#
✕ nick:lina
#asahi-gpu · 2023-06-04 05:34
linaBut if there's something else triggering a leak of around that size every 3-4 minutes that would start to add up
#asahi-gpu · 2023-06-04 05:34
linaOn this machine, a single fullscreen framebuffer leak per minute (when the locker clock ticks) is still too much...
#asahi-gpu · 2023-06-04 05:33
linaThe smallest possible 60Hz leak is about 1MB/s (a single 16K page), which is a lot more than what I'm seeing. But maybe it has something to do with the screen locker clock.
#asahi-gpu · 2023-06-04 05:28
linaIt could be a kwin bug for all I know...
#asahi-gpu · 2023-06-04 05:28
linaI think there is a GEM object leak somewhere still, but it's not clear what causes it. I noticed shmem usage increasing, and killing kwin resets it. But it's not exported dmabufs. And I think I've only seen it happen with the lid closed?
#asahi-gpu · 2023-06-02 09:16
linaI was using a debug build to work around that GCC bug...
#asahi-gpu · 2023-06-02 09:16
linaLooks like it works on release builds ^^
#asahi-gpu · 2023-06-02 08:40
linaI think Carla is still leaking memory... and why do I get the feeling this one might not be Mesa..
#asahi-gpu · 2023-06-02 08:39
linaLet's see if it survives a few hours...
#asahi-gpu · 2023-06-02 08:39
linaOK, between that and the fd leak (which was probably keeping dmabufs alive in the wrong places) the baseline memory usage for my streaming machine looks a lot more reasonable
#asahi-gpu · 2023-06-02 08:28
linaJust boring old heap leaks...
#asahi-gpu · 2023-06-02 08:28
linajannau: It was at least 3 leaks in mesa, nothing to do with buffers
#asahi-gpu · 2023-06-02 05:11
linaOK, so Inochi2D is leaking a megabyte every few seconds, I guess I need to sort that out... wonder if it's its own thing or Mesa...
#asahi-gpu · 2023-06-02 04:58
linaOh but now it's fine... looks like it was kinda random?
#asahi-gpu · 2023-06-02 04:57
linaAnd kwin_wayland is using 80% CPU doing screen capture which also doesn't sound right...
#asahi-gpu · 2023-06-02 04:52
linaI finally updated my streaming machine and 4 hours in it my entire session got OOM-killed ^^;;
#asahi-gpu · 2023-06-02 04:51
linaalyssa: Umm... I think we have a memory leak somewhere...
#asahi-gpu · 2023-05-12 15:44
linaInteresting... I'll have to look at that, I wonder if all the Mesa changes that are queued made any difference so far (we did fix a while bunch of synchronization bugs... ^^)
#asahi-gpu · 2023-05-07 17:20
linaalyssa: If you're OK with me looking at that instead of the dEQP regression with the masking thing and we just disable it for now, I think that will work ^^
#asahi-gpu · 2023-05-06 15:29
linaAlso apparently something is weird with ARB_multi_draw_indirect that breaks Darwinia, but as far as I know Mesa just lowers that for us so I have no idea how it would break... I'll have to look at it next time...
#asahi-gpu · 2023-05-06 15:24
linaalyssa: Nevermind, she didn't read the spec and forgot the shader qualifier...
#asahi-gpu · 2023-05-06 15:04
linaalyssa: Did we ever test fbfetch (the GL side, not lower_blend)? Luna says advanced blend modes don't work, and from what I can see those are lowered on the GLSL side to framebuffer fetch ops...
#asahi-gpu · 2023-05-03 05:01
linaFor the GPU anyway, but it sounds like for DCP it is used ^^
#asahi-gpu · 2023-05-03 05:00
linajannau: the PMPDoorbell register range is only in the mappings for t8103, I think it's not used on any other SoC ^^
#asahi-gpu · 2023-04-26 17:40
lina(the one with a too large TIB stride)
#asahi-gpu · 2023-04-26 17:40
linaSpeaking of, I thought we were passing all of GLES3 but there's still that one test that needs eMRT?
