OLD | NEW |
---|---|
1 // Copyright (c) 2012, the Dart project authors. Please see the AUTHORS file | 1 // Copyright (c) 2012, the Dart project authors. Please see the AUTHORS file |
2 // for details. All rights reserved. Use of this source code is governed by a | 2 // for details. All rights reserved. Use of this source code is governed by a |
3 // BSD-style license that can be found in the LICENSE file. | 3 // BSD-style license that can be found in the LICENSE file. |
4 | 4 |
5 #include "vm/flow_graph_allocator.h" | 5 #include "vm/flow_graph_allocator.h" |
6 | 6 |
7 #include "vm/bit_vector.h" | 7 #include "vm/bit_vector.h" |
8 #include "vm/intermediate_language.h" | 8 #include "vm/intermediate_language.h" |
9 #include "vm/il_printer.h" | 9 #include "vm/il_printer.h" |
10 #include "vm/flow_graph_builder.h" | 10 #include "vm/flow_graph_builder.h" |
(...skipping 10 matching lines...) Expand all Loading... | |
21 #define TRACE_ALLOC(m) do { \ | 21 #define TRACE_ALLOC(m) do { \ |
22 if (FLAG_trace_ssa_allocator) OS::Print m ; \ | 22 if (FLAG_trace_ssa_allocator) OS::Print m ; \ |
23 } while (0) | 23 } while (0) |
24 #else | 24 #else |
25 #define TRACE_ALLOC(m) | 25 #define TRACE_ALLOC(m) |
26 #endif | 26 #endif |
27 | 27 |
28 | 28 |
29 static const intptr_t kNoVirtualRegister = -1; | 29 static const intptr_t kNoVirtualRegister = -1; |
30 static const intptr_t kTempVirtualRegister = -2; | 30 static const intptr_t kTempVirtualRegister = -2; |
31 static UseInterval* const kPermanentlyBlocked = | |
32 reinterpret_cast<UseInterval*>(-1); | |
33 static const intptr_t kIllegalPosition = -1; | 31 static const intptr_t kIllegalPosition = -1; |
34 static const intptr_t kMaxPosition = 0x7FFFFFFF; | 32 static const intptr_t kMaxPosition = 0x7FFFFFFF; |
35 | 33 |
36 | 34 |
35 static intptr_t MinPosition(intptr_t a, intptr_t b) { | |
36 return (a < b) ? a : b; | |
37 } | |
38 | |
39 | |
40 static bool IsParallelMovePosition(intptr_t pos) { | |
41 return (pos & 1) == 0; | |
42 } | |
43 | |
44 | |
45 static bool IsInstructionPosition(intptr_t pos) { | |
46 return (pos & 1) == 1; | |
47 } | |
48 | |
49 | |
50 static intptr_t ToParallelMove(intptr_t pos) { | |
51 return (pos & ~1); | |
52 } | |
53 | |
37 FlowGraphAllocator::FlowGraphAllocator( | 54 FlowGraphAllocator::FlowGraphAllocator( |
38 const GrowableArray<BlockEntryInstr*>& block_order, | 55 const GrowableArray<BlockEntryInstr*>& block_order, |
39 FlowGraphBuilder* builder) | 56 FlowGraphBuilder* builder) |
40 : builder_(builder), | 57 : builder_(builder), |
41 block_order_(block_order), | 58 block_order_(block_order), |
42 postorder_(builder->postorder_block_entries()), | 59 postorder_(builder->postorder_block_entries()), |
43 live_out_(block_order.length()), | 60 live_out_(block_order.length()), |
44 kill_(block_order.length()), | 61 kill_(block_order.length()), |
45 live_in_(block_order.length()), | 62 live_in_(block_order.length()), |
46 vreg_count_(builder->current_ssa_temp_index()), | 63 vreg_count_(builder->current_ssa_temp_index()), |
47 live_ranges_(builder->current_ssa_temp_index()) { | 64 live_ranges_(builder->current_ssa_temp_index()) { |
48 for (intptr_t i = 0; i < vreg_count_; i++) live_ranges_.Add(NULL); | 65 for (intptr_t i = 0; i < vreg_count_; i++) live_ranges_.Add(NULL); |
49 | 66 |
50 for (intptr_t reg = 0; reg < kNumberOfCpuRegisters; reg++) { | 67 for (intptr_t reg = 0; reg < kNumberOfCpuRegisters; reg++) { |
51 cpu_regs_[reg] = NULL; | 68 blocked_cpu_regs_[reg] = false; |
52 } | 69 } |
53 | 70 |
54 cpu_regs_[CTX] = kPermanentlyBlocked; | 71 blocked_cpu_regs_[CTX] = true; |
55 if (TMP != kNoRegister) { | 72 if (TMP != kNoRegister) { |
56 cpu_regs_[TMP] = kPermanentlyBlocked; | 73 blocked_cpu_regs_[TMP] = true; |
57 } | 74 } |
58 cpu_regs_[SPREG] = kPermanentlyBlocked; | 75 blocked_cpu_regs_[SPREG] = true; |
59 cpu_regs_[FPREG] = kPermanentlyBlocked; | 76 blocked_cpu_regs_[FPREG] = true; |
60 } | 77 } |
61 | 78 |
62 | 79 |
63 void FlowGraphAllocator::ComputeInitialSets() { | 80 void FlowGraphAllocator::ComputeInitialSets() { |
64 const intptr_t block_count = postorder_.length(); | 81 const intptr_t block_count = postorder_.length(); |
65 for (intptr_t i = 0; i < block_count; i++) { | 82 for (intptr_t i = 0; i < block_count; i++) { |
66 BlockEntryInstr* block = postorder_[i]; | 83 BlockEntryInstr* block = postorder_[i]; |
67 | 84 |
68 BitVector* kill = kill_[i]; | 85 BitVector* kill = kill_[i]; |
69 BitVector* live_in = live_in_[i]; | 86 BitVector* live_in = live_in_[i]; |
(...skipping 136 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... | |
206 } | 223 } |
207 OS::Print("\n"); | 224 OS::Print("\n"); |
208 | 225 |
209 PrintBitVector(" live out", live_out_[i]); | 226 PrintBitVector(" live out", live_out_[i]); |
210 PrintBitVector(" kill", kill_[i]); | 227 PrintBitVector(" kill", kill_[i]); |
211 PrintBitVector(" live in", live_in_[i]); | 228 PrintBitVector(" live in", live_in_[i]); |
212 } | 229 } |
213 } | 230 } |
214 | 231 |
215 | 232 |
216 void UseInterval::Print() { | 233 void LiveRange::AddUse(intptr_t pos, |
217 OS::Print(" [%d, %d) uses {", start_, end_); | 234 Location* location_slot) { |
srdjan
2012/07/22 23:27:33
Args should fit into one line.
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
218 for (UsePosition* use_pos = uses_; | 235 ASSERT((first_use_interval_->start_ <= pos) && |
219 use_pos != NULL && use_pos->pos() <= end(); | 236 (pos <= first_use_interval_->end_)); |
220 use_pos = use_pos->next()) { | |
221 if (use_pos != uses_) OS::Print(", "); | |
222 OS::Print("%d", use_pos->pos()); | |
223 } | |
224 OS::Print("}\n"); | |
225 } | |
226 | |
227 | |
228 void UseInterval::AddUse(Instruction* instr, | |
229 intptr_t pos, | |
230 Location* location_slot) { | |
231 ASSERT((start_ <= pos) && (pos <= end_)); | |
232 ASSERT((instr == NULL) || (instr->lifetime_position() == pos)); | |
233 if ((uses_ != NULL) && (uses_->pos() == pos)) { | 237 if ((uses_ != NULL) && (uses_->pos() == pos)) { |
234 if ((location_slot == NULL) || (uses_->location_slot() == location_slot)) { | 238 if ((location_slot == NULL) || (uses_->location_slot() == location_slot)) { |
235 return; | 239 return; |
236 } else if ((uses_->location_slot() == NULL) && (instr == NULL)) { | 240 } else if (uses_->location_slot() == NULL) { |
237 uses_->set_location_slot(location_slot); | 241 uses_->set_location_slot(location_slot); |
238 return; | 242 return; |
239 } | 243 } |
240 } | 244 } |
241 uses_ = new UsePosition(instr, pos, uses_, location_slot); | 245 uses_ = new UsePosition(pos, uses_, location_slot); |
242 } | 246 } |
243 | 247 |
244 | 248 |
245 void LiveRange::Print() { | 249 void LiveRange::AddUseInterval(intptr_t start, intptr_t end) { |
246 OS::Print("vreg %d live intervals:\n", vreg_); | 250 if (first_use_interval_ != NULL) { |
srdjan
2012/07/22 23:27:33
Add comment: do not add if the first_use_interval_
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Added comment.
| |
247 for (UseInterval* interval = head_; | 251 if (first_use_interval_->start_ == start) { |
248 interval != NULL; | 252 ASSERT(end <= first_use_interval_->end_); |
249 interval = interval->next_) { | 253 return; |
250 interval->Print(); | 254 } else if (first_use_interval_->start_ == end) { |
255 first_use_interval_->start_ = start; | |
srdjan
2012/07/22 23:27:33
Do you want to do that even if start < first_use_i
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
This is not shrinking this is expansion. UseInterv
| |
256 return; | |
257 } | |
258 } | |
259 | |
260 first_use_interval_ = new UseInterval(start, end, first_use_interval_); | |
261 if (last_use_interval_ == NULL) last_use_interval_ = first_use_interval_; | |
srdjan
2012/07/22 23:27:33
ASSERT(last_use_interval_->next() == NULL) thus ma
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
262 } | |
263 | |
264 | |
265 void LiveRange::DefineAt(intptr_t pos) { | |
266 if (first_use_interval_ != NULL) { | |
srdjan
2012/07/22 23:27:33
I would revert, test for positive first, makes it
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
267 ASSERT(first_use_interval_->start_ <= pos); | |
268 first_use_interval_->start_ = pos; | |
269 } else { | |
270 // Definition without a use. | |
271 first_use_interval_ = new UseInterval(pos, pos + 1, NULL); | |
272 last_use_interval_ = first_use_interval_; | |
251 } | 273 } |
252 } | 274 } |
253 | 275 |
254 | 276 |
255 void LiveRange::AddUseInterval(intptr_t start, intptr_t end) { | |
256 if ((head_ != NULL) && (head_->start_ == end)) { | |
257 head_->start_ = start; | |
258 return; | |
259 } | |
260 | |
261 head_ = new UseInterval(vreg_, start, end, head_); | |
262 } | |
263 | |
264 | |
265 void LiveRange::DefineAt(Instruction* instr, intptr_t pos, Location* loc) { | |
266 if (head_ != NULL) { | |
267 ASSERT(head_->start_ <= pos); | |
268 head_->start_ = pos; | |
269 } else { | |
270 // Definition without a use. | |
271 head_ = new UseInterval(vreg_, pos, pos + 1, NULL); | |
272 } | |
273 head_->AddUse(instr, pos, loc); | |
274 } | |
275 | |
276 | |
277 // TODO(vegorov): encode use_at_start vs. use_at_end in the location itself? | |
278 void LiveRange::UseAt(Instruction* instr, | |
279 intptr_t def, intptr_t use, | |
280 bool use_at_end, | |
281 Location* loc) { | |
282 if (head_ == NULL || head_->start_ != def) { | |
283 AddUseInterval(def, use + (use_at_end ? 1 : 0)); | |
284 } | |
285 head_->AddUse(instr, use, loc); | |
286 } | |
287 | |
288 | |
289 LiveRange* FlowGraphAllocator::GetLiveRange(intptr_t vreg) { | 277 LiveRange* FlowGraphAllocator::GetLiveRange(intptr_t vreg) { |
290 if (live_ranges_[vreg] == NULL) { | 278 if (live_ranges_[vreg] == NULL) { |
291 live_ranges_[vreg] = new LiveRange(vreg); | 279 live_ranges_[vreg] = new LiveRange(vreg); |
292 } | 280 } |
293 return live_ranges_[vreg]; | 281 return live_ranges_[vreg]; |
294 } | 282 } |
295 | 283 |
296 | 284 |
297 void FlowGraphAllocator::BlockLocation(Location loc, intptr_t pos) { | 285 void FlowGraphAllocator::BlockLocation(Location loc, |
286 intptr_t from, | |
287 intptr_t to) { | |
298 ASSERT(loc.IsRegister()); | 288 ASSERT(loc.IsRegister()); |
299 const Register reg = loc.reg(); | 289 const Register reg = loc.reg(); |
300 UseInterval* last = cpu_regs_[reg]; | 290 if (blocked_cpu_regs_[reg]) return; |
srdjan
2012/07/22 23:27:33
== NULL
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
blocked_cpu_regs_[reg] is a bool value.
