1 /*
2  * Copyright (C) 2014 The Android Open Source Project
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *      http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #ifndef ART_COMPILER_OPTIMIZING_PARALLEL_MOVE_RESOLVER_H_
18 #define ART_COMPILER_OPTIMIZING_PARALLEL_MOVE_RESOLVER_H_
19 
20 #include "base/arena_containers.h"
21 #include "base/value_object.h"
22 #include "data_type.h"
23 #include "locations.h"
24 
25 namespace art {
26 
27 class HParallelMove;
28 class MoveOperands;
29 
30 // Helper classes to resolve a set of parallel moves. Architecture dependent code generator must
31 // have their own subclass that implements corresponding virtual functions.
32 class ParallelMoveResolver : public ValueObject {
33  public:
ParallelMoveResolver(ArenaAllocator * allocator)34   explicit ParallelMoveResolver(ArenaAllocator* allocator)
35       : moves_(allocator->Adapter(kArenaAllocParallelMoveResolver)) {
36     moves_.reserve(32);
37   }
~ParallelMoveResolver()38   virtual ~ParallelMoveResolver() {}
39 
40   // Resolve a set of parallel moves, emitting assembler instructions.
41   virtual void EmitNativeCode(HParallelMove* parallel_move) = 0;
42 
43  protected:
44   // Build the initial list of moves.
45   void BuildInitialMoveList(HParallelMove* parallel_move);
46 
47   ArenaVector<MoveOperands*> moves_;
48 
49  private:
50   DISALLOW_COPY_AND_ASSIGN(ParallelMoveResolver);
51 };
52 
53 // This helper class uses swap to resolve dependencies and may emit swap.
54 class ParallelMoveResolverWithSwap : public ParallelMoveResolver {
55  public:
ParallelMoveResolverWithSwap(ArenaAllocator * allocator)56   explicit ParallelMoveResolverWithSwap(ArenaAllocator* allocator)
57       : ParallelMoveResolver(allocator) {}
~ParallelMoveResolverWithSwap()58   virtual ~ParallelMoveResolverWithSwap() {}
59 
60   // Resolve a set of parallel moves, emitting assembler instructions.
61   void EmitNativeCode(HParallelMove* parallel_move) override;
62 
63  protected:
64   class ScratchRegisterScope : public ValueObject {
65    public:
66     ScratchRegisterScope(ParallelMoveResolverWithSwap* resolver,
67                          int blocked,
68                          int if_scratch,
69                          int number_of_registers);
70     ~ScratchRegisterScope();
71 
GetRegister()72     int GetRegister() const { return reg_; }
IsSpilled()73     bool IsSpilled() const { return spilled_; }
74 
75    private:
76     ParallelMoveResolverWithSwap* resolver_;
77     int reg_;
78     bool spilled_;
79   };
80 
81   // Return true if the location can be scratched.
82   bool IsScratchLocation(Location loc);
83 
84   // Allocate a scratch register for performing a move. The method will try to use
85   // a register that is the destination of a move, but that move has not been emitted yet.
86   int AllocateScratchRegister(int blocked, int if_scratch, int register_count, bool* spilled);
87 
88   // Emit a move.
89   virtual void EmitMove(size_t index) = 0;
90 
91   // Execute a move by emitting a swap of two operands.
92   virtual void EmitSwap(size_t index) = 0;
93 
94   virtual void SpillScratch(int reg) = 0;
95   virtual void RestoreScratch(int reg) = 0;
96 
97   static constexpr int kNoRegister = -1;
98 
99  private:
100   // Perform the move at the moves_ index in question (possibly requiring
101   // other moves to satisfy dependencies).
102   //
103   // Return whether another move in the dependency cycle needs to swap. This
104   // is to handle 64bits swaps:
105   // 1) In the case of register pairs, where we want the pair to swap first to avoid
106   //    building pairs that are unexpected by the code generator. For example, if
107   //    we were to swap R1 with R2, we would need to update all locations using
108   //    R2 to R1. So a (R2,R3) pair register could become (R1,R3). We could make
109   //    the code generator understand such pairs, but it's easier and cleaner to
110   //    just not create such pairs and exchange pairs in priority.
