00:27 duttasankha: I was wondering if there is any documentation/link about some brief description about the code inside a directory and it's sub-directories.. e.g. the brief description of the code inside nvkm/subdev and it's sub-directories like bar/bios....In some of them the name is kind of self-explanatory..but it would be helpful to go through some sort of description...
03:24 imirkin: karolherbst: 330d0607ed60fd3edca192e54b4246310f06652f received a bunch of fixups in nouveau after-the-fact
03:25 imirkin: hm, you're right it's still broken though
03:25 imirkin: ARGH!
03:32 imirkin: karolherbst: looks like something in the push hint logic... of course
03:46 imirkin: aaand fixed.
03:56 imirkin: ugh. and it's all broken on nv50 too =/
03:59 imirkin: oh that's nice. fix is the same.
07:58 karolherbst: imirkin: nice, thanks
07:59 karolherbst: uhh, my first though was actually that the offset was wrong, but I thought not enough was added, so I looked the other way...
14:33 tobijk: imirkin: scilab is an open source clone of matlab, you can download it at their website if you are interested :)
14:33 imirkin_: ah
15:19 aphirst: is the scilab language even similar to matlab?
15:19 aphirst: I'd've thought that the title "open source clone of matlab" would better belong to octave
15:46 imirkin_: i wish there were an open-source mathematica...
15:48 karolherbst: that would be usefull, yes
15:48 imirkin_: it was always so much better at math than me
15:48 imirkin_: kinda sad, really
15:49 imirkin_: some day i'll know how it does Integrate[Log[Sin[x]], {x, 0, Pi}]
15:50 karolherbst: :D
15:50 imirkin_: fun. it's even able to compute an indefinite integral, with the help of a polylogarithm function. which i have no clue what it is.
15:51 karolherbst: it's a little crazy but oh well
15:51 imirkin_: but the final result is that it's equal to -Pi * Log[2], which it's able to compute symbolically.
15:57 karolherbst: Li is weird stuff: http://mathworld.wolfram.com/Polylogarithm.html
15:59 karolherbst: ohhhh
15:59 karolherbst: it kind of makes sense though
16:29 imirkin_: karolherbst: any other regressions?
17:22 karolherbst: next regression: spec@ext_framebuffer_object@fbo-maxsize
17:40 imirkin_: i didn't run that one
17:40 imirkin_: but i suspect it's not really a regression
17:41 imirkin_: what is it bitching about?
17:49 duttasankha: I was trying to understand the VM and the information provided in http://envytools.readthedocs.io/en/latest/hw/memory/g80-vm.html ....For that I was planning to go through subdev/mmu ....I just want if I am going trough the right code...
18:13 karolherbst: imirkin_: 743ad599a97d09b119d26b99f6b79e41b567e421 broke it
18:13 karolherbst: wrong pixel values
18:14 karolherbst: Probe color at (192,64) Expected: 0.750000 0.250000 0.250000 Observed: 1.000000 1.000000 1.000000
18:14 imirkin_: ok
18:15 imirkin_: probably some over-flushing made it work
18:15 karolherbst: weird, first run with llvmpipe also failed, but not the second one
19:31 karolherbst: why is GL45-CTS.shader_atomic_counters.basic-buffer-bind failing now
19:31 karolherbst: meh... random fail
19:31 karolherbst: *sigh*
19:33 airlied: no that's my fault
19:34 karolherbst: well it passes if I run this test only
19:36 karolherbst: https://trello.com/c/16KVl2ZG/16-cts-4540-status
19:38 airlied: wierd
19:39 karolherbst: uhm... airlied you have the same color as imirkin_, I thought you were him
19:40 karolherbst: now I am thinking, either fixing 3d images or taking care of the sampling issues :/
19:46 karolherbst: okay, the blit is already wrong in the GL45-CTS.gtf30.GL3Tests.packed_depth_stencil.packed_depth_stencil_blit test
19:53 karolherbst: seems like we finally need to fix those spec@!opengl 1.1@depthstencil* tests
19:54 duttasankha: what is the purpose of 0x100cbc and 0x100cb8 registers? Does anyone knows?
19:54 karolherbst: check with lookup
19:55 duttasankha: I did git grep in rnndb...didn't get it...
19:55 RSpliet: duttasankha: that's a "Tesla" generation GPU right?
19:55 duttasankha: yes
19:56 duttasankha: no sorry fermi
19:56 RSpliet: oh... hmm, hard to say what that might be on Fermi, on Tesla I recall this being the area where the NISO poller and some other caches are configured
19:57 RSpliet: pmoreau: you touched these regs, didn't you?
