00:46rhyskidd: it's 20 years from filing for the post 1995 patents
00:46rhyskidd: so should be up in June 2018
03:32duttasankha: Is Ptimer code is present inside drm/nvkm/subdev/pmu? or is it only contained in drm/nvkm/engine/pm?
03:34duttasankha: sorry Pcounter..not Ptimer
19:00tobijk: mhhh, something is fishy with pascal and nvc0_draw_elements(), nvc0_state_validate_3d() - they both miss something, adding a PUSH_KICK() sledgehammer "resolves" the issue
19:01tobijk: imirkin: any idea what pascal does different there in comparison to kepler? -^
19:01imirkin_: well, maxwell+
19:01imirkin_: and it's draw-elements-related
19:02imirkin_: it never uses the immediate path
19:03imirkin_:trying to find
19:05imirkin_: the code has changed relative to what i remember
19:06tobijk: where do you look at exactly? not sure where the actual differences are :/
19:07imirkin_: this feels totally broken... hold on.
19:07imirkin_: well, you can't feed it vertex data via pushbuf anymore
19:11imirkin_: or maybe i'm confusing myself?
19:11imirkin_: definite possibility...
19:11tobijk: not sure about that, sry
19:29tobijk: imirkin: the vertices are not structured btw: info->has_user_indices = false, maybe that is helpful :)
19:36imirkin_: doesn't matter
19:36imirkin_: is it a small draw out of a large vbo?
19:36imirkin_: does the vbo_push_hint get set?
19:38tobijk_: imirkin_: not sure, i have yet to find the "real" call, i just know its a glDrawElements or glDrawRangeElements
19:39tobijk_: the problem is visible in unigine valley, just mentioning
19:39imirkin_: well - try killing the translate thing
19:39imirkin_: oh yeah - i know that problem - on maxwell too right?
19:39tobijk_: imirkin_: possible, i don't have a maxwell around
19:39imirkin_: iirc MESA_DEBUG=flush helped it
19:40tobijk_: imirkin_: yep thats where i started and worked down to nouveau :/
19:40imirkin_: well i never had any doubt about nouveau being broken :)
19:41imirkin_: i wonder if it's some sort of misuse of the scratch buffer
19:41imirkin_: that'd be funny.
19:43tobijk_: maybe i should have another look at nvc0_state_validate_3d, as having a flush in front of that, removes another rendering issue
19:44tobijk_: flush as in PUSH_KICK
19:45imirkin_: this is probably obvious
19:46imirkin_: but trying to find a smaller test-case than 'unigine valley' would be quite useful ;)
19:46tobijk_: yep :)
19:53duttasankha: I was wondering if someone could help me regarding the calling mechanism of gf100_perfctr_X (X : init) functions? I don't see it get called...
19:53imirkin_: 'git grep'
19:56duttasankha: Actually I have created a function call graph using doxygen...but I don't see it getting called...it would try git grep
20:00imirkin_: i doubt doxygen would do a good with a function-pointer-heavy codebase like nouveau's
20:03duttasankha: okay...I see what you are saying and I am facing the challenge while I am digging...what do you suggest if I want to track the function calls?
20:07imirkin_: 'git grep'
20:09tobijk_: look for the function pointer name and see where it gets assigned :)
20:09tobijk_: hence grep
20:12tobijk_:was silly enough to hope for a maxwell+ vertex validate somewhere :/
20:12duttasankha: okay..thanks a lot for the suggestions..let me try it...