| 1 | /* |
| 2 | * Copyright (C) 2015-2016 Apple Inc. All rights reserved. |
| 3 | * |
| 4 | * Redistribution and use in source and binary forms, with or without |
| 5 | * modification, are permitted provided that the following conditions |
| 6 | * are met: |
| 7 | * 1. Redistributions of source code must retain the above copyright |
| 8 | * notice, this list of conditions and the following disclaimer. |
| 9 | * 2. Redistributions in binary form must reproduce the above copyright |
| 10 | * notice, this list of conditions and the following disclaimer in the |
| 11 | * documentation and/or other materials provided with the distribution. |
| 12 | * |
| 13 | * THIS SOFTWARE IS PROVIDED BY APPLE INC. AND ITS CONTRIBUTORS ``AS IS'' |
| 14 | * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, |
| 15 | * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
| 16 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR ITS CONTRIBUTORS |
| 17 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
| 18 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
| 19 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
| 20 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
| 21 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
| 22 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF |
| 23 | * THE POSSIBILITY OF SUCH DAMAGE. |
| 24 | */ |
| 25 | |
| 26 | #include "config.h" |
| 27 | #include "DFGStoreBarrierInsertionPhase.h" |
| 28 | |
| 29 | #if ENABLE(DFG_JIT) |
| 30 | |
| 31 | #include "DFGAbstractInterpreterInlines.h" |
| 32 | #include "DFGBlockMapInlines.h" |
| 33 | #include "DFGDoesGC.h" |
| 34 | #include "DFGGraph.h" |
| 35 | #include "DFGInPlaceAbstractState.h" |
| 36 | #include "DFGInsertionSet.h" |
| 37 | #include "DFGPhase.h" |
| 38 | #include "JSCInlines.h" |
| 39 | #include <wtf/CommaPrinter.h> |
| 40 | #include <wtf/HashSet.h> |
| 41 | |
| 42 | namespace JSC { namespace DFG { |
| 43 | |
| 44 | namespace { |
| 45 | |
| 46 | namespace DFGStoreBarrierInsertionPhaseInternal { |
| 47 | static const bool verbose = false; |
| 48 | } |
| 49 | |
| 50 | enum class PhaseMode { |
| 51 | // Does only a local analysis for store barrier insertion and assumes that pointers live |
| 52 | // from predecessor blocks may need barriers. Assumes CPS conventions. Does not use AI for |
| 53 | // eliminating store barriers, but does a best effort to eliminate barriers when you're |
| 54 | // storing a non-cell value by using Node::result() and by looking at constants. The local |
| 55 | // analysis is based on GC epochs, so it will eliminate a lot of locally redundant barriers. |
| 56 | Fast, |
| 57 | |
| 58 | // Does a global analysis for store barrier insertion. Reuses the GC-epoch-based analysis |
| 59 | // used by Fast, but adds a conservative merge rule for propagating information from one |
| 60 | // block to the next. This will ensure for example that if a value V coming from multiple |
| 61 | // predecessors in B didn't need any more barriers at the end of each predecessor (either |
| 62 | // because it was the last allocated object in that predecessor or because it just had a |
| 63 | // barrier executed), then until we hit another GC point in B, we won't need another barrier |