111   // 2) Even when the architecture does not have pairs, we must handle 64bits swaps
112   //    first. Consider the case: (R0->R1) (R1->S) (S->R0), where 'S' is a single
113   //    stack slot. If we end up swapping S and R0, S will only contain the low bits
114   //    of R0. If R0->R1 is for a 64bits instruction, R1 will therefore not contain
115   //    the right value.
116   MoveOperands* PerformMove(size_t index);
117 
118   DISALLOW_COPY_AND_ASSIGN(ParallelMoveResolverWithSwap);
119 };
120 
121 // This helper class uses additional scratch registers to resolve dependencies. It supports all kind
122 // of dependency cycles and does not care about the register layout.
123 class ParallelMoveResolverNoSwap : public ParallelMoveResolver {
124  public:
ParallelMoveResolverNoSwap(ArenaAllocator * allocator)125   explicit ParallelMoveResolverNoSwap(ArenaAllocator* allocator)
126       : ParallelMoveResolver(allocator),
127         scratches_(allocator->Adapter(kArenaAllocParallelMoveResolver)),
128         pending_moves_(allocator->Adapter(kArenaAllocParallelMoveResolver)),
129         allocator_(allocator) {
130     scratches_.reserve(32);
131     pending_moves_.reserve(8);
132   }
~ParallelMoveResolverNoSwap()133   virtual ~ParallelMoveResolverNoSwap() {}
134 
135   // Resolve a set of parallel moves, emitting assembler instructions.
136   void EmitNativeCode(HParallelMove* parallel_move) override;
137 
138  protected:
139   // Called at the beginning of EmitNativeCode(). A subclass may put some architecture dependent
140   // initialization here.
141   virtual void PrepareForEmitNativeCode() = 0;
142 
143   // Called at the end of EmitNativeCode(). A subclass may put some architecture dependent cleanup
144   // here. All scratch locations will be removed after this call.
145   virtual void FinishEmitNativeCode() = 0;
146 
147   // Allocate a scratch location to perform a move from input kind of location. A subclass should
148   // implement this to get the best fit location. If there is no suitable physical register, it can
149   // also return a stack slot.
150   virtual Location AllocateScratchLocationFor(Location::Kind kind) = 0;
151 
152   // Called after a move which takes a scratch location as source. A subclass can defer the cleanup
153   // to FinishEmitNativeCode().
154   virtual void FreeScratchLocation(Location loc) = 0;
155 
156   // Emit a move.
157   virtual void EmitMove(size_t index) = 0;
158 
159   // Return a scratch location from the moves which exactly matches the kind.
160   // Return Location::NoLocation() if no matching scratch location can be found.
161   Location GetScratchLocation(Location::Kind kind);
162 
163   // Add a location to the scratch list which can be returned from GetScratchLocation() to resolve
164   // dependency cycles.
165   void AddScratchLocation(Location loc);
166 
167   // Remove a location from the scratch list.
168   void RemoveScratchLocation(Location loc);
169 
170   // List of scratch locations.
171   ArenaVector<Location> scratches_;
172 
173  private:
174   // Perform the move at the given index in `moves_` (possibly requiring other moves to satisfy
175   // dependencies).
176   void PerformMove(size_t index);
177 
178   void UpdateMoveSource(Location from, Location to);
179 
180   void AddPendingMove(Location source, Location destination, DataType::Type type);
181 
182   void DeletePendingMove(MoveOperands* move);
183 
184   // Find a move that may be unblocked after (loc -> XXX) is performed.
185   MoveOperands* GetUnblockedPendingMove(Location loc);
186 
187   // Return true if the location is blocked by outstanding moves.
188   bool IsBlockedByMoves(Location loc);
189 
190   // Return the number of pending moves.
191   size_t GetNumberOfPendingMoves();
192 
193   // Additional pending moves which might be added to resolve dependency cycle.
194   ArenaVector<MoveOperands*> pending_moves_;
195 
196   // Used to allocate pending MoveOperands.
197   ArenaAllocator* const allocator_;
198 
199   DISALLOW_COPY_AND_ASSIGN(ParallelMoveResolverNoSwap);
200 };
201 
202 }  // namespace art
203 
204 #endif  // ART_COMPILER_OPTIMIZING_PARALLEL_MOVE_RESOLVER_H_
205