| |
301 if (last == kPermanentlyBlocked) return; | 291 if (cpu_regs_[reg].length() == 0) { |
302 if ((last != NULL) && (last->start() == pos)) return; | 292 cpu_regs_[reg].Add(new LiveRange(kNoVirtualRegister)); |
303 cpu_regs_[reg] = new UseInterval(kNoVirtualRegister, pos, pos + 1, last); | 293 } |
304 } | 294 cpu_regs_[reg][0]->AddUseInterval(from, to); |
305 | 295 } |
306 | 296 |
307 void FlowGraphAllocator::Define(Instruction* instr, | 297 |
308 intptr_t pos, | 298 void LiveRange::Print() { |
309 intptr_t vreg, | 299 OS::Print(" live range v%d [%d, %d)\n", vreg(), Start(), End()); |
310 Location* loc) { | 300 UsePosition* use_pos = uses_; |
311 LiveRange* range = GetLiveRange(vreg); | 301 for (UseInterval* interval = first_use_interval_; |
312 ASSERT(loc != NULL); | 302 interval != NULL; |
313 if (loc->IsRegister()) { | 303 interval = interval->next()) { |
314 BlockLocation(*loc, pos); | 304 OS::Print(" use interval [%d, %d)\n", |
315 range->DefineAt(instr, pos + 1, loc); | 305 interval->start(), |
316 } else if (loc->IsUnallocated()) { | 306 interval->end()); |
317 range->DefineAt(instr, pos, loc); | 307 while ((use_pos != NULL) && (use_pos->pos() <= interval->end())) { |
318 } else { | 308 OS::Print(" use at %d as %s\n", |
319 UNREACHABLE(); | 309 use_pos->pos(), |
320 } | 310 (use_pos->location_slot() == NULL) |
321 | 311 ? "-" : use_pos->location_slot()->Name()); |
322 AddToUnallocated(range->head()); | 312 use_pos = use_pos->next(); |
323 } | 313 } |
324 | 314 } |
325 | 315 |
326 void FlowGraphAllocator::UseValue(Instruction* instr, | 316 if (next_sibling() != NULL) { |
327 intptr_t def_pos, | 317 next_sibling()->Print(); |
328 intptr_t use_pos, | |
329 intptr_t vreg, | |
330 Location* loc, | |
331 bool use_at_end) { | |
332 LiveRange* range = GetLiveRange(vreg); | |
333 if (loc == NULL) { | |
334 range->UseAt(NULL, def_pos, use_pos, true, loc); | |
335 } else if (loc->IsRegister()) { | |
336 // We have a fixed use. | |
337 BlockLocation(*loc, use_pos); | |
338 range->UseAt(instr, def_pos, use_pos, false, loc); | |
339 } else if (loc->IsUnallocated()) { | |
340 ASSERT(loc->policy() == Location::kRequiresRegister); | |
341 range->UseAt(use_at_end ? NULL : instr, def_pos, use_pos, use_at_end, loc); | |
342 } | |
343 } | |
344 | |
345 | |
346 static void PrintChain(UseInterval* chain) { | |
347 if (chain == kPermanentlyBlocked) { | |
348 OS::Print(" not for allocation\n"); | |
349 return; | |
350 } | |
351 | |
352 while (chain != NULL) { | |
353 chain->Print(); | |
354 chain = chain->next(); | |
355 } | 318 } |
356 } | 319 } |
357 | 320 |
358 | 321 |
359 void FlowGraphAllocator::PrintLiveRanges() { | 322 void FlowGraphAllocator::PrintLiveRanges() { |
360 for (intptr_t i = 0; i < unallocated_.length(); i++) { | 323 for (intptr_t i = 0; i < unallocated_.length(); i++) { |
361 OS::Print("unallocated chain for vr%d\n", unallocated_[i]->vreg()); | 324 unallocated_[i]->Print(); |
362 PrintChain(unallocated_[i]); | |
363 } | 325 } |
364 | 326 |
365 for (intptr_t reg = 0; reg < kNumberOfCpuRegisters; reg++) { | 327 for (intptr_t reg = 0; reg < kNumberOfCpuRegisters; reg++) { |
366 OS::Print("blocking chain for %s\n", | 328 if (blocked_cpu_regs_[reg]) continue; |
329 if (cpu_regs_[reg].length() == 0) continue; | |
330 | |
331 ASSERT(cpu_regs_[reg].length() == 1); | |
332 OS::Print("blocking live range for %s\n", | |
367 Location::RegisterLocation(static_cast<Register>(reg)).Name()); | 333 Location::RegisterLocation(static_cast<Register>(reg)).Name()); |
368 PrintChain(cpu_regs_[reg]); | 334 cpu_regs_[reg][0]->Print(); |
369 } | 335 } |
336 } | |
337 | |
338 | |
339 Instruction* FlowGraphAllocator::ConnectOutgoingPhiMoves( | |
340 BlockEntryInstr* block) { | |
srdjan
2012/07/22 15:09:24
indent 4 spaces
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
341 Instruction* last = block->last_instruction(); | |
342 | |
343 GotoInstr* goto_instr = last->AsGoto(); | |
344 if (goto_instr == NULL) return last; | |
srdjan
2012/07/22 23:27:33
What about branches?
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Branches cannot have outgoing phi moves this viola
| |
345 | |
346 // If we have a parallel move here then the successor block must be a | |
347 // join with phis. The phi inputs contribute uses to each predecessor | |
348 // block (and the phi outputs contribute definitions in the successor | |
349 // block). | |
350 ParallelMoveInstr* parallel_move = goto_instr->previous()->AsParallelMove(); | |
351 if (parallel_move == NULL) return last->previous(); | |
srdjan
2012/07/22 23:27:33
Since last can survive here only as goto_instr, re
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
352 | |
353 // All uses are recorded at the position of parallel move preceding goto. | |
354 const intptr_t pos = goto_instr->lifetime_position() - 1; | |
355 ASSERT((pos >= 0) && IsParallelMovePosition(pos)); | |
356 | |
357 JoinEntryInstr* join = goto_instr->successor(); | |
358 ASSERT(join != NULL); | |
359 | |
360 // Search for the index of the current block in the predecessors of | |
361 // the join. | |
362 // TODO(kmillikin): record the predecessor index in the goto when | |
363 // building the predecessor list to avoid this search. | |
364 intptr_t pred_idx = 0; | |
365 for (; pred_idx < join->PredecessorCount(); pred_idx++) { | |
366 if (join->PredecessorAt(pred_idx) == block) break; | |
367 } | |
368 ASSERT(pred_idx < join->PredecessorCount()); | |
srdjan
2012/07/22 23:27:33
class JoinInstr has a method that does the above:
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
I guess this code predates the method. Thanks.
| |
369 | |
370 // Record the corresponding phi input use for each phi. | |
371 ZoneGrowableArray<PhiInstr*>* phis = join->phis(); | |
372 intptr_t move_idx = 0; | |
373 for (intptr_t phi_idx = 0; phi_idx < phis->length(); phi_idx++) { | |
374 PhiInstr* phi = (*phis)[phi_idx]; | |
375 if (phi == NULL) continue; | |
376 | |
377 Value* val = phi->InputAt(pred_idx); | |
378 MoveOperands* move = parallel_move->MoveOperandsAt(move_idx); | |
379 if (val->IsUse()) { | |
380 // Expected shape of live ranges (m is parallel move, g is goto | |
381 // instruction): | |
382 // | |
383 // m g | |
384 // value --* | |
385 // | |
srdjan
2012/07/22 15:09:24
I do not understand the comment above. I see you a
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
386 | |
387 LiveRange* range = GetLiveRange( | |
388 val->AsUse()->definition()->ssa_temp_index()); | |
srdjan
2012/07/22 15:09:24
4 spaces indent. 2 spaces is used only when starti
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
389 | |
390 Location* slot = move->src_slot(); | |
391 *slot = Location::PrefersRegister(); | |
392 | |
393 range->AddUseInterval(block->start_pos(), pos); | |
394 range->AddUse(pos, slot); | |
srdjan
2012/07/22 23:27:33
Rewrite:
move->set_src(Location::Prefersregister(
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
395 } else { | |
396 ASSERT(val->IsConstant()); | |
397 move->set_src(Location::Constant(val->AsConstant()->value())); | |
398 } | |
399 move_idx++; | |
400 } | |
401 | |
402 // Begin backward iteration with the instruction before the parallel | |
403 // move. | |
404 return parallel_move->previous(); | |
405 } | |
406 | |
407 | |
408 void FlowGraphAllocator::ConnectIncommingPhiMoves(BlockEntryInstr* block) { | |
srdjan
2012/07/22 23:27:33
s/Incomming/Incoming/
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
409 // If this block is a join we need to add destinations of phi | |
410 // resolution moves to phi's live range so that register allocator will | |
411 // fill them with moves. | |
412 JoinEntryInstr* join = block->AsJoinEntry(); | |
413 if (join == NULL) return; | |
414 | |
415 // All uses are recorded at the start position in the block. | |
416 const intptr_t pos = block->start_pos(); | |
417 | |
srdjan
2012/07/22 23:27:33
s/block/join/
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
418 ZoneGrowableArray<PhiInstr*>* phis = join->phis(); | |
419 if (phis != NULL) { | |
420 intptr_t move_idx = 0; | |
421 for (intptr_t j = 0; j < phis->length(); j++) { | |
422 PhiInstr* phi = (*phis)[j]; | |
423 if (phi == NULL) continue; | |
424 | |
425 const intptr_t vreg = phi->ssa_temp_index(); | |
426 ASSERT(vreg != -1); | |
427 | |
428 // Expected shape of live range (B is the block start): | |
429 // | |
430 // B | |
431 // phi [-------- | |
432 // | |
srdjan
2012/07/22 15:09:24
ditto
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
433 | |
434 LiveRange* range = GetLiveRange(vreg); | |
435 range->DefineAt(pos); // Shorten live range. | |
436 | |
437 for (intptr_t k = 0; k < phi->InputCount(); k++) { | |
438 BlockEntryInstr* pred = block->PredecessorAt(k); | |
439 ASSERT(pred->last_instruction()->IsGoto()); | |
440 Instruction* move_instr = pred->last_instruction()->previous(); | |
441 ASSERT(move_instr->IsParallelMove()); | |
442 | |
443 MoveOperands* move = | |
444 move_instr->AsParallelMove()->MoveOperandsAt(move_idx); | |
srdjan
2012/07/22 23:27:33
indent 4
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
445 move->set_dest(Location::PrefersRegister()); | |
446 range->AddUse(pos, move->dest_slot()); | |
447 } | |
448 | |
449 // All phi resolution moves are connected. Phi's live range is | |
450 // complete. | |
451 AddToUnallocated(range); | |
452 | |
453 move_idx++; | |
454 } | |
455 } | |
456 } | |
457 | |
458 | |
459 void FlowGraphAllocator::ProcessOneInstruction(BlockEntryInstr* block, | |
460 Instruction* current) { | |
srdjan
2012/07/22 23:27:33
ProcessOneInstruction is not self explanatory, pel
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
461 const intptr_t pos = current->lifetime_position(); | |
462 ASSERT(IsInstructionPosition(pos)); | |
463 | |
464 LocationSummary* locs = current->locs(); | |
465 | |
466 // TODO(vegorov): number of inputs must match number of input locations. | |
467 if (locs->input_count() != current->InputCount()) { | |
468 builder_->Bailout("ssa allocator: number of input locations mismatch"); | |
469 } | |
470 | |
471 const bool output_same_as_first_input = | |
472 locs->out().IsUnallocated() && | |
473 (locs->out().policy() == Location::kSameAsFirstInput); | |
srdjan
2012/07/22 23:27:33
Indent 4
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
474 | |
475 // Add uses from the deoptimization environment. | |
476 if (current->env() != NULL) { | |
477 // Any value mentioned in the deoptimization environment should survive | |
478 // until the end of instruction but it does not need to be in the register. | |
479 // Expected shape of live range: | |
480 // | |
481 // m i m | |
482 // value -----*--) | |
483 // | |
484 | |
485 const GrowableArray<Value*>& values = current->env()->values(); | |
