A couple of years ago High Quality resizing (Lanczos resampling) was introduced (https://bugzilla.mozilla.org/show_bug.cgi?id=486918) to Firefox. It was done via borrowing corresponding code from Google Chrome's skia and Chrome itself. Apparently, by that time Chrome's implementation was suffering from out-of-bound errors in SIMD case (https://code.google.com/p/chromium/issues/detail?id=181072 don't let the title confuse you, it was changed), so SSE support in FF was deliberately turned off since the very beginning. And still remains off. Apparently, currently Chrome supports SSE for all major OS, so we should be able to turn it on quite cheaply: SSE implementation of all major functions is already in the source tree and (almost) up to date.

tbpl: https://tbpl.mozilla.org/?tree=Try&rev=b05a968d8a75

Comment on attachment 8480711 [details] [diff] [review] This patch turns SSE2 on for image resizing Review of attachment 8480711 [details] [diff] [review]: ----------------------------------------------------------------- This needs to check for SSE2 support at run time otherwise we'll get crashes on machines that don't have it. i.e. Factory::HasSSE2()

All tests: https://tbpl.mozilla.org/?tree=Try&rev=7f80198d59df

Alright, everything looks good on try! Thanks so much for working on this, Artem! Pushed: https://hg.mozilla.org/integration/mozilla-inbound/rev/16c3207fa04c

Looking at this again. Where did the 'int begin, int end' arguments come from? I don't see anything like that in the chrome code.