编译点滴 » WebKit

JIT Compiler and Virtual Machine SW Engineer for Web Browsers in Qualcomm

erlv — Sun, 09 Oct 2011 08:03:53 +0000

SOURCE: https://jobs.qualcomm.com/public/jobDetails.xhtml?requisitionId=1887529&page=jobSearch

Division	Qualcomm CDMA Technology
Job Area	Engineering – Software
Location	California – San Diego
Job Function	The JIT compiler and Virtual Machines team in the Web Technologies group at Qualcomm Innovation Center focuses on development and optimization of software virtual machines (both interpreter and JIT compiler based), for languages like Java, JavaScript, Action Script, etc, commonly used in Web Browsers, for Qualcomms various MSM platforms. This role requires the following essential duties and responsibilities: Research, design, develop, enhance, and implement the different components of the software in a virtual machine by efficiently utilizing/interfacing the HW in Qualcomms SOCs, leading to optimized performance, power, memory usage, and user experience. Collaborate with chipset architecture and hardware engineers, and come up with software solutions best suited for Qualcomms SOCs. Develop and execute benchmark and test scenarios for the virtual machines based on the different use-cases in a web browser. Work with Product and Systems teams to understand and improve the design and working of the software involved in the virtual machines. Analyze and identify system level integration issues, interface with the browser integration and test teams. Follow Qualcomms open source design practices Keep up with the web technology landscape (HTML5, CSS, JavaScript, etc) and help drive design aspects of Qualcomm CDMA Technology’s (QCT) chipsets necessary for supporting these technologies Plan and integrate new features into the software for virtual machines based on open web standards such as W3C and the language standards (e.g. ECMA for JavaScript).
Skills/Experience	Experience in compiler development – compiler intermediate representations, machine independent and machine dependent optimizations, instruction selection, scheduling, register-allocation, back-end optimizations. Working knowledge of Operating System concepts, process/thread, memory management, garbage collection algorithms, threads scheduling. Working knowledge of assemblers, linkers, loaders, object file formats, instruction set simulators and experience using them. Knowledge of the structure & function of the virtual machine internals byte-code formats, heap, constant pool resolution, garbage collection, execution mechanisms, etc. Understanding of the compilation challenges and potential solutions for languages like Java (e.g., dynamic dispatch, polymorphism, inline caching) and JavaScript (e.g., dynamic types, dynamic addition/deletion of properties/values to objects, prototype based inheritance). Knowledge of computer architectures – pipeline structure & hazards, cache & memory organization, etc. Knowledge of the software architecture and internals of a web browser, JavaScript to Webkit bindings, JavaScript engines interaction with DOM and Plugins. Strong background in embedded software development, experience in assembly language programming and optimization. 5 to 10 years of programming experience in C/C++. Should have exposure to large scale software build, integration and test environments Strong communication and interpersonal skills required
Responsibilities	Knowledge and experience in Android or Chrome OS mobile platforms. ARM architecture knowledge and assembly code optimization experience. Experience with scripting languages used in build systems and hands-on experience on debugging embedded systems. Experience with JTAG and ICE debuggers
Education Requirements	Masters degree in Computer Science, Computer Engineering or related field, with experience in the technical domain. PhD degree and research work in the related technical area also encouraged. Candidates close to graduating with Masters or PhD in Computer Science/Engineering with project experience in the technical area can also apply.

