1 /*
2  * Copyright (C) 2013 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 #include "mir_graph.h"
18 
19 #include <inttypes.h>
20 #include <queue>
21 #include <unistd.h>
22 
23 #include "base/bit_vector-inl.h"
24 #include "base/logging.h"
25 #include "base/stl_util.h"
26 #include "base/stringprintf.h"
27 #include "base/scoped_arena_containers.h"
28 #include "compiler_ir.h"
29 #include "dex_file-inl.h"
30 #include "dex_flags.h"
31 #include "dex_instruction-inl.h"
32 #include "driver/compiler_driver.h"
33 #include "driver/dex_compilation_unit.h"
34 #include "dex/quick/quick_compiler.h"
35 #include "leb128.h"
36 #include "pass_driver_me_post_opt.h"
37 #include "stack.h"
38 #include "utils.h"
39 
40 namespace art {
41 
42 #define MAX_PATTERN_LEN 5
43 
44 const char* MIRGraph::extended_mir_op_names_[kMirOpLast - kMirOpFirst] = {
45   "Phi",
46   "Copy",
47   "FusedCmplFloat",
48   "FusedCmpgFloat",
49   "FusedCmplDouble",
50   "FusedCmpgDouble",
51   "FusedCmpLong",
52   "Nop",
53   "OpNullCheck",
54   "OpRangeCheck",
55   "OpDivZeroCheck",
56   "Check",
57   "Select",
58   "ConstVector",
59   "MoveVector",
60   "PackedMultiply",
61   "PackedAddition",
62   "PackedSubtract",
63   "PackedShiftLeft",
64   "PackedSignedShiftRight",
65   "PackedUnsignedShiftRight",
66   "PackedAnd",
67   "PackedOr",
68   "PackedXor",
69   "PackedAddReduce",
70   "PackedReduce",
71   "PackedSet",
72   "ReserveVectorRegisters",
73   "ReturnVectorRegisters",
74   "MemBarrier",
75   "PackedArrayGet",
76   "PackedArrayPut",
77   "MaddInt",
78   "MsubInt",
79   "MaddLong",
80   "MsubLong",
81 };
82 
MIRGraph(CompilationUnit * cu,ArenaAllocator * arena)83 MIRGraph::MIRGraph(CompilationUnit* cu, ArenaAllocator* arena)
84     : reg_location_(nullptr),
85       block_id_map_(std::less<unsigned int>(), arena->Adapter()),
86       cu_(cu),
87       ssa_base_vregs_(arena->Adapter(kArenaAllocSSAToDalvikMap)),
88       ssa_subscripts_(arena->Adapter(kArenaAllocSSAToDalvikMap)),
89       vreg_to_ssa_map_(nullptr),
90       ssa_last_defs_(nullptr),
91       is_constant_v_(nullptr),
92       constant_values_(nullptr),
93       use_counts_(arena->Adapter()),
94       raw_use_counts_(arena->Adapter()),
95       num_reachable_blocks_(0),
96       max_num_reachable_blocks_(0),
97       dfs_orders_up_to_date_(false),
98       domination_up_to_date_(false),
99       mir_ssa_rep_up_to_date_(false),
100       topological_order_up_to_date_(false),
101       dfs_order_(arena->Adapter(kArenaAllocDfsPreOrder)),
102       dfs_post_order_(arena->Adapter(kArenaAllocDfsPostOrder)),
103       dom_post_order_traversal_(arena->Adapter(kArenaAllocDomPostOrder)),
104       topological_order_(arena->Adapter(kArenaAllocTopologicalSortOrder)),
105       topological_order_loop_ends_(arena->Adapter(kArenaAllocTopologicalSortOrder)),
106       topological_order_indexes_(arena->Adapter(kArenaAllocTopologicalSortOrder)),
107       topological_order_loop_head_stack_(arena->Adapter(kArenaAllocTopologicalSortOrder)),
108       max_nested_loops_(0u),
109       i_dom_list_(nullptr),
110       temp_scoped_alloc_(),
111       block_list_(arena->Adapter(kArenaAllocBBList)),
112       try_block_addr_(nullptr),
113       entry_block_(nullptr),
114       exit_block_(nullptr),
115       current_code_item_(nullptr),
116       m_units_(arena->Adapter()),
117       method_stack_(arena->Adapter()),
118       current_method_(kInvalidEntry),
119       current_offset_(kInvalidEntry),
120       def_count_(0),
121       opcode_count_(nullptr),
122       num_ssa_regs_(0),
123       extended_basic_blocks_(arena->Adapter()),
124       method_sreg_(0),
125       attributes_(METHOD_IS_LEAF),  // Start with leaf assumption, change on encountering invoke.
126       checkstats_(nullptr),
127       arena_(arena),
128       backward_branches_(0),
129       forward_branches_(0),
130       num_non_special_compiler_temps_(0),
131       max_available_special_compiler_temps_(1),  // We only need the method ptr as a special temp for now.
132       requested_backend_temp_(false),
133       compiler_temps_committed_(false),
134       punt_to_interpreter_(false),
135       merged_df_flags_(0u),
136       ifield_lowering_infos_(arena->Adapter(kArenaAllocLoweringInfo)),
137       sfield_lowering_infos_(arena->Adapter(kArenaAllocLoweringInfo)),
138       method_lowering_infos_(arena->Adapter(kArenaAllocLoweringInfo)),
139       suspend_checks_in_loops_(nullptr) {
140   memset(&temp_, 0, sizeof(temp_));
141   use_counts_.reserve(256);
142   raw_use_counts_.reserve(256);
143   block_list_.reserve(100);
144   try_block_addr_ = new (arena_) ArenaBitVector(arena_, 0, true /* expandable */);
145 
146 
147   if (cu_->instruction_set == kX86 || cu_->instruction_set == kX86_64) {
148     // X86 requires a temp to keep track of the method address.
149     // TODO For x86_64, addressing can be done with RIP. When that is implemented,
150     // this needs to be updated to reserve 0 temps for BE.
151     max_available_non_special_compiler_temps_ = cu_->target64 ? 2 : 1;
152     reserved_temps_for_backend_ = max_available_non_special_compiler_temps_;
153   } else {
154     // Other architectures do not have a known lower bound for non-special temps.
155     // We allow the update of the max to happen at BE initialization stage and simply set 0 for now.
156     max_available_non_special_compiler_temps_ = 0;
157     reserved_temps_for_backend_ = 0;
158   }
159 }
160 
~MIRGraph()161 MIRGraph::~MIRGraph() {
162   STLDeleteElements(&block_list_);
163   STLDeleteElements(&m_units_);
164 }
165 
166 /*
167  * Parse an instruction, return the length of the instruction
168  */
ParseInsn(const uint16_t * code_ptr,MIR::DecodedInstruction * decoded_instruction)169 int MIRGraph::ParseInsn(const uint16_t* code_ptr, MIR::DecodedInstruction* decoded_instruction) {
170   const Instruction* inst = Instruction::At(code_ptr);
171   decoded_instruction->opcode = inst->Opcode();
172   decoded_instruction->vA = inst->HasVRegA() ? inst->VRegA() : 0;
173   decoded_instruction->vB = inst->HasVRegB() ? inst->VRegB() : 0;
174   decoded_instruction->vB_wide = inst->HasWideVRegB() ? inst->WideVRegB() : 0;
175   decoded_instruction->vC = inst->HasVRegC() ?  inst->VRegC() : 0;
176   if (inst->HasVarArgs()) {
177     inst->GetVarArgs(decoded_instruction->arg);
178   }
179   return inst->SizeInCodeUnits();
180 }
181 
182 
183 /* Split an existing block from the specified code offset into two */
SplitBlock(DexOffset code_offset,BasicBlock * orig_block,BasicBlock ** immed_pred_block_p)184 BasicBlock* MIRGraph::SplitBlock(DexOffset code_offset,
185                                  BasicBlock* orig_block, BasicBlock** immed_pred_block_p) {
186   DCHECK_GT(code_offset, orig_block->start_offset);
187   MIR* insn = orig_block->first_mir_insn;
188   MIR* prev = nullptr;  // Will be set to instruction before split.
189   while (insn) {
190     if (insn->offset == code_offset) break;
191     prev = insn;
192     insn = insn->next;
193   }
194   if (insn == nullptr) {
195     LOG(FATAL) << "Break split failed";
196   }
197   // Now insn is at the instruction where we want to split, namely
198   // insn will be the first instruction of the "bottom" block.
199   // Similarly, prev will be the last instruction of the "top" block
200 
201   BasicBlock* bottom_block = CreateNewBB(kDalvikByteCode);
202 
203   bottom_block->start_offset = code_offset;
204   bottom_block->first_mir_insn = insn;
205   bottom_block->last_mir_insn = orig_block->last_mir_insn;
206 
207   /* If this block was terminated by a return, conditional branch or throw,
208    * the flag needs to go with the bottom block
209    */
210   bottom_block->terminated_by_return = orig_block->terminated_by_return;
211   orig_block->terminated_by_return = false;
212 
213   bottom_block->conditional_branch = orig_block->conditional_branch;
214   orig_block->conditional_branch = false;
215 
216   bottom_block->explicit_throw = orig_block->explicit_throw;
217   orig_block->explicit_throw = false;
218 
219   /* Handle the taken path */
220   bottom_block->taken = orig_block->taken;
221   if (bottom_block->taken != NullBasicBlockId) {
222     orig_block->taken = NullBasicBlockId;
223     BasicBlock* bb_taken = GetBasicBlock(bottom_block->taken);
224     bb_taken->ErasePredecessor(orig_block->id);
225     bb_taken->predecessors.push_back(bottom_block->id);
226   }
227 
228   /* Handle the fallthrough path */
229   bottom_block->fall_through = orig_block->fall_through;
230   orig_block->fall_through = bottom_block->id;
231   bottom_block->predecessors.push_back(orig_block->id);
232   if (bottom_block->fall_through != NullBasicBlockId) {
233     BasicBlock* bb_fall_through = GetBasicBlock(bottom_block->fall_through);
234     bb_fall_through->ErasePredecessor(orig_block->id);
235     bb_fall_through->predecessors.push_back(bottom_block->id);
236   }
237 
238   /* Handle the successor list */
239   if (orig_block->successor_block_list_type != kNotUsed) {
240     bottom_block->successor_block_list_type = orig_block->successor_block_list_type;
241     bottom_block->successor_blocks.swap(orig_block->successor_blocks);
242     orig_block->successor_block_list_type = kNotUsed;
243     DCHECK(orig_block->successor_blocks.empty());  // Empty after the swap() above.
244     for (SuccessorBlockInfo* successor_block_info : bottom_block->successor_blocks) {
245       BasicBlock* bb = GetBasicBlock(successor_block_info->block);
246       if (bb != nullptr) {
247         bb->ErasePredecessor(orig_block->id);
248         bb->predecessors.push_back(bottom_block->id);
249       }
250     }
251   }
252 
253   orig_block->last_mir_insn = prev;
254   prev->next = nullptr;
255 
256   /*
257    * Update the immediate predecessor block pointer so that outgoing edges
258    * can be applied to the proper block.
259    */
260   if (immed_pred_block_p) {
261     DCHECK_EQ(*immed_pred_block_p, orig_block);
262     *immed_pred_block_p = bottom_block;
263   }
264 
265   // Associate dex instructions in the bottom block with the new container.
266   DCHECK(insn != nullptr);
267   DCHECK(insn != orig_block->first_mir_insn);
268   DCHECK(insn == bottom_block->first_mir_insn);
269   DCHECK_EQ(insn->offset, bottom_block->start_offset);
270   // Scan the "bottom" instructions, remapping them to the
271   // newly created "bottom" block.
272   MIR* p = insn;
273   p->bb = bottom_block->id;
274   while (p != bottom_block->last_mir_insn) {
275     p = p->next;
276     DCHECK(p != nullptr);
277     p->bb = bottom_block->id;
278   }
279 
280   return bottom_block;
281 }
282 
283 /*
284  * Given a code offset, find out the block that starts with it. If the offset
285  * is in the middle of an existing block, split it into two.  If immed_pred_block_p
286  * is not non-null and is the block being split, update *immed_pred_block_p to
287  * point to the bottom block so that outgoing edges can be set up properly
288  * (by the caller)
289  * Utilizes a map for fast lookup of the typical cases.
290  */
FindBlock(DexOffset code_offset,bool create,BasicBlock ** immed_pred_block_p,ScopedArenaVector<uint16_t> * dex_pc_to_block_map)291 BasicBlock* MIRGraph::FindBlock(DexOffset code_offset, bool create,
292                                 BasicBlock** immed_pred_block_p,
293                                 ScopedArenaVector<uint16_t>* dex_pc_to_block_map) {
294   if (UNLIKELY(code_offset >= current_code_item_->insns_size_in_code_units_)) {
295     // There can be a fall-through out of the method code. We shall record such a block
296     // here (assuming create==true) and check that it's dead at the end of InlineMethod().
297     // Though we're only aware of the cases where code_offset is exactly the same as
298     // insns_size_in_code_units_, treat greater code_offset the same just in case.
299     code_offset = current_code_item_->insns_size_in_code_units_;
300   }
301 
302   int block_id = (*dex_pc_to_block_map)[code_offset];
303   BasicBlock* bb = GetBasicBlock(block_id);
304 
305   if ((bb != nullptr) && (bb->start_offset == code_offset)) {
306     // Does this containing block start with the desired instruction?
307     return bb;
308   }
309 
310   // No direct hit.
311   if (!create) {
312     return nullptr;
313   }
314 
315   if (bb != nullptr) {
316     // The target exists somewhere in an existing block.
317     BasicBlock* bottom_block = SplitBlock(code_offset, bb, bb == *immed_pred_block_p ?  immed_pred_block_p : nullptr);
318     DCHECK(bottom_block != nullptr);
319     MIR* p = bottom_block->first_mir_insn;
320     BasicBlock* orig_block = bb;
321     DCHECK_EQ((*dex_pc_to_block_map)[p->offset], orig_block->id);
322     // Scan the "bottom" instructions, remapping them to the
323     // newly created "bottom" block.
324     (*dex_pc_to_block_map)[p->offset] = bottom_block->id;
325     while (p != bottom_block->last_mir_insn) {
326       p = p->next;
327       DCHECK(p != nullptr);
328       int opcode = p->dalvikInsn.opcode;
329       /*
330        * Some messiness here to ensure that we only enter real opcodes and only the
331        * first half of a potentially throwing instruction that has been split into
332        * CHECK and work portions. Since the 2nd half of a split operation is always
333        * the first in a BasicBlock, we can't hit it here.
334        */
335       if ((opcode == kMirOpCheck) || !MIR::DecodedInstruction::IsPseudoMirOp(opcode)) {
336         BasicBlockId mapped_id = (*dex_pc_to_block_map)[p->offset];
337         // At first glance the instructions should all be mapped to orig_block.
338         // However, multiple instructions may correspond to the same dex, hence an earlier
339         // instruction may have already moved the mapping for dex to bottom_block.
340         DCHECK((mapped_id == orig_block->id) || (mapped_id == bottom_block->id));
341         (*dex_pc_to_block_map)[p->offset] = bottom_block->id;
342       }
343     }
344     return bottom_block;
345   }
346 
347   // Create a new block.