486 GrowableArray<Location>* locations = current->env()->locations(); | |
487 | |
488 for (intptr_t j = 0; j < values.length(); j++) { | |
489 Value* val = values[j]; | |
490 if (val->IsUse()) { | |
491 locations->Add(Location::Any()); | |
492 const intptr_t vreg = val->AsUse()->definition()->ssa_temp_index(); | |
493 | |
494 LiveRange* range = GetLiveRange(vreg); | |
495 range->AddUseInterval(block->start_pos(), pos + 1); | |
496 range->AddUse(pos, &(*locations)[j]); | |
497 } else { | |
498 ASSERT(val->IsConstant()); | |
499 locations->Add(Location::NoLocation()); | |
500 } | |
501 } | |
502 } | |
503 | |
504 // Process inputs. | |
505 // Skip the first input if output is specified with kSameAsFirstInput policy, | |
506 // they will be processed together at the very end. | |
507 for (intptr_t j = output_same_as_first_input ? 1 : 0; | |
508 j < current->InputCount(); | |
509 j++) { | |
510 Value* input = current->InputAt(j); | |
511 ASSERT(input->IsUse()); // Can not be a constant currently. | |
512 | |
513 const intptr_t vreg = input->AsUse()->definition()->ssa_temp_index(); | |
514 LiveRange* range = GetLiveRange(vreg); | |
515 | |
516 Location* in_ref = locs->in_slot(j); | |
517 | |
518 if (in_ref->IsRegister()) { | |
519 // Input is expected in a fixed register. Expected shape of | |
520 // live ranges: | |
521 // | |
522 // m i m | |
523 // value --* | |
524 // register [-----) | |
525 // | |
526 MoveOperands* move = | |
527 AddMoveAt(pos - 1, *in_ref, Location::PrefersRegister()); | |
srdjan
2012/07/22 23:27:33
4 spaces
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
528 BlockLocation(*in_ref, pos - 1, pos + 1); | |
529 range->AddUseInterval(block->start_pos(), pos - 1); | |
530 range->AddUse(pos - 1, move->src_slot()); | |
531 } else { | |
532 // Normal unallocated input. Expected shape of | |
533 // live ranges: | |
534 // | |
535 // m i m | |
536 // value -----*--) | |
537 // | |
538 ASSERT(in_ref->IsUnallocated()); | |
539 range->AddUseInterval(block->start_pos(), pos + 1); | |
540 range->AddUse(pos, in_ref); | |
541 } | |
542 } | |
543 | |
544 // Process temps. | |
545 for (intptr_t j = 0; j < locs->temp_count(); j++) { | |
546 // Expected shape of live range: | |
547 // | |
548 // m i m | |
549 // [--) | |
550 // | |
551 | |
552 Location temp = locs->temp(j); | |
553 if (temp.IsRegister()) { | |
554 BlockLocation(temp, pos, pos + 1); | |
555 } else if (temp.IsUnallocated()) { | |
556 LiveRange* range = new LiveRange(kTempVirtualRegister); | |
557 range->AddUseInterval(pos, pos + 1); | |
558 range->AddUse(pos, locs->temp_slot(j)); | |
559 AddToUnallocated(range); | |
560 } else { | |
561 UNREACHABLE(); | |
562 } | |
563 } | |
564 | |
565 // Block all allocatable registers for calls. | |
566 if (locs->is_call()) { | |
567 // Expected shape of live range: | |
568 // | |
569 // m i m | |
570 // [--) | |
571 // | |
572 | |
573 for (intptr_t reg = 0; reg < kNumberOfCpuRegisters; reg++) { | |
574 BlockLocation(Location::RegisterLocation(static_cast<Register>(reg)), | |
575 pos, | |
576 pos + 1); | |
577 } | |
578 | |
579 #ifdef DEBUG | |
580 // Verify that temps, inputs and output were specified as fixed | |
581 // locations. Every register is blocked now so attempt to | |
582 // allocate will not succeed. | |
583 for (intptr_t j = 0; j < locs->temp_count(); j++) { | |
584 ASSERT(!locs->temp(j).IsUnallocated()); | |
585 } | |
586 | |
587 for (intptr_t j = 0; j < locs->input_count(); j++) { | |
588 ASSERT(!locs->in(j).IsUnallocated()); | |
589 } | |
590 | |
591 ASSERT(!locs->out().IsUnallocated()); | |
592 #endif | |
593 } | |
594 | |
595 Definition* def = current->AsDefinition(); | |
596 if ((def == NULL) || (def->ssa_temp_index() < 0)) { | |
597 ASSERT(locs->out().IsInvalid()); | |
598 return; | |
599 } | |
600 | |
601 LiveRange* range = GetLiveRange(def->ssa_temp_index()); | |
602 Location* out = locs->out_slot(); | |
603 | |
604 // Process output and finalize its liverange. | |
605 if (out->IsRegister()) { | |
606 // Fixed output location. Expected shape of live range: | |
607 // | |
608 // m i m | |
609 // register [--) | |
610 // output [------- | |
611 // | |
612 BlockLocation(*out, pos, pos + 1); | |
613 // We need to emit move connecting out to some other place. | |
614 // Interesting case is: fixed output + fixed input last use: | |
615 | |
616 UsePosition* use = range->first_use(); | |
617 if (use->pos() == (pos + 1)) { | |
618 // We have a use position on the parallel move. | |
619 ASSERT(use->location_slot()->IsUnallocated()); | |
620 *(use->location_slot()) = *out; | |
621 } | |
622 | |
623 // Shorten live range to the point of definition, this might make the range | |
624 // empty (if the only use immediately follows). If range is not empty add | |
625 // move from a fixed register to an unallocated location. | |
626 range->DefineAt(pos + 1); | |
627 if (range->Start() == range->End()) return; | |
628 | |
629 MoveOperands* move = AddMoveAt(pos + 1, Location::PrefersRegister(), *out); | |
630 range->AddUse(pos - 1, move->dest_slot()); | |
631 } else if (output_same_as_first_input) { | |
632 // Output register will contain a value of the first input at instruction's | |
633 // start. Expected shape of live ranges: | |
634 // | |
635 // m i m | |
636 // input #0 --* | |
637 // output [--*---- | |
638 // | |
639 ASSERT(locs->in_slot(0)->Equals(Location::RequiresRegister())); | |
640 | |
641 // Create move that will copy value between input and output. | |
642 locs->set_out(Location::RequiresRegister()); | |
643 MoveOperands* move = AddMoveAt(pos - 1, | |
644 Location::RequiresRegister(), | |
645 Location::PrefersRegister()); | |
646 | |
647 // Add uses to the live range of the input. | |
648 Value* input = current->InputAt(0); | |
649 ASSERT(input->IsUse()); // Can not be a constant currently. | |
650 LiveRange* input_range = GetLiveRange( | |
651 input->AsUse()->definition()->ssa_temp_index()); | |
652 input_range->AddUseInterval(block->start_pos(), pos - 1); | |
653 input_range->AddUse(pos - 1, move->src_slot()); | |
654 | |
655 // Shorten output live range to the point of definition and add both input | |
656 // and output uses slots to be filled by allocator. | |
657 range->DefineAt(pos - 1); | |
658 range->AddUse(pos - 1, out); | |
659 range->AddUse(pos - 1, move->dest_slot()); | |
660 range->AddUse(pos, locs->in_slot(0)); | |
661 } else { | |
662 // Normal unallocated location that requires a register. Expected shape of | |
663 // live range: | |
664 // | |
665 // m i m | |
666 // output [------- | |
667 // | |
668 ASSERT(out->IsUnallocated() && | |
669 (out->policy() == Location::kRequiresRegister)); | |
670 | |
671 // Shorten live range to the point of definition and add use to be filled by | |
672 // allocator. | |
673 range->DefineAt(pos); | |
674 range->AddUse(pos, out); | |
675 } | |
676 | |
677 AddToUnallocated(range); | |
370 } | 678 } |
371 | 679 |
372 | 680 |
373 void FlowGraphAllocator::BuildLiveRanges() { | 681 void FlowGraphAllocator::BuildLiveRanges() { |
374 NumberInstructions(); | 682 NumberInstructions(); |
375 | 683 |
376 const intptr_t block_count = postorder_.length(); | 684 const intptr_t block_count = postorder_.length(); |
377 for (intptr_t i = 0; i < block_count; i++) { | 685 ASSERT(postorder_[block_count - 1]->IsGraphEntry()); |
686 for (intptr_t i = 0; i < (block_count - 1); i++) { | |
378 BlockEntryInstr* block = postorder_[i]; | 687 BlockEntryInstr* block = postorder_[i]; |
379 | 688 |
380 // For every SSA value that is live out of this block create an interval | 689 // For every SSA value that is live out of this block create an interval |
381 // that covers the hole block. It will be shortened if we encounter a | 690 // that covers the hole block. It will be shortened if we encounter a |
382 // definition of this value in this block. | 691 // definition of this value in this block. |
383 for (BitVector::Iterator it(live_out_[i]); !it.Done(); it.Advance()) { | 692 for (BitVector::Iterator it(live_out_[i]); !it.Done(); it.Advance()) { |
384 LiveRange* range = GetLiveRange(it.Current()); | 693 LiveRange* range = GetLiveRange(it.Current()); |
385 range->AddUseInterval(block->start_pos(), block->end_pos()); | 694 range->AddUseInterval(block->start_pos(), block->end_pos()); |
386 } | 695 } |
387 | 696 |
388 // Position corresponding to the beginning of the last instruction in the | 697 // Connect outgoing phi-moves that were created in NumberInstructions |
389 // block. | 698 // and find last instruction that contributes to liveness. |
390 intptr_t pos = block->end_pos() - 1; | 699 Instruction* current = ConnectOutgoingPhiMoves(block); |
391 Instruction* current = block->last_instruction(); | 700 |
392 | 701 // Now process all instructions in reverse order. |
393 // Goto instructions do not contribute liveness information. | 702 while (current != block) { |
394 GotoInstr* goto_instr = current->AsGoto(); | 703 // Skip parallel moves that we insert while processing instructions. |
395 if (goto_instr != NULL) { | 704 if (!current->IsParallelMove()) { |
705 ProcessOneInstruction(block, current); | |
706 } | |
396 current = current->previous(); | 707 current = current->previous(); |
397 // If we have a parallel move here then the successor block must be a | 708 } |
398 // join with phis. The phi inputs contribute uses to each predecessor | 709 |
399 // block (and the phi outputs contribute definitions in the successor | 710 ConnectIncommingPhiMoves(block); |
400 // block). | 711 } |
401 // | 712 } |
402 // We record those uses at the end of the instruction preceding the | 713 |
403 // parallel move. This position is 'pos', because we do not assign | 714 |
404 // instruction numbers to parallel moves. | 715 static ParallelMoveInstr* CreateParallelMoveBefore(Instruction* instr, |
405 ParallelMoveInstr* parallel_move = current->AsParallelMove(); | 716 intptr_t pos) { |
406 if (parallel_move != NULL) { | 717 Instruction* prev = instr->previous(); |
407 JoinEntryInstr* join = goto_instr->successor(); | 718 ParallelMoveInstr* move = prev->AsParallelMove(); |
408 ASSERT(join != NULL); | 719 if ((move == NULL) || (move->lifetime_position() != pos)) { |
409 | 720 move = new ParallelMoveInstr(); |
410 // Search for the index of the current block in the predecessors of | 721 move->set_next(prev->next()); |
411 // the join. | 722 prev->set_next(move); |
412 // TODO(kmillikin): record the predecessor index in the goto when | 723 move->next()->set_previous(move); |
413 // building the predecessor list to avoid this search. | 724 move->set_previous(prev); |
414 intptr_t pred_idx = 0; | 725 move->set_lifetime_position(pos); |
415 for (; pred_idx < join->PredecessorCount(); pred_idx++) { | 726 } |
416 if (join->PredecessorAt(pred_idx) == block) break; | 727 return move; |
