1 | /* |
2 | * Copyright (C) 2016-2019 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 "WasmB3IRGenerator.h" |
28 | |
29 | #if ENABLE(WEBASSEMBLY) |
30 | |
31 | #include "AllowMacroScratchRegisterUsageIf.h" |
32 | #include "B3BasicBlockInlines.h" |
33 | #include "B3CCallValue.h" |
34 | #include "B3Compile.h" |
35 | #include "B3ConstPtrValue.h" |
36 | #include "B3FixSSA.h" |
37 | #include "B3Generate.h" |
38 | #include "B3InsertionSet.h" |
39 | #include "B3SlotBaseValue.h" |
40 | #include "B3StackmapGenerationParams.h" |
41 | #include "B3SwitchValue.h" |
42 | #include "B3UpsilonValue.h" |
43 | #include "B3Validate.h" |
44 | #include "B3ValueInlines.h" |
45 | #include "B3ValueKey.h" |
46 | #include "B3Variable.h" |
47 | #include "B3VariableValue.h" |
48 | #include "B3WasmAddressValue.h" |
49 | #include "B3WasmBoundsCheckValue.h" |
50 | #include "DisallowMacroScratchRegisterUsage.h" |
51 | #include "JSCInlines.h" |
52 | #include "ScratchRegisterAllocator.h" |
53 | #include "VirtualRegister.h" |
54 | #include "WasmCallingConvention.h" |
55 | #include "WasmContextInlines.h" |
56 | #include "WasmExceptionType.h" |
57 | #include "WasmFunctionParser.h" |
58 | #include "WasmInstance.h" |
59 | #include "WasmMemory.h" |
60 | #include "WasmOMGPlan.h" |
61 | #include "WasmOpcodeOrigin.h" |
62 | #include "WasmSignatureInlines.h" |
63 | #include "WasmThunks.h" |
64 | #include <limits> |
65 | #include <wtf/Optional.h> |
66 | #include <wtf/StdLibExtras.h> |
67 | |
68 | void dumpProcedure(void* ptr) |
69 | { |
70 | JSC::B3::Procedure* proc = static_cast<JSC::B3::Procedure*>(ptr); |
71 | proc->dump(WTF::dataFile()); |
72 | } |
73 | |
74 | namespace JSC { namespace Wasm { |
75 | |
76 | using namespace B3; |
77 | |
78 | namespace { |
79 | namespace WasmB3IRGeneratorInternal { |
80 | static const bool verbose = false; |
81 | } |
82 | } |
83 | |
84 | class B3IRGenerator { |
85 | public: |
86 | struct ControlData { |
87 | ControlData(Procedure& proc, Origin origin, Type signature, BlockType type, BasicBlock* continuation, BasicBlock* special = nullptr) |
88 | : blockType(type) |
89 | , continuation(continuation) |
90 | , special(special) |
91 | { |
92 | if (signature != Void) |
93 | result.append(proc.add<Value>(Phi, toB3Type(signature), origin)); |
94 | } |
95 | |
96 | ControlData() |
97 | { |
98 | } |
99 | |
100 | void dump(PrintStream& out) const |
101 | { |
102 | switch (type()) { |
103 | case BlockType::If: |
104 | out.print("If: " ); |
105 | break; |
106 | case BlockType::Block: |
107 | out.print("Block: " ); |
108 | break; |
109 | case BlockType::Loop: |
110 | out.print("Loop: " ); |
111 | break; |
112 | case BlockType::TopLevel: |
113 | out.print("TopLevel: " ); |
114 | break; |
115 | } |
116 | out.print("Continuation: " , *continuation, ", Special: " ); |
117 | if (special) |
118 | out.print(*special); |
119 | else |
120 | out.print("None" ); |
121 | } |
122 | |
123 | BlockType type() const { return blockType; } |
124 | |
125 | bool hasNonVoidSignature() const { return result.size(); } |
126 | |
127 | BasicBlock* targetBlockForBranch() |
128 | { |
129 | if (type() == BlockType::Loop) |
130 | return special; |
131 | return continuation; |
132 | } |
133 | |
134 | void convertIfToBlock() |
135 | { |
136 | ASSERT(type() == BlockType::If); |
137 | blockType = BlockType::Block; |
138 | special = nullptr; |
139 | } |
140 | |
141 | using ResultList = Vector<Value*, 1>; // Value must be a Phi |
142 | |
143 | ResultList resultForBranch() const |
144 | { |
145 | if (type() == BlockType::Loop) |
146 | return ResultList(); |
147 | return result; |
148 | } |
149 | |
150 | private: |
151 | friend class B3IRGenerator; |
152 | BlockType blockType; |
153 | BasicBlock* continuation; |
154 | BasicBlock* special; |
155 | ResultList result; |
156 | }; |
157 | |
158 | typedef Value* ExpressionType; |
159 | typedef ControlData ControlType; |
160 | typedef Vector<ExpressionType, 1> ExpressionList; |
161 | typedef ControlData::ResultList ResultList; |
162 | typedef FunctionParser<B3IRGenerator>::ControlEntry ControlEntry; |
163 | |
164 | static constexpr ExpressionType emptyExpression() { return nullptr; } |
165 | |
166 | typedef String ErrorType; |
167 | typedef Unexpected<ErrorType> UnexpectedResult; |
168 | typedef Expected<std::unique_ptr<InternalFunction>, ErrorType> Result; |
169 | typedef Expected<void, ErrorType> PartialResult; |
170 | template <typename ...Args> |
171 | NEVER_INLINE UnexpectedResult WARN_UNUSED_RETURN fail(Args... args) const |
172 | { |
173 | using namespace FailureHelper; // See ADL comment in WasmParser.h. |
174 | return UnexpectedResult(makeString("WebAssembly.Module failed compiling: "_s , makeString(args)...)); |
175 | } |
176 | #define WASM_COMPILE_FAIL_IF(condition, ...) do { \ |
177 | if (UNLIKELY(condition)) \ |
178 | return fail(__VA_ARGS__); \ |
179 | } while (0) |
180 | |
181 | B3IRGenerator(const ModuleInformation&, Procedure&, InternalFunction*, Vector<UnlinkedWasmToWasmCall>&, MemoryMode, CompilationMode, unsigned functionIndex, TierUpCount*, ThrowWasmException); |
182 | |
183 | PartialResult WARN_UNUSED_RETURN addArguments(const Signature&); |
184 | PartialResult WARN_UNUSED_RETURN addLocal(Type, uint32_t); |
185 | ExpressionType addConstant(Type, uint64_t); |
186 | |
187 | PartialResult WARN_UNUSED_RETURN addRefIsNull(ExpressionType& value, ExpressionType& result); |
188 | |
189 | // Locals |
190 | PartialResult WARN_UNUSED_RETURN getLocal(uint32_t index, ExpressionType& result); |
191 | PartialResult WARN_UNUSED_RETURN setLocal(uint32_t index, ExpressionType value); |
192 | |
193 | // Globals |
194 | PartialResult WARN_UNUSED_RETURN getGlobal(uint32_t index, ExpressionType& result); |
195 | PartialResult WARN_UNUSED_RETURN setGlobal(uint32_t index, ExpressionType value); |
196 | |
197 | // Memory |
198 | PartialResult WARN_UNUSED_RETURN load(LoadOpType, ExpressionType pointer, ExpressionType& result, uint32_t offset); |
199 | PartialResult WARN_UNUSED_RETURN store(StoreOpType, ExpressionType pointer, ExpressionType value, uint32_t offset); |
200 | PartialResult WARN_UNUSED_RETURN addGrowMemory(ExpressionType delta, ExpressionType& result); |
201 | PartialResult WARN_UNUSED_RETURN addCurrentMemory(ExpressionType& result); |
202 | |
203 | // Basic operators |
204 | template<OpType> |
205 | PartialResult WARN_UNUSED_RETURN addOp(ExpressionType arg, ExpressionType& result); |
206 | template<OpType> |
207 | PartialResult WARN_UNUSED_RETURN addOp(ExpressionType left, ExpressionType right, ExpressionType& result); |
208 | PartialResult WARN_UNUSED_RETURN addSelect(ExpressionType condition, ExpressionType nonZero, ExpressionType zero, ExpressionType& result); |
209 | |
210 | // Control flow |
211 | ControlData WARN_UNUSED_RETURN addTopLevel(Type signature); |
212 | ControlData WARN_UNUSED_RETURN addBlock(Type signature); |
213 | ControlData WARN_UNUSED_RETURN addLoop(Type signature); |
214 | PartialResult WARN_UNUSED_RETURN addIf(ExpressionType condition, Type signature, ControlData& result); |
215 | PartialResult WARN_UNUSED_RETURN addElse(ControlData&, const ExpressionList&); |
216 | PartialResult WARN_UNUSED_RETURN addElseToUnreachable(ControlData&); |
217 | |
218 | PartialResult WARN_UNUSED_RETURN addReturn(const ControlData&, const ExpressionList& returnValues); |
219 | PartialResult WARN_UNUSED_RETURN addBranch(ControlData&, ExpressionType condition, const ExpressionList& returnValues); |
220 | PartialResult WARN_UNUSED_RETURN addSwitch(ExpressionType condition, const Vector<ControlData*>& targets, ControlData& defaultTargets, const ExpressionList& expressionStack); |
221 | PartialResult WARN_UNUSED_RETURN endBlock(ControlEntry&, ExpressionList& expressionStack); |
222 | PartialResult WARN_UNUSED_RETURN addEndToUnreachable(ControlEntry&); |
223 | |
224 | // Calls |
225 | PartialResult WARN_UNUSED_RETURN addCall(uint32_t calleeIndex, const Signature&, Vector<ExpressionType>& args, ExpressionType& result); |
226 | PartialResult WARN_UNUSED_RETURN addCallIndirect(const Signature&, Vector<ExpressionType>& args, ExpressionType& result); |
227 | PartialResult WARN_UNUSED_RETURN addUnreachable(); |
228 | |
229 | void dump(const Vector<ControlEntry>& controlStack, const ExpressionList* expressionStack); |
230 | void setParser(FunctionParser<B3IRGenerator>* parser) { m_parser = parser; }; |
231 | |
232 | Value* constant(B3::Type, uint64_t bits, Optional<Origin> = WTF::nullopt); |
233 | void insertConstants(); |
234 | |
235 | ALWAYS_INLINE void didKill(ExpressionType) { } |
236 | |
237 | private: |
238 | void emitExceptionCheck(CCallHelpers&, ExceptionType); |
239 | |
240 | void emitTierUpCheck(uint32_t decrementCount, Origin); |
241 | |
242 | ExpressionType emitCheckAndPreparePointer(ExpressionType pointer, uint32_t offset, uint32_t sizeOfOp); |
243 | B3::Kind memoryKind(B3::Opcode memoryOp); |
244 | ExpressionType emitLoadOp(LoadOpType, ExpressionType pointer, uint32_t offset); |
245 | void emitStoreOp(StoreOpType, ExpressionType pointer, ExpressionType value, uint32_t offset); |
246 | |
247 | void unify(const ExpressionType phi, const ExpressionType source); |
248 | void unifyValuesWithBlock(const ExpressionList& resultStack, const ResultList& stack); |
249 | |
250 | void emitChecksForModOrDiv(B3::Opcode, ExpressionType left, ExpressionType right); |
251 | |
252 | int32_t WARN_UNUSED_RETURN fixupPointerPlusOffset(ExpressionType&, uint32_t); |
253 | |
254 | void restoreWasmContextInstance(Procedure&, BasicBlock*, Value*); |
255 | enum class RestoreCachedStackLimit { No, Yes }; |
256 | void restoreWebAssemblyGlobalState(RestoreCachedStackLimit, const MemoryInformation&, Value* instance, Procedure&, BasicBlock*); |
257 | |
258 | Origin origin(); |
259 | |
260 | FunctionParser<B3IRGenerator>* m_parser { nullptr }; |
261 | const ModuleInformation& m_info; |
262 | const MemoryMode m_mode { MemoryMode::BoundsChecking }; |
263 | const CompilationMode m_compilationMode { CompilationMode::BBQMode }; |
264 | const unsigned m_functionIndex { UINT_MAX }; |
265 | const TierUpCount* m_tierUp { nullptr }; |
266 | |
267 | Procedure& m_proc; |
268 | BasicBlock* m_currentBlock { nullptr }; |
269 | Vector<Variable*> m_locals; |
270 | Vector<UnlinkedWasmToWasmCall>& m_unlinkedWasmToWasmCalls; // List each call site and the function index whose address it should be patched with. |
271 | HashMap<ValueKey, Value*> m_constantPool; |
272 | InsertionSet m_constantInsertionValues; |
273 | GPRReg m_memoryBaseGPR { InvalidGPRReg }; |
274 | GPRReg m_memorySizeGPR { InvalidGPRReg }; |
275 | GPRReg m_wasmContextInstanceGPR { InvalidGPRReg }; |
276 | bool m_makesCalls { false }; |
277 | |
278 | Value* m_instanceValue { nullptr }; // Always use the accessor below to ensure the instance value is materialized when used. |
279 | bool m_usesInstanceValue { false }; |
280 | Value* instanceValue() |
281 | { |
282 | m_usesInstanceValue = true; |
283 | return m_instanceValue; |
284 | } |
285 | |
286 | uint32_t m_maxNumJSCallArguments { 0 }; |
287 | }; |
288 | |
289 | // Memory accesses in WebAssembly have unsigned 32-bit offsets, whereas they have signed 32-bit offsets in B3. |
290 | int32_t B3IRGenerator::fixupPointerPlusOffset(ExpressionType& ptr, uint32_t offset) |
291 | { |
292 | if (static_cast<uint64_t>(offset) > static_cast<uint64_t>(std::numeric_limits<int32_t>::max())) { |
293 | ptr = m_currentBlock->appendNew<Value>(m_proc, Add, origin(), ptr, m_currentBlock->appendNew<Const64Value>(m_proc, origin(), offset)); |
294 | return 0; |
295 | } |
296 | return offset; |
297 | } |
298 | |
299 | void B3IRGenerator::restoreWasmContextInstance(Procedure& proc, BasicBlock* block, Value* arg) |
300 | { |
301 | if (Context::useFastTLS()) { |
302 | PatchpointValue* patchpoint = block->appendNew<PatchpointValue>(proc, B3::Void, Origin()); |
303 | if (CCallHelpers::storeWasmContextInstanceNeedsMacroScratchRegister()) |
304 | patchpoint->clobber(RegisterSet::macroScratchRegisters()); |
