Hi Honggyu The focus for Ignition to date has been getting off the ground. It passes ~90% of test262, but there's still a ways to go. There has been zero optimization effort beyond identifying opportunities. There are plenty of thoughts on this, but it'll be measurement driven.
We don't expect Ignition to match FullCode anytime soon, but there's a wider set of engineering concerns than speed alone. For example, Ignition could be easier to maintain as it has less architecture specific code and this may make it easier to bring language features out faster. Ditto TurboFan vs Crankshaft. Thanks Orion On Tue, Jan 5, 2016 at 2:28 PM, nexus226 <[email protected]> wrote: > Hi, > > I would like to ask someone here regarding the progress of current > TurboFan and Ignition interpreter. I have run v8 benchmark using TurboFan > and Ignition but the performace scores of most of benchmarks are much lower > than default, which is not as expected. (I didn't put the score numbers > because I don't want to put possible wrong numbers.) > Can anyone please verify if I test it correctly? > > I have executed benchmarks using x64.release version of d8 binary. > > 1. default execution is tested as below: > $ cd benchmarks > $ ../out/x64.release/d8 run.js > > 2. TurboFan is tested as below: > $ cd benchmarks > $ ../out/x64.release/d8 --turbo run.js > > As a result, TurboFan scores were much slower than default CrankShaft > execution scores. > > I also tested ignition interpreter mode as well but it seems that still > many bytecode generater functions have to be implemented. Because of this > unimplmentation issue, I was not able to execute the entire benchmark with > the driver run.js script. So I splitted all the v8 benchmarks with > seperate run scripts(e.g. run-richards.js, run-raytrace.js, etc.) to see > Scores for each benchmark. > > 3. ignition is tested as below: > $ cd benchmarks > $ ../out/x64.release/d8 --ignition --ignition_filter=* > --ignition_fake_try_catch run-richards.js > $ ../out/x64.release/d8 --ignition --ignition_filter=* > --ignition_fake_try_catch run-deltablue.js > $ ../out/x64.release/d8 --ignition --ignition_filter=* > --ignition_fake_try_catch run-crypto.js > ... > It still fails when I run early-boyer.js and regexp.js, but I could get > other benchmarks result. I was able to see the memory consumption is > reduced but performance was to much behind compared to the default ones. > > Since I've got these unexpected performance results of TurboFan and > Ignition, I would like to verify if I made some mistakes when I tested > benchmarks. Please let me know if I made some terrible mistakes regarding > execution options or other environments. > I appreciate all kind of answers. > > Thanks, > Honggyu > > -- > -- > 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.