19:58 pmoreau: RSpliet: Yes, for NVAC
19:58 imirkin_: duttasankha: you can't grep rnndb
19:58 imirkin_: duttasankha: use the 'lookup' tool
19:58 RSpliet: oh... no that
19:58 RSpliet: s 100c10-100c24
19:58 pmoreau: 0x100c14, 0x100c18, 0x100c24 IIRC
19:59 RSpliet: oh, heh, yes, and what imirkin_ says
19:59 imirkin_: lookup -a c0 100cbc
19:59 imirkin_: PFFB.VM.TLB_FLUSH_TRIGGER => { PAGE = SPECIFIC | VSPACE = SPECIFIC | SET = HUB_AND_GPC }
20:00 imirkin_: lookup -a c0 100cb8
20:00 imirkin_: PFFB.VM.TLB_FLUSH_VSPACE => { TARGET = VRAM | ADDR = 0 }
20:00 imirkin_: you can also feed it a value which will tell you how that decoder -- this is decoding with the implicit value of 0
20:01 duttasankha: imirkin_: okay.. thanks for mentioning the lookup...
20:01 imirkin_: so if e.g. you saw that value 123456 was being written, you might do
20:02 imirkin_: lookup -a c0 100cb8 123456
20:02 imirkin_: PFFB.VM.TLB_FLUSH_VSPACE => { TARGET = SYSRAM | ADDR = 0x12345000 | 0x4 }
20:03 imirkin_: you might do -a c8 for your gf110, although the results should be rahter similar
20:05 karolherbst: imirkin_: every looked into those opengl1.1 blit piglit fails?
20:05 imirkin_: probably
20:06 karolherbst: I am sure we need to fix those to pass at least one of those gtf tests
20:10 karolherbst: uhh
20:10 karolherbst: interesting
20:10 karolherbst: I disabled that "flags |= NV50_TEXVIEW_ACCESS_RESOLVE;" in nvc0_blit_set_src and now more subtest fail with the exactly same error
20:11 imirkin_: "don't do that" :)
20:11 karolherbst: right
20:11 karolherbst: there is also this if (filter && res->nr_samples == 8) flags |= NV50_TEXVIEW_FILTER_MSAA8; thing, but mhh
20:12 imirkin_: why are you looking at msaa stuff?
20:12 karolherbst: it's the line after the other line
20:12 imirkin_: look away while you still can
20:12 karolherbst: well...
20:12 karolherbst: I think we need to fix it to pass cts
20:13 karolherbst: or I mix up those terms
20:13 karolherbst: ohh wait
20:13 karolherbst: I guess I should ask this now before wasting time
20:13 imirkin_: not sure what msaa has to do with depth
20:13 karolherbst: what does "samples" exactly mean here? spec@!opengl 1.1@depthstencil-default_fb-blit samples=2
20:13 imirkin_: unless the depth failures only happen when we're doing msaa or something
20:14 imirkin_: that means you're in a world of pain
20:14 karolherbst: well righto
20:14 imirkin_: you have a MSAA *visual*
20:14 karolherbst: GTF output: https://gist.github.com/karolherbst/f2d867245d87dd9559ee9c3002a7c2f3
20:14 imirkin_: which means that there's implicit resolve junk that needs to happen but doesn't
20:14 karolherbst: so I figured, I need to fix multisample stuff
20:14 karolherbst: guess I wasn't wrong
20:15 imirkin_: can you do a trace on that cts test?
20:15 imirkin_: and see if it's requesting a msaa visual
20:15 karolherbst: apitrace?
20:15 imirkin_: ya
20:16 karolherbst: which EGL call I have to look for?
20:16 imirkin_: well, mainly see which config it picks
20:17 karolherbst: mhh eglGetConfigAttrib(dpy = 0x2d171f0, config = 0x2d52190, attribute = EGL_SAMPLES, value = [0])
20:17 karolherbst: this?
20:17 imirkin_: maybe? i'm not so familiar with egl
20:17 karolherbst: me neither
20:18 karolherbst: well, it is the only egl* call with samples inside one of the arguments
20:18 karolherbst: (ignoring eglGetProcAddress)
20:18 imirkin_: i don't have time to look now
20:18 karolherbst: mhh or this? eglGetConfigAttrib(dpy = 0x2d171f0, config = 0x2d52190, attribute = EGL_NATIVE_VISUAL_ID, value = [33]) = EGL_TRUE
20:22 karolherbst: ohh okay
20:22 karolherbst: it chooses a config with samples==0
20:24 karolherbst: it does enable multisampling though