305 | patchpoint->append(ConstrainedValue(arg, ValueRep::SomeRegister)); |
306 | patchpoint->setGenerator( |
307 | [=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
308 | AllowMacroScratchRegisterUsageIf allowScratch(jit, CCallHelpers::storeWasmContextInstanceNeedsMacroScratchRegister()); |
309 | jit.storeWasmContextInstance(params[0].gpr()); |
310 | }); |
311 | return; |
312 | } |
313 | |
314 | // FIXME: Because WasmToWasm call clobbers wasmContextInstance register and does not restore it, we need to restore it in the caller side. |
315 | // This prevents us from using ArgumentReg to this (logically) immutable pinned register. |
316 | PatchpointValue* patchpoint = block->appendNew<PatchpointValue>(proc, B3::Void, Origin()); |
317 | Effects effects = Effects::none(); |
318 | effects.writesPinned = true; |
319 | effects.reads = B3::HeapRange::top(); |
320 | patchpoint->effects = effects; |
321 | patchpoint->clobberLate(RegisterSet(m_wasmContextInstanceGPR)); |
322 | patchpoint->append(arg, ValueRep::SomeRegister); |
323 | GPRReg wasmContextInstanceGPR = m_wasmContextInstanceGPR; |
324 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& param) { |
325 | jit.move(param[0].gpr(), wasmContextInstanceGPR); |
326 | }); |
327 | } |
328 | |
329 | B3IRGenerator::B3IRGenerator(const ModuleInformation& info, Procedure& procedure, InternalFunction* compilation, Vector<UnlinkedWasmToWasmCall>& unlinkedWasmToWasmCalls, MemoryMode mode, CompilationMode compilationMode, unsigned functionIndex, TierUpCount* tierUp, ThrowWasmException throwWasmException) |
330 | : m_info(info) |
331 | , m_mode(mode) |
332 | , m_compilationMode(compilationMode) |
333 | , m_functionIndex(functionIndex) |
334 | , m_tierUp(tierUp) |
335 | , m_proc(procedure) |
336 | , m_unlinkedWasmToWasmCalls(unlinkedWasmToWasmCalls) |
337 | , m_constantInsertionValues(m_proc) |
338 | { |
339 | m_currentBlock = m_proc.addBlock(); |
340 | |
341 | // FIXME we don't really need to pin registers here if there's no memory. It makes wasm -> wasm thunks simpler for now. https://bugs.webkit.org/show_bug.cgi?id=166623 |
342 | const PinnedRegisterInfo& pinnedRegs = PinnedRegisterInfo::get(); |
343 | |
344 | m_memoryBaseGPR = pinnedRegs.baseMemoryPointer; |
345 | m_proc.pinRegister(m_memoryBaseGPR); |
346 | |
347 | m_wasmContextInstanceGPR = pinnedRegs.wasmContextInstancePointer; |
348 | if (!Context::useFastTLS()) |
349 | m_proc.pinRegister(m_wasmContextInstanceGPR); |
350 | |
351 | if (mode != MemoryMode::Signaling) { |
352 | m_memorySizeGPR = pinnedRegs.sizeRegister; |
353 | m_proc.pinRegister(m_memorySizeGPR); |
354 | } |
355 | |
356 | if (throwWasmException) |
357 | Thunks::singleton().setThrowWasmException(throwWasmException); |
358 | |
359 | if (info.memory) { |
360 | m_proc.setWasmBoundsCheckGenerator([=] (CCallHelpers& jit, GPRReg pinnedGPR) { |
361 | AllowMacroScratchRegisterUsage allowScratch(jit); |
362 | switch (m_mode) { |
363 | case MemoryMode::BoundsChecking: |
364 | ASSERT_UNUSED(pinnedGPR, m_memorySizeGPR == pinnedGPR); |
365 | break; |
366 | case MemoryMode::Signaling: |
367 | ASSERT_UNUSED(pinnedGPR, InvalidGPRReg == pinnedGPR); |
368 | break; |
369 | } |
370 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsMemoryAccess); |
371 | }); |
372 | |
373 | switch (m_mode) { |
374 | case MemoryMode::BoundsChecking: |
375 | break; |
376 | case MemoryMode::Signaling: |
377 | // Most memory accesses in signaling mode don't do an explicit |
378 | // exception check because they can rely on fault handling to detect |
379 | // out-of-bounds accesses. FaultSignalHandler nonetheless needs the |
380 | // thunk to exist so that it can jump to that thunk. |
381 | if (UNLIKELY(!Thunks::singleton().stub(throwExceptionFromWasmThunkGenerator))) |
382 | CRASH(); |
383 | break; |
384 | } |
385 | } |
386 | |
387 | wasmCallingConvention().setupFrameInPrologue(&compilation->calleeMoveLocation, m_proc, Origin(), m_currentBlock); |
388 | |
389 | { |
390 | B3::Value* framePointer = m_currentBlock->appendNew<B3::Value>(m_proc, B3::FramePointer, Origin()); |
391 | B3::PatchpointValue* stackOverflowCheck = m_currentBlock->appendNew<B3::PatchpointValue>(m_proc, pointerType(), Origin()); |
392 | m_instanceValue = stackOverflowCheck; |
393 | stackOverflowCheck->appendSomeRegister(framePointer); |
394 | stackOverflowCheck->clobber(RegisterSet::macroScratchRegisters()); |
395 | if (!Context::useFastTLS()) { |
396 | // FIXME: Because WasmToWasm call clobbers wasmContextInstance register and does not restore it, we need to restore it in the caller side. |
397 | // This prevents us from using ArgumentReg to this (logically) immutable pinned register. |
398 | stackOverflowCheck->effects.writesPinned = false; |
399 | stackOverflowCheck->effects.readsPinned = true; |
400 | stackOverflowCheck->resultConstraint = ValueRep::reg(m_wasmContextInstanceGPR); |
401 | } |
402 | stackOverflowCheck->numGPScratchRegisters = 2; |
403 | stackOverflowCheck->setGenerator([=] (CCallHelpers& jit, const B3::StackmapGenerationParams& params) { |
404 | const Checked<int32_t> wasmFrameSize = params.proc().frameSize(); |
405 | const unsigned minimumParentCheckSize = WTF::roundUpToMultipleOf(stackAlignmentBytes(), 1024); |
406 | const unsigned = WTF::roundUpToMultipleOf(stackAlignmentBytes(), std::max<uint32_t>( |
407 | // This allows us to elide stack checks for functions that are terminal nodes in the call |
408 | // tree, (e.g they don't make any calls) and have a small enough frame size. This works by |
409 | // having any such terminal node have its parent caller include some extra size in its |
410 | // own check for it. The goal here is twofold: |
411 | // 1. Emit less code. |
412 | // 2. Try to speed things up by skipping stack checks. |
413 | minimumParentCheckSize, |
414 | // This allows us to elide stack checks in the Wasm -> Embedder call IC stub. Since these will |
415 | // spill all arguments to the stack, we ensure that a stack check here covers the |
416 | // stack that such a stub would use. |
417 | (Checked<uint32_t>(m_maxNumJSCallArguments) * sizeof(Register) + jscCallingConvention().headerSizeInBytes()).unsafeGet() |
418 | )); |
419 | const int32_t checkSize = m_makesCalls ? (wasmFrameSize + extraFrameSize).unsafeGet() : wasmFrameSize.unsafeGet(); |
420 | bool needUnderflowCheck = static_cast<unsigned>(checkSize) > Options::reservedZoneSize(); |
421 | bool needsOverflowCheck = m_makesCalls || wasmFrameSize >= minimumParentCheckSize || needUnderflowCheck; |
422 | |
423 | GPRReg contextInstance = Context::useFastTLS() ? params[0].gpr() : m_wasmContextInstanceGPR; |
424 | |
425 | // This allows leaf functions to not do stack checks if their frame size is within |
426 | // certain limits since their caller would have already done the check. |
427 | if (needsOverflowCheck) { |
428 | AllowMacroScratchRegisterUsage allowScratch(jit); |
429 | GPRReg fp = params[1].gpr(); |
430 | GPRReg scratch1 = params.gpScratch(0); |
431 | GPRReg scratch2 = params.gpScratch(1); |
432 | |
433 | if (Context::useFastTLS()) |
434 | jit.loadWasmContextInstance(contextInstance); |
435 | |
436 | jit.loadPtr(CCallHelpers::Address(contextInstance, Instance::offsetOfCachedStackLimit()), scratch2); |
437 | jit.addPtr(CCallHelpers::TrustedImm32(-checkSize), fp, scratch1); |
438 | MacroAssembler::JumpList overflow; |
439 | if (UNLIKELY(needUnderflowCheck)) |
440 | overflow.append(jit.branchPtr(CCallHelpers::Above, scratch1, fp)); |
441 | overflow.append(jit.branchPtr(CCallHelpers::Below, scratch1, scratch2)); |
442 | jit.addLinkTask([overflow] (LinkBuffer& linkBuffer) { |
443 | linkBuffer.link(overflow, CodeLocationLabel<JITThunkPtrTag>(Thunks::singleton().stub(throwStackOverflowFromWasmThunkGenerator).code())); |
444 | }); |
445 | } else if (m_usesInstanceValue && Context::useFastTLS()) { |
446 | // No overflow check is needed, but the instance values still needs to be correct. |
447 | AllowMacroScratchRegisterUsageIf allowScratch(jit, CCallHelpers::loadWasmContextInstanceNeedsMacroScratchRegister()); |
448 | jit.loadWasmContextInstance(contextInstance); |
449 | } else { |
450 | // We said we'd return a pointer. We don't actually need to because it isn't used, but the patchpoint conservatively said it had effects (potential stack check) which prevent it from getting removed. |
451 | } |
452 | }); |
453 | } |
454 | |
455 | emitTierUpCheck(TierUpCount::functionEntryDecrement(), Origin()); |
456 | } |
457 | |
458 | void B3IRGenerator::restoreWebAssemblyGlobalState(RestoreCachedStackLimit restoreCachedStackLimit, const MemoryInformation& memory, Value* instance, Procedure& proc, BasicBlock* block) |
459 | { |
460 | restoreWasmContextInstance(proc, block, instance); |
461 | |
462 | if (restoreCachedStackLimit == RestoreCachedStackLimit::Yes) { |
463 | // The Instance caches the stack limit, but also knows where its canonical location is. |
464 | Value* pointerToActualStackLimit = block->appendNew<MemoryValue>(m_proc, Load, pointerType(), origin(), instanceValue(), safeCast<int32_t>(Instance::offsetOfPointerToActualStackLimit())); |
465 | Value* actualStackLimit = block->appendNew<MemoryValue>(m_proc, Load, pointerType(), origin(), pointerToActualStackLimit); |
466 | block->appendNew<MemoryValue>(m_proc, Store, origin(), actualStackLimit, instanceValue(), safeCast<int32_t>(Instance::offsetOfCachedStackLimit())); |
467 | } |
468 | |
469 | if (!!memory) { |
470 | const PinnedRegisterInfo* pinnedRegs = &PinnedRegisterInfo::get(); |
471 | RegisterSet clobbers; |
472 | clobbers.set(pinnedRegs->baseMemoryPointer); |
473 | clobbers.set(pinnedRegs->sizeRegister); |
474 | if (!isARM64()) |
475 | clobbers.set(RegisterSet::macroScratchRegisters()); |
476 | |
477 | B3::PatchpointValue* patchpoint = block->appendNew<B3::PatchpointValue>(proc, B3::Void, origin()); |
478 | Effects effects = Effects::none(); |
479 | effects.writesPinned = true; |
480 | effects.reads = B3::HeapRange::top(); |
481 | patchpoint->effects = effects; |
482 | patchpoint->clobber(clobbers); |
483 | patchpoint->numGPScratchRegisters = Gigacage::isEnabled(Gigacage::Primitive) ? 1 : 0; |
484 | |
485 | patchpoint->append(instance, ValueRep::SomeRegister); |
486 | patchpoint->setGenerator([pinnedRegs] (CCallHelpers& jit, const B3::StackmapGenerationParams& params) { |
487 | RELEASE_ASSERT(!Gigacage::isEnabled(Gigacage::Primitive) || !isARM64()); |
488 | AllowMacroScratchRegisterUsageIf allowScratch(jit, !isARM64()); |
489 | GPRReg baseMemory = pinnedRegs->baseMemoryPointer; |
490 | GPRReg scratchOrSize = Gigacage::isEnabled(Gigacage::Primitive) ? params.gpScratch(0) : pinnedRegs->sizeRegister; |
491 | |
492 | jit.loadPtr(CCallHelpers::Address(params[0].gpr(), Instance::offsetOfCachedMemorySize()), pinnedRegs->sizeRegister); |
493 | jit.loadPtr(CCallHelpers::Address(params[0].gpr(), Instance::offsetOfCachedMemory()), baseMemory); |
494 | |
495 | jit.cageConditionally(Gigacage::Primitive, baseMemory, scratchOrSize); |
496 | }); |
497 | } |
498 | } |
499 | |
500 | void B3IRGenerator::emitExceptionCheck(CCallHelpers& jit, ExceptionType type) |
501 | { |
502 | jit.move(CCallHelpers::TrustedImm32(static_cast<uint32_t>(type)), GPRInfo::argumentGPR1); |
503 | auto jumpToExceptionStub = jit.jump(); |
504 | |
505 | jit.addLinkTask([jumpToExceptionStub] (LinkBuffer& linkBuffer) { |
506 | linkBuffer.link(jumpToExceptionStub, CodeLocationLabel<JITThunkPtrTag>(Thunks::singleton().stub(throwExceptionFromWasmThunkGenerator).code())); |
507 | }); |
508 | } |
509 | |
510 | Value* B3IRGenerator::constant(B3::Type type, uint64_t bits, Optional<Origin> maybeOrigin) |
511 | { |
512 | auto result = m_constantPool.ensure(ValueKey(opcodeForConstant(type), type, static_cast<int64_t>(bits)), [&] { |
513 | Value* result = m_proc.addConstant(maybeOrigin ? *maybeOrigin : origin(), type, bits); |
514 | m_constantInsertionValues.insertValue(0, result); |
515 | return result; |
516 | }); |
517 | return result.iterator->value; |