348   bb = CreateNewBB(kDalvikByteCode);
349   bb->start_offset = code_offset;
350   (*dex_pc_to_block_map)[bb->start_offset] = bb->id;
351   return bb;
352 }
353 
354 
355 /* Identify code range in try blocks and set up the empty catch blocks */
ProcessTryCatchBlocks(ScopedArenaVector<uint16_t> * dex_pc_to_block_map)356 void MIRGraph::ProcessTryCatchBlocks(ScopedArenaVector<uint16_t>* dex_pc_to_block_map) {
357   int tries_size = current_code_item_->tries_size_;
358   DexOffset offset;
359 
360   if (tries_size == 0) {
361     return;
362   }
363 
364   for (int i = 0; i < tries_size; i++) {
365     const DexFile::TryItem* pTry =
366         DexFile::GetTryItems(*current_code_item_, i);
367     DexOffset start_offset = pTry->start_addr_;
368     DexOffset end_offset = start_offset + pTry->insn_count_;
369     for (offset = start_offset; offset < end_offset; offset++) {
370       try_block_addr_->SetBit(offset);
371     }
372   }
373 
374   // Iterate over each of the handlers to enqueue the empty Catch blocks.
375   const uint8_t* handlers_ptr = DexFile::GetCatchHandlerData(*current_code_item_, 0);
376   uint32_t handlers_size = DecodeUnsignedLeb128(&handlers_ptr);
377   for (uint32_t idx = 0; idx < handlers_size; idx++) {
378     CatchHandlerIterator iterator(handlers_ptr);
379     for (; iterator.HasNext(); iterator.Next()) {
380       uint32_t address = iterator.GetHandlerAddress();
381       FindBlock(address, true /*create*/, /* immed_pred_block_p */ nullptr, dex_pc_to_block_map);
382     }
383     handlers_ptr = iterator.EndDataPointer();
384   }
385 }
386 
IsBadMonitorExitCatch(NarrowDexOffset monitor_exit_offset,NarrowDexOffset catch_offset)387 bool MIRGraph::IsBadMonitorExitCatch(NarrowDexOffset monitor_exit_offset,
388                                      NarrowDexOffset catch_offset) {
389   // Catches for monitor-exit during stack unwinding have the pattern
390   //   move-exception (move)* (goto)? monitor-exit throw
391   // In the currently generated dex bytecode we see these catching a bytecode range including
392   // either its own or an identical monitor-exit, http://b/15745363 . This function checks if
393   // it's the case for a given monitor-exit and catch block so that we can ignore it.
394   // (We don't want to ignore all monitor-exit catches since one could enclose a synchronized
395   // block in a try-block and catch the NPE, Error or Throwable and we should let it through;
396   // even though a throwing monitor-exit certainly indicates a bytecode error.)
397   const Instruction* monitor_exit = Instruction::At(current_code_item_->insns_ + monitor_exit_offset);
398   DCHECK(monitor_exit->Opcode() == Instruction::MONITOR_EXIT);
399   int monitor_reg = monitor_exit->VRegA_11x();
400   const Instruction* check_insn = Instruction::At(current_code_item_->insns_ + catch_offset);
401   if (check_insn->Opcode() == Instruction::MOVE_EXCEPTION) {
402     if (check_insn->VRegA_11x() == monitor_reg) {
403       // Unexpected move-exception to the same register. Probably not the pattern we're looking for.
404       return false;
405     }
406     check_insn = check_insn->Next();
407   }
408   while (true) {
409     int dest = -1;
410     bool wide = false;
411     switch (check_insn->Opcode()) {
412       case Instruction::MOVE_WIDE:
413         wide = true;
414         FALLTHROUGH_INTENDED;
415       case Instruction::MOVE_OBJECT:
416       case Instruction::MOVE:
417         dest = check_insn->VRegA_12x();
418         break;
419 
420       case Instruction::MOVE_WIDE_FROM16:
421         wide = true;
422         FALLTHROUGH_INTENDED;
423       case Instruction::MOVE_OBJECT_FROM16:
424       case Instruction::MOVE_FROM16:
425         dest = check_insn->VRegA_22x();
426         break;
427 
428       case Instruction::MOVE_WIDE_16:
429         wide = true;
430         FALLTHROUGH_INTENDED;
431       case Instruction::MOVE_OBJECT_16:
432       case Instruction::MOVE_16:
433         dest = check_insn->VRegA_32x();
434         break;
435 
436       case Instruction::GOTO:
437       case Instruction::GOTO_16:
438       case Instruction::GOTO_32:
439         check_insn = check_insn->RelativeAt(check_insn->GetTargetOffset());
440         FALLTHROUGH_INTENDED;
441       default:
442         return check_insn->Opcode() == Instruction::MONITOR_EXIT &&
443             check_insn->VRegA_11x() == monitor_reg;
444     }
445 
446     if (dest == monitor_reg || (wide && dest + 1 == monitor_reg)) {
447       return false;
448     }
449 
450     check_insn = check_insn->Next();
451   }
452 }
453 
454 /* Process instructions with the kBranch flag */
ProcessCanBranch(BasicBlock * cur_block,MIR * insn,DexOffset cur_offset,int width,int flags,const uint16_t * code_ptr,const uint16_t * code_end,ScopedArenaVector<uint16_t> * dex_pc_to_block_map)455 BasicBlock* MIRGraph::ProcessCanBranch(BasicBlock* cur_block, MIR* insn, DexOffset cur_offset,
456                                        int width, int flags, const uint16_t* code_ptr,
457                                        const uint16_t* code_end,
458                                        ScopedArenaVector<uint16_t>* dex_pc_to_block_map) {
459   DexOffset target = cur_offset;
460   switch (insn->dalvikInsn.opcode) {
461     case Instruction::GOTO:
462     case Instruction::GOTO_16:
463     case Instruction::GOTO_32:
464       target += insn->dalvikInsn.vA;
465       break;
466     case Instruction::IF_EQ:
467     case Instruction::IF_NE:
468     case Instruction::IF_LT:
469     case Instruction::IF_GE:
470     case Instruction::IF_GT:
471     case Instruction::IF_LE:
472       cur_block->conditional_branch = true;
473       target += insn->dalvikInsn.vC;
474       break;
475     case Instruction::IF_EQZ:
476     case Instruction::IF_NEZ:
477     case Instruction::IF_LTZ:
478     case Instruction::IF_GEZ:
479     case Instruction::IF_GTZ:
480     case Instruction::IF_LEZ:
481       cur_block->conditional_branch = true;
482       target += insn->dalvikInsn.vB;
483       break;
484     default:
485       LOG(FATAL) << "Unexpected opcode(" << insn->dalvikInsn.opcode << ") with kBranch set";
486   }
487   CountBranch(target);
488   BasicBlock* taken_block = FindBlock(target, /* create */ true,
489                                       /* immed_pred_block_p */ &cur_block,
490                                       dex_pc_to_block_map);
491   DCHECK(taken_block != nullptr);
492   cur_block->taken = taken_block->id;
493   taken_block->predecessors.push_back(cur_block->id);
494 
495   /* Always terminate the current block for conditional branches */
496   if (flags & Instruction::kContinue) {
497     BasicBlock* fallthrough_block = FindBlock(cur_offset +  width,
498                                              /* create */
499                                              true,
500                                              /* immed_pred_block_p */
501                                              &cur_block,
502                                              dex_pc_to_block_map);
503     DCHECK(fallthrough_block != nullptr);
504     cur_block->fall_through = fallthrough_block->id;
505     fallthrough_block->predecessors.push_back(cur_block->id);
506   } else if (code_ptr < code_end) {
507     FindBlock(cur_offset + width, /* create */ true, /* immed_pred_block_p */ nullptr, dex_pc_to_block_map);
508   }
509   return cur_block;
510 }
511 
512 /* Process instructions with the kSwitch flag */
ProcessCanSwitch(BasicBlock * cur_block,MIR * insn,DexOffset cur_offset,int width,int flags,ScopedArenaVector<uint16_t> * dex_pc_to_block_map)513 BasicBlock* MIRGraph::ProcessCanSwitch(BasicBlock* cur_block, MIR* insn, DexOffset cur_offset,
514                                        int width, int flags,
515                                        ScopedArenaVector<uint16_t>* dex_pc_to_block_map) {
516   UNUSED(flags);
517   const uint16_t* switch_data =
518       reinterpret_cast<const uint16_t*>(GetCurrentInsns() + cur_offset +
519           static_cast<int32_t>(insn->dalvikInsn.vB));
520   int size;
521   const int* keyTable;
522   const int* target_table;
523   int i;
524   int first_key;
525 
526   /*
527    * Packed switch data format:
528    *  ushort ident = 0x0100   magic value
529    *  ushort size             number of entries in the table
530    *  int first_key           first (and lowest) switch case value
531    *  int targets[size]       branch targets, relative to switch opcode
532    *
533    * Total size is (4+size*2) 16-bit code units.
534    */
535   if (insn->dalvikInsn.opcode == Instruction::PACKED_SWITCH) {
536     DCHECK_EQ(static_cast<int>(switch_data[0]),
537               static_cast<int>(Instruction::kPackedSwitchSignature));
538     size = switch_data[1];
539     first_key = switch_data[2] | (switch_data[3] << 16);
540     target_table = reinterpret_cast<const int*>(&switch_data[4]);
541     keyTable = nullptr;        // Make the compiler happy.
542   /*
543    * Sparse switch data format:
544    *  ushort ident = 0x0200   magic value
545    *  ushort size             number of entries in the table; > 0
546    *  int keys[size]          keys, sorted low-to-high; 32-bit aligned
547    *  int targets[size]       branch targets, relative to switch opcode
548    *
549    * Total size is (2+size*4) 16-bit code units.
550    */
551   } else {
552     DCHECK_EQ(static_cast<int>(switch_data[0]),
553               static_cast<int>(Instruction::kSparseSwitchSignature));
554     size = switch_data[1];
555     keyTable = reinterpret_cast<const int*>(&switch_data[2]);
556     target_table = reinterpret_cast<const int*>(&switch_data[2 + size*2]);
557     first_key = 0;   // To make the compiler happy.
558   }
559 
560   if (cur_block->successor_block_list_type != kNotUsed) {
561     LOG(FATAL) << "Successor block list already in use: "
562                << static_cast<int>(cur_block->successor_block_list_type);
563   }
564   cur_block->successor_block_list_type =
565       (insn->dalvikInsn.opcode == Instruction::PACKED_SWITCH) ?  kPackedSwitch : kSparseSwitch;
566   cur_block->successor_blocks.reserve(size);
567 
568   for (i = 0; i < size; i++) {
569     BasicBlock* case_block = FindBlock(cur_offset + target_table[i],  /* create */ true,
570                                        /* immed_pred_block_p */ &cur_block,
571                                        dex_pc_to_block_map);
572     DCHECK(case_block != nullptr);
573     SuccessorBlockInfo* successor_block_info =
574         static_cast<SuccessorBlockInfo*>(arena_->Alloc(sizeof(SuccessorBlockInfo),
575                                                        kArenaAllocSuccessor));
576     successor_block_info->block = case_block->id;
577     successor_block_info->key =
578         (insn->dalvikInsn.opcode == Instruction::PACKED_SWITCH) ?
579         first_key + i : keyTable[i];
580     cur_block->successor_blocks.push_back(successor_block_info);
581     case_block->predecessors.push_back(cur_block->id);
582   }
583 
584   /* Fall-through case */
585   BasicBlock* fallthrough_block = FindBlock(cur_offset +  width, /* create */ true,
586                                             /* immed_pred_block_p */ nullptr,
587                                             dex_pc_to_block_map);
588   DCHECK(fallthrough_block != nullptr);
589   cur_block->fall_through = fallthrough_block->id;
590   fallthrough_block->predecessors.push_back(cur_block->id);
591   return cur_block;
592 }
593 
594 /* Process instructions with the kThrow flag */
ProcessCanThrow(BasicBlock * cur_block,MIR * insn,DexOffset cur_offset,int width,int flags,ArenaBitVector * try_block_addr,const uint16_t * code_ptr,const uint16_t * code_end,ScopedArenaVector<uint16_t> * dex_pc_to_block_map)595 BasicBlock* MIRGraph::ProcessCanThrow(BasicBlock* cur_block, MIR* insn, DexOffset cur_offset,
596                                       int width, int flags, ArenaBitVector* try_block_addr,
597                                       const uint16_t* code_ptr, const uint16_t* code_end,
598                                       ScopedArenaVector<uint16_t>* dex_pc_to_block_map) {
599   UNUSED(flags);
600   bool in_try_block = try_block_addr->IsBitSet(cur_offset);
601   bool is_throw = (insn->dalvikInsn.opcode == Instruction::THROW);
602 
603   /* In try block */
604   if (in_try_block) {
605     CatchHandlerIterator iterator(*current_code_item_, cur_offset);
606 
607     if (cur_block->successor_block_list_type != kNotUsed) {
608       LOG(INFO) << PrettyMethod(cu_->method_idx, *cu_->dex_file);
609       LOG(FATAL) << "Successor block list already in use: "
610                  << static_cast<int>(cur_block->successor_block_list_type);
611     }
612 
613     for (; iterator.HasNext(); iterator.Next()) {
614       BasicBlock* catch_block = FindBlock(iterator.GetHandlerAddress(), false /* create */,
615                                           nullptr /* immed_pred_block_p */,
616                                           dex_pc_to_block_map);
617       if (insn->dalvikInsn.opcode == Instruction::MONITOR_EXIT &&
618           IsBadMonitorExitCatch(insn->offset, catch_block->start_offset)) {
619         // Don't allow monitor-exit to catch its own exception, http://b/15745363 .
620         continue;
621       }
622       if (cur_block->successor_block_list_type == kNotUsed) {
623         cur_block->successor_block_list_type = kCatch;
624       }
625       catch_block->catch_entry = true;
626       if (kIsDebugBuild) {
627         catches_.insert(catch_block->start_offset);
628       }
629       SuccessorBlockInfo* successor_block_info = reinterpret_cast<SuccessorBlockInfo*>
630           (arena_->Alloc(sizeof(SuccessorBlockInfo), kArenaAllocSuccessor));
631       successor_block_info->block = catch_block->id;
632       successor_block_info->key = iterator.GetHandlerTypeIndex();
633       cur_block->successor_blocks.push_back(successor_block_info);
634       catch_block->predecessors.push_back(cur_block->id);
635     }
636     in_try_block = (cur_block->successor_block_list_type != kNotUsed);
637   }
638   bool build_all_edges =
639       (cu_->disable_opt & (1 << kSuppressExceptionEdges)) || is_throw || in_try_block;
640   if (!in_try_block && build_all_edges) {
641     BasicBlock* eh_block = CreateNewBB(kExceptionHandling);
642     cur_block->taken = eh_block->id;
643     eh_block->start_offset = cur_offset;
644     eh_block->predecessors.push_back(cur_block->id);
645   }
646 
647   if (is_throw) {
648     cur_block->explicit_throw = true;
649     if (code_ptr < code_end) {
650       // Force creation of new block following THROW via side-effect.
651       FindBlock(cur_offset + width, /* create */ true, /* immed_pred_block_p */ nullptr, dex_pc_to_block_map);
652     }
653     if (!in_try_block) {
654        // Don't split a THROW that can't rethrow - we're done.
