[bug#38403] gnu: Add intel-vaapi-driver-g45-h264.

Hey Guix,

I am not sure if this package offers any practical benefit or not, so I
defer to someone who actually knows what they are talking about to make
a judgment call ;-).

This variant of intel-vaapi-driver is a backport of the ancient g45-h264
branch that used to be maintained by the intel-vaapi-driver team. As far
as I know, the reason they don't maintain this branch anymore is that
hardware accelerated h264 decoding on the G45 chipset was much too slow
for 1080p video, as by default it only had 32MB of VRAM to work with.

It so happens that the librebooted Thinkpad T400 has one of these
G45-based iGPUs, with one change: the iGPU has 256MB VRAM
available. Using this package and a `hwdec=vaapi' setting in
~/.config/mpv/mpv.conf, I enjoy smooth video playback and low CPU usage
for 1080p video using mpv.

Jelle,

Jelle Licht 写道：
> I am not sure if this package offers any practical benefit or 
> not, so I
> defer to someone who actually knows what they are talking about 
> to make
> a judgment call ;-).

I don't have a G45, but if this driver verifiably works for you 
(and the regular version does not) I think it's good to to include 
it.  Also because I expect this card to be overrepresented amongst 
Guix users: there aren't that many common Librebootable machines.

It needs its own clear synopsis, description, and home page 
though.  Is there a home page?  I'm not good at navigating 
BitBucket.

> This variant of intel-vaapi-driver is a backport of the ancient 
> g45-h264
> branch that used to be maintained by the intel-vaapi-driver 
> team. As far
> as I know, the reason they don't maintain this branch anymore is 
> that
> hardware accelerated h264 decoding on the G45 chipset was much 
> too slow
> for 1080p video, as by default it only had 32MB of VRAM to work 
> with.

These devices don't have their own RAM chips, so ‘by default’ here 
means the BIOS menu setting, right?  If so, it should be mentioned 
in the description for lack of upstream documentation.

I'm guessing that your card identifies as an Intel® GMA X4500MHD. 
Is that correct?  We should include the marketing name in the 
synopsis & description too, even if it might not map 1:1 to the 
chipset.

> This variant of intel-vaapi-driver is a backport of the ancient 
> g45-h264
> branch that used to be maintained by the intel-vaapi-driver 
> team. As far
> as I know, the reason they don't maintain this branch anymore is 
> that
> hardware accelerated h264 decoding on the G45 chipset was much 
> too slow
> for 1080p video, as by default it only had 32MB of VRAM to work 
> with.

This is good! :-)  It should be slightly edited and used as a 
description together with the original intel-vaapi-driver text.

Apart from that, the package LGTM.  Thanks!

Kind regards,

T G-R

Hey Tobias,

Tobias Geerinckx-Rice via Guix-patches via <guix-patches@gnu.org>
writes:

> Jelle,
>
> Jelle Licht 写道：
>> I am not sure if this package offers any practical benefit or 
>> not, so I
>> defer to someone who actually knows what they are talking about 
>> to make
>> a judgment call ;-).
>
> I don't have a G45, but if this driver verifiably works for you 
> (and the regular version does not) I think it's good to to include 
> it.  Also because I expect this card to be overrepresented amongst 
> Guix users: there aren't that many common Librebootable machines.
>
> It needs its own clear synopsis, description, and home page 
> though.  Is there a home page?  I'm not good at navigating 
> BitBucket.

I am in contact with the upstream author (of this tarball): They do not
have access to a G45-based chipset, so we are currently looking into
options for the future. There is currently no home page.

The actual bitbucket repo has been made public only yesterday, so it was
not that weird that you could not find it :-).
>
>> This variant of intel-vaapi-driver is a backport of the ancient 
>> g45-h264
>> branch that used to be maintained by the intel-vaapi-driver 
>> team. As far
>> as I know, the reason they don't maintain this branch anymore is 
>> that
>> hardware accelerated h264 decoding on the G45 chipset was much 
>> too slow
>> for 1080p video, as by default it only had 32MB of VRAM to work 
>> with.
>
> These devices don't have their own RAM chips, so ‘by default’ here 
> means the BIOS menu setting, right?  If so, it should be mentioned 
> in the description for lack of upstream documentation.
This is correct. I think it makes sense to point people to the right
documentation for at least libreboot and coreboot, and refer them to
their BIOS supplier for other instructions.

OTOH, I do not feel that the package description is the right place to
do this. Where would such instructions normally exist on GNU/Linux
machines?

> I'm guessing that your card identifies as an Intel® GMA X4500MHD. 
> Is that correct?  We should include the marketing name in the 
> synopsis & description too, even if it might not map 1:1 to the 
> chipset.
Yes on all fronts.

- Jelle

Hello,

I have an old laptop (Vaio VGN-TT90S) with such a chipset,
that I can get out of the attic, if you need some testing.

Hey,

Vincent Legoll <vincent.legoll@gmail.com> writes:

> Hello,
>
> I have an old laptop (Vaio VGN-TT90S) with such a chipset,
> that I can get out of the attic, if you need some testing.

That would be nice! I am waiting until we remove all linux-libre kernels
between 5.0 and 5.5, as these lead to pretty bad crashes with this
package.

I have it on my agenda for early May to backport the latest
intel-vaapi-driver changes to the g45-h264 branch and subsequently push
it to Guix.

 - Jelle

On 13/04/2020 22:01, Jelle Licht wrote:
> I have it on my agenda for early May to backport the latest
> intel-vaapi-driver changes to the g45-h264 branch and subsequently push
> it to Guix.

Just ping me when needed

Necro tiem (thanks Vincent),

Jelle Licht 写道：
>> These devices don't have their own RAM chips, so ‘by default’ 
>> here 
>> means the BIOS menu setting, right?  If so, it should be 
>> mentioned 
>> in the description for lack of upstream documentation.
> This is correct. I think it makes sense to point people to the 
> right
> documentation for at least libreboot and coreboot, and refer 
> them to
> their BIOS supplier for other instructions.
>
> OTOH, I do not feel that the package description is the right 
> place to
> do this.

I agree it's not ideal.

> Where would such instructions normally exist on GNU/Linux 
> machines?

Well…  /share/doc.  I check it; I don't know how many people 
(still) do.

It seems to be less well known than it used to be, which is a 
shame, but it's technically the answer to your question.

Kind regards,

T G-R

Vincent Legoll <vincent.legoll@gmail.com> writes:

> On 13/04/2020 22:01, Jelle Licht wrote:
>> I have it on my agenda for early May to backport the latest
>> intel-vaapi-driver changes to the g45-h264 branch and subsequently push
>> it to Guix.
>
> Just ping me when needed

Some changes made in linux kernel 5.X made it harder to get this to work
reliably. Since getting this to work allows me to watch videos on my
laptop without melting my pants off, I'm still very much motivated to
get it to work in the long run ;-).

I been made aware that the 5.x series of kernels have seen a lot of
regressions (and improvements, probably) for intel-drm related code
paths; my hope is that this still somehow stabilizes in the coming
months.

For now, it seems to work on linux kernel 4.19.143. See [1] for the
maintainer of the AUR package that I based my work on.

Is there some way to 'park' this debbugs issue? As in, keep it open, but
rather as a reminder to revisit this issue in the future?

Thanks,
Jelle

[1] https://aur.archlinux.org/packages/libva-intel-driver-g45-h264

Still doesn't work on linux 5.15. Situation same as in comments to aur 
package: version 2.4.1 hangs gpu, 2.4.1-experimental doesn't work.