| 1 | /* |
| 2 | * Copyright (C) 2013-2018 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. ``AS IS'' AND ANY |
| 14 | * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 15 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
| 16 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR |
| 17 | * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
| 18 | * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
| 19 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR |
| 20 | * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY |
| 21 | * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 22 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 23 | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 24 | */ |
| 25 | |
| 26 | #include "config.h" |
| 27 | #include "DFGLICMPhase.h" |
| 28 | |
| 29 | #if ENABLE(DFG_JIT) |
| 30 | |
| 31 | #include "DFGAbstractInterpreterInlines.h" |
| 32 | #include "DFGAtTailAbstractState.h" |
| 33 | #include "DFGBasicBlockInlines.h" |
| 34 | #include "DFGClobberSet.h" |
| 35 | #include "DFGClobberize.h" |
| 36 | #include "DFGControlEquivalenceAnalysis.h" |
| 37 | #include "DFGEdgeDominates.h" |
| 38 | #include "DFGGraph.h" |
| 39 | #include "DFGInsertionSet.h" |
| 40 | #include "DFGMayExit.h" |
| 41 | #include "DFGNaturalLoops.h" |
| 42 | #include "DFGPhase.h" |
| 43 | #include "DFGSafeToExecute.h" |
| 44 | #include "JSCInlines.h" |
| 45 | |
| 46 | namespace JSC { namespace DFG { |
| 47 | |
| 48 | class LICMPhase : public Phase { |
| 49 | static const bool verbose = false; |
| 50 | |
| 51 | using NaturalLoop = SSANaturalLoop; |
| 52 | |
| 53 | struct LoopData { |
| 54 | ClobberSet writes; |
| 55 | BasicBlock* { nullptr }; |
| 56 | }; |
| 57 | |
| 58 | public: |
| 59 | LICMPhase(Graph& graph) |
| 60 | : Phase(graph, "LICM" ) |
| 61 | , m_state(graph) |
| 62 | , m_interpreter(graph, m_state) |
| 63 | { |
| 64 | } |
| 65 | |
| 66 | bool run() |
| 67 | { |
| 68 | DFG_ASSERT(m_graph, nullptr, m_graph.m_form == SSA); |
| 69 | |
| 70 | m_graph.ensureSSADominators(); |
| 71 | m_graph.ensureSSANaturalLoops(); |
| 72 | m_graph.ensureControlEquivalenceAnalysis(); |
| 73 | |
| 74 | if (verbose) { |
| 75 | dataLog("Graph before LICM:\n" ); |
| 76 | m_graph.dump(); |
| 77 | } |
| 78 | |
| 79 | m_data.resize(m_graph.m_ssaNaturalLoops->numLoops()); |
| 80 | |
| 81 | // Figure out the set of things each loop writes to, not including blocks that |
| 82 | // belong to inner loops. We fix this later. |
| 83 | for (BlockIndex blockIndex = m_graph.numBlocks(); blockIndex--;) { |
| 84 | BasicBlock* block = m_graph.block(blockIndex); |
| 85 | if (!block) |
| 86 | continue; |
| 87 | |
| 88 | // Skip blocks that are proved to not execute. |
| 89 | // FIXME: This shouldn't be needed. |
| 90 | // https://bugs.webkit.org/show_bug.cgi?id=128584 |
| 91 | if (!block->cfaHasVisited) |
| 92 | continue; |
| 93 | |
| 94 | const NaturalLoop* loop = m_graph.m_ssaNaturalLoops->innerMostLoopOf(block); |
| 95 | if (!loop) |
| 96 | continue; |
| 97 | LoopData& data = m_data[loop->index()]; |
| 98 | for (auto* node : *block) { |
| 99 | // Don't look beyond parts of the code that definitely always exit. |
| 100 | // FIXME: This shouldn't be needed. |