417 } | 728 } |
418 ASSERT(pred_idx < join->PredecessorCount()); | 729 |
419 | 730 |
420 // Record the corresponding phi input use for each phi. | 731 static ParallelMoveInstr* CreateParallelMoveAfter(Instruction* instr, |
421 ZoneGrowableArray<PhiInstr*>* phis = join->phis(); | 732 intptr_t pos) { |
422 intptr_t move_idx = 0; | 733 Instruction* next = instr->next(); |
423 for (intptr_t phi_idx = 0; phi_idx < phis->length(); phi_idx++) { | 734 if (next->IsParallelMove() && (next->lifetime_position() == pos)) { |
424 PhiInstr* phi = (*phis)[phi_idx]; | 735 return next->AsParallelMove(); |
425 if (phi == NULL) continue; | 736 } |
426 | 737 return CreateParallelMoveBefore(next, pos); |
427 Value* val = phi->InputAt(pred_idx); | 738 } |
428 MoveOperands* move = parallel_move->MoveOperandsAt(move_idx); | 739 |
429 if (val->IsUse()) { | 740 |
430 const intptr_t virtual_register = | |
431 val->AsUse()->definition()->ssa_temp_index(); | |
432 move->set_src(Location::RequiresRegister()); | |
433 GetLiveRange( | |
434 virtual_register)->head()->AddUse(NULL, pos, move->src_slot()); | |
435 } else { | |
436 ASSERT(val->IsConstant()); | |
437 move->set_src(Location::Constant(val->AsConstant()->value())); | |
438 } | |
439 move_idx++; | |
440 } | |
441 | |
442 // Begin backward iteration with the instruction before the parallel | |
443 // move. | |
444 current = current->previous(); | |
445 } | |
446 } | |
447 | |
448 // Now process all instructions in reverse order. | |
449 --pos; // 'pos' is now the start position for the current instruction. | |
450 while (current != block) { | |
451 LocationSummary* locs = current->locs(); | |
452 | |
453 const bool output_same_as_first_input = | |
454 locs->out().IsUnallocated() && | |
455 locs->out().policy() == Location::kSameAsFirstInput; | |
456 | |
457 // TODO(vegorov): number of inputs should match number of input locations. | |
458 // TODO(vegorov): generic support for writable registers? | |
459 for (intptr_t j = 0; j < current->InputCount(); j++) { | |
460 Value* input = current->InputAt(j); | |
461 if (input->IsUse()) { | |
462 const intptr_t use = input->AsUse()->definition()->ssa_temp_index(); | |
463 | |
464 Location* in_ref = (j < locs->input_count()) ? | |
465 locs->in_slot(j) : NULL; | |
466 const bool use_at_end = (j > 0) || (in_ref == NULL) || | |
467 !output_same_as_first_input; | |
468 UseValue(current, block->start_pos(), pos, use, in_ref, use_at_end); | |
469 } | |
470 } | |
471 | |
472 // Add uses from the deoptimization environment. | |
473 // TODO(vegorov): these uses should _not_ require register but for now | |
474 // they do because we don't support spilling at all. | |
475 if (current->env() != NULL) { | |
476 const GrowableArray<Value*>& values = current->env()->values(); | |
477 GrowableArray<Location>* locations = current->env()->locations(); | |
478 | |
479 for (intptr_t j = 0; j < values.length(); j++) { | |
480 Value* val = values[j]; | |
481 if (val->IsUse()) { | |
482 locations->Add(Location::RequiresRegister()); | |
483 const intptr_t use = val->AsUse()->definition()->ssa_temp_index(); | |
484 UseValue(current, | |
485 block->start_pos(), | |
486 pos, | |
487 use, | |
488 &(*locations)[j], | |
489 true); | |
490 } else { | |
491 locations->Add(Location::NoLocation()); | |
492 } | |
493 } | |
494 } | |
495 | |
496 // Process temps. | |
497 for (intptr_t j = 0; j < locs->temp_count(); j++) { | |
498 Location temp = locs->temp(j); | |
499 if (temp.IsRegister()) { | |
500 BlockLocation(temp, pos); | |
501 } else if (temp.IsUnallocated()) { | |
502 UseInterval* temp_interval = new UseInterval( | |
503 kTempVirtualRegister, pos, pos + 1, NULL); | |
504 temp_interval->AddUse(NULL, pos, locs->temp_slot(j)); | |
505 AddToUnallocated(temp_interval); | |
506 } else { | |
507 UNREACHABLE(); | |
508 } | |
509 } | |
510 | |
511 // Block all allocatable registers for calls. | |
512 if (locs->is_call()) { | |
513 for (intptr_t reg = 0; reg < kNumberOfCpuRegisters; reg++) { | |
514 BlockLocation(Location::RegisterLocation(static_cast<Register>(reg)), | |
515 pos); | |
516 } | |
517 } | |
518 | |
519 if (locs->out().IsRegister()) { | |
520 builder_->Bailout("ssa allocator: fixed outputs are not supported"); | |
521 } | |
522 | |
523 Definition* def = current->AsDefinition(); | |
524 if ((def != NULL) && (def->ssa_temp_index() >= 0)) { | |
525 Define(output_same_as_first_input ? current : NULL, | |
526 pos, | |
527 def->ssa_temp_index(), | |
528 locs->out_slot()); | |
529 } | |
530 | |
531 current = current->previous(); | |
532 pos -= 2; | |
533 } | |
534 | |
535 // If this block is a join we need to add destinations of phi | |
536 // resolution moves to phi's live range so that register allocator will | |
537 // fill them with moves. | |
538 JoinEntryInstr* join = block->AsJoinEntry(); | |
539 if (join != NULL) { | |
540 ZoneGrowableArray<PhiInstr*>* phis = join->phis(); | |
541 if (phis != NULL) { | |
542 intptr_t move_idx = 0; | |
543 for (intptr_t j = 0; j < phis->length(); j++) { | |
544 PhiInstr* phi = (*phis)[j]; | |
545 if (phi == NULL) continue; | |
546 | |
547 const intptr_t virtual_register = phi->ssa_temp_index(); | |
548 ASSERT(virtual_register != -1); | |
549 | |
550 LiveRange* range = GetLiveRange(virtual_register); | |
551 range->DefineAt(NULL, pos, NULL); | |
552 UseInterval* interval = GetLiveRange(virtual_register)->head(); | |
553 | |
554 for (intptr_t k = 0; k < phi->InputCount(); k++) { | |
555 BlockEntryInstr* pred = block->PredecessorAt(k); | |
556 ASSERT(pred->last_instruction()->IsGoto()); | |
557 Instruction* move_instr = pred->last_instruction()->previous(); | |
558 ParallelMoveInstr* pmove = move_instr->AsParallelMove(); | |
559 ASSERT(pmove != NULL); | |
560 | |
561 MoveOperands* move_operands = pmove->MoveOperandsAt(move_idx); | |
562 move_operands->set_dest(Location::RequiresRegister()); | |
563 interval->AddUse(NULL, pos, move_operands->dest_slot()); | |
564 } | |
565 | |
566 // All phi resolution moves are connected. Phi's live range is | |
567 // complete. | |
568 AddToUnallocated(interval); | |
569 | |
570 move_idx++; | |
571 } | |
572 } | |
573 } | |
574 } | |
575 } | |
576 | |
577 | |
578 // Linearize the control flow graph. The chosen order will be used by the | 741 // Linearize the control flow graph. The chosen order will be used by the |
579 // linear-scan register allocator. Number most instructions with a pair of | 742 // linear-scan register allocator. Number most instructions with a pair of |
580 // numbers representing lifetime positions. Introduce explicit parallel | 743 // numbers representing lifetime positions. Introduce explicit parallel |
581 // move instructions in the predecessors of join nodes. The moves are used | 744 // move instructions in the predecessors of join nodes. The moves are used |
582 // for phi resolution. | 745 // for phi resolution. |
583 void FlowGraphAllocator::NumberInstructions() { | 746 void FlowGraphAllocator::NumberInstructions() { |
584 intptr_t pos = 0; | 747 intptr_t pos = 0; |
585 | 748 |
586 // The basic block order is reverse postorder. | 749 // The basic block order is reverse postorder. |
587 const intptr_t block_count = postorder_.length(); | 750 const intptr_t block_count = postorder_.length(); |
588 for (intptr_t i = block_count - 1; i >= 0; i--) { | 751 for (intptr_t i = block_count - 1; i >= 0; i--) { |
589 BlockEntryInstr* block = postorder_[i]; | 752 BlockEntryInstr* block = postorder_[i]; |
753 | |
754 instructions_.Add(block); | |
590 block->set_start_pos(pos); | 755 block->set_start_pos(pos); |
591 block->set_lifetime_position(pos); | 756 block->set_lifetime_position(pos + 1); |
592 pos += 2; | 757 pos += 2; |
758 | |
593 for (ForwardInstructionIterator it(block); !it.Done(); it.Advance()) { | 759 for (ForwardInstructionIterator it(block); !it.Done(); it.Advance()) { |
594 Instruction* current = it.Current(); | 760 Instruction* current = it.Current(); |
595 // Do not assign numbers to parallel moves or goto instructions. | 761 // Do not assign numbers to parallel move instructions. |
596 if (!current->IsParallelMove() && !current->IsGoto()) { | 762 if (!current->IsParallelMove()) { |
597 current->set_lifetime_position(pos); | 763 instructions_.Add(current); |
764 current->set_lifetime_position(pos + 1); | |
598 pos += 2; | 765 pos += 2; |
599 } | 766 } |
600 } | 767 } |
601 block->set_end_pos(pos); | 768 block->set_end_pos(pos); |
602 | 769 |
603 // For join entry predecessors create phi resolution moves if | 770 // For join entry predecessors create phi resolution moves if |
604 // necessary. They will be populated by the register allocator. | 771 // necessary. They will be populated by the register allocator. |
605 JoinEntryInstr* join = block->AsJoinEntry(); | 772 JoinEntryInstr* join = block->AsJoinEntry(); |
606 if ((join != NULL) && (join->phi_count() > 0)) { | 773 if ((join != NULL) && (join->phi_count() > 0)) { |
607 const intptr_t phi_count = join->phi_count(); | 774 const intptr_t phi_count = join->phi_count(); |
608 for (intptr_t i = 0; i < block->PredecessorCount(); i++) { | 775 for (intptr_t i = 0; i < block->PredecessorCount(); i++) { |
609 ParallelMoveInstr* move = new ParallelMoveInstr(); | 776 // Insert the move between the last two instructions of the |
777 // predecessor block (all such blocks have at least two instructions: | |
778 // the block entry and goto instructions.) | |
779 Instruction* last = block->PredecessorAt(i)->last_instruction(); | |
780 ParallelMoveInstr* move = | |
781 CreateParallelMoveBefore(last, last->lifetime_position() - 1); | |
782 | |
610 // Populate the ParallelMove with empty moves. | 783 // Populate the ParallelMove with empty moves. |
611 for (intptr_t j = 0; j < phi_count; j++) { | 784 for (intptr_t j = 0; j < phi_count; j++) { |
612 move->AddMove(Location::NoLocation(), Location::NoLocation()); | 785 move->AddMove(Location::NoLocation(), Location::NoLocation()); |
613 } | 786 } |
614 | |
615 // Insert the move between the last two instructions of the | |
616 // predecessor block (all such blocks have at least two instructions: | |
617 // the block entry and goto instructions.) | |
618 BlockEntryInstr* pred = block->PredecessorAt(i); | |
619 Instruction* next = pred->last_instruction(); | |
620 Instruction* previous = next->previous(); | |
621 ASSERT(next->IsGoto()); | |
622 ASSERT(!previous->IsParallelMove()); | |
623 previous->set_next(move); | |
624 move->set_previous(previous); | |
625 move->set_next(next); | |
626 next->set_previous(move); | |
627 } | 787 } |
628 } | 788 } |
629 } | 789 } |
630 } | 790 } |
631 | 791 |
632 | 792 |
793 Instruction* FlowGraphAllocator::InstructionAt(intptr_t pos) { | |
794 return instructions_[pos >> 1]; | |
srdjan
2012/07/22 23:27:33
pos / 2 is clearer, IMHO.
