This started as a desperate attempt to lower the memory requirement
of error diffusion, but later it turns out that this change also
improved the rendering performance a lot (by 40% as I tested).
Errors was stored in three uint before this change, each with 24bit
precision. This change encoded them into a single uint, each with 8bit
precision. This reduced the shared memory usage, as well as number of
atomic operations, all by three times.
Before this change, with the minimum required 32kb shared memory, only
the `simple` kernel can be used to render 1080p video, which is mostly
useless compare to `--dither=fruit`. After this change, 32kb can
handle `burkes` kernel for 1080p, or `sierra-lite` for 4K resolution.
This is a straightforward parallel implementation of error diffusion
algorithms in compute shader. Basically we use single work group with
maximal possible size to process the whole image. After a shift
mapping we are able to process all pixels column by column.
A large ring buffer are allocated in shared memory to speed things up.
However the size of required shared memory depends linearly on the
height of video window (or screen height in fullscreen mode). In case
there is no enough shared memory, it will fallback to `--dither=fruit`.
The maximal allowed work group size is hardcoded as 1024. Ideally we
could query `GL_MAX_COMPUTE_WORK_GROUP_INVOCATIONS`. But for whatever
reason, it seems most high end card from nvidia and amd support only
the minimal required value, so I guess we can stick to it for now.
Bin Jin