<bbrezillon>
alyssa: any plan/WIP for veci8 support? my min/max index search shader is using such ops for 8bit indexes
mifritscher has joined #panfrost
<alyssa>
bbrezillon: i8vec4, that is?
<bbrezillon>
yep
<alyssa>
it's an "eventually" issue right now
<alyssa>
when I start playing around with OpenCL, sure
<alyssa>
in the mean time, getting it right for Bifrost will take a -lot- of work and there are frankly much more impactful opts
<alyssa>
i16vec2 is complicated enough as it is right now
<macc24_>
alyssa: is performance on kevin still kinda bad?
<alyssa>
macc24_: Yeah although I'm less and less sure whether that's purely a mesa issue..
<alyssa>
italove: Seems heavy-handed but there's not an obviously better solution so... meh.
<macc24_>
well on cros it's kind of better, no idea if it's ui or something else
<macc24_>
trying to drive 2400x1600 screen with a midgard was a mistake...
<alyssa>
heh
<alyssa>
1366x768 with t760 otoh is pretty nice :p
<macc24_>
yeah, my c201pa is gonna get used more after i transplant it into chassis of a thinkpad x201
<bbrezillon>
alyssa: can we at least lower those in the meantime?
<alyssa>
bbrezillon: It should be getting lowered now?
<bbrezillon>
things are scalarized, yes
<alyssa>
lower_alu_to_scalar should be scalarizing it, and we should be handling scalar i8
<alyssa>
[If we're breaking on *scalar* i8, yes I do consider that a bug that does merit my attention now instead of later.]
<bbrezillon>
but i8vec4 component extraction and packing are failing
<alyssa>
I see. Could you push the repro case? (even if it's just "my indirect draw branch, this deqp case")
<alyssa>
will look later today
<italove>
alyssa: I thought you'd say that, that's why I mentioned in the commit that we might need to optimize this later, I agree that there isn't a better solution that's quick to implement
<macc24_>
i wonder if i can glue duet to kevin and use kevin as just virtualgl client