Vyacheslav Egorov (Google)
2012/07/24 12:26:42
Done.
| |
795 } | |
796 | |
797 | |
798 bool FlowGraphAllocator::IsBlockEntry(intptr_t pos) { | |
799 return InstructionAt(pos)->IsBlockEntry(); | |
800 } | |
801 | |
802 | |
803 static UsePosition* FirstUseAfter(UsePosition* use, intptr_t after) { | |
804 while ((use != NULL) && (use->pos() < after)) { | |
805 use = use->next(); | |
806 } | |
807 return use; | |
808 } | |
809 | |
810 | |
811 Location AllocationFinger::FirstHint() { | |
812 UsePosition* use = first_hinted_use_; | |
813 | |
814 while (use != NULL) { | |
815 if (use->HasHint()) return use->hint(); | |
816 use = use->next(); | |
817 } | |
818 | |
819 return Location::NoLocation(); | |
820 } | |
821 | |
822 | |
823 UsePosition* AllocationFinger::FirstRegisterUse(intptr_t after) { | |
824 for (UsePosition* use = FirstUseAfter(first_register_use_, after); | |
825 use != NULL; | |
826 use = use->next()) { | |
827 Location* loc = use->location_slot(); | |
828 if ((loc != NULL) && | |
829 loc->IsUnallocated() && | |
830 (loc->policy() == Location::kRequiresRegister)) { | |
831 first_register_use_ = use; | |
832 return use; | |
833 } | |
834 } | |
835 return NULL; | |
836 } | |
837 | |
838 | |
839 UsePosition* AllocationFinger::FirstRegisterBeneficialUse(intptr_t after) { | |
840 for (UsePosition* use = FirstUseAfter(first_register_beneficial_use_, after); | |
841 use != NULL; | |
842 use = use->next()) { | |
843 Location* loc = use->location_slot(); | |
844 if ((loc != NULL) && | |
845 (loc->IsRegister() || | |
846 (loc->IsUnallocated() && loc->IsRegisterBeneficial()))) { | |
847 first_register_beneficial_use_ = use; | |
848 return use; | |
849 } | |
850 } | |
851 return NULL; | |
852 } | |
853 | |
854 | |
633 intptr_t UseInterval::Intersect(UseInterval* other) { | 855 intptr_t UseInterval::Intersect(UseInterval* other) { |
634 if (this->start() <= other->start()) { | 856 if (this->start() <= other->start()) { |
635 if (other->start() < this->end()) return other->start(); | 857 if (other->start() < this->end()) return other->start(); |
636 } else if (this->start() < other->end()) { | 858 } else if (this->start() < other->end()) { |
637 return this->start(); | 859 return this->start(); |
638 } | 860 } |
639 return kIllegalPosition; | 861 return kIllegalPosition; |
640 } | 862 } |
641 | 863 |
642 | 864 |
643 static intptr_t FirstIntersection(UseInterval* a, UseInterval* u) { | 865 static intptr_t FirstIntersection(UseInterval* a, UseInterval* u) { |
644 while (a != NULL && u != NULL) { | 866 while (a != NULL && u != NULL) { |
645 const intptr_t pos = a->Intersect(u); | 867 const intptr_t pos = a->Intersect(u); |
646 if (pos != kIllegalPosition) return pos; | 868 if (pos != kIllegalPosition) return pos; |
647 | 869 |
648 if (a->start() < u->start()) { | 870 if (a->start() < u->start()) { |
649 a = a->next_allocated(); | 871 a = a->next(); |
650 } else { | 872 } else { |
651 u = u->next(); | 873 u = u->next(); |
652 } | 874 } |
653 } | 875 } |
654 | 876 |
655 return kMaxPosition; | 877 return kMaxPosition; |
656 } | 878 } |
657 | 879 |
658 | 880 |
659 static Location LookAheadForHint(UseInterval* interval) { | 881 LiveRange* LiveRange::MakeTemp(intptr_t pos, Location* location_slot) { |
660 UsePosition* use = interval->first_use(); | 882 UNREACHABLE(); |
661 | 883 return NULL; |
662 while (use != NULL) { | |
663 if (use->HasHint()) return use->hint(); | |
664 use = use->next(); | |
665 } | |
666 | |
667 return Location::NoLocation(); | |
668 } | 884 } |
669 | 885 |
670 | 886 |
671 bool FlowGraphAllocator::AllocateFreeRegister(UseInterval* unallocated) { | 887 LiveRange* LiveRange::SplitAt(intptr_t split_pos) { |
888 if (Start() == split_pos) return this; | |
889 | |
890 // Ranges can only be connected by parallel moves. | |
891 split_pos = ToParallelMove(split_pos); | |
892 | |
893 UseInterval* interval = finger_.first_pending_use_interval(); | |
894 ASSERT(interval->start() < split_pos); | |
895 | |
896 // Corner case. We need to start over to find previous interval. | |
897 if (interval->start() == split_pos) interval = first_use_interval_; | |
898 | |
899 UseInterval* last_before_split = NULL; | |
900 while (interval->end() <= split_pos) { | |
901 last_before_split = interval; | |
902 interval = interval->next(); | |
903 } | |
904 | |
905 const bool split_at_start = (interval->start() == split_pos); | |
906 | |
907 UseInterval* first_after_split = interval; | |
908 if (!split_at_start && interval->Contains(split_pos)) { | |
909 first_after_split = new UseInterval(split_pos, | |
910 interval->end(), | |
911 interval->next()); | |
912 interval->end_ = split_pos; | |
913 interval->next_ = first_after_split; | |
914 last_before_split = interval; | |
915 } | |
916 | |
917 ASSERT(last_before_split->next() == first_after_split); | |
918 ASSERT(last_before_split->end() <= split_pos); | |
919 ASSERT(split_pos <= first_after_split->start()); | |
920 | |
921 UsePosition* last_use_before_split = NULL; | |
922 UsePosition* use = uses_; | |
923 if (split_at_start) { | |
924 while ((use != NULL) && (use->pos() < split_pos)) { | |
925 last_use_before_split = use; | |
926 use = use->next(); | |
927 } | |
928 } else { | |
929 while ((use != NULL) && (use->pos() <= split_pos)) { | |
930 last_use_before_split = use; | |
931 use = use->next(); | |
932 } | |
933 } | |
934 UsePosition* first_use_after_split = use; | |
935 | |
936 if (last_use_before_split == NULL) { | |
937 uses_ = NULL; | |
938 } else { | |
939 last_use_before_split->set_next(NULL); | |
940 } | |
941 | |
942 UseInterval* last_use_interval = (last_before_split == last_use_interval_) ? | |
943 first_after_split : last_use_interval_; | |
944 next_sibling_ = new LiveRange(vreg(), | |
945 first_use_after_split, | |
946 first_after_split, | |
947 last_use_interval, | |
948 next_sibling_); | |
949 | |
950 TRACE_ALLOC((" split sibling [%d, %d)\n", | |
951 next_sibling_->Start(), next_sibling_->End())); | |
952 | |
953 // Split sibling can only start at a parallel move. | |
954 ASSERT(IsParallelMovePosition(next_sibling_->Start())); | |
955 | |
956 last_use_interval_ = last_before_split; | |
957 last_use_interval_->next_ = NULL; | |
958 return next_sibling_; | |
959 } | |
960 | |
961 | |
962 LiveRange* FlowGraphAllocator::SplitBetween(LiveRange* range, | |
963 intptr_t from, | |
964 intptr_t to) { | |
965 // TODO(vegorov): select optimal split position based on loop structure. | |
966 TRACE_ALLOC(("split %d [%d, %d) between [%d, %d)\n", | |
967 range->vreg(), range->Start(), range->End(), from, to)); | |
968 return range->SplitAt(to); | |
969 } | |
970 | |
971 | |
972 void FlowGraphAllocator::SpillBetween(LiveRange* range, | |
973 intptr_t from, | |
974 intptr_t to) { | |
975 ASSERT(from < to); | |
976 TRACE_ALLOC(("spill %d [%d, %d) between [%d, %d)\n", | |
977 range->vreg(), range->Start(), range->End(), from, to)); | |
978 LiveRange* tail = range->SplitAt(from); | |
979 | |
980 if (tail->Start() < to) { | |
981 // There is an intersection of tail and [from, to). | |
982 LiveRange* tail_tail = SplitBetween(tail, tail->Start(), to); | |
983 Spill(tail); | |
984 AddToUnallocated(tail_tail); | |
985 } else { | |
986 // No intersection between tail and [from, to). | |
987 AddToUnallocated(tail); | |
988 } | |
989 } | |
990 | |
991 | |
992 void FlowGraphAllocator::SpillAfter(LiveRange* range, intptr_t from) { | |
993 TRACE_ALLOC(("spill %d [%d, %d) after %d\n", | |
994 range->vreg(), range->Start(), range->End(), from)); | |
995 LiveRange* tail = range->SplitAt(from); | |
996 Spill(tail); | |
997 } | |
998 | |
999 | |
1000 intptr_t FlowGraphAllocator::AllocateSpillSlotFor(LiveRange* range) { | |
1001 for (intptr_t i = 0; i < spill_slots_.length(); i++) { | |
1002 if (spill_slots_[i] <= range->Start()) { | |
1003 return i; | |
1004 } | |
1005 } | |
1006 spill_slots_.Add(0); | |
1007 return spill_slots_.length() - 1; | |
1008 } | |
1009 | |
1010 | |
1011 void FlowGraphAllocator::Spill(LiveRange* range) { | |
1012 const intptr_t spill_index = AllocateSpillSlotFor(range); | |
1013 ASSERT(spill_slots_[spill_index] < range->Start()); | |
1014 spill_slots_[spill_index] = range->End(); | |
1015 range->set_assigned_location(Location::SpillSlot(spill_index)); | |
1016 ConvertAllUses(range); | |
1017 } | |
1018 | |
1019 | |
1020 intptr_t FlowGraphAllocator::FirstIntersectionWithAllocated( | |
1021 Register reg, LiveRange* unallocated) { | |
1022 intptr_t intersection = kMaxPosition; | |
1023 for (intptr_t i = 0; i < cpu_regs_[reg].length(); i++) { | |
1024 LiveRange* allocated = cpu_regs_[reg][i]; | |
1025 if (allocated == NULL) continue; | |
1026 | |
1027 UseInterval* allocated_head = | |
1028 allocated->finger()->first_pending_use_interval(); | |
1029 if (allocated_head->start() >= intersection) continue; | |
