Save blend time and memory when rendering distant pieces of text

[JustAMan]

This happens when e.g. subtitles are displayed simultaneously on top and at bottom.
Before the change subtitles were rendered in the whole bounding rect, which could end up taking the whole viewport.

Now I change this to splitted rendering (splitting the viewport in 3x3 grid and then merging intersecting rects). This reduces memory footprint and blending time for subtitles which have far spaced parts.

Notes

• src/SubOctpInterface.cpp and src/SubOctpInterface.js are generated
• renderBlendPart() is almost a cut-paste portion of renderBlend(), with pretty minimal changes inside

[dmitrylyzo]

Almost didn't look on blend itself, because I believe that you just moved it.

[JustAMan]

Almost didn't look on blend itself, because I believe that you just moved it.

Here are the relevant changes:
https://github.com/jellyfin/JavascriptSubtitlesOctopus/pull/9/files#diff-a1bb6e1327d5352744b2f8253e342183R587-R589
https://github.com/jellyfin/JavascriptSubtitlesOctopus/pull/9/files#diff-a1bb6e1327d5352744b2f8253e342183R625-R651

[dmitrylyzo]

https://github.com/jellyfin/JavascriptSubtitlesOctopus/pull/9/files#diff-a1bb6e1327d5352744b2f8253e342183R625-R651

Yeah, I looked this.
You search for biggest buffer to minimize allocation, right?
Suppose we have render chunks from smaller to bigger. Can't be that a smaller chunk will steal biggest buffer from next bigger chunk. In worst case, all buffers will be screen size, and this will become irrelevant. But maybe search for smallest bigger ("just enough size") and biggest smaller buffer.

Nm, I read it wrong.

[JustAMan]

Not exactly. I always search for smallest buffer, but actually detect two kinds of them: one (called bigBuffer) is big enough to hold the needed amount without growing it, and another (smallBuffer) is the smallest available (so that I don't take some bigger which could be re-used by other parts without growing that).

So the strategy is: take the smallest buffer which can hold desired memory without growing; if it doesn't exist, take just the smallest buffer to maximize the ability of other parts to re-use other buffers.

Essentially re-allocating the buffer should be avoided at all costs to minimize the work for GC IMHO. They're all small enough to not cause significant memory consumption compared to pre-rendered cache later on.