| 64 | // on V. Uses AI for eliminating barriers when we know that the value being stored is not a |
| 65 | // cell. Assumes SSA conventions. |
| 66 | Global |
| 67 | }; |
| 68 | |
| 69 | template<PhaseMode mode> |
| 70 | class StoreBarrierInsertionPhase : public Phase { |
| 71 | public: |
| 72 | StoreBarrierInsertionPhase(Graph& graph) |
| 73 | : Phase(graph, mode == PhaseMode::Fast ? "fast store barrier insertion" : "global store barrier insertion" ) |
| 74 | , m_insertionSet(graph) |
| 75 | { |
| 76 | } |
| 77 | |
| 78 | bool run() |
| 79 | { |
| 80 | if (DFGStoreBarrierInsertionPhaseInternal::verbose) { |
| 81 | dataLog("Starting store barrier insertion:\n" ); |
| 82 | m_graph.dump(); |
| 83 | } |
| 84 | |
| 85 | switch (mode) { |
| 86 | case PhaseMode::Fast: { |
| 87 | DFG_ASSERT(m_graph, nullptr, m_graph.m_form != SSA); |
| 88 | |
| 89 | m_graph.clearEpochs(); |
| 90 | for (BasicBlock* block : m_graph.blocksInNaturalOrder()) |
| 91 | handleBlock(block); |
| 92 | return true; |
| 93 | } |
| 94 | |
| 95 | case PhaseMode::Global: { |
| 96 | DFG_ASSERT(m_graph, nullptr, m_graph.m_form == SSA); |
| 97 | |
| 98 | m_state = std::make_unique<InPlaceAbstractState>(m_graph); |
| 99 | m_interpreter = std::make_unique<AbstractInterpreter<InPlaceAbstractState>>(m_graph, *m_state); |
| 100 | |
| 101 | m_isConverged = false; |
| 102 | |
| 103 | // First run the analysis. Inside basic blocks we use an epoch-based analysis that |
| 104 | // is very precise. At block boundaries, we just propagate which nodes may need a |
| 105 | // barrier. This gives us a very nice bottom->top fixpoint: we start out assuming |
| 106 | // that no node needs any barriers at block boundaries, and then we converge |
| 107 | // towards believing that all nodes need barriers. "Needing a barrier" is like |
| 108 | // saying that the node is in a past epoch. "Not needing a barrier" is like saying |
| 109 | // that the node is in the current epoch. |
| 110 | m_stateAtHead = std::make_unique<BlockMap<HashSet<Node*>>>(m_graph); |
| 111 | m_stateAtTail = std::make_unique<BlockMap<HashSet<Node*>>>(m_graph); |
| 112 | |
| 113 | BlockList postOrder = m_graph.blocksInPostOrder(); |
| 114 | |
| 115 | bool changed = true; |
| 116 | while (changed) { |
| 117 | changed = false; |
| 118 | |
| 119 | // Intentional backwards loop because we are using RPO. |
| 120 | for (unsigned blockIndex = postOrder.size(); blockIndex--;) { |
| 121 | BasicBlock* block = postOrder[blockIndex]; |
| 122 | |
| 123 | if (!handleBlock(block)) { |
| 124 | // If the block didn't finish, then it cannot affect the fixpoint. |
| 125 | continue; |
| 126 | } |
| 127 | |
| 128 | // Construct the state-at-tail based on the epochs of live nodes and the |
| 129 | // current epoch. We grow state-at-tail monotonically to ensure convergence. |
| 130 | bool thisBlockChanged = false; |
| 131 | for (NodeFlowProjection node : block->ssa->liveAtTail) { |
| 132 | if (node.kind() == NodeFlowProjection::Shadow) |
| 133 | continue; |
| 134 | if (node->epoch() != m_currentEpoch) { |
| 135 | // If the node is older than the current epoch, then we may need to |
| 136 | // run a barrier on it in the future. So, add it to the state. |
| 137 | thisBlockChanged |= m_stateAtTail->at(block).add(node.node()).isNewEntry; |
| 138 | } |
| 139 | } |
| 140 | |
