chocolate42

> I chose ZLib because its libraries are so widely available and for simplicity of adoption. LZ4 and ZSTD are widely available too, the only likely places that they aren't...

It's unfortunate that ungetc only guarantees one byte of pushback, otherwise with a minor change to libflac external programs like flac could implement basic chaining support with libflac relatively easily....

I'm having a play trying to get this to work, but as I don't know OpenCL or why sum64 is used this is a crapshoot. My initial read was that...

Probably there is a bug in the fp64 implementation, it's still experimental after all. I'll look into it ability permitting and add an issue to rusticl's tracker if appropriate. Probably...

Mesa 23.3.5 the latest arch packages. But I'm not even on arch host this is through a temp arch environment with distrobox which you could do too. It's as simple...

This is with the latest sources: ``` 📦[f40@f39 f39]$ RUSTICL_ENABLE=radeonsi RUSTICL_FEATURES=fp64 meson devenv -C ../mesabuild ../f39/gpuowl/build-debug/gpuowl -prp 77936867 20240222 12:49:34 GpuOwl VERSION v7.5-6-gd522826 20240222 12:49:34 GpuOwl VERSION v7.5-6-gd522826 20240222 12:49:34...

If I'm interpreting things correctly, rusticl doesn't yet support the opencl 2.0 extension that adds things like work_group_reduce_add() https://registry.khronos.org/OpenCL/sdk/3.0/docs/man/html/workGroupFunctions.html Corresponding to this feature on the tracker: https://mesamatrix.net/#RusticlOpenCL2.0_Extension__Workgroup_Collective_Functions

I'm pretty sure opencl 2 is required but could be wrong.

> Is using variable blocksize adding an overhead to frames (or subframes) of the FLAC file? Only a minor overhead of a few bytes per frame, thanks to fixed blocksize...

Also tried setting a 10MiB+8 buffer in the decoder reader (10mb8r). Same re-encode test, average of 10 runs: ``` HDD SSD 10mb8 9.561 7.244 10mb8r 9.552 7.083 ``` There might...