On Mon, Aug 11, 2025 at 6:08 PM Alexandre Demers
<alexandre.f.dem...@gmail.com> wrote:
>
> Hi,
>
> For those who know, I'm still working on VCE1 enablement under AMDGPU.
> Progress is happening, slowly but surely. While investigating the ring init
> calls, a few elements catched my attention and I'd like some help in figuring
> out the differences between AMDGPU's ring sizes and Radeon's ones.
>
> 1- I understand that the size parameter changed from bytes under
> radeon_ring_init to dword under amdgpu_ring_init. That being said, some
> values don't seem to be equivalent between Radeon and AMDGPU. For example,
> GFX ring size went from 1024 * 1024 bytes to 1024 dwords (for most GFX
> versions), which seems off even when taking into account how amdgpu_ring_init
> calculates the final allocated size. This question is more about
> understanding than a problem strictly speaking.
You can make the ring any size you want. It's specified in the queue
descriptor. I don't remember why we picked the sizes we did in radeon
off hand.
>
> 2- Under AMDGPU, SI's GFX (GFX6) ring size is 2048, while this value is 1024
> for all the other GFX versions. Under Radeon, the GFX ring size values are
> all the same (1024 * 1024) under Evergreen/SI/CIK/NI and others. Is there any
> reason why SI's GFX6 ring size would be twice the size of the other values
> under AMDGPU?
>
>From the git history:
commit 97041ed37718dc9ba30aa23ca74093dc93ac89fb
Author: Bas Nieuwenhuizen <b...@basnieuwenhuizen.nl>
Date: Thu Apr 13 16:22:51 2023 +0200
drm/amdgpu: Increase GFX6 graphics ring size.
To ensure it supports 192 IBs per submission, so we can keep a
simplified IB limit in the follow up patch without having to
look at IP or GPU version.
Reviewed-by: Christian König <christian.koe...@amd.com>
Signed-off-by: Bas Nieuwenhuizen <b...@basnieuwenhuizen.nl>
Signed-off-by: Alex Deucher <alexander.deuc...@amd.com>
> 3- Would it be acceptable to add names to rings under Radeon, the same names
> as the ones used under AMDGPU? I think it is more talkative for the average
> user and for debugging purposes to deal with ring names than indexes. I
> already have patches in my code to address this suggestion.
>
If you think there is value there. It seems like it would generate a
lot of churn in that driver for little gain.
Alex
> That's all for now.
>
> Cheers
> Alexandre Demers
