| 1 | /* |
| 2 | * Copyright (C) 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. ``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 "B3InferSwitches.h" |
| 28 | |
| 29 | #if ENABLE(B3_JIT) |
| 30 | |
| 31 | #include "B3BasicBlockInlines.h" |
| 32 | #include "B3CaseCollectionInlines.h" |
| 33 | #include "B3InsertionSetInlines.h" |
| 34 | #include "B3PhaseScope.h" |
| 35 | #include "B3ProcedureInlines.h" |
| 36 | #include "B3SwitchValue.h" |
| 37 | #include "B3UseCounts.h" |
| 38 | #include "B3ValueInlines.h" |
| 39 | #include <wtf/ListDump.h> |
| 40 | |
| 41 | namespace JSC { namespace B3 { |
| 42 | |
| 43 | namespace { |
| 44 | |
| 45 | namespace B3InferSwitchesInternal { |
| 46 | static const bool verbose = false; |
| 47 | } |
| 48 | |
| 49 | class InferSwitches { |
| 50 | public: |
| 51 | InferSwitches(Procedure& proc) |
| 52 | : m_proc(proc) |
| 53 | , m_insertionSet(proc) |
| 54 | , m_useCounts(proc) |
| 55 | { |
| 56 | } |
| 57 | |
| 58 | bool run() |
| 59 | { |
| 60 | if (B3InferSwitchesInternal::verbose) |
| 61 | dataLog("B3 before inferSwitches:\n" , m_proc); |
| 62 | |
| 63 | bool changed = true; |
| 64 | bool everChanged = false; |
| 65 | while (changed) { |
| 66 | changed = false; |
| 67 | |
| 68 | if (B3InferSwitchesInternal::verbose) |
| 69 | dataLog("Performing fixpoint iteration:\n" ); |
| 70 | |
| 71 | for (BasicBlock* block : m_proc) |
| 72 | changed |= attemptToMergeWithPredecessor(block); |
| 73 | |
| 74 | everChanged |= changed; |
| 75 | } |
| 76 | |
| 77 | if (everChanged) { |
| 78 | m_proc.resetReachability(); |
| 79 | m_proc.invalidateCFG(); |
| 80 | |
| 81 | m_proc.deleteOrphans(); |
| 82 | |
| 83 | if (B3InferSwitchesInternal::verbose) |
| 84 | dataLog("B3 after inferSwitches:\n" , m_proc); |
| 85 | return true; |
| 86 | } |
| 87 | |
| 88 | return false; |
| 89 | } |
| 90 | |
| 91 | private: |
| 92 | bool attemptToMergeWithPredecessor(BasicBlock* block) |
| 93 | { |
| 94 | // No point in considering the root block. We also don't consider blocks with multiple |
| 95 | // predecessors, but we could handle this if we made this code a bit more general and we were |
| 96 | // not afraid of code bloat. |
| 97 | if (block->numPredecessors() != 1) |
| 98 | return false; |
| 99 | |
| 100 | SwitchDescription description = describe(block); |
| 101 | if (B3InferSwitchesInternal::verbose) |
| 102 | dataLog("Description of primary block " , *block, ": " , description, "\n" ); |
| 103 | if (!description) { |
| 104 | if (B3InferSwitchesInternal::verbose) |
| 105 | dataLog(" Bailing because not switch-like.\n" ); |
| 106 | return false; |
| 107 | } |
| 108 | |
| 109 | // We know that this block behaves like a switch. But we need to verify that it doesn't also |
| 110 | // perform any effects or do expensive things. We don't want to create a switch if that will |
| 111 | // make expensive things execute unconditionally. We're very conservative about how we define |
| 112 | // "expensive". |
| 113 | for (Value* value : *block) { |
| 114 | if (value->isFree()) |
| 115 | continue; |
| 116 | if (value == description.extra) |
| 117 | continue; |
| 118 | if (value == description.branch) |
| 119 | continue; |
| 120 | if (B3InferSwitchesInternal::verbose) |
| 121 | dataLog(" Bailing because of " , deepDump(m_proc, value), "\n" ); |
| 122 | return false; |
| 123 | } |
| 124 | |
| 125 | BasicBlock* predecessor = block->predecessor(0); |
| 126 | SwitchDescription predecessorDescription = describe(predecessor); |
| 127 | if (B3InferSwitchesInternal::verbose) |
| 128 | dataLog(" Description of predecessor block " , *predecessor, ": " , predecessorDescription, "\n" ); |
| 129 | if (!predecessorDescription) { |
| 130 | if (B3InferSwitchesInternal::verbose) |
| 131 | dataLog(" Bailing because not switch-like.\n" ); |
| 132 | return false; |
| 133 | } |
| 134 | |
| 135 | // Both us and the predecessor are switch-like, but that doesn't mean that we're compatible. |
| 136 | // We may be switching on different values! |
| 137 | if (description.source != predecessorDescription.source) { |
