User Agent: Mozilla/5.0 (Windows NT 6.1; WOW64) AppleWebKit/535.19 (KHTML, like Gecko) Chrome/18.0.1025.168 Safari/535.19 Steps to reproduce: Went to http://jsperf.com/canvas-global-composite-operations Actual results: The "source-atop" test achieved over 248,000 ops/sec. Every other global composite operation achieved under 2,000 ops/sec Expected results: The other GCO's should have achieved higher than 2,000; in Firefox 11, they most achieved in the 60,000 range (see attached graph) This issue affects FF 12 and FF 13

Oops, I said "source-atop", but meant "source-over" "source-over" is the default, and the one that's really fast.

I use "lighter" extensively because it's the only remotely efficient way to approximate the GL_MULTIPLY blend equation in canvas. IIRC based on discussions with the GFX guys this problem is caused by Direct2D not exposing any control over blend operations, so any non-source-atop blend operation requires an expensive Direct3D fallback.

For a real-world example, the level editor in http://playescapegoat.com/ gets absolutely murdered by this. Even caching the rendered output isn't enough.

Note: This is still slow with Firefox 23 + Direct2D/Azure canvas.

Note also: CompositeOperations like "lighter" works pretty fast in other browsers like Chrome. It would be a nice feature for particle effects.

This is still an issue. I've checked the original test case at http://jsperf.com/canvas-global-composite-operations Current beta (38.0.5) renders all composition modes other than "source-over" at 4% of the speed of "source-over". In contrast, Chrome 42 renders at 40%-70% of the speed of "source-over" (which is 10% faster than Firefox on same PC anyway). Thus, Chrome renders upto 10-20 times faster.