“They also want to improve threading, accelerate dav2d for less common architectures, such as PPC, SSE2, RISC-V, or AVX-512, improve the C code base with various tweaks, use more GPU decoding when possible, provide a usable API for devs to integrate dav2d into their projects, and port dav2d to most platforms.”
I’m not making demands of free software; all of that development is appreciated.
But it does seem to be CPU-first, like dav1d and previous efforts. And I know from previous projects, GPUs have many constraints that make offloading a CPU-centric project difficult, unless it’s the #1 priority from the start.
I’m just saying it’d be interesting another AV2 project that shot for pure, or as-pure-as-possible, GPU shader decoding. It would make AV2 a whole lot more accessible than CPU-centric decoding.
What about some generic GPU decode?
Like: https://aomedia.googlesource.com/av1-xbox-one/
“They also want to improve threading, accelerate dav2d for less common architectures, such as PPC, SSE2, RISC-V, or AVX-512, improve the C code base with various tweaks, use more GPU decoding when possible, provide a usable API for devs to integrate dav2d into their projects, and port dav2d to most platforms.”
It’s in the plans (Step 15) but not forgotten.
It’s the last step, though.
I’m not making demands of free software; all of that development is appreciated.
But it does seem to be CPU-first, like dav1d and previous efforts. And I know from previous projects, GPUs have many constraints that make offloading a CPU-centric project difficult, unless it’s the #1 priority from the start.
I’m just saying it’d be interesting another AV2 project that shot for pure, or as-pure-as-possible, GPU shader decoding. It would make AV2 a whole lot more accessible than CPU-centric decoding.