| 101 | // https://bugs.webkit.org/show_bug.cgi?id=128584 |
| 102 | if (node->op() == ForceOSRExit) |
| 103 | break; |
| 104 | |
| 105 | addWrites(m_graph, node, data.writes); |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | // For each loop: |
| 110 | // - Identify its pre-header. |
| 111 | // - Make sure its outer loops know what it clobbers. |
| 112 | for (unsigned loopIndex = m_graph.m_ssaNaturalLoops->numLoops(); loopIndex--;) { |
| 113 | const NaturalLoop& loop = m_graph.m_ssaNaturalLoops->loop(loopIndex); |
| 114 | LoopData& data = m_data[loop.index()]; |
| 115 | |
| 116 | for ( |
| 117 | const NaturalLoop* outerLoop = m_graph.m_ssaNaturalLoops->innerMostOuterLoop(loop); |
| 118 | outerLoop; |
| 119 | outerLoop = m_graph.m_ssaNaturalLoops->innerMostOuterLoop(*outerLoop)) |
| 120 | m_data[outerLoop->index()].writes.addAll(data.writes); |
| 121 | |
| 122 | BasicBlock* = loop.header(); |
| 123 | BasicBlock* = nullptr; |
| 124 | unsigned = 0; // We're cool if this is 1. |
| 125 | |
| 126 | // This is guaranteed because we expect the CFG not to have unreachable code. Therefore, a |
| 127 | // loop header must have a predecessor. (Also, we don't allow the root block to be a loop, |
| 128 | // which cuts out the one other way of having a loop header with only one predecessor.) |
| 129 | DFG_ASSERT(m_graph, header->at(0), header->predecessors.size() > 1, header->predecessors.size()); |
| 130 | |
| 131 | for (unsigned i = header->predecessors.size(); i--;) { |
| 132 | BasicBlock* predecessor = header->predecessors[i]; |
| 133 | if (m_graph.m_ssaDominators->dominates(header, predecessor)) |
| 134 | continue; |
| 135 | |
| 136 | preHeader = predecessor; |
| 137 | ++numberOfPreHeaders; |
| 138 | } |
| 139 | |
| 140 | // We need to validate the pre-header. There are a bunch of things that could be wrong |
| 141 | // about it: |
| 142 | // |
| 143 | // - There might be more than one. This means that pre-header creation either did not run, |
| 144 | // or some CFG transformation destroyed the pre-headers. |
| 145 | // |
| 146 | // - It may not be legal to exit at the pre-header. That would be a real bummer. Currently, |
| 147 | // LICM assumes that it can always hoist checks. See |
| 148 | // https://bugs.webkit.org/show_bug.cgi?id=148545. Though even with that fixed, we anyway |
| 149 | // would need to check if it's OK to exit at the pre-header since if we can't then we |
| 150 | // would have to restrict hoisting to non-exiting nodes. |
| 151 | |
| 152 | if (numberOfPreHeaders != 1) |
| 153 | continue; |
| 154 | |
| 155 | // This is guaranteed because the header has multiple predecessors and critical edges are |
| 156 | // broken. Therefore the predecessors must all have one successor, which implies that they |
| 157 | // must end in a Jump. |
| 158 | DFG_ASSERT(m_graph, preHeader->terminal(), preHeader->terminal()->op() == Jump, preHeader->terminal()->op()); |
| 159 | |
| 160 | if (!preHeader->terminal()->origin.exitOK) |
| 161 | continue; |
| 162 | |
| 163 | data.preHeader = preHeader; |
| 164 | } |
| 165 | |
| 166 | m_graph.initializeNodeOwners(); |
| 167 | |
| 168 | // Walk all basic blocks that belong to loops, looking for hoisting opportunities. |
