Thanks, great testcase! Are those links going to remain up? I'll file a chromium bug if so, as I also see the chrome performance on asyncify as slower than I'd expect.
Otherwise I looked into the code, and most of the itime is spent inside a single huge function that asyncify has modified. Perhaps the main interpreter loop? It's hard to see more optimization opportunities there. I think that means we need a whitelist like in emterpreter. Btw, did you use the whitelist or blacklist there - which would be better? - Alon On Mon, Jul 22, 2019 at 10:03 PM Александр Гурьянов <[email protected]> wrote: > wdosbox.wasm.js - is wasm binary > wdosbox.js.symbols - symbols > > вт, 23 июл. 2019 г. в 12:03, Александр Гурьянов <[email protected]>: > > > > I am sorry guys, my estimation of sizes/preformance was wrong. > > > > Ubuntu 19.04 / Chrome 72.0.3626.121 > > > > O3: > > DEFAULT 2.0Mb 60 FPS / 22 FPS > > EMTERPRETIFY 2.2Mb 60 FPS / 40 - 50 FPS > > ASYNCIFY 3.0Mb 44-55 FPS / 11 - 44 FPS > > > > Oz: > > DEFAULT 1.8Mb 55-60 FPS / 22 FPS > > EMTERPRETIFY 2.1Mb 55-60 FPS / 22 - 47 FPS > > ASYNCIFY 2.8Mb 50-60 FPS / 13 - 35 FPS > > > > NOTES: > > - First fps number is for single dosbox instance on page > (index.html?game=0) > > - Second fps number is for 9 dosbox instances on page (index.html) > > - Default version used emscripten_set_main_loop, I think low fps for 9 > > instances is because stats.js can't calculate it correctly, cause > > everything work very smooth in Default mode. > > - In FF every version works near 60 FPS, but game expirence is not > > same (need to find better metric then FPS) > > > > From my point of view, ASYNCIFY is slower on ~10 FPS I tried to do > > basic profiling but didn't find any spikes. I think this slowness is > > related to bigger file size. Hope, option like EMTERPRETIFY_WHITELIST > > can solve this problem. > > ASYNCIFY have dramatic slow down (against emterpreter {in chrome?}) > > when you run multiple instance of same app on single page (maybe there > > is some synchronization or inside chrome). > > > > I swithced to emterpreter, becaue for now it's better. You can try > > this test here: > > ASYNCIFY: > > http://js-dos.com/6.22/64k/index-upstream.html?game=0 (1 game) > > http://js-dos.com/6.22/64k/index-upstream.html > > EMTERPRETIFY: > > http://js-dos.com/6.22/64k/index-fastcomp.html?game=0 (1 game) > > http://js-dos.com/6.22/64k/index-fastcomp.html > > > > Also you can download binaries here: > > https://github.com/caiiiycuk/js-dos/tree/gh-pages/6.22/latest-fastcomp > > https://github.com/caiiiycuk/js-dos/tree/gh-pages/6.22/latest-upstream > > > > Where > > > > пн, 22 июл. 2019 г. в 18:15, Александр Гурьянов <[email protected]>: > > > > > > Even digger works very poor here http://js-dos.com/ (in chrome). I > > > just changed emscripten_sleep_with_yield with emscripten_sleep and > > > that's all. Don't know why Chrome goes crazy. > > > > > > пн, 22 июл. 2019 г. в 18:04, Александр Гурьянов <[email protected]>: > > > > > > > > Just updated js-dos to latest upsteram, now you can compare > performance: > > > > http://js-dos.com/6.22/64k/index.html - for Asyncify > > > > http://js-dos.com/6.22/64k/index-fastcomp.html - for Emterpreter > > > > > > > > In FF performance is same, but in Chrome asyncify version looks > > > > significant slowly. Please give a try. > > > > > > > > пт, 19 июл. 2019 г. в 20:43, Alon Zakai <[email protected]>: > > > > > > > > > > > > > > > > > > > > On Thu, Jul 18, 2019 at 11:52 PM Александр Гурьянов < > [email protected]> wrote: > > > > >> > > > > >> Works like