518 | } |
519 | |
520 | void B3IRGenerator::insertConstants() |
521 | { |
522 | m_constantInsertionValues.execute(m_proc.at(0)); |
523 | } |
524 | |
525 | auto B3IRGenerator::addLocal(Type type, uint32_t count) -> PartialResult |
526 | { |
527 | Checked<uint32_t, RecordOverflow> totalBytesChecked = count; |
528 | totalBytesChecked += m_locals.size(); |
529 | uint32_t totalBytes; |
530 | WASM_COMPILE_FAIL_IF((totalBytesChecked.safeGet(totalBytes) == CheckedState::DidOverflow) || !m_locals.tryReserveCapacity(totalBytes), "can't allocate memory for " , totalBytes, " locals" ); |
531 | |
532 | for (uint32_t i = 0; i < count; ++i) { |
533 | Variable* local = m_proc.addVariable(toB3Type(type)); |
534 | m_locals.uncheckedAppend(local); |
535 | m_currentBlock->appendNew<VariableValue>(m_proc, Set, Origin(), local, constant(toB3Type(type), 0, Origin())); |
536 | } |
537 | return { }; |
538 | } |
539 | |
540 | auto B3IRGenerator::addArguments(const Signature& signature) -> PartialResult |
541 | { |
542 | ASSERT(!m_locals.size()); |
543 | WASM_COMPILE_FAIL_IF(!m_locals.tryReserveCapacity(signature.argumentCount()), "can't allocate memory for " , signature.argumentCount(), " arguments" ); |
544 | |
545 | m_locals.grow(signature.argumentCount()); |
546 | wasmCallingConvention().loadArguments(signature, m_proc, m_currentBlock, Origin(), |
547 | [=] (ExpressionType argument, unsigned i) { |
548 | Variable* argumentVariable = m_proc.addVariable(argument->type()); |
549 | m_locals[i] = argumentVariable; |
550 | m_currentBlock->appendNew<VariableValue>(m_proc, Set, Origin(), argumentVariable, argument); |
551 | }); |
552 | return { }; |
553 | } |
554 | |
555 | auto B3IRGenerator::addRefIsNull(ExpressionType& value, ExpressionType& result) -> PartialResult |
556 | { |
557 | result = m_currentBlock->appendNew<Value>(m_proc, B3::Equal, origin(), value, m_currentBlock->appendNew<Const64Value>(m_proc, origin(), JSValue::encode(jsNull()))); |
558 | return { }; |
559 | } |
560 | |
561 | auto B3IRGenerator::getLocal(uint32_t index, ExpressionType& result) -> PartialResult |
562 | { |
563 | ASSERT(m_locals[index]); |
564 | result = m_currentBlock->appendNew<VariableValue>(m_proc, B3::Get, origin(), m_locals[index]); |
565 | return { }; |
566 | } |
567 | |
568 | auto B3IRGenerator::addUnreachable() -> PartialResult |
569 | { |
570 | B3::PatchpointValue* unreachable = m_currentBlock->appendNew<B3::PatchpointValue>(m_proc, B3::Void, origin()); |
571 | unreachable->setGenerator([this] (CCallHelpers& jit, const B3::StackmapGenerationParams&) { |
572 | this->emitExceptionCheck(jit, ExceptionType::Unreachable); |
573 | }); |
574 | unreachable->effects.terminal = true; |
575 | return { }; |
576 | } |
577 | |
578 | auto B3IRGenerator::addGrowMemory(ExpressionType delta, ExpressionType& result) -> PartialResult |
579 | { |
580 | int32_t (*growMemory)(void*, Instance*, int32_t) = [] (void* callFrame, Instance* instance, int32_t delta) -> int32_t { |
581 | instance->storeTopCallFrame(callFrame); |
582 | |
583 | if (delta < 0) |
584 | return -1; |
585 | |
586 | auto grown = instance->memory()->grow(PageCount(delta)); |
587 | if (!grown) { |
588 | switch (grown.error()) { |
589 | case Memory::GrowFailReason::InvalidDelta: |
590 | case Memory::GrowFailReason::InvalidGrowSize: |
591 | case Memory::GrowFailReason::WouldExceedMaximum: |
592 | case Memory::GrowFailReason::OutOfMemory: |
593 | return -1; |
594 | } |
595 | RELEASE_ASSERT_NOT_REACHED(); |
596 | } |
597 | |
598 | return grown.value().pageCount(); |
599 | }; |
600 | |
601 | result = m_currentBlock->appendNew<CCallValue>(m_proc, Int32, origin(), |
602 | m_currentBlock->appendNew<ConstPtrValue>(m_proc, origin(), tagCFunctionPtr<void*>(growMemory, B3CCallPtrTag)), |
603 | m_currentBlock->appendNew<B3::Value>(m_proc, B3::FramePointer, origin()), instanceValue(), delta); |
604 | |
605 | restoreWebAssemblyGlobalState(RestoreCachedStackLimit::No, m_info.memory, instanceValue(), m_proc, m_currentBlock); |
606 | |
607 | return { }; |
608 | } |
609 | |
610 | auto B3IRGenerator::addCurrentMemory(ExpressionType& result) -> PartialResult |
611 | { |
612 | static_assert(sizeof(decltype(static_cast<Memory*>(nullptr)->size())) == sizeof(uint64_t), "codegen relies on this size" ); |
613 | Value* size = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, Int64, origin(), instanceValue(), safeCast<int32_t>(Instance::offsetOfCachedMemorySize())); |
614 | |
615 | constexpr uint32_t shiftValue = 16; |
616 | static_assert(PageCount::pageSize == 1ull << shiftValue, "This must hold for the code below to be correct." ); |
617 | Value* numPages = m_currentBlock->appendNew<Value>(m_proc, ZShr, origin(), |
618 | size, m_currentBlock->appendNew<Const32Value>(m_proc, origin(), shiftValue)); |
619 | |
620 | result = m_currentBlock->appendNew<Value>(m_proc, Trunc, origin(), numPages); |
621 | |
622 | return { }; |
623 | } |
624 | |
625 | auto B3IRGenerator::setLocal(uint32_t index, ExpressionType value) -> PartialResult |
626 | { |
627 | ASSERT(m_locals[index]); |
628 | m_currentBlock->appendNew<VariableValue>(m_proc, B3::Set, origin(), m_locals[index], value); |
629 | return { }; |
630 | } |
631 | |
632 | auto B3IRGenerator::getGlobal(uint32_t index, ExpressionType& result) -> PartialResult |
633 | { |
634 | Value* globalsArray = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, pointerType(), origin(), instanceValue(), safeCast<int32_t>(Instance::offsetOfGlobals())); |
635 | result = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, toB3Type(m_info.globals[index].type), origin(), globalsArray, safeCast<int32_t>(index * sizeof(Register))); |
636 | return { }; |
637 | } |
638 | |
639 | auto B3IRGenerator::setGlobal(uint32_t index, ExpressionType value) -> PartialResult |
640 | { |
641 | ASSERT(toB3Type(m_info.globals[index].type) == value->type()); |
642 | Value* globalsArray = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, pointerType(), origin(), instanceValue(), safeCast<int32_t>(Instance::offsetOfGlobals())); |
643 | m_currentBlock->appendNew<MemoryValue>(m_proc, Store, origin(), value, globalsArray, safeCast<int32_t>(index * sizeof(Register))); |
644 | return { }; |
645 | } |
646 | |
647 | inline Value* B3IRGenerator::emitCheckAndPreparePointer(ExpressionType pointer, uint32_t offset, uint32_t sizeOfOperation) |
648 | { |
649 | ASSERT(m_memoryBaseGPR); |
650 | |
651 | switch (m_mode) { |
652 | case MemoryMode::BoundsChecking: { |
653 | // We're not using signal handling at all, we must therefore check that no memory access exceeds the current memory size. |
654 | ASSERT(m_memorySizeGPR); |
655 | ASSERT(sizeOfOperation + offset > offset); |
656 | m_currentBlock->appendNew<WasmBoundsCheckValue>(m_proc, origin(), m_memorySizeGPR, pointer, sizeOfOperation + offset - 1); |
657 | break; |
658 | } |
659 | |
660 | case MemoryMode::Signaling: { |
661 | // We've virtually mapped 4GiB+redzone for this memory. Only the user-allocated pages are addressable, contiguously in range [0, current], |
662 | // and everything above is mapped PROT_NONE. We don't need to perform any explicit bounds check in the 4GiB range because WebAssembly register |
663 | // memory accesses are 32-bit. However WebAssembly register + offset accesses perform the addition in 64-bit which can push an access above |
664 | // the 32-bit limit (the offset is unsigned 32-bit). The redzone will catch most small offsets, and we'll explicitly bounds check any |
665 | // register + large offset access. We don't think this will be generated frequently. |
666 | // |
667 | // We could check that register + large offset doesn't exceed 4GiB+redzone since that's technically the limit we need to avoid overflowing the |
668 | // PROT_NONE region, but it's better if we use a smaller immediate because it can codegens better. We know that anything equal to or greater |
669 | // than the declared 'maximum' will trap, so we can compare against that number. If there was no declared 'maximum' then we still know that |
670 | // any access equal to or greater than 4GiB will trap, no need to add the redzone. |
671 | if (offset >= Memory::fastMappedRedzoneBytes()) { |
672 | size_t maximum = m_info.memory.maximum() ? m_info.memory.maximum().bytes() : std::numeric_limits<uint32_t>::max(); |
673 | m_currentBlock->appendNew<WasmBoundsCheckValue>(m_proc, origin(), pointer, sizeOfOperation + offset - 1, maximum); |
674 | } |
675 | break; |
676 | } |
677 | } |
678 | |
679 | pointer = m_currentBlock->appendNew<Value>(m_proc, ZExt32, origin(), pointer); |
680 | return m_currentBlock->appendNew<WasmAddressValue>(m_proc, origin(), pointer, m_memoryBaseGPR); |
681 | } |
682 | |
683 | inline uint32_t sizeOfLoadOp(LoadOpType op) |
684 | { |
685 | switch (op) { |
686 | case LoadOpType::I32Load8S: |
687 | case LoadOpType::I32Load8U: |
688 | case LoadOpType::I64Load8S: |
689 | case LoadOpType::I64Load8U: |
690 | return 1; |
691 | case LoadOpType::I32Load16S: |
692 | case LoadOpType::I64Load16S: |
693 | case LoadOpType::I32Load16U: |
694 | case LoadOpType::I64Load16U: |
695 | return 2; |
696 | case LoadOpType::I32Load: |
697 | case LoadOpType::I64Load32S: |
698 | case LoadOpType::I64Load32U: |
699 | case LoadOpType::F32Load: |
700 | return 4; |
701 | case LoadOpType::I64Load: |
702 | case LoadOpType::F64Load: |
703 | return 8; |
704 | } |
705 | RELEASE_ASSERT_NOT_REACHED(); |
706 | } |
707 | |
708 | inline B3::Kind B3IRGenerator::memoryKind(B3::Opcode memoryOp) |
709 | { |
710 | if (m_mode == MemoryMode::Signaling) |
711 | return trapping(memoryOp); |
712 | return memoryOp; |
713 | } |
714 | |
715 | inline Value* B3IRGenerator::emitLoadOp(LoadOpType op, ExpressionType pointer, uint32_t uoffset) |
716 | { |
717 | int32_t offset = fixupPointerPlusOffset(pointer, uoffset); |
718 | |
719 | switch (op) { |
720 | case LoadOpType::I32Load8S: { |
721 | return m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load8S), origin(), pointer, offset); |
722 | } |
723 | |
724 | case LoadOpType::I64Load8S: { |
725 | Value* value = m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load8S), origin(), pointer, offset); |
726 | return m_currentBlock->appendNew<Value>(m_proc, SExt32, origin(), value); |
727 | } |
728 | |
729 | case LoadOpType::I32Load8U: { |
730 | return m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load8Z), origin(), pointer, offset); |
731 | } |
732 | |
733 | case LoadOpType::I64Load8U: { |
734 | Value* value = m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load8Z), origin(), pointer, offset); |
735 | return m_currentBlock->appendNew<Value>(m_proc, ZExt32, origin(), value); |
736 | } |
737 | |
738 | case LoadOpType::I32Load16S: { |
739 | return m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load16S), origin(), pointer, offset); |
740 | } |
741 | |
742 | case LoadOpType::I64Load16S: { |
743 | Value* value = m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load16S), origin(), pointer, offset); |
744 | return m_currentBlock->appendNew<Value>(m_proc, SExt32, origin(), value); |
745 | } |
746 | |
747 | case LoadOpType::I32Load16U: { |
748 | return m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load16Z), origin(), pointer, offset); |
749 | } |
750 | |
751 | case LoadOpType::I64Load16U: { |
752 | Value* value = m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load16Z), origin(), pointer, offset); |
753 | return m_currentBlock->appendNew<Value>(m_proc, ZExt32, origin(), value); |
754 | } |
755 | |
756 | case LoadOpType::I32Load: { |
757 | return m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load), Int32, origin(), pointer, offset); |
758 | } |
759 | |
760 | case LoadOpType::I64Load32U: { |
761 | Value* value = m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load), Int32, origin(), pointer, offset); |
762 | return m_currentBlock->appendNew<Value>(m_proc, ZExt32, origin(), value); |
763 | } |
764 | |
765 | case LoadOpType::I64Load32S: { |
