v8/test/fuzzer/wasm-fuzzer-common.h
Clemens Backes 918e478354 [wasm][fuzzer] Reduce number of instructions in simulator
It turned out that executing 1 million (wasm) instructions can take way
over a minute on a simulator.
Thus changing this back to 16k instructions (as it was before
https://crrev.com/c/4078983). This takes around 5 seconds on my machine.

R=jkummerow@chromium.org

Bug: chromium:1396351, chromium:1396237
Change-Id: Ibf2ae2c81816be93807e564876b75aec18a57c42
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/4079195
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
Cr-Commit-Position: refs/heads/main@{#84691}
2022-12-06 16:14:23 +00:00

60 lines
2.2 KiB
C++

// Copyright 2016 the V8 project authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#ifndef WASM_FUZZER_COMMON_H_
#define WASM_FUZZER_COMMON_H_
#include <stddef.h>
#include <stdint.h>
#include <memory>
#include "src/wasm/module-decoder.h"
#include "src/wasm/wasm-module-builder.h"
namespace v8::internal::wasm::fuzzer {
// A default value for {max_executed_instructions} in {ExecuteAgainstReference}.
#ifdef USE_SIMULATOR
constexpr int kDefaultMaxFuzzerExecutedInstructions = 16'000;
#else
constexpr int kDefaultMaxFuzzerExecutedInstructions = 1'000'000;
#endif
// First creates a reference module fully compiled with Liftoff, with
// instrumentation to stop after a given number of steps and to record any
// nondeterminism while executing. If execution finishes within {max_steps},
// {module_object} is instantiated, its "main" function is executed, and the
// result is compared against the reference execution. If non-determinism was
// detected during the reference execution, the result is allowed to differ.
void ExecuteAgainstReference(Isolate* isolate,
Handle<WasmModuleObject> module_object,
int32_t max_executed_instructions);
void GenerateTestCase(Isolate* isolate, ModuleWireBytes wire_bytes,
bool compiles);
// On the first call, enables all staged wasm features and experimental features
// that are ready for fuzzing. All subsequent calls are no-ops. This avoids race
// conditions with threads reading the flags. Fuzzers are executed in their own
// process anyway, so this should not interfere with anything.
void EnableExperimentalWasmFeatures(v8::Isolate* isolate);
class WasmExecutionFuzzer {
public:
virtual ~WasmExecutionFuzzer() = default;
void FuzzWasmModule(base::Vector<const uint8_t> data,
bool require_valid = false);
virtual size_t max_input_size() const { return 512; }
protected:
virtual bool GenerateModule(Isolate* isolate, Zone* zone,
base::Vector<const uint8_t> data,
ZoneBuffer* buffer) = 0;
};
} // namespace v8::internal::wasm::fuzzer
#endif // WASM_FUZZER_COMMON_H_