| 169 | // We try to hoist to the outer-most loop that permits it. Hoisting is valid if: |
| 170 | // - The node doesn't write anything. |
| 171 | // - The node doesn't read anything that the loop writes. |
| 172 | // - The preHeader is valid (i.e. it passed the validation above). |
| 173 | // - The preHeader's state at tail makes the node safe to execute. |
| 174 | // - The loop's children all belong to nodes that strictly dominate the loop header. |
| 175 | // - The preHeader's state at tail is still valid. This is mostly to save compile |
| 176 | // time and preserve some kind of sanity, if we hoist something that must exit. |
| 177 | // |
| 178 | // Also, we need to remember to: |
| 179 | // - Update the state-at-tail with the node we hoisted, so future hoist candidates |
| 180 | // know about any type checks we hoisted. |
| 181 | // |
| 182 | // For maximum profit, we walk blocks in DFS order to ensure that we generally |
| 183 | // tend to hoist dominators before dominatees. |
| 184 | Vector<const NaturalLoop*> loopStack; |
| 185 | bool changed = false; |
| 186 | for (BasicBlock* block : m_graph.blocksInPreOrder()) { |
| 187 | if (!block->cfaHasVisited) |
| 188 | continue; |
| 189 | |
| 190 | const NaturalLoop* loop = m_graph.m_ssaNaturalLoops->innerMostLoopOf(block); |
| 191 | if (!loop) |
| 192 | continue; |
| 193 | |
| 194 | loopStack.shrink(0); |
| 195 | for ( |
| 196 | const NaturalLoop* current = loop; |
| 197 | current; |
| 198 | current = m_graph.m_ssaNaturalLoops->innerMostOuterLoop(*current)) |
| 199 | loopStack.append(current); |
| 200 | |
| 201 | // Remember: the loop stack has the inner-most loop at index 0, so if we want |
| 202 | // to bias hoisting to outer loops then we need to use a reverse loop. |
| 203 | |
| 204 | if (verbose) { |
| 205 | dataLog( |
| 206 | "Attempting to hoist out of block " , *block, " in loops:\n" ); |
| 207 | for (unsigned stackIndex = loopStack.size(); stackIndex--;) { |
| 208 | dataLog( |
| 209 | " " , *loopStack[stackIndex], ", which writes " , |
| 210 | m_data[loopStack[stackIndex]->index()].writes, "\n" ); |
| 211 | } |
| 212 | } |
| 213 | |
| 214 | for (unsigned nodeIndex = 0; nodeIndex < block->size(); ++nodeIndex) { |
| 215 | Node*& nodeRef = block->at(nodeIndex); |
| 216 | if (nodeRef->op() == ForceOSRExit) |
| 217 | break; |
| 218 | for (unsigned stackIndex = loopStack.size(); stackIndex--;) |
| 219 | changed |= attemptHoist(block, nodeRef, loopStack[stackIndex]); |
| 220 | } |
| 221 | } |
| 222 | |
| 223 | return changed; |
| 224 | } |
| 225 | |
| 226 | private: |
| 227 | bool attemptHoist(BasicBlock* fromBlock, Node*& nodeRef, const NaturalLoop* loop) |
| 228 | { |
| 229 | Node* node = nodeRef; |
| 230 | LoopData& data = m_data[loop->index()]; |
| 231 | |
| 232 | if (!data.preHeader) { |
| 233 | if (verbose) |
| 234 | dataLog(" Not hoisting " , node, " because the pre-header is invalid.\n" ); |
| 235 | return false; |
| 236 | } |
| 237 | |
| 238 | if (!data.preHeader->cfaDidFinish) { |
| 239 | if (verbose) |
| 240 | dataLog(" Not hoisting " , node, " because CFA is invalid.\n" ); |
| 241 | return false; |
| 242 | } |
| 243 | |
| 244 | m_state.initializeTo(data.preHeader); |
| 245 | ASSERT(m_state.isValid()); |