龙芯N32系统上的Squirrelfish性能

erlv — Mon, 28 Jun 2010 08:58:37 +0000

缘起：写这篇文章因为报名参加了龙芯开源软件大赛，项目是《Webkit-龙》,把webkit移植到龙芯N32系统上，起初是想做chromium的移植。但在起步的过程中，发现一是Google的V8引擎，对mips的支持基本处于不可用状态，而且chromium为了追求性能，使用的方式太过灵巧，如Native client(Google的野心–Native Client+LLVM),凭自己的一己之力，靠业余时间，很难完成，只好先做webkit了。

Squirrelfish 是Webkit的Javascript引擎。针对龙芯平台(MIPS)已经有了JIT支持，但仅仅对O32系统才有，本博曾经介绍过相关内容（WebKit和Firefox的JavaScript性能对比）。

切入正题，N32系统下，龙芯平台上，Sunspider性能测试结果如下：
============================================ RESULTS (means and 95% confidence intervals) -------------------------------------------- Total: 9125.6ms +/- 0.1% -------------------------------------------- 3d: 1058.6ms +/- 0.2% cube: 326.9ms +/- 0.1% morph: 380.9ms +/- 0.6% raytrace: 350.8ms +/- 0.1% access: 1441.2ms +/- 0.1% binary-trees: 108.4ms +/- 0.3% fannkuch: 829.2ms +/- 0.2% nbody: 249.0ms +/- 0.0% nsieve: 254.6ms +/- 0.4% bitops: 1088.6ms +/- 0.2% 3bit-bits-in-byte: 196.2ms +/- 0.3% bits-in-byte: 250.1ms +/- 0.2% bitwise-and: 224.9ms +/- 0.3% nsieve-bits: 417.4ms +/- 0.3% controlflow: 138.2ms +/- 0.2% recursive: 138.2ms +/- 0.2% crypto: 621.5ms +/- 0.1% aes: 323.0ms +/- 0.1% md5: 145.3ms +/- 0.2% sha1: 153.2ms +/- 0.3% date: 523.6ms +/- 0.6% format-tofte: 302.4ms +/- 0.6% format-xparb: 221.2ms +/- 1.3% math: 931.3ms +/- 0.1% cordic: 356.0ms +/- 0.2% partial-sums: 324.5ms +/- 0.1% spectral-norm: 250.8ms +/- 0.2% regexp: 1681.6ms +/- 0.1% dna: 1681.6ms +/- 0.1% string: 1641.0ms +/- 0.2% base64: 232.9ms +/- 0.2% fasta: 261.9ms +/- 0.2% tagcloud: 394.9ms +/- 0.2% unpack-code: 525.2ms +/- 0.2% validate-input: 226.1ms +/- 0.7%

与O32系统的性能相比：9125ms vs 6355ms，慢了一半。一个原因是N32现在没有JIT支持。

正确性方面，使用webkit的 Javascript测试集测试。有7个例子出错：
ecma/Date/15.9.5.14.js ecma/Date/15.9.5.28-1.js ecma/Date/15.9.5.29-1.js ecma/Date/15.9.5.31-1.js ecma/Date/15.9.5.34-1.js ecma/Date/15.9.5.35-1.js ecma/Math/15.8.2.17.js

WebKit-龙项目方面：

主要优化JavaScript引擎，
现在WebKit虽然能编译，但运行还会报错：

Can't find built framework at "/root/WebKit/WebKitBuild/Release/lib/libQtWebKit.so".

这个项目刚刚启动，欢迎加入

WebKit和Firefox的JavaScript性能对比

erlv — Fri, 22 Jan 2010 14:37:22 +0000

严格的说WebKit仅仅是个浏览器核心，采用该核心的浏览器很多，如国内的搜狗浏览器，遨游浏览器。其他的如google的chrome(Windows平台，linux平台下为chromium),epiphany(linux平台下，gnome2.28版本之后),苹果的Safari 都采用了webkit的内核。Firefox则是采用Gecko的内核，这是NetScape公司开发的内核，后来开源，mozilla继续开发。另外，现在还有另外两种常见的浏览器内核，Trident主要用在IE系列上，Presto主要用在Opera上。

这篇文章仅仅针对浏览器处理JavaScript的性能作比较，主要在X86平台和龙芯平台。先来解释一下JavaScript，JavaScript是互联网内较为常用的脚本语言，面向对象，主要在浏览器内解释执行，用于生成动态网页，因为很多语言特性受Java影响，所以叫JavaScript。通过JavaScript，浏览器可以运行服务器想要在访问者终端上运行的一些计算程序，以达到更好的浏览体验。

