#
…

2023-01-15

00:07linaalyssa: What did you see when you tried compute stuff without the magic shader?
00:07linaI'm not using it right now and I *think* it's for context switching only, but I also have this cache issue...
01:34faruk has quit [Quit: Textual IRC Client: www.textualapp.com]
01:47systwi has joined #asahi-gpu
02:11ah- has joined #asahi-gpu
02:58possiblemeatball has joined #asahi-gpu
03:00cylm has quit [Ping timeout: 480 seconds]
03:04mini0n has quit [Quit: Leaving]
03:12pthariensflame has joined #asahi-gpu
03:23Stary has quit [Ping timeout: 480 seconds]
03:37Stary has joined #asahi-gpu
04:00pthariensflame has quit [Quit: Textual IRC Client: www.textualapp.com]
04:11Stary has quit [Ping timeout: 480 seconds]
04:11linaI'm looking at the magic shader code and it seems to be some kind of dispatcher based on a special register value...
04:12linaIt definitely interacts with the compute buffer struct I found
04:12linaSome codepaths do save/restore stuff (and I think we might be missing some looping instruction decoding? I see unknown instructions right after the load/store groups...)
04:12linaBut at least one op, 0xf, just runs 5 unknown instructions. I wonder if that is cache maintenance...
04:21linaAnd then there's a magic store to address 0x2d822acc in here... is this some kind of MMIO or doorbell?
04:54Stary has joined #asahi-gpu
06:02linaAaaa, I just tried doing a faulting write in that shader and it locks up the whole GPU... the compute part runs fine, but a subsequent 3D job hangs. I can see the fault status but the firmware is just sitting there...
06:02linaI think what happens is that since we're not flushing caches, the fault happens after the compute job is complete, in some state where the firmware doesn't implement timeouts properly...
06:26possiblemeatball has quit [Quit: Leaving]
06:52SSJ_GZ has joined #asahi-gpu
07:43hertz has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
08:18tim has joined #asahi-gpu
08:18tim has quit [Remote host closed the connection]
09:07le0n has quit [Quit: see you later, alligator]
09:08le0n has joined #asahi-gpu
09:36LinuxM1 has joined #asahi-gpu
09:52SSJ_GZ has quit [Ping timeout: 480 seconds]
10:02kit_ty_kate has quit [Quit: WeeChat 3.6]
10:56LinuxM1 has quit [Ping timeout: 480 seconds]
11:06bcrumb has joined #asahi-gpu
11:06bcrumb has quit []
12:02hertz has joined #asahi-gpu
12:16mkurz has quit [Quit: Leaving]
13:28bcrumb has joined #asahi-gpu
13:49possiblemeatball has joined #asahi-gpu
14:05bcrumb has quit [Quit: WeeChat 3.7.1]
14:11SSJ_GZ has joined #asahi-gpu
14:19cylm has joined #asahi-gpu
14:25cr1901 has quit [Ping timeout: 480 seconds]
14:38Tramtrist has quit [Remote host closed the connection]
14:42Tramtrist has joined #asahi-gpu
15:03amarioguy has joined #asahi-gpu
15:52alyssa has joined #asahi-gpu
15:52alyssathis should be _writes, not _reads
15:54alyssathis should be predicated on stage == COMPUTE, you're writing into a union
15:55balor has quit [Quit: balor]
15:56balor has joined #asahi-gpu
15:56ella-0alyssa: With respect to the system values MR I am currently porting AGXV to it. Will let you know if anything doesn't work well with vulkan but so far so good :3
16:05bcrumb has joined #asahi-gpu
16:07alyssaella-0: awesome!
16:07alyssaIs it, like... better than what we had before?
16:07alyssaI've never written a VK driver lol
16:14ella-0It simplifies the command buffer code somewhat and makes it possible to allocate a chunk of the register file for push constants.
16:15alyssaI still don't really get how
16:15alyssabut if you say it's right and Jason says it's right that is good for me
16:15alyssaand obviously it solves a problem I had in GL lol
16:18bcrumb has quit [Quit: WeeChat 3.7.1]
16:22alyssaby the way.. at what point will we be bottlenecked on agxv specific stuff
16:22alyssai.e. when should I pull your tree and start hacking on deqp-vk, versus continuing to support more and better gl in the interest of agxv
16:28mini0n has joined #asahi-gpu
16:28ella-0uhh not sure
16:29ella-0I was hoping to get some basic compute running on agxv before that I think
16:29alyssamakes sense
16:30alyssaI was hoping the combination of you, Lina, Karol, and Dougall would be able to finish off compute support without me ^.^
16:33alyssamake sure my baby can grow without me, yknow
16:41faruk has joined #asahi-gpu
16:57cr1901 has joined #asahi-gpu
16:58ella-0That makes sense. I'm happy to work on compute stuff :3 currently the main thing holding agxv back is not having vkCmdCopy* and other transfer operations implemented. I tried and failed to implement them using the tilebuffer load/store shaders
17:00alyssavkCmdCopy* sucks.
17:01alyssavkCmdCopyImageToBuffer should probably be a compute kernel
17:01alyssavkCmdCopyBufferToImage should be a fragment shader
17:01alyssaprobably
17:01alyssaexcept maybe some little details about ASTC/BCn formats? but I think you can just munge the dimensions?