655       return cur_block;
656     }
657   }
658 
659   if (!build_all_edges) {
660     /*
661      * Even though there is an exception edge here, control cannot return to this
662      * method.  Thus, for the purposes of dataflow analysis and optimization, we can
663      * ignore the edge.  Doing this reduces compile time, and increases the scope
664      * of the basic-block level optimization pass.
665      */
666     return cur_block;
667   }
668 
669   /*
670    * Split the potentially-throwing instruction into two parts.
671    * The first half will be a pseudo-op that captures the exception
672    * edges and terminates the basic block.  It always falls through.
673    * Then, create a new basic block that begins with the throwing instruction
674    * (minus exceptions).  Note: this new basic block must NOT be entered into
675    * the block_map.  If the potentially-throwing instruction is the target of a
676    * future branch, we need to find the check psuedo half.  The new
677    * basic block containing the work portion of the instruction should
678    * only be entered via fallthrough from the block containing the
679    * pseudo exception edge MIR.  Note also that this new block is
680    * not automatically terminated after the work portion, and may
681    * contain following instructions.
682    *
683    * Note also that the dex_pc_to_block_map entry for the potentially
684    * throwing instruction will refer to the original basic block.
685    */
686   BasicBlock* new_block = CreateNewBB(kDalvikByteCode);
687   new_block->start_offset = insn->offset;
688   cur_block->fall_through = new_block->id;
689   new_block->predecessors.push_back(cur_block->id);
690   MIR* new_insn = NewMIR();
691   *new_insn = *insn;
692   insn->dalvikInsn.opcode = static_cast<Instruction::Code>(kMirOpCheck);
693   // Associate the two halves.
694   insn->meta.throw_insn = new_insn;
695   new_block->AppendMIR(new_insn);
696   return new_block;
697 }
698 
699 /* Parse a Dex method and insert it into the MIRGraph at the current insert point. */
InlineMethod(const DexFile::CodeItem * code_item,uint32_t access_flags,InvokeType invoke_type ATTRIBUTE_UNUSED,uint16_t class_def_idx,uint32_t method_idx,jobject class_loader,const DexFile & dex_file)700 void MIRGraph::InlineMethod(const DexFile::CodeItem* code_item, uint32_t access_flags,
701                            InvokeType invoke_type ATTRIBUTE_UNUSED, uint16_t class_def_idx,
702                            uint32_t method_idx, jobject class_loader, const DexFile& dex_file) {
703   current_code_item_ = code_item;
704   method_stack_.push_back(std::make_pair(current_method_, current_offset_));
705   current_method_ = m_units_.size();
706   current_offset_ = 0;
707   // TODO: will need to snapshot stack image and use that as the mir context identification.
708   m_units_.push_back(new (arena_) DexCompilationUnit(
709       cu_, class_loader, Runtime::Current()->GetClassLinker(), dex_file,
710       current_code_item_, class_def_idx, method_idx, access_flags,
711       cu_->compiler_driver->GetVerifiedMethod(&dex_file, method_idx)));
712   const uint16_t* code_ptr = current_code_item_->insns_;
713   const uint16_t* code_end =
714       current_code_item_->insns_ + current_code_item_->insns_size_in_code_units_;
715 
716   // TODO: need to rework expansion of block list & try_block_addr when inlining activated.
717   // TUNING: use better estimate of basic blocks for following resize.
718   block_list_.reserve(block_list_.size() + current_code_item_->insns_size_in_code_units_);
719   // FindBlock lookup cache.
720   ScopedArenaAllocator allocator(&cu_->arena_stack);
721   ScopedArenaVector<uint16_t> dex_pc_to_block_map(allocator.Adapter());
722   dex_pc_to_block_map.resize(current_code_item_->insns_size_in_code_units_ +
723                              1 /* Fall-through on last insn; dead or punt to interpreter. */);
724 
725   // TODO: replace with explicit resize routine.  Using automatic extension side effect for now.
726   try_block_addr_->SetBit(current_code_item_->insns_size_in_code_units_);
727   try_block_addr_->ClearBit(current_code_item_->insns_size_in_code_units_);
728 
729   // If this is the first method, set up default entry and exit blocks.
730   if (current_method_ == 0) {
731     DCHECK(entry_block_ == nullptr);
732     DCHECK(exit_block_ == nullptr);
733     DCHECK_EQ(GetNumBlocks(), 0U);
734     // Use id 0 to represent a null block.
735     BasicBlock* null_block = CreateNewBB(kNullBlock);
736     DCHECK_EQ(null_block->id, NullBasicBlockId);
737     null_block->hidden = true;
738     entry_block_ = CreateNewBB(kEntryBlock);
739     exit_block_ = CreateNewBB(kExitBlock);
740   } else {
741     UNIMPLEMENTED(FATAL) << "Nested inlining not implemented.";
742     /*
743      * Will need to manage storage for ins & outs, push prevous state and update
744      * insert point.
745      */
746   }
747 
748   /* Current block to record parsed instructions */
749   BasicBlock* cur_block = CreateNewBB(kDalvikByteCode);
750   DCHECK_EQ(current_offset_, 0U);
751   cur_block->start_offset = current_offset_;
752   // TODO: for inlining support, insert at the insert point rather than entry block.
753   entry_block_->fall_through = cur_block->id;
754   cur_block->predecessors.push_back(entry_block_->id);
755 
756   /* Identify code range in try blocks and set up the empty catch blocks */
757   ProcessTryCatchBlocks(&dex_pc_to_block_map);
758 
759   uint64_t merged_df_flags = 0u;
760 
761   /* Parse all instructions and put them into containing basic blocks */
762   while (code_ptr < code_end) {
763     MIR *insn = NewMIR();
764     insn->offset = current_offset_;
765     insn->m_unit_index = current_method_;
766     int width = ParseInsn(code_ptr, &insn->dalvikInsn);
767     Instruction::Code opcode = insn->dalvikInsn.opcode;
768     if (opcode_count_ != nullptr) {
769       opcode_count_[static_cast<int>(opcode)]++;
770     }
771 
772     int flags = insn->dalvikInsn.FlagsOf();
773     int verify_flags = Instruction::VerifyFlagsOf(insn->dalvikInsn.opcode);
774 
775     uint64_t df_flags = GetDataFlowAttributes(insn);
776     merged_df_flags |= df_flags;
777 
778     if (df_flags & DF_HAS_DEFS) {
779       def_count_ += (df_flags & DF_A_WIDE) ? 2 : 1;
780     }
781 
782     if (df_flags & DF_LVN) {
783       cur_block->use_lvn = true;  // Run local value numbering on this basic block.
784     }
785 
786     // Check for inline data block signatures.
787     if (opcode == Instruction::NOP) {
788       // A simple NOP will have a width of 1 at this point, embedded data NOP > 1.
789       if ((width == 1) && ((current_offset_ & 0x1) == 0x1) && ((code_end - code_ptr) > 1)) {
790         // Could be an aligning nop.  If an embedded data NOP follows, treat pair as single unit.
791         uint16_t following_raw_instruction = code_ptr[1];
792         if ((following_raw_instruction == Instruction::kSparseSwitchSignature) ||
793             (following_raw_instruction == Instruction::kPackedSwitchSignature) ||
794             (following_raw_instruction == Instruction::kArrayDataSignature)) {
795           width += Instruction::At(code_ptr + 1)->SizeInCodeUnits();
796         }
797       }
798       if (width == 1) {
799         // It is a simple nop - treat normally.
800         cur_block->AppendMIR(insn);
801       } else {
802         DCHECK(cur_block->fall_through == NullBasicBlockId);
803         DCHECK(cur_block->taken == NullBasicBlockId);
804         // Unreachable instruction, mark for no continuation and end basic block.
805         flags &= ~Instruction::kContinue;
806         FindBlock(current_offset_ + width, /* create */ true,
807                   /* immed_pred_block_p */ nullptr, &dex_pc_to_block_map);
808       }
809     } else {
810       cur_block->AppendMIR(insn);
811     }
812 
813     // Associate the starting dex_pc for this opcode with its containing basic block.
814     dex_pc_to_block_map[insn->offset] = cur_block->id;
815 
816     code_ptr += width;
817 
818     if (flags & Instruction::kBranch) {
819       cur_block = ProcessCanBranch(cur_block, insn, current_offset_,
820                                    width, flags, code_ptr, code_end, &dex_pc_to_block_map);
821     } else if (flags & Instruction::kReturn) {
822       cur_block->terminated_by_return = true;
823       cur_block->fall_through = exit_block_->id;
824       exit_block_->predecessors.push_back(cur_block->id);
825       /*
826        * Terminate the current block if there are instructions
827        * afterwards.
828        */
829       if (code_ptr < code_end) {
830         /*
831          * Create a fallthrough block for real instructions
832          * (incl. NOP).
833          */
834          FindBlock(current_offset_ + width, /* create */ true,
835                    /* immed_pred_block_p */ nullptr, &dex_pc_to_block_map);
836       }
837     } else if (flags & Instruction::kThrow) {
838       cur_block = ProcessCanThrow(cur_block, insn, current_offset_, width, flags, try_block_addr_,
839                                   code_ptr, code_end, &dex_pc_to_block_map);
840     } else if (flags & Instruction::kSwitch) {
841       cur_block = ProcessCanSwitch(cur_block, insn, current_offset_, width,
842                                    flags, &dex_pc_to_block_map);
843     }
844     if (verify_flags & Instruction::kVerifyVarArgRange ||
845         verify_flags & Instruction::kVerifyVarArgRangeNonZero) {
846       /*
847        * The Quick backend's runtime model includes a gap between a method's
848        * argument ("in") vregs and the rest of its vregs.  Handling a range instruction
849        * which spans the gap is somewhat complicated, and should not happen
850        * in normal usage of dx.  Punt to the interpreter.
851        */
852       int first_reg_in_range = insn->dalvikInsn.vC;
853       int last_reg_in_range = first_reg_in_range + insn->dalvikInsn.vA - 1;
854       if (IsInVReg(first_reg_in_range) != IsInVReg(last_reg_in_range)) {
855         punt_to_interpreter_ = true;
856       }
857     }
858     current_offset_ += width;
859     BasicBlock* next_block = FindBlock(current_offset_, /* create */ false,
860                                        /* immed_pred_block_p */ nullptr,
861                                        &dex_pc_to_block_map);
862     if (next_block) {
863       /*
864        * The next instruction could be the target of a previously parsed
865        * forward branch so a block is already created. If the current
866        * instruction is not an unconditional branch, connect them through
867        * the fall-through link.
868        */
869       DCHECK(cur_block->fall_through == NullBasicBlockId ||
870              GetBasicBlock(cur_block->fall_through) == next_block ||
871              GetBasicBlock(cur_block->fall_through) == exit_block_);
872 
873       if ((cur_block->fall_through == NullBasicBlockId) && (flags & Instruction::kContinue)) {
874         cur_block->fall_through = next_block->id;
875         next_block->predecessors.push_back(cur_block->id);
876       }
877       cur_block = next_block;
878     }
879   }
880   merged_df_flags_ = merged_df_flags;
881 
882   if (cu_->enable_debug & (1 << kDebugDumpCFG)) {
883     DumpCFG("/sdcard/1_post_parse_cfg/", true);
884   }
885 
886   if (cu_->verbose) {
887     DumpMIRGraph();
888   }
889 
890   // Check if there's been a fall-through out of the method code.
891   BasicBlockId out_bb_id = dex_pc_to_block_map[current_code_item_->insns_size_in_code_units_];
892   if (UNLIKELY(out_bb_id != NullBasicBlockId)) {
893     // Eagerly calculate DFS order to determine if the block is dead.
894     DCHECK(!DfsOrdersUpToDate());
895     ComputeDFSOrders();
896     BasicBlock* out_bb = GetBasicBlock(out_bb_id);
897     DCHECK(out_bb != nullptr);
898     if (out_bb->block_type != kDead) {
899       LOG(WARNING) << "Live fall-through out of method in " << PrettyMethod(method_idx, dex_file);
900       SetPuntToInterpreter(true);
901     }
902   }
903 }
904 
ShowOpcodeStats()905 void MIRGraph::ShowOpcodeStats() {
906   DCHECK(opcode_count_ != nullptr);
907   LOG(INFO) << "Opcode Count";
908   for (int i = 0; i < kNumPackedOpcodes; i++) {
909     if (opcode_count_[i] != 0) {
910       LOG(INFO) << "-C- " << Instruction::Name(static_cast<Instruction::Code>(i))
911                 << " " << opcode_count_[i];
912     }
913   }
914 }
915 
GetDataFlowAttributes(Instruction::Code opcode)916 uint64_t MIRGraph::GetDataFlowAttributes(Instruction::Code opcode) {
917   DCHECK_LT((size_t) opcode, (sizeof(oat_data_flow_attributes_) / sizeof(oat_data_flow_attributes_[0])));
918   return oat_data_flow_attributes_[opcode];
919 }
920 
GetDataFlowAttributes(MIR * mir)921 uint64_t MIRGraph::GetDataFlowAttributes(MIR* mir) {
922   DCHECK(mir != nullptr);
923   Instruction::Code opcode = mir->dalvikInsn.opcode;
924   return GetDataFlowAttributes(opcode);
925 }
926 
927 // The path can easily surpass FS limits because of parameters etc. Use pathconf to get FS
928 // restrictions here. Note that a successful invocation will return an actual value. If the path
929 // is too long for some reason, the return will be ENAMETOOLONG. Then cut off part of the name.
930 //
931 // It's possible the path is not valid, or some other errors appear. In that case return false.
CreateDumpFile(std::string & fname,const char * dir_prefix,NarrowDexOffset start_offset,const char * suffix,int nr,std::string * output)932 static bool CreateDumpFile(std::string& fname, const char* dir_prefix, NarrowDexOffset start_offset,
933                            const char *suffix, int nr, std::string* output) {
934   std::string dir = StringPrintf("./%s", dir_prefix);
935   int64_t max_name_length = pathconf(dir.c_str(), _PC_NAME_MAX);
936   if (max_name_length <= 0) {
937     PLOG(ERROR) << "Could not get file name restrictions for " << dir;
938     return false;
939   }
940 
941   std::string name = StringPrintf("%s%x%s_%d.dot", fname.c_str(), start_offset,
942                                   suffix == nullptr ? "" : suffix, nr);
943   std::string fpath;
944   if (static_cast<int64_t>(name.size()) > max_name_length) {
945     std::string suffix_str = StringPrintf("_%d.dot", nr);
946     name = name.substr(0, static_cast<size_t>(max_name_length) - suffix_str.size()) + suffix_str;
947   }
948   // Sanity check.
949   DCHECK_LE(name.size(), static_cast<size_t>(max_name_length));
950 
951   *output = StringPrintf("%s%s", dir_prefix, name.c_str());
952   return true;
953 }
954 
955 // TODO: use a configurable base prefix, and adjust callers to supply pass name.
956 /* Dump the CFG into a DOT graph */
DumpCFG(const char * dir_prefix,bool all_blocks,const char * suffix)957 void MIRGraph::DumpCFG(const char* dir_prefix, bool all_blocks, const char *suffix) {
958   FILE* file;
959   static AtomicInteger cnt(0);
960 
961   // Increment counter to get a unique file number.