766 | Value* value = m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load), Int32, origin(), pointer, offset); |
767 | return m_currentBlock->appendNew<Value>(m_proc, SExt32, origin(), value); |
768 | } |
769 | |
770 | case LoadOpType::I64Load: { |
771 | return m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load), Int64, origin(), pointer, offset); |
772 | } |
773 | |
774 | case LoadOpType::F32Load: { |
775 | return m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load), Float, origin(), pointer, offset); |
776 | } |
777 | |
778 | case LoadOpType::F64Load: { |
779 | return m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Load), Double, origin(), pointer, offset); |
780 | } |
781 | } |
782 | RELEASE_ASSERT_NOT_REACHED(); |
783 | } |
784 | |
785 | auto B3IRGenerator::load(LoadOpType op, ExpressionType pointer, ExpressionType& result, uint32_t offset) -> PartialResult |
786 | { |
787 | ASSERT(pointer->type() == Int32); |
788 | |
789 | if (UNLIKELY(sumOverflows<uint32_t>(offset, sizeOfLoadOp(op)))) { |
790 | // FIXME: Even though this is provably out of bounds, it's not a validation error, so we have to handle it |
791 | // as a runtime exception. However, this may change: https://bugs.webkit.org/show_bug.cgi?id=166435 |
792 | B3::PatchpointValue* throwException = m_currentBlock->appendNew<B3::PatchpointValue>(m_proc, B3::Void, origin()); |
793 | throwException->setGenerator([this] (CCallHelpers& jit, const B3::StackmapGenerationParams&) { |
794 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsMemoryAccess); |
795 | }); |
796 | |
797 | switch (op) { |
798 | case LoadOpType::I32Load8S: |
799 | case LoadOpType::I32Load16S: |
800 | case LoadOpType::I32Load: |
801 | case LoadOpType::I32Load16U: |
802 | case LoadOpType::I32Load8U: |
803 | result = constant(Int32, 0); |
804 | break; |
805 | case LoadOpType::I64Load8S: |
806 | case LoadOpType::I64Load8U: |
807 | case LoadOpType::I64Load16S: |
808 | case LoadOpType::I64Load32U: |
809 | case LoadOpType::I64Load32S: |
810 | case LoadOpType::I64Load: |
811 | case LoadOpType::I64Load16U: |
812 | result = constant(Int64, 0); |
813 | break; |
814 | case LoadOpType::F32Load: |
815 | result = constant(Float, 0); |
816 | break; |
817 | case LoadOpType::F64Load: |
818 | result = constant(Double, 0); |
819 | break; |
820 | } |
821 | |
822 | } else |
823 | result = emitLoadOp(op, emitCheckAndPreparePointer(pointer, offset, sizeOfLoadOp(op)), offset); |
824 | |
825 | return { }; |
826 | } |
827 | |
828 | inline uint32_t sizeOfStoreOp(StoreOpType op) |
829 | { |
830 | switch (op) { |
831 | case StoreOpType::I32Store8: |
832 | case StoreOpType::I64Store8: |
833 | return 1; |
834 | case StoreOpType::I32Store16: |
835 | case StoreOpType::I64Store16: |
836 | return 2; |
837 | case StoreOpType::I32Store: |
838 | case StoreOpType::I64Store32: |
839 | case StoreOpType::F32Store: |
840 | return 4; |
841 | case StoreOpType::I64Store: |
842 | case StoreOpType::F64Store: |
843 | return 8; |
844 | } |
845 | RELEASE_ASSERT_NOT_REACHED(); |
846 | } |
847 | |
848 | |
849 | inline void B3IRGenerator::emitStoreOp(StoreOpType op, ExpressionType pointer, ExpressionType value, uint32_t uoffset) |
850 | { |
851 | int32_t offset = fixupPointerPlusOffset(pointer, uoffset); |
852 | |
853 | switch (op) { |
854 | case StoreOpType::I64Store8: |
855 | value = m_currentBlock->appendNew<Value>(m_proc, Trunc, origin(), value); |
856 | FALLTHROUGH; |
857 | |
858 | case StoreOpType::I32Store8: |
859 | m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Store8), origin(), value, pointer, offset); |
860 | return; |
861 | |
862 | case StoreOpType::I64Store16: |
863 | value = m_currentBlock->appendNew<Value>(m_proc, Trunc, origin(), value); |
864 | FALLTHROUGH; |
865 | |
866 | case StoreOpType::I32Store16: |
867 | m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Store16), origin(), value, pointer, offset); |
868 | return; |
869 | |
870 | case StoreOpType::I64Store32: |
871 | value = m_currentBlock->appendNew<Value>(m_proc, Trunc, origin(), value); |
872 | FALLTHROUGH; |
873 | |
874 | case StoreOpType::I64Store: |
875 | case StoreOpType::I32Store: |
876 | case StoreOpType::F32Store: |
877 | case StoreOpType::F64Store: |
878 | m_currentBlock->appendNew<MemoryValue>(m_proc, memoryKind(Store), origin(), value, pointer, offset); |
879 | return; |
880 | } |
881 | RELEASE_ASSERT_NOT_REACHED(); |
882 | } |
883 | |
884 | auto B3IRGenerator::store(StoreOpType op, ExpressionType pointer, ExpressionType value, uint32_t offset) -> PartialResult |
885 | { |
886 | ASSERT(pointer->type() == Int32); |
887 | |
888 | if (UNLIKELY(sumOverflows<uint32_t>(offset, sizeOfStoreOp(op)))) { |
889 | // FIXME: Even though this is provably out of bounds, it's not a validation error, so we have to handle it |
890 | // as a runtime exception. However, this may change: https://bugs.webkit.org/show_bug.cgi?id=166435 |
891 | B3::PatchpointValue* throwException = m_currentBlock->appendNew<B3::PatchpointValue>(m_proc, B3::Void, origin()); |
892 | throwException->setGenerator([this] (CCallHelpers& jit, const B3::StackmapGenerationParams&) { |
893 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsMemoryAccess); |
894 | }); |
895 | } else |
896 | emitStoreOp(op, emitCheckAndPreparePointer(pointer, offset, sizeOfStoreOp(op)), value, offset); |
897 | |
898 | return { }; |
899 | } |
900 | |
901 | auto B3IRGenerator::addSelect(ExpressionType condition, ExpressionType nonZero, ExpressionType zero, ExpressionType& result) -> PartialResult |
902 | { |
903 | result = m_currentBlock->appendNew<Value>(m_proc, B3::Select, origin(), condition, nonZero, zero); |
904 | return { }; |
905 | } |
906 | |
907 | B3IRGenerator::ExpressionType B3IRGenerator::addConstant(Type type, uint64_t value) |
908 | { |
909 | return constant(toB3Type(type), value); |
910 | } |
911 | |
912 | void B3IRGenerator::emitTierUpCheck(uint32_t decrementCount, Origin origin) |
913 | { |
914 | if (!m_tierUp) |
915 | return; |
916 | |
917 | ASSERT(m_tierUp); |
918 | Value* countDownLocation = constant(pointerType(), reinterpret_cast<uint64_t>(m_tierUp), origin); |
919 | Value* oldCountDown = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, Int32, origin, countDownLocation); |
920 | Value* newCountDown = m_currentBlock->appendNew<Value>(m_proc, Sub, origin, oldCountDown, constant(Int32, decrementCount, origin)); |
921 | m_currentBlock->appendNew<MemoryValue>(m_proc, Store, origin, newCountDown, countDownLocation); |
922 | |
923 | PatchpointValue* patch = m_currentBlock->appendNew<PatchpointValue>(m_proc, B3::Void, origin); |
924 | Effects effects = Effects::none(); |
925 | // FIXME: we should have a more precise heap range for the tier up count. |
926 | effects.reads = B3::HeapRange::top(); |
927 | effects.writes = B3::HeapRange::top(); |
928 | patch->effects = effects; |
929 | |
930 | patch->append(newCountDown, ValueRep::SomeRegister); |
931 | patch->append(oldCountDown, ValueRep::SomeRegister); |
932 | patch->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
933 | MacroAssembler::Jump tierUp = jit.branch32(MacroAssembler::Above, params[0].gpr(), params[1].gpr()); |
934 | MacroAssembler::Label tierUpResume = jit.label(); |
935 | |
936 | params.addLatePath([=] (CCallHelpers& jit) { |
937 | tierUp.link(&jit); |
938 | |
939 | const unsigned = 0; |
940 | RegisterSet registersToSpill = { }; |
941 | registersToSpill.add(GPRInfo::argumentGPR1); |
942 | unsigned numberOfStackBytesUsedForRegisterPreservation = ScratchRegisterAllocator::preserveRegistersToStackForCall(jit, registersToSpill, extraPaddingBytes); |
943 | |
944 | jit.move(MacroAssembler::TrustedImm32(m_functionIndex), GPRInfo::argumentGPR1); |
945 | MacroAssembler::Call call = jit.nearCall(); |
946 | |
947 | ScratchRegisterAllocator::restoreRegistersFromStackForCall(jit, registersToSpill, RegisterSet(), numberOfStackBytesUsedForRegisterPreservation, extraPaddingBytes); |
948 | jit.jump(tierUpResume); |
949 | |
950 | jit.addLinkTask([=] (LinkBuffer& linkBuffer) { |
951 | MacroAssembler::repatchNearCall(linkBuffer.locationOfNearCall<NoPtrTag>(call), CodeLocationLabel<JITThunkPtrTag>(Thunks::singleton().stub(triggerOMGTierUpThunkGenerator).code())); |
952 | |
953 | }); |
954 | }); |
955 | }); |
956 | } |
957 | |
958 | B3IRGenerator::ControlData B3IRGenerator::addLoop(Type signature) |
959 | { |
960 | BasicBlock* body = m_proc.addBlock(); |
961 | BasicBlock* continuation = m_proc.addBlock(); |
962 | |
963 | m_currentBlock->appendNewControlValue(m_proc, Jump, origin(), body); |
964 | |
965 | m_currentBlock = body; |
966 | emitTierUpCheck(TierUpCount::loopDecrement(), origin()); |
967 | |
968 | return ControlData(m_proc, origin(), signature, BlockType::Loop, continuation, body); |
969 | } |
970 | |
971 | B3IRGenerator::ControlData B3IRGenerator::addTopLevel(Type signature) |
972 | { |
973 | return ControlData(m_proc, Origin(), signature, BlockType::TopLevel, m_proc.addBlock()); |
974 | } |
975 | |
976 | B3IRGenerator::ControlData B3IRGenerator::addBlock(Type signature) |
977 | { |
978 | return ControlData(m_proc, origin(), signature, BlockType::Block, m_proc.addBlock()); |
979 | } |
980 | |
981 | auto B3IRGenerator::addIf(ExpressionType condition, Type signature, ControlType& result) -> PartialResult |
982 | { |
983 | // FIXME: This needs to do some kind of stack passing. |
984 | |
985 | BasicBlock* taken = m_proc.addBlock(); |
986 | BasicBlock* notTaken = m_proc.addBlock(); |
987 | BasicBlock* continuation = m_proc.addBlock(); |
988 | |
989 | m_currentBlock->appendNew<Value>(m_proc, B3::Branch, origin(), condition); |
990 | m_currentBlock->setSuccessors(FrequentedBlock(taken), FrequentedBlock(notTaken)); |
991 | taken->addPredecessor(m_currentBlock); |
992 | notTaken->addPredecessor(m_currentBlock); |
993 | |
994 | m_currentBlock = taken; |
995 | result = ControlData(m_proc, origin(), signature, BlockType::If, continuation, notTaken); |
996 | return { }; |
997 | } |
998 | |
999 | auto B3IRGenerator::addElse(ControlData& data, const ExpressionList& currentStack) -> PartialResult |
1000 | { |
1001 | unifyValuesWithBlock(currentStack, data.result); |
1002 | m_currentBlock->appendNewControlValue(m_proc, Jump, origin(), data.continuation); |
1003 | return addElseToUnreachable(data); |
1004 | } |
1005 | |
1006 | auto B3IRGenerator::addElseToUnreachable(ControlData& data) -> PartialResult |
1007 | { |
1008 | ASSERT(data.type() == BlockType::If); |
1009 | m_currentBlock = data.special; |
1010 | data.convertIfToBlock(); |
1011 | return { }; |
1012 | } |
1013 | |
1014 | auto B3IRGenerator::addReturn(const ControlData&, const ExpressionList& returnValues) -> PartialResult |
1015 | { |
1016 | ASSERT(returnValues.size() <= 1); |
1017 | if (returnValues.size()) |
1018 | m_currentBlock->appendNewControlValue(m_proc, B3::Return, origin(), returnValues[0]); |
1019 | else |
1020 | m_currentBlock->appendNewControlValue(m_proc, B3::Return, origin()); |
1021 | return { }; |
1022 | } |
1023 | |
1024 | auto B3IRGenerator::addBranch(ControlData& data, ExpressionType condition, const ExpressionList& returnValues) -> PartialResult |
1025 | { |
1026 | unifyValuesWithBlock(returnValues, data.resultForBranch()); |
1027 | |
1028 | BasicBlock* target = data.targetBlockForBranch(); |
1029 | if (condition) { |
1030 | BasicBlock* continuation = m_proc.addBlock(); |
1031 | m_currentBlock->appendNew<Value>(m_proc, B3::Branch, origin(), condition); |
1032 | m_currentBlock->setSuccessors(FrequentedBlock(target), FrequentedBlock(continuation)); |
1033 | target->addPredecessor(m_currentBlock); |
1034 | continuation->addPredecessor(m_currentBlock); |
1035 | m_currentBlock = continuation; |
1036 | } else { |
1037 | m_currentBlock->appendNewControlValue(m_proc, Jump, origin(), FrequentedBlock(target)); |
1038 | target->addPredecessor(m_currentBlock); |
1039 | } |
1040 | |
1041 | return { }; |
1042 | } |
1043 | |