#asahi-gpu · 2023-04-26 17:39
linaI opened a PR with the UAPI changes anyway ^^
#asahi-gpu · 2023-04-26 17:39
linaI just ended the stream! I'll have a bit of dinner and sleep ^^
#asahi-gpu · 2023-04-26 17:39
linaThe only interesting things I found are that if large_tib and the TVB isn't large enough I get a message and the job hangs (means I need to update the min_tvb_size for that case) and one of the registers has two bits which need to be set or a ton of things fault. I have a feeling it might be related to cache control/flushing?
#asahi-gpu · 2023-04-26 17:38
linaalyssa: I don't know if this is a good thing or a bad thing, but almost all the unknowns I exposed seem to do nothing to dEQP2/3... ^^;;
#asahi-gpu · 2023-04-13 09:09
linaI also suspect some of the register stuff is an actual union in the work command structure, and they still have the NULL pointers to the old fields in the new setup
#asahi-gpu · 2023-04-13 09:09
linaIt's possible that Apple themselves have a single codebase mostly, and just ifdefs for the features/modes...
#asahi-gpu · 2023-04-13 09:08
linajannau: Yeah, so now that I squashed my fixes in, other than a suspected padding field which has weird logic, 100% of the 13.2 -> 13.3 changes in initdata affected both the G14X and the G14G branches identically.
#asahi-gpu · 2023-04-13 08:50
linaUnfortuntely no, the register list thing didn't get backported to g14g (or at least not yet)... but at least some/many 13.2 -> 13.3 changes apply to both branches.
#asahi-gpu · 2023-04-13 08:00
linaYeah, I have 13.0b4 and 13.2 right now for everything and it's getting unwieldy, including changes that got reverted. So I think once I figure out if this is worth it, I'll drop those which were more of a PoC.
#asahi-gpu · 2023-04-13 07:57
linaI'm going to look at H14G next, we'll find out as soon as macOS installs ^^
#asahi-gpu · 2023-04-13 07:56
linaI get the feeling they might have unified this or plan to, but I still need to check other platforms
#asahi-gpu · 2023-04-13 07:55
linaI just worked out the struct changes, besides more initdata shuffling there's a few changes to 3D fields too...
#asahi-gpu · 2023-04-13 07:55
linajannau: All the firmwares changed significantly in 13.3 vs 13.2...
#asahi-gpu · 2023-04-13 07:04
linaThe firmwares did change, though 13.3 and 13.3.1 are the same and they have the same iBoot version, so we might want to target 13.3 if it makes sense over 13.2 I think
#asahi-gpu · 2023-04-13 06:29
linajannau: Soooo Apple made some major changes in the t602x GPU ABI that aren't in the other GPU variants, so I'm going to try 13.3.1 to see if any of that got unified by any chance
#asahi-gpu · 2023-04-13 06:16
linadottedmag: Each client has its own VA space (or actually as many separate VA spaces as it wants), so there is nothing too coordinate ^^
#asahi-gpu · 2023-04-12 09:15
linajannau: Ahhh! That explains it! (re agx_bo_free())
#asahi-gpu · 2023-04-12 05:20
linaOoooh, that's good to know! Either we're leaking those objects or we're leaking the address space.
#asahi-gpu · 2023-04-11 08:56
linarust-wip if possible!
#asahi-gpu · 2023-04-11 07:39
linajannau: PR works! ^^
#asahi-gpu · 2023-04-11 04:09
linaNeed more RAM to run Steam!!!
#asahi-gpu · 2023-04-11 04:01
linaLet's see how fast M2 Max goes! ^^
#asahi-gpu · 2023-04-11 03:57
linaalyssa: Wednesday is M2 Max but I'll see if I can fit those in (maybe offstream?) ^^
#asahi-gpu · 2023-03-29 12:30
linaalyssa_: ^^

plus ›