| 141 | if (!thisBlockChanged) { |
| 142 | // This iteration didn't learn anything new about this block. |
| 143 | continue; |
| 144 | } |
| 145 | |
| 146 | // Changed things. Make sure that we loop one more time. |
| 147 | changed = true; |
| 148 | |
| 149 | for (BasicBlock* successor : block->successors()) { |
| 150 | for (Node* node : m_stateAtTail->at(block)) |
| 151 | m_stateAtHead->at(successor).add(node); |
| 152 | } |
| 153 | } |
| 154 | } |
| 155 | |
| 156 | // Tell handleBlock() that it's time to actually insert barriers for real. |
| 157 | m_isConverged = true; |
| 158 | |
| 159 | for (BasicBlock* block : m_graph.blocksInNaturalOrder()) |
| 160 | handleBlock(block); |
| 161 | |
| 162 | return true; |
| 163 | } } |
| 164 | |
| 165 | RELEASE_ASSERT_NOT_REACHED(); |
| 166 | return false; |
| 167 | } |
| 168 | |
| 169 | private: |
| 170 | bool handleBlock(BasicBlock* block) |
| 171 | { |
| 172 | if (DFGStoreBarrierInsertionPhaseInternal::verbose) { |
| 173 | dataLog("Dealing with block " , pointerDump(block), "\n" ); |
| 174 | if (reallyInsertBarriers()) |
| 175 | dataLog(" Really inserting barriers.\n" ); |
| 176 | } |
| 177 | |
| 178 | m_currentEpoch = Epoch::first(); |
| 179 | |
| 180 | if (mode == PhaseMode::Global) { |
| 181 | if (!block->cfaHasVisited) |
| 182 | return false; |
| 183 | m_state->beginBasicBlock(block); |
| 184 | |
| 185 | for (NodeFlowProjection node : block->ssa->liveAtHead) { |
| 186 | if (node.kind() == NodeFlowProjection::Shadow) |
| 187 | continue; |
| 188 | if (m_stateAtHead->at(block).contains(node.node())) { |
| 189 | // If previous blocks tell us that this node may need a barrier in the |
| 190 | // future, then put it in the ancient primordial epoch. This forces us to |
| 191 | // emit a barrier on any possibly-cell store, regardless of the epoch of the |
| 192 | // stored value. |
| 193 | node->setEpoch(Epoch()); |
| 194 | } else { |
| 195 | // If previous blocks aren't requiring us to run a barrier on this node, |
| 196 | // then put it in the current epoch. This means that we will skip barriers |
| 197 | // on this node so long as we don't allocate. It also means that we won't |
| 198 | // run barriers on stores to on one such node into another such node. That's |
| 199 | // fine, because nodes would be excluded from the state set if at the tails |
| 200 | // of all predecessors they always had the current epoch. |
| 201 | node->setEpoch(m_currentEpoch); |
| 202 | } |
| 203 | } |
| 204 | } |
| 205 | |
| 206 | bool result = true; |
| 207 | |
| 208 | for (m_nodeIndex = 0; m_nodeIndex < block->size(); ++m_nodeIndex) { |
| 209 | m_node = block->at(m_nodeIndex); |
| 210 | |
| 211 | if (DFGStoreBarrierInsertionPhaseInternal::verbose) { |
| 212 | dataLog( |
| 213 | " " , m_currentEpoch, ": Looking at node " , m_node, " with children: " ); |
| 214 | CommaPrinter comma; |
| 215 | m_graph.doToChildren( |
| 216 | m_node, |
| 217 | [&] (Edge edge) { |
| 218 | dataLog(comma, edge, " (" , edge->epoch(), ")" ); |
| 219 | }); |
| 220 | dataLog("\n" ); |
| 221 | } |
| 222 | |
| 223 | if (mode == PhaseMode::Global) { |
| 224 | // Execute edges separately because we don't want to insert barriers if the |
| 225 | // operation doing the store does a check that ensures that the child is not |
| 226 | // a cell. |
