While stepping through the bowels of the GL implementation for bug 1072145, I noticed that on the ReadTexImage() path for reading back a renderbuffer, we end up on a fallback copy path that does a pixel-by-pixel copy with a scissor test per pixel. That's going to be ... slow. We're hitting this code through layerscope in my testing, so I don't care about that for now. But IIRC, on platforms without GL fences, we do approximately the same readback for webgl canvases. If we end up on the same slow path, webgl will probably suffer noticeably. (We end up on the slow path because the source texture is stored in a special tiled format for the VideoCore GPU, but the destination surface is normal GL_RGBA.)
This bug isn't relevant anymore since we'll be targeting a different GL stack in the future.
Status: NEW → RESOLVED
Last Resolved: 4 years ago
Resolution: --- → WONTFIX
You need to log in before you can comment on or make changes to this bug.