| 246 | NodeOrigin originalOrigin = node->origin; |
| 247 | bool canSpeculateBlindly = !m_graph.hasGlobalExitSite(originalOrigin.semantic, HoistingFailed); |
| 248 | |
| 249 | // NOTE: We could just use BackwardsDominators here directly, since we already know that the |
| 250 | // preHeader dominates fromBlock. But we wouldn't get anything from being so clever, since |
| 251 | // dominance checks are O(1) and only a few integer compares. |
| 252 | bool isControlEquivalent = m_graph.m_controlEquivalenceAnalysis->dominatesEquivalently(data.preHeader, fromBlock); |
| 253 | |
| 254 | bool addsBlindSpeculation = !isControlEquivalent; |
| 255 | NodeOrigin terminalOrigin = data.preHeader->terminal()->origin; |
| 256 | Vector<Node*, 2> hoistedNodes; // This is sorted in the program order they will appear in the basic block we're hoisting to. |
| 257 | |
| 258 | auto insertHoistedNode = [&] (Node* node) { |
| 259 | data.preHeader->insertBeforeTerminal(node); |
| 260 | node->owner = data.preHeader; |
| 261 | node->origin = terminalOrigin.withSemantic(node->origin.semantic); |
| 262 | node->origin.wasHoisted |= addsBlindSpeculation; |
| 263 | hoistedNodes.append(node); |
| 264 | }; |
| 265 | |
| 266 | auto updateAbstractState = [&] { |
| 267 | auto invalidate = [&] (const NaturalLoop* loop) { |
| 268 | LoopData& data = m_data[loop->index()]; |
| 269 | data.preHeader->cfaDidFinish = false; |
| 270 | |
| 271 | for (unsigned bodyIndex = loop->size(); bodyIndex--;) { |
| 272 | BasicBlock* block = loop->at(bodyIndex); |
| 273 | if (block != data.preHeader) |
| 274 | block->cfaHasVisited = false; |
| 275 | block->cfaDidFinish = false; |
| 276 | } |
| 277 | }; |
| 278 | |
| 279 | // We can trust what AI proves about edge proof statuses when hoisting to the preheader. |
| 280 | m_state.trustEdgeProofs(); |
| 281 | for (unsigned i = 0; i < hoistedNodes.size(); ++i) { |
| 282 | if (!m_interpreter.execute(hoistedNodes[i])) { |
| 283 | invalidate(loop); |
| 284 | return; |
| 285 | } |
| 286 | } |
| 287 | |
| 288 | // However, when walking various inner loops below, the proof status of |
| 289 | // an edge may be trivially true, even if it's not true in the preheader |
| 290 | // we hoist to. We don't allow the below node executions to change the |
| 291 | // state of edge proofs. An example of where a proof is trivially true |
| 292 | // is if we have two loops, L1 and L2, where L2 is nested inside L1. The |
| 293 | // header for L1 dominates L2. We hoist a Check from L1's header into L1's |
| 294 | // preheader. However, inside L2's preheader, we can't trust that AI will |
| 295 | // tell us this edge is proven. It's proven in L2's preheader because L2 |
| 296 | // is dominated by L1's header. However, the edge is not guaranteed to be |
| 297 | // proven inside L1's preheader. |
| 298 | m_state.dontTrustEdgeProofs(); |
| 299 | |
| 300 | // Modify the states at the end of the preHeader of the loop we hoisted to, |
| 301 | // and all pre-headers inside the loop. This isn't a stability bottleneck right now |
| 302 | // because most loops are small and most blocks belong to few loops. |
| 303 | for (unsigned bodyIndex = loop->size(); bodyIndex--;) { |
| 304 | BasicBlock* subBlock = loop->at(bodyIndex); |
