Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

[Audio Effect] Does Web Audio have a way to use HW acceleration path?

43 views
Skip to first unread message

Marco Chen

unread,
Apr 17, 2013, 8:42:18 AM4/17/13
to dev-...@lists.mozilla.org
Hi all,

As I know that Web Audio provides the capability of applying audio effect like equalizer, reverb..etc.
And is there any ABI for utilizing the HW capability like DSP core or advanced codec?

Thanks,
Sincerely yours.

-------------------------------------
Marco Chen
System Engineer in Firefox OS
Mozilla Taiwan

Tel: +886-2-87861100 # 352
mail: mc...@mozilla.com

Thinker K.F. Li

unread,
Apr 17, 2013, 9:07:37 AM4/17/13
to mc...@mozilla.com, dev-...@lists.mozilla.org
From: Marco Chen <mc...@mozilla.com>
Subject: [Audio Effect] Does Web Audio have a way to use HW acceleration path?
Date: Wed, 17 Apr 2013 05:42:18 -0700 (PDT)

> Hi all,
>
> As I know that Web Audio provides the capability of applying audio effect like equalizer, reverb..etc.
> And is there any ABI for utilizing the HW capability like DSP core or advanced codec?

OpenAL? Are you talking about a lower layer API for using HW?

>
> Thanks,
> Sincerely yours.
>
> -------------------------------------
> Marco Chen
> System Engineer in Firefox OS
> Mozilla Taiwan
>
> Tel: +886-2-87861100 # 352
> mail: mc...@mozilla.com
>
> _______________________________________________
> dev-media mailing list
> dev-...@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-media

Paul Adenot

unread,
Apr 17, 2013, 9:06:53 AM4/17/13
to dev-...@lists.mozilla.org
On 04/17/2013 02:42 PM, Marco Chen wrote:
> As I know that Web Audio provides the capability of applying audio effect like equalizer, reverb..etc.
> And is there any ABI for utilizing the HW capability like DSP core or advanced codec?

We are in the early stages of the implemention of Web Audio. It would be
certainly possible to use SIMD or even a specialized DSP chip on mobile,
but the focus is on implementing the standard rather than optimizing the
current implementation for the moment.

fwiw, we don't have equalizer (biquad) and reverb (convolver) support yet.

Cheers,
Paul.

Marco Chen

unread,
Apr 17, 2013, 9:17:25 AM4/17/13
to Paul Adenot, dev-...@lists.mozilla.org
Hi Paul,

Thanks for your reply first.

I know that Web Audio is on early stage now but just want to know whether
the architecture can allow this extension to be performed from Web Audio to HW Acceleration chip.

Sincerely yours.

Paul Adenot

unread,
Apr 17, 2013, 9:58:49 AM4/17/13
to dev-...@lists.mozilla.org
I can't think of anything on the architectural side that would prevent
us from optimizing Web Audio using a DSP or SIMD, because it boils down
to implementing pure signal processing functions using SIMD intrinsics
or assembly.

Simply implementing the specialized functions in a separate file and
calling the right function based on the chip capabilities (like we do
for SIMD) should be enough. iirc, we tried to group all the signal
processing functions in the same file, or could do so rather easily.

But maybe you are talking about something else, in which case, what kind
of chip do you have in mind?

Paul.

Sotaro Ikeda

unread,
Apr 22, 2013, 9:53:17 AM4/22/13
to Marco Chen, dev-...@lists.mozilla.org, Paul Adenot
Hi Marco,

There is one problem to handle "Acceleration chip" on Web Audio. Application's process do not have a privilege to directly access/load Acceleration chip.

sotaro

Andreas Gal

unread,
Apr 22, 2013, 9:58:21 AM4/22/13
to Sotaro Ikeda, dev-...@lists.mozilla.org, Marco Chen, Paul Adenot
As Chrome does for GL we can remote access to the audio hardware to a
privileged process (not necessarily the parent process).

Andreas

Sent from Mobile.

Ehsan Akhgari

unread,
Apr 22, 2013, 11:32:08 AM4/22/13
to Paul Adenot, dev-...@lists.mozilla.org
On 2013-04-17 9:58 AM, Paul Adenot wrote:
> I can't think of anything on the architectural side that would prevent
> us from optimizing Web Audio using a DSP or SIMD, because it boils down
> to implementing pure signal processing functions using SIMD intrinsics
> or assembly.

I have not looked into this at all, but using hardware support for
effects such as convolution would only be an option if the hardware
processing uses an algorithm with compatible results to the software
implementation (but this is of course not an issue for SIMD-ifying our
processing code.)

> Simply implementing the specialized functions in a separate file and
> calling the right function based on the chip capabilities (like we do
> for SIMD) should be enough. iirc, we tried to group all the signal
> processing functions in the same file, or could do so rather easily.

Yes, see
<http://mxr.mozilla.org/mozilla-central/source/content/media/AudioNodeEngine.cpp>.

Cheers,
Ehsan
0 new messages