1044 | auto B3IRGenerator::addSwitch(ExpressionType condition, const Vector<ControlData*>& targets, ControlData& defaultTarget, const ExpressionList& expressionStack) -> PartialResult |
1045 | { |
1046 | for (size_t i = 0; i < targets.size(); ++i) |
1047 | unifyValuesWithBlock(expressionStack, targets[i]->resultForBranch()); |
1048 | unifyValuesWithBlock(expressionStack, defaultTarget.resultForBranch()); |
1049 | |
1050 | SwitchValue* switchValue = m_currentBlock->appendNew<SwitchValue>(m_proc, origin(), condition); |
1051 | switchValue->setFallThrough(FrequentedBlock(defaultTarget.targetBlockForBranch())); |
1052 | for (size_t i = 0; i < targets.size(); ++i) |
1053 | switchValue->appendCase(SwitchCase(i, FrequentedBlock(targets[i]->targetBlockForBranch()))); |
1054 | |
1055 | return { }; |
1056 | } |
1057 | |
1058 | auto B3IRGenerator::endBlock(ControlEntry& entry, ExpressionList& expressionStack) -> PartialResult |
1059 | { |
1060 | ControlData& data = entry.controlData; |
1061 | |
1062 | unifyValuesWithBlock(expressionStack, data.result); |
1063 | m_currentBlock->appendNewControlValue(m_proc, Jump, origin(), data.continuation); |
1064 | data.continuation->addPredecessor(m_currentBlock); |
1065 | |
1066 | return addEndToUnreachable(entry); |
1067 | } |
1068 | |
1069 | |
1070 | auto B3IRGenerator::addEndToUnreachable(ControlEntry& entry) -> PartialResult |
1071 | { |
1072 | ControlData& data = entry.controlData; |
1073 | m_currentBlock = data.continuation; |
1074 | |
1075 | if (data.type() == BlockType::If) { |
1076 | data.special->appendNewControlValue(m_proc, Jump, origin(), m_currentBlock); |
1077 | m_currentBlock->addPredecessor(data.special); |
1078 | } |
1079 | |
1080 | for (Value* result : data.result) { |
1081 | m_currentBlock->append(result); |
1082 | entry.enclosedExpressionStack.append(result); |
1083 | } |
1084 | |
1085 | // TopLevel does not have any code after this so we need to make sure we emit a return here. |
1086 | if (data.type() == BlockType::TopLevel) |
1087 | return addReturn(entry.controlData, entry.enclosedExpressionStack); |
1088 | |
1089 | return { }; |
1090 | } |
1091 | |
1092 | auto B3IRGenerator::addCall(uint32_t functionIndex, const Signature& signature, Vector<ExpressionType>& args, ExpressionType& result) -> PartialResult |
1093 | { |
1094 | ASSERT(signature.argumentCount() == args.size()); |
1095 | |
1096 | m_makesCalls = true; |
1097 | |
1098 | Type returnType = signature.returnType(); |
1099 | Vector<UnlinkedWasmToWasmCall>* unlinkedWasmToWasmCalls = &m_unlinkedWasmToWasmCalls; |
1100 | |
1101 | if (m_info.isImportedFunctionFromFunctionIndexSpace(functionIndex)) { |
1102 | m_maxNumJSCallArguments = std::max(m_maxNumJSCallArguments, static_cast<uint32_t>(args.size())); |
1103 | |
1104 | // FIXME imports can be linked here, instead of generating a patchpoint, because all import stubs are generated before B3 compilation starts. https://bugs.webkit.org/show_bug.cgi?id=166462 |
1105 | Value* targetInstance = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, pointerType(), origin(), instanceValue(), safeCast<int32_t>(Instance::offsetOfTargetInstance(functionIndex))); |
1106 | // The target instance is 0 unless the call is wasm->wasm. |
1107 | Value* isWasmCall = m_currentBlock->appendNew<Value>(m_proc, NotEqual, origin(), targetInstance, m_currentBlock->appendNew<Const64Value>(m_proc, origin(), 0)); |
1108 | |
1109 | BasicBlock* isWasmBlock = m_proc.addBlock(); |
1110 | BasicBlock* isEmbedderBlock = m_proc.addBlock(); |
1111 | BasicBlock* continuation = m_proc.addBlock(); |
1112 | m_currentBlock->appendNewControlValue(m_proc, B3::Branch, origin(), isWasmCall, FrequentedBlock(isWasmBlock), FrequentedBlock(isEmbedderBlock)); |
1113 | |
1114 | Value* wasmCallResult = wasmCallingConvention().setupCall(m_proc, isWasmBlock, origin(), args, toB3Type(returnType), |
1115 | [=] (PatchpointValue* patchpoint) { |
1116 | patchpoint->effects.writesPinned = true; |
1117 | patchpoint->effects.readsPinned = true; |
1118 | // We need to clobber all potential pinned registers since we might be leaving the instance. |
1119 | // We pessimistically assume we could be calling to something that is bounds checking. |
1120 | // FIXME: We shouldn't have to do this: https://bugs.webkit.org/show_bug.cgi?id=172181 |
1121 | patchpoint->clobberLate(PinnedRegisterInfo::get().toSave(MemoryMode::BoundsChecking)); |
1122 | patchpoint->setGenerator([unlinkedWasmToWasmCalls, functionIndex] (CCallHelpers& jit, const B3::StackmapGenerationParams&) { |
1123 | AllowMacroScratchRegisterUsage allowScratch(jit); |
1124 | CCallHelpers::Call call = jit.threadSafePatchableNearCall(); |
1125 | jit.addLinkTask([unlinkedWasmToWasmCalls, call, functionIndex] (LinkBuffer& linkBuffer) { |
1126 | unlinkedWasmToWasmCalls->append({ linkBuffer.locationOfNearCall<WasmEntryPtrTag>(call), functionIndex }); |
1127 | }); |
1128 | }); |
1129 | }); |
1130 | UpsilonValue* wasmCallResultUpsilon = returnType == Void ? nullptr : isWasmBlock->appendNew<UpsilonValue>(m_proc, origin(), wasmCallResult); |
1131 | isWasmBlock->appendNewControlValue(m_proc, Jump, origin(), continuation); |
1132 | |
1133 | // FIXME: Let's remove this indirection by creating a PIC friendly IC |
1134 | // for calls out to the embedder. This shouldn't be that hard to do. We could probably |
1135 | // implement the IC to be over Context*. |
1136 | // https://bugs.webkit.org/show_bug.cgi?id=170375 |
1137 | Value* jumpDestination = isEmbedderBlock->appendNew<MemoryValue>(m_proc, |
1138 | Load, pointerType(), origin(), instanceValue(), safeCast<int32_t>(Instance::offsetOfWasmToEmbedderStub(functionIndex))); |
1139 | |
1140 | Value* embedderCallResult = wasmCallingConvention().setupCall(m_proc, isEmbedderBlock, origin(), args, toB3Type(returnType), |
1141 | [=] (PatchpointValue* patchpoint) { |
1142 | patchpoint->effects.writesPinned = true; |
1143 | patchpoint->effects.readsPinned = true; |
1144 | patchpoint->append(jumpDestination, ValueRep::SomeRegister); |
1145 | // We need to clobber all potential pinned registers since we might be leaving the instance. |
1146 | // We pessimistically assume we could be calling to something that is bounds checking. |
1147 | // FIXME: We shouldn't have to do this: https://bugs.webkit.org/show_bug.cgi?id=172181 |
1148 | patchpoint->clobberLate(PinnedRegisterInfo::get().toSave(MemoryMode::BoundsChecking)); |
1149 | patchpoint->setGenerator([returnType] (CCallHelpers& jit, const B3::StackmapGenerationParams& params) { |
1150 | AllowMacroScratchRegisterUsage allowScratch(jit); |
1151 | jit.call(params[returnType == Void ? 0 : 1].gpr(), WasmEntryPtrTag); |
1152 | }); |
1153 | }); |
1154 | UpsilonValue* embedderCallResultUpsilon = returnType == Void ? nullptr : isEmbedderBlock->appendNew<UpsilonValue>(m_proc, origin(), embedderCallResult); |
1155 | isEmbedderBlock->appendNewControlValue(m_proc, Jump, origin(), continuation); |
1156 | |
1157 | m_currentBlock = continuation; |
1158 | |
1159 | if (returnType == Void) |
1160 | result = nullptr; |
1161 | else { |
1162 | result = continuation->appendNew<Value>(m_proc, Phi, toB3Type(returnType), origin()); |
1163 | wasmCallResultUpsilon->setPhi(result); |
1164 | embedderCallResultUpsilon->setPhi(result); |
1165 | } |
1166 | |
1167 | // The call could have been to another WebAssembly instance, and / or could have modified our Memory. |
1168 | restoreWebAssemblyGlobalState(RestoreCachedStackLimit::Yes, m_info.memory, instanceValue(), m_proc, continuation); |
1169 | } else { |
1170 | result = wasmCallingConvention().setupCall(m_proc, m_currentBlock, origin(), args, toB3Type(returnType), |
1171 | [=] (PatchpointValue* patchpoint) { |
1172 | patchpoint->effects.writesPinned = true; |
1173 | patchpoint->effects.readsPinned = true; |
1174 | |
1175 | patchpoint->setGenerator([unlinkedWasmToWasmCalls, functionIndex] (CCallHelpers& jit, const B3::StackmapGenerationParams&) { |
1176 | AllowMacroScratchRegisterUsage allowScratch(jit); |
1177 | CCallHelpers::Call call = jit.threadSafePatchableNearCall(); |
1178 | jit.addLinkTask([unlinkedWasmToWasmCalls, call, functionIndex] (LinkBuffer& linkBuffer) { |
1179 | unlinkedWasmToWasmCalls->append({ linkBuffer.locationOfNearCall<WasmEntryPtrTag>(call), functionIndex }); |
1180 | }); |
1181 | }); |
1182 | }); |
1183 | } |
1184 | |
1185 | return { }; |
1186 | } |
1187 | |
1188 | auto B3IRGenerator::addCallIndirect(const Signature& signature, Vector<ExpressionType>& args, ExpressionType& result) -> PartialResult |
1189 | { |
1190 | ExpressionType calleeIndex = args.takeLast(); |
1191 | ASSERT(signature.argumentCount() == args.size()); |
1192 | |
1193 | m_makesCalls = true; |
1194 | // Note: call indirect can call either WebAssemblyFunction or WebAssemblyWrapperFunction. Because |
1195 | // WebAssemblyWrapperFunction is like calling into the embedder, we conservatively assume all call indirects |
1196 | // can be to the embedder for our stack check calculation. |
1197 | m_maxNumJSCallArguments = std::max(m_maxNumJSCallArguments, static_cast<uint32_t>(args.size())); |
1198 | |
1199 | ExpressionType callableFunctionBuffer; |
1200 | ExpressionType instancesBuffer; |
1201 | ExpressionType callableFunctionBufferLength; |
1202 | ExpressionType mask; |
1203 | { |
1204 | ExpressionType table = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, pointerType(), origin(), |
1205 | instanceValue(), safeCast<int32_t>(Instance::offsetOfTable())); |
1206 | callableFunctionBuffer = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, pointerType(), origin(), |
1207 | table, safeCast<int32_t>(Table::offsetOfFunctions())); |
1208 | instancesBuffer = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, pointerType(), origin(), |
1209 | table, safeCast<int32_t>(Table::offsetOfInstances())); |
1210 | callableFunctionBufferLength = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, Int32, origin(), |
1211 | table, safeCast<int32_t>(Table::offsetOfLength())); |
1212 | mask = m_currentBlock->appendNew<Value>(m_proc, ZExt32, origin(), |
1213 | m_currentBlock->appendNew<MemoryValue>(m_proc, Load, Int32, origin(), |
1214 | table, safeCast<int32_t>(Table::offsetOfMask()))); |
1215 | } |
1216 | |
1217 | // Check the index we are looking for is valid. |
1218 | { |
1219 | CheckValue* check = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), |
1220 | m_currentBlock->appendNew<Value>(m_proc, AboveEqual, origin(), calleeIndex, callableFunctionBufferLength)); |
1221 | |
1222 | check->setGenerator([=] (CCallHelpers& jit, const B3::StackmapGenerationParams&) { |
1223 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsCallIndirect); |
1224 | }); |
1225 | } |
1226 | |
1227 | calleeIndex = m_currentBlock->appendNew<Value>(m_proc, ZExt32, origin(), calleeIndex); |
1228 | |
1229 | if (Options::enableSpectreMitigations()) |
1230 | calleeIndex = m_currentBlock->appendNew<Value>(m_proc, BitAnd, origin(), mask, calleeIndex); |
1231 | |
1232 | ExpressionType callableFunction; |
1233 | { |
1234 | // Compute the offset in the table index space we are looking for. |
1235 | ExpressionType offset = m_currentBlock->appendNew<Value>(m_proc, Mul, origin(), |
1236 | calleeIndex, constant(pointerType(), sizeof(WasmToWasmImportableFunction))); |
1237 | callableFunction = m_currentBlock->appendNew<Value>(m_proc, Add, origin(), callableFunctionBuffer, offset); |
1238 | |
1239 | // Check that the WasmToWasmImportableFunction is initialized. We trap if it isn't. An "invalid" SignatureIndex indicates it's not initialized. |
1240 | // FIXME: when we have trap handlers, we can just let the call fail because Signature::invalidIndex is 0. https://bugs.webkit.org/show_bug.cgi?id=177210 |
1241 | static_assert(sizeof(WasmToWasmImportableFunction::signatureIndex) == sizeof(uint64_t), "Load codegen assumes i64" ); |
1242 | ExpressionType calleeSignatureIndex = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, Int64, origin(), callableFunction, safeCast<int32_t>(WasmToWasmImportableFunction::offsetOfSignatureIndex())); |
