#
✕ nick:daniels
#asahi-gpu · 2022-02-10 15:33
daniels(at least it was in IMG GPUs)
#asahi-gpu · 2022-02-10 15:33
danielsTA -> Tile Accelerator
#asahi-gpu · 2021-05-30 20:53
danielswe can hear!
#asahi-gpu · 2021-05-30 19:55
danielsit's less latent than push + look at patch :P
#asahi-gpu · 2021-05-30 19:55
danielsbloom: you should also embrace tig
#asahi-gpu · 2021-05-30 19:52
danielsrebase ftw
#asahi-gpu · 2021-05-30 18:14
danielsapparently I have an oddly-shaped blob on my screen, because that read as 'I would witch it', which seems apt
#asahi-gpu · 2021-03-25 20:46
danielsnope, via a backend IR - I would recommend reading the blog post above to get a reasonable idea of the lines between the three (NIR, backend IR, bytecode)
#asahi-gpu · 2021-03-25 20:43
danielsit it has substantial changes, your abstraction layer is just going to get in the way
#asahi-gpu · 2021-03-25 20:43
danielsif the new architecture has trivial changes, you don't need enough adaptation to have built an abstraction layer
#asahi-gpu · 2021-03-25 20:43
danielsoh no, please don't think about future architecture generations
#asahi-gpu · 2021-03-25 20:38
danielsso the short answer is that LLVM is not useful in and of itself. the driver won't use LLVM to implement a full compiler at any point; and for code generation and analysis in development, the driver will ultimately have its own bespoke backend IR, so a (dis)assembler will ultimately be targeted closer to that
#asahi-gpu · 2021-03-25 20:36
danielshttps://www.collabora.com/news-and-blog/blog/2020/04/23/from-bifrost-to-panfrost-deep-dive-into-the-first-render/ is probably the best introduction of this I can think of
#asahi-gpu · 2021-03-25 20:36
danielseach driver (and generation within the driver) though has its own custom backend IR, which is a much closer fit to assembly and disassembly
#asahi-gpu · 2021-03-25 20:36
danielsNIR (like SSA) is a mostly-architecture-independent SSA IR which is good for transformations and optimisations
#asahi-gpu · 2021-03-25 20:30
danielsDarkShadow44: it depends on what your end goal is - ultimately the results of the GPU reverse engineering will result in a Mesa driver which uses Mesa's NIR compiler framework rather than LLVM
#asahi-gpu · 2021-03-18 16:43
danielsbalrog: it's strong differentiator for every SoC vendor, there's basically no commonality in ISP
#asahi-gpu · 2021-03-18 16:31
danielsright sorry, I misread 'integrated' as 'on-CPU' rather than 'on-die'
#asahi-gpu · 2021-03-18 16:30
danielsas EL3?
#asahi-gpu · 2021-03-18 16:29
danielsmarcan: SEP -> Secure Enclave Processor? -> ... T2?
#asahi-gpu · 2021-03-18 16:27
danielsany idea if it's in-house or licensed?
#asahi-gpu · 2021-03-18 16:26
danielsjust a big old userspace blob
#asahi-gpu · 2021-03-18 16:26
danielsmarcan: presumably it's a stateless impl -> very possibly no firmware?
#asahi-gpu · 2021-03-18 16:22
daniels(yes it's a lot of out-of-band storage, but cheaper in power than keeping DDR lit up full)
#asahi-gpu · 2021-03-18 16:21
danielskeep an internal camera -> ISP -> codec pipeline which spits out a downscaled YUV/RGB frame for display, and then the codec goes straight to H.264/HEVC/whatever, ideally without touching sysmem
#asahi-gpu · 2021-03-18 16:20
danielsbut given that your phone doesn't have a 4K display, rather than compressing 4K output, the secret is to just ... not have it
#asahi-gpu · 2021-03-18 16:19
daniels(I've also not seen it in the wild)
#asahi-gpu · 2021-03-18 16:19
danielsArm never got their media-codec IP in any shipping SoC TTBOMK, but they will license AFBC to you if you want to do that
#asahi-gpu · 2021-03-18 16:11
danielsbut yeah, enabling things like PSR is a seriously frustrating long-tail effort
#asahi-gpu · 2021-03-18 16:10
daniels(not surprising that Apple actually did it properly, but it is sadly uncommon for GPU render targets and display controller fetch to share tiling formats ...)
#asahi-gpu · 2021-03-18 16:08
danielsrealistically you don't need too many more if you do the HWC thing of using the GPU to flatten your static-ish content to a single buffer, then using the display controller for the more dynamic content
#asahi-gpu · 2021-03-18 16:07
danielsmarcan: 3 is plenty: one for system UI, one for browser chrome, one for Netflix content
#asahi-gpu · 2021-03-18 16:03
danielsgiven their battery life figures, I have a hard time believing that they're using the GPU to composite every frame (at least when not doing fullscreen desktop)
#asahi-gpu · 2021-03-18 16:03
danielschrisf: if I had to guess, more complex than desktop, less complex than Qualcomm