01:28imirkin: urgh... all this breakage
01:28imirkin: i can't keep up =/
01:28imirkin: looks like vdpau died
01:31imirkin: offf course. dri3.
01:31imirkin: and there's a SECOND issue in there which is not dri3.
06:31sneil: I tried to trace the 381.22 driver on the blob for the arb_compute_shader-indirect-compute piglit case that locks my desktop under nouveau
06:31sneil: this is all I got: https://paste.fedoraproject.org/paste/4DTQOo7PojHoo5DbXiY7jF5M1UNdIGYhyRLivL9gydE=
06:34sneil: I'm assuming something didn't work or isn't working
06:48sneil: ah... reading the mmiotrace doc, I'm probably not getting anything because I need to load the driver while under mmiotrace
07:56sneil: OK, I think I have a full mmiotrace (and I also ran draw-pixels -samples=2 -auto as well)
07:56sneil: File is ~194M, .xz is ~5.8M
07:57sneil: It's late here so if anyone is interested in it, let me know. I read the irc logs
10:29karolherbst: I think cmp b8 is broken
10:38karolherbst: ohh wait, silly interface
10:38karolherbst: R14 is 0x80
10:48karolherbst: cmp b8 0x0 0x80 != cmp b32 0c0 0x80
10:49karolherbst: cmp b8 0x0 0x80 != cmp b32 0x0 0x80
11:00karolherbst: yeah, something is broken with cmp b8
11:01karolherbst: uhh wait... I need to use cmpu
11:12karolherbst: mwk, mupuf: did I do something wrong here? Or is the falcon doing silly things indeed? https://gist.github.com/karolherbst/d9c9b2282a920daa701bbd1930f34929
11:14karolherbst: ohhh, meh
11:16karolherbst: ge doesn't trigger as well
11:16karolherbst: so we have a broken envyas
11:21karolherbst: RSpliet, skeggsb: basically cmpu is broken currently, if used with bra l/g/le/ge it produces wrong results
11:21karolherbst: there might be more issues
11:22karolherbst: maybe context errors are due to that?
11:22karolherbst: ne is also broken \o/
11:22karolherbst: cmpu b32 $r0 $r0 ; bra ne #symbol -> it branches
11:23karolherbst: ohh wait, I wrote be
11:23karolherbst: okay, ne works, but still
13:22karolherbst: mhh I am silly. cmpu just doesn't work like this and I need to use those a/b bra codes
16:57nuovolnx: Currently i use debian 9 with nvidia legacy owner driver. Are the nouveau open source drivers good for my graphics card? It's a gt 320. The pc has blocked on google maps with the default nouveau drivers so I installed those nvidia legacy. Does Nouveau work with gt320?
16:58karolherbst: nuovolnx: it should, yes
17:01nuovolnx: karolherbst, Sorry for my English, I'm Italian. Now I'm using the nvidia-legacy-340xx-driver drivers because the PC was blocked with nouveau. I have an acer aspire x3900.
17:02nuovolnx: This proprietary driver works fine but I would be use the open driver, but I don't understand what wrong
17:02karolherbst: maybe something went wrong with the switch?
17:02karolherbst: no idea how that stuff works on debian 9 now, you should ask in a debian channel for distribution support
17:03karolherbst: but depending on your desktop it might block, because there are some unresolved multi-context problems in nouveau
17:06nuovolnx: I have installed minimal debian, than I have installed mate, apt-get install mate-desktop-environment, than apt-get install lightdm
17:09nuovolnx: the same problem there is with debian 8.8 but with nvidia legacy driver works fine
17:09nuovolnx: I wouldn't use the close driver
17:19nuovolnx: karolherbst, in debian-it are tell me to use close driver
17:20karolherbst: how does this even fit with the debian philosophy
17:20karolherbst: tell them that you want to use the open driver and they should stop discussing this
17:22nuovolnx: I would like to buy a graphics card. Is the r7 250 automatically recognized by the open source driver with debian 8.8 or 9 without making any configuration?
17:25AndrewR: ..I partially stay on old monitor/card combo due to ..new stuff having tendency to fail in new and wonderful ways (displayport, from recent yers ....but also a bit surprizingly VGA connected via ..something separate)
17:28nuovolnx: AndrewR, to me?
17:37AndrewR: nuovolnx, for everyone ... I think any card (or integrated GPU) will show its problematic aspects at some point,... just don't want to spread false hopes (but my view often colored by constant read of bugzilla ..and some strange case from my friend, who had unexplained display failre from new amd card ...thanfully it was gone in more recent ver. of distribution ...)
17:37AndrewR: argh for typos :/
21:32pmoreau: Hum… with radeonsi maybe moving to a NIR backend, maybe should I reconsider the SPIR-V to NVIR approach, and try to go with a NIR to NVIR instead.
21:34airlied: pmoreau: the problem with nir for opencl spir-v is the flow control
21:34pmoreau: I know it doesn’t have to be structured for OpenCL SPIR-V modules.
21:35pmoreau: Does that break some of NIR assumptions?
21:36airlied: nir only allows structured flow control afaik
21:36airlied: though jason has looked into supporting unstructured
21:38pmoreau: I know he also looked into adding pointers.
21:41pmoreau: There are a few hurdles currently, but going with NIR could be a better long-term choice? However, it will take some extra time to add what is missing.
21:41pmoreau: But it will benefit other drivers like freedreno.
21:42karolherbst: pmoreau: Doing it simply as an experiment may be worth it for nouveau already
21:43karolherbst: especially comparing the generated code with both paths might give us interesting results
21:44pmoreau: Sure, but then the question becomes: how much more time should I spend in the SPIR-V backend, as it might be scrapped in the future?
21:53smusiland: Sun is not doing, allah is doing
21:53smusiland: Skies are not doing, allah is doing
21:53smusiland: boss is not doing Allah is doing
21:53smusiland: #nouveau is not doing, allah is foing
21:53smusiland: Trump is not doing, allah is doing
21:54smusiland: Galaxies are not doing, allah is doing
21:55RSpliet: imirkin: could you ehh... do, well
23:16pmoreau: xexaxo1: There is something weird going on with the build system: I just updated to HEAD, made a clean compilation (and re-ran automake/configure from scratch), but when I modify sources in clover, it does not trigger a recompilation.
23:19pmoreau: xexaxo1: Even though it did compile them initially. It does not look like it tests state_trackers at all on the latter runs (https://hastebin.com/labahudati.go). My configure options: https://hastebin.com/izeworeyus.js