| 305 | const NaturalLoop* subLoop = m_graph.m_ssaNaturalLoops->headerOf(subBlock); |
| 306 | if (!subLoop) |
| 307 | continue; |
| 308 | BasicBlock* = m_data[subLoop->index()].preHeader; |
| 309 | // We may not have given this loop a pre-header because either it didn't have exitOK |
| 310 | // or the header had multiple predecessors that it did not dominate. In that case the |
| 311 | // loop wouldn't be a hoisting candidate anyway, so we don't have to do anything. |
| 312 | if (!subPreHeader) |
| 313 | continue; |
| 314 | // The pre-header's tail may be unreachable, in which case we have nothing to do. |
| 315 | if (!subPreHeader->cfaDidFinish) |
| 316 | continue; |
| 317 | // We handled this above. |
| 318 | if (subPreHeader == data.preHeader) |
| 319 | continue; |
| 320 | m_state.initializeTo(subPreHeader); |
| 321 | for (unsigned i = 0; i < hoistedNodes.size(); ++i) { |
| 322 | if (!m_interpreter.execute(hoistedNodes[i])) { |
| 323 | invalidate(subLoop); |
| 324 | break; |
| 325 | } |
| 326 | } |
| 327 | } |
| 328 | }; |
| 329 | |
| 330 | auto tryHoistChecks = [&] { |
| 331 | if (addsBlindSpeculation && !canSpeculateBlindly) |
| 332 | return false; |
| 333 | |
| 334 | ASSERT(hoistedNodes.isEmpty()); |
| 335 | |
| 336 | Vector<Edge, 3> checks; |
| 337 | m_graph.doToChildren(node, [&] (Edge edge) { |
| 338 | if (!m_graph.m_ssaDominators->dominates(edge.node()->owner, data.preHeader)) |
| 339 | return; |
| 340 | |
| 341 | if (!edge.willHaveCheck()) |
| 342 | return; |
| 343 | |
| 344 | if ((m_state.forNode(edge).m_type & SpecEmpty) && checkMayCrashIfInputIsEmpty(edge.useKind())) { |
| 345 | if (!canSpeculateBlindly) |
| 346 | return; |
| 347 | Node* checkNotEmpty = m_graph.addNode(CheckNotEmpty, originalOrigin, Edge(edge.node(), UntypedUse)); |
| 348 | insertHoistedNode(checkNotEmpty); |
| 349 | } |
| 350 | |
| 351 | checks.append(edge); |
| 352 | }); |
| 353 | |
| 354 | if (checks.isEmpty()) |
| 355 | return false; |
| 356 | |
| 357 | AdjacencyList children; |
| 358 | NodeType checkOp = Check; |
| 359 | if (checks.size() <= AdjacencyList::Size) { |
| 360 | children = AdjacencyList(AdjacencyList::Fixed); |
| 361 | for (unsigned i = 0; i < checks.size(); ++i) |
| 362 | children.setChild(i, checks[i]); |
| 363 | } else { |
| 364 | checkOp = CheckVarargs; |
| 365 | unsigned firstChild = m_graph.m_varArgChildren.size(); |
| 366 | for (Edge edge : checks) |
| 367 | m_graph.m_varArgChildren.append(edge); |
| 368 | children = AdjacencyList(AdjacencyList::Variable, firstChild, checks.size()); |
| 369 | } |
| 370 | |
| 371 | Node* check = m_graph.addNode(checkOp, originalOrigin, children); |
| 372 | insertHoistedNode(check); |
| 373 | updateAbstractState(); |
| 374 | |
| 375 | if (verbose) |
| 376 | dataLogLn(" Hoisted some checks from " , node, " and created a new Check " , check, ". Hoisted from " , *fromBlock, " to " , *data.preHeader); |
| 377 | |
| 378 | return true; |
| 379 | }; |
| 380 | |
| 381 | if (!edgesDominate(m_graph, node, data.preHeader)) { |
| 382 | if (verbose) { |
| 383 | dataLog( |
| 384 | " Not hoisting " , node, " because it isn't loop invariant.\n" ); |
| 385 | } |
| 386 | return tryHoistChecks(); |
| 387 | } |
| 388 | |
| 389 | if (doesWrites(m_graph, node)) { |
