x264 XOP and FMA4 optimisations ?

[Romario]

Hi guys, I read somewhere that our Dark Shikari, our dearest developer, started work on heavy optimizations for AMD FX Bulldozer processor.

Dark Shikari, is it true, and can you explain us the method and tell us what is your goal here ?

[Dark Shikari]

It's the same method as always: look at the new instructions, think about how to use them, write code, benchmark, repeat. This happens every time a new instruction set is released.

[Romario]

Ok, but tell us when we can expect XOP optimised x264 encoder, any time soon ? Some builds maybe ?

About speeding-up the encoding process, what you expect ?

[Dark Shikari]

Some changes are already up on github. Don't expect much, maybe 1-2%; microarchitecture changes are vastly more important and valuable than new instructions, particularly in an application with functions as complex and diverse as x264.

[mandarinka]

It should however benefit from all the ssse3/sse4 code that is already in, too :) At least let's hope so.

[Romario]

Well, tomorrow is BIG day: official launching of the AMD FX processors (Bulldozer). I somewhat sense that our dear dev don't like to speak about XOP optimisations before it's ready. Am I right, Dark Shikari ?

[nm]

I somewhat sense that our dear dev don't like to speak about XOP optimisations before it's ready.

Didn't he already tell you what to expect.

[Romario]

Well, I heard various rumors, so...