| 138 | if (B3InferSwitchesInternal::verbose) |
| 139 | dataLog(" Bailing because sources don't match.\n" ); |
| 140 | return false; |
| 141 | } |
| 142 | |
| 143 | // We expect that we are the fall-through destination of the predecessor. This is a bit of a |
| 144 | // goofy condition. If we were not the fall-through destination then our switch is probably |
| 145 | // just totally redundant and we should be getting rid of it. But we don't handle that here, |
| 146 | // yet. |
| 147 | if (predecessorDescription.fallThrough.block() != block) { |
| 148 | if (B3InferSwitchesInternal::verbose) |
| 149 | dataLog(" Bailing because fall-through of predecessor is not the primary block.\n" ); |
| 150 | return false; |
| 151 | } |
| 152 | |
| 153 | // Make sure that there ain't no loops. |
| 154 | if (description.fallThrough.block() == block |
| 155 | || description.fallThrough.block() == predecessor) { |
| 156 | if (B3InferSwitchesInternal::verbose) |
| 157 | dataLog(" Bailing because of fall-through loop.\n" ); |
| 158 | return false; |
| 159 | } |
| 160 | for (SwitchCase switchCase : description.cases) { |
| 161 | if (switchCase.targetBlock() == block |
| 162 | || switchCase.targetBlock() == predecessor) { |
| 163 | if (B3InferSwitchesInternal::verbose) |
| 164 | dataLog(" Bailing because of loop in primary cases.\n" ); |
| 165 | return false; |
| 166 | } |
| 167 | } |
| 168 | for (SwitchCase switchCase : predecessorDescription.cases) { |
| 169 | if (switchCase.targetBlock() == block |
| 170 | || switchCase.targetBlock() == predecessor) { |
| 171 | if (B3InferSwitchesInternal::verbose) |
| 172 | dataLog(" Bailing because of loop in predecessor cases.\n" ); |
| 173 | return false; |
| 174 | } |
| 175 | } |
| 176 | |
| 177 | if (B3InferSwitchesInternal::verbose) |
| 178 | dataLog(" Doing it!\n" ); |
| 179 | // We're committed to doing the thing. |
| 180 | |
| 181 | // Delete the extra value from the predecessor, since that would break downstream inference |
| 182 | // on the next fixpoint iteration. We would think that this block is too expensive to merge |
| 183 | // because of the Equal or NotEqual value even though that value is dead! We know it's dead |
| 184 | // so we kill it ourselves. |
| 185 | for (Value* value : *predecessor) { |
| 186 | if (value == predecessorDescription.extra) |
| 187 | value->replaceWithNopIgnoringType(); |
| 188 | } |
| 189 | |
| 190 | // Insert all non-terminal values from our block into our predecessor. We definitely need to |
| 191 | // do this for constants. We must not do it for the extra value, since that would break |
| 192 | // downstream inference on the next fixpoint iteration. As a bonus, we don't do it for nops, |
| 193 | // so that we limit how big blocks get in this phase. |
| 194 | for (unsigned i = 0; i < block->size() - 1; ++i) { |
| 195 | Value* value = block->at(i); |
| 196 | if (value != description.extra && value->opcode() != Nop) |
| 197 | m_insertionSet.insertValue(predecessor->size() - 1, value); |
| 198 | } |
| 199 | m_insertionSet.execute(predecessor); |
| 200 | block->values().shrink(0); |
| 201 | block->appendNew<Value>(m_proc, Oops, description.branch->origin()); |
| 202 | block->removePredecessor(predecessor); |
| 203 | |
| 204 | for (BasicBlock* successorBlock : description.block->successorBlocks()) |
| 205 | successorBlock->replacePredecessor(block, predecessor); |
| 206 | |
| 207 | block->clearSuccessors(); |
| 208 | |
| 209 | SwitchValue* switchValue = predecessor->replaceLastWithNew<SwitchValue>( |
| 210 | m_proc, predecessor->last()->origin(), description.source); |
| 211 | predecessor->clearSuccessors(); |
| 212 | switchValue->setFallThrough(description.fallThrough); |
| 213 | |
| 214 | Vector<int64_t> predecessorCases; |
| 215 | for (SwitchCase switchCase : predecessorDescription.cases) { |
| 216 | switchValue->appendCase(switchCase); |
| 217 | predecessorCases.append(switchCase.caseValue()); |
| 218 | } |
| 219 | std::sort(predecessorCases.begin(), predecessorCases.end()); |