1243 | { |
1244 | CheckValue* check = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), |
1245 | m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), |
1246 | calleeSignatureIndex, |
1247 | m_currentBlock->appendNew<Const64Value>(m_proc, origin(), Signature::invalidIndex))); |
1248 | |
1249 | check->setGenerator([=] (CCallHelpers& jit, const B3::StackmapGenerationParams&) { |
1250 | this->emitExceptionCheck(jit, ExceptionType::NullTableEntry); |
1251 | }); |
1252 | } |
1253 | |
1254 | // Check the signature matches the value we expect. |
1255 | { |
1256 | ExpressionType expectedSignatureIndex = m_currentBlock->appendNew<Const64Value>(m_proc, origin(), SignatureInformation::get(signature)); |
1257 | CheckValue* check = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), |
1258 | m_currentBlock->appendNew<Value>(m_proc, NotEqual, origin(), calleeSignatureIndex, expectedSignatureIndex)); |
1259 | |
1260 | check->setGenerator([=] (CCallHelpers& jit, const B3::StackmapGenerationParams&) { |
1261 | this->emitExceptionCheck(jit, ExceptionType::BadSignature); |
1262 | }); |
1263 | } |
1264 | } |
1265 | |
1266 | // Do a context switch if needed. |
1267 | { |
1268 | Value* offset = m_currentBlock->appendNew<Value>(m_proc, Mul, origin(), |
1269 | calleeIndex, constant(pointerType(), sizeof(Instance*))); |
1270 | Value* newContextInstance = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, pointerType(), origin(), |
1271 | m_currentBlock->appendNew<Value>(m_proc, Add, origin(), instancesBuffer, offset)); |
1272 | |
1273 | BasicBlock* continuation = m_proc.addBlock(); |
1274 | BasicBlock* doContextSwitch = m_proc.addBlock(); |
1275 | |
1276 | Value* isSameContextInstance = m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), |
1277 | newContextInstance, instanceValue()); |
1278 | m_currentBlock->appendNewControlValue(m_proc, B3::Branch, origin(), |
1279 | isSameContextInstance, FrequentedBlock(continuation), FrequentedBlock(doContextSwitch)); |
1280 | |
1281 | PatchpointValue* patchpoint = doContextSwitch->appendNew<PatchpointValue>(m_proc, B3::Void, origin()); |
1282 | patchpoint->effects.writesPinned = true; |
1283 | // We pessimistically assume we're calling something with BoundsChecking memory. |
1284 | // FIXME: We shouldn't have to do this: https://bugs.webkit.org/show_bug.cgi?id=172181 |
1285 | patchpoint->clobber(PinnedRegisterInfo::get().toSave(MemoryMode::BoundsChecking)); |
1286 | patchpoint->clobber(RegisterSet::macroScratchRegisters()); |
1287 | patchpoint->append(newContextInstance, ValueRep::SomeRegister); |
1288 | patchpoint->append(instanceValue(), ValueRep::SomeRegister); |
1289 | patchpoint->numGPScratchRegisters = Gigacage::isEnabled(Gigacage::Primitive) ? 1 : 0; |
1290 | |
1291 | patchpoint->setGenerator([=] (CCallHelpers& jit, const B3::StackmapGenerationParams& params) { |
1292 | AllowMacroScratchRegisterUsage allowScratch(jit); |
1293 | GPRReg newContextInstance = params[0].gpr(); |
1294 | GPRReg oldContextInstance = params[1].gpr(); |
1295 | const PinnedRegisterInfo& pinnedRegs = PinnedRegisterInfo::get(); |
1296 | GPRReg baseMemory = pinnedRegs.baseMemoryPointer; |
1297 | ASSERT(newContextInstance != baseMemory); |
1298 | jit.loadPtr(CCallHelpers::Address(oldContextInstance, Instance::offsetOfCachedStackLimit()), baseMemory); |
1299 | jit.storePtr(baseMemory, CCallHelpers::Address(newContextInstance, Instance::offsetOfCachedStackLimit())); |
1300 | jit.storeWasmContextInstance(newContextInstance); |
1301 | ASSERT(pinnedRegs.sizeRegister != baseMemory); |
1302 | // FIXME: We should support more than one memory size register |
1303 | // see: https://bugs.webkit.org/show_bug.cgi?id=162952 |
1304 | ASSERT(pinnedRegs.sizeRegister != newContextInstance); |
1305 | GPRReg scratchOrSize = Gigacage::isEnabled(Gigacage::Primitive) ? params.gpScratch(0) : pinnedRegs.sizeRegister; |
1306 | |
1307 | jit.loadPtr(CCallHelpers::Address(newContextInstance, Instance::offsetOfCachedMemorySize()), pinnedRegs.sizeRegister); // Memory size. |
1308 | jit.loadPtr(CCallHelpers::Address(newContextInstance, Instance::offsetOfCachedMemory()), baseMemory); // Memory::void*. |
1309 | |
1310 | jit.cageConditionally(Gigacage::Primitive, baseMemory, scratchOrSize); |
1311 | }); |
1312 | doContextSwitch->appendNewControlValue(m_proc, Jump, origin(), continuation); |
1313 | |
1314 | m_currentBlock = continuation; |
1315 | } |
1316 | |
1317 | ExpressionType calleeCode = m_currentBlock->appendNew<MemoryValue>(m_proc, Load, pointerType(), origin(), |
1318 | m_currentBlock->appendNew<MemoryValue>(m_proc, Load, pointerType(), origin(), callableFunction, |
1319 | safeCast<int32_t>(WasmToWasmImportableFunction::offsetOfEntrypointLoadLocation()))); |
1320 | |
1321 | Type returnType = signature.returnType(); |
1322 | result = wasmCallingConvention().setupCall(m_proc, m_currentBlock, origin(), args, toB3Type(returnType), |
1323 | [=] (PatchpointValue* patchpoint) { |
1324 | patchpoint->effects.writesPinned = true; |
1325 | patchpoint->effects.readsPinned = true; |
1326 | // We need to clobber all potential pinned registers since we might be leaving the instance. |
1327 | // We pessimistically assume we're always calling something that is bounds checking so |
1328 | // because the wasm->wasm thunk unconditionally overrides the size registers. |
1329 | // FIXME: We should not have to do this, but the wasm->wasm stub assumes it can |
1330 | // use all the pinned registers as scratch: https://bugs.webkit.org/show_bug.cgi?id=172181 |
1331 | patchpoint->clobberLate(PinnedRegisterInfo::get().toSave(MemoryMode::BoundsChecking)); |
1332 | |
1333 | patchpoint->append(calleeCode, ValueRep::SomeRegister); |
1334 | patchpoint->setGenerator([=] (CCallHelpers& jit, const B3::StackmapGenerationParams& params) { |
1335 | AllowMacroScratchRegisterUsage allowScratch(jit); |
1336 | jit.call(params[returnType == Void ? 0 : 1].gpr(), WasmEntryPtrTag); |
1337 | }); |
1338 | }); |
1339 | |
1340 | // The call could have been to another WebAssembly instance, and / or could have modified our Memory. |
1341 | restoreWebAssemblyGlobalState(RestoreCachedStackLimit::Yes, m_info.memory, instanceValue(), m_proc, m_currentBlock); |
1342 | |
1343 | return { }; |
1344 | } |
1345 | |
1346 | void B3IRGenerator::unify(const ExpressionType phi, const ExpressionType source) |
1347 | { |
1348 | m_currentBlock->appendNew<UpsilonValue>(m_proc, origin(), source, phi); |
1349 | } |
1350 | |
1351 | void B3IRGenerator::unifyValuesWithBlock(const ExpressionList& resultStack, const ResultList& result) |
1352 | { |
1353 | ASSERT(result.size() <= resultStack.size()); |
1354 | |
1355 | for (size_t i = 0; i < result.size(); ++i) |
1356 | unify(result[result.size() - 1 - i], resultStack[resultStack.size() - 1 - i]); |
1357 | } |
1358 | |
1359 | static void dumpExpressionStack(const CommaPrinter& comma, const B3IRGenerator::ExpressionList& expressionStack) |
1360 | { |
1361 | dataLog(comma, "ExpressionStack:" ); |
1362 | for (const auto& expression : expressionStack) |
1363 | dataLog(comma, *expression); |
1364 | } |
1365 | |
1366 | void B3IRGenerator::dump(const Vector<ControlEntry>& controlStack, const ExpressionList* expressionStack) |
1367 | { |
1368 | dataLogLn("Constants:" ); |
1369 | for (const auto& constant : m_constantPool) |
1370 | dataLogLn(deepDump(m_proc, constant.value)); |
1371 | |
1372 | dataLogLn("Processing Graph:" ); |
1373 | dataLog(m_proc); |
1374 | dataLogLn("With current block:" , *m_currentBlock); |
1375 | dataLogLn("Control stack:" ); |
1376 | ASSERT(controlStack.size()); |
1377 | for (size_t i = controlStack.size(); i--;) { |
1378 | dataLog(" " , controlStack[i].controlData, ": " ); |
1379 | CommaPrinter comma(", " , "" ); |
1380 | dumpExpressionStack(comma, *expressionStack); |
1381 | expressionStack = &controlStack[i].enclosedExpressionStack; |
1382 | dataLogLn(); |
1383 | } |
1384 | dataLogLn(); |
1385 | } |
1386 | |
1387 | auto B3IRGenerator::origin() -> Origin |
1388 | { |
1389 | OpcodeOrigin origin(m_parser->currentOpcode(), m_parser->currentOpcodeStartingOffset()); |
1390 | ASSERT(isValidOpType(static_cast<uint8_t>(origin.opcode()))); |
1391 | return bitwise_cast<Origin>(origin); |
1392 | } |
1393 | |
1394 | Expected<std::unique_ptr<InternalFunction>, String> parseAndCompile(CompilationContext& compilationContext, const uint8_t* functionStart, size_t functionLength, const Signature& signature, Vector<UnlinkedWasmToWasmCall>& unlinkedWasmToWasmCalls, const ModuleInformation& info, MemoryMode mode, CompilationMode compilationMode, uint32_t functionIndex, TierUpCount* tierUp, ThrowWasmException throwWasmException) |
1395 | { |
1396 | auto result = std::make_unique<InternalFunction>(); |
1397 | |
1398 | compilationContext.embedderEntrypointJIT = std::make_unique<CCallHelpers>(); |
1399 | compilationContext.wasmEntrypointJIT = std::make_unique<CCallHelpers>(); |
1400 | |
1401 | Procedure procedure; |
1402 | |
1403 | procedure.setOriginPrinter([] (PrintStream& out, Origin origin) { |
1404 | if (origin.data()) |
1405 | out.print("Wasm: " , bitwise_cast<OpcodeOrigin>(origin)); |
1406 | }); |
1407 | |
1408 | // This means we cannot use either StackmapGenerationParams::usedRegisters() or |
1409 | // StackmapGenerationParams::unavailableRegisters(). In exchange for this concession, we |
1410 | // don't strictly need to run Air::reportUsedRegisters(), which saves a bit of CPU time at |
1411 | // optLevel=1. |
1412 | procedure.setNeedsUsedRegisters(false); |
1413 | |
1414 | procedure.setOptLevel(compilationMode == CompilationMode::BBQMode |
1415 | ? Options::webAssemblyBBQOptimizationLevel() |
1416 | : Options::webAssemblyOMGOptimizationLevel()); |
1417 | |
1418 | B3IRGenerator irGenerator(info, procedure, result.get(), unlinkedWasmToWasmCalls, mode, compilationMode, functionIndex, tierUp, throwWasmException); |
1419 | FunctionParser<B3IRGenerator> parser(irGenerator, functionStart, functionLength, signature, info); |
1420 | WASM_FAIL_IF_HELPER_FAILS(parser.parse()); |
1421 | |
1422 | irGenerator.insertConstants(); |
1423 | |
1424 | procedure.resetReachability(); |
1425 | if (!ASSERT_DISABLED) |
1426 | validate(procedure, "After parsing:\n" ); |
1427 | |
1428 | dataLogIf(WasmB3IRGeneratorInternal::verbose, "Pre SSA: " , procedure); |
1429 | fixSSA(procedure); |
1430 | dataLogIf(WasmB3IRGeneratorInternal::verbose, "Post SSA: " , procedure); |
1431 | |
1432 | { |
1433 | B3::prepareForGeneration(procedure); |
1434 | B3::generate(procedure, *compilationContext.wasmEntrypointJIT); |
1435 | compilationContext.wasmEntrypointByproducts = procedure.releaseByproducts(); |
1436 | result->entrypoint.calleeSaveRegisters = procedure.calleeSaveRegisterAtOffsetList(); |
1437 | } |
1438 | |
1439 | return result; |
1440 | } |
1441 | |
1442 | // Custom wasm ops. These are the ones too messy to do in wasm.json. |
1443 | |
1444 | void B3IRGenerator::emitChecksForModOrDiv(B3::Opcode operation, ExpressionType left, ExpressionType right) |
1445 | { |
1446 | ASSERT(operation == Div || operation == Mod || operation == UDiv || operation == UMod); |
1447 | const B3::Type type = left->type(); |
1448 | |
1449 | { |
1450 | CheckValue* check = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), |
1451 | m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), right, constant(type, 0))); |
1452 | |
1453 | check->setGenerator([=] (CCallHelpers& jit, const B3::StackmapGenerationParams&) { |
1454 | this->emitExceptionCheck(jit, ExceptionType::DivisionByZero); |
1455 | }); |
1456 | } |
1457 | |
1458 | if (operation == Div) { |
1459 | int64_t min = type == Int32 ? std::numeric_limits<int32_t>::min() : std::numeric_limits<int64_t>::min(); |
1460 | |
1461 | CheckValue* check = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), |
1462 | m_currentBlock->appendNew<Value>(m_proc, BitAnd, origin(), |
1463 | m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), left, constant(type, min)), |
