This is an optimization that unfortunately breaks some assumptions related to how interning works, which causes more tile invalidations than needed. Since this optimization was added, the DL encoding format has been optimized significantly and it doesn't appear to regress anything from the limited perf testing so far. If it does show up as a regression anywhere, we can apply a similar optimization but in a different way that doesn't break interning (by removing the leaf of a prim clip-chain when it's redundant). Differential Revision: https://phabricator.services.mozilla.com/D280809
3 lines
79 B
INI
3 lines
79 B
INI
[perspective-split-by-zero-w.html]
|
|
fuzzy: maxDifference=0-1;totalPixels=0-40
|