[PATCH 1/2] drm/amdgpu: Enable scatter gather display support

[Christian König]

That's what I meant with use up the otherwise unused VRAM. I don't see 
any disadvantage of always setting GTT as second domain on APUs.

My assumption was that we dropped this in userspace for displayable 
surfaces, but Marek proved that wrong.

So what we should do is actually to add GTT as fallback to all BOs on 
APUs in Mesa and only make sure that the kernel is capable of handling 
GTT with optimal performance (e.g. have user huge pages etc..).

Christian.

Am 20.03.2018 um 21:04 schrieb Deucher, Alexander:
>
> I think we can also have the case of systems with similar amounts of 
> carve out and system ram.  E.g., on a system with 4 GB of system 
> memory with 1 GB carved out for vram.  It would be a big waste not to 
> use VRAM.  We'll probably need a heuristic at some point.
>
>
> Alex
>
> ------------------------------------------------------------------------
> *From:* Christian König <ckoenig.leichtzumerken at gmail.com>
> *Sent:* Tuesday, March 20, 2018 2:32:49 PM
> *To:* Marek Olšák; Koenig, Christian
> *Cc:* Alex Deucher; Deucher, Alexander; Michel Dänzer; Li, Samuel; 
> amd-gfx list
> *Subject:* Re: [PATCH 1/2] drm/amdgpu: Enable scatter gather display 
> support
> I don't think that is a good idea.
>
> Ideally GTT should now have the same performance as VRAM on APUs and 
> we should use VRAM only for things where we absolutely have to and to 
> actually use up the otherwise unused VRAM.
>
> Can you run some tests with all BOs forced to GTT and see if there is 
> any performance regression?
>
> Christian.
>
> Am 20.03.2018 um 15:51 schrieb Marek Olšák:
>> On Tue, Mar 20, 2018 at 9:55 AM, Christian König 
>> <ckoenig.leichtzumerken at gmail.com 
>> <mailto:ckoenig.leichtzumerken at gmail.com>> wrote:
>>
>>     Yes, exactly. And if I remember correctly Mesa used to always set
>>     GTT as fallback on APUs, correct?
>>
>>
>> "used to" is the key part. Mesa doesn't force GTT on APUs anymore. It 
>> expects that the combination of BO priorities and BO move throttling 
>> will result in optimal BO placements over time.
>>
>> Marek
>>
>>
>>     The problem seems to be that this fallback isn't set for
>>     displayable BOs.
>>
>>     So what needs to be done is to just enable this fallback for
>>     displayable BOs as well if the kernel can handle it.
>>
>>     Christian.
>>
>>
>>     Am 20.03.2018 um 00:01 schrieb Marek Olšák:
>>>     In theory, Mesa doesn't have to do anything. It can continue
>>>     setting VRAM and if the kernel has to put a display buffer into
>>>     GTT, it doesn't matter (for Mesa). Whether the VRAM placement is
>>>     really used is largely determined by BO priorities.
>>>
>>>     The way I understand scather/gather is that it only allows the
>>>     GTT placement. It doesn't force the GTT placement. Mesa also
>>>     doesn't force the GTT placement.
>>>
>>>     Marek
>>>
>>>     On Mon, Mar 19, 2018 at 5:12 PM, Alex Deucher
>>>     <alexdeucher at gmail.com <mailto:alexdeucher at gmail.com>> wrote:
>>>
>>>         On Mon, Mar 19, 2018 at 4:29 PM, Li, Samuel
>>>         <Samuel.Li at amd.com <mailto:Samuel.Li at amd.com>> wrote:
>>>         >>to my earlier point, there may be cases where it is
>>>         advantageous to put
>>>         >> display buffers in vram even if s/g display is supported
>>>         >
>>>         > Agreed. That is also why the patch has the options to let
>>>         user select where
>>>         > to put display buffers.
>>>         >
>>>         > As whether to put the option in Mesa or kernel, it seems
>>>         the difference is
>>>         > not much. Also, since amdgpufb can request even without
>>>         mesa, kernel might
>>>         > be a better choice. In addition, putting in the kernel can
>>>         save client’s
>>>         > duplicate work(mesa, ogl, vulkan, 2d, kernel…)
>>>
>>>         Why do we even expose different memory pools to the UMDs in
>>>         the first
>>>         place ;)  Each pool has performance characteristics that may be
>>>         relevant for a particular work load. Only the UMDs really
>>>         know the
>>>         finer points of those workloads. In general, you don't want
>>>         the kernel
>>>         dictating policy if you can avoid it. The kernel exposes
>>>         functionality and userspace sets the policy.  With the
>>>         location set in
>>>         userspace, each app/user can have whatever policy makes
>>>         sense for
>>>         their use case all at the same time without needing to tweak
>>>         their
>>>         kernel for every use case.
>>>
>>>         Alex
>>>
>>>
>>>
>>>     _______________________________________________
>>>     amd-gfx mailing list
>>>     amd-gfx at lists.freedesktop.org <mailto:amd-gfx at lists.freedesktop.org>
>>>     https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>>>     <https://lists.freedesktop.org/mailman/listinfo/amd-gfx>
>>
>>
>>
>>
>> _______________________________________________
>> amd-gfx mailing list
>> amd-gfx at lists.freedesktop.org <mailto:amd-gfx at lists.freedesktop.org>
>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20180320/2532078d/attachment.html>