Quote:
|
Like I said, when we did WME x64 edition, it was a whole lot faster than the 32-bit version. But once we looked at the optimizations that were done, there was very little that was 64-bit specific in that.
|
It's also the case for x264. Only a handful of SIMD functions make use of more than 8 SIMD registers. So most of the roughly 10% boost is brought simply by compiling for 64 bits (more registers, faster calling convention). Imho, the only scenario in which x64 wouldn't help that much is when a piece of code is memory bound.
Finally, since x264 is both open source and already insanely optimized for 32bits, it's a good project to test what speed ups x64 can manage. As you said, if a closed source application gains 50% speed up in x64, we can't conclude anything since the whole application may have been rewritten while being converted to x64.