On Wed, Jun 10, 2020 at 10:23:15PM +0200, Paul Menzel via blfs-dev wrote:
> Dear Beyond Linux From Scratch folks,
>
>
> Mozilla Firefox bug #1644610 ([X11] Implement ffmpeg/VAAPI video playback,
> pixmap-based image sharing, basic compositor) [1] has three patches to
> support video playback using FFmpeg/VA-API.
>
> Maybe we want to add a note for those, as it’ll save some energy and reduce
> noise, if the GPU can be used for that.
>
> From Rinat:
>
> > (In reply to Jean-Yves Avenard [:jya] from comment #6)
> >
> > There's no gain in performance as the GPU readback eats all the gain
> > made by the HW decoder.
> >
> > Basic, no accel, jumps from 3.4 W to 4.0 W, with about 3.5 W most of the
> > time
> > Basic, with accel, jumps from 2.5 W to 3.1 W, with about 2.7 W most of the
> > time.
> >
> > There is a clear power consumption benefit.
> >
> > Also, not for the sake of debates, but for comparison:
> > WebRender, no accel, jumps from 4.4 W to 4.7 W, with about 4.5 W most of
> > the time.
>
>
> Kind regards,
>
> Paul
>
>
> [1]: https://bugzilla.mozilla.org/show_bug.cgi?id=1644610
WebRender ? I think that is not used on any of my machines - ISTR
it gets enabled for some nvidia chips, but I have read that modern
nvidia GPUs tend to work poorly in nouveau. I bought a low-end
nvidia card a couple of years ago to bring up a first-generation
ryzen, it turned out to be very problematic in the tty (flooding the
syslog) and more so in X (often locking up, or really flooding the
syslog).
Based on the comments in that bug, I don't think it is going
anywhere soon. I've no idea what tree the patches are against
(probably nightly, wherever that is kept in mercurial?) and I have
no interest in it.
If someone wants to provide a headed patch which applies to the
version in the book, that can be uploaded. But like some other
patches, I see no point in mentioning it in the book unless someone
is going to regularly test it and upload fixes for newer versions.
I intend to move the book to firefox-78.0 (the next ESR series) when
that is released at the end of this month.
The system-graphite2-harfbuzz patch had an obvious benefit (use
system libs), but maintaining it was a lot of work and it died after
about firefox-73. I'm dubious about the reported benefits from this
set of patches.
While I understand 'save energy' in terms of an end-user running
something on a laptop, for packages which we ourselves compile I'm
not sure that there is a positive balance between making multiple
compiles to test with/without the patch, and actually running it.
And the figures quoted above appear to be averages _with_ the
patches, but not the corresponding averages from the unpatched
build.
And I don't understand what you mean by 'reduce noise'. Do you mean
that a specific video looks noisy without these patches but less
noisey with them applied ? Or do you mean that the audio sounds
noisy without these patches ? Details of your own experience in
using these patches (together with the hardware, toolchain versions
and whether you used clang{,++} rather than gcc,g++, and package
versions) would be helpful in letting people know if this is likely
to be beneficial.
ĸen
--
+++ OUT OF CHEESE ERROR. REDO FROM START +++
--
http://lists.linuxfromscratch.org/listinfo/blfs-dev
FAQ: http://www.linuxfromscratch.org/blfs/faq.html
Unsubscribe: See the above information page
