitting an edge or block?
184 /// As soon as ::getPoint is called and thus, the point
185 /// materialized, the point will not require splitting anymore,
186 /// i.e., this will return false.
187 virtual bool isSplit() const { return false; }
189 /// Frequency of the insertion point.
190 /// \p P is used to access the various analysis that will help to
191 /// get that information, like MachineBlockFrequencyInfo. If \p P
192 /// does not contain enough enough to return the actual frequency,
193 /// this returns 1.
194 virtual uint64_t frequency(const Pass &P) const { return 1; }
196 /// Check whether this insertion point can be materialized.
197 /// As soon as ::getPoint is called and thus, the point materialized
198 /// calling this method does not make sense.
199 virtual bool canMaterialize() const { return false; }
202 /// Insertion point before or after an instruction.
203 class InstrInsertPoint : public InsertPoint {
204 private:
205 /// Insertion point.
206 MachineInstr &Instr;
208 /// Does the insertion point is before or after Instr.
209 bool Before;
211 void materialize() override;
213 MachineBasicBlock::iterator getPointImpl() override {
214 if (Before)
215 return Instr;
216 return Instr.getNextNode() ? *Instr.getNextNode()
217 : Instr.getParent()->end();
220 MachineBasicBlock &getInsertMBBImpl() override {
221 return *Instr.getParent();
224 public:
225 /// Create an insertion point before (\p Before=true) or after \p Instr.
226 InstrInsertPoint(MachineInstr &Instr, bool Before = true);
228 bool isSplit() const override;
229 uint64_t frequency(const Pass &P) const override;
231 // Worst case, we need to slice the basic block, but that is still doable.
232 bool canMaterialize() const override { return true; }
235 /// Insertion point at the beginning or end of a basic block.
236 class MBBInsertPoint : public InsertPoint {
237 private:
238 /// Insertion point.
239 MachineBasicBlock &MBB;
241 /// Does the insertion point is at the beginning or end of MBB.
242 bool Beginning;
244 void materialize() override { /*Nothing to do to materialize*/
247 MachineBasicBlock::iterator getPointImpl() override {
248 return Beginning ? MBB.begin() : MBB.end();
251 MachineBasicBlock &getInsertMBBImpl() override { return MBB; }
253 public:
254 MBBInsertPoint(MachineBasicBlock &MBB, bool Beginning = true)
255 : InsertPoint(), MBB(MBB), Beginning(Beginning) {
256 // If we try to insert before phis, we should use the insertion
257 // points on the incoming edges.
258 assert((!Beginning || MBB.getFirstNonPHI() == MBB.begin()) &&
259 "Invalid beginning point");
260 // If we try to insert after the terminators, we should use the
261 // points on the outcoming edges.
262 assert((Beginning || MBB.getFirstTerminator() == MBB.end()) &&
263 "Invalid end point");
266 bool isSplit() const override { return false; }
267 uint64_t frequency(const Pass &P) const override;
268 bool canMaterialize() const override { return true; };
271 /// Insertion point on an edge.
272 class EdgeInsertPoint : public InsertPoint {
273 private:
274 /// Source of the edge.
275 MachineBasicBlock &Src;
277 /// Destination of the edge.
278 /// After the materialization is done, this hold the basic block
279 /// that resulted from the splitting.
280 MachineBasicBlock *DstOrSplit;
282 /// P is used to update the analysis passes as applicable.
283 Pass &P;
285 void materialize() override;
287 MachineBasicBlock::iterator getPointImpl() override {
288 // DstOrSplit should be the Split block at this point.
289 // I.e., it should have one predecessor, Src, and one successor,
290 // the original Dst.
291 assert(DstOrSplit && DstOrSplit->isPredecessor(&Src) &&
292 DstOrSplit->pred_size() == 1 && DstOrSplit->succ_size() == 1 &&
293 "Did not split?!");
294 return DstOrSplit->begin();
297 MachineBasicBlock &getInsertMBBImpl() override { return *DstOrSplit; }
299 public:
300 EdgeInsertPoint(MachineBasicBlock &Src, MachineBasicBlock &Dst, Pass &P)
301 : InsertPoint(), Src(Src), DstOrSplit(&Dst), P(P) {}
303 bool isSplit() const override {
304 return Src.succ_size() > 1 && DstOrSplit->pred_size() > 1;
307 uint64_t frequency(const Pass &P) const override;
308 bool canMaterialize() const override;
311 /// Struct used to represent the placement of a repairing point for
312 /// a given operand.
313 class RepairingPlacement {
314 public:
315 /// Define the kind of action this repairing needs.
316 enum RepairingKind {
317 /// Nothing to repair, just drop this action.
318 None,
319 /// Reparing code needs to happen before InsertPoints.
320 Insert,
321 /// (Re)assign the register bank of the operand.
