Chrome Flash Pepper reports V1 0.891s V2 0.828s V3 0.276s Great results! This will clearly improve large ArrayCollections deserialization! Thank you for this!
On 19 October 2013 01:12, Maurice Amsellem <maurice.amsel...@systar.com>wrote: > Forgot about this :-( > You are right, FF is running debugger version of FP, so that explains the > difference. > Thanks for the check. > > BTW, thanks also for the static recycled ByteArray idea, this is an easy > and efficient optimization that I could also use in my own code. > > Maurice > > -----Message d'origine----- > De : Alex Harui [mailto:aha...@adobe.com] > Envoyé : samedi 19 octobre 2013 02:07 > À : dev@flex.apache.org > Objet : Re: UID performance > > I assume you double-checked in flashplayerversion.com that FF is running > a release player? > > It could be that bytearray got optimized over time as it was a key piece > of gaming strategy and Alchemy. > > -Alex > > On 10/18/13 5:04 PM, "Maurice Amsellem" <maurice.amsel...@systar.com> > wrote: > > >FYI, I tried with Chrome (FP 11.9) and Firefox (FP 11.4) and the > >results are quite different: > > > >Chrome FP 11.9: > >1.29 > >1.24 > >0.405 > > > >FireFox FP 11.4: > >2.16 > >1.93 > >1.07 > > > >Is the difference expected ? That is, is FP 11.9 twice as fast as FP > >11.4, or is it just for these computations. > > > >Maurice > > > >-----Message d'origine----- > >De : Maurice Amsellem [mailto:maurice.amsel...@systar.com] > >Envoyé : samedi 19 octobre 2013 01:59 > >À : dev@flex.apache.org > >Objet : RE: UID performance > > > >I tried already and the results are consistent. > > > >Here is the link: > > > >https://dl.dropboxusercontent.com/u/12169005/ApacheFlex/TestUIDPerf_rel > >eas > >e.swf > > > >Maurice > > > >-----Message d'origine----- > >De : Alex Harui [mailto:aha...@adobe.com] Envoyé : samedi 19 octobre > >2013 > >01:56 À : dev@flex.apache.org Objet : Re: UID performance > > > >Actually, I thought of one more test that needs to be run which I don't > >have time to do right now: Release player with release SWF. Sometimes > >you can get very different results on the debugger players. > > > >Thanks, > >-Alex > > > >On 10/18/13 4:53 PM, "Alex Harui" <aha...@adobe.com> wrote: > > > >>Yup, V3 wins on mac and windows for me. > >> > >>Thanks for doing it. > >> > >>-Alex > >> > >>On 10/18/13 4:45 PM, "Maurice Amsellem" <maurice.amsel...@systar.com> > >>wrote: > >> > >>>I have posted FB project with the source s + compiled SWF. > >>> > >>>https://dl.dropboxusercontent.com/u/12169005/ApacheFlex/TestUIDPerf.f > >>>x > >>>p > >>>https://dl.dropboxusercontent.com/u/12169005/ApacheFlex/TestUIDPerf.s > >>>w > >>>f > >>> > >>>Algorithm 1= Original > >>>Algorithm 2 = using Array.join() and other optimizations by JMcLean > >>>Algo 3: using single static byteArray (Maurice) > >>> > >>>These are the results I get for 100,000 iterations: > >>> > >>>Original: 2.74s > >>>V2: 2.5s > >>>V3: 1.70s > >>> > >>>Please can you test on your side as well... > >>> > >>>Maurice > >>> > >>>-----Message d'origine----- > >>>De : Maurice Amsellem [mailto:maurice.amsel...@systar.com] > >>>Envoyé : samedi 19 octobre 2013 00:56 À : dev@flex.apache.org Objet : > >>>RE: UID performance > >>> > >>>I have used a static ByteArray as suggested: much better. > >>>Thanks for the suggestion. > >>> > >>> > >>>-----Message d'origine----- > >>>De : Maurice Amsellem [mailto:maurice.amsel...