#asahi-gpu · 2023-02-04 02:28
linaDo we know how the mipmaps work in the Z direction? I don't know what OpenGL says, but I imagine if you're mipmapping in 3D you'd also decimate in the Z direction... obviously in memory you don't do that, but do we still have to do it during mip generation (and are we doing it properly now)?#asahi-gpu · 2023-02-04 02:20
linaSo does that mean 3D and arrays are *not* the same as far as the hardware is concerned?#asahi-gpu · 2023-02-04 02:20
linaOhhhh so for 3D textures the Z dimension also goes into the mipmaps. That makes perfect sense and answers some questions I had.#asahi-gpu · 2023-02-04 02:14
linaalyssa: we need to align to the "full" miptree <- didn't we do this already? I'm pretty sure I sent that patch to always allocate full miptrees...#asahi-gpu · 2023-02-04 02:14
linaalyssa: ^^#asahi-gpu · 2023-02-02 11:51
linajannau: Whoops! Pushed to lina/gpu-wip now, sorry about that!#asahi-gpu · 2023-01-25 15:01
linaWell I'm going to fix that soon! ^^#asahi-gpu · 2023-01-25 14:09
linaAh...#asahi-gpu · 2023-01-25 13:58
linaI thought we were? At least at some point...#asahi-gpu · 2023-01-25 11:24
linaGot it!#asahi-gpu · 2023-01-25 03:43
linaSorry ^^;#asahi-gpu · 2023-01-23 05:10
linaOn the other hand it's rock solid right now! ^^#asahi-gpu · 2023-01-23 05:09
linaSome of it is just "Rust paranoia" and generally architectural. Some of it can be fixed with refactoring, some with more unsafe... but I guess that can wait a bit ^^#asahi-gpu · 2023-01-23 05:08
linaSome of it will probably improve after review too... I *am* a Rust newbie, I'm sure I'm doing more than one silly thing.#asahi-gpu · 2023-01-23 05:07
linaI've kind of been punting the performance issue until after the general design is finished, which I guess means soon once fences/scheduling is in.#asahi-gpu · 2023-01-23 05:07
linaI haven't, but I'm not too surprised... There's a lot of stuff going on and more locking/etc complexity than I originally intended.#asahi-gpu · 2023-01-21 19:21
linaalyssa: :linakira:#asahi-gpu · 2023-01-20 16:35
linaI managed to make it work, now it's down to 4-5 seconds for a simple change ^^#asahi-gpu · 2023-01-20 16:35
linaalyssa: So apparently Rust supports incremental compilation and multiple codegen units... it all just wasn't hooked up into the kernel.#asahi-gpu · 2023-01-20 15:09
linaAbout 14 seconds for a single version#asahi-gpu · 2023-01-20 15:08
linaRight now it takes about 30 seconds to compile, which isn't great, but for normal dev you can disable some versions. I should probably add config options for that...#asahi-gpu · 2023-01-20 15:02
linas/managed/manage/#asahi-gpu · 2023-01-20 15:01
linaI did managed to insulate it from the event module though (yay traits and dynamic dispatch)#asahi-gpu · 2023-01-20 15:01
linaThough I did have to add versioning to the whole workqueue module and a bunch of commands with this bump so that increased version divergence a bit#asahi-gpu · 2023-01-20 15:00
linaI'm pretty sure Nouveau still takes longer to compile ^^;;#asahi-gpu · 2023-01-20 14:59
linaHaven't tested G13X/13.2 but given the intersection of structure changes I'd be surprised if it doesn't work.#asahi-gpu · 2023-01-20 14:58
linaThe current code is tested on G13G/12.3, G14G/12.4, G13G/13.2, G14G/13.2 (3D and compute)#asahi-gpu · 2023-01-20 14:58
linaIt's not a big deal though, it just increases compile time#asahi-gpu · 2023-01-20 14:57
linaThe M2 firmware is missing the clustering fields, so either it's still a different branch (which means we will need at least a G14G/G14X split, at least for this version) or they did something weird and don't need that any more#asahi-gpu · 2023-01-20 14:56
linaalyssa: Port to 13.2 done ^^ (both M1 and M2)#asahi-gpu · 2023-01-19 16:54
linaalyssa, jannau: Doesn't mean they're not compatible, g13g and g13x have different firmwares but the structures are compatible (only runtime changes, same version tag in the driver)#asahi-gpu · 2023-01-17 16:22
linaAlso this is all assuming I'm reading the shader asm correctly... I'm not exactly used to it ^^;;#asahi-gpu · 2023-01-17 16:22
linaThose are the only SRs it reads#asahi-gpu · 2023-01-17 16:19
linaAlso I'm pretty sure 20 is core ID#asahi-gpu · 2023-01-17 16:19
linaI only ever saw command 0xf, which is the one that runs some mystery instructions and terminates, and never saw 146/147 changing, but that would make sense if there is only one item in the FIFO here#asahi-gpu · 2023-01-17 16:18
lina146/147 is the save/restore context descriptor pointer, but then 144 seems to be a command FIFO, and after reading that it reads 146 again and (for some commands) treats it as an offset into a pointer which only makes any sense if reading 144 pops the FIFO and 146/147 are a data payload.#asahi-gpu · 2023-01-17 16:14
linaIt's funky, I think one of the special registers is actually a FIFO? The outer loop and the way it reads one register twice and adds together the values makes no sense otherwise.#asahi-gpu · 2023-01-17 16:13
linaThe main reason I wanted a higher level actual language is that it has to deal with data structures to some extent, and a big outer switch statement inside a loop, and it just feels pretty painful to write all that in NIR...#asahi-gpu · 2023-01-17 16:12
linaI'm more interested in what it's doing at a higher level and how to replace it#asahi-gpu · 2023-01-17 16:12
linadougall might want to look at the mystery instructions though#asahi-gpu · 2023-01-17 16:12
linaalyssa: I already looked through some of it, I'll probably cover it during tomorrow's stream ^^#asahi-gpu · 2023-01-17 07:09
linaFor the compute context save/restore/misc blob, the logic is complicated enough I don't want to be messing around with assembly/binary (or building it in raw NIR), but I need some way of inserting at the very least some special register loads and some still-mysterious instructions...#asahi-gpu · 2023-01-17 07:08
linaalyssa: Is there some way of compiling GLSL for an internal shader in mesa, and some reasonable way of adding intrinsics to it?#asahi-gpu · 2023-01-16 16:06
linayes#asahi-gpu · 2023-01-16 16:05
linaSome registers are still the same or very similar#asahi-gpu · 2023-01-16 16:05
linaAGX is a PowerVR descendant, forked quite a while ago#asahi-gpu · 2023-01-16 16:03
linaalyssa: Oooooo that is good, their code drop was missing a ton of stuff. I'm sure we can dig out interesting things from here ^^#asahi-gpu · 2023-01-15 06:02
linaI 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...#asahi-gpu · 2023-01-15 06:02
linaAaaa, 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...