V8 引入了新的非优化编译器以提高性能 (#8870)

* update v8-sparkplug-compiler article

* update v8-sparkplug-compiler article

* update v8-sparkplug-compiler article

* update v8-sparkplug-compiler article

* update v8-sparkplug-compiler article

* update v8-sparkplug-compiler article

* update v8-sparkplug-compiler article

* Update v8-sparkplug-compiler.md
Co-authored-by: Nlsvih <lsvih@qq.com>

article/2021/v8-sparkplug-compiler.md

@@ -2,32 +2,32 @@
 > * 原文作者：[Sergio De Simone](https://www.infoq.com/profile/Sergio-De-Simone/)
 > * 译文出自：[掘金翻译计划](https://github.com/xitu/gold-miner)
 > * 本文永久链接：[https://github.com/xitu/gold-miner/blob/master/article/2021/v8-sparkplug-compiler.md](https://github.com/xitu/gold-miner/blob/master/article/2021/v8-sparkplug-compiler.md)
-> * 译者：
-> * 校对者：
+> * 译者：[CarlosChenN](https://github.com/CarlosChenN)
+> * 校对者：[Chorer](https://github.com/Chorer)，[PassionPenguin](https://github.com/PassionPenguin)
 
-# V8 Gets a Non-Optimizing Compiler Stage to Improve Performance
+# V8 引入了新的非优化编译器以提高性能
 
-The latest version of the JavaScript V8 engine, V8 9.1, introduces a new intermediate compiler stage, called [Sparkplug](https://v8.dev/blog/sparkplug), that improves performance on real-world benchmarks by 5-15%, says V8 engineer [Leszek Swirski](https://twitter.com/leszekswirski). It will be available in the upcoming Chrome 91.
+最新版本的 JavaScript V8 引擎，V8 9.1 ，引入一个过渡编译阶段，叫 [Sparkplug ](https://v8.dev/blog/sparkplug) 。据 V8 开发者 [Leszek Swirski](https://twitter.com/leszekswirski) 介绍，它能在现实基准中提升 5~15% 的性能。它会在即将发布的 Chrome 91 中正式使用。
 
-The old V8 architecture included just two stages: Ignition, a JavaScript interpreter, and TurboFan, a highly optimising compiler. Ignition takes a JavaScript AST and generates V8 bytecode, while TurboFan is able to generate machine code from that. The reason for the introduction of Sparkplug is clearly explained in the [Sparkplug design](https://docs.google.com/document/d/1NeOsqjPPAcDWbuHxW5MobzVQgj9qZd6NqKUnz0h-fOw/edit) overview:
+旧的 V8 架构中包括了两个阶段： Ignition （一个 JavaScript 解释器），和 TurboFan （一个高度优化的编译器），Ignition 使用 JavaScript 抽象语法树去生成 V8 字节码，而 TurboFan 使用这些字节码生成机器码。引入 Sparkplug 的原因，在  [Sparkplug 设计](https://docs.google.com/document/d/1NeOsqjPPAcDWbuHxW5MobzVQgj9qZd6NqKUnz0h-fOw/edit) 的概述中有详细的解释。
 
-> There is a big performance cliff between \[Ignition and TurboFan\]; staying too long in the interpreter means we don’t take advantage of optimisation, but calling TurboFan too early might mean we “waste” time optimising functions that aren’t actually hot — or worse, it means we deopt. We can reduce this gap with a simple, fast, non-optimising compiler, that can quickly and cheaply tier-up from the interpreter by linearly walking the bytecode and spitting out machine code. We call this compiler Sparkplug.
+> 在 Ignition 和 TurboFan 之间有一个巨大的性能差异；长时间驻留在解释器中，意味着我们无法获得优化的好处，但是太早调用 TurboFan 又意味着，我们在优化那些不属于热点代码的函数，甚至更糟，它意味着我们在做些不必要的事。我们可以用一个简单且快速的非优化编译器去缩小这个差距，这样可以通过线性遍历字节码并吐出机器码，快速地、低成本地从解释器中分层。我们叫这个编译器为 Sparkplug 。
 
-Before the introduction of Ignition and TurboFan, V8 used to have a fast JIT compiler called Full-CodeGen (FCG). As Leszek [explains in a Hacker News thread](https://news.ycombinator.com/item?id=27307862), this was ditched in favour of a new interpreter architecture, called Ignition, which did not include FCG.
+在引入 Ignition 和 TurboFan 之前，V8 曾经使用一个快速即时编译器，叫 Full-CodeGen（FCG）。正如 Leszek [在黑客新闻中解释](https://news.ycombinator.com/item?id=27307862) 到，新的解释器架构 Ignition 抛弃了这种编译器，不再包含 FCG 。
 
