18:06Lyude: mupuf: btw: do we need any more reviews on the clockgating patches?
18:09mupuf: not from me
18:09mupuf: Lyude: now, it is up to skeggsb to accept the code or not
19:00fr33will: Hi all, has there been any breakthough on NV138 (GP108)? The FeatureMatrix was "Last edited Mon 31 Jul 2017". This post suggests that it has working 3D hardware acceleration with no-reclocking support https://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-GP108-Firmware
19:00imirkin: support should land in 4.16
19:05karolherbst: imirkin: welcome back!
19:06fr33will: Does NV130 use the NVC0 driver? If so will GP108 have the same feature set as indicated on https://mesamatrix.net/ in ~4.16?
19:06imirkin: fr33will: yes. although note that the nvc0 driver offers different levels of support for different hw.
19:06imirkin: mesamatrix is not good at representing that
19:09karolherbst: imirkin: there is one odd thing I found in codegen regarding texture query lod. For 1D_Array textures for example we get one coord argument, even in TGSI, but from_tgsi reads out two, because getArgCount() returns 2. In the end we just end up with undefined values, but in nir we only get one and while handling texture query lod we would end up with an cvt without a source, because it doesn't exist. In the end I would just add
19:09karolherbst: fake sources, but I was wondering if we could drop those entirely
19:09imirkin: (maxwell+ does not differ too materially from kepler, but it does have a handful of small differences)
19:09imirkin: karolherbst: that's wrong. should only take a single arg for 1d_array.
19:09imirkin: karolherbst: it shouldn't be using getArgCount
19:09imirkin: it should be using getDims() + isCube()
19:09karolherbst: makes sense
19:09karolherbst: I try to fix codegen then in this regard
19:10imirkin: i think that was one of the first extensions i added support for
19:10karolherbst: I only found this issue running tests/all.py in piglit ....
19:10imirkin: and my understanding of texturing was much weaker then
19:10imirkin: and all this stuff seemed random and confusing
19:11imirkin: 423f64e83ab5b1ea7de475ae80300a8408522743 and its parent -- April 2014.
19:12karolherbst: I think the issue is caused by code within handleTEX though
19:12imirkin: the issue is that it shouldn't be using handleTEX :)
19:12karolherbst: because there getArgsCount is used for iterating over the srcs
19:19fr33will: thanks imirkin that comparison is really good :-)
19:21imirkin: fr33will: note that even with the patches slated for 4.16, you won't get reclocking, so it will all still be quite slow
19:27imirkin: note that the page i linked to is for mesa 17.1.0 ... been a bit lazy about keeping it up to date
19:34fr33will: imirkin: I want to try https://github.com/gnif/LookingGlass on a Geforce 1030, AFAIK it does not require a lot of processing power. However it does require GL_ARB_buffer_storage
19:36fr33will: (for the host part of the project)
19:36imirkin: that should be supporte
19:37imirkin: at least the ext - dunno about the software