Nikolay Vabishchevich

> There's 5 unused values I had a similar idea, but with those values (plus center zero) used for permutations of (±2, 0, 0). If it's reusing, better to reuse...

> If I'm understanding correctly this means that for example 5 of the 9 GB states could have an R of +2. Is there a way to detect those 5...

So my intuition about value distribution is right. Maybe even better would be to use original 64 values and distribute them more symmetrically.

I've tried it under different OSes and it works fine with OpenMPI under Linux (ubuntu 18.04). Looks like most MPI implementations can't handle this.

I'm definitely against aliasing seemingly different variables with each other. Also we have `ASS_DVector` for x/y-pairs, better to use it. Also, I'm not sure, but isn't [anamorphic blur branch](https://github.com/libass/libass/commits/anamorphic-blur) related...

I'm planning to work on rough clipping at the rasterization stage. Tiled rendering can help in this case, but only if each tile is cached and processed individually which can...

I should mention that there are several optimizations that can induce shift of boundaries and this line is only one of it. So this PR is incomplete as is and...

What I think should be done: - Instead of `CONFIG_RENDER_LQ` something like `CONFIG_LARGE_SCREEN` or `CONFIG_HI_DPI` will be better. - Spatial precision constants should be multiplied by the same factor (2...

> Given that the original motivation was to provide accurate bitmap boundaries, isn’t naming it `...LARGE_SCREEN` or `...HIGH_DPI` kinda missing the point? The original motivation is only about pixel-perfect axis-aligned...

Try `buf[x] = val > 127 ? 255 : 0`. But note that it's not the only place that should be modified, similar change should be made for `ass_fill_halfplane_tile*` functions.