#asahi-gpu · 2023-08-25 14:23
i509vcbI've had the opinion that rust punishes bad architecture#asahi-gpu · 2023-08-25 14:21
i509vcbCertainly not as bad as some cases I've seen where people have done inheritence via that...#asahi-gpu · 2023-08-25 14:20
i509vcbI guess Into<u64> if you really need the raw value#asahi-gpu · 2023-08-25 14:20
i509vcbYeah probably a bit of Deref abuse there#asahi-gpu · 2023-08-25 13:21
i509vcbAnd then implement Deref<Self::Target = u64> for GpuPtr#asahi-gpu · 2023-08-25 13:20
i509vcbYeah in rust I'd probably make a newtype for that#asahi-gpu · 2023-08-25 04:19
i509vcbCould there be any stylistic improvement to have a sort of "agx_gpu_ptr" typedef to make it more clear that a uint64_t should be semantically treated as a gpu pointer in code?#asahi-gpu · 2023-08-23 22:58
i509vcbs/intrinsic/sysval#asahi-gpu · 2023-08-23 22:47
i509vcb(now I see why the gallium driver doesn't lower every intrinsic the same way)#asahi-gpu · 2023-08-23 22:34
i509vcbI'll do that later#asahi-gpu · 2023-08-23 22:34
i509vcbAnd I realize I did not consider indirect yet in my MR#asahi-gpu · 2023-08-23 21:30
i509vcbat least one is done#asahi-gpu · 2023-08-23 21:29
i509vcbs/unsued/unused#asahi-gpu · 2023-08-23 21:29
i509vcbI guess all the stuff I built to make the sysval stuff expansible in the future is going to sit unsued for a bit...#asahi-gpu · 2023-08-23 21:28
i509vcbit clobbers the currently bound pipeline#asahi-gpu · 2023-08-23 21:28
i509vcbHmm I seem to have found a problem with the vk_meta_copy stuff#asahi-gpu · 2023-08-23 20:57
i509vcbOk now it passes#asahi-gpu · 2023-08-23 20:53
i509vcbnope premature victory it passes some of the tests...#asahi-gpu · 2023-08-23 20:50
i509vcbalyssa: thank you for the help so far#asahi-gpu · 2023-08-23 20:50
i509vcbWell dEQP-VK.compute.builtin_var.num_work_groups passes#asahi-gpu · 2023-08-23 20:50
i509vcbFound my problem, forgot to copy a struct...#asahi-gpu · 2023-08-23 20:39
i509vcbI guess blame my code for uploading the uniforms?#asahi-gpu · 2023-08-23 20:39
i509vcbhttps://gist.github.com/i509VCB/1e8f18e69ba980cd37e641eed56a8900#asahi-gpu · 2023-08-23 20:39
i509vcbThe shader looks fine, I can see the load from the right uniform registers and the store into the buffer where the dEQP test compares the result#asahi-gpu · 2023-08-23 20:27
i509vcbsomething a bit simpler, dumping nir?#asahi-gpu · 2023-08-23 20:22
i509vcbI guess I'll have to guess where I went wrong since the dEQP says I have the wrong values#asahi-gpu · 2023-08-23 20:16
i509vcbis there a way I can dump the state of uniform registers?#asahi-gpu · 2023-08-23 17:36
i509vcb(well I mean what buffer to read from for indirect)#asahi-gpu · 2023-08-23 17:35
i509vcbfor indirect, probably the same thing but I just tell where to load the values from?#asahi-gpu · 2023-08-23 17:33
i509vcb(how do I go from the parameters passed into vkCmdDispatch to putting those in the uniform registers I have reserved for that sysval)#asahi-gpu · 2023-08-23 17:32
i509vcbI believe I've done lowering for base_vertex and num_workgroups correctly. But I need to have these values be set before the main shader executes somehow? I'd imagine I need to do something when recording the command to load the uniform registers with the expected values?#asahi-gpu · 2023-08-22 23:57
i509vcbbut for a 64-bit value, it appears the two registers just need to be neighbors? (u4 and u3 for example?)#asahi-gpu · 2023-08-22 23:56
i509vcbJust to confirm, it not possible to do unaligned reads? i.e. read a 32-bit value from u1h and u2l#asahi-gpu · 2023-08-22 23:24
i509vcbregister index of sorts?#asahi-gpu · 2023-08-22 23:19
i509vcbwhat does the "base" mean in load_preamble?#asahi-gpu · 2023-08-22 20:19
i509vcbConceptually I guess I can just consider the whole thing one pass even though I do a sort of inner pass to figure out what sysvals need to be lowered and another inner pass to actually lower#asahi-gpu · 2023-08-22 19:45
i509vcbSounds about right#asahi-gpu · 2023-08-22 19:44
i509vcbWell relative to the rest of the passes#asahi-gpu · 2023-08-22 19:44
i509vcbI guess I wonder where do I throw the passes to replace the intrinsics for things like load_num_workgroups with loads from uniform registers#asahi-gpu · 2023-08-22 19:41
i509vcbagxv seemed to be fine after the upgrade, although I am a bit deep into sysval stuff in agxv #asahi-gpu · 2023-08-22 03:56
i509vcbOver nir as in the entire shader#asahi-gpu · 2023-08-22 03:56
i509vcbWell... I already reinvented the wheel, but it could be square #asahi-gpu · 2023-08-22 03:56
i509vcbBefore I reinvent the wheel, is there some sort of accepted way to do a sort of read only pass over nir?#asahi-gpu · 2023-08-22 02:16
i509vcbwhat does the bool return value mean for nir passes? #asahi-gpu · 2023-08-21 22:55
i509vcbElla and I did discuss this a bit and it seems we agreed that agxv would probably differ a bit there#asahi-gpu · 2023-08-21 22:54
i509vcbok#asahi-gpu · 2023-08-21 21:54
i509vcbalyssa: I imagine some of the stuff in agx_nir_lower_sysvals.c will need to be replicated in agxv, is that something you'd require or really like to be common between vk and gallium or is agxv free to do what it wants?#asahi-gpu · 2023-08-17 21:29
i509vcbok so agxv can't use that#asahi-gpu · 2023-08-17 21:17
i509vcb(v3dv and a few other drivers use lower_compute_system_values)#asahi-gpu · 2023-08-17 21:17
i509vcbis there a reason the gallium driver has it's own bespoke thing there with AGX_SYSVAL_TABLE_GRID?