1030 | |
1031 const intptr_t pos = FirstIntersection( | |
1032 unallocated->finger()->first_pending_use_interval(), | |
1033 allocated_head); | |
1034 if (pos < intersection) intersection = pos; | |
1035 } | |
1036 return intersection; | |
1037 } | |
1038 | |
1039 | |
1040 | |
1041 bool FlowGraphAllocator::AllocateFreeRegister(LiveRange* unallocated) { | |
672 Register candidate = kNoRegister; | 1042 Register candidate = kNoRegister; |
673 intptr_t free_until = 0; | 1043 intptr_t free_until = 0; |
674 | 1044 |
675 // If hint is available try hint first. | 1045 // If hint is available try hint first. |
676 // TODO(vegorov): ensure that phis are hinted on the backedge. | 1046 // TODO(vegorov): ensure that phis are hinted on the back edge. |
677 Location hint = LookAheadForHint(unallocated); | 1047 Location hint = unallocated->finger()->FirstHint(); |
678 if (!hint.IsInvalid()) { | 1048 if (!hint.IsInvalid()) { |
679 ASSERT(hint.IsRegister()); | 1049 ASSERT(hint.IsRegister()); |
680 | 1050 |
681 if (cpu_regs_[hint.reg()] != kPermanentlyBlocked) { | 1051 if (!blocked_cpu_regs_[hint.reg()]) { |
682 free_until = FirstIntersection(cpu_regs_[hint.reg()], unallocated); | 1052 free_until = FirstIntersectionWithAllocated(hint.reg(), unallocated); |
683 candidate = hint.reg(); | 1053 candidate = hint.reg(); |
684 } | 1054 } |
685 | 1055 |
686 TRACE_ALLOC(("found hint %s for %d: free until %d\n", | 1056 TRACE_ALLOC(("found hint %s for %d: free until %d\n", |
687 hint.Name(), unallocated->vreg(), free_until)); | 1057 hint.Name(), unallocated->vreg(), free_until)); |
688 } | 1058 } |
689 | 1059 |
690 if (free_until != kMaxPosition) { | 1060 if (free_until != kMaxPosition) { |
691 for (int reg = 0; reg < kNumberOfCpuRegisters; ++reg) { | 1061 for (intptr_t reg = 0; reg < kNumberOfCpuRegisters; ++reg) { |
692 if (cpu_regs_[reg] == NULL) { | 1062 if (!blocked_cpu_regs_[reg] && cpu_regs_[reg].length() == 0) { |
693 candidate = static_cast<Register>(reg); | 1063 candidate = static_cast<Register>(reg); |
694 free_until = kMaxPosition; | 1064 free_until = kMaxPosition; |
695 break; | 1065 break; |
696 } | 1066 } |
697 } | 1067 } |
698 } | 1068 } |
699 | 1069 |
700 ASSERT(0 <= kMaxPosition); | 1070 ASSERT(0 <= kMaxPosition); |
701 if (free_until != kMaxPosition) { | 1071 if (free_until != kMaxPosition) { |
702 for (int reg = 0; reg < kNumberOfCpuRegisters; ++reg) { | 1072 for (intptr_t reg = 0; reg < kNumberOfCpuRegisters; ++reg) { |
703 if (cpu_regs_[reg] == kPermanentlyBlocked) continue; | 1073 if (blocked_cpu_regs_[reg] || (reg == candidate)) continue; |
704 if (reg == candidate) continue; | 1074 |
705 | 1075 const intptr_t intersection = |
706 const intptr_t pos = FirstIntersection(cpu_regs_[reg], unallocated); | 1076 FirstIntersectionWithAllocated(static_cast<Register>(reg), unallocated); |
707 | 1077 |
708 if (pos > free_until) { | 1078 if (intersection > free_until) { |
709 candidate = static_cast<Register>(reg); | 1079 candidate = static_cast<Register>(reg); |
710 free_until = pos; | 1080 free_until = intersection; |
711 if (free_until == kMaxPosition) break; | 1081 if (free_until == kMaxPosition) break; |
712 } | 1082 } |
713 } | 1083 } |
714 } | 1084 } |
715 | 1085 |
1086 if (free_until != kMaxPosition) free_until = ToParallelMove(free_until); | |
1087 | |
716 // All registers are blocked by active ranges. | 1088 // All registers are blocked by active ranges. |
717 if (free_until <= unallocated->start()) return false; | 1089 if (free_until <= unallocated->Start()) return false; |
718 | 1090 |
719 AssignFreeRegister(unallocated, candidate); | 1091 TRACE_ALLOC(("assigning free register %s to %d\n", |
1092 Location::RegisterLocation(candidate).Name(), | |
1093 unallocated->vreg())); | |
1094 | |
1095 if (free_until != kMaxPosition) { | |
1096 // There was an intersection. Split unallocated. | |
1097 TRACE_ALLOC((" splitting at %d\n", free_until)); | |
1098 LiveRange* tail = unallocated->SplitAt(free_until); | |
1099 AddToUnallocated(tail); | |
1100 } | |
1101 | |
1102 cpu_regs_[candidate].Add(unallocated); | |
1103 unallocated->set_assigned_location(Location::RegisterLocation(candidate)); | |
1104 | |
720 return true; | 1105 return true; |
721 } | 1106 } |
722 | 1107 |
723 | 1108 |
724 UseInterval* UseInterval::Split(intptr_t pos) { | 1109 void FlowGraphAllocator::AllocateAnyRegister(LiveRange* unallocated) { |
725 if (pos == start()) return this; | 1110 UsePosition* register_use = |
726 ASSERT(Contains(pos)); | 1111 unallocated->finger()->FirstRegisterUse(unallocated->Start()); |
727 UseInterval* tail = new UseInterval(vreg(), pos, end(), next()); | 1112 if (register_use == NULL) { |
728 | 1113 Spill(unallocated); |
729 UsePosition* use = uses_; | 1114 return; |
730 while (use != NULL && use->pos() <= pos) { | 1115 } |
731 use = use->next(); | 1116 |
732 } | 1117 Register candidate = kNoRegister; |
733 | 1118 intptr_t free_until = 0; |
734 tail->uses_ = use; | 1119 intptr_t blocked_at = kMaxPosition; |
735 | 1120 |
736 end_ = pos; | 1121 for (int reg = 0; reg < kNumberOfCpuRegisters; ++reg) { |
737 | 1122 if (blocked_cpu_regs_[reg]) continue; |
738 return tail; | 1123 if (UpdateFreeUntil(static_cast<Register>(reg), |
739 } | 1124 unallocated, |
740 | 1125 &free_until, |
741 | 1126 &blocked_at)) { |
742 void FlowGraphAllocator::AssignFreeRegister(UseInterval* unallocated, | 1127 candidate = static_cast<Register>(reg); |
743 Register reg) { | 1128 } |
744 TRACE_ALLOC(("assigning free register %s to %d\n", | 1129 } |
1130 | |
1131 if (free_until < register_use->pos()) { | |
1132 // Can't acquire free register. Spill until we really need one. | |
1133 ASSERT(unallocated->Start() < ToParallelMove(register_use->pos())); | |
1134 SpillBetween(unallocated, unallocated->Start(), register_use->pos()); | |
1135 return; | |
1136 } | |
1137 | |
1138 if (blocked_at < unallocated->End()) { | |
1139 LiveRange* tail = SplitBetween(unallocated, | |
1140 unallocated->Start(), | |
1141 blocked_at); | |
1142 AddToUnallocated(tail); | |
1143 } | |
1144 | |
1145 AssignBlockedRegister(unallocated, candidate); | |
1146 } | |
1147 | |
1148 | |
1149 bool FlowGraphAllocator::UpdateFreeUntil(Register reg, | |
1150 LiveRange* unallocated, | |
1151 intptr_t* cur_free_until, | |
1152 intptr_t* cur_blocked_at) { | |
1153 intptr_t free_until = kMaxPosition; | |
1154 intptr_t blocked_at = kMaxPosition; | |
1155 const intptr_t start = unallocated->Start(); | |
1156 | |
1157 for (intptr_t i = 0; i < cpu_regs_[reg].length(); i++) { | |
1158 LiveRange* allocated = cpu_regs_[reg][i]; | |
1159 | |
1160 UseInterval* first_pending_use_interval = | |
1161 allocated->finger()->first_pending_use_interval(); | |
1162 if (first_pending_use_interval->Contains(start)) { | |
1163 // This is an active interval. | |
1164 if (allocated->vreg() <= 0) { | |
1165 // This register blocked by an interval that | |
1166 // can't be spilled. | |
1167 return false; | |
1168 } | |
1169 | |
1170 const UsePosition* use = | |
1171 allocated->finger()->FirstRegisterBeneficialUse(unallocated->Start()); | |
1172 | |
1173 if ((use != NULL) && ((use->pos() - start) <= 1)) { | |
1174 // This register is blocked by interval that is used | |
1175 // as register in the current instruction and can't | |
1176 // be spilled. | |
1177 return false; | |
1178 } | |
1179 | |
1180 const intptr_t use_pos = (use != NULL) ? use->pos() | |
1181 : allocated->End(); | |
1182 | |
1183 if (use_pos < free_until) free_until = use_pos; | |
1184 } else { | |
1185 // This is inactive interval. | |
1186 const intptr_t intersection = FirstIntersection( | |
1187 first_pending_use_interval, unallocated->first_use_interval()); | |
1188 if (intersection != kMaxPosition) { | |
1189 if (intersection < free_until) free_until = intersection; | |
1190 if (allocated->vreg() == kNoVirtualRegister) blocked_at = intersection; | |
1191 } | |
1192 } | |
1193 | |
1194 if (free_until <= *cur_free_until) { | |
1195 return false; | |
1196 } | |
1197 } | |
1198 | |
1199 ASSERT(free_until > *cur_free_until); | |
1200 *cur_free_until = free_until; | |
1201 *cur_blocked_at = blocked_at; | |
1202 return true; | |
1203 } | |
1204 | |
1205 | |
1206 void FlowGraphAllocator::RemoveEvicted(Register reg, intptr_t first_evicted) { | |
1207 intptr_t to = first_evicted; | |
1208 intptr_t from = first_evicted + 1; | |
1209 while (from < cpu_regs_[reg].length()) { | |
1210 LiveRange* allocated = cpu_regs_[reg][from++]; | |
1211 if (allocated != NULL) cpu_regs_[reg][to++] = allocated; | |
1212 } | |
1213 cpu_regs_[reg].TruncateTo(to); | |
1214 } | |
1215 | |
1216 | |
1217 void FlowGraphAllocator::AssignBlockedRegister(LiveRange* unallocated, | |
1218 Register reg) { | |
1219 TRACE_ALLOC(("assigning blocked register %s to live range %d\n", | |
