dedandClaude Opus 5.5 ba04ab70a0 3D text and shapes rasterise only the part the output shows (#1388) ...

A 3D text or shape layer close to the camera was rasterised whole at its on-screen scale, up to
~16 million pixels (a quarter gigabyte as f32) per frame worker, though the frame showed a small
part of it; behind the camera it was rasterised and not drawn. With 16 preview workers one such
layer took the app to 8-14 GB.

Renderer::raster_clip projects the output frame back onto the layer's plane (as the Classic 3D
compositor draws it) and rasterises only those cells of an 8 x 8 grid over the content, in a
window of the whole raster's pixel grid; nothing when the content is behind the camera. Cells
keep a moving camera on one cached raster. Layers whose off-frame pixels can matter (effects,
styles, masks, motion blur, depth of field, lights, Advanced 3D, per-character 3D, collapsed 2D
transforms) are rasterised whole as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

2026-10-20 18:11:20 +03:00

..

src

3D text and shapes rasterise only the part the output shows (#1388)

2026-10-20 18:11:20 +03:00

tests

Pixel Motion Blur: synthesize the blur from motion vectors and honour Vector Detail (#535).

2026-10-11 13:32:13 +11:00

Cargo.toml

Merge main into IO-2/SHP-5/OUT-2/PRV-2 work

2026-10-02 22:39:47 -04:00