322 Reassign,
323 /// Mark this repairing placement as impossible.
324 Impossible
327 /// \name Convenient types for a list of insertion points.
328 /// @{
329 using InsertionPoints = SmallVector<std::unique_ptr<InsertPoint>, 2>;
330 using insertpt_iterator = InsertionPoints::iterator;
331 using const_insertpt_iterator = InsertionPoints::const_iterator;
332 /// @}
334 private:
335 /// Kind of repairing.
336 RepairingKind Kind;
337 /// Index of the operand that will be repaired.
338 unsigned OpIdx;
339 /// Are all the insert points materializeable?
340 bool CanMaterialize;
341 /// Is there any of the insert points needing splitting?
342 bool HasSplit = false;
343 /// Insertion point for the repair code.
344 /// The repairing code needs to happen just before these points.
345 InsertionPoints InsertPoints;
346 /// Some insertion points may need to update the liveness and such.
347 Pass &P;
349 public:
350 /// Create a repairing placement for the \p OpIdx-th operand of
351 /// \p MI. \p TRI is used to make some checks on the register aliases
352 /// if the machine operand is a physical register. \p P is used to
353 /// to update liveness information and such when materializing the
354 /// points.
355 RepairingPlacement(MachineInstr &MI, unsigned OpIdx,
356 const TargetRegisterInfo &TRI, Pass &P,
357 RepairingKind Kind = RepairingKind::Insert);
359 /// \name Getters.
360 /// @{
361 RepairingKind getKind() const { return Kind; }
362 unsigned getOpIdx() const { return OpIdx; }
363 bool canMaterialize() const { return CanMaterialize; }
364 bool hasSplit() { return HasSplit; }
365 /// @}
367 /// \name Overloaded methods to add an insertion point.
368 /// @{
369 /// Add a MBBInsertionPoint to the list of InsertPoints.
370 void addInsertPoint(MachineBasicBlock &MBB, bool Beginning);
371 /// Add a InstrInsertionPoint to the list of InsertPoints.
372 void addInsertPoint(MachineInstr &MI, bool Before);
373 /// Add an EdgeInsertionPoint (\p Src, \p Dst) to the list of InsertPoints.
374 void addInsertPoint(MachineBasicBlock &Src, MachineBasicBlock &Dst);
375 /// Add an InsertPoint to the list of insert points.
376 /// This method takes the ownership of &\p Point.
377 void addInsertPoint(InsertPoint &Point);
378 /// @}
380 /// \name Accessors related to the insertion points.
381 /// @{
382 insertpt_iterator begin() { return InsertPoints.begin(); }
383 insertpt_iterator end() { return InsertPoints.end(); }
385 const_insertpt_iterator begin() const { return InsertPoints.begin(); }
386 const_insertpt_iterator end() const { return InsertPoints.end(); }
388 unsigned getNumInsertPoints() const { return InsertPoints.size(); }
389 /// @}
391 /// Change the type of this repairing placement to \p NewKind.
392 /// It is not possible to switch a repairing placement to the
393 /// RepairingKind::Insert. There is no fundamental problem with
394 /// that, but no uses as well, so do not support it for now.
396 /// \pre NewKind != RepairingKind::Insert
397 /// \post getKind() == NewKind
398 void switchTo(RepairingKind NewKind) {
399 assert(NewKind != Kind && "Already of the right Kind");
400 Kind = NewKind;
401 InsertPoints.clear();
402 CanMaterialize = NewKind != RepairingKind::Impossible;
403 HasSplit = false;
404 assert(NewKind != RepairingKind::Insert &&
405 "We would need more MI to switch to Insert");
409 private:
410 /// Helper class used to represent the cost for mapping an instruction.
411 /// When mapping an instruction, we may introduce some repairing code.
412 /// In most cases, the repairing code is local to the instruction,
413 /// thus, we can omit the basic block frequency from the cost.
414 /// However, some alternatives may produce non-local cost, e.g., when
415 /// repairing a phi, and thus we then need to scale the local cost
416 /// to the non-local cost. This class does this for us.
417 /// \note: We could simply always scale the cost. The problem is that
418 /// there are higher chances that we saturate the cost easier and end
419 /// up having the same cost for actually different alternatives.
420 /// Another option would be to use APInt everywhere.
421 class MappingCost {
422 private:
423 /// Cost of the local instructions.
424 /// This cost is free of basic block frequency.
425 uint64_t LocalCost = 0;
426 /// Cost of the non-local instructions.
427 /// This cost should include the frequency of the related blocks.
428 uint64_t NonLocalCost = 0;
429 /// Frequency of the block where the local instructions live.
430 uint64_t LocalFreq;
432 MappingCost(uint64_t LocalCost, uint64_t NonLocalCost, uint64_t LocalFreq)
433 : LocalCost(LocalCost), NonLocalCost(NonLocalCost),
434 LocalFreq(LocalFreq) {}
436 /// Check if this cost is saturated.
