01:52 alyssa: jadahl: "blacker than black" like in retrocomputing days =)
13:21 alyssa:is continuing to wish Gallium exposed shader linking to drivers
13:54 Venemo: alyssa: what kind of shader linking?
14:00 Venemo: There are a bunch of NIR linking helpers that you could use, not sure if that's what you're referring to
14:55 alyssa: Venemo: Ideally, I'd like to get the whole VS/GS/TCS/TES/FS chain together as a single CSO where possible, and an explicit "SSO in use" flag on the separate CSOs when not.
14:57 alyssa: For handling things like varyings correctly (instead of having to essentially relink the shade ron every draw)
15:02 alyssa: https://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/etnaviv/etnaviv_shader.c#n110 etna is already doing this sort of thing
15:08 alyssa: looks like vc4 may be as well
19:43 robclark: alyssa: not sure if it helps, but maybe look at ir3_cache (in the gallium driver).. we map the individual shaders to a single hw "state object".. (and note, only do the lookup if program state is dirty, rather than every draw.. doing things on every draw is not advised ;-)
19:45 robclark: hmm, interestingly, I do see "Handoff smr" traces on flex-5g boot (I think this is still using ACPI boot.. was just trying a f33 image, but no keyboard).. I guess time to swap in a new kernel..
19:48 robclark: hmm, but seems like my usb-drive has become "read only"..
20:04 alyssa: robclark: will look on tuesday, alright
20:05 robclark: fair, it will still be there on tues
20:05 alyssa: Currently panfrost does no dirty tracking since it's not necessary for the hardware (i.e. it's a pure CPU win, not GPU)
20:05 alyssa: and I'm too scared of introducing subtle bugs in random places that won't show up in deqp
20:06 robclark: ouch.. that has gotta hurt webgl aquarium :-P
20:07 alyssa: 🤷
20:07 alyssa: we still manage to be gpu bound everywhere that matters :(