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