437 bool isSaturated() const;
439 public:
440 /// Create a MappingCost assuming that most of the instructions
441 /// will occur in a basic block with \p LocalFreq frequency.
442 MappingCost(const BlockFrequency &LocalFreq);
444 /// Add \p Cost to the local cost.
445 /// \return true if this cost is saturated, false otherwise.
446 bool addLocalCost(uint64_t Cost);
448 /// Add \p Cost to the non-local cost.
449 /// Non-local cost should reflect the frequency of their placement.
450 /// \return true if this cost is saturated, false otherwise.
451 bool addNonLocalCost(uint64_t Cost);
453 /// Saturate the cost to the maximal representable value.
454 void saturate();
456 /// Return an instance of MappingCost that represents an
457 /// impossible mapping.
458 static MappingCost ImpossibleCost();
460 /// Check if this is less than \p Cost.
461 bool operator<(const MappingCost &Cost) const;
462 /// Check if this is equal to \p Cost.
463 bool operator==(const MappingCost &Cost) const;
464 /// Check if this is not equal to \p Cost.
465 bool operator!=(const MappingCost &Cost) const { return !(*this == Cost); }
466 /// Check if this is greater than \p Cost.
467 bool operator>(const MappingCost &Cost) const {
468 return *this != Cost && Cost < *this;
471 /// Print this on dbgs() stream.
472 void dump() const;
474 /// Print this on \p OS;
475 void print(raw_ostream &OS) const;
477 /// Overload the stream operator for easy debug printing.
478 friend raw_ostream &operator<<(raw_ostream &OS, const MappingCost &Cost) {
479 Cost.print(OS);
480 return OS;
484 /// Interface to the target lowering info related
485 /// to register banks.
486 const RegisterBankInfo *RBI = nullptr;
488 /// MRI contains all the register class/bank information that this
489 /// pass uses and updates.
490 MachineRegisterInfo *MRI = nullptr;
492 /// Information on the register classes for the current function.
493 const TargetRegisterInfo *TRI = nullptr;
495 /// Get the frequency of blocks.
496 /// This is required for non-fast mode.
497 MachineBlockFrequencyInfo *MBFI = nullptr;
499 /// Get the frequency of the edges.
500 /// This is required for non-fast mode.
501 MachineBranchProbabilityInfo *MBPI = nullptr;
503 /// Current optimization remark emitter. Used to report failures.
504 std::unique_ptr<MachineOptimizationRemarkEmitter> MORE;
506 /// Helper class used for every code morphing.
507 MachineIRBuilder MIRBuilder;
509 /// Optimization mode of the pass.
510 Mode OptMode;
512 /// Current target configuration. Controls how the pass handles errors.
513 const TargetPassConfig *TPC;
515 /// Assign the register bank of each operand of \p MI.
516 /// \return True on success, false otherwise.
517 bool assignInstr(MachineInstr &MI);
519 /// Initialize the field members using \p MF.
520 void init(MachineFunction &MF);
522 /// Check if \p Reg is already assigned what is described by \p ValMapping.
523 /// \p OnlyAssign == true means that \p Reg just needs to be assigned a
524 /// register bank. I.e., no repairing is necessary to have the
525 /// assignment match.
526 bool assignmentMatch(Register Reg,
527 const RegisterBankInfo::ValueMapping &ValMapping,
528 bool &OnlyAssign) const;
530 /// Insert repairing code for \p Reg as specified by \p ValMapping.
531 /// The repairing placement is specified by \p RepairPt.
532 /// \p NewVRegs contains all the registers required to remap \p Reg.
533 /// In other words, the number of registers in NewVRegs must be equal
534 /// to ValMapping.BreakDown.size().
536 /// The transformation could be sketched as:
537 /// \code
538 /// ... = op Reg
539 /// \endcode
540 /// Becomes
541 /// \code
542 /// <NewRegs> = COPY or extract Reg
543 /// ... = op Reg
544 /// \endcode
546 /// and
547 /// \code
548 /// Reg = op ...
549 /// \endcode
550 /// Becomes
551 /// \code
552 /// Reg = op ...
553 /// Reg = COPY or build_sequence <NewRegs>
554 /// \endcode
556 /// \pre NewVRegs.size() == ValMapping.BreakDown.size()
558 /// \note The caller is supposed to do the rewriting of op if need be.
559 /// I.e., Reg = op ... => <NewRegs> = NewOp ...
561 /// \return True if the repairing worked, false otherwise.
562 bool repairReg(MachineOperand &MO,
563 const RegisterBankInfo::ValueMapping &ValMapping,
564 RegBankSelect::RepairingPlacement &RepairPt,
565 const iterator_range<SmallVectorImpl<Register>::const_iterator>
566 &NewVRegs);
568 /// Return the cost of the instruction needed to map \p MO to \p ValMapping.
