15:04 KungFuJesus: hah, yep, I don't think this is specifically an big endian issue like that macro would imply
15:22 KungFuJesus: awesome, I think I'm also to confirm the bugs that I'm seeing that I've worked around already are also _not_ endianness specific
15:24 KungFuJesus: and just like the 7800 GT, framerates seem abysmal with vsync enabled - suggesting something dodgy around fencing mechanics, maybe
15:26 KungFuJesus: though I kind of wonder if it's just the engine for this game having really bad swap interval mechanics. Maybe I should try enabling vsync with a compositor instead to see if it improves things
15:26 KungFuJesus: I can't seem to reproduce the issue of randomly missing VBOs, though
15:26 KungFuJesus: well, textures, anyway
15:35 KungFuJesus: hmm, maybe that bug is endianess specific, or something only seen when paired with a weak CPU
15:35 KungFuJesus: in any case, the "endianness workaround" applied was probably the only reason nv30 was at all behaved on big endian and why x86 was all kinds of jank
15:37 KungFuJesus: or somehow NV49 is magically better behaved than NV47
15:39 KitsuWhooa: I thought I'd offer running things on my card if you wanted to verify anything on little endian, but turns out I have an NV86 :p
15:51 KungFuJesus: Yeah everything just became usable with NV30-NV40 with that patch. Here's the MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20974
15:54 KungFuJesus: I guess one circumstance I never encountered was something that didn't hit the "push_hint" heuristic on big endian
16:33 karolherbst: KungFuJesus: nice
16:33 karolherbst: also tried like the newest gnome with that?
16:33 karolherbst: guess I'll give it a try and see how much it helps wiht everything
16:42 KungFuJesus: imirkin's comments seem to imply maybe there was some inverted logic happening here?
16:42 KungFuJesus: perhaps I misunderstood when and how that hint was supposed to be honored
16:42 KungFuJesus: vbo_fifo being set to "not 0" definitely bypasses an upload there, though
16:43 KungFuJesus: you can tell because the framerate drops kind of precipitously there, too
17:16 KungFuJesus: karolherbst: Am I interpreting the original author's logic here badly? Does it not seem like VBOs are reuploaded if vbo_fifo is 0?
17:25 karolherbst: I'd have to dake a deeper look. Will do so tomorrow
17:39 KungFuJesus: that other goofy issue with nv30_screen.c also alludes me, only seems to happen on my PPC64 big endian machine. Makes me wonder if it has anything to do with the missing textures when VBOs get used, too
17:55 KungFuJesus: the GL 1.3 version of the game suffers no such missing texture issues, even at the full framerate. Geometry is handled by glDrawArrays there
18:23 karolherbst: you could also do apitraces and compare them, but not sure how easy that would be for x86 vs ppc, but worst case you just "bisect" it by hand and find the first frame/call being different
18:24 KungFuJesus: well, part of it I think is there's filtering being handled in shaders
18:26 karolherbst: hey
18:26 karolherbst: uhh.. wrong chat