<rellla>
it shows me a benefit of ~ 1.64 microseconds per single run :/
<enunes>
rellla: do you get better results if you scale it more for very large buffers and large deltas (index 1,2,3, then 10000,10001,10002) and accumulated time differences between thousands of runs?
<rellla>
enunes: i will give it a try
<rellla>
btw, is there any better way to measure this? though, if we have an avarage of many runs and different scalings/ large buffers we should at least see, whats better...
<enunes>
I think the arm kernel driver had some performance counters, but we don't have that implemented
<enunes>
maybe just 'perf' could help showing where most of the time is spent
paulk-leonov has quit [Excess Flood]
monstr has quit [Remote host closed the connection]
<enunes>
rellla: by the way, I tried to compile deqp to run with the blob some days ago and I was not very successful at that. do you still have a working image for the A20? (cubieboard 2?)
<enunes>
I ran into many issues with different deqp targets, at the end I remember it kept trying to link with libGL.so or fail because some of the gl functions were not in libGLESv2, even though I tried my best to make it only gles2...
<enunes>
my final goal was to test a change enabling some fake shader-db like report for number of instructions, so I could get some comparison data between ppir and the disassembles from the full deqp runs on blob
adjtm_ has joined #lima
adjtm has quit [Ping timeout: 265 seconds]
adjtm_ has quit [Remote host closed the connection]
adjtm_ has joined #lima
<rellla>
enunes: i have. can send it tomorrow
<rellla>
it's built for wayland target and uses the wayland/gbm blob
<enunes>
rellla: ah, that is helpful too, I never tried that target. well don't even know if the distro I installed has a wayland blob
<rellla>
surfaceless target doesn't work with blob btw
<enunes>
cool, thanks
<enunes>
and you use some debian as base os?
<enunes>
I tried surfaceless and some targets for x11
<rellla>
a clean debian buster from scratch with own kernel, yes. no armbian.
<enunes>
ok that is supposedly what I have too
<enunes>
I'll give a try to those blobs
<enunes>
I'm also thinking of an alternate solution, which would be a horrible hack on shader-db run.c to initialize a window in some other way than /dev/dri/renderD%d so that it might run on shader files themselves even with the blob
<rellla>
after you got deqp build you need a running weston instance started locally
<rellla>
then you have to export XDG_RUNTIME_DIR
<rellla>
that was a bit tricky for me to find out but lets me rin deqp via ssh now
<enunes>
looks like I have a wayland blob but weston doesn't start, maybe I have to find out which net of symlinks end up pointing to one blob or another
<rellla>
yeah. thats why i dont have lima and mali and debian-mesa on one machine. its a horror.
anarsoul has quit [Read error: Connection reset by peer]