17:01alyssaalso I would expect vk_meta can do those because they're pretty general
17:02alyssavkCmdCopyBufferToBuffer is also a totally generic compute kernel
17:02faruk has quit [Quit: Textual IRC Client: www.textualapp.com]
17:02alyssathe only spicy thing is vkCmdCopyImageToImage
17:02alyssaimplementing that efficiently requires being able to "cast" framebuffer compressed images to different formats
17:03ella-0Yup it's pain :<
17:04alyssaright, so, there are basically 2 classes of hardware
17:06alyssa1. hw that can do compressed texture views. in this case, img2img copies are just blits! you might need to munge the dest format into something renderable with the same size, maybe munge the dimensions in some evil cases, but by and large this is generic blitting on the fragment pipe, where a texture view (with view format != original format) on the sampling image for casting
17:08alyssa2. hw that can't. obviously this class of hw can still do uncompressed -- at worst, linear -- texture views. so the sane option here is to eat the bandwidth hit -- allocate an uncompressed staging resource and lower compressed->compressed copies to a pair of copies compressed->uncompressed + uncompressed->compressed because you can do views for each of those. and those simpler copies are easy and
17:08alyssageneric on the fragment pipe.
17:08alyssaI am unsure which class of hardware AGX is.
17:08alyssaMali is #1 on Valhall, but #2 on anything older
17:09alyssapanvk 1.0 has some extremely delicate logic to try to do compressed->compressed image copies with format conversion in one shot with format packing/unpacking in the copy shaders
17:10alyssawhile that's almost certainly faster, it isn't worth the complexity ... vkCmdCopyImageToImage with compressed<--->compressed and incompatible formats is an abomination that should never have been added to the spec and bad perf is to be expected
17:10alyssaI don't know if any real apps would even hit that. I'm doubful.
17:10alyssaGL drivers won't even try and will just decompressed your image in this case.
17:13hightower2 has joined #asahi-gpu
17:15alyssa3 classes i guess
17:16alyssa3. hardware blitters that trivialize the problem
17:16alyssaNVIDIA is #3 so NVK doesn't need meta shaders for this
17:16ella-0interesting
17:17alyssabut architecturally vk_meta should be able to support both #1 and #2 with some work
17:17alyssaif you don't get to it I probably will in, like, June
17:17alyssaJuly maybe
17:17alyssafor $DAY_JOB
17:24bluetail8 has joined #asahi-gpu
17:29faruk has joined #asahi-gpu
17:30landscape15 has joined #asahi-gpu
17:31bluetail has quit [Ping timeout: 480 seconds]
17:31bluetail8 is now known as bluetail
17:41faruk has quit [Quit: Textual IRC Client: www.textualapp.com]
18:10landscape15 has quit [Quit: WeeChat 3.8]
18:21DarkShadow4444 has quit [Quit: ZNC - https://znc.in]
18:21hotbbqsauce has joined #asahi-gpu
18:22hotbbqsauce has quit []
18:25DarkShadow4444 has joined #asahi-gpu
18:26DarkShadow4444 has quit [Remote host closed the connection]
18:26DarkShadow4444 has joined #asahi-gpu
18:33dg_ has joined #asahi-gpu
18:34dg_Error: PageSize configuation is wrong: configured with 4096, but got 16384
18:34dg_ has quit []
18:57m1nion has joined #asahi-gpu
19:03mini0n has quit [Ping timeout: 480 seconds]
19:28tim has joined #asahi-gpu
19:29tim is now known as Guest1427
19:32zkrx has quit [Remote host closed the connection]
19:33zkrx has joined #asahi-gpu
19:35possiblemeatball has quit [Ping timeout: 480 seconds]
19:39possiblemeatball has joined #asahi-gpu
20:12bcrumb has joined #asahi-gpu
20:12bcrumb has quit []
20:15bcrumb has joined #asahi-gpu
20:22bcrumb has quit [Quit: WeeChat 3.7.1]
20:27bcrumb has joined #asahi-gpu
20:43bcrumb has quit [Quit: WeeChat 3.7.1]
20:47SSJ_GZ has quit []
20:50bluetail has quit [Quit: The Lounge - https://thelounge.chat]
20:53bluetail has joined #asahi-gpu
21:06bluetail has quit [Quit: The Lounge - https://thelounge.chat]
21:09bluetail has joined #asahi-gpu
21:16balor has quit [Quit: balor]
21:16balor has joined #asahi-gpu
21:20krbtgt has joined #asahi-gpu
21:26possiblemeatball has quit [Quit: Leaving]
21:31Guest1427 has quit [Quit: Guest1427]
21:54Dementor has quit [Remote host closed the connection]
21:54possiblemeatball has joined #asahi-gpu
21:55Dementor has joined #asahi-gpu
23:00yrlf has quit [Quit: The Lounge - https://thelounge.chat]
23:20irth has joined #asahi-gpu
23:44m1nion has quit []
23:53ju has quit [Remote host closed the connection]
23:54possiblemeatball has quit [Ping timeout: 480 seconds]
23:56possiblemeatball has joined #asahi-gpu
…