XviD and SSE 4.1

[gurabli]

Hi!

I have an Intel E8400 CPU that has the SSE 4.1 instructions. I wonder that XviD can use these new SSE functions or not? Is there any benefit at all?

Regards,
gurabli

[Dark Shikari]

Xvid is no longer being developed, so it certainly won't make use of them regardless.

More importantly, Xvid doesn't use an exhaustive motion search for any purpose, so it wouldn't be useful anyways.

And even more importantly, if it did, it'd use x264's SEA algorithm, which is much faster than the SSE4 equivalent.

[gurabli]

Xvid is no longer being developed, so it certainly won't make use of them regardless.

More importantly, Xvid doesn't use an exhaustive motion search for any purpose, so it wouldn't be useful anyways.

And even more importantly, if it did, it'd use x264's SEA algorithm, which is much faster than the SSE4 equivalent.

Thank you for your answer!

[henryho_hk]

I recall that DivX has that exhaustive motion search especially designed to demonstrate the power of SSE 4.1.

[Dark Shikari]

I recall that DivX has that exhaustive motion search especially designed to demonstrate the power of SSE 4.1.Or, equivalently, lack of power. :D

[LoRd_MuldeR]

I recall that DivX has that exhaustive motion search especially designed to demonstrate the power of SSE 4.1.

1. The benefit of an "exhaustive" motion search is very small and usually not worth the heavy speed-drop

2. As mentioned before, x264's SEA (using SSE2) is much faster than SSE4 ESA ...

So there's no benefit in SSE4 for video compression...

[Manao]

So there's no benefit in SSE4 for video compression... SSE4 isn't summarized to mpsadbw. And video compression isn't summarized to x264.

[LoRd_MuldeR]

SSE4 isn't summarized to mpsadbw. And video compression isn't summarized to x264.

Well, I'm not an expert of x86 assembler coding.

But it was proven that "exhaustive" motion search can be implemented more efficient than using the SSE4 instruction that was especially designed for that purpose (MPSADBW).
And are there other instructions in SSE4 that are actually helpful for the task of video encoding? I did not read anything about such yet. Only the "exhaustive" search is mentioned all the time...

[Dark Shikari]

Well, I'm not an expert of x86 assembler coding.

But it was proven that "exhaustive" motion search can be implemented more efficient than using the SSE4 instruction that was especially designed for that purpose (MPSADBW).
And are there other instructions in SSE4 that are actually helpful for the task of video encoding?Yes, but as far as I can tell, only one, and one has to be rather creative with it, and not on Penryn.

[Manao]

Only the "exhaustive" search is mentioned all the time... Thus my jumping in. There are no "must have" / killer instructions in SSE4, but pblend should help deblocking, pmovzxbw should help for adding back residual, and I must that I have already loathed two or three times MMX for not having a packusdw and more generally instructions working on unsigned words.

[Dark Shikari]

Thus my jumping in. There are no "must have" / killer instructions in SSE4, but pblend should help deblocking, pmovzxbw should help for adding back residual, and I must that I have already loathed two or three times MMX for not having a packusdw and more generally instructions working on unsigned words."Should" doesn't mean "does". pmovzxbw is by a wide margin slower than movq+punpcklbw, for example, despite being a 1/1 latency/throughput instruction.

I'm still trying to get Intel to explain this one to me :D