#asahi-dev · 2023-07-14 23:33
eilnget that here file.io/UclKLM41n4bw#asahi-dev · 2023-07-14 23:33
eilnand there's an ipc.bin of (constant) boot args i haven't parsed yet#asahi-dev · 2023-07-14 23:32
eilnthis code is fucked in every way possible#asahi-dev · 2023-07-14 23:32
eilnChaosPrincess: https://pastebin.com/jaS9Kz9h#asahi-dev · 2023-07-14 20:36
eilnid be happy with just grabbing the raw frames, bc the ipc stuff is looking tough#asahi-dev · 2023-07-14 20:35
eilnthis thing is massive. like why is CISP_CMD_CH_FACE_DETECTION_IMAGE_ORIENTATION_SET in the fw?#asahi-dev · 2023-07-14 20:35
eilnnow to talk to the firmware...#asahi-dev · 2023-07-14 20:35
eiln5:30am but i got the isp fw booting on the asc hehe#asahi-dev · 2023-07-14 17:00
eiln(i.e. extremely sporadically and unreliably, sorry)#asahi-dev · 2023-07-14 16:55
eilni was poking at the isp yesterday. it's a beast but similar to the ane. my priority's the ane, but the isp seems stale, so i'll look at it in my downtime.#asahi-dev · 2023-06-16 00:51
eilni'd assume they should result in the same dev without chmod 666#asahi-dev · 2023-06-16 00:51
eilnmight anyone know why accel chrdev is crw------- whereas drm is crw-rw-rw- ?#asahi-dev · 2023-05-09 23:41
eilntake 2 today, gonna try to split the model#asahi-dev · 2023-05-09 23:40
eilnand ragequit lol#asahi-dev · 2023-05-09 23:40
eilnkept running out of ram in converting the image (unet) part from pytorch though#asahi-dev · 2023-05-09 23:40
eilnturns out they didn't really add much in the ventura fw update#asahi-dev · 2023-05-09 23:40
eilndid surgery on stable diffusion & got the text encoder/decoder part working yesterday#asahi-dev · 2023-05-05 03:57
eilnfixed#asahi-dev · 2023-05-05 03:48
eilnwait i counted wrong hold on a min#asahi-dev · 2023-05-05 03:47
eilnlmk if experiments/ane.py still runs fine#asahi-dev · 2023-05-05 03:46
eilnmerged & commented out the pd patch in ane-t6000 branch https://github.com/eiln/m1n1.git #asahi-dev · 2023-05-05 03:43
eilnt8103 is unaffected, jannau/sasha has been testing ane0/ane2 for me#asahi-dev · 2023-05-05 03:35
eilnah, ty. looking into it#asahi-dev · 2023-05-05 03:33
eilnmarcan: on some runs it was b0f, is that different? seems to work after shutting it down though#asahi-dev · 2023-05-04 22:43
eilnjannau: woohoo! the first run is expected since it's not pwrgated yet#asahi-dev · 2023-05-04 22:19
eilnif it does i should fix src/pmgr.c to handle pwrgating#asahi-dev · 2023-05-04 22:18
eilntake 2, lmk if experiments/ane.py under ane-t6000 branch works (either ane0/ane2)#asahi-dev · 2023-05-04 21:40
eilnjannau: ah finally looks right!! thank you sm :)#asahi-dev · 2023-05-04 21:36
eilnjannau: do you mind attaching the mon.poll()? the other one wasn't fully on then#asahi-dev · 2023-05-04 21:17
eilnplease lmk if anyone w/ a t600x ane0 can test#asahi-dev · 2023-05-04 21:17
eilnthe updated experiments/ane_t600x_power.py in the ane-power branch (https://github.com/eiln/m1n1.git) tries to manually powergate ane_sys_cpu before turning it on#asahi-dev · 2023-05-04 21:16
eilnwhich seems to be the PMGR_WAS_PWRGATED flag#asahi-dev · 2023-05-04 21:16
eilnTTY> pmgr: timeout while trying to set mode f for device at 0x28e0802c8: 10f#asahi-dev · 2023-05-04 21:16
eilnthe problem w/ t6000 seems to be:#asahi-dev · 2023-04-29 06:09
eilnstarted the HW:ANE page#asahi-dev · 2023-04-29 06:09
eilnfinally getting to documentation#asahi-dev · 2023-04-27 05:45
eilnand trace_device("/arm-io/error-handler", False)#asahi-dev · 2023-04-27 05:45
eilnjannau: yes please! do you mind tracing trace_device(node, mode=TraceMode.SYNC) for ane0, ane1, ane2, ane3#asahi-dev · 2023-04-27 05:12
eilni should move into tree w/ the next release. was waiting on the accel config#asahi-dev · 2023-04-02 11:55
eilnwith the accel patches i can move in-tree#asahi-dev · 2023-04-02 11:55
eilnmarcan: was wondering how the ane stuff should proceed?#asahi-dev · 2023-03-31 04:22
eilnexposing a render node for each seems straightforward, don't think it's the best though#asahi-dev · 2023-03-31 04:21
eilnwhat is the best way to bind/manage devices in this case?#asahi-dev · 2023-03-31 04:21
eilnso to exploit the "double" power requires running two models simultaneously#asahi-dev · 2023-03-31 04:21
eilni predict this influenced the "disabled" ane1/ane3 & lack of ane1 in m2 max decision#asahi-dev · 2023-03-31 04:21
eilncoreml compiles all h13 the same, which says everything :P#asahi-dev · 2023-03-31 04:21
eilnidentical firmwares across h13, no shared cache, DMA travels encoded at compile time#asahi-dev · 2023-03-31 04:21
eilnthere's no evidence multi-ane systems can "load balance" a single job (model)#asahi-dev · 2023-03-28 14:25
eilnmaz: if a patch holds that much importance the focus should be on simplifying the process #asahi-dev · 2023-03-28 14:22
eilne.g. intel vpu had to scrape userptrs in like v2; if qaic also does the patch becomes meaningless