| 227 | m_interpreter->startExecuting(); |
| 228 | m_interpreter->executeEdges(m_node); |
| 229 | } |
| 230 | |
| 231 | switch (m_node->op()) { |
| 232 | case PutByValDirect: |
| 233 | case PutByVal: |
| 234 | case PutByValAlias: { |
| 235 | switch (m_node->arrayMode().modeForPut().type()) { |
| 236 | case Array::Contiguous: |
| 237 | case Array::ArrayStorage: |
| 238 | case Array::SlowPutArrayStorage: { |
| 239 | Edge child1 = m_graph.varArgChild(m_node, 0); |
| 240 | Edge child3 = m_graph.varArgChild(m_node, 2); |
| 241 | considerBarrier(child1, child3); |
| 242 | break; |
| 243 | } |
| 244 | default: |
| 245 | break; |
| 246 | } |
| 247 | break; |
| 248 | } |
| 249 | |
| 250 | case ArrayPush: { |
| 251 | switch (m_node->arrayMode().type()) { |
| 252 | case Array::Contiguous: |
| 253 | case Array::ArrayStorage: { |
| 254 | unsigned elementOffset = 2; |
| 255 | unsigned elementCount = m_node->numChildren() - elementOffset; |
| 256 | Edge& arrayEdge = m_graph.varArgChild(m_node, 1); |
| 257 | for (unsigned i = 0; i < elementCount; ++i) { |
| 258 | Edge& element = m_graph.varArgChild(m_node, i + elementOffset); |
| 259 | considerBarrier(arrayEdge, element); |
| 260 | } |
| 261 | break; |
| 262 | } |
| 263 | default: |
| 264 | break; |
| 265 | } |
| 266 | break; |
| 267 | } |
| 268 | |
| 269 | case PutById: |
| 270 | case PutByIdFlush: |
| 271 | case PutByIdDirect: |
| 272 | case PutStructure: { |
| 273 | considerBarrier(m_node->child1()); |
| 274 | break; |
| 275 | } |
| 276 | |
| 277 | case RecordRegExpCachedResult: { |
| 278 | considerBarrier(m_graph.varArgChild(m_node, 0)); |
| 279 | break; |
| 280 | } |
| 281 | |
| 282 | case PutClosureVar: |
| 283 | case PutToArguments: |
| 284 | case SetRegExpObjectLastIndex: { |
| 285 | considerBarrier(m_node->child1(), m_node->child2()); |
| 286 | break; |
| 287 | } |
| 288 | |
| 289 | case MultiPutByOffset: { |
| 290 | considerBarrier(m_node->child1()); |
| 291 | break; |
| 292 | } |
| 293 | |
| 294 | case PutByOffset: { |
| 295 | considerBarrier(m_node->child2(), m_node->child3()); |
| 296 | break; |
| 297 | } |
| 298 | |
| 299 | case PutGlobalVariable: { |
| 300 | considerBarrier(m_node->child1(), m_node->child2()); |
| 301 | break; |
| 302 | } |
| 303 | |
| 304 | case SetFunctionName: { |
| 305 | considerBarrier(m_node->child1(), m_node->child2()); |
| 306 | break; |
| 307 | } |
| 308 | |
| 309 | case NukeStructureAndSetButterfly: { |
| 310 | considerBarrier(m_node->child1()); |
| 311 | break; |
| 312 | } |
| 313 | |
| 314 | default: |
| 315 | break; |
| 316 | } |
| 317 | |
| 318 | if (doesGC(m_graph, m_node)) |
| 319 | m_currentEpoch.bump(); |
| 320 | |
| 321 | switch (m_node->op()) { |
| 322 | case NewObject: |
| 323 | case NewArray: |
| 324 | case NewArrayWithSize: |
| 325 | case NewArrayBuffer: |
| 326 | case NewTypedArray: |
| 327 | case NewRegexp: |
| 328 | case NewStringObject: |
| 329 | case NewSymbol: |
| 330 | case MaterializeNewObject: |
| 331 | case MaterializeCreateActivation: |
| 332 | case MakeRope: |
| 333 | case CreateActivation: |
| 334 | case CreateDirectArguments: |
| 335 | case CreateScopedArguments: |
| 336 | case CreateClonedArguments: |
| 337 | case NewFunction: |
| 338 | case NewGeneratorFunction: |
| 339 | case NewAsyncGeneratorFunction: |