再说说浏览器内核和JavaScript的关系，其实JavaScrip脚本的执行仅仅是浏览器内核的一部分，其他的还有Html语言的解释执行，网页的呈现等等也是内核要做的。之所以这么关心浏览器核心对JavaScript脚本的处理情况，是因为现在的很多应用不再是简单的网页浏览，如gmail，google reader，gooe wave还有一些网页3D特效等等都需要在客户机上作计算，这就需要JavaScript大显身手了。而且越来越多的应用依赖JavaScript，所以现在浏览器对JavaScript的处理速度直接影响着用户体验。目前WebKit的JavaScript引擎SquirrelFish，JavaScript的引擎是SpiderMonkey

目前JavaScript在大多数平台上的处理是靠解释执行的，又因为是动态类型，面向对象。。。，决定了JavaScript执行效率低，所以就诞生了各种针对JavaScript的优化，也有了测试JavaScipt解释器性能的BenchMark，目前常用的是Sunspider和google V8，这篇文章中的对比针对Sunspider测试集。

另外还要说说JIT，Just In time，转换，将部分程序代码直接转换成机器码执行，这种技术在运行时优化中比较常用，JavaScript是解释器，所以JIT在解释器中也是很重要的优化手段。目前X86的webkit和Firefox默认就有JIT支持，但龙芯平台上还没有，本博也是最早对龙芯2F平台有JIT支持的Webkit和firefox JavaScript引擎作对比的，目前这两个JIT都还没有进入官方的代码库中，感兴趣的朋友可以在下面的前两个链接中找到相关源码。其中webkit的补丁还要做些宏的修改，改动不大。这两个都是从官方源码库checkout出来并修改之后的。其中firefox的JIT部分是由ZSC大牛写出来的，详细的讨论贴和测试结果可以在龙芯论坛看到，见下面的第三个链接。
x86下epiphany(webkit内核)和firefox开了jit之后的性能，如下

webkit Total:                  833.6ms +/- 2.1%
firefox Total:                 1774.4ms +/- 2.7%

（测试环境是我的笔记本 intel Pentium t2390,开了频率调节，所以波动较大)、两个差距：

TEST                   COMPARISON            FROM                 TO             DETAILS

=============================================================================

** TOTAL **:           2.13x as fast     1774.4ms +/- 2.7%   833.6ms +/- 2.1%     significant

=============================================================================

  3d:                  2.96x as fast      257.2ms +/- 2.1%    87.0ms +/- 1.0%     significant
    cube:              3.36x as fast       79.2ms +/- 1.7%    23.6ms +/- 2.9%     significant
    morph:             1.52x as fast       50.4ms +/- 1.4%    33.2ms +/- 3.1%     significant
    raytrace:          4.23x as fast      127.6ms +/- 3.0%    30.2ms +/- 3.4%     significant

  access:              3.26x as fast      237.0ms +/- 4.7%    72.8ms +/- 2.8%     significant
    binary-trees:      8.82x as fast       67.0ms +/- 4.4%     7.6ms +/- 9.0%     significant
    fannkuch:          3.48x as fast       98.2ms +/- 0.6%    28.2ms +/- 3.7%     significant
    nbody:             2.50x as fast       51.4ms +/- 10.3%    20.6ms +/- 3.3%     significant
    nsieve:            -                   20.4ms +/- 56.6%    16.4ms +/- 4.2% 

  bitops:              1.42x as fast       53.8ms +/- 1.9%    38.0ms +/- 0.0%     significant
    3bit-bits-in-byte: *3.10x as slow*      2.0ms +/- 0.0%     6.2ms +/- 9.0%     significant
    bits-in-byte:      -                   11.0ms +/- 8.0%    10.4ms +/- 6.5%
    bitwise-and:       *2.33x as slow*      3.0ms +/- 0.0%     7.0ms +/- 0.0%     significant
    nsieve-bits:       2.62x as fast       37.8ms +/- 1.5%    14.4ms +/- 4.7%     significant

