#
#asahi-gpu · 2021-02-16 01:10
bloomI'd offer to do the plumbing but uh there's a reason I haven't written any asahi code in a month :(
#asahi-gpu · 2021-02-16 01:10
bloomdougall: most of `demo` is gfx-specific hacks but compute kernels will be simpler
#asahi-gpu · 2021-02-16 00:51
chrisfdougall: neat!
#asahi-gpu · 2021-02-16 00:28
dougallAlso, I knew I had a reason to be suspicious of these loads and stores! Looks like they can do float<->int and sRGB<->RGB (most of the "Pack and Unpack Functions"). Hopefully that will explain most of the bits that were worrying me
#asahi-gpu · 2021-02-16 00:27
dougallYeah, it just needs to hold together for long enough to replace the testbed with non-Metal code, and your 'demo' stuff feels close (I haven't looked at command stream stuff yet, though)
#asahi-gpu · 2021-02-15 13:57
bloomI know this feeling uncomfortably well lol
#asahi-gpu · 2021-02-15 13:56
bloomdougall: I'm generally afraid to modify the code in hwtestbed because any changes to how it configures the GPU might break all the tests"
#asahi-gpu · 2021-02-15 12:43
dougallHey - I cleaned up and pushed my hardware testing code, based on wrap.dylib. It's uncomfortably hacky, but it's been my most useful tool in verifying instruction behaviour so far. Happy to help if other people want to mess with it :)
#asahi-gpu · 2021-02-15 02:07
chrisfso can we do big strided & masked stores to uniform space just like device space stores?
#asahi-gpu · 2021-02-15 01:58
bloomon Mali, the equivalent operation is genuinely a general-purpose store, mapping the uniform register file as a generic output for a compute shader
#asahi-gpu · 2021-02-15 01:57
bloom". This is encoded like (and possibly is) a store to device memory, and can move one 16-bit register to initialise a 16-bit uniform, or two consecutive 16-bit registers to initialise a 32-bit uniform"
#asahi-gpu · 2021-02-15 01:52
bloomnever_released: 🎉
#asahi-gpu · 2021-02-14 23:05
chrisfthis looks like another good resource if usable
#asahi-gpu · 2021-02-14 22:55
never_releasedbloom: it's there now
#asahi-gpu · 2021-02-14 17:47
bloomLicense?
#asahi-gpu · 2021-02-14 12:23
j`eyhttps://github.com/dougallj/applegpu
#asahi-gpu · 2021-02-12 13:57
marcanopticron: jfyi, chrisf works on mesa/bifrost, I assume he's aware of the background
#asahi-gpu · 2021-02-12 12:51
opticronchrisf, that has been discussed before and I think the outcome was that b.loom was more familiar with gallium, so that would be the first GPU driver
#asahi-gpu · 2021-02-12 04:18
chrisfone potentially contentious idea would be that it shouldnt exist, and we'll just do the vulkan driver. if you want GL, use zink or angle over it.
#asahi-gpu · 2021-02-12 03:16
dhewgimpossible!
#asahi-gpu · 2021-02-12 03:14
marcandhewg: I think she means you got the name wrong :p
#asahi-gpu · 2021-02-12 02:49
dhewgawwww
#asahi-gpu · 2021-02-12 01:10
bloom;P
#asahi-gpu · 2021-02-12 01:10
bloomdhewg: notabug wontfix
#asahi-gpu · 2021-02-11 17:02
dhewgthat's just my weird kind of humor
#asahi-gpu · 2021-02-11 17:02
dhewg:P
#asahi-gpu · 2021-02-11 16:59
chrisfdhewg, this is going to take just a little patience ;)
#asahi-gpu · 2021-02-11 16:57
dhewgplz2fix
#asahi-gpu · 2021-02-11 16:57
dhewgls: cannot access 'src/gallium/drivers/sandstorm': No such file or directory
#asahi-gpu · 2021-01-23 15:04
bloomhthh_: your first one's always memorable :>
#asahi-gpu · 2021-01-23 14:35
hthh_heh, it did rather - so many good mysteries (and just things to learn about how gpus work)
#asahi-gpu · 2021-01-23 14:23
bloomhthh_: heard you're doing ISA stuff now, the nerdsniping worked! ;P
#asahi-gpu · 2021-01-23 14:04
bloomnever_released: Even now, I review a lot of code :)
#asahi-gpu · 2021-01-23 14:04
bloomdanilonc: Compute would have made more sense (and I have notes on it), but graphics was a lot splashier
#asahi-gpu · 2021-01-23 13:07
xerpi[m]Do we know if the display engine supports scanout of tiled/compressed framebuffers (which I would guess it does), or is there a de-tiling/de-compressing job on the swapchain before presenting?
#asahi-gpu · 2021-01-23 11:08
never_releasedbloom: once we start doing GPU stuff or even before that?
#asahi-gpu · 2021-01-23 03:12
davidrysk[m]bloom: would it make sense to do testing with compute kernels too?
#asahi-gpu · 2021-01-23 03:04
davidrysk[m]wonder what these buffers do
#asahi-gpu · 2021-01-23 03:04
davidrysk[m]yeah no triangle
#asahi-gpu · 2021-01-23 03:03
davidrysk[m]ha I changed demo_unk10[15] to 0x00 and now I get a single solid color
#asahi-gpu · 2021-01-23 03:02
davidrysk[m]anything else to poke at before I leave for the night?
#asahi-gpu · 2021-01-23 02:58
davidrysk[m]yeah I'd say merge that and do the same for the demo
#asahi-gpu · 2021-01-23 02:57
davidrysk[m]ah that's mostly what I meant to do :)
#asahi-gpu · 2021-01-23 02:56
bloomyes, but see PR
#asahi-gpu · 2021-01-23 02:56
davidrysk[m]bloom: your makefiles aren't quite right -- can I fix them?
#asahi-gpu · 2021-01-23 02:48
davidrysk[m]yes
#asahi-gpu · 2021-01-23 02:47
bloomas stated, I expect the framebuffer to be tiled/compressed
#asahi-gpu · 2021-01-23 02:46
davidrysk[m]It was abused for a bunch of jailbreaks though
#asahi-gpu · 2021-01-23 02:46
davidrysk[m]before iOS 9, people regularly used it to record the screen
#asahi-gpu · 2021-01-23 02:46
davidrysk[m]btw it should be possible to get the framebuffer through IOMobileFramebuffer's private APIs but Apple has made that much more difficult than it used to be

plus ›