745 Location::RegisterLocation(reg).Name(), | 1220 Location::RegisterLocation(reg).Name(), |
746 unallocated->vreg())); | 1221 unallocated->vreg())); |
747 | 1222 |
748 UseInterval* a = cpu_regs_[reg]; | 1223 intptr_t first_evicted = -1; |
749 if (a == NULL) { | 1224 for (intptr_t i = cpu_regs_[reg].length() - 1; i >= 0; i--) { |
750 // Register is completely free. | 1225 LiveRange* allocated = cpu_regs_[reg][i]; |
751 cpu_regs_[reg] = unallocated; | 1226 if (allocated->vreg() < 0) continue; // Can't be evicted. |
1227 if (EvictIntersection(allocated, | |
1228 unallocated)) { | |
1229 cpu_regs_[reg][i] = NULL; | |
1230 first_evicted = i; | |
1231 } | |
1232 } | |
1233 | |
1234 // Remove evicted ranges from the array. | |
1235 if (first_evicted != -1) RemoveEvicted(reg, first_evicted); | |
1236 | |
1237 cpu_regs_[reg].Add(unallocated); | |
1238 unallocated->set_assigned_location(Location::RegisterLocation(reg)); | |
1239 } | |
1240 | |
1241 | |
1242 bool FlowGraphAllocator::EvictIntersection(LiveRange* allocated, | |
1243 LiveRange* unallocated) { | |
1244 UseInterval* first_unallocated = | |
1245 unallocated->finger()->first_pending_use_interval(); | |
1246 const intptr_t intersection = FirstIntersection( | |
1247 allocated->finger()->first_pending_use_interval(), | |
1248 first_unallocated); | |
1249 if (intersection == kMaxPosition) return false; | |
1250 | |
1251 const intptr_t spill_position = first_unallocated->start(); | |
1252 UsePosition* use = allocated->finger()->FirstRegisterUse(spill_position); | |
1253 if (use == NULL) { | |
1254 // No register uses after this point. | |
1255 SpillAfter(allocated, spill_position); | |
1256 } else { | |
1257 const intptr_t restore_position = | |
1258 (spill_position < intersection) ? MinPosition(intersection, use->pos()) | |
1259 : use->pos(); | |
1260 | |
1261 SpillBetween(allocated, spill_position, restore_position); | |
1262 } | |
1263 | |
1264 return true; | |
1265 } | |
1266 | |
1267 | |
1268 MoveOperands* FlowGraphAllocator::AddMoveAt(intptr_t pos, | |
1269 Location to, | |
1270 Location from) { | |
1271 ASSERT(IsParallelMovePosition(pos)); | |
1272 Instruction* instr = InstructionAt(pos); | |
1273 ASSERT(!instr->IsBlockEntry()); | |
1274 return CreateParallelMoveBefore(instr, pos)->AddMove(to, from); | |
1275 } | |
1276 | |
1277 | |
1278 void FlowGraphAllocator::ConvertUseTo(UsePosition* use, Location loc) { | |
1279 ASSERT(use->location_slot() != NULL); | |
1280 Location* slot = use->location_slot(); | |
1281 ASSERT(slot->IsUnallocated()); | |
1282 ASSERT((slot->policy() == Location::kRequiresRegister) || | |
1283 (slot->policy() == Location::kPrefersRegister) || | |
1284 (slot->policy() == Location::kAny)); | |
1285 TRACE_ALLOC((" use at %d converted to %s\n", use->pos(), loc.Name())); | |
1286 *slot = loc; | |
1287 } | |
1288 | |
1289 | |
1290 void FlowGraphAllocator::ConvertAllUses(LiveRange* range) { | |
1291 if (range->vreg() == kNoVirtualRegister) return; | |
1292 TRACE_ALLOC(("range [%d, %d) for v%d has been allocated to %s:\n", | |
1293 range->Start(), | |
1294 range->End(), | |
1295 range->vreg(), | |
1296 range->assigned_location().Name())); | |
1297 ASSERT(!range->assigned_location().IsInvalid()); | |
1298 const Location loc = range->assigned_location(); | |
1299 for (UsePosition* use = range->first_use(); use != NULL; use = use->next()) { | |
1300 ConvertUseTo(use, loc); | |
1301 } | |
1302 } | |
1303 | |
1304 | |
1305 bool AllocationFinger::Advance(const intptr_t start) { | |
1306 UseInterval* a = first_pending_use_interval_; | |
1307 while (a != NULL && a->end() <= start) a = a->next(); | |
1308 first_pending_use_interval_ = a; | |
1309 if (first_pending_use_interval_ == NULL) { | |
1310 return true; | |
1311 } | |
1312 return false; | |
1313 } | |
1314 | |
1315 | |
1316 void FlowGraphAllocator::AdvanceActiveIntervals(const intptr_t start) { | |
1317 for (intptr_t reg = 0; reg < kNumberOfCpuRegisters; reg++) { | |
1318 if (cpu_regs_[reg].length() == 0) continue; | |
1319 | |
1320 intptr_t first_evicted = -1; | |
1321 for (intptr_t i = cpu_regs_[reg].length() - 1; i >= 0; i--) { | |
1322 LiveRange* range = cpu_regs_[reg][i]; | |
1323 if (range->finger()->Advance(start)) { | |
1324 ConvertAllUses(range); | |
1325 cpu_regs_[reg][i] = NULL; | |
1326 first_evicted = i; | |
1327 } | |
1328 } | |
1329 | |
1330 if (first_evicted != -1) { | |
1331 RemoveEvicted(static_cast<Register>(reg), first_evicted); | |
1332 } | |
1333 } | |
1334 } | |
1335 | |
1336 | |
1337 void AllocationFinger::Initialize(LiveRange* range) { | |
1338 first_pending_use_interval_ = range->first_use_interval(); | |
1339 first_register_use_ = range->first_use(); | |
1340 first_register_beneficial_use_ = range->first_use(); | |
1341 first_hinted_use_ = range->first_use(); | |
1342 } | |
1343 | |
1344 | |
1345 static inline bool ShouldBeAllocatedBefore(LiveRange* a, LiveRange* b) { | |
1346 return a->Start() <= b->Start(); | |
1347 } | |
1348 | |
1349 | |
1350 void FlowGraphAllocator::AddToUnallocated(LiveRange* range) { | |
1351 range->finger()->Initialize(range); | |
1352 | |
1353 if (unallocated_.is_empty()) { | |
1354 unallocated_.Add(range); | |
752 return; | 1355 return; |
753 } | 1356 } |
754 | 1357 |
755 UseInterval* u = unallocated; | 1358 for (intptr_t i = unallocated_.length() - 1; i >= 0; i--) { |
756 ASSERT(u->start() < a->start()); // Register is free. | 1359 if (ShouldBeAllocatedBefore(range, unallocated_[i])) { |
757 cpu_regs_[reg] = u; | 1360 unallocated_.InsertAt(i + 1, range); |
758 if (u->next() == NULL || u->next()->start() >= a->start()) { | |
759 u->set_next_allocated(a); | |
760 } | |
761 | |
762 while (a != NULL && u != NULL) { | |
763 const intptr_t pos = a->Intersect(u); | |
764 if (pos != kIllegalPosition) { | |
765 // TODO(vegorov): split live ranges might require control flow resolution | |
766 // which is not implemented yet. | |
767 builder_->Bailout("ssa allocator: control flow resolution required"); | |
768 | |
769 TRACE_ALLOC((" splitting at %d\n", pos)); | |
770 // Reached intersection | |
771 UseInterval* tail = u->Split(pos); | |
772 AddToUnallocated(tail); | |
773 ASSERT(tail == u || u->next_allocated() == a); | |
774 return; | 1361 return; |
775 } | 1362 } |
776 | 1363 } |
777 if (a->start() < u->start()) { | 1364 unallocated_.InsertAt(0, range); |
778 if (a->next_allocated() == NULL) { | |
779 a->set_next_allocated(u); | |
780 break; | |
781 } | |
782 | |
783 UseInterval* next = a->next_allocated(); | |
784 if (next->start() > u->start()) { | |
785 a->set_next_allocated(u); | |
786 u->set_next_allocated(next); | |
787 } | |
788 | |
789 a = next; | |
790 } else { | |
791 UseInterval* next = u->next(); | |
792 | |
793 if (next == NULL || next->start() >= a->start()) { | |
794 u->set_next_allocated(a); | |
795 } | |
796 u = next; | |
797 } | |
798 } | |
799 } | |
800 | |
801 | |
802 static void InsertMoveBefore(Instruction* instr, Location to, Location from) { | |
803 Instruction* prev = instr->previous(); | |
804 ParallelMoveInstr* move = prev->AsParallelMove(); | |
805 if (move == NULL) { | |
806 move = new ParallelMoveInstr(); | |
807 move->set_next(prev->next()); | |
808 prev->set_next(move); | |
809 move->next()->set_previous(move); | |
810 move->set_previous(prev); | |
811 } | |
812 move->AddMove(to, from); | |
813 } | |
814 | |
815 | |
816 void UsePosition::AssignLocation(Location loc) { | |
817 if (location_slot_ == NULL) return; | |
818 | |
819 if (location_slot_->IsUnallocated()) { | |
820 if (location_slot_->policy() == Location::kSameAsFirstInput) { | |
821 Instruction* instr = this->instr(); | |
822 LocationSummary* locs = instr->locs(); | |
823 if (!locs->in(0).IsUnallocated()) { | |
824 InsertMoveBefore(instr, loc, locs->in(0)); | |
825 } | |
826 locs->set_in(0, loc); | |
827 } | |
828 TRACE_ALLOC((" use at %d converted to %s\n", pos(), loc.Name())); | |
829 *location_slot_ = loc; | |
830 } else if (location_slot_->IsRegister()) { | |
831 InsertMoveBefore(this->instr(), *location_slot_, loc); | |
832 } | |
833 } | |
834 | |
835 | |
836 void FlowGraphAllocator::FinalizeInterval(UseInterval* interval, Location loc) { | |
837 if (interval->vreg() == kNoVirtualRegister) return; | |
838 | |
839 TRACE_ALLOC(("assigning location %s to interval [%d, %d)\n", loc.Name(), | |
840 interval->start(), interval->end())); | |
841 | |
842 for (UsePosition* use = interval->first_use(); | |
843 use != NULL && use->pos() <= interval->end(); | |
844 use = use->next()) { | |
845 use->AssignLocation(loc); | |
846 } | |
847 } | |
848 | |
849 | |
850 void FlowGraphAllocator::AdvanceActiveIntervals(const intptr_t start) { | |
851 for (int reg = 0; reg < kNumberOfCpuRegisters; reg++) { | |
852 if (cpu_regs_[reg] == NULL) continue; | |
853 if (cpu_regs_[reg] == kPermanentlyBlocked) continue; | |