| 220 | auto isPredecessorCase = [&] (int64_t value) -> bool { |
| 221 | return !!tryBinarySearch<int64_t>( |
| 222 | predecessorCases, predecessorCases.size(), value, |
| 223 | [] (int64_t* element) -> int64_t { return *element; }); |
| 224 | }; |
| 225 | |
| 226 | for (SwitchCase switchCase : description.cases) { |
| 227 | if (!isPredecessorCase(switchCase.caseValue())) |
| 228 | switchValue->appendCase(switchCase); |
| 229 | } |
| 230 | return true; |
| 231 | } |
| 232 | |
| 233 | struct SwitchDescription { |
| 234 | SwitchDescription() |
| 235 | { |
| 236 | } |
| 237 | |
| 238 | explicit operator bool() { return !!block; } |
| 239 | |
| 240 | void dump(PrintStream& out) const |
| 241 | { |
| 242 | out.print( |
| 243 | "{block = " , pointerDump(block), |
| 244 | ", branch = " , pointerDump(branch), |
| 245 | ", extra = " , pointerDump(extra), |
| 246 | ", source = " , pointerDump(source), |
| 247 | ", cases = " , listDump(cases), |
| 248 | ", fallThrough = " , fallThrough, "}" ); |
| 249 | } |
| 250 | |
| 251 | BasicBlock* block { nullptr }; |
| 252 | Value* branch { nullptr }; |
| 253 | Value* { nullptr }; // This is the Equal or NotEqual value, if applicable. |
| 254 | Value* source { nullptr }; |
| 255 | Vector<SwitchCase, 1> cases; |
| 256 | FrequentedBlock fallThrough; |
| 257 | }; |
| 258 | |
| 259 | SwitchDescription describe(BasicBlock* block) |
| 260 | { |
| 261 | SwitchDescription result; |
| 262 | result.block = block; |
| 263 | result.branch = block->last(); |
| 264 | |
| 265 | switch (result.branch->opcode()) { |
| 266 | case Branch: { |
| 267 | Value* predicate = result.branch->child(0); |
| 268 | FrequentedBlock taken = result.block->taken(); |
| 269 | FrequentedBlock notTaken = result.block->notTaken(); |
| 270 | bool handled = false; |
| 271 | // NOTE: This uses UseCounts that we computed before any transformation. This is fine |
| 272 | // because although we may have mutated the IR, we would not have added any new |
| 273 | // predicates. |
| 274 | if (predicate->numChildren() == 2 |
| 275 | && predicate->child(1)->hasInt() |
| 276 | && m_useCounts.numUses(predicate) == 1) { |
| 277 | switch (predicate->opcode()) { |
| 278 | case Equal: |
| 279 | result.source = predicate->child(0); |
| 280 | result.extra = predicate; |
| 281 | result.cases.append(SwitchCase(predicate->child(1)->asInt(), taken)); |
| 282 | result.fallThrough = notTaken; |
| 283 | handled = true; |
| 284 | break; |
| 285 | case NotEqual: |
| 286 | result.source = predicate->child(0); |
| 287 | result.extra = predicate; |
| 288 | result.cases.append(SwitchCase(predicate->child(1)->asInt(), notTaken)); |
| 289 | result.fallThrough = taken; |
| 290 | handled = true; |
| 291 | break; |
| 292 | default: |
| 293 | break; |
| 294 | } |
| 295 | } |
| 296 | if (handled) |
| 297 | break; |
| 298 | result.source = predicate; |
| 299 | result.cases.append(SwitchCase(0, notTaken)); |
| 300 | result.fallThrough = taken; |
| 301 | break; |
| 302 | } |
| 303 | |
| 304 | case Switch: { |
| 305 | SwitchValue* switchValue = result.branch->as<SwitchValue>(); |
| 306 | result.source = switchValue->child(0); |
| 307 | for (SwitchCase switchCase : switchValue->cases(result.block)) |
| 308 | result.cases.append(switchCase); |
| 309 | result.fallThrough = result.block->fallThrough(); |
| 310 | break; |
| 311 | } |
| 312 | |
| 313 | default: |
| 314 | result.block = nullptr; |
| 315 | result.branch = nullptr; |
| 316 | break; |
| 317 | } |
| 318 | |
| 319 | return result; |
| 320 | } |
| 321 | |
| 322 | Procedure& m_proc; |
| 323 | InsertionSet m_insertionSet; |
| 324 | UseCounts m_useCounts; |
| 325 | }; |
| 326 | |
| 327 | } // anonymous namespace |
| 328 | |
| 329 | bool inferSwitches(Procedure& proc) |
| 330 | { |
| 331 | PhaseScope phaseScope(proc, "inferSwitches" ); |
| 332 | InferSwitches inferSwitches(proc); |
| 333 | return inferSwitches.run(); |
| 334 | } |
| 335 | |
| 336 | } } // namespace JSC::B3 |
| 337 | |
| 338 | #endif // ENABLE(B3_JIT) |
| 339 | |
| 340 | |