x264-> is there a optimized Athlon64 x2 (VFW-)Build?

[Morpheus_xx]

as subject says, I have an Athlon 64 x2 dual core cpu.

Is there a special compiled version for those cpu?

And does the vfw-Version support multithreading?

Btw, the cli-version only get used up to 80% of cpu when using --threads 2 option?

[Doom9]

x264, just like xvid and many other open source software, contains mostly non CPU specific optimizations. Consequently, it's fast on all platforms whereas the intel sponsored SSE optimized software generally runs rather badly on older hardware.

Btw, the cli-version only get used up to 80% of cpu when using --threads 2 option?
http://forum.doom9.org/showthread.php?t=104535

[akupenguin]

x264, just like xvid and many other open source software, contains mostly non CPU specific optimizations. Consequently, it's fast on all platforms whereas the intel sponsored SSE optimized software generally runs rather badly on older hardware.
It's not intel vs amd. In the few functions where that matters, x264 has both versions and selects at runtime.
But bob0r and Sharktooth build for win32 only. 64bit x264 is about 10% faster.

[Doom9]

64bit x264 is about 10% faster.Since you mention it, I've always wondered why exactly that is (it seems to apply to many video related things). I presume one thing is that longs are natively supported to reading/writing them is more efficient, but does that really give a whopping 10% boost? Or is the architecture more efficient? Or are there other reasons?

[squid_80]

Possible reasons:
- Fastcall calling convention. Parameters passed via registers instead of in memory via the stack.
- More general purpose registers, so again less memory usage.

[Morpheus_xx]

64bit x264 is about 10% faster.

Where can I download such versions?

Does it require Windows 64bit or does it run on Win32?

[Doom9]

a little searching wouldn't hurt, now would it?

[Morpheus_xx]

a little searching wouldn't hurt, now would it?
of course it wouldn't hurt, but I get 22 threads when searching for "x264 64bit".

And most of them are questions and no answers.

So a direct link to 64 bit builds would be useful (maybe also in the sticky's)

PS: the only link I found ftp://squid80.no-ip.com/ is currently not available

[max-holz]

Beg pardon perhaps I'am out of topic, I want to know which are the differences between Sharktooth's build and bob0r's?

[omion]

of course it wouldn't hurt, but I get 22 threads when searching for "x264 64bit".

And most of them are questions and no answers.

So a direct link to 64 bit builds would be useful (maybe also in the sticky's)

PS: the only link I found ftp://squid80.no-ip.com/ is currently not available
I just put up some of my x64 builds here (http://omion.dyndns.org/x264x64/), and as akupengin says, it's about 10% faster than the 32-bit builds. My builds only output to raw and Matroska files, not MP4. Also, getting it to play nicely with AVIsynth takes a bit of work. See this thread (http://forum.doom9.org/showthread.php?p=767659).

Note, however, that 64-bit builds REQUIRE 64-bit Windows. Having a 64-bit processor is not enough.

[akupenguin]

Possible reasons:
- Fastcall calling convention. Parameters passed via registers instead of in memory via the stack.
- More general purpose registers, so again less memory usage.
Agreed.
e.g. x264_pixel_sad_16x16_mmxext() on an athlon64 in 64bit mode is 60 cycles, in 32bit mode is 66 cycles. They're identical except for calling convention.

[squid_80]

PS: the only link I found ftp://squid80.no-ip.com/ is currently not available
The server should be available again now, although the connection seems very slow. Methinks one of the other computers in the house might be zombified, gonna go pull some plugs and see if things pick up.

Agreed.
e.g. x264_pixel_sad_16x16_mmxext() on an athlon64 in 64bit mode is 60 cycles, in 32bit mode is 66 cycles. They're identical except for calling convention.Isn't it possible to declare the functions as fastcall and modify the assembly code to match? I thought xvid did this with some functions but I might be imagining things.

[akupenguin]

I don't see any such in xvid, but it is possible. Only uses 2 registers for arguments on x86-32, but should be better than nothing.

Edit: nope, tried fastcall on sad,ssd,satd,mc,avg and there was <0.2% difference in overall speed.