Ok, the main code is in src/proxy*.js. You can see how basic canvas
operations are forwarded there.

The proxying there listens to DOM events and sends them over. For SDL
audio, there are a few options and I am not sure which is best. One is to
proxy at the C level - replace MIX_* calls in the worker with functions
that proxy over. Or, another is to proxy at the HTML Audio object, replace
it with a shim that proxies. First step is to look at both of those
options, I think.

- Alon



On Sun, Dec 29, 2013 at 6:53 PM, Richard Janicek <[email protected]> wrote:

> Any guidelines you could provide would be much appreciated. I will attempt
> to implement.
>
>
>
> On Sun, Dec 29, 2013 at 9:30 PM, Alon Zakai <[email protected]> wrote:
>
>> Yes, audio is possible as well. I don't have plans to work on this myself
>> but happy to give guidelines if anyone else wants to.
>>
>> - Alon
>>
>>
>>
>> On Sun, Dec 29, 2013 at 4:54 PM, Richard Janicek <[email protected]>wrote:
>>
>>> Hello,
>>>
>>> I am testing --proxy-to-worker with VICE.js. Looks like it's getting
>>> +10% performance boost on Chrome. Very nice.
>>>
>>> Would it be possible to also proxy SDL audio?
>>>
>>> Best regards,
>>> ~Richard Janicek
>>>
>>> On Saturday, September 14, 2013 12:37:21 AM UTC-4, azakai wrote:
>>>
>>>> Close to that, yes. SDL_Delay can be implemented as a busy-wait both on
>>>> the main thread and in workers, but on the main thread _it blocks
>>>> rendering_. On workers, it just blocks event handling; the worker can still
>>>> send out paint events.
>>>>
>>>> This actually happens in Doom, it does screen wipe effects
>>>> synchronously. Those are not shown when running on the main thread, because
>>>> nothing renders until the event loop is exited. But when running in a
>>>> worker, they are shown perfectly. Input events can't be handled during that
>>>> time, but in that case it doesn't matter - this might be a problem in other
>>>> ones.
>>>>
>>>> I see it refresh the canvas very quickly, and this is without much
>>>> optimization (no typed array transfers, for example). No reason it can't be
>>>> 60fps.
>>>>
>>>> - Alon
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> On Fri, Sep 13, 2013 at 6:34 PM, Richard Janicek <[email protected]>wrote:
>>>>
>>>>>  Oh wow, I think this may work for my app. Does this mean it is
>>>>> possible to allow synchronous blocking flow control in the worker thread
>>>>> (SDL_Delay) ? How fast can it refresh the canvas, is 60 fps possible?
>>>>>
>>>>> Thank you very much for this!
>>>>>
>>>>> § Richard Janicek <http://janicek.co> ¦ mappur <http://mappur.com> ¦
>>>>> player <http://player.janicek.co/#player> ¦ 
>>>>> clock2d<http://www.clock2d.com/>
>>>>>
>>>>>
>>>>> On Fri, Sep 13, 2013 at 8:47 PM, Alon Zakai <[email protected]>wrote:
>>>>>
>>>>>>  If you use HTML canvas commands like lines, shapes, fills, etc.,
>>>>>> all those need a 2D canvas context, and do not work in workers (yet). 
>>>>>> Right
>>>>>> now what works is if all you do is render pixels to a buffer and do
>>>>>> SDL_UnlockSurface to push them to the screen.
>>>>>>
>>>>>> But this is temporary, there is spec work going on to get canvas and
>>>>>> WebGL to workers, it will happen eventually, see e.g.
>>>>>>
>>>>>> https://bugzilla.mozilla.org/show_bug.cgi?id=801176
>>>>>>
>>>>>> - Alon
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Fri, Sep 13, 2013 at 5:44 PM, Richard Janicek <[email protected]>wrote:
>>>>>>
>>>>>>> What do you mean by "software renderer" ? My app 
>>>>>>> (VICE.js<http://vice.janicek.co/>)
>>>>>>> renders frames to HTML canvas, does this mean I can't take advantage of
>>>>>>> this new technique?
>>>>>>>
>>>>>>> Thank you,
>>>>>>>
>>>>>>> § Richard Janicek <http://janicek.co> ¦ mappur <http://mappur.com>
>>>>>>>  ¦ clock2d <http://www.clock2d.com/>
>>>>>>>
>>>>>>>
>>>>>>> On Fri, Sep 13, 2013 at 7:49 PM, Alon Zakai <[email protected]>wrote:
>>>>>>>
>>>>>>>> 1. Proxing to/from a worker. With
>>>>>>>>
>>>>>>>> emcc --proxy-to-worker
>>>>>>>>
>>>>>>>> the compiler will generate html and js files, where the html is a
>>>>>>>> small shim that just proxies events to the js, which runs in a worker, 
>>>>>>>> and
>>>>>>>> the js sends back the SDL screen which the html then renders. So 
>>>>>>>> basically
>>>>>>>> this lets you run simple SDL apps in a worker, which is usually good 
>>>>>>>> for
>>>>>>>> performance (doing less on the main thread).
>>>>>>>>
>>>>>>>> This just supports code that does not use HTML canvas or WebGL,
>>>>>>>> because those don't run in workers in browsers (yet). So it will work 
>>>>>>>> right
>>>>>>>> now for a software renderer, for example. I've tested this on Doom.
>>>>>>>>
>>>>>>>> 2. emscripten_async_load_script (see emscripten.h)
>>>>>>>>
>>>>>>>> This loads a url into a newly-created HTML script element. This is
>>>>>>>> the most efficient way to load a new large amount of code into your app
>>>>>>>> (emscripten_run_script converts binary to text, then evals, which is
>>>>>>>> worse). It also integrates with the run dependencies system, which 
>>>>>>>> means
>>>>>>>> that you can generate code to preload some files using the file 
>>>>>>>> packager,
>>>>>>>> and emscripten_async_load_script of that will asynchronously load the
>>>>>>>> script as well as fetch the file data and prepare it; the callback will
>>>>>>>> happen when all the data is ready. This can be used e.g. to load the 
>>>>>>>> next
>>>>>>>> level in your game, at the right time, and in general it can make it 
>>>>>>>> easier
>>>>>>>> to avoid preloading everything at the beginning, which can cause a
>>>>>>>> noticeable pause.
>>>>>>>>
>>>>>>>> - Alon
>>>>>>>>
>>>>>>>>  --
>>>>>>>> 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].
>>>>>>>> For more options, visit https://groups.google.com/groups/opt_out.
>>>>>>>>
>>>>>>>
>>>>>>>  --
>>>>>>> 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].
>>>>>>> For more options, visit https://groups.google.com/groups/opt_out.
>>>>>>>
>>>>>>
>>>>>>  --
>>>>>> 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].
>>>>>> For more options, visit https://groups.google.com/groups/opt_out.
>>>>>>
>>>>>
>>>>>  --
>>>>> 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].
>>>>> For more options, visit https://groups.google.com/groups/opt_out.
>>>>>
>>>>
>>>>  --
>>> 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].
>>> For more options, visit https://groups.google.com/groups/opt_out.
>>>
>>
>>  --
>> 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].
>> For more options, visit https://groups.google.com/groups/opt_out.
>>
>
>  --
> 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].
> For more options, visit https://groups.google.com/groups/opt_out.
>

-- 
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].
For more options, visit https://groups.google.com/groups/opt_out.

Reply via email to