WebGL Architecture in FlashFX

FlashFX's rendering pipeline operates in the following stages:

Each operation that submits geometry to the GPU is a draw call. Fewer draw calls means less CPU-GPU communication overhead and faster rendering.

FlashFX batches elements that share:

• -The same blend mode
• -No intervening alpha compositing requirements
• -Compatible shader programs

Elements that break batching:

• -Any element with a non-Normal blend mode (requires isolation)
• -Groups set to Isolated compositing mode
• -Elements with certain filter types (require intermediate offscreen render passes)
• -Transparent elements composited over other transparent elements (order-dependent)

When an image is imported or a rasterized text/shape is generated, it is uploaded to GPU memory as a texture. This upload happens:

• -Once per unique image asset
• -When an asset's content changes (e.g., after a filter parameter changes)
• -On session start when loading a project with embedded assets

Texture upload is expensive. Avoid situations where the GPU texture must be invalidated and re-uploaded on every frame (e.g., animating a filter that requires a full re-rasterize of vector content).

Certain operations require rendering to an intermediate offscreen framebuffer before compositing into the final output. Each offscreen buffer consumes GPU memory equal to its dimensions x 4 bytes per pixel.

Operations requiring offscreen buffers:

• -Non-Normal blend modes (the layer must be isolated before blending)
• -Blur effects (blur requires a copy of the pre-blur pixels)
• -Some distortion effects (displacement needs both the target and the map simultaneously)
• -Groups with isolated compositing
• -Alpha masks

A complex composition with many elements using non-Normal blend modes or blur effects may require many simultaneous offscreen buffers. On devices with limited VRAM, this can cause rendering slowdowns, quality reduction, or in extreme cases, a tab crash.