| 340 | case NewAsyncFunction: |
| 341 | case AllocatePropertyStorage: |
| 342 | case ReallocatePropertyStorage: |
| 343 | // Nodes that allocate get to set their epoch because for those nodes we know |
| 344 | // that they will be the newest object in the heap. |
| 345 | m_node->setEpoch(m_currentEpoch); |
| 346 | break; |
| 347 | |
| 348 | case Upsilon: |
| 349 | // Assume the worst for Phis so that we don't have to worry about Phi shadows. |
| 350 | m_node->phi()->setEpoch(Epoch()); |
| 351 | m_node->setEpoch(Epoch()); |
| 352 | break; |
| 353 | |
| 354 | default: |
| 355 | // For nodes that aren't guaranteed to allocate, we say that their return value |
| 356 | // (if there is one) could be arbitrarily old. |
| 357 | m_node->setEpoch(Epoch()); |
| 358 | break; |
| 359 | } |
| 360 | |
| 361 | if (DFGStoreBarrierInsertionPhaseInternal::verbose) { |
| 362 | dataLog( |
| 363 | " " , m_currentEpoch, ": Done with node " , m_node, " (" , m_node->epoch(), |
| 364 | ") with children: " ); |
| 365 | CommaPrinter comma; |
| 366 | m_graph.doToChildren( |
| 367 | m_node, |
| 368 | [&] (Edge edge) { |
| 369 | dataLog(comma, edge, " (" , edge->epoch(), ")" ); |
| 370 | }); |
| 371 | dataLog("\n" ); |
| 372 | } |
| 373 | |
| 374 | if (mode == PhaseMode::Global) { |
| 375 | if (!m_interpreter->executeEffects(m_nodeIndex, m_node)) { |
| 376 | result = false; |
| 377 | break; |
| 378 | } |
| 379 | } |
| 380 | } |
| 381 | |
| 382 | if (mode == PhaseMode::Global) |
| 383 | m_state->reset(); |
| 384 | |
| 385 | if (reallyInsertBarriers()) |
| 386 | m_insertionSet.execute(block); |
| 387 | |
| 388 | return result; |
| 389 | } |
| 390 | |
| 391 | void considerBarrier(Edge base, Edge child) |
| 392 | { |
| 393 | if (DFGStoreBarrierInsertionPhaseInternal::verbose) |
| 394 | dataLog(" Considering adding barrier " , base, " => " , child, "\n" ); |
| 395 | |
| 396 | // We don't need a store barrier if the child is guaranteed to not be a cell. |
| 397 | switch (mode) { |
| 398 | case PhaseMode::Fast: { |
| 399 | // Don't try too hard because it's too expensive to run AI. |
| 400 | if (child->hasConstant()) { |
| 401 | if (!child->asJSValue().isCell()) { |
| 402 | if (DFGStoreBarrierInsertionPhaseInternal::verbose) |
| 403 | dataLog(" Rejecting because of constant type.\n" ); |
| 404 | return; |
| 405 | } |
| 406 | } else { |
| 407 | switch (child->result()) { |
| 408 | case NodeResultNumber: |
| 409 | case NodeResultDouble: |
| 410 | case NodeResultInt32: |
| 411 | case NodeResultInt52: |
| 412 | case NodeResultBoolean: |
| 413 | if (DFGStoreBarrierInsertionPhaseInternal::verbose) |
| 414 | dataLog(" Rejecting because of result type.\n" ); |
| 415 | return; |
| 416 | default: |
| 417 | break; |
| 418 | } |
| 419 | } |
| 420 | break; |
| 421 | } |
| 422 | |
| 423 | case PhaseMode::Global: { |
| 424 | // Go into rage mode to eliminate any chance of a barrier with a non-cell child. We |
| 425 | // can afford to keep around AI in Global mode. |
| 426 | if (!m_interpreter->needsTypeCheck(child, ~SpecCell)) { |
| 427 | if (DFGStoreBarrierInsertionPhaseInternal::verbose) |
| 428 | dataLog(" Rejecting because of AI type.\n" ); |
| 429 | return; |
| 430 | } |
| 431 | break; |
| 432 | } } |
| 433 | |
| 434 | considerBarrier(base); |
| 435 | } |
| 436 | |
