[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Updates to rocfft and hipfft



Hi Kari,

Thank you for your help.

On June 28, 2024 4:02:47 p.m. GMT+02:00, Kari Pahula <kaol@debian.org> wrote:
>Upstream git had only tags for ROCm 6.1.2 for the projects.  I didn't
>want to pick a version number that wasn't there so I went with that
>instead.

It seems that ROCm 6.1.3 was a Windows release and those are not being tagged. There's a bit of weirdness because the Windows and Linux releases are managed by different groups.

>I'm thinking just go with unstable
>and not bother with experimental.

Sounds good.

>However, what's your opinion on testing? I ran tests but my 8GB GPU
>couldn't do all of them.  Would someone having more memory give them a
>spin?  Moreover, a new feature with these is support for multi-GPU
>operation, I don't know if anyone has that available for testing.

I believe you have been granted access to Pinwheel, which has two MI210 GPUs with 64GB of VRAM each. If not, I can arrange it.

>A funny detail about the multi-GPU support: at least for rocfft's
>test, it made it try to run it on my GPU as well as the Ryzen 9
>7900X's internal GPU.  Do the developers just use Threadrippers for
>everything?

Yes. They were using Ryzen for dev machines during the Zen 2 era, then switched to Threadripper for Zen 3. There are also Xeon and EPYC test machines.

AFAIK, there are no upstream developers using Zen 4 Ryzen CPUs, which I think results in a poor experience for most desktop users. I'm glad Debian is highlighting these issues, as they are relatively simple problems but are confusing and frustrating for novice users.

The problem you encountered also affects rocblas. I'm not sure what should be done about it. What would the ideal behaviour be?

>Doxygen stuff didn't quite work as expected out of the box and I had
>to wrangle with that for both projects.  I made one PR to upstream's
>hipFFT git.  Shortly put, TYPEDEF_HIDES_STRUCTS = YES means that only
>typedef names are available and struct and enum names are hidden and
>there was some confusion about that.

Thank you for the fix, and especially for reporting the issues upstream.

Sincerely,
Cory Bloor

Reply to: