You can turn on --serialize-eager to make sure you compile as much as possible before creating the code cache. We currently haven't implemented a way to incrementally add to the code cache.
Cheers, Yang On Wed, Apr 5, 2017 at 1:40 PM 'Daniel Vogelheim' via v8-dev < [email protected]> wrote: > I agree with your analysis, and I agree we could do better. Getting there > will take a bit, though. > > The reason why we currently cache only the immediately compiled functions > (which are top-level functions to IIFEs) is that the serialization is > fairly expensive. Just compiling more would help the cache-hit case, but > would have significant performance penalty for the cache-miss case(s), and > those are fairly common. I'd like to fix these issues. Here > <https://docs.google.com/document/d/1-HhAgqRpHs9YQF9Zs7pTovUw8EE06tR3_zMIxaeCyOw/edit#heading=h.xgjl2srtytjt>'s > a suggestion on how, plus some background info. The feedback I've gotten is > that this is not the best way to proceed, so I'll need to do a bit more > work and adapt the design. I'll update the doc when I know more. As said... > it'll take a bit. > > Daniel > > > On Wed, Apr 5, 2017 at 1:15 PM, 阿炳 <[email protected]> wrote: > > Hi, V8's code cache in Chrome can significantly improve WebApp's startup > time, code caching is very important to modern WebApp as WebApp using more > and more js. > As I dig a little deeper in V8 code caching, I wonder whether we can do > more improvement on it. > Below is the trace result of > http://m.sm.cn/s?q=%E9%A9%AC%E4%BA%91&from=smor&safe=1&snum=6, before I > capture this trace, I have browse this url 4 times to make sure chrome > generated code cache.(You can find V8.CompileDeserialize in the table.) > > Though we have code cache, we still need 501ms in CompileCode, 1/3 of > total JS time. > What's the problem? > I think it's because we only cache the top level functions and IIFEs compile > result. > Is there any possible to store the code cache not just after the script is > compiled, but later like the page is loaded or time to interactive? > If we can store the code cache a little later, I think we can store much > more functions those are not in top level but will be called shortly after, > therefore we can significantly shorten the startup time of WebApp next time. > > > > Function Wall Wall self CPU CPU self wall average count > v8.callFunction 716.022 ms 616.025 ms 2.257 ms 1.809 ms 9.625 ms 64 > v8.compile 227.817 ms 151.827 ms 9.691 ms 7.658 ms 0.791 ms 192 > V8.Compile 49.760 ms 30.138 ms 35.384 ms 22.936 ms 0.158 ms 191 > V8.CompileCode 501.925 ms 378.723 ms 143.903 ms 117.921 ms 0.379 ms 1000 > V8.CompileDeserialize 1.436 ms 1.186 ms 1.436 ms 1.186 ms 0.237 ms 5 > V8.CompileIgnition 48.886 ms 35.360 ms 48.886 ms 35.360 ms 0.035 ms 1018 > V8.CompileScript 208.867 ms 137.379 ms 13.413 ms 9.672 ms 0.716 ms 192 > V8.DeoptimizeCode 0.038 ms 0.035 ms 0.038 ms 0.035 ms 0.035 ms 1 > V8.Execute 2,083.692 ms 1,662.409 ms 1,003.724 ms 764.488 ms 5.523 ms 301 > V8.GC_Objects_Stats 0.000 ms 0.000 ms 0.000 ms 0.000 ms 0.000 ms 2 > V8.GCCompactor 74.109 ms 66.842 ms 0.179 ms 0.171 ms 33.421 ms 2 > V8.GCPhantomHandleProcessingCallback 0.244 ms 0.210 ms 0.105 ms 0.070 ms > 0.105 ms 2 > V8.GCScavenger 22.378 ms 19.512 ms 0.371 ms 0.219 ms 9.756 ms 2 > V8.ICStats 0.000 ms 0.000 ms 0.000 ms 0.000 ms 0.000 ms 1 > V8.NewContext 97.478 ms 87.695 ms 19.344 ms 17.471 ms 87.695 ms 1 > v8.newInstance 1.829 ms 1.521 ms 0.782 ms 0.567 ms 0.035 ms 43 > V8.OptimizeCode 6.630 ms 6.548 ms 0.118 ms 0.118 ms 6.548 ms 1 > V8.ParseFunction 144.272 ms 116.927 ms 144.272 ms 116.927 ms 0.159 ms 735 > V8.ParseProgram 129.480 ms 85.579 ms 96.469 ms 58.753 ms 0.448 ms 191 > V8.PreParse 33.011 ms 26.826 ms 33.011 ms 26.826 ms 0.363 ms 74 > V8.RecompileSynchronous 6.512 ms 6.430 ms 6.512 ms 6.430 ms 6.430 ms 1 > v8.run 1,376.877 ms 1,053.397 ms 6.397 ms 5.232 ms 5.486 ms 192 > V8.RunMicrotasks 2.101 ms 1.444 ms 2.101 ms 1.444 ms 0.206 ms 7 > V8.ScriptCompiler 218.126 ms 144.169 ms 9.259 ms 6.790 ms 0.751 ms 192 > > -- > -- > v8-dev mailing list > [email protected] > http://groups.google.com/group/v8-dev > --- > You received this message because you are subscribed to the Google Groups > "v8-dev" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > > > -- > -- > v8-dev mailing list > [email protected] > http://groups.google.com/group/v8-dev > --- > You received this message because you are subscribed to the Google Groups > "v8-dev" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- * • * *Yang Guo** • **Google Germany GmbH* * • *Erika-Mann-Str. 33 * • *80636 Munich • [email protected] Geschäftsführer: Matthew Scott Sucherman, Paul Terence Manicle Registergericht und -nummer: Hamburg, HRB 86891 Sitz der Gesellschaft: Hamburg Diese E-Mail ist vertraulich. Wenn Sie nicht der richtige Adressat sind, leiten Sie diese bitte nicht weiter, informieren Sie den Absender und löschen Sie die E-Mail und alle Anhänge. Vielen Dank. This e-mail is confidential. If you are not the right addressee please do not forward it, please inform the sender, and please erase this e-mail including any attachments. Thanks. -- -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev --- You received this message because you are subscribed to the Google Groups "v8-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
