When the same WebAssembly module is fetched again from the same URL then the cached code can be used immediately without additional compilation. If the WebAssembly module was compiled with pileStreaming then the TurboFan-generated machine code will also get cached. This means that if TurboFan code becomes available after the function was called, the function call will complete its execution with Liftoff code. Note though that we don’t do on-stack-replacement.
Any new calls to that function will then use the new, optimized code produced by TurboFan, not the Liftoff code. Once compilation is finished, the new code gets registered with the WebAssembly module, replacing the existing Liftoff code. Once a function reaches a certain threshold, the function is considered hot, and re-compilation gets triggered on a background thread. V8 monitors how often WebAssembly functions get called. These additional internal representations allow optimizations and better register allocations, resulting in significantly faster code. TurboFan is a multi-pass compiler, which means that it builds multiple internal representations of the compiled code before emitting machine code. This is why hot functions, which are functions that get executed often, get re-compiled with TurboFan, the optimizing compiler in V8 for both WebAssembly and JavaScript. However, because it emits code for each WebAssembly instruction independently, there is very little room for optimizations, like improving register allocations or common compiler optimizations like redundant load elimination, strength reduction, or function inlining. Liftoff emits decently fast machine code in a very short period of time. Once Liftoff compilation is finished, the the resulting machine code gets registered with the WebAssembly module, so that for future calls to the function the compiled code can be used immediately. Indeed, Liftoff can compile WebAssembly code very fast, tens of megabytes per second. One-pass compilers excel at fast code generation, but can only apply a small set of optimizations. Liftoff is a one-pass compiler, which means it iterates over the WebAssembly code once and emits machine code immediately for each WebAssembly instruction. Instead, functions get compiled lazily with the baseline compiler Liftoff when the function gets called for the first time.
Initially, V8 does not compile any functions in a WebAssembly module. In this document we dive into the WebAssembly compilation pipeline in V8 and explain how we use the different compilers to provide good performance. WebAssembly is a binary format that allows you to run code from programming languages other than JavaScript on the web efficiently and securely. Research WebAssembly compilation pipeline.
0 kommentar(er)