1464 | m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), right, constant(type, -1)))); |
1465 | |
1466 | check->setGenerator([=] (CCallHelpers& jit, const B3::StackmapGenerationParams&) { |
1467 | this->emitExceptionCheck(jit, ExceptionType::IntegerOverflow); |
1468 | }); |
1469 | } |
1470 | } |
1471 | |
1472 | template<> |
1473 | auto B3IRGenerator::addOp<OpType::I32DivS>(ExpressionType left, ExpressionType right, ExpressionType& result) -> PartialResult |
1474 | { |
1475 | const B3::Opcode op = Div; |
1476 | emitChecksForModOrDiv(op, left, right); |
1477 | result = m_currentBlock->appendNew<Value>(m_proc, op, origin(), left, right); |
1478 | return { }; |
1479 | } |
1480 | |
1481 | template<> |
1482 | auto B3IRGenerator::addOp<OpType::I32RemS>(ExpressionType left, ExpressionType right, ExpressionType& result) -> PartialResult |
1483 | { |
1484 | const B3::Opcode op = Mod; |
1485 | emitChecksForModOrDiv(op, left, right); |
1486 | result = m_currentBlock->appendNew<Value>(m_proc, chill(op), origin(), left, right); |
1487 | return { }; |
1488 | } |
1489 | |
1490 | template<> |
1491 | auto B3IRGenerator::addOp<OpType::I32DivU>(ExpressionType left, ExpressionType right, ExpressionType& result) -> PartialResult |
1492 | { |
1493 | const B3::Opcode op = UDiv; |
1494 | emitChecksForModOrDiv(op, left, right); |
1495 | result = m_currentBlock->appendNew<Value>(m_proc, op, origin(), left, right); |
1496 | return { }; |
1497 | } |
1498 | |
1499 | template<> |
1500 | auto B3IRGenerator::addOp<OpType::I32RemU>(ExpressionType left, ExpressionType right, ExpressionType& result) -> PartialResult |
1501 | { |
1502 | const B3::Opcode op = UMod; |
1503 | emitChecksForModOrDiv(op, left, right); |
1504 | result = m_currentBlock->appendNew<Value>(m_proc, op, origin(), left, right); |
1505 | return { }; |
1506 | } |
1507 | |
1508 | template<> |
1509 | auto B3IRGenerator::addOp<OpType::I64DivS>(ExpressionType left, ExpressionType right, ExpressionType& result) -> PartialResult |
1510 | { |
1511 | const B3::Opcode op = Div; |
1512 | emitChecksForModOrDiv(op, left, right); |
1513 | result = m_currentBlock->appendNew<Value>(m_proc, op, origin(), left, right); |
1514 | return { }; |
1515 | } |
1516 | |
1517 | template<> |
1518 | auto B3IRGenerator::addOp<OpType::I64RemS>(ExpressionType left, ExpressionType right, ExpressionType& result) -> PartialResult |
1519 | { |
1520 | const B3::Opcode op = Mod; |
1521 | emitChecksForModOrDiv(op, left, right); |
1522 | result = m_currentBlock->appendNew<Value>(m_proc, chill(op), origin(), left, right); |
1523 | return { }; |
1524 | } |
1525 | |
1526 | template<> |
1527 | auto B3IRGenerator::addOp<OpType::I64DivU>(ExpressionType left, ExpressionType right, ExpressionType& result) -> PartialResult |
1528 | { |
1529 | const B3::Opcode op = UDiv; |
1530 | emitChecksForModOrDiv(op, left, right); |
1531 | result = m_currentBlock->appendNew<Value>(m_proc, op, origin(), left, right); |
1532 | return { }; |
1533 | } |
1534 | |
1535 | template<> |
1536 | auto B3IRGenerator::addOp<OpType::I64RemU>(ExpressionType left, ExpressionType right, ExpressionType& result) -> PartialResult |
1537 | { |
1538 | const B3::Opcode op = UMod; |
1539 | emitChecksForModOrDiv(op, left, right); |
1540 | result = m_currentBlock->appendNew<Value>(m_proc, op, origin(), left, right); |
1541 | return { }; |
1542 | } |
1543 | |
1544 | template<> |
1545 | auto B3IRGenerator::addOp<OpType::I32Ctz>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1546 | { |
1547 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int32, origin()); |
1548 | patchpoint->append(arg, ValueRep::SomeRegister); |
1549 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1550 | jit.countTrailingZeros32(params[1].gpr(), params[0].gpr()); |
1551 | }); |
1552 | patchpoint->effects = Effects::none(); |
1553 | result = patchpoint; |
1554 | return { }; |
1555 | } |
1556 | |
1557 | template<> |
1558 | auto B3IRGenerator::addOp<OpType::I64Ctz>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1559 | { |
1560 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int64, origin()); |
1561 | patchpoint->append(arg, ValueRep::SomeRegister); |
1562 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1563 | jit.countTrailingZeros64(params[1].gpr(), params[0].gpr()); |
1564 | }); |
1565 | patchpoint->effects = Effects::none(); |
1566 | result = patchpoint; |
1567 | return { }; |
1568 | } |
1569 | |
1570 | template<> |
1571 | auto B3IRGenerator::addOp<OpType::I32Popcnt>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1572 | { |
1573 | #if CPU(X86_64) |
1574 | if (MacroAssembler::supportsCountPopulation()) { |
1575 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int32, origin()); |
1576 | patchpoint->append(arg, ValueRep::SomeRegister); |
1577 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1578 | jit.countPopulation32(params[1].gpr(), params[0].gpr()); |
1579 | }); |
1580 | patchpoint->effects = Effects::none(); |
1581 | result = patchpoint; |
1582 | return { }; |
1583 | } |
1584 | #endif |
1585 | |
1586 | uint32_t (*popcount)(int32_t) = [] (int32_t value) -> uint32_t { return __builtin_popcount(value); }; |
1587 | Value* funcAddress = m_currentBlock->appendNew<ConstPtrValue>(m_proc, origin(), tagCFunctionPtr<void*>(popcount, B3CCallPtrTag)); |
1588 | result = m_currentBlock->appendNew<CCallValue>(m_proc, Int32, origin(), Effects::none(), funcAddress, arg); |
1589 | return { }; |
1590 | } |
1591 | |
1592 | template<> |
1593 | auto B3IRGenerator::addOp<OpType::I64Popcnt>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1594 | { |
1595 | #if CPU(X86_64) |
1596 | if (MacroAssembler::supportsCountPopulation()) { |
1597 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int64, origin()); |
1598 | patchpoint->append(arg, ValueRep::SomeRegister); |
1599 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1600 | jit.countPopulation64(params[1].gpr(), params[0].gpr()); |
1601 | }); |
1602 | patchpoint->effects = Effects::none(); |
1603 | result = patchpoint; |
1604 | return { }; |
1605 | } |
1606 | #endif |
1607 | |
1608 | uint64_t (*popcount)(int64_t) = [] (int64_t value) -> uint64_t { return __builtin_popcountll(value); }; |
1609 | Value* funcAddress = m_currentBlock->appendNew<ConstPtrValue>(m_proc, origin(), tagCFunctionPtr<void*>(popcount, B3CCallPtrTag)); |
1610 | result = m_currentBlock->appendNew<CCallValue>(m_proc, Int64, origin(), Effects::none(), funcAddress, arg); |
1611 | return { }; |
1612 | } |
1613 | |
1614 | template<> |
1615 | auto B3IRGenerator::addOp<F64ConvertUI64>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1616 | { |
1617 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Double, origin()); |
1618 | if (isX86()) |
1619 | patchpoint->numGPScratchRegisters = 1; |
1620 | patchpoint->clobber(RegisterSet::macroScratchRegisters()); |
1621 | patchpoint->append(ConstrainedValue(arg, ValueRep::SomeRegister)); |
1622 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1623 | AllowMacroScratchRegisterUsage allowScratch(jit); |
1624 | #if CPU(X86_64) |
1625 | jit.convertUInt64ToDouble(params[1].gpr(), params[0].fpr(), params.gpScratch(0)); |
1626 | #else |
1627 | jit.convertUInt64ToDouble(params[1].gpr(), params[0].fpr()); |
1628 | #endif |
1629 | }); |
1630 | patchpoint->effects = Effects::none(); |
1631 | result = patchpoint; |
1632 | return { }; |
1633 | } |
1634 | |
1635 | template<> |
1636 | auto B3IRGenerator::addOp<OpType::F32ConvertUI64>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1637 | { |
1638 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Float, origin()); |
1639 | if (isX86()) |
1640 | patchpoint->numGPScratchRegisters = 1; |
1641 | patchpoint->clobber(RegisterSet::macroScratchRegisters()); |
1642 | patchpoint->append(ConstrainedValue(arg, ValueRep::SomeRegister)); |
1643 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1644 | AllowMacroScratchRegisterUsage allowScratch(jit); |
1645 | #if CPU(X86_64) |
1646 | jit.convertUInt64ToFloat(params[1].gpr(), params[0].fpr(), params.gpScratch(0)); |
1647 | #else |
1648 | jit.convertUInt64ToFloat(params[1].gpr(), params[0].fpr()); |
1649 | #endif |
1650 | }); |
1651 | patchpoint->effects = Effects::none(); |
1652 | result = patchpoint; |
1653 | return { }; |
1654 | } |
1655 | |
1656 | template<> |
1657 | auto B3IRGenerator::addOp<OpType::F64Nearest>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1658 | { |
1659 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Double, origin()); |
1660 | patchpoint->append(arg, ValueRep::SomeRegister); |
1661 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1662 | jit.roundTowardNearestIntDouble(params[1].fpr(), params[0].fpr()); |
1663 | }); |
1664 | patchpoint->effects = Effects::none(); |
1665 | result = patchpoint; |
1666 | return { }; |
1667 | } |
1668 | |
1669 | template<> |
1670 | auto B3IRGenerator::addOp<OpType::F32Nearest>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1671 | { |
1672 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Float, origin()); |
1673 | patchpoint->append(arg, ValueRep::SomeRegister); |
1674 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1675 | jit.roundTowardNearestIntFloat(params[1].fpr(), params[0].fpr()); |
1676 | }); |
1677 | patchpoint->effects = Effects::none(); |
1678 | result = patchpoint; |
1679 | return { }; |
1680 | } |
1681 | |
1682 | template<> |
1683 | auto B3IRGenerator::addOp<OpType::F64Trunc>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1684 | { |
1685 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Double, origin()); |
1686 | patchpoint->append(arg, ValueRep::SomeRegister); |
1687 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1688 | jit.roundTowardZeroDouble(params[1].fpr(), params[0].fpr()); |
1689 | }); |
1690 | patchpoint->effects = Effects::none(); |
1691 | result = patchpoint; |
1692 | return { }; |
1693 | } |
1694 | |
1695 | template<> |
1696 | auto B3IRGenerator::addOp<OpType::F32Trunc>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1697 | { |
1698 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Float, origin()); |
1699 | patchpoint->append(arg, ValueRep::SomeRegister); |
1700 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1701 | jit.roundTowardZeroFloat(params[1].fpr(), params[0].fpr()); |
1702 | }); |
1703 | patchpoint->effects = Effects::none(); |
1704 | result = patchpoint; |
1705 | return { }; |
1706 | } |
1707 | |
1708 | template<> |
1709 | auto B3IRGenerator::addOp<OpType::I32TruncSF64>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1710 | { |
1711 | Value* max = constant(Double, bitwise_cast<uint64_t>(-static_cast<double>(std::numeric_limits<int32_t>::min()))); |
1712 | Value* min = constant(Double, bitwise_cast<uint64_t>(static_cast<double>(std::numeric_limits<int32_t>::min()))); |
1713 | Value* outOfBounds = m_currentBlock->appendNew<Value>(m_proc, BitAnd, origin(), |
1714 | m_currentBlock->appendNew<Value>(m_proc, LessThan, origin(), arg, max), |
1715 | m_currentBlock->appendNew<Value>(m_proc, GreaterEqual, origin(), arg, min)); |
1716 | outOfBounds = m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), outOfBounds, constant(Int32, 0)); |
1717 | CheckValue* trap = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), outOfBounds); |
1718 | trap->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams&) { |
1719 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsTrunc); |
1720 | }); |
1721 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int32, origin()); |
1722 | patchpoint->append(arg, ValueRep::SomeRegister); |
1723 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1724 | jit.truncateDoubleToInt32(params[1].fpr(), params[0].gpr()); |
1725 | }); |
1726 | patchpoint->effects = Effects::none(); |
1727 | result = patchpoint; |
1728 | return { }; |
1729 | } |
1730 | |
1731 | template<> |
1732 | auto B3IRGenerator::addOp<OpType::I32TruncSF32>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1733 | { |