| 390 | if (verbose) |
| 391 | dataLog(" Not hoisting " , node, " because it writes things.\n" ); |
| 392 | return tryHoistChecks(); |
| 393 | } |
| 394 | |
| 395 | // It's not safe to consult the AbstractState inside mayExit until we prove all edges |
| 396 | // dominate the pre-header we're hoisting to. We are more conservative above when assigning |
| 397 | // to this variable since we hadn't yet proven all edges dominate the pre-header. Above, we |
| 398 | // just assume mayExit is true. We refine that here since we can now consult the AbstractState. |
| 399 | addsBlindSpeculation = mayExit(m_graph, node, m_state) && !isControlEquivalent; |
| 400 | |
| 401 | if (readsOverlap(m_graph, node, data.writes)) { |
| 402 | if (verbose) { |
| 403 | dataLog( |
| 404 | " Not hoisting " , node, |
| 405 | " because it reads things that the loop writes.\n" ); |
| 406 | } |
| 407 | return tryHoistChecks(); |
| 408 | } |
| 409 | |
| 410 | if (addsBlindSpeculation && !canSpeculateBlindly) { |
| 411 | if (verbose) { |
| 412 | dataLog( |
| 413 | " Not hoisting " , node, " because it may exit and the pre-header (" , |
| 414 | *data.preHeader, ") is not control equivalent to the node's original block (" , |
| 415 | *fromBlock, ") and hoisting had previously failed.\n" ); |
| 416 | } |
| 417 | return tryHoistChecks(); |
| 418 | } |
| 419 | |
| 420 | if (!safeToExecute(m_state, m_graph, node)) { |
| 421 | // See if we can rescue the situation by inserting blind speculations. |
| 422 | bool ignoreEmptyChildren = true; |
| 423 | if (canSpeculateBlindly |
| 424 | && safeToExecute(m_state, m_graph, node, ignoreEmptyChildren)) { |
| 425 | if (verbose) { |
| 426 | dataLog( |
| 427 | " Rescuing hoisting by inserting empty checks.\n" ); |
| 428 | } |
| 429 | m_graph.doToChildren( |
| 430 | node, |
| 431 | [&] (Edge& edge) { |
| 432 | if (!(m_state.forNode(edge).m_type & SpecEmpty)) |
| 433 | return; |
| 434 | |
| 435 | Node* check = m_graph.addNode(CheckNotEmpty, originalOrigin, Edge(edge.node(), UntypedUse)); |
| 436 | insertHoistedNode(check); |
| 437 | }); |
| 438 | } else { |
| 439 | if (verbose) { |
| 440 | dataLog( |
| 441 | " Not hoisting " , node, " because it isn't safe to execute.\n" ); |
| 442 | } |
| 443 | return tryHoistChecks(); |
| 444 | } |
| 445 | } |
| 446 | |
| 447 | if (verbose) { |
| 448 | dataLog( |
| 449 | " Hoisting " , node, " from " , *fromBlock, " to " , *data.preHeader, |
| 450 | "\n" ); |
| 451 | } |
| 452 | |
| 453 | insertHoistedNode(node); |
| 454 | updateAbstractState(); |
| 455 | |
| 456 | if (node->flags() & NodeHasVarArgs) |
| 457 | nodeRef = m_graph.addNode(CheckVarargs, originalOrigin, m_graph.copyVarargChildren(node)); |
| 458 | else |
| 459 | nodeRef = m_graph.addNode(Check, originalOrigin, node->children); |
| 460 | |
| 461 | return true; |
| 462 | } |
| 463 | |
| 464 | AtTailAbstractState m_state; |
| 465 | AbstractInterpreter<AtTailAbstractState> m_interpreter; |
| 466 | Vector<LoopData> m_data; |
| 467 | }; |
| 468 | |
| 469 | bool performLICM(Graph& graph) |
| 470 | { |
| 471 | return runPhase<LICMPhase>(graph); |
| 472 | } |
| 473 | |
| 474 | } } // namespace JSC::DFG |
| 475 | |
| 476 | #endif // ENABLE(DFG_JIT) |
| 477 | |
| 478 | |