@systar.com] > >>>Envoyé : samedi 19 octobre 2013 00:49 À : dev@flex.apache.org Objet : > >>>RE: UID performance > >>> > >>>Sure, I will do that. > >>> > >>>I will also post the source so that you can recompile at your > >>>convenience. > >>> > >>>Maurice > >>> > >>> > >>>-----Message d'origine----- > >>>De : Alex Harui [mailto:aha...@adobe.com] Envoyé : samedi 19 octobre > >>>2013 > >>>00:13 À : dev@flex.apache.org Objet : Re: UID performance > >>> > >>>I'm a bit surprised that ByteArray.writeByte is faster but maybe > >>>Array.join isn't that fast. I would also like to see a test of a > >>>single static bytearray and resetting it and/or overwriting the old > >>>bytes. > >>> > >>>Maybe if you put a SWF with various algorithms in your people.a.o > >>>folder folks can hit it and make sure we all get the same results and > >>>then we'll know which algorithm to check in. > >>> > >>>Thanks for this idea as well. > >>> > >>>-Alex > >>> > >>>On 10/18/13 3:05 PM, "Maurice Amsellem" <maurice.amsel...@systar.com> > >>>wrote: > >>> > >>>>Any answer ? > >>>> > >>>>-----Message d'origine----- > >>>>De : Maurice Amsellem [mailto:maurice.amsel...@systar.com] > >>>>Envoyé : vendredi 18 octobre 2013 14:39 À : dev@flex.apache.org Objet > : > >>>>RE: UID performance > >>>> > >>>>Hi, > >>>> > >>>>I shouldn't have mixed the two options, because it's confusing. > >>>> > >>>>Actually, V3 option generates uppercase, exactly as the current > >>>>UUID, and it's 2x faster. > >>>>So let's forget about V4... > >>>> > >>>>Here is the code: > >>>> > >>>> private static const HEX_CHARS:String = "0123456789ABCDEF"; > >>>> private static const DASH:int = 45; // "-" > >>>> > >>>>public static function createUID():String > >>>> { > >>>> var ba:ByteArray = new ByteArray(); > >>>> var i:int; > >>>> var j:int; > >>>> > >>>> for (i = 0; i < 8; i++) { > >>>> ba.writeByte(HEX_CHARS.charCodeAt(Math.random() * 16)); > >>>> } > >>>> > >>>> for (i = 0; i < 3; i++) > >>>> { > >>>> ba.writeByte(DASH); > >>>> for (j = 0; j < 4; j++) > >>>> { > >>>> ba.writeByte(HEX_CHARS.charCodeAt(Math.random() * 16)); > >>>> } > >>>> } > >>>> > >>>> ba.writeByte(DASH); > >>>> > >>>> var time:Number = new Date().getTime(); > >>>> > >>>> var timeString:String = ("0000000" + > >>>>time.toString(16).toUpperCase()).substr(-8); > >>>> ba.writeUTFBytes(timeString); > >>>> > >>>> for (i = 0; i < 4; i++) > >>>> { > >>>> ba.writeByte(HEX_CHARS.charCodeAt(Math.random() * 16)); > >>>> } > >>>> > >>>> return ba.toString(); > >>>> } > >>>> > >>>>-----Message d'origine----- > >>>>De : Justin Mclean [mailto:jus...@classsoftware.com] Envoyé : > >>>>vendredi > >>>>18 octobre 2013 14:34 À : dev@flex.apache.org Objet : Re: UID > >>>>performance > >>>> > >>>>Hi, > >>>> > >>>>> V3: generates the random hex bytes one by one, and writes them to > >>>>> the BA > >>>>> V4: generates 4 or 8 random bytes in one shot, and writes them to > >>>>> the BA (hexa in lowercase, to avoid call toUppercase() ) > >>>>Probably best not too fiddle too much with the UID construction. > >>>>While I can't think of any major issues it would cause there could > >>>>be subtle issues due to seeding of random numbers and the like. > >>>> > >>>>> I don't know if this is acceptable (are we breaking some code if > >>>>>we use lowercase instead of uppercase). > >>>> > >>>>Again can't think of any issues but probably safer if we kept it > >>>>upper case. > >>>> > >>>>Go ahead and post the code it could be interesting/useful to someone. > >>>> > >>>>Thanks, > >>>>Justin > >>> > >> > > > > -- João Fernandes
