<anarsoul>
tomeu: could you please update kernel on boards with lima in lava lab to 5.5-rc? Looks like error recovery is broken due to some drm_sched bugs in 5.4
fysa has quit [Read error: Connection reset by peer]
stikonas has joined #panfrost
adjtm_ has joined #panfrost
adjtm has quit [Ping timeout: 258 seconds]
BenG83 has joined #panfrost
BenG83 has quit [Ping timeout: 240 seconds]
<icecream95>
It might be possible to do S3TC decode in shaders at a reasonable speed, using something like both a full-res RG1 and 1 pixel/block RGBA16 texture.
<icecream95>
The main problem is working out how (and where) to split the texture into two.
<icecream95>
Tiling a 2 bits per pixel format might be interesting as well...
forkbomb has joined #panfrost
<icecream95>
It might be easier to just use one texture with 1 pixel/block, though that might be slightly slower.
icecream95 has quit [Ping timeout: 245 seconds]
yann has quit [Ping timeout: 240 seconds]
yann has joined #panfrost
megi has joined #panfrost
stikonas has quit [Remote host closed the connection]
davidlt has quit [Ping timeout: 265 seconds]
raster has joined #panfrost
_whitelogger has joined #panfrost
davidlt has joined #panfrost
stikonas has joined #panfrost
TheKit has quit [Read error: Connection reset by peer]
stikonas has quit [Remote host closed the connection]
nerdboy has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
TheKit has joined #panfrost
davidlt has quit [Ping timeout: 260 seconds]
megi has quit [Ping timeout: 240 seconds]
icecream95 has joined #panfrost
<anarsoul>
alyssa: ping
<anarsoul>
alyssa: is this FIXED_16 used a lot in midgard?
<anarsoul>
we may have something similar for min_lod/max_lod/lod_bias in texture descriptor in utgard
<anarsoul>
but it's 8 bit for unsigned and 9 bits for signed
<anarsoul>
and it seems to be 4.4
<anarsoul>
i.e. 4 bits for integer part, 4 for fraction
<anarsoul>
1 for sign
<anarsoul>
also blob seems to set lod_bias depending on combination of min/mag/mipmap filtering, it can be 0/0.5/-1
<anarsoul>
hm, however I'm not sure that 0x1ff is -1
<urjaman>
... i just got a kernel fucked up in a way where commanding it to reboot -f doesnt do anything, but i can ssh into the system
<urjaman>
i'm like ... me:"IT WAS REBOOT -F" kernel: "grumble grumble stuck workqueues..."
<urjaman>
ah well, atleast i got my work saved :P (and power-cycled the thing with the power button already)
<icecream95>
The formula for decoding the MALI_FORMAT_SPECIAL2 formats from a signed N-bit int: (<c> + 0.5) / (0.5 * 2^<N> - 0.5)
<icecream95>
It returns a value in [-1, 1] and is similar to SNORM, except that for SNORM -127 and -128 are the same, and this format doesn't have an exact 0.