  controlflow:         8.76x as fast       50.8ms +/- 2.0%     5.8ms +/- 9.6%     significant
    recursive:         8.76x as fast       50.8ms +/- 2.0%     5.8ms +/- 9.6%     significant

  crypto:              2.34x as fast       93.8ms +/- 1.7%    40.0ms +/- 2.2%     significant
    aes:               2.57x as fast       54.0ms +/- 3.6%    21.0ms +/- 4.2%     significant
    md5:               2.62x as fast       26.2ms +/- 2.1%    10.0ms +/- 0.0%     significant
    sha1:              1.51x as fast       13.6ms +/- 5.0%     9.0ms +/- 0.0%     significant

  date:                1.87x as fast      275.2ms +/- 2.0%   147.0ms +/- 1.0%     significant
    format-tofte:      2.75x as fast      149.0ms +/- 1.0%    54.2ms +/- 1.0%     significant
    format-xparb:      1.36x as fast      126.2ms +/- 4.0%    92.8ms +/- 1.7%     significant

  math:                1.12x as fast       89.8ms +/- 1.8%    80.2ms +/- 21.7%     significant
    cordic:            2.13x as fast       43.4ms +/- 3.3%    20.4ms +/- 9.2%     significant
    partial-sums:      *1.41x as slow*     34.4ms +/- 2.0%    48.6ms +/- 38.2%     significant
    spectral-norm:     1.07x as fast       12.0ms +/- 0.0%    11.2ms +/- 12.2%     significant

  regexp:              3.73x as fast      125.4ms +/- 15.5%    33.6ms +/- 2.0%     significant
    dna:               3.73x as fast      125.4ms +/- 15.5%    33.6ms +/- 2.0%     significant

  string:              1.80x as fast      591.4ms +/- 3.3%   329.2ms +/- 1.6%     significant
    base64:            *1.24x as slow*     28.4ms +/- 2.4%    35.2ms +/- 1.6%     significant
    fasta:             1.71x as fast      133.0ms +/- 2.3%    78.0ms +/- 4.6%     significant
    tagcloud:          2.81x as fast      169.4ms +/- 3.8%    60.2ms +/- 8.2%     significant
    unpack-code:       1.91x as fast      193.8ms +/- 4.4%   101.6ms +/- 2.7%     significant
    validate-input:    1.23x as fast       66.8ms +/- 5.8%    54.2ms +/- 6.4%     significant

龙芯上的JIT具体对比数据不便透漏，可以从下面第三个链接中找到第一版发布时的测试数据。这里仅仅给出一个对比。webkit之所以with GUI是使用编译出的迷你小浏览器中测试结果，no GUI是使用测试脚本跑出来的结果。

 Webkit with GUI Total:                  6723.4ms +/- 0.5%
        no GUI   Total:                  6355.8ms +/- 0.5%
 Firfox Total 7000 ms 左右。

注：龙芯中的JIT现在还不够成熟，其中Webkit的JIT仅是针对MIPS III做的，没有针对龙芯的特殊优化。

另外，X86中firefox的版本是3.5.7，龙芯中firefox的版本是3.7pre1.WebKit在X86中版本是epiphany的2.28中的版本，龙芯中为SVN中的2010-01-12 r53114。X86运行环境为Gentoo，Gcc4.4.2,Glibc2.11,龙芯运行环境是玲珑一体机，GCC4.3(龙梦修改版）,Glibc2.10。

