16:44 mohamexiety[d]: skeggsb9778[d]: what should I be looking for in OpenRM for compression tags? (as in, what are the relevant details we need? I am still looking but found quite a few so far; just not sure which of it is relevant/helpful or not)
21:31 airlied[d]: not sure anyone knows openrm that well, you might have to trace the vulkan driver ioctl calls to see what msgs it might send to the gsp etc
21:50 dragongirlsandtrucks: Hi, guys. I have an NVS 4200M in my Dell XFR E6420 - I'm running off of the Nouveau drivers, but keep running into issues with the display freezing if I scroll or do anything too quickly. Any recommendations?
22:02 dragongirlsandtrucks: I tried following these directions: https://nouveau.freedesktop.org/VideoDecoding.html
22:02 dragongirlsandtrucks: But it keeps fialing at this step
22:03 dragongirlsandtrucks: $ python extract_firmware.py
22:03 dragongirlsandtrucks: Skipping gzip blob at 0x5d92e4 (33440 bytes), wrong magic: 0x2
22:03 dragongirlsandtrucks: Skipping gzip blob at 0x5da444 (7648 bytes), wrong magic: 0x100c
22:03 Ermine: this script doesn't work on newer nvidia drivers
22:04 dragongirlsandtrucks: Is this script meant for the non-free drivers, then?
22:05 airlied[d]: it's just to extract mpeg2/h264 decoding firmware, it's not going to solve display freezeing
22:06 dragongirlsandtrucks: I wasn't sure if video acceleration would have any ties to that. Thanks.
22:06 skeggsb9778[d]: mohamexiety[d]: I think you should mostly just need to find where GSP-RM tells CPU-RM how many tags there are
22:06 skeggsb9778[d]: But again, I'm not sure if/what else changed for Turing+
22:09 dragongirlsandtrucks: Unfortunately, this territory is new to me. I'm happy to go looking, but I'm 1.not quite sure where to start, 2.not quite sure what to do with that information if I find it. I apologize.
22:13 dragongirlsandtrucks: I was directed here by someone else, I spent some time digging through the site and found a bugs section associated with GitLab, I will give that a shot. I believe this would be a 2D driver issue, I was able to run emulators for games without much issue, it mainly only happened with web browsers and other similar programs.
22:13 dragongirlsandtrucks: Thank you.
22:20 airlied[d]: it's probably not fixable, we don't have reclocking
22:20 airlied[d]: though fermi might have options to manually reclock
22:21 skeggsb9778[d]: mohamexiety[d]: I'm not 100% sure, I only had a quick look, but:
22:21 skeggsb9778[d]: - it appears the comptag can be calculated from the *physical* (vram) address directly on turing (https://github.com/NVIDIA/open-gpu-kernel-modules/blob/main/src/nvidia/src/kernel/gpu/mem_mgr/arch/turing/mem_mgr_tu102.c#L464)
22:22 skeggsb9778[d]: And there's some hints here about detecting support for it at all: https://github.com/NVIDIA/open-gpu-kernel-modules/blob/main/src/nvidia/src/kernel/gpu/mem_mgr/arch/turing/mem_mgr_tu102.c#L590
22:23 skeggsb9778[d]: airlied[d]: That doesn't work so well on Fermi. I had a branch somewhere many years ago with a couple of attempts at it (for different RAM types), but I never finished it. Kepler is probably the only place it has a decent chance of working
22:35 mohamexiety[d]: skeggsb9778[d]: thanks! yeah I saw this one among the stuff I found. did not see the other one though about detecting support
22:36 mohamexiety[d]: what's next? I guess the large page stuff? I thought to search around openrm first since it seemed easier
22:37 skeggsb9778[d]: yeah, i think all you should need to do (famous last words) is to fix the large page support, then fill in the PTE comptags field when mapping with a kind that supports compression
22:41 mohamexiety[d]: got it, thanks. will look around tomorrow then for large page. I am honestly still not sure how to begin with it but I probably need to look harder