-> The big difference is that Sparkplug compiles from bytecode, not from source, and thus the bytecode stays the source of truth for the program. Back in the FCG days, the optimising compiler had to re-parse the source code to AST, and compile from there - even worse, to be able to deoptimise back to FCG, it had to kind of "replay" FCG compilation to get the deopted stack frame right.
+> 一个大的不同是，Sparkplug 是从字节码中编译的，而不是源码，因此，字节码才是程序的真实来源。回到 FCG 时代，做优化的编译器必须重解释源码成抽象语法树，然后从那里编译，甚至更糟，能反优化退化至 FCG ，它不得不有点像重复 FCG 编译去纠正错选的栈帧。
 
-Leszek [mentions](https://news.ycombinator.com/item?id=27312037) that FCG was ditched instead of evolved because of its massive technical debt, specifically related to the fact that both the Ignition compiler and FCG had to compile from source and provide their output to TurboFan, which led to all kind of complexities. This also [made it harder to keep up with JavaScript new features](https://v8project.blogspot.com/2017/05/launching-ignition-and-turbofan.html).
+Leszek [提到](https://news.ycombinator.com/item?id=27312037) FCG 被抛弃而不是改进的原因是它巨大的技术负担，特别是与 Sparkplug 编译器和 FCG 不得不从源码编译，然后把输出提供给 TurboFan 有关，因为这导致了各种各样的复杂情况。这同样也[让 FCG 更难跟上 JavaScript 的新特性](https://v8project.blogspot.com/2017/05/launching-ignition-and-turbofan.html) 。
 
-Sparkplug speed comes from two factors, says Swirski. First, it relies on bytecode generated by Ignition, which means a bunch of work has already been done, including variable resolution, figuring out if parentheses are actually arrow functions, desugaring destructuring statements, and so on. Additionally, Sparkplug does not generate any intermediate representation (IR), rather it outputs machine code in a single linear pass. This approach means Sparkplug cannot do any advanced optimizations based on IR and it must be entirely ported to each new platform.
+Swirski 称，Sparkplug 的速度来源于两个因素。首先，它依赖于 TurboFan 生成的字节码，这意味着大量的工作已经完成了，包括变量解析、判断圆括号是否是箭头函数、为解构声明语句进行脱糖等等。此外，Sparkplug 并不生成任何中间产物，而是在一个单一的线性通道中输出机器码。这种方式意味着 Sparkplug 不能做任何基于中间产物的高级优化，它必须全面移植到每个新的平台上。
 
-When it comes to performance, Sparkplug improves both the [Speedometer](https://browserbench.org/Speedometer2.0/) as well as a set of real-world benchmarks used by the V8 team. The improvement is in the range of 5-15% depending of test machine and website. Google has not released official low-level benchmarks comparing the various pipeline components' performance. Leszek, though, [explains](https://news.ycombinator.com/item?id=27308038) that:
+在性能方面，Sparkplug 改进了 [Speedometer](https://browserbench.org/Speedometer2.0) 以及 V8 团队使用的一组现实的基准测试。根据测试机器和网站的不同，性能的提高范围在 5~15% 之间。谷歌尚未发布各种管道组件性能的官方低级基准。不过 Leszek [在这解释过](https://news.ycombinator.com/item?id=27308038) ：
 
-> Compile time is on roughly the same order of magnitude as Ignition compilation (just AST to bytecode, so excluding parsing), and roughly two to three orders of magnitude faster that TurboFan. The relative performance to the interpreter, as the sibling comments point out, varies **wildly** by workload, but around 4x is probably a decent approximation for something not entirely dominated by property loads.
+> 编译时间和 Ignition 编译器大致相同（因为只是抽象语法树到字节码，不包括解析），大概比 TurboFan 快二到三个量级。正如评论里所指出的，解释器的相对性能会由于工作量的不同而产生很大的差别，在不完全由属性负荷来控制的情况下，大概 4 倍就算是个不错的近似值了。
 
-Besides performance, another key benefit of Sparkplug is reduced CPU usage, which can improve battery usage on mobile devices as well as reduce the bill on pay-per-cycle servers, according to V8 designers.
+据 V8 的设计者说，除了性能，Sparkplug 带来的另一个关键收益是减少 CPU 的占用，而这可以减少移动设备的电量消耗，还可以减少按周期付费服务器的费用。
 
-As mentioned, Sparkplug is being [currently rolled out in Chrome 91](https://developer.chrome.com/blog/new-in-chrome-91/) so you will soon be able to try it out.  If you are interested in the nitty-gritty details of Sparkplug internals and the way it interfaces with Ignition and TurboFan, do not miss Swirski's writeup.
+正如提及到的，Sparkplug 会在 [Chrome 91 中推出](https://developer.chrome.com/blog/new-in-chrome-91/) 所以，你很快就能自己体验一下。如果你对 Sparkplug 的内核细节，以及它与 Ignition 和 TurboFan 的联系感兴趣。那么千万不要错过 Swirski 的文章。
 
 > 如果发现译文存在错误或其他需要改进的地方，欢迎到 [掘金翻译计划](https://github.com/xitu/gold-miner) 对译文进行修改并 PR，也可获得相应奖励积分。文章开头的 **本文永久链接** 即为本文在 GitHub 上的 MarkDown 链接。