| 437 | void considerBarrier(Edge base) |
| 438 | { |
| 439 | if (DFGStoreBarrierInsertionPhaseInternal::verbose) |
| 440 | dataLog(" Considering adding barrier on " , base, "\n" ); |
| 441 | |
| 442 | // We don't need a store barrier if the epoch of the base is identical to the current |
| 443 | // epoch. That means that we either just allocated the object and so it's guaranteed to |
| 444 | // be in newgen, or we just ran a barrier on it so it's guaranteed to be remembered |
| 445 | // already. |
| 446 | if (base->epoch() == m_currentEpoch) { |
| 447 | if (DFGStoreBarrierInsertionPhaseInternal::verbose) |
| 448 | dataLog(" Rejecting because it's in the current epoch.\n" ); |
| 449 | return; |
| 450 | } |
| 451 | |
| 452 | if (DFGStoreBarrierInsertionPhaseInternal::verbose) |
| 453 | dataLog(" Inserting barrier.\n" ); |
| 454 | insertBarrier(m_nodeIndex + 1, base); |
| 455 | } |
| 456 | |
| 457 | void insertBarrier(unsigned nodeIndex, Edge base) |
| 458 | { |
| 459 | // This is just our way of saying that barriers are not redundant with each other according |
| 460 | // to forward analysis: if we proved one time that a barrier was necessary then it'll for |
| 461 | // sure be necessary next time. |
| 462 | base->setEpoch(Epoch()); |
| 463 | |
| 464 | // If we're in global mode, we should only insert the barriers once we have converged. |
| 465 | if (!reallyInsertBarriers()) |
| 466 | return; |
| 467 | |
| 468 | // FIXME: We could support StoreBarrier(UntypedUse:). That would be sort of cool. |
| 469 | // But right now we don't need it. |
| 470 | |
| 471 | DFG_ASSERT(m_graph, m_node, isCell(base.useKind()), m_node->op(), base.useKind()); |
| 472 | |
| 473 | // Barriers are always inserted after the node that they service. Therefore, we always know |
| 474 | // that the thing is a cell now. |
| 475 | base.setUseKind(KnownCellUse); |
| 476 | |
| 477 | NodeOrigin origin = m_node->origin; |
| 478 | if (clobbersExitState(m_graph, m_node)) |
| 479 | origin = origin.withInvalidExit(); |
| 480 | |
| 481 | m_insertionSet.insertNode(nodeIndex, SpecNone, FencedStoreBarrier, origin, base); |
| 482 | } |
| 483 | |
| 484 | bool reallyInsertBarriers() |
| 485 | { |
| 486 | return mode == PhaseMode::Fast || m_isConverged; |
| 487 | } |
| 488 | |
| 489 | InsertionSet m_insertionSet; |
| 490 | Epoch m_currentEpoch; |
| 491 | unsigned m_nodeIndex; |
| 492 | Node* m_node; |
| 493 | |
| 494 | // Things we only use in Global mode. |
| 495 | std::unique_ptr<InPlaceAbstractState> m_state; |
| 496 | std::unique_ptr<AbstractInterpreter<InPlaceAbstractState>> m_interpreter; |
| 497 | std::unique_ptr<BlockMap<HashSet<Node*>>> m_stateAtHead; |
| 498 | std::unique_ptr<BlockMap<HashSet<Node*>>> m_stateAtTail; |
| 499 | bool m_isConverged; |
| 500 | }; |
| 501 | |
| 502 | } // anonymous namespace |
| 503 | |
| 504 | bool performFastStoreBarrierInsertion(Graph& graph) |
| 505 | { |
| 506 | return runPhase<StoreBarrierInsertionPhase<PhaseMode::Fast>>(graph); |
| 507 | } |
| 508 | |
| 509 | bool performGlobalStoreBarrierInsertion(Graph& graph) |
| 510 | { |
| 511 | return runPhase<StoreBarrierInsertionPhase<PhaseMode::Global>>(graph); |
| 512 | } |
| 513 | |
| 514 | } } // namespace JSC::DFG |
| 515 | |
| 516 | #endif // ENABLE(DFG_JIT) |
| 517 | |
| 518 | |