15:13 <rellla> i'm testing with deqp now but have an issue:

15:14 <rellla> i use https://gitlab.freedesktop.org/mesa/mesa/blob/master/.gitlab-ci/deqp-lima-fails.txt as expected-fails list, but these file has the "Fail" suffix at the end of each line.

15:15 <rellla> doing the comparison with "comm" with the fails.txt (which doesn't have the "Fail" suffix) doesn't give me a right comparison result.

15:15 <rellla> what bit am i missing?

15:17 <rellla> i think the "Fail" in https://gitlab.freedesktop.org/mesa/mesa/blob/master/.gitlab-ci/deqp-lima-fails.txt should be deleted?

15:28 <anarsoul> rellla: they're using some running in lava CI that takes this list

15:28 <anarsoul> s/running/runner

15:28 <anarsoul> deqp-volt

15:28 <rellla> anarsoul: i think i have to insert some sed in my runner script, that deletes the "Fail" ...

15:29 <anarsoul> good idea :)

15:39 <rellla> anarsoul: regarding the vertex shader color outputs, i did https://gitlab.freedesktop.org/rellla/mesa/commits/lima-color-clamp

15:40 <rellla> and got some fixes in piglit http://imkreisrum.de/piglit/mali450/fb07fd4..fb07fd4-lima-color-clamp/

15:40 <anarsoul> these are random fixes because tests are likely not good

15:40 <rellla> the glsl-clamp-vertex-color one?

15:41 <anarsoul> yes

15:43 <anarsoul> it can be emulated though

15:43 <anarsoul> see nir_lower_two_sided_color

15:51 <rellla> i still don't get it what you mean with random?! :p

15:52 <anarsoul> rellla: test should check that fragment shader gets right color in gl_Color depending on what face it's shading - front or back

15:52 <anarsoul> but it doesn't

15:53 <anarsoul> so you get a pass by pure luck :)

15:53 <anarsoul> because test devs decided that they don't want to test this particular case :)

15:56 <rellla> ok, but anyway. with this nir pass i get 2 unexpected deqp fails:

15:56 <rellla> dEQP-GLES2.functional.negative_api.shader.uniform_matrixfv_invalid_transpose

15:57 <rellla> dEQP-GLES2.functional.negative_api.texture.generatemipmap_zero_level_array_compressed

15:57 <rellla> are they supposed to pass with master?

15:58 <anarsoul> check what's the difference?

16:03 <rellla> deqp is kind of confusing... where do i find the test's sources to see what it does?

16:03 <anarsoul> in modules/gles2

16:04 <anarsoul> I'd suggest to run this test separately

16:04 <anarsoul> you can do that with "-n dEQP-GLES2.functional.negative_api.shader.uniform_matrixfv_invalid_transpose" argument

16:04 <rellla> ok, will try

16:17 <rellla> http://imkreisrum.de/deqp/TestResults.qpa

16:17 <anarsoul> convert them to xml

16:17 <anarsoul> and put xsl and css in the same dir

16:17 <anarsoul> :)

16:17 <rellla> deqp convert xml

16:18 <anarsoul> testlog-to-xml

16:18 <rellla> yeah, #linux-sunxi != google :)

16:27 <rellla> hm, sth is wrong: Runtime check failed: 'out.good()' at /home/andreas/nfs/opipc2/VK-GL-CTS/executor/tools/xeBatchResultToXml.cpp:209

16:29 <rellla> maybe i deleted to much from the deqp build ...

16:30 <anarsoul> :)

23:31 <mastermart> well, i know i am behind my real capabilities cause starting my research late and sacrificing lot of my talent and life to do so. The position i was at demanded it , i had limited choices, hence i became an annoyoance to people who were having a better line, luckily the code is still so wrong that i still have a mission to fill in though.

23:34 <mastermart> But that is also common sense, if you torcher some known talents and capable thinkers, and let second hand people handle the stack of linus torvalds popularity growing LINUX os, well then shit happens also. Though it may have happened cause as always you did not take things seriously enough.

23:38 <mastermart> I did not expect to get any mission and involvement, i did not expect even making the breakthroughs i have done, i expected the way things went before in my life, that this time around i am dead and buried.

23:40 <mastermart> but definitely i sense in myself some internal content to come out lagging so far behind and at old age be still having that type of resillience to fight a mission to myself.

23:45 <mastermart> to beat all those nutters who terrorised me and made fun of me, and blocked all my chanches to become someone is , is the least of consolation, not in any way as cherry on the cake , though due to i was made sick, it is some satisfaction still.

23:49 <mastermart> It took me a while to realise that modern chips also have an arbiter for texture units, which is also out-of-order can be preempted by other requests which turn ready

23:50 <mastermart> because when you calculate the texture units on earlier gpus which is aligned with simd arbiter line width, it matches, but new chips have queues present per texunits.

23:50 <mastermart> so the simd line is larger then physical texture unit count.

23:51 <mastermart> i was going crazy and i am still on meds, and my mind plays tricks hence. I have been there before, but i took another look at this all.

23:52 <mastermart> by googling certain keywords, from attila-sim

23:53 <mastermart> this led me to several patents among which I link one here.

23:53 <mastermart> https://patents.google.com/patent/US7027062B2/en

23:54 <anarsoul> mastermart: that's nvidia, not arm

23:55 <mastermart> anarsoul: yeah, but it is the same on new arm cards definitely, as 2005 released attila-sim does the same roughly.