I've found that one of the the slow points for my video-decoding use case on IE 11 is the Math.imul polyfill. libvpx's VP8 and VP9 decoders do a lot of integer multiplication, as it turns out, but never (?) in ways that would trigger 32-bit integer overflows...
I added a post-processing step to the asm.js builds of my VP8 and VP9 codecs, which finds all uses of the minified var assigned from global.Math.imul and replaces it with straight multiplication: https://github.com/brion/ogv.js/blob/master/buildscripts/strip-imul.js With this in, I'm seeing roughly a 15-25% speedup in decoding in IE11 versus using the Math.imul polyfill! (Safari and Edge versions which include Math.imul natively don't get any speedup, and newer versions will get the Web Assembly builds anyway.) This of course is a dangerous trade-off for general code -- it's possible something will mis-optimize a legit multiplication that way -- but I thought it might be an interesting tidbit some others might need. -- brion -- You received this message because you are subscribed to the Google Groups "emscripten-discuss" group. To unsubscribe from this group and stop receiving emails from it, send an email to emscripten-discuss+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
