08:24 <q4a> hi. Somebody knows: does first GPU with Valhall architecture is Mali-G77 ? Does rk3588 with Mali “Odin” MP4 GPU has Valhall architecture?

08:31 <q4a> ok. https://en.wikipedia.org/wiki/Mali_(GPU) says that Mali-G77 and Mali-G57 was "Valhall 1st gen", but question about Mali “Odin” from rk3588 - still remains

09:14 <alarumbe> jekstrand: o/

09:21 <q4a> jekstrand: Hi. What skills (or HW) I need to help with panvk? I have C++ skills, CI (mostly github actions, but also travis CI and AppVeyor), my "3D" skills is not so good, but I'm trying to improve it. I have rk3288 (with Mali T760) and I'm waiting for rk3588 Mali Odin.

09:22 <q4a> I also really want to understand what you are using to improve panvk: just the documentation from Khronos/ARM or you also checking how other Vulkan drivers work / are implemented?

09:24 <q4a> Or may be checking and understanding why some CTS test failed is helping?

09:40 <robmur01> q4a: RK3588's G610 is 2nd-gen Valhall ("v10" in developer-speak here)

09:44 <robmur01> at this point its biggest blocker is kernel/Mesa support for the new Command Stream Frontend

09:49 <q4a> I saw that some Valhall support just landed to mesa: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16035

09:50 <q4a> robmur01: where I can read about Command Stream Frontend? (docs, src,..?)

09:50 <q4a> or only reversing?

09:55 <robmur01> RE from the downstream kbase driver, I'm afraid

10:18 <icecream95> The ac

10:19 <icecream95> The actual command stream for most parts of the job descriptors has been reverse engineered, but the other parts of the command stream which do not correspond directly to GPU descriptors are still unknown

10:21 <icecream95> Most of what CSF does is just writing 4 or 6 byte values to an internal buffer that is copied somewhere for other parts of the GPU to read

10:24 <icecream95> For example, the command 0x0132ffdeff09380 would write the GPU address 0xffdeff09380 to word 0x32 of the buffer, which specifies the address of the blend descriptor

10:25 <icecream95> Erm, it'd be more like 0x0132ffdeff093800 and 0xffdeff093800

10:26 <icecream95> q4a: ^^

10:28 <q4a> Thanks!

10:32 <icecream95> Eventually we may decide to rewrite the firmware, so then some other format for specifying commands could be chosen. but it's likely that only the blob firmware will be supported initially

10:34 <icecream95> (Luckily I have not yet seen any indication that firmware is signed)

11:09 <q4a> A bit weird question, but who can fix rss feed on collabora.com ? https://validator.w3.org/feed/check.cgi?url=https%3A%2F%2Fwww.collabora.com%2Fnewsroom-rss-feed.rss

11:19 <daniels> q4a: fun, I've passed it on, thanks

11:32 <robmur01> FWIW, looking at all the stuff the CSF firmware does, I know *I* wouldn't want to try rewriting it :P

15:04 <jekstrand> alarumbe: Found more buffer offset bugs: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16216

15:06 <jekstrand> alyssa: ^^

15:06 <jekstrand> Ugh...

15:06 <jekstrand> Need to stop doing that. :)

15:06 <jekstrand> Also, bbrezillon ^^

17:40 <q4a> Is that correct, that rbug is not suitable for debugging panvk, because only applies to gallium drivers? Is there something similar that can be used for panvk?

17:44 <robclark> that is correct

20:06 <q4a> daniels: rss fixed, thanks!

20:10 <daniels> q4a: \o/