962   cnt++;
963   int nr = cnt.LoadRelaxed();
964 
965   std::string fname(PrettyMethod(cu_->method_idx, *cu_->dex_file));
966   ReplaceSpecialChars(fname);
967   std::string fpath;
968   if (!CreateDumpFile(fname, dir_prefix, GetBasicBlock(GetEntryBlock()->fall_through)->start_offset,
969                       suffix, nr, &fpath)) {
970     LOG(ERROR) << "Could not create dump file name for " << fname;
971     return;
972   }
973   file = fopen(fpath.c_str(), "w");
974   if (file == nullptr) {
975     PLOG(ERROR) << "Could not open " << fpath << " for DumpCFG.";
976     return;
977   }
978   fprintf(file, "digraph G {\n");
979 
980   fprintf(file, "  rankdir=TB\n");
981 
982   int num_blocks = all_blocks ? GetNumBlocks() : num_reachable_blocks_;
983   int idx;
984 
985   for (idx = 0; idx < num_blocks; idx++) {
986     int block_idx = all_blocks ? idx : dfs_order_[idx];
987     BasicBlock* bb = GetBasicBlock(block_idx);
988     if (bb == nullptr) continue;
989     if (bb->block_type == kDead) continue;
990     if (bb->hidden) continue;
991     if (bb->block_type == kEntryBlock) {
992       fprintf(file, "  entry_%d [shape=Mdiamond];\n", bb->id);
993     } else if (bb->block_type == kExitBlock) {
994       fprintf(file, "  exit_%d [shape=Mdiamond];\n", bb->id);
995     } else if (bb->block_type == kDalvikByteCode) {
996       fprintf(file, "  block%04x_%d [shape=record,label = \"{ \\\n",
997               bb->start_offset, bb->id);
998       const MIR* mir;
999         fprintf(file, "    {block id %d\\l}%s\\\n", bb->id,
1000                 bb->first_mir_insn ? " | " : " ");
1001         for (mir = bb->first_mir_insn; mir; mir = mir->next) {
1002             int opcode = mir->dalvikInsn.opcode;
1003             fprintf(file, "    {%04x %s %s %s %s %s %s %s %s %s\\l}%s\\\n", mir->offset,
1004                       mir->ssa_rep ? GetDalvikDisassembly(mir) :
1005                       !MIR::DecodedInstruction::IsPseudoMirOp(opcode) ?
1006                         Instruction::Name(mir->dalvikInsn.opcode) :
1007                         extended_mir_op_names_[opcode - kMirOpFirst],
1008                       (mir->optimization_flags & MIR_IGNORE_RANGE_CHECK) != 0 ? " no_rangecheck" : " ",
1009                       (mir->optimization_flags & MIR_IGNORE_NULL_CHECK) != 0 ? " no_nullcheck" : " ",
1010                       (mir->optimization_flags & MIR_IGNORE_SUSPEND_CHECK) != 0 ? " no_suspendcheck" : " ",
1011                       (mir->optimization_flags & MIR_STORE_NON_TEMPORAL) != 0 ? " non_temporal" : " ",
1012                       (mir->optimization_flags & MIR_CALLEE) != 0 ? " inlined" : " ",
1013                       (mir->optimization_flags & MIR_CLASS_IS_INITIALIZED) != 0 ? " cl_inited" : " ",
1014                       (mir->optimization_flags & MIR_CLASS_IS_IN_DEX_CACHE) != 0 ? " cl_in_cache" : " ",
1015                       (mir->optimization_flags & MIR_IGNORE_DIV_ZERO_CHECK) != 0 ? " no_div_check" : " ",
1016                       mir->next ? " | " : " ");
1017         }
1018         fprintf(file, "  }\"];\n\n");
1019     } else if (bb->block_type == kExceptionHandling) {
1020       char block_name[BLOCK_NAME_LEN];
1021 
1022       GetBlockName(bb, block_name);
1023       fprintf(file, "  %s [shape=invhouse];\n", block_name);
1024     }
1025 
1026     char block_name1[BLOCK_NAME_LEN], block_name2[BLOCK_NAME_LEN];
1027 
1028     if (bb->taken != NullBasicBlockId) {
1029       GetBlockName(bb, block_name1);
1030       GetBlockName(GetBasicBlock(bb->taken), block_name2);
1031       fprintf(file, "  %s:s -> %s:n [style=dotted]\n",
1032               block_name1, block_name2);
1033     }
1034     if (bb->fall_through != NullBasicBlockId) {
1035       GetBlockName(bb, block_name1);
1036       GetBlockName(GetBasicBlock(bb->fall_through), block_name2);
1037       fprintf(file, "  %s:s -> %s:n\n", block_name1, block_name2);
1038     }
1039 
1040     if (bb->successor_block_list_type != kNotUsed) {
1041       fprintf(file, "  succ%04x_%d [shape=%s,label = \"{ \\\n",
1042               bb->start_offset, bb->id,
1043               (bb->successor_block_list_type == kCatch) ?  "Mrecord" : "record");
1044 
1045       int last_succ_id = static_cast<int>(bb->successor_blocks.size() - 1u);
1046       int succ_id = 0;
1047       for (SuccessorBlockInfo* successor_block_info : bb->successor_blocks) {
1048         BasicBlock* dest_block = GetBasicBlock(successor_block_info->block);
1049         fprintf(file, "    {<f%d> %04x: %04x\\l}%s\\\n",
1050                 succ_id,
1051                 successor_block_info->key,
1052                 dest_block->start_offset,
1053                 (succ_id != last_succ_id) ? " | " : " ");
1054         ++succ_id;
1055       }
1056       fprintf(file, "  }\"];\n\n");
1057 
1058       GetBlockName(bb, block_name1);
1059       fprintf(file, "  %s:s -> succ%04x_%d:n [style=dashed]\n",
1060               block_name1, bb->start_offset, bb->id);
1061 
1062       // Link the successor pseudo-block with all of its potential targets.
1063       succ_id = 0;
1064       for (SuccessorBlockInfo* successor_block_info : bb->successor_blocks) {
1065         BasicBlock* dest_block = GetBasicBlock(successor_block_info->block);
1066 
1067         GetBlockName(dest_block, block_name2);
1068         fprintf(file, "  succ%04x_%d:f%d:e -> %s:n\n", bb->start_offset,
1069                 bb->id, succ_id++, block_name2);
1070       }
1071     }
1072     fprintf(file, "\n");
1073 
1074     if (cu_->verbose) {
1075       /* Display the dominator tree */
1076       GetBlockName(bb, block_name1);
1077       fprintf(file, "  cfg%s [label=\"%s\", shape=none];\n",
1078               block_name1, block_name1);
1079       if (bb->i_dom) {
1080         GetBlockName(GetBasicBlock(bb->i_dom), block_name2);
1081         fprintf(file, "  cfg%s:s -> cfg%s:n\n\n", block_name2, block_name1);
1082       }
1083     }
1084   }
1085   fprintf(file, "}\n");
1086   fclose(file);
1087 }
1088 
1089 /* Insert an MIR instruction to the end of a basic block. */
AppendMIR(MIR * mir)1090 void BasicBlock::AppendMIR(MIR* mir) {
1091   // Insert it after the last MIR.
1092   InsertMIRListAfter(last_mir_insn, mir, mir);
1093 }
1094 
AppendMIRList(MIR * first_list_mir,MIR * last_list_mir)1095 void BasicBlock::AppendMIRList(MIR* first_list_mir, MIR* last_list_mir) {
1096   // Insert it after the last MIR.
1097   InsertMIRListAfter(last_mir_insn, first_list_mir, last_list_mir);
1098 }
1099 
AppendMIRList(const std::vector<MIR * > & insns)1100 void BasicBlock::AppendMIRList(const std::vector<MIR*>& insns) {
1101   for (std::vector<MIR*>::const_iterator it = insns.begin(); it != insns.end(); it++) {
1102     MIR* new_mir = *it;
1103 
1104     // Add a copy of each MIR.
1105     InsertMIRListAfter(last_mir_insn, new_mir, new_mir);
1106   }
1107 }
1108 
1109 /* Insert a MIR instruction after the specified MIR. */
InsertMIRAfter(MIR * current_mir,MIR * new_mir)1110 void BasicBlock::InsertMIRAfter(MIR* current_mir, MIR* new_mir) {
1111   InsertMIRListAfter(current_mir, new_mir, new_mir);
1112 }
1113 
InsertMIRListAfter(MIR * insert_after,MIR * first_list_mir,MIR * last_list_mir)1114 void BasicBlock::InsertMIRListAfter(MIR* insert_after, MIR* first_list_mir, MIR* last_list_mir) {
1115   // If no MIR, we are done.
1116   if (first_list_mir == nullptr || last_list_mir == nullptr) {
1117     return;
1118   }
1119 
1120   // If insert_after is null, assume BB is empty.
1121   if (insert_after == nullptr) {
1122     first_mir_insn = first_list_mir;
1123     last_mir_insn = last_list_mir;
1124     last_list_mir->next = nullptr;
1125   } else {
1126     MIR* after_list = insert_after->next;
1127     insert_after->next = first_list_mir;
1128     last_list_mir->next = after_list;
1129     if (after_list == nullptr) {
1130       last_mir_insn = last_list_mir;
1131     }
1132   }
1133 
1134   // Set this BB to be the basic block of the MIRs.
1135   MIR* last = last_list_mir->next;
1136   for (MIR* mir = first_list_mir; mir != last; mir = mir->next) {
1137     mir->bb = id;
1138   }
1139 }
1140 
1141 /* Insert an MIR instruction to the head of a basic block. */
PrependMIR(MIR * mir)1142 void BasicBlock::PrependMIR(MIR* mir) {
1143   InsertMIRListBefore(first_mir_insn, mir, mir);
1144 }
1145 
PrependMIRList(MIR * first_list_mir,MIR * last_list_mir)1146 void BasicBlock::PrependMIRList(MIR* first_list_mir, MIR* last_list_mir) {
1147   // Insert it before the first MIR.
1148   InsertMIRListBefore(first_mir_insn, first_list_mir, last_list_mir);
1149 }
1150 
PrependMIRList(const std::vector<MIR * > & to_add)1151 void BasicBlock::PrependMIRList(const std::vector<MIR*>& to_add) {
1152   for (std::vector<MIR*>::const_iterator it = to_add.begin(); it != to_add.end(); it++) {
1153     MIR* mir = *it;
1154 
1155     InsertMIRListBefore(first_mir_insn, mir, mir);
1156   }
1157 }
1158 
1159 /* Insert a MIR instruction before the specified MIR. */
InsertMIRBefore(MIR * current_mir,MIR * new_mir)1160 void BasicBlock::InsertMIRBefore(MIR* current_mir, MIR* new_mir) {
1161   // Insert as a single element list.
1162   return InsertMIRListBefore(current_mir, new_mir, new_mir);
1163 }
1164 
FindPreviousMIR(MIR * mir)1165 MIR* BasicBlock::FindPreviousMIR(MIR* mir) {
1166   MIR* current = first_mir_insn;
1167 
1168   while (current != nullptr) {
1169     MIR* next = current->next;
1170 
1171     if (next == mir) {
1172       return current;
1173     }
1174 
1175     current = next;
1176   }
1177 
1178   return nullptr;
1179 }
1180 
InsertMIRListBefore(MIR * insert_before,MIR * first_list_mir,MIR * last_list_mir)1181 void BasicBlock::InsertMIRListBefore(MIR* insert_before, MIR* first_list_mir, MIR* last_list_mir) {
1182   // If no MIR, we are done.
1183   if (first_list_mir == nullptr || last_list_mir == nullptr) {
1184     return;
1185   }
1186 
1187   // If insert_before is null, assume BB is empty.
1188   if (insert_before == nullptr) {
1189     first_mir_insn = first_list_mir;
1190     last_mir_insn = last_list_mir;
1191     last_list_mir->next = nullptr;
1192   } else {
1193     if (first_mir_insn == insert_before) {
1194       last_list_mir->next = first_mir_insn;
1195       first_mir_insn = first_list_mir;
1196     } else {
1197       // Find the preceding MIR.
1198       MIR* before_list = FindPreviousMIR(insert_before);
1199       DCHECK(before_list != nullptr);
1200       before_list->next = first_list_mir;
1201       last_list_mir->next = insert_before;
1202     }
1203   }
1204 
1205   // Set this BB to be the basic block of the MIRs.
1206   for (MIR* mir = first_list_mir; mir != last_list_mir->next; mir = mir->next) {
1207     mir->bb = id;
1208   }
1209 }
1210 
RemoveMIR(MIR * mir)1211 bool BasicBlock::RemoveMIR(MIR* mir) {
1212   // Remove as a single element list.
1213   return RemoveMIRList(mir, mir);
1214 }
1215 
RemoveMIRList(MIR * first_list_mir,MIR * last_list_mir)1216 bool BasicBlock::RemoveMIRList(MIR* first_list_mir, MIR* last_list_mir) {
1217   if (first_list_mir == nullptr) {
1218     return false;
1219   }
1220 
1221   // Try to find the MIR.
1222   MIR* before_list = nullptr;
1223   MIR* after_list = nullptr;
1224 
1225   // If we are removing from the beginning of the MIR list.
1226   if (first_mir_insn == first_list_mir) {
1227     before_list = nullptr;
1228   } else {
1229     before_list = FindPreviousMIR(first_list_mir);
1230     if (before_list == nullptr) {
1231       // We did not find the mir.
1232       return false;
1233     }
1234   }
1235 
1236   // Remove the BB information and also find the after_list.
1237   for (MIR* mir = first_list_mir; mir != last_list_mir->next; mir = mir->next) {
1238     mir->bb = NullBasicBlockId;
1239   }
1240 
1241   after_list = last_list_mir->next;
1242 
1243   // If there is nothing before the list, after_list is the first_mir.
1244   if (before_list == nullptr) {
1245     first_mir_insn = after_list;
1246   } else {
1247     before_list->next = after_list;
1248   }
1249 
1250   // If there is nothing after the list, before_list is last_mir.
1251   if (after_list == nullptr) {
1252     last_mir_insn = before_list;
1253   }
1254 
1255   return true;
1256 }
1257 
GetFirstNonPhiInsn()1258 MIR* BasicBlock::GetFirstNonPhiInsn() {
1259   MIR* mir = first_mir_insn;
1260   while (mir != nullptr && static_cast<int>(mir->dalvikInsn.opcode) == kMirOpPhi) {
1261     mir = mir->next;
1262   }
1263   return mir;
1264 }
1265 
GetNextUnconditionalMir(MIRGraph * mir_graph,MIR * current)1266 MIR* BasicBlock::GetNextUnconditionalMir(MIRGraph* mir_graph, MIR* current) {
1267   MIR* next_mir = nullptr;
1268 
1269   if (current != nullptr) {
1270     next_mir = current->next;
1271   }
1272 
1273   if (next_mir == nullptr) {
1274     // Only look for next MIR that follows unconditionally.
1275     if ((taken == NullBasicBlockId) && (fall_through != NullBasicBlockId)) {
1276       next_mir = mir_graph->GetBasicBlock(fall_through)->first_mir_insn;
1277     }
1278   }
1279 
1280   return next_mir;
1281 }
1282 
FillTypeSizeString(uint32_t type_size,std::string * decoded_mir)1283 static void FillTypeSizeString(uint32_t type_size, std::string* decoded_mir) {
1284   DCHECK(decoded_mir != nullptr);
1285   OpSize type = static_cast<OpSize>(type_size >> 16);
1286   uint16_t vect_size = (type_size & 0xFFFF);
1287 
1288   // Now print the type and vector size.
1289   std::stringstream ss;
1290   ss << " (type:";
1291   ss << type;
1292   ss << " vectsize:";
1293   ss << vect_size;
1294   ss << ")";
1295 
1296   decoded_mir->append(ss.str());
1297 }
1298 
DisassembleExtendedInstr(const MIR * mir,std::string * decoded_mir)1299 void MIRGraph::DisassembleExtendedInstr(const MIR* mir, std::string* decoded_mir) {
1300   DCHECK(decoded_mir != nullptr);
1301   int opcode = mir->dalvikInsn.opcode;
1302   SSARepresentation* ssa_rep = mir->ssa_rep;
1303   int defs = (ssa_rep != nullptr) ? ssa_rep->num_defs : 0;
1304   int uses = (ssa_rep != nullptr) ? ssa_rep->num_uses : 0;
1305 
1306   if (opcode < kMirOpFirst) {
1307     return;  // It is not an extended instruction.
1308   }
1309 
1310   decoded_mir->append(extended_mir_op_names_[opcode - kMirOpFirst]);
1311 
1312   switch (opcode) {
1313     case kMirOpPhi: {
1314       if (defs > 0 && uses > 0) {
1315         BasicBlockId* incoming = mir->meta.phi_incoming;
1316         decoded_mir->append(StringPrintf(" %s = (%s",
1317                            GetSSANameWithConst(ssa_rep->defs[0], true).c_str(),
1318                            GetSSANameWithConst(ssa_rep->uses[0], true).c_str()));
1319         decoded_mir->append(StringPrintf(":%d", incoming[0]));
1320         for (int i = 1; i < uses; i++) {
1321           decoded_mir->append(StringPrintf(", %s:%d", GetSSANameWithConst(ssa_rep->uses[i], true).c_str(), incoming[i]));
1322         }
1323         decoded_mir->append(")");
1324       }
1325       break;
1326     }
1327     case kMirOpCopy:
1328       if (ssa_rep != nullptr) {
1329         decoded_mir->append(" ");
1330         decoded_mir->append(GetSSANameWithConst(ssa_rep->defs[0], false));
1331         if (defs > 1) {
1332           decoded_mir->append(", ");
1333           decoded_mir->append(GetSSANameWithConst(ssa_rep->defs[1], false));
1334         }
1335         decoded_mir->append(" = ");
1336         decoded_mir->append(GetSSANameWithConst(ssa_rep->uses[0], false));
1337         if (uses > 1) {
1338           decoded_mir->append(", ");
1339           decoded_mir->append(GetSSANameWithConst(ssa_rep->uses[1], false));
1340         }
1341       } else {
1342         decoded_mir->append(StringPrintf(" v%d = v%d", mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1343       }
1344       break;
1345     case kMirOpFusedCmplFloat:
1346     case kMirOpFusedCmpgFloat:
1347     case kMirOpFusedCmplDouble:
1348     case kMirOpFusedCmpgDouble:
1349     case kMirOpFusedCmpLong:
1350       if (ssa_rep != nullptr) {
1351         decoded_mir->append(" ");
1352         decoded_mir->append(GetSSANameWithConst(ssa_rep->uses[0], false));
1353         for (int i = 1; i < uses; i++) {
1354           decoded_mir->append(", ");
1355           decoded_mir->append(GetSSANameWithConst(ssa_rep->uses[i], false));
1356         }
1357       } else {
1358         decoded_mir->append(StringPrintf(" v%d, v%d", mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1359       }
1360       break;
1361     case kMirOpMoveVector:
1362       decoded_mir->append(StringPrintf(" vect%d = vect%d", mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1363       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1364       break;
1365     case kMirOpPackedAddition:
1366       decoded_mir->append(StringPrintf(" vect%d = vect%d + vect%d", mir->dalvikInsn.vA, mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1367       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1368       break;
1369     case kMirOpPackedMultiply:
1370       decoded_mir->append(StringPrintf(" vect%d = vect%d * vect%d", mir->dalvikInsn.vA, mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1371       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1372       break;
1373     case kMirOpPackedSubtract:
1374       decoded_mir->append(StringPrintf(" vect%d = vect%d - vect%d", mir->dalvikInsn.vA, mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1375       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1376       break;
1377     case kMirOpPackedAnd:
1378       decoded_mir->append(StringPrintf(" vect%d = vect%d & vect%d", mir->dalvikInsn.vA, mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1379       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1380       break;
1381     case kMirOpPackedOr:
1382       decoded_mir->append(StringPrintf(" vect%d = vect%d \\| vect%d", mir->dalvikInsn.vA, mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1383       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1384       break;
1385     case kMirOpPackedXor:
1386       decoded_mir->append(StringPrintf(" vect%d = vect%d ^ vect%d", mir->dalvikInsn.vA, mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1387       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1388       break;
1389     case kMirOpPackedShiftLeft:
1390       decoded_mir->append(StringPrintf(" vect%d = vect%d \\<\\< %d", mir->dalvikInsn.vA, mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1391       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1392       break;
1393     case kMirOpPackedUnsignedShiftRight:
1394       decoded_mir->append(StringPrintf(" vect%d = vect%d \\>\\>\\> %d", mir->dalvikInsn.vA, mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1395       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1396       break;
1397     case kMirOpPackedSignedShiftRight:
1398       decoded_mir->append(StringPrintf(" vect%d = vect%d \\>\\> %d", mir->dalvikInsn.vA, mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1399       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1400       break;
1401     case kMirOpConstVector:
1402       decoded_mir->append(StringPrintf(" vect%d = %x, %x, %x, %x", mir->dalvikInsn.vA, mir->dalvikInsn.arg[0],
1403                                       mir->dalvikInsn.arg[1], mir->dalvikInsn.arg[2], mir->dalvikInsn.arg[3]));
1404       break;
1405     case kMirOpPackedSet:
1406       if (ssa_rep != nullptr) {
1407         decoded_mir->append(StringPrintf(" vect%d = %s", mir->dalvikInsn.vA,
1408               GetSSANameWithConst(ssa_rep->uses[0], false).c_str()));
1409         if (uses > 1) {
1410           decoded_mir->append(", ");
1411           decoded_mir->append(GetSSANameWithConst(ssa_rep->uses[1], false));
1412         }
1413       } else {
1414         decoded_mir->append(StringPrintf(" vect%d = v%d", mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1415       }
1416       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1417       break;
1418     case kMirOpPackedAddReduce:
1419       if (ssa_rep != nullptr) {
1420         decoded_mir->append(" ");
1421         decoded_mir->append(GetSSANameWithConst(ssa_rep->defs[0], false));
1422         if (defs > 1) {
1423           decoded_mir->append(", ");
1424           decoded_mir->append(GetSSANameWithConst(ssa_rep->defs[1], false));
1425         }
1426         decoded_mir->append(StringPrintf(" = vect%d + %s", mir->dalvikInsn.vB,
1427             GetSSANameWithConst(ssa_rep->uses[0], false).c_str()));
1428         if (uses > 1) {
1429           decoded_mir->append(", ");
1430           decoded_mir->append(GetSSANameWithConst(ssa_rep->uses[1], false));
1431         }
1432       } else {
1433         decoded_mir->append(StringPrintf("v%d = vect%d + v%d", mir->dalvikInsn.vA, mir->dalvikInsn.vB, mir->dalvikInsn.vA));
1434       }
1435       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1436       break;
1437     case kMirOpPackedReduce:
1438       if (ssa_rep != nullptr) {
1439         decoded_mir->append(" ");
1440         decoded_mir->append(GetSSANameWithConst(ssa_rep->defs[0], false));
1441         if (defs > 1) {
1442           decoded_mir->append(", ");
1443           decoded_mir->append(GetSSANameWithConst(ssa_rep->defs[1], false));
1444         }
1445         decoded_mir->append(StringPrintf(" = vect%d (extr_idx:%d)", mir->dalvikInsn.vB, mir->dalvikInsn.arg[0]));
1446       } else {
1447         decoded_mir->append(StringPrintf(" v%d = vect%d (extr_idx:%d)", mir->dalvikInsn.vA,
1448                                          mir->dalvikInsn.vB, mir->dalvikInsn.arg[0]));
1449       }
1450       FillTypeSizeString(mir->dalvikInsn.vC, decoded_mir);
1451       break;
1452     case kMirOpReserveVectorRegisters:
1453     case kMirOpReturnVectorRegisters:
1454       decoded_mir->append(StringPrintf(" vect%d - vect%d", mir->dalvikInsn.vA, mir->dalvikInsn.vB));
1455       break;
1456     case kMirOpMemBarrier: {
1457       decoded_mir->append(" type:");
1458       std::stringstream ss;
1459       ss << static_cast<MemBarrierKind>(mir->dalvikInsn.vA);
1460       decoded_mir->append(ss.str());
1461       break;
1462     }
1463     case kMirOpPackedArrayGet:
1464     case kMirOpPackedArrayPut:
1465       decoded_mir->append(StringPrintf(" vect%d", mir->dalvikInsn.vA));
1466       if (ssa_rep != nullptr) {
1467         decoded_mir->append(StringPrintf(", %s[%s]",
1468                                         GetSSANameWithConst(ssa_rep->uses[0], false).c_str(),
1469                                         GetSSANameWithConst(ssa_rep->uses[1], false).c_str()));
1470       } else {
1471         decoded_mir->append(StringPrintf(", v%d[v%d]", mir->dalvikInsn.vB, mir->dalvikInsn.vC));
1472       }
1473       FillTypeSizeString(mir->dalvikInsn.arg[0], decoded_mir);
1474       break;
1475     case kMirOpMaddInt:
1476     case kMirOpMsubInt:
1477     case kMirOpMaddLong:
1478     case kMirOpMsubLong:
1479       if (ssa_rep != nullptr) {
1480         decoded_mir->append(" ");
1481         decoded_mir->append(GetSSANameWithConst(ssa_rep->defs[0], false));
1482         if (defs > 1) {
1483           decoded_mir->append(", ");
1484           decoded_mir->append(GetSSANameWithConst(ssa_rep->defs[1], false));
1485         }
1486         for (int i = 0; i < uses; i++) {
1487           decoded_mir->append(", ");
1488           decoded_mir->append(GetSSANameWithConst(ssa_rep->uses[i], false));
1489         }
1490       } else {
1491         decoded_mir->append(StringPrintf(" v%d, v%d, v%d, v%d",
1492                                          mir->dalvikInsn.vA, mir->dalvikInsn.vB,
1493                                          mir->dalvikInsn.vC, mir->dalvikInsn.arg[0]));
1494       }
1495       break;
1496     default:
1497       break;
1498   }
1499 }
1500 
GetDalvikDisassembly(const MIR * mir)1501 char* MIRGraph::GetDalvikDisassembly(const MIR* mir) {
1502   MIR::DecodedInstruction insn = mir->dalvikInsn;
1503   std::string str;
1504   int flags = 0;
1505   int opcode = insn.opcode;
1506   char* ret;
1507   bool nop = false;
1508   SSARepresentation* ssa_rep = mir->ssa_rep;
1509   Instruction::Format dalvik_format = Instruction::k10x;  // Default to no-operand format.
1510 
1511   // Handle special cases that recover the original dalvik instruction.
1512   if (opcode == kMirOpCheck) {
1513     str.append(extended_mir_op_names_[opcode - kMirOpFirst]);
1514     str.append(": ");
1515     // Recover the original Dex instruction.
1516     insn = mir->meta.throw_insn->dalvikInsn;
1517     ssa_rep = mir->meta.throw_insn->ssa_rep;
1518     opcode = insn.opcode;
1519   } else if (opcode == kMirOpNop) {
1520     str.append("[");
1521     if (mir->offset < current_code_item_->insns_size_in_code_units_) {
1522       // Recover original opcode.
1523       insn.opcode = Instruction::At(current_code_item_->insns_ + mir->offset)->Opcode();
1524       opcode = insn.opcode;
1525     }
1526     nop = true;
1527   }
1528   int defs = (ssa_rep != nullptr) ? ssa_rep->num_defs : 0;
1529   int uses = (ssa_rep != nullptr) ? ssa_rep->num_uses : 0;
1530 
1531   if (MIR::DecodedInstruction::IsPseudoMirOp(opcode)) {
1532     // Note that this does not check the MIR's opcode in all cases. In cases where it
1533     // recovered dalvik instruction, it uses opcode of that instead of the extended one.
1534     DisassembleExtendedInstr(mir, &str);
1535   } else {
1536     dalvik_format = Instruction::FormatOf(insn.opcode);
1537     flags = insn.FlagsOf();
1538     str.append(Instruction::Name(insn.opcode));
1539 
1540     // For invokes-style formats, treat wide regs as a pair of singles.
1541     bool show_singles = ((dalvik_format == Instruction::k35c) ||
1542                          (dalvik_format == Instruction::k3rc));
1543     if (defs != 0) {
1544       str.append(" ");
1545       str.append(GetSSANameWithConst(ssa_rep->defs[0], false));
1546       if (defs > 1) {
1547         str.append(", ");
1548         str.append(GetSSANameWithConst(ssa_rep->defs[1], false));
1549       }
1550       if (uses != 0) {
1551         str.append(", ");
1552       }
1553     }
1554     for (int i = 0; i < uses; i++) {
1555       str.append(" ");
1556       str.append(GetSSANameWithConst(ssa_rep->uses[i], show_singles));
1557       if (!show_singles && (reg_location_ != nullptr) && reg_location_[i].wide) {
1558         // For the listing, skip the high sreg.
1559         i++;
1560       }
1561       if (i != (uses - 1)) {
1562         str.append(",");
1563       }
1564     }
1565 
1566     switch (dalvik_format) {
1567       case Instruction::k11n:  // Add one immediate from vB.
1568       case Instruction::k21s:
1569       case Instruction::k31i:
1570       case Instruction::k21h:
1571         str.append(StringPrintf(", #0x%x", insn.vB));
1572         break;
1573       case Instruction::k51l:  // Add one wide immediate.
1574         str.append(StringPrintf(", #%" PRId64, insn.vB_wide));
1575         break;
1576       case Instruction::k21c:  // One register, one string/type/method index.
1577       case Instruction::k31c:
1578         str.append(StringPrintf(", index #0x%x", insn.vB));
1579         break;
1580       case Instruction::k22c:  // Two registers, one string/type/method index.
1581         str.append(StringPrintf(", index #0x%x", insn.vC));
1582         break;
1583       case Instruction::k22s:  // Add one immediate from vC.
1584       case Instruction::k22b:
1585         str.append(StringPrintf(", #0x%x", insn.vC));
1586         break;
1587       default:
1588         // Nothing left to print.
1589         break;
1590     }
1591 
1592     if ((flags & Instruction::kBranch) != 0) {
1593       // For branches, decode the instructions to print out the branch targets.
