13:29haasn: how stupid is it to create a unique VkCommandPool for every VkCommandBuffer
13:30haasn: or would I be better just taking a lock at that point
13:32ishitatsuyuki: haasn: since it's #radeon, a command pool is basically no-op on radv
13:32ishitatsuyuki: but do remember to reuse the command buffers themselves via reset, because creating each of them do a GPU-side allocation which is costly
13:36haasn: right, I can't imagine other drivers are too different here, apparently vulkano does the same thing (one cmdpool per cmdbuf)
13:37haasn: I suppose I can implement it in a way where I can test both options without too much pain
13:37haasn: problem is that I have an application where the conventional wisdom of having a commandpool per frame per thread doesn't work, because due to API limitations I can't access any thread-specific struct from the frame-specific struct
13:37haasn: but I also need concurrent access to the same frame from multiple threads, so
16:25Caterpillar: 2 over 2 AMD GPU machines I own (RX480 and 680M), they both have driver problems that lead to loss of data by simply watching a video in the browser. Damn. ( https://gitlab.freedesktop.org/drm/amd/-/issues/2220 and https://gitlab.freedesktop.org/drm/amd/-/issues/2272 )
16:32Caterpillar: I just found out that there is only one developer in such repository. Damn
17:59Caterpillar: I was wrong, the commits on Gitlab are not about drivers source code. The commits are on Github, and there there are plenty of developers
17:59Caterpillar: https://github.com/torvalds/linux/commits/master?before=3ca112b71f35dd5d99fc4571a56b5fc6f0c15814+35&branch=master&path%5B%5D=drivers&path%5B%5D=gpu&path%5B%5D=drm&path%5B%5D=amd&path%5B%5D=amdgpu&qualified_name=refs%2Fheads%2Fmaster
23:30kode54: which specific video
23:30kode54: also are you using any addons in firefox like that one that makes youtube do a smart blur extended around the video frame
23:30kode54: I had an issue with that one causing at least firefox to crash on this one ancient 320x240 video