11:30marshmellow: hi all. i'm trying to learn more about GPUs and very new to this world. i don't quite understand how different pieces like X11, Wayland, nouveau, 2D/3D acceleration on firefox, the proprietary nvidia drivers and the "Nvidia X Server" fit together. could anyone recommend good reading materials to learn more?
11:43linkmauve: marshmellow, maybe you’d be interested in this blog post? https://blog.mecheye.net/2016/01/dri/
11:49marshmellow: linkmauve: thanks, this looks like a good blog to read
15:31karolherbst: imirkin: if end is 0 in nvc0_set_global_bindings we can skip the entire function, no?
15:33karolherbst: or well.. actually if nr is 0
15:33imirkin: could mean "reset everything"
15:34karolherbst: I have this with clover when all inputs are SVM pointers
15:34karolherbst: so we actually have 0 buffers
15:34karolherbst: but clover also calls "q.pipe->set_global_binding(q.pipe, 0, exec.g_buffers.size(), NULL, NULL)" for... no reason at all?
15:34karolherbst: dunno how much "reset everything" makes sense
15:34karolherbst: you can keep stale state and just update what you care about
15:35karolherbst: at least this is usually the kind of opt mareko seems to be working on if he comes up with perf opts
15:35karolherbst: this clear everthing makes no sense... clover sets the state before launching anyway.. mhhh
15:37karolherbst: imirkin: it's documented actually
15:38karolherbst: "Any buffers that were previously bound between [first, first + count - 1] are unbound after this call."
15:38karolherbst: I think the intention was that it's never called with count 0 :)
15:39karolherbst: and then resources being null unbounds the new ones, etc...
15:39karolherbst: well.. clears the slot rather
15:39karolherbst: it's more of a "nothing new gets bound" rather than "unbind if it's null"
15:41karolherbst: anyway.. this is a clover only function
15:41karolherbst: so.. I just decide what it means now :p
16:13pmoreau: Yeah, I think similar is happening for samplers as well. Had to handle that case even though I am not using any samplers or textures.
16:18karolherbst: all SVM test are passing now :)
16:18karolherbst: they got rid of the generic pointer req
16:18karolherbst: uhm.. wait
16:18karolherbst: it skipped all tests
16:21imirkin: best way to pass them
16:21imirkin: most reliable
16:21imirkin: doesn't need hw either
16:21karolherbst: there is a bug
16:22karolherbst: I wish we would rate limit our prints