1594       int offset = 0;
1595       switch (dalvik_format) {
1596         case Instruction::k21t:
1597           offset = insn.vB;
1598           break;
1599         case Instruction::k22t:
1600           offset = insn.vC;
1601           break;
1602         case Instruction::k10t:
1603         case Instruction::k20t:
1604         case Instruction::k30t:
1605           offset = insn.vA;
1606           break;
1607         default:
1608           LOG(FATAL) << "Unexpected branch format " << dalvik_format << " from " << insn.opcode;
1609           break;
1610       }
1611       str.append(StringPrintf(", 0x%x (%c%x)", mir->offset + offset,
1612                               offset > 0 ? '+' : '-', offset > 0 ? offset : -offset));
1613     }
1614 
1615     if (nop) {
1616       str.append("]--optimized away");
1617     }
1618   }
1619   int length = str.length() + 1;
1620   ret = arena_->AllocArray<char>(length, kArenaAllocDFInfo);
1621   strncpy(ret, str.c_str(), length);
1622   return ret;
1623 }
1624 
1625 /* Turn method name into a legal Linux file name */
ReplaceSpecialChars(std::string & str)1626 void MIRGraph::ReplaceSpecialChars(std::string& str) {
1627   static const struct { const char before; const char after; } match[] = {
1628     {'/', '-'}, {';', '#'}, {' ', '#'}, {'$', '+'},
1629     {'(', '@'}, {')', '@'}, {'<', '='}, {'>', '='}
1630   };
1631   for (unsigned int i = 0; i < sizeof(match)/sizeof(match[0]); i++) {
1632     std::replace(str.begin(), str.end(), match[i].before, match[i].after);
1633   }
1634 }
1635 
GetSSAName(int ssa_reg)1636 std::string MIRGraph::GetSSAName(int ssa_reg) {
1637   // TODO: This value is needed for debugging. Currently, we compute this and then copy to the
1638   //       arena. We should be smarter and just place straight into the arena, or compute the
1639   //       value more lazily.
1640   int vreg = SRegToVReg(ssa_reg);
1641   if (vreg >= static_cast<int>(GetFirstTempVR())) {
1642     return StringPrintf("t%d_%d", SRegToVReg(ssa_reg), GetSSASubscript(ssa_reg));
1643   } else {
1644     return StringPrintf("v%d_%d", SRegToVReg(ssa_reg), GetSSASubscript(ssa_reg));
1645   }
1646 }
1647 
1648 // Similar to GetSSAName, but if ssa name represents an immediate show that as well.
GetSSANameWithConst(int ssa_reg,bool singles_only)1649 std::string MIRGraph::GetSSANameWithConst(int ssa_reg, bool singles_only) {
1650   if (reg_location_ == nullptr) {
1651     // Pre-SSA - just use the standard name.
1652     return GetSSAName(ssa_reg);
1653   }
1654   if (IsConst(reg_location_[ssa_reg])) {
1655     if (!singles_only && reg_location_[ssa_reg].wide &&
1656         !reg_location_[ssa_reg].high_word) {
1657       return StringPrintf("v%d_%d#0x%" PRIx64, SRegToVReg(ssa_reg), GetSSASubscript(ssa_reg),
1658                           ConstantValueWide(reg_location_[ssa_reg]));
1659     } else {
1660       return StringPrintf("v%d_%d#0x%x", SRegToVReg(ssa_reg), GetSSASubscript(ssa_reg),
1661                           ConstantValue(reg_location_[ssa_reg]));
1662     }
1663   } else {
1664     int vreg = SRegToVReg(ssa_reg);
1665     if (vreg >= static_cast<int>(GetFirstTempVR())) {
1666       return StringPrintf("t%d_%d", SRegToVReg(ssa_reg), GetSSASubscript(ssa_reg));
1667     } else {
1668       return StringPrintf("v%d_%d", SRegToVReg(ssa_reg), GetSSASubscript(ssa_reg));
1669     }
1670   }
1671 }
1672 
GetBlockName(BasicBlock * bb,char * name)1673 void MIRGraph::GetBlockName(BasicBlock* bb, char* name) {
1674   switch (bb->block_type) {
1675     case kEntryBlock:
1676       snprintf(name, BLOCK_NAME_LEN, "entry_%d", bb->id);
1677       break;
1678     case kExitBlock:
1679       snprintf(name, BLOCK_NAME_LEN, "exit_%d", bb->id);
1680       break;
1681     case kDalvikByteCode:
1682       snprintf(name, BLOCK_NAME_LEN, "block%04x_%d", bb->start_offset, bb->id);
1683       break;
1684     case kExceptionHandling:
1685       snprintf(name, BLOCK_NAME_LEN, "exception%04x_%d", bb->start_offset,
1686                bb->id);
1687       break;
1688     default:
1689       snprintf(name, BLOCK_NAME_LEN, "_%d", bb->id);
1690       break;
1691   }
1692 }
1693 
GetShortyFromMethodReference(const MethodReference & target_method)1694 const char* MIRGraph::GetShortyFromMethodReference(const MethodReference& target_method) {
1695   const DexFile::MethodId& method_id =
1696       target_method.dex_file->GetMethodId(target_method.dex_method_index);
1697   return target_method.dex_file->GetShorty(method_id.proto_idx_);
1698 }
1699 
1700 /* Debug Utility - dump a compilation unit */
DumpMIRGraph()1701 void MIRGraph::DumpMIRGraph() {
1702   const char* block_type_names[] = {
1703     "Null Block",
1704     "Entry Block",
1705     "Code Block",
1706     "Exit Block",
1707     "Exception Handling",
1708     "Catch Block"
1709   };
1710 
1711   LOG(INFO) << "Compiling " << PrettyMethod(cu_->method_idx, *cu_->dex_file);
1712   LOG(INFO) << GetInsns(0) << " insns";
1713   LOG(INFO) << GetNumBlocks() << " blocks in total";
1714 
1715   for (BasicBlock* bb : block_list_) {
1716     LOG(INFO) << StringPrintf("Block %d (%s) (insn %04x - %04x%s)",
1717         bb->id,
1718         block_type_names[bb->block_type],
1719         bb->start_offset,
1720         bb->last_mir_insn ? bb->last_mir_insn->offset : bb->start_offset,
1721         bb->last_mir_insn ? "" : " empty");
1722     if (bb->taken != NullBasicBlockId) {
1723       LOG(INFO) << "  Taken branch: block " << bb->taken
1724                 << "(0x" << std::hex << GetBasicBlock(bb->taken)->start_offset << ")";
1725     }
1726     if (bb->fall_through != NullBasicBlockId) {
1727       LOG(INFO) << "  Fallthrough : block " << bb->fall_through
1728                 << " (0x" << std::hex << GetBasicBlock(bb->fall_through)->start_offset << ")";
1729     }
1730   }
1731 }
1732 
1733 /*
1734  * Build an array of location records for the incoming arguments.
1735  * Note: one location record per word of arguments, with dummy
1736  * high-word loc for wide arguments.  Also pull up any following
1737  * MOVE_RESULT and incorporate it into the invoke.
1738  */
NewMemCallInfo(BasicBlock * bb,MIR * mir,InvokeType type,bool is_range)1739 CallInfo* MIRGraph::NewMemCallInfo(BasicBlock* bb, MIR* mir, InvokeType type, bool is_range) {
1740   CallInfo* info = static_cast<CallInfo*>(arena_->Alloc(sizeof(CallInfo),
1741                                                         kArenaAllocMisc));
1742   MIR* move_result_mir = FindMoveResult(bb, mir);
1743   if (move_result_mir == nullptr) {
1744     info->result.location = kLocInvalid;
1745   } else {
1746     info->result = GetRawDest(move_result_mir);
1747     move_result_mir->dalvikInsn.opcode = static_cast<Instruction::Code>(kMirOpNop);
1748   }
1749   info->num_arg_words = mir->ssa_rep->num_uses;
1750   info->args = (info->num_arg_words == 0) ? nullptr :
1751       arena_->AllocArray<RegLocation>(info->num_arg_words, kArenaAllocMisc);
1752   for (size_t i = 0; i < info->num_arg_words; i++) {
1753     info->args[i] = GetRawSrc(mir, i);
1754   }
1755   info->opt_flags = mir->optimization_flags;
1756   info->type = type;
1757   info->is_range = is_range;
1758   if (IsInstructionQuickInvoke(mir->dalvikInsn.opcode)) {
1759     const auto& method_info = GetMethodLoweringInfo(mir);
1760     info->method_ref = method_info.GetTargetMethod();
1761   } else {
1762     info->method_ref = MethodReference(GetCurrentDexCompilationUnit()->GetDexFile(),
1763                                        mir->dalvikInsn.vB);
1764   }
1765   info->index = mir->dalvikInsn.vB;
1766   info->offset = mir->offset;
1767   info->mir = mir;
1768   return info;
1769 }
1770 
1771 // Allocate a new MIR.
NewMIR()1772 MIR* MIRGraph::NewMIR() {
1773   MIR* mir = new (arena_) MIR();
1774   return mir;
1775 }
1776 
1777 // Allocate a new basic block.
NewMemBB(BBType block_type,int block_id)1778 BasicBlock* MIRGraph::NewMemBB(BBType block_type, int block_id) {
1779   BasicBlock* bb = new (arena_) BasicBlock(block_id, block_type, arena_);
1780 
1781   // TUNING: better estimate of the exit block predecessors?
1782   bb->predecessors.reserve((block_type == kExitBlock) ? 2048 : 2);
1783   block_id_map_.Put(block_id, block_id);
1784   return bb;
1785 }
1786 
InitializeConstantPropagation()1787 void MIRGraph::InitializeConstantPropagation() {
1788   is_constant_v_ = new (arena_) ArenaBitVector(arena_, GetNumSSARegs(), false);
1789   constant_values_ = arena_->AllocArray<int>(GetNumSSARegs(), kArenaAllocDFInfo);
1790 }
1791 
InitializeMethodUses()1792 void MIRGraph::InitializeMethodUses() {
1793   // The gate starts by initializing the use counts.
1794   int num_ssa_regs = GetNumSSARegs();
1795   use_counts_.clear();
1796   use_counts_.reserve(num_ssa_regs + 32);
1797   use_counts_.resize(num_ssa_regs, 0u);
1798   raw_use_counts_.clear();
1799   raw_use_counts_.reserve(num_ssa_regs + 32);
1800   raw_use_counts_.resize(num_ssa_regs, 0u);
1801 }
1802 
SSATransformationStart()1803 void MIRGraph::SSATransformationStart() {
1804   DCHECK(temp_scoped_alloc_.get() == nullptr);
1805   temp_scoped_alloc_.reset(ScopedArenaAllocator::Create(&cu_->arena_stack));
1806   temp_.ssa.num_vregs = GetNumOfCodeAndTempVRs();
1807   temp_.ssa.work_live_vregs = new (temp_scoped_alloc_.get()) ArenaBitVector(
1808       temp_scoped_alloc_.get(), temp_.ssa.num_vregs, false, kBitMapRegisterV);
1809 }
1810 
SSATransformationEnd()1811 void MIRGraph::SSATransformationEnd() {
1812   // Verify the dataflow information after the pass.
1813   if (cu_->enable_debug & (1 << kDebugVerifyDataflow)) {
1814     VerifyDataflow();
1815   }
1816 
1817   temp_.ssa.num_vregs = 0u;
1818   temp_.ssa.work_live_vregs = nullptr;
1819   DCHECK(temp_.ssa.def_block_matrix == nullptr);
1820   temp_.ssa.phi_node_blocks = nullptr;
1821   DCHECK(temp_scoped_alloc_.get() != nullptr);
1822   temp_scoped_alloc_.reset();
1823 
1824   // Update the maximum number of reachable blocks.
1825   max_num_reachable_blocks_ = num_reachable_blocks_;
1826 
1827   // Mark MIR SSA representations as up to date.
1828   mir_ssa_rep_up_to_date_ = true;
1829 }
1830 
GetNumDalvikInsns() const1831 size_t MIRGraph::GetNumDalvikInsns() const {
1832   size_t cumulative_size = 0u;
1833   bool counted_current_item = false;
1834   const uint8_t size_for_null_code_item = 2u;
1835 
1836   for (auto it : m_units_) {
1837     const DexFile::CodeItem* code_item = it->GetCodeItem();
1838     // Even if the code item is null, we still count non-zero value so that
1839     // each m_unit is counted as having impact.
1840     cumulative_size += (code_item == nullptr ?
1841         size_for_null_code_item : code_item->insns_size_in_code_units_);
1842     if (code_item == current_code_item_) {
1843       counted_current_item = true;
1844     }
1845   }
1846 
1847   // If the current code item was not counted yet, count it now.
1848   // This can happen for example in unit tests where some fields like m_units_
1849   // are not initialized.
1850   if (counted_current_item == false) {
1851     cumulative_size += (current_code_item_ == nullptr ?
1852         size_for_null_code_item : current_code_item_->insns_size_in_code_units_);
1853   }
1854 
1855   return cumulative_size;
1856 }
1857 
SelectTopologicalSortOrderFallBack(MIRGraph * mir_graph,const ArenaBitVector * current_loop,const ScopedArenaVector<size_t> * visited_cnt_values,ScopedArenaAllocator * allocator,ScopedArenaVector<BasicBlockId> * tmp_stack)1858 static BasicBlock* SelectTopologicalSortOrderFallBack(
1859     MIRGraph* mir_graph, const ArenaBitVector* current_loop,
1860     const ScopedArenaVector<size_t>* visited_cnt_values, ScopedArenaAllocator* allocator,
1861     ScopedArenaVector<BasicBlockId>* tmp_stack) {
1862   // No true loop head has been found but there may be true loop heads after the mess we need
1863   // to resolve. To avoid taking one of those, pick the candidate with the highest number of
1864   // reachable unvisited nodes. That candidate will surely be a part of a loop.
1865   BasicBlock* fall_back = nullptr;
1866   size_t fall_back_num_reachable = 0u;
1867   // Reuse the same bit vector for each candidate to mark reachable unvisited blocks.
1868   ArenaBitVector candidate_reachable(allocator, mir_graph->GetNumBlocks(), false, kBitMapMisc);
1869   AllNodesIterator iter(mir_graph);
1870   for (BasicBlock* candidate = iter.Next(); candidate != nullptr; candidate = iter.Next()) {
1871     if (candidate->hidden ||                            // Hidden, or
1872         candidate->visited ||                           // already processed, or
1873         (*visited_cnt_values)[candidate->id] == 0u ||   // no processed predecessors, or
1874         (current_loop != nullptr &&                     // outside current loop.
1875          !current_loop->IsBitSet(candidate->id))) {
1876       continue;
1877     }
1878     DCHECK(tmp_stack->empty());
1879     tmp_stack->push_back(candidate->id);
1880     candidate_reachable.ClearAllBits();
1881     size_t num_reachable = 0u;
1882     while (!tmp_stack->empty()) {
1883       BasicBlockId current_id = tmp_stack->back();
1884       tmp_stack->pop_back();
1885       BasicBlock* current_bb = mir_graph->GetBasicBlock(current_id);
1886       DCHECK(current_bb != nullptr);
1887       ChildBlockIterator child_iter(current_bb, mir_graph);
1888       BasicBlock* child_bb = child_iter.Next();
1889       for ( ; child_bb != nullptr; child_bb = child_iter.Next()) {
1890         DCHECK(!child_bb->hidden);
1891         if (child_bb->visited ||                            // Already processed, or
1892             (current_loop != nullptr &&                     // outside current loop.
1893              !current_loop->IsBitSet(child_bb->id))) {
1894           continue;
1895         }
1896         if (!candidate_reachable.IsBitSet(child_bb->id)) {
1897           candidate_reachable.SetBit(child_bb->id);
1898           tmp_stack->push_back(child_bb->id);
1899           num_reachable += 1u;
1900         }
1901       }
1902     }
1903     if (fall_back_num_reachable < num_reachable) {
1904       fall_back_num_reachable = num_reachable;
1905       fall_back = candidate;
1906     }
1907   }
1908   return fall_back;
1909 }
1910 
1911 // Compute from which unvisited blocks is bb_id reachable through unvisited blocks.