对比结果可见：X86中WebKit浏览器的性能要明显优于Firefox，达到两倍之多。龙芯相比起来还有些差距，不过相信优化后龙芯的性能会更好。

参考

https://bugs.webkit.org/show_bug.cgi?id=30144

http://dev.lemote.com/code/firefox-3.7-loongson-jit

http://lemote.com/bbs/viewthread.php?tid=26687&extra=page%3D1

http://en.wikipedia.org/wiki/Google_Chrome

http://ineolee.com/pc/several-browser-core-introduction/

GNOME 2.22 抢先预览

erlv — Mon, 29 Nov 1999 16:00:00 +0000

2008-03-12 Toy Posted in Featured Post, GNOMERSSTrackback

GNOME 2.22 是 GNOME 桌面环境目前最新的稳定版，该版本计划在今天正式发布。根据 GNOME 项目团队的开发周期，每隔 6 个月，GNOME 都会发布一个新的版本。可以预见的是，在 Ubuntu 8.04、Fedora 9、openSUSE 11 等 Linux 发行版中都将装载 GNOME 2.22。在 GNOME 2.22 中都有哪些新特性，让我们一起来关注。

Cheese：摄像头程序

Cheese 是一个有趣的小程序，它允许用户从摄像头拍摄照片或录制视频。之所以说它有趣，是因为该程序自带了许多神奇的效果，你只需轻点鼠标就可以将其应用到所拍摄的照片或录制的视频上。试试 Cheese，相信你会喜欢它的。

Metacity：加入了窗口混合特性

GNOME 桌面环境的默认窗口管理器 Metacity 已经具有了混合特性，这使得你的桌面即使不用 Compiz，也可以获得诸如下拉阴影、窗口透明、以及 Alt+Tab 切换效果。

Totem：支持 DVD 及 DVB 播放

在 GNOME 中的电影播放器 Totem 已改进了对 DVD 及 DVB (数字电视) 播放的支持。

GVFS

GNOME 2.22 引入了 GVFS。GVFS 为文件传输/复制/移动对话框提供更多信息，并支持使用诸如 cdda://、gphoto2:// 等之类的新协议来访问。

Clock applet：支持显示多个时区

时钟面板程序引入了从世界各地显示多个时区的新功能。

Evolution：更多改善

Evolution 不仅支持 Google Calendar，而且能够为邮件打 tag。此外，在垃圾邮件过滤方面也有新的进展。

Vinagre：远程桌面

GNOME 2.22 加入了新的远程桌面程序 Vinagre，它支持浏览服务器网络，并允许收藏经常使用的连接。

键盘设置：更加简单

GNOME 2.22 对 Keyboard Preferences 窗口进行了重新设计，使键盘的设置更加简单。

其他特性

在 GNOME 2.22 中增加的其他特性也值得注意，包括：改善了屏幕阅读和放大功能，新的 Mousetweaks 鼠标工具，集成了 PolicyKit，更佳的系统工具，在 Deskbar 中可以直接打开 Evolution 联系人，改进了文本编辑器 gedit 的打印功能，Chess 支持网络多玩家对战，Epiphany 支持 Webkit 后端和下载通知功能，存档管理器支持 LZMA (7-zip)，当烧录 CD 时能禁止自动挂机和休眠，文档查看器 Evince 具有更快的执行速度、更低的内存占用及支持页面过渡效果 (PDF slideshows)，Tomboy 支持分类整理笔记，Sound Juicer 支持更多的元数据，计算器程序的精度和可用性都得到了改进，Keyring Manager 被 Seahorse 完全取代，等等。

您可能也喜欢：
webkit-龙小组成立	zzMozilla的疯狂构想 Firefox 4将开拓浏览器新疆界	龙芯N32系统上的Squirrelfish性能	JIT Compiler and Virtual Machine SW Engineer for Web Browsers in Qualcomm	擦亮眼睛 “百度Hi”用户协议有猫腻
无觅

编译点滴 » WebKit

JIT Compiler and Virtual Machine SW Engineer for Web Browsers in Qualcomm

相关文章：

龙芯N32系统上的Squirrelfish性能

相关文章：

WebKit和Firefox的JavaScript性能对比

相关文章：

GNOME 2.22 抢先预览

相关文章：