a charm! Even file size is smaller on 300Kb (2.8Mb > instead > > > > >> of 3.1Mb) > > > > > > > > > > > > > > > Great! > > > > > > > > > >> - Is Asyncify have simmilar conecpt of normal state like > emterpreter > > > > >> (EmterpreterAsync.state === 0)? > > > > > > > > > > > > > > > Yes, there is Asyncify.state, and 0 means "normal" there too. > > > > > > > > > >> - Is EMTERPRETIFY_WHITELIST (list of functions that neet to be > > > > >> emterpret) can be useful for Asyncify? I tried -s > > > > >> ASYNCIFY_IGNORE_INDIRECT output size is 1,8M, but didn't work > > > > >> > > > > > > > > > > We are considering that, see > > > > > > > > > > https://github.com/WebAssembly/binaryen/issues/2218 > > > > > > > > > > But the overhead tends to be fairly small, and I hope we can > optimize it more without adding manual lists. > > > > > > > > > > Sounds like you see a 55% code size increase - I'm also curious > what is the slowdown, if you can measure that? > > > > > > > > > > - Alon > > > > > > > > > >> чт, 18 июл. 2019 г. в 23:28, Alon Zakai <[email protected]>: > > > > >> > > > > > >> > You may need to do "emsdk update" instead of "emsdk > update-tags", depending on if you get the emsdk using git or not. > > > > >> > > > > > >> > Latest version should be 1.38.39. > > > > >> > > > > > >> > About those errors - yes, the APIs for old Asyncify, > Emterpreter, and new Asyncify are not identical. Emterpreter has the > "yield" concept, which is not needed in Asyncify (so as you said, just call > emscripten_sleep instead of _with_yield). Module.setAsync is an internal > API of old Asyncify - if you have custom JS code using it, you need to > update it to the new Asyncify's API (see notes on handleSleep in the > blogpost: https://kripken.github.io/blog/wasm/2019/07/16/asyncify.html ). > > > > >> > > > > > >> > On Thu, Jul 18, 2019 at 1:01 AM Александр Гурьянов < > [email protected]> wrote: > > > > >> >> > > > > >> >> I am not sure that I used latest emscripten (through emsdk), > don't > > > > >> >> know how to check. > > > > >> >> > > > > >> >> emcc (Emscripten gcc/clang-like replacement) 1.38.37 (commit > > > > >> >> d33f7a29002a2463a7956cd53a559b97a52f9560) > > > > >> >> Copyright (C) 2014 the Emscripten authors (see AUTHORS.txt) > > > > >> >> This is free and open source software under the MIT license. > > > > >> >> There is NO warranty; not even for MERCHANTABILITY or FITNESS > FOR A > > > > >> >> PARTICULAR PURPOSE. > > > > >> >> > > > > >> >> My debug log didn't contain wasm-opt --asyncify which should > be I think. > > > > >> >> > > > > >> >> Btw, I used this to update emscripten: > > > > >> >> emsdk update-tags && emsdk install latest-upstream && emsdk > activate > > > > >> >> -embedded latest-upstream > > > > >> >> > > > > >> >> чт, 18 июл. 2019 г. в 14:52, Александр Гурьянов < > [email protected]>: > > > > >> >> > > > > > >> >> > After replacing emscripten_sleep_with_yield on > emscripten_sleep I have > > > > >> >> > new error: > > > > >> >> > > > > > >> >> > exception thrown: TypeError: Module.setAsync is not a > > > > >> >> > function,_emscripten_sleep@http://127.0.0.1:8080/js-dos.js > line 1 > > > > > >> >> > eval:8:182429 > > > > >> >> > GUI_StartUp(Section*)@http://127.0.0.1:8080/js-dos.js line > 1 > > > > > >> >> > WebAssembly.compile:wasm-function[3244]:0x11dccf > > > > >> >> > Section::ExecuteInit(bool)@http://127.0.0.1:8080/js-dos.js > line 1 > > > > > >> >> > WebAssembly.compile:wasm-function[1025]:0x64869 > > > > >> >> > Config::Init()@http://127.0.0.1:8080/js-dos.js line 1 > > > > > >> >> > WebAssembly.compile:wasm-function[3634]:0x1339dd > > > > >> >> > main@http://127.0.0.1:8080/js-dos.js line 1 > > > > > >> >> > WebAssembly.compile:wasm-function[2183]:0xc6c59 > > > > >> >> > WDOSBOX/Module._main@http://127.0.0.1:8080/js-dos.js line 1 > > eval:8:220000 > > > > >> >> > callMain@http://127.0.0.1:8080/js-dos.js line 1 > > eval:8:243221 > > > > >> >> > main@http://127.0.0.1:8080/js-dos.js:1:24720 > > > > >> >> > @http://127.0.0.1:8080/index.html:21:9 > > > > >> >> > > > > > >> >> > чт, 18 июл. 2019 г. в 14:47, Александр Гурьянов < > [email protected]>: > > > > >> >> > > > > > > >> >> > > Hi. Trying to compile dosbox with new latest-upstream. > It's compiles > > > > >> >> > > without error, but when I run it, I have this error > > > > >> >> > > > > > > >> >> > > missing function: emscripten_sleep_with_yield > > > > >> >> > > > > > > >> >> > > I just replaced all emterpreter related flags with > > > > >> >> > > -s ASYNCIFY > > > > >> >> > > > > >> >> -- > > > > >> >> You received this message because you are subscribed to the > Google Groups "emscripten-discuss" group. > > > > >> >> To unsubscribe from this group and stop receiving emails from > it, send an email to [email protected]. > > > > >> >> To view this discussion on the web visit > https://groups.google.com/d/msgid/emscripten-discuss/CAKOm%3DVG2PF2MET9GXVeRZTGvSTMMRa5RtawkzMRYvO6rvo1T-A%40mail.gmail.com > . > > > > >> >> For more options, visit https://groups.google.com/d/optout. > > > > >> > > > > > >> > -- > > > > >> > You received this message because you are subscribed to the > Google Groups "emscripten-discuss" group. > > > > >> > To unsubscribe from this group and stop receiving emails from > it, send an email to [email protected]. > > > > >> > To view this discussion on the web visit > https://groups.google.com/d/msgid/emscripten-discuss/CAEX4NpSciPtuE4ry3V108OC9iZfy4%2BeEvrCYfd2Ht9UTm%2BPrtA%40mail.gmail.com > . > > > > >> > For more options, visit https://groups.google.com/d/optout. > > > > >> > > > > >> -- > > > > >> You received this message because you are subscribed to the > Google Groups "emscripten-discuss" group. > > > > >> To unsubscribe from this group and stop receiving emails from it, > send an email to [email protected]. > > > > >> To view this discussion on the web visit > https://groups.google.com/d/msgid/emscripten-discuss/CAKOm%3DVFt2fLPxnZJLoapy3WoLP7Yt%2Ba85SthTLPktwRsfCoJHw%40mail.gmail.com > . > > > > >> For more options, visit https://groups.google.com/d/optout. > > > > > > > > > > -- > > > > > You received this message because you are subscribed to the Google > Groups "emscripten-discuss" group. > > > > > To unsubscribe from this group and stop receiving emails from it, > send an email to [email protected]. > > > > > To view this discussion on the web visit > https://groups.google.com/d/msgid/emscripten-discuss/CAEX4NpQsNRR1ydC4GLVjjkYvjJ8Z5R%3DaF4FSLzYXUPRqwMao7w%40mail.gmail.com > . > > > > > For more options, visit https://groups.google.com/d/optout. > > -- > You received this message because you are subscribed to the Google Groups > "emscripten-discuss" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To view this discussion on the web visit > https://groups.google.com/d/msgid/emscripten-discuss/CAKOm%3DVEqX2W0Kvym32VYfBqPu6qXMzrffkQEXfaNWs2yCLciQQ%40mail.gmail.com > . > -- You received this message because you are subscribed to the Google Groups "emscripten-discuss" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/emscripten-discuss/CAEX4NpS_v24602kd5t1mm9MkAaj4LydBmFc7xi%2BnAck4nZzNXw%40mail.gmail.com.