1734 | Value* max = constant(Float, bitwise_cast<uint32_t>(-static_cast<float>(std::numeric_limits<int32_t>::min()))); |
1735 | Value* min = constant(Float, bitwise_cast<uint32_t>(static_cast<float>(std::numeric_limits<int32_t>::min()))); |
1736 | Value* outOfBounds = m_currentBlock->appendNew<Value>(m_proc, BitAnd, origin(), |
1737 | m_currentBlock->appendNew<Value>(m_proc, LessThan, origin(), arg, max), |
1738 | m_currentBlock->appendNew<Value>(m_proc, GreaterEqual, origin(), arg, min)); |
1739 | outOfBounds = m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), outOfBounds, constant(Int32, 0)); |
1740 | CheckValue* trap = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), outOfBounds); |
1741 | trap->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams&) { |
1742 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsTrunc); |
1743 | }); |
1744 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int32, origin()); |
1745 | patchpoint->append(arg, ValueRep::SomeRegister); |
1746 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1747 | jit.truncateFloatToInt32(params[1].fpr(), params[0].gpr()); |
1748 | }); |
1749 | patchpoint->effects = Effects::none(); |
1750 | result = patchpoint; |
1751 | return { }; |
1752 | } |
1753 | |
1754 | |
1755 | template<> |
1756 | auto B3IRGenerator::addOp<OpType::I32TruncUF64>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1757 | { |
1758 | Value* max = constant(Double, bitwise_cast<uint64_t>(static_cast<double>(std::numeric_limits<int32_t>::min()) * -2.0)); |
1759 | Value* min = constant(Double, bitwise_cast<uint64_t>(-1.0)); |
1760 | Value* outOfBounds = m_currentBlock->appendNew<Value>(m_proc, BitAnd, origin(), |
1761 | m_currentBlock->appendNew<Value>(m_proc, LessThan, origin(), arg, max), |
1762 | m_currentBlock->appendNew<Value>(m_proc, GreaterThan, origin(), arg, min)); |
1763 | outOfBounds = m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), outOfBounds, constant(Int32, 0)); |
1764 | CheckValue* trap = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), outOfBounds); |
1765 | trap->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams&) { |
1766 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsTrunc); |
1767 | }); |
1768 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int32, origin()); |
1769 | patchpoint->append(arg, ValueRep::SomeRegister); |
1770 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1771 | jit.truncateDoubleToUint32(params[1].fpr(), params[0].gpr()); |
1772 | }); |
1773 | patchpoint->effects = Effects::none(); |
1774 | result = patchpoint; |
1775 | return { }; |
1776 | } |
1777 | |
1778 | template<> |
1779 | auto B3IRGenerator::addOp<OpType::I32TruncUF32>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1780 | { |
1781 | Value* max = constant(Float, bitwise_cast<uint32_t>(static_cast<float>(std::numeric_limits<int32_t>::min()) * static_cast<float>(-2.0))); |
1782 | Value* min = constant(Float, bitwise_cast<uint32_t>(static_cast<float>(-1.0))); |
1783 | Value* outOfBounds = m_currentBlock->appendNew<Value>(m_proc, BitAnd, origin(), |
1784 | m_currentBlock->appendNew<Value>(m_proc, LessThan, origin(), arg, max), |
1785 | m_currentBlock->appendNew<Value>(m_proc, GreaterThan, origin(), arg, min)); |
1786 | outOfBounds = m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), outOfBounds, constant(Int32, 0)); |
1787 | CheckValue* trap = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), outOfBounds); |
1788 | trap->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams&) { |
1789 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsTrunc); |
1790 | }); |
1791 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int32, origin()); |
1792 | patchpoint->append(arg, ValueRep::SomeRegister); |
1793 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1794 | jit.truncateFloatToUint32(params[1].fpr(), params[0].gpr()); |
1795 | }); |
1796 | patchpoint->effects = Effects::none(); |
1797 | result = patchpoint; |
1798 | return { }; |
1799 | } |
1800 | |
1801 | template<> |
1802 | auto B3IRGenerator::addOp<OpType::I64TruncSF64>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1803 | { |
1804 | Value* max = constant(Double, bitwise_cast<uint64_t>(-static_cast<double>(std::numeric_limits<int64_t>::min()))); |
1805 | Value* min = constant(Double, bitwise_cast<uint64_t>(static_cast<double>(std::numeric_limits<int64_t>::min()))); |
1806 | Value* outOfBounds = m_currentBlock->appendNew<Value>(m_proc, BitAnd, origin(), |
1807 | m_currentBlock->appendNew<Value>(m_proc, LessThan, origin(), arg, max), |
1808 | m_currentBlock->appendNew<Value>(m_proc, GreaterEqual, origin(), arg, min)); |
1809 | outOfBounds = m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), outOfBounds, constant(Int32, 0)); |
1810 | CheckValue* trap = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), outOfBounds); |
1811 | trap->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams&) { |
1812 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsTrunc); |
1813 | }); |
1814 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int64, origin()); |
1815 | patchpoint->append(arg, ValueRep::SomeRegister); |
1816 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1817 | jit.truncateDoubleToInt64(params[1].fpr(), params[0].gpr()); |
1818 | }); |
1819 | patchpoint->effects = Effects::none(); |
1820 | result = patchpoint; |
1821 | return { }; |
1822 | } |
1823 | |
1824 | template<> |
1825 | auto B3IRGenerator::addOp<OpType::I64TruncUF64>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1826 | { |
1827 | Value* max = constant(Double, bitwise_cast<uint64_t>(static_cast<double>(std::numeric_limits<int64_t>::min()) * -2.0)); |
1828 | Value* min = constant(Double, bitwise_cast<uint64_t>(-1.0)); |
1829 | Value* outOfBounds = m_currentBlock->appendNew<Value>(m_proc, BitAnd, origin(), |
1830 | m_currentBlock->appendNew<Value>(m_proc, LessThan, origin(), arg, max), |
1831 | m_currentBlock->appendNew<Value>(m_proc, GreaterThan, origin(), arg, min)); |
1832 | outOfBounds = m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), outOfBounds, constant(Int32, 0)); |
1833 | CheckValue* trap = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), outOfBounds); |
1834 | trap->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams&) { |
1835 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsTrunc); |
1836 | }); |
1837 | |
1838 | Value* signBitConstant; |
1839 | if (isX86()) { |
1840 | // Since x86 doesn't have an instruction to convert floating points to unsigned integers, we at least try to do the smart thing if |
1841 | // the numbers are would be positive anyway as a signed integer. Since we cannot materialize constants into fprs we have b3 do it |
1842 | // so we can pool them if needed. |
1843 | signBitConstant = constant(Double, bitwise_cast<uint64_t>(static_cast<double>(std::numeric_limits<uint64_t>::max() - std::numeric_limits<int64_t>::max()))); |
1844 | } |
1845 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int64, origin()); |
1846 | patchpoint->append(arg, ValueRep::SomeRegister); |
1847 | if (isX86()) { |
1848 | patchpoint->append(signBitConstant, ValueRep::SomeRegister); |
1849 | patchpoint->numFPScratchRegisters = 1; |
1850 | } |
1851 | patchpoint->clobber(RegisterSet::macroScratchRegisters()); |
1852 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1853 | AllowMacroScratchRegisterUsage allowScratch(jit); |
1854 | FPRReg scratch = InvalidFPRReg; |
1855 | FPRReg constant = InvalidFPRReg; |
1856 | if (isX86()) { |
1857 | scratch = params.fpScratch(0); |
1858 | constant = params[2].fpr(); |
1859 | } |
1860 | jit.truncateDoubleToUint64(params[1].fpr(), params[0].gpr(), scratch, constant); |
1861 | }); |
1862 | patchpoint->effects = Effects::none(); |
1863 | result = patchpoint; |
1864 | return { }; |
1865 | } |
1866 | |
1867 | template<> |
1868 | auto B3IRGenerator::addOp<OpType::I64TruncSF32>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1869 | { |
1870 | Value* max = constant(Float, bitwise_cast<uint32_t>(-static_cast<float>(std::numeric_limits<int64_t>::min()))); |
1871 | Value* min = constant(Float, bitwise_cast<uint32_t>(static_cast<float>(std::numeric_limits<int64_t>::min()))); |
1872 | Value* outOfBounds = m_currentBlock->appendNew<Value>(m_proc, BitAnd, origin(), |
1873 | m_currentBlock->appendNew<Value>(m_proc, LessThan, origin(), arg, max), |
1874 | m_currentBlock->appendNew<Value>(m_proc, GreaterEqual, origin(), arg, min)); |
1875 | outOfBounds = m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), outOfBounds, constant(Int32, 0)); |
1876 | CheckValue* trap = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), outOfBounds); |
1877 | trap->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams&) { |
1878 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsTrunc); |
1879 | }); |
1880 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int64, origin()); |
1881 | patchpoint->append(arg, ValueRep::SomeRegister); |
1882 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1883 | jit.truncateFloatToInt64(params[1].fpr(), params[0].gpr()); |
1884 | }); |
1885 | patchpoint->effects = Effects::none(); |
1886 | result = patchpoint; |
1887 | return { }; |
1888 | } |
1889 | |
1890 | template<> |
1891 | auto B3IRGenerator::addOp<OpType::I64TruncUF32>(ExpressionType arg, ExpressionType& result) -> PartialResult |
1892 | { |
1893 | Value* max = constant(Float, bitwise_cast<uint32_t>(static_cast<float>(std::numeric_limits<int64_t>::min()) * static_cast<float>(-2.0))); |
1894 | Value* min = constant(Float, bitwise_cast<uint32_t>(static_cast<float>(-1.0))); |
1895 | Value* outOfBounds = m_currentBlock->appendNew<Value>(m_proc, BitAnd, origin(), |
1896 | m_currentBlock->appendNew<Value>(m_proc, LessThan, origin(), arg, max), |
1897 | m_currentBlock->appendNew<Value>(m_proc, GreaterThan, origin(), arg, min)); |
1898 | outOfBounds = m_currentBlock->appendNew<Value>(m_proc, Equal, origin(), outOfBounds, constant(Int32, 0)); |
1899 | CheckValue* trap = m_currentBlock->appendNew<CheckValue>(m_proc, Check, origin(), outOfBounds); |
1900 | trap->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams&) { |
1901 | this->emitExceptionCheck(jit, ExceptionType::OutOfBoundsTrunc); |
1902 | }); |
1903 | |
1904 | Value* signBitConstant; |
1905 | if (isX86()) { |
1906 | // Since x86 doesn't have an instruction to convert floating points to unsigned integers, we at least try to do the smart thing if |
1907 | // the numbers would be positive anyway as a signed integer. Since we cannot materialize constants into fprs we have b3 do it |
1908 | // so we can pool them if needed. |
1909 | signBitConstant = constant(Float, bitwise_cast<uint32_t>(static_cast<float>(std::numeric_limits<uint64_t>::max() - std::numeric_limits<int64_t>::max()))); |
1910 | } |
1911 | PatchpointValue* patchpoint = m_currentBlock->appendNew<PatchpointValue>(m_proc, Int64, origin()); |
1912 | patchpoint->append(arg, ValueRep::SomeRegister); |
1913 | if (isX86()) { |
1914 | patchpoint->append(signBitConstant, ValueRep::SomeRegister); |
1915 | patchpoint->numFPScratchRegisters = 1; |
1916 | } |
1917 | patchpoint->clobber(RegisterSet::macroScratchRegisters()); |
1918 | patchpoint->setGenerator([=] (CCallHelpers& jit, const StackmapGenerationParams& params) { |
1919 | AllowMacroScratchRegisterUsage allowScratch(jit); |
1920 | FPRReg scratch = InvalidFPRReg; |
1921 | FPRReg constant = InvalidFPRReg; |
1922 | if (isX86()) { |
1923 | scratch = params.fpScratch(0); |
1924 | constant = params[2].fpr(); |
1925 | } |
1926 | jit.truncateFloatToUint64(params[1].fpr(), params[0].gpr(), scratch, constant); |
1927 | }); |
1928 | patchpoint->effects = Effects::none(); |
1929 | result = patchpoint; |
1930 | return { }; |
1931 | } |
1932 | |
1933 | } } // namespace JSC::Wasm |
1934 | |
1935 | #include "WasmB3IRGeneratorInlines.h" |
1936 | |
1937 | #endif // ENABLE(WEBASSEMBLY) |
1938 | |