854 | |
855 UseInterval* a = cpu_regs_[reg]; | |
856 while (a != NULL && a->end() <= start) { | |
857 FinalizeInterval(a, | |
858 Location::RegisterLocation(static_cast<Register>(reg))); | |
859 a = a->next_allocated(); | |
860 } | |
861 | |
862 cpu_regs_[reg] = a; | |
863 } | |
864 } | |
865 | |
866 | |
867 static inline bool ShouldBeAllocatedBefore(UseInterval* a, UseInterval* b) { | |
868 return a->start() <= b->start(); | |
869 } | |
870 | |
871 | |
872 void FlowGraphAllocator::AddToUnallocated(UseInterval* chain) { | |
873 if (unallocated_.is_empty()) { | |
874 unallocated_.Add(chain); | |
875 return; | |
876 } | |
877 | |
878 for (intptr_t i = unallocated_.length() - 1; i >= 0; i--) { | |
879 if (ShouldBeAllocatedBefore(chain, unallocated_[i])) { | |
880 unallocated_.InsertAt(i + 1, chain); | |
881 return; | |
882 } | |
883 } | |
884 unallocated_.InsertAt(0, chain); | |
885 } | 1365 } |
886 | 1366 |
887 | 1367 |
888 bool FlowGraphAllocator::UnallocatedIsSorted() { | 1368 bool FlowGraphAllocator::UnallocatedIsSorted() { |
889 for (intptr_t i = unallocated_.length() - 1; i >= 1; i--) { | 1369 for (intptr_t i = unallocated_.length() - 1; i >= 1; i--) { |
890 UseInterval* a = unallocated_[i]; | 1370 LiveRange* a = unallocated_[i]; |
891 UseInterval* b = unallocated_[i - 1]; | 1371 LiveRange* b = unallocated_[i - 1]; |
892 if (!ShouldBeAllocatedBefore(a, b)) return false; | 1372 if (!ShouldBeAllocatedBefore(a, b)) return false; |
893 } | 1373 } |
894 return true; | 1374 return true; |
895 } | 1375 } |
896 | 1376 |
897 | 1377 |
898 void FlowGraphAllocator::AllocateCPURegisters() { | 1378 void FlowGraphAllocator::AllocateCPURegisters() { |
899 ASSERT(UnallocatedIsSorted()); | 1379 ASSERT(UnallocatedIsSorted()); |
900 | 1380 |
1381 for (intptr_t i = 0; i < kNumberOfCpuRegisters; i++) { | |
1382 if (cpu_regs_[i].length() == 1) { | |
1383 LiveRange* range = cpu_regs_[i][0]; | |
1384 range->finger()->Initialize(range); | |
1385 } | |
1386 } | |
1387 | |
901 while (!unallocated_.is_empty()) { | 1388 while (!unallocated_.is_empty()) { |
902 UseInterval* range = unallocated_.Last(); | 1389 LiveRange* range = unallocated_.Last(); |
903 unallocated_.RemoveLast(); | 1390 unallocated_.RemoveLast(); |
904 const intptr_t start = range->start(); | 1391 const intptr_t start = range->Start(); |
905 TRACE_ALLOC(("Processing interval chain for vreg %d starting at %d\n", | 1392 TRACE_ALLOC(("Processing live range for vreg %d starting at %d\n", |
906 range->vreg(), | 1393 range->vreg(), |
907 start)); | 1394 start)); |
908 | 1395 |
909 // TODO(vegorov): eagerly spill liveranges without register uses. | 1396 // TODO(vegorov): eagerly spill liveranges without register uses. |
910 AdvanceActiveIntervals(start); | 1397 AdvanceActiveIntervals(start); |
911 | 1398 |
912 if (!AllocateFreeRegister(range)) { | 1399 if (!AllocateFreeRegister(range)) { |
913 builder_->Bailout("ssa allocator: spilling required"); | 1400 AllocateAnyRegister(range); |
914 return; | |
915 } | 1401 } |
916 } | 1402 } |
917 | 1403 |
918 // All allocation decisions were done. | 1404 // All allocation decisions were done. |
919 ASSERT(unallocated_.is_empty()); | 1405 ASSERT(unallocated_.is_empty()); |
920 | 1406 |
921 // Finish allocation. | 1407 // Finish allocation. |
922 AdvanceActiveIntervals(kMaxPosition); | 1408 AdvanceActiveIntervals(kMaxPosition); |
923 TRACE_ALLOC(("Allocation completed\n")); | 1409 TRACE_ALLOC(("Allocation completed\n")); |
924 } | 1410 } |
925 | 1411 |
926 | 1412 |
1413 void FlowGraphAllocator::ConnectSplitSiblings(LiveRange* range, | |
1414 BlockEntryInstr* source_block, | |
1415 BlockEntryInstr* target_block) { | |
1416 if (range->next_sibling() == NULL) { | |
1417 // Nothing to connect. The whole range was allocated to the same location. | |
1418 TRACE_ALLOC(("range %d has no siblings\n", range->vreg())); | |
1419 return; | |
1420 } | |
1421 | |
1422 const intptr_t source_pos = source_block->end_pos() - 1; | |
1423 ASSERT(IsInstructionPosition(source_pos)); | |
1424 | |
1425 const intptr_t target_pos = target_block->start_pos(); | |
1426 | |
1427 Location target; | |
1428 Location source; | |
1429 | |
1430 #ifdef DEBUG | |
1431 LiveRange* source_cover = NULL; | |
1432 LiveRange* target_cover = NULL; | |
1433 #endif | |
1434 | |
1435 while ((range != NULL) && (source.IsInvalid() || target.IsInvalid())) { | |
1436 if (range->CanCover(source_pos)) { | |
1437 ASSERT(source.IsInvalid()); | |
1438 source = range->assigned_location(); | |
1439 #ifdef DEBUG | |
1440 source_cover = range; | |
1441 #endif | |
1442 } | |
1443 if (range->CanCover(target_pos)) { | |
1444 ASSERT(target.IsInvalid()); | |
1445 target = range->assigned_location(); | |
1446 #ifdef DEBUG | |
1447 target_cover = range; | |
1448 #endif | |
1449 } | |
1450 | |
1451 range = range->next_sibling(); | |
1452 } | |
1453 | |
1454 TRACE_ALLOC(("connecting [%d, %d) [%s] to [%d, %d) [%s]\n", | |
1455 source_cover->Start(), source_cover->End(), source.Name(), | |
1456 target_cover->Start(), target_cover->End(), target.Name())); | |
1457 | |
1458 // Siblings were allocated to the same register. | |
1459 if (source.Equals(target)) return; | |
1460 | |
1461 Instruction* last = source_block->last_instruction(); | |
1462 if (last->SuccessorCount() == 1) { | |
1463 CreateParallelMoveBefore(last, last->lifetime_position() - 1)-> | |
1464 AddMove(target, source); | |
1465 } else { | |
1466 CreateParallelMoveAfter(target_block, target_block->start_pos())-> | |
1467 AddMove(target, source); | |
1468 } | |
1469 } | |
1470 | |
1471 | |
1472 void FlowGraphAllocator::ResolveControlFlow() { | |
1473 // Resolve linear control flow between touching split siblings | |
1474 // inside basic blocks. | |
1475 for (intptr_t vreg = 0; vreg < live_ranges_.length(); vreg++) { | |
1476 LiveRange* range = live_ranges_[vreg]; | |
1477 if (range == NULL) continue; | |
1478 | |
1479 while (range->next_sibling() != NULL) { | |
1480 LiveRange* sibling = range->next_sibling(); | |
1481 if ((range->End() == sibling->Start()) && | |
1482 !range->assigned_location().Equals(sibling->assigned_location()) && | |
1483 !IsBlockEntry(range->End())) { | |
1484 AddMoveAt(sibling->Start(), | |
1485 sibling->assigned_location(), | |
1486 range->assigned_location()); | |
1487 } | |
1488 range = sibling; | |
1489 } | |
1490 } | |
1491 | |
1492 // Resolve non-linear control flow across branches. | |
1493 for (intptr_t i = 1; i < block_order_.length(); i++) { | |
1494 BlockEntryInstr* block = block_order_[i]; | |
1495 BitVector* live = live_in_[block->postorder_number()]; | |
1496 for (BitVector::Iterator it(live); !it.Done(); it.Advance()) { | |
1497 LiveRange* range = GetLiveRange(it.Current()); | |
1498 for (intptr_t j = 0; j < block->PredecessorCount(); j++) { | |
1499 ConnectSplitSiblings(range, block->PredecessorAt(j), block); | |
1500 } | |
1501 } | |
1502 } | |
1503 } | |
1504 | |
1505 | |
927 void FlowGraphAllocator::AllocateRegisters() { | 1506 void FlowGraphAllocator::AllocateRegisters() { |
928 GraphEntryInstr* entry = block_order_[0]->AsGraphEntry(); | 1507 GraphEntryInstr* entry = block_order_[0]->AsGraphEntry(); |
929 ASSERT(entry != NULL); | 1508 ASSERT(entry != NULL); |
930 | 1509 |
931 for (intptr_t i = 0; i < entry->start_env()->values().length(); i++) { | 1510 for (intptr_t i = 0; i < entry->start_env()->values().length(); i++) { |
932 if (entry->start_env()->values()[i]->IsUse()) { | 1511 if (entry->start_env()->values()[i]->IsUse()) { |
933 builder_->Bailout("ssa allocator: unsupported start environment"); | 1512 builder_->Bailout("ssa allocator: unsupported start environment"); |
934 } | 1513 } |
935 } | 1514 } |
936 | 1515 |
937 AnalyzeLiveness(); | 1516 AnalyzeLiveness(); |
938 | 1517 |
939 BuildLiveRanges(); | 1518 BuildLiveRanges(); |
940 | 1519 |
941 if (FLAG_print_ssa_liveness) { | 1520 if (FLAG_print_ssa_liveness) { |
942 DumpLiveness(); | 1521 DumpLiveness(); |
943 } | 1522 } |
944 | 1523 |
945 if (FLAG_trace_ssa_allocator) { | 1524 if (FLAG_trace_ssa_allocator) { |
946 PrintLiveRanges(); | 1525 PrintLiveRanges(); |
947 } | 1526 } |
948 | 1527 |
949 AllocateCPURegisters(); | 1528 AllocateCPURegisters(); |
950 | 1529 |
1530 ResolveControlFlow(); | |
1531 | |
951 if (FLAG_trace_ssa_allocator) { | 1532 if (FLAG_trace_ssa_allocator) { |
952 OS::Print("-- ir after allocation -------------------------\n"); | 1533 OS::Print("-- ir after allocation -------------------------\n"); |
953 FlowGraphPrinter printer(Function::Handle(), block_order_, true); | 1534 FlowGraphPrinter printer(Function::Handle(), block_order_, true); |
954 printer.PrintBlocks(); | 1535 printer.PrintBlocks(); |
955 } | 1536 } |
956 } | 1537 } |
957 | 1538 |
958 | 1539 |
959 } // namespace dart | 1540 } // namespace dart |
OLD | NEW |