01:11mlankhorst: imirkin: I just grabbed the mmiotrace, wrote a parser for it
01:15RSpliet: ... I added support for decoding these scripts to demmio last summer
01:16RSpliet: (only for the scripts, not the full "PPWR" protocol)
01:25mlankhorst: imirkin: create a mmiotrace for going through all the clock states, isolate the bits that touch memory, replay those
01:25mlankhorst: i have one for boot to max speed, max speed to 300 mhz, and 300 mhz to max speed
01:42xexaxo: mupuf: ouch that piece of code ... :-)
02:06mlankhorst: but reclocking seems to make my card crash at a random point :/
02:06mlankhorst: even without memory
03:32JoshSyn: I am thinking to get myself a laptop with gtx 980m
03:32JoshSyn: I wonder if the opengl next will be compatible with it
03:32JoshSyn: what are the chances?
03:33JoshSyn: It's compatible with directx 12, so I assume features are gonna be same with glnext
03:34RSpliet: JoshSyn: without seeing the specs of this "gl-next", it's hard to judge
03:34JoshSyn: I agree though
03:34RSpliet: but I don't think Mesa will be in good shape to support this any time soon
03:35JoshSyn: just hope that nvidia are not crap like amd,
03:35JoshSyn: amd pretty much discontinued their support for older cards for the amdgpu drivers
03:35JoshSyn: and they don't even both to update their broken catalyst
03:36RSpliet: NVIDIA does not support the nouveau driver currently for anything besides Tegra K1 and soon Tegra X1
03:36RSpliet: doesn't mean the rest of the GPUs don't benefit from their work, but that's their focus
03:36RSpliet: for desktop cards, I believe NVIDIA has provided a pretty good closed source driver (in terms of performance and stability)
03:36JoshSyn: I just hope there is some sort of support for us gamers at least, does not matter if its proprietory
03:37RSpliet: then #nouveau is probably the wrong chan to ask ;-)
03:38RSpliet: all I can tell you is that I wouldn
03:38RSpliet: t expect it to be supported in nouveau any time soon
03:39RSpliet: for all the open source drivers, most of the front-end work is carried by AMD and Intel as they have the biggest manpower available
03:40RSpliet: without the front-end in order, the individual drivers can't start to implement
03:40RSpliet: nouveau will therefore most likely slightly lag behind AMD and Intels open source drivers
03:41RSpliet: the official NVIDIA driver, developed by folks that already have insight in the current draft specs, is likely to be quicker than the Mesa work *if* NVIDIA decides to commit to the standard
03:42RSpliet: this if is something I can't answer, as I'm not inside NVIDIA ;-)
03:42RSpliet: and of course, if it doesn't demand something from HW that it can't do... but NVIDIA has a say in the specs, so that's unlikely to happen
06:03Lucretia: JoshSyn: glnext will most likely need a completely new driver
06:04Lucretia: and a new "mesa" like project
06:04Lucretia:is expecting a slightly higher level command list api
06:11RSpliet: Lucretia: I don't know how big the impact is on the gallium-based backends
06:11Lucretia: er, dunno, but i'm betting they can reuse them
06:11Lucretia: in part
06:11Lucretia: at least the llvm part :D
07:23kisak: imirkin: a followup from a while back -- I finally got around to playing with the nouveau-wayland test box I intended to play with, and the GT330 I threw in it had favorable behavior with RBos, thanks for your opinion the other day
07:24kisak: (instead of a 7800GT)
09:21imirkin: kisak: yeah, nv50 is supported way better than nv40
09:22imirkin: mlankhorst: i was _hoping_ for something a bit more specific =/
10:58pmoreau: Does someone with a Riva TNT would have some time to look at https://bugs.freedesktop.org/show_bug.cgi?id=28095?
10:59imirkin: pmoreau: not a lot of nv04's running around... i have a nv05 but it works fine for me...
11:00pmoreau: How old are those cards?
11:00imirkin: 1999 or so
11:01imirkin: The first RIVA TNT based card released on June 15, 1998
11:01pmoreau: "Quite" old then! :D
11:01imirkin: The RIVA TNT2 was a graphics processing unit manufactured by Nvidia starting in early 1999
13:26mlankhorst: imirkin: what exactly are you trying on what card?
13:28imirkin: mlankhorst: nothing on any card... what i'd like to do is take, e.g., a mmiotrace (perhaps carefully-constructed) from my GF108 and write a script to reclock that works on that one specific card
13:29imirkin: mlankhorst: and then ideally, based on that experience, be able to write instructions so that others can do the same thing
13:51mlankhorst: imirkin: you need to grab the PDAEMON stuff and the uploads :P
13:51mlankhorst: the opcodes might be slightly different though
13:52imirkin: mlankhorst: well, i don't really care to try to understand what's going on
13:52imirkin: mlankhorst: just extracting the relevant bits and shoving them in again.
13:53imirkin: mlankhorst: do you have a training sequence in yours, or do you just "cache" training results?
13:54mlankhorst: imirkin: there's the scripted part that uploads something hardcoded :P
13:55mlankhorst: and the scripts around it use nvapeek/poke to do the preparation part from the host
13:55imirkin: mlankhorst: right... so how exactly do i extract that hardcoded stuff? do you have the mmiotrace you got that from?
13:56imirkin: iow what do i look for? :)
13:56mlankhorst: PDAEMON pokes :P
13:56mlankhorst: particularly look at the reclock stuff, there's a training pattern there..
13:57imirkin: ok. and do i have to upload the nvidia PDAEMON code blob as well?
13:57mlankhorst: look in my reclocking repo
13:57imirkin: i saw you had your own pdaemon blob... right?
13:57imirkin: i assume it was the nvidia one but decoded?
13:57mlankhorst: it's the one I wrote by hand
13:58mlankhorst: your take card from the bus function might be different
14:00imirkin: _could_ i just take the nvidia one?
14:00imirkin: or will that not work well with nouveau?
14:01mlankhorst: no idea
14:01imirkin: put another way -- why didn't you take the nvidia one?
14:01mlankhorst: I overwrote the blob with a custom hack
14:01mlankhorst: didn't want to RE it, I just replay the opcodes..
14:01imirkin: why would you have to RE it?
14:02imirkin: (i think i'm missing something)
14:02mlankhorst: erm because the nvidia blob probably needs to be loaded somehow and then you need to force it in reclcoking mode
14:02mlankhorst: no idea tbh
14:02mlankhorst: plus it was a good way to learn fuc
14:03imirkin: don't you just load it the same way you load all the other blobs?
14:03mlankhorst: https://github.com/mlankhorst/nvc0-memory-reclocking/blob/master/in.fuc ops aren't that hard..
14:04imirkin: yeah, but for all you knwo they've changed
14:05mlankhorst: enter_lock and leave_lock might be changed :s
14:05mlankhorst: reclocking is very card specific
14:05imirkin: i just mean that newer blob may have changed opcode numbers/functions around
14:06mlankhorst: I think they're pretty constant
14:07mlankhorst: besides if it's garbage you will know..
14:08mlankhorst: if for no other reason than assert(out[outpos - 2] == nvc0_pdaemon_data[fuc_ops_done]);
14:11imirkin: this would all go a lot smoother on a second dev box =/ might do it on mupuf's or something