569 /// The cost is free of basic block frequencies.
570 /// \pre MO.isReg()
571 /// \pre MO is assigned to a register bank.
572 /// \pre ValMapping is a valid mapping for MO.
573 uint64_t
574 getRepairCost(const MachineOperand &MO,
575 const RegisterBankInfo::ValueMapping &ValMapping) const;
577 /// Find the best mapping for \p MI from \p PossibleMappings.
578 /// \return a reference on the best mapping in \p PossibleMappings.
579 const RegisterBankInfo::InstructionMapping &
580 findBestMapping(MachineInstr &MI,
581 RegisterBankInfo::InstructionMappings &PossibleMappings,
582 SmallVectorImpl<RepairingPlacement> &RepairPts);
584 /// Compute the cost of mapping \p MI with \p InstrMapping and
585 /// compute the repairing placement for such mapping in \p
586 /// RepairPts.
587 /// \p BestCost is used to specify when the cost becomes too high
588 /// and thus it is not worth computing the RepairPts. Moreover if
589 /// \p BestCost == nullptr, the mapping cost is actually not
590 /// computed.
591 MappingCost
592 computeMapping(MachineInstr &MI,
593 const RegisterBankInfo::InstructionMapping &InstrMapping,
594 SmallVectorImpl<RepairingPlacement> &RepairPts,
595 const MappingCost *BestCost = nullptr);
597 /// When \p RepairPt involves splitting to repair \p MO for the
598 /// given \p ValMapping, try to change the way we repair such that
599 /// the splitting is not required anymore.
601 /// \pre \p RepairPt.hasSplit()
602 /// \pre \p MO == MO.getParent()->getOperand(\p RepairPt.getOpIdx())
603 /// \pre \p ValMapping is the mapping of \p MO for MO.getParent()
604 /// that implied \p RepairPt.
605 void tryAvoidingSplit(RegBankSelect::RepairingPlacement &RepairPt,
606 const MachineOperand &MO,
607 const RegisterBankInfo::ValueMapping &ValMapping) const;
609 /// Apply \p Mapping to \p MI. \p RepairPts represents the different
610 /// mapping action that need to happen for the mapping to be
611 /// applied.
612 /// \return True if the mapping was applied sucessfully, false otherwise.
613 bool applyMapping(MachineInstr &MI,
614 const RegisterBankInfo::InstructionMapping &InstrMapping,
615 SmallVectorImpl<RepairingPlacement> &RepairPts);
617 public:
618 /// Create a RegBankSelect pass with the specified \p RunningMode.
619 RegBankSelect(Mode RunningMode = Fast);
621 StringRef getPassName() const override { return "RegBankSelect"; }
623 void getAnalysisUsage(AnalysisUsage &AU) const override;
625 MachineFunctionProperties getRequiredProperties() const override {
626 return MachineFunctionProperties()
627 .set(MachineFunctionProperties::Property::IsSSA)
628 .set(MachineFunctionProperties::Property::Legalized);
631 MachineFunctionProperties getSetProperties() const override {
632 return MachineFunctionProperties().set(
633 MachineFunctionProperties::Property::RegBankSelected);
636 MachineFunctionProperties getClearedProperties() const override {
637 return MachineFunctionProperties()
638 .set(MachineFunctionProperties::Property::NoPHIs);
641 /// Walk through \p MF and assign a register bank to every virtual register
642 /// that are still mapped to nothing.
643 /// The target needs to provide a RegisterBankInfo and in particular
644 /// override RegisterBankInfo::getInstrMapping.
646 /// Simplified algo:
647 /// \code
648 /// RBI = MF.subtarget.getRegBankInfo()
649 /// MIRBuilder.setMF(MF)
650 /// for each bb in MF
651 /// for each inst in bb
652 /// MIRBuilder.setInstr(inst)
653 /// MappingCosts = RBI.getMapping(inst);
654 /// Idx = findIdxOfMinCost(MappingCosts)
655 /// CurRegBank = MappingCosts[Idx].RegBank
656 /// MRI.setRegBank(inst.getOperand(0).getReg(), CurRegBank)
657 /// for each argument in inst
658 /// if (CurRegBank != argument.RegBank)
659 /// ArgReg = argument.getReg()
660 /// Tmp = MRI.createNewVirtual(MRI.getSize(ArgReg), CurRegBank)
661 /// MIRBuilder.buildInstr(COPY, Tmp, ArgReg)
662 /// inst.getOperand(argument.getOperandNo()).setReg(Tmp)
663 /// \endcode
664 bool runOnMachineFunction(MachineFunction &MF) override;
667 } // end namespace llvm
669 #endif // LLVM_CODEGEN_GLOBALISEL_REGBANKSELECT_H