03:59 Lynne: most drivers spin up a shader for vkCmdFillBuffer, right?
04:00 Lynne: can that shader be assumed to be faster or as fast as the best users can manually do via compute shaders?
07:41 Kayden: probably so (and if it isn't, a bug report is likely warranted) - the driver can probably play tricks with the formats or compression that might not normally be allowed
10:27 dj-death: Lynne: Anv will either use a fragment or compute shader to avoid the cost of pipeline switches on most HW
10:28 dj-death: Lynne: writes will go through the render target cache with fragment shader
14:08 Lynne: thanks, my compute shader should be fast enough to not really warrant replacing it
14:09 Lynne: by the way, are you still interested in a minimal example to reproduce the ffv1 encoder issues I've had?
14:09 Lynne: the code got merged in ffmpeg last week, so it would be nice to get it running under anv
14:09 Lynne: lavapipe's also partially broken, but radv works greast
14:10 Lynne: *great
15:27 Lynne: does vulkan expose a limit for the maximum single allocation a user can ask for?
16:11 glehmann: Lynne: do you mean the largest single VkDeviceMemory? then VK_KHR_maintenance3/Vulkan1.1 maxMemoryAllocationSize
16:16 Lynne: yup, that, thanks
16:17 Lynne: weird its in 1.1
16:31 glehmann: it's a soft limit, if you try to go larger, the allocation will just fail and no UB will happen
16:32 glehmann: oh and ofc you might not be able to allocate maxMemoryAllocationSize because of OOM
20:35 HdkR: Looks like 24.3.0 breaks vulkaninfo on non-pi devices if the broadcom icd is installed. main doesn't have the issue anymore, but 24.3.0 is still an issue. This is the third time this has occurred over the past couple years, could a CI job install all mesa vulkan ICDs and run vulkaninfo or something so this stops happening?
20:51 dj-death: Lynne: definitely interested
20:54 kisak: HdkR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32266 looks like it barely missed the release timing.
20:55 HdkR: Nice
21:21 airlied: HdkR: fixes had cc stable, should make the next release I assume
21:21 HdkR: Nice
21:21 airlied: it definitely seems like something CI could do better at
21:22 HdkR: Luckily I've been pushing out updating my images to mesa 24.3, so waiting for a point release is easy :D
21:22 HdkR: Would be nice, it's bitten me quite a few times