ComputeUnvisitedReachableFrom(MIRGraph * mir_graph,BasicBlockId bb_id,ArenaBitVector * reachable,ScopedArenaVector<BasicBlockId> * tmp_stack)1912 static void ComputeUnvisitedReachableFrom(MIRGraph* mir_graph, BasicBlockId bb_id,
1913                                           ArenaBitVector* reachable,
1914                                           ScopedArenaVector<BasicBlockId>* tmp_stack) {
1915   // NOTE: Loop heads indicated by the "visited" flag.
1916   DCHECK(tmp_stack->empty());
1917   reachable->ClearAllBits();
1918   tmp_stack->push_back(bb_id);
1919   while (!tmp_stack->empty()) {
1920     BasicBlockId current_id = tmp_stack->back();
1921     tmp_stack->pop_back();
1922     BasicBlock* current_bb = mir_graph->GetBasicBlock(current_id);
1923     DCHECK(current_bb != nullptr);
1924     for (BasicBlockId pred_id : current_bb->predecessors) {
1925       BasicBlock* pred_bb = mir_graph->GetBasicBlock(pred_id);
1926       DCHECK(pred_bb != nullptr);
1927       if (!pred_bb->visited && !reachable->IsBitSet(pred_bb->id)) {
1928         reachable->SetBit(pred_bb->id);
1929         tmp_stack->push_back(pred_bb->id);
1930       }
1931     }
1932   }
1933 }
1934 
ComputeTopologicalSortOrder()1935 void MIRGraph::ComputeTopologicalSortOrder() {
1936   ScopedArenaAllocator allocator(&cu_->arena_stack);
1937   unsigned int num_blocks = GetNumBlocks();
1938 
1939   ScopedArenaQueue<BasicBlock*> q(allocator.Adapter());
1940   ScopedArenaVector<size_t> visited_cnt_values(num_blocks, 0u, allocator.Adapter());
1941   ScopedArenaVector<BasicBlockId> loop_head_stack(allocator.Adapter());
1942   size_t max_nested_loops = 0u;
1943   ArenaBitVector loop_exit_blocks(&allocator, num_blocks, false, kBitMapMisc);
1944   loop_exit_blocks.ClearAllBits();
1945 
1946   // Count the number of blocks to process and add the entry block(s).
1947   unsigned int num_blocks_to_process = 0u;
1948   for (BasicBlock* bb : block_list_) {
1949     if (bb->hidden == true) {
1950       continue;
1951     }
1952 
1953     num_blocks_to_process += 1u;
1954 
1955     if (bb->predecessors.size() == 0u) {
1956       // Add entry block to the queue.
1957       q.push(bb);
1958     }
1959   }
1960 
1961   // Clear the topological order arrays.
1962   topological_order_.clear();
1963   topological_order_.reserve(num_blocks);
1964   topological_order_loop_ends_.clear();
1965   topological_order_loop_ends_.resize(num_blocks, 0u);
1966   topological_order_indexes_.clear();
1967   topological_order_indexes_.resize(num_blocks, static_cast<uint16_t>(-1));
1968 
1969   // Mark all blocks as unvisited.
1970   ClearAllVisitedFlags();
1971 
1972   // For loop heads, keep track from which blocks they are reachable not going through other
1973   // loop heads. Other loop heads are excluded to detect the heads of nested loops. The children
1974   // in this set go into the loop body, the other children are jumping over the loop.
1975   ScopedArenaVector<ArenaBitVector*> loop_head_reachable_from(allocator.Adapter());
1976   loop_head_reachable_from.resize(num_blocks, nullptr);
1977   // Reuse the same temp stack whenever calculating a loop_head_reachable_from[loop_head_id].
1978   ScopedArenaVector<BasicBlockId> tmp_stack(allocator.Adapter());
1979 
1980   while (num_blocks_to_process != 0u) {
1981     BasicBlock* bb = nullptr;
1982     if (!q.empty()) {
1983       num_blocks_to_process -= 1u;
1984       // Get top.
1985       bb = q.front();
1986       q.pop();
1987       if (bb->visited) {
1988         // Loop head: it was already processed, mark end and copy exit blocks to the queue.
1989         DCHECK(q.empty()) << PrettyMethod(cu_->method_idx, *cu_->dex_file);
1990         uint16_t idx = static_cast<uint16_t>(topological_order_.size());
1991         topological_order_loop_ends_[topological_order_indexes_[bb->id]] = idx;
1992         DCHECK_EQ(loop_head_stack.back(), bb->id);
1993         loop_head_stack.pop_back();
1994         ArenaBitVector* reachable =
1995             loop_head_stack.empty() ? nullptr : loop_head_reachable_from[loop_head_stack.back()];
1996         for (BasicBlockId candidate_id : loop_exit_blocks.Indexes()) {
1997           if (reachable == nullptr || reachable->IsBitSet(candidate_id)) {
1998             q.push(GetBasicBlock(candidate_id));
1999             // NOTE: The BitVectorSet::IndexIterator will not check the pointed-to bit again,
2000             // so clearing the bit has no effect on the iterator.
2001             loop_exit_blocks.ClearBit(candidate_id);
2002           }
2003         }
2004         continue;
2005       }
2006     } else {
2007       // Find the new loop head.
2008       AllNodesIterator iter(this);
2009       while (true) {
2010         BasicBlock* candidate = iter.Next();
2011         if (candidate == nullptr) {
2012           // We did not find a true loop head, fall back to a reachable block in any loop.
2013           ArenaBitVector* current_loop =
2014               loop_head_stack.empty() ? nullptr : loop_head_reachable_from[loop_head_stack.back()];
2015           bb = SelectTopologicalSortOrderFallBack(this, current_loop, &visited_cnt_values,
2016                                                   &allocator, &tmp_stack);
2017           DCHECK(bb != nullptr) << PrettyMethod(cu_->method_idx, *cu_->dex_file);
2018           if (kIsDebugBuild && cu_->dex_file != nullptr) {
2019             LOG(INFO) << "Topological sort order: Using fall-back in "
2020                 << PrettyMethod(cu_->method_idx, *cu_->dex_file) << " BB #" << bb->id
2021                 << " @0x" << std::hex << bb->start_offset
2022                 << ", num_blocks = " << std::dec << num_blocks;
2023           }
2024           break;
2025         }
2026         if (candidate->hidden ||                            // Hidden, or
2027             candidate->visited ||                           // already processed, or
2028             visited_cnt_values[candidate->id] == 0u ||      // no processed predecessors, or
2029             (!loop_head_stack.empty() &&                    // outside current loop.
2030              !loop_head_reachable_from[loop_head_stack.back()]->IsBitSet(candidate->id))) {
2031           continue;
2032         }
2033 
2034         for (BasicBlockId pred_id : candidate->predecessors) {
2035           BasicBlock* pred_bb = GetBasicBlock(pred_id);
2036           DCHECK(pred_bb != nullptr);
2037           if (pred_bb != candidate && !pred_bb->visited &&
2038               !pred_bb->dominators->IsBitSet(candidate->id)) {
2039             candidate = nullptr;  // Set candidate to null to indicate failure.
2040             break;
2041           }
2042         }
2043         if (candidate != nullptr) {
2044           bb = candidate;
2045           break;
2046         }
2047       }
2048       // Compute blocks from which the loop head is reachable and process those blocks first.
2049       ArenaBitVector* reachable =
2050           new (&allocator) ArenaBitVector(&allocator, num_blocks, false, kBitMapMisc);
2051       loop_head_reachable_from[bb->id] = reachable;
2052       ComputeUnvisitedReachableFrom(this, bb->id, reachable, &tmp_stack);
2053       // Now mark as loop head. (Even if it's only a fall back when we don't find a true loop.)
2054       loop_head_stack.push_back(bb->id);
2055       max_nested_loops = std::max(max_nested_loops, loop_head_stack.size());
2056     }
2057 
2058     DCHECK_EQ(bb->hidden, false);
2059     DCHECK_EQ(bb->visited, false);
2060     bb->visited = true;
2061     bb->nesting_depth = loop_head_stack.size();
2062 
2063     // Now add the basic block.
2064     uint16_t idx = static_cast<uint16_t>(topological_order_.size());
2065     topological_order_indexes_[bb->id] = idx;
2066     topological_order_.push_back(bb->id);
2067 
2068     // Update visited_cnt_values for children.
2069     ChildBlockIterator succIter(bb, this);
2070     BasicBlock* successor = succIter.Next();
2071     for ( ; successor != nullptr; successor = succIter.Next()) {
2072       if (successor->hidden) {
2073         continue;
2074       }
2075 
2076       // One more predecessor was visited.
2077       visited_cnt_values[successor->id] += 1u;
2078       if (visited_cnt_values[successor->id] == successor->predecessors.size()) {
2079         if (loop_head_stack.empty() ||
2080             loop_head_reachable_from[loop_head_stack.back()]->IsBitSet(successor->id)) {
2081           q.push(successor);
2082         } else {
2083           DCHECK(!loop_exit_blocks.IsBitSet(successor->id));
2084           loop_exit_blocks.SetBit(successor->id);
2085         }
2086       }
2087     }
2088   }
2089 
2090   // Prepare the loop head stack for iteration.
2091   topological_order_loop_head_stack_.clear();
2092   topological_order_loop_head_stack_.reserve(max_nested_loops);
2093   max_nested_loops_ = max_nested_loops;
2094   topological_order_up_to_date_ = true;
2095 }
2096 
IsExceptionBlock() const2097 bool BasicBlock::IsExceptionBlock() const {
2098   if (block_type == kExceptionHandling) {
2099     return true;
2100   }
2101   return false;
2102 }
2103 
ChildBlockIterator(BasicBlock * bb,MIRGraph * mir_graph)2104 ChildBlockIterator::ChildBlockIterator(BasicBlock* bb, MIRGraph* mir_graph)
2105     : basic_block_(bb), mir_graph_(mir_graph), visited_fallthrough_(false),
2106       visited_taken_(false), have_successors_(false) {
2107   // Check if we actually do have successors.
2108   if (basic_block_ != 0 && basic_block_->successor_block_list_type != kNotUsed) {
2109     have_successors_ = true;
2110     successor_iter_ = basic_block_->successor_blocks.cbegin();
2111   }
2112 }
2113 
Next()2114 BasicBlock* ChildBlockIterator::Next() {
2115   // We check if we have a basic block. If we don't we cannot get next child.
2116   if (basic_block_ == nullptr) {
2117     return nullptr;
2118   }
2119 
2120   // If we haven't visited fallthrough, return that.
2121   if (visited_fallthrough_ == false) {
2122     visited_fallthrough_ = true;
2123 
2124     BasicBlock* result = mir_graph_->GetBasicBlock(basic_block_->fall_through);
2125     if (result != nullptr) {
2126       return result;
2127     }
2128   }
2129 
2130   // If we haven't visited taken, return that.
2131   if (visited_taken_ == false) {
2132     visited_taken_ = true;
2133 
2134     BasicBlock* result = mir_graph_->GetBasicBlock(basic_block_->taken);
2135     if (result != nullptr) {
2136       return result;
2137     }
2138   }
2139 
2140   // We visited both taken and fallthrough. Now check if we have successors we need to visit.
2141   if (have_successors_ == true) {
2142     // Get information about next successor block.
2143     auto end = basic_block_->successor_blocks.cend();
2144     while (successor_iter_ != end) {
2145       SuccessorBlockInfo* successor_block_info = *successor_iter_;
2146       ++successor_iter_;
2147       // If block was replaced by zero block, take next one.
2148       if (successor_block_info->block != NullBasicBlockId) {
2149         return mir_graph_->GetBasicBlock(successor_block_info->block);
2150       }
2151     }
2152   }
2153 
2154   // We do not have anything.
2155   return nullptr;
2156 }
2157 
Copy(CompilationUnit * c_unit)2158 BasicBlock* BasicBlock::Copy(CompilationUnit* c_unit) {
2159   MIRGraph* mir_graph = c_unit->mir_graph.get();
2160   return Copy(mir_graph);
2161 }
2162 
Copy(MIRGraph * mir_graph)2163 BasicBlock* BasicBlock::Copy(MIRGraph* mir_graph) {
2164   BasicBlock* result_bb = mir_graph->CreateNewBB(block_type);
2165 
2166   // We don't do a memcpy style copy here because it would lead to a lot of things
2167   // to clean up. Let us do it by hand instead.
2168   // Copy in taken and fallthrough.
2169   result_bb->fall_through = fall_through;
2170   result_bb->taken = taken;
2171 
2172   // Copy successor links if needed.
2173   ArenaAllocator* arena = mir_graph->GetArena();
2174 
2175   result_bb->successor_block_list_type = successor_block_list_type;
2176   if (result_bb->successor_block_list_type != kNotUsed) {
2177     result_bb->successor_blocks.reserve(successor_blocks.size());
2178     for (SuccessorBlockInfo* sbi_old : successor_blocks) {
2179       SuccessorBlockInfo* sbi_new = static_cast<SuccessorBlockInfo*>(
2180           arena->Alloc(sizeof(SuccessorBlockInfo), kArenaAllocSuccessor));
2181       memcpy(sbi_new, sbi_old, sizeof(SuccessorBlockInfo));
2182       result_bb->successor_blocks.push_back(sbi_new);
2183     }
2184   }
2185 
2186   // Copy offset, method.
2187   result_bb->start_offset = start_offset;
2188 
2189   // Now copy instructions.
2190   for (MIR* mir = first_mir_insn; mir != 0; mir = mir->next) {
2191     // Get a copy first.
2192     MIR* copy = mir->Copy(mir_graph);
2193 
2194     // Append it.
2195     result_bb->AppendMIR(copy);
2196   }
2197 
2198   return result_bb;
2199 }
2200 
Copy(MIRGraph * mir_graph)2201 MIR* MIR::Copy(MIRGraph* mir_graph) {
2202   MIR* res = mir_graph->NewMIR();
2203   *res = *this;
2204 
2205   // Remove links
2206   res->next = nullptr;
2207   res->bb = NullBasicBlockId;
2208   res->ssa_rep = nullptr;
2209 
2210   return res;
2211 }
2212 
Copy(CompilationUnit * c_unit)2213 MIR* MIR::Copy(CompilationUnit* c_unit) {
2214   return Copy(c_unit->mir_graph.get());
2215 }
2216 
GetStartUseIndex(Instruction::Code opcode)2217 uint32_t SSARepresentation::GetStartUseIndex(Instruction::Code opcode) {
2218   // Default result.
2219   int res = 0;
2220 
2221   // We are basically setting the iputs to their igets counterparts.
2222   switch (opcode) {
2223     case Instruction::IPUT:
2224     case Instruction::IPUT_OBJECT:
2225     case Instruction::IPUT_BOOLEAN:
2226     case Instruction::IPUT_BYTE:
2227     case Instruction::IPUT_CHAR:
2228     case Instruction::IPUT_SHORT:
2229     case Instruction::IPUT_QUICK:
2230     case Instruction::IPUT_OBJECT_QUICK:
2231     case Instruction::IPUT_BOOLEAN_QUICK:
2232     case Instruction::IPUT_BYTE_QUICK:
2233     case Instruction::IPUT_CHAR_QUICK:
2234     case Instruction::IPUT_SHORT_QUICK:
2235     case Instruction::APUT:
2236     case Instruction::APUT_OBJECT:
2237     case Instruction::APUT_BOOLEAN:
2238     case Instruction::APUT_BYTE:
2239     case Instruction::APUT_CHAR:
2240     case Instruction::APUT_SHORT:
2241     case Instruction::SPUT:
2242     case Instruction::SPUT_OBJECT:
2243     case Instruction::SPUT_BOOLEAN:
2244     case Instruction::SPUT_BYTE:
2245     case Instruction::SPUT_CHAR:
2246     case Instruction::SPUT_SHORT:
2247       // Skip the VR containing what to store.
2248       res = 1;
2249       break;
2250     case Instruction::IPUT_WIDE:
2251     case Instruction::IPUT_WIDE_QUICK:
2252     case Instruction::APUT_WIDE:
2253     case Instruction::SPUT_WIDE:
2254       // Skip the two VRs containing what to store.
2255       res = 2;
2256       break;
2257     default:
2258       // Do nothing in the general case.
2259       break;
2260   }
2261 
2262   return res;
2263 }
2264 
2265 /**
2266  * @brief Given a decoded instruction, it checks whether the instruction
2267  * sets a constant and if it does, more information is provided about the
2268  * constant being set.
2269  * @param ptr_value pointer to a 64-bit holder for the constant.
2270  * @param wide Updated by function whether a wide constant is being set by bytecode.
2271  * @return Returns false if the decoded instruction does not represent a constant bytecode.
2272  */
GetConstant(int64_t * ptr_value,bool * wide) const2273 bool MIR::DecodedInstruction::GetConstant(int64_t* ptr_value, bool* wide) const {
2274   bool sets_const = true;
2275   int64_t value = vB;
2276 
2277   DCHECK(ptr_value != nullptr);
2278   DCHECK(wide != nullptr);
2279 
2280   switch (opcode) {
2281     case Instruction::CONST_4:
2282     case Instruction::CONST_16:
2283     case Instruction::CONST:
2284       *wide = false;
2285       value <<= 32;      // In order to get the sign extend.
2286       value >>= 32;
2287       break;
2288     case Instruction::CONST_HIGH16:
2289       *wide = false;
2290       value <<= 48;      // In order to get the sign extend.
2291       value >>= 32;
2292       break;
2293     case Instruction::CONST_WIDE_16:
2294     case Instruction::CONST_WIDE_32:
2295       *wide = true;
2296       value <<= 32;      // In order to get the sign extend.
2297       value >>= 32;
2298       break;
2299     case Instruction::CONST_WIDE:
2300       *wide = true;
2301       value = vB_wide;
2302       break;
2303     case Instruction::CONST_WIDE_HIGH16:
2304       *wide = true;
2305       value <<= 48;      // In order to get the sign extend.
2306       break;
2307     default:
2308       sets_const = false;
2309       break;
2310   }
2311 
2312   if (sets_const) {
2313     *ptr_value = value;
2314   }
2315 
2316   return sets_const;
2317 }
2318 
ResetOptimizationFlags(uint16_t reset_flags)2319 void BasicBlock::ResetOptimizationFlags(uint16_t reset_flags) {
2320   // Reset flags for all MIRs in bb.
2321   for (MIR* mir = first_mir_insn; mir != nullptr; mir = mir->next) {
2322     mir->optimization_flags &= (~reset_flags);
2323   }
2324 }
2325 
Kill(MIRGraph * mir_graph)2326 void BasicBlock::Kill(MIRGraph* mir_graph) {
2327   for (BasicBlockId pred_id : predecessors) {
2328     BasicBlock* pred_bb = mir_graph->GetBasicBlock(pred_id);
2329     DCHECK(pred_bb != nullptr);
2330 
2331     // Sadly we have to go through the children by hand here.
2332     pred_bb->ReplaceChild(id, NullBasicBlockId);
2333   }
2334   predecessors.clear();
2335 
2336   // Mark as dead and hidden.
2337   block_type = kDead;
2338   hidden = true;
2339 
2340   // Detach it from its MIRs so we don't generate code for them. Also detached MIRs
2341   // are updated to know that they no longer have a parent.
2342   for (MIR* mir = first_mir_insn; mir != nullptr; mir = mir->next) {
2343     mir->bb = NullBasicBlockId;
2344   }
2345   first_mir_insn = nullptr;
2346   last_mir_insn = nullptr;
2347 
2348   data_flow_info = nullptr;
2349 
2350   // Erase this bb from all children's predecessors and kill unreachable children.
2351   ChildBlockIterator iter(this, mir_graph);
2352   for (BasicBlock* succ_bb = iter.Next(); succ_bb != nullptr; succ_bb = iter.Next()) {
2353     succ_bb->ErasePredecessor(id);
2354   }
2355 
2356   // Remove links to children.
2357   fall_through = NullBasicBlockId;
2358   taken = NullBasicBlockId;
2359   successor_block_list_type = kNotUsed;
2360 
2361   if (kIsDebugBuild) {
2362     if (catch_entry) {
2363       DCHECK_EQ(mir_graph->catches_.count(start_offset), 1u);
2364       mir_graph->catches_.erase(start_offset);
2365     }
2366   }
2367 }
2368 
IsSSALiveOut(const CompilationUnit * c_unit,int ssa_reg)2369 bool BasicBlock::IsSSALiveOut(const CompilationUnit* c_unit, int ssa_reg) {
2370   // In order to determine if the ssa reg is live out, we scan all the MIRs. We remember
2371   // the last SSA number of the same dalvik register. At the end, if it is different than ssa_reg,
2372   // then it is not live out of this BB.
2373   int dalvik_reg = c_unit->mir_graph->SRegToVReg(ssa_reg);
2374 
2375   int last_ssa_reg = -1;
2376 
2377   // Walk through the MIRs backwards.
2378   for (MIR* mir = first_mir_insn; mir != nullptr; mir = mir->next) {
2379     // Get ssa rep.
2380     SSARepresentation *ssa_rep = mir->ssa_rep;
2381 
2382     // Go through the defines for this MIR.
2383     for (int i = 0; i < ssa_rep->num_defs; i++) {
2384       DCHECK(ssa_rep->defs != nullptr);
2385 
2386       // Get the ssa reg.
2387       int def_ssa_reg = ssa_rep->defs[i];
2388 
2389       // Get dalvik reg.
2390       int def_dalvik_reg = c_unit->mir_graph->SRegToVReg(def_ssa_reg);
2391 
2392       // Compare dalvik regs.
2393       if (dalvik_reg == def_dalvik_reg) {
2394         // We found a def of the register that we are being asked about.
2395         // Remember it.
2396         last_ssa_reg = def_ssa_reg;
2397       }
2398     }
2399   }
2400 
2401   if (last_ssa_reg == -1) {
2402     // If we get to this point we couldn't find a define of register user asked about.
2403     // Let's assume the user knows what he's doing so we can be safe and say that if we
2404     // couldn't find a def, it is live out.
2405     return true;
2406   }
2407 
2408   // If it is not -1, we found a match, is it ssa_reg?
2409   return (ssa_reg == last_ssa_reg);
2410 }
2411 
ReplaceChild(BasicBlockId old_bb,BasicBlockId new_bb)2412 bool BasicBlock::ReplaceChild(BasicBlockId old_bb, BasicBlockId new_bb) {
2413   // We need to check taken, fall_through, and successor_blocks to replace.
2414   bool found = false;
2415   if (taken == old_bb) {
2416     taken = new_bb;
2417     found = true;
2418   }
2419 
2420   if (fall_through == old_bb) {
2421     fall_through = new_bb;
2422     found = true;
2423   }
2424 
2425   if (successor_block_list_type != kNotUsed) {
2426     for (SuccessorBlockInfo* successor_block_info : successor_blocks) {
2427       if (successor_block_info->block == old_bb) {
2428         successor_block_info->block = new_bb;
2429         found = true;
2430       }
2431     }
2432   }
2433 
2434   return found;
2435 }
2436 
ErasePredecessor(BasicBlockId old_pred)2437 void BasicBlock::ErasePredecessor(BasicBlockId old_pred) {
2438   auto pos = std::find(predecessors.begin(), predecessors.end(), old_pred);
2439   DCHECK(pos != predecessors.end());
2440   // It's faster to move the back() to *pos than erase(pos).
2441   *pos = predecessors.back();
2442   predecessors.pop_back();
2443   size_t idx = std::distance(predecessors.begin(), pos);
2444   for (MIR* mir = first_mir_insn; mir != nullptr; mir = mir->next) {
2445     if (static_cast<int>(mir->dalvikInsn.opcode) != kMirOpPhi) {
2446       break;
2447     }
2448     DCHECK_EQ(mir->ssa_rep->num_uses - 1u, predecessors.size());
2449     DCHECK_EQ(mir->meta.phi_incoming[idx], old_pred);
2450     mir->meta.phi_incoming[idx] = mir->meta.phi_incoming[predecessors.size()];
2451     mir->ssa_rep->uses[idx] = mir->ssa_rep->uses[predecessors.size()];
2452     mir->ssa_rep->num_uses = predecessors.size();
2453   }
2454 }
2455 
UpdatePredecessor(BasicBlockId old_pred,BasicBlockId new_pred)2456 void BasicBlock::UpdatePredecessor(BasicBlockId old_pred, BasicBlockId new_pred) {
2457   DCHECK_NE(new_pred, NullBasicBlockId);
2458   auto pos = std::find(predecessors.begin(), predecessors.end(), old_pred);
2459   DCHECK(pos != predecessors.end());
2460   *pos = new_pred;
2461   size_t idx = std::distance(predecessors.begin(), pos);
2462   for (MIR* mir = first_mir_insn; mir != nullptr; mir = mir->next) {
2463     if (static_cast<int>(mir->dalvikInsn.opcode) != kMirOpPhi) {
2464       break;
2465     }
2466     DCHECK_EQ(mir->meta.phi_incoming[idx], old_pred);
2467     mir->meta.phi_incoming[idx] = new_pred;
2468   }
2469 }
2470 
2471 // Create a new basic block with block_id as num_blocks_ that is
2472 // post-incremented.
CreateNewBB(BBType block_type)2473 BasicBlock* MIRGraph::CreateNewBB(BBType block_type) {
2474   BasicBlockId id = static_cast<BasicBlockId>(block_list_.size());
2475   BasicBlock* res = NewMemBB(block_type, id);
2476   block_list_.push_back(res);
2477   return res;
2478 }
2479 
CalculateBasicBlockInformation(const PassManager * const post_opt_pass_manager)2480 void MIRGraph::CalculateBasicBlockInformation(const PassManager* const post_opt_pass_manager) {
2481   /* Create the pass driver and launch it */
2482   PassDriverMEPostOpt driver(post_opt_pass_manager, cu_);
2483   driver.Launch();
2484 }
2485 
FlagsOf() const2486 int MIR::DecodedInstruction::FlagsOf() const {
2487   // Calculate new index.
2488   int idx = static_cast<int>(opcode) - kNumPackedOpcodes;
2489 
2490   // Check if it is an extended or not.
2491   if (idx < 0) {
2492     return Instruction::FlagsOf(opcode);
2493   }
2494 
2495   // For extended, we use a switch.
2496   switch (static_cast<int>(opcode)) {
2497     case kMirOpPhi:
2498       return Instruction::kContinue;
2499     case kMirOpCopy:
2500       return Instruction::kContinue;
2501     case kMirOpFusedCmplFloat:
2502       return Instruction::kContinue | Instruction::kBranch;
2503     case kMirOpFusedCmpgFloat:
2504       return Instruction::kContinue | Instruction::kBranch;
2505     case kMirOpFusedCmplDouble:
2506       return Instruction::kContinue | Instruction::kBranch;
2507     case kMirOpFusedCmpgDouble:
2508       return Instruction::kContinue | Instruction::kBranch;
2509     case kMirOpFusedCmpLong:
2510       return Instruction::kContinue | Instruction::kBranch;
2511     case kMirOpNop:
2512       return Instruction::kContinue;
2513     case kMirOpNullCheck:
2514       return Instruction::kContinue | Instruction::kThrow;
2515     case kMirOpRangeCheck:
2516       return Instruction::kContinue | Instruction::kThrow;
2517     case kMirOpDivZeroCheck:
2518       return Instruction::kContinue | Instruction::kThrow;
2519     case kMirOpCheck:
2520       return Instruction::kContinue | Instruction::kThrow;
2521     case kMirOpSelect:
2522       return Instruction::kContinue;
2523     case kMirOpConstVector:
2524       return Instruction::kContinue;
2525     case kMirOpMoveVector:
2526       return Instruction::kContinue;
2527     case kMirOpPackedMultiply:
2528       return Instruction::kContinue;
2529     case kMirOpPackedAddition:
2530       return Instruction::kContinue;
2531     case kMirOpPackedSubtract:
2532       return Instruction::kContinue;
2533     case kMirOpPackedShiftLeft:
2534       return Instruction::kContinue;
2535     case kMirOpPackedSignedShiftRight:
2536       return Instruction::kContinue;
2537     case kMirOpPackedUnsignedShiftRight:
2538       return Instruction::kContinue;
2539     case kMirOpPackedAnd:
2540       return Instruction::kContinue;
2541     case kMirOpPackedOr:
2542       return Instruction::kContinue;
2543     case kMirOpPackedXor:
2544       return Instruction::kContinue;
2545     case kMirOpPackedAddReduce:
2546       return Instruction::kContinue;
2547     case kMirOpPackedReduce:
2548       return Instruction::kContinue;
2549     case kMirOpPackedSet:
2550       return Instruction::kContinue;
2551     case kMirOpReserveVectorRegisters:
2552       return Instruction::kContinue;
2553     case kMirOpReturnVectorRegisters:
2554       return Instruction::kContinue;
2555     case kMirOpMemBarrier:
2556       return Instruction::kContinue;
2557     case kMirOpPackedArrayGet:
2558       return Instruction::kContinue | Instruction::kThrow;
2559     case kMirOpPackedArrayPut:
2560       return Instruction::kContinue | Instruction::kThrow;
2561     case kMirOpMaddInt:
2562     case kMirOpMsubInt:
2563     case kMirOpMaddLong:
2564     case kMirOpMsubLong:
2565       return Instruction::kContinue;
2566     default:
2567       LOG(WARNING) << "ExtendedFlagsOf: Unhandled case: " << static_cast<int> (opcode);
2568       return 0;
2569   }
2570 }
2571 
GetInsns(int m_unit_index) const2572 const uint16_t* MIRGraph::GetInsns(int m_unit_index) const {
2573   return m_units_[m_unit_index]->GetCodeItem()->insns_;
2574 }
2575 
SetPuntToInterpreter(bool val)2576 void MIRGraph::SetPuntToInterpreter(bool val) {
2577   punt_to_interpreter_ = val;
2578   if (val) {
2579     // Disable all subsequent optimizations. They may not be safe to run. (For example,
2580     // LVN/GVN assumes there are no conflicts found by the type inference pass.)
2581     cu_->disable_opt = ~static_cast<decltype(cu_->disable_opt)>(0);
2582   }
2583 }
2584 
2585 }  // namespace art
2586