do ffdshow support sse4 or sse4a ?

[221122]

do ffdshow support sse4 or sse4a ?
( on cpu )

do ffdshow support 4 core ?

[Ryokurin]

There is a pretty long thread on ffdshow tryouts in the alternative codecs forum available here (http://forum.doom9.org/showthread.php?t=120465)

It may have changed but in a older thread sse4 didn't really make that big of a difference in speed.

[221122]

i am not see where the sse4 or sse4a ?
not reference to sse4 or sse4a or sse5 ..

sse4 or sse4a can bring the video to new line

[clsid]

ffdshow does not use SSE4/SSE4a, nor do any other video decoders that I know of.

SSE4 is NOT very useful to make video encoding/decoder faster. Whatever you have heard or read is wrong.

[LoRd_MuldeR]

Also people have to understand, that the "ffdshow(-tryouts)" project does not develop any decoders or encoders. The purpose of that project is to build a DirectShow wrapper around libavcodec and a number of other decoder/encoder libraries. This way people can use the famous OpenSource encoders/decoders in DirectShow-based application. Hence ffdshow will not "support" SSE4/SSE4a nor anything else, unless it's implemented in libavcodec/ffmpeg...

[221122]

AND WHAT ABUT "SSSE3" WHAT CPU support that ?

this "ssse3 " how this make more good the video or audio ?

[221122]

you see in the "jpg "

that sse4 good for video and hdtv

[221122]

you see in the "jpg "

that sse4 good for video and hdtv

i find where the SSSE3 IS IN CORE
http://images.anandtech.com/reviews/motherboards/asrock/1600penryn/650i9x435e8400a.png

[Sharktooth]

please... stop it...
ffshow is a "wrapper" for libavcodec and no, libavcodec doesnt support SSE4 coz speedwise, they're almost useless for video encoding/decoding since better routines do not even need those pesky instructions.
if you dont believe it, learn how to code and find it by yourself.
next time :search: and read up before asking (as per forum rules (http://forum.doom9.org/forum-rules.htm))

[221122]

Sharktooth please... stop it...



Sharktooth - if you don't want to answer don't answer !
:devil:
i am not you friend .

please " Sharktooth " don't answer me again .

[Sharktooth]

read again... i have answered... just stop asking the same questions which have already been answered in this and other threads then.
as i said :search: and read up before asking as per FORUM RULES (http://forum.doom9.org/forum-rules.htm).
your questions and persistence are just an offense for the ffdshow and libavcodec developers since you're supposing you know more about coding than them...

[Konrad Klar]

I wonder where such instructionts would be actually usable (if not in audio/video processing).

[LoRd_MuldeR]

From Wiki:
Unlike all previous iterations of SSE, SSE4 contains instructions that do operations not specific to multimedia applications. It features a number of instructions whose action is determined by a constant field, and, in a rather surprising move, a set of instructions which take XMM0 as an implicit third operand. In addition, SSE4 totally lacks support for operations on 64-bit MMX registers; SIMD integer operations can be carried out on 128-bit XMM registers only.

Also the new instructions can be useful for "dumb" implementations, but at the same time more smart implementations (not using any of the new instructions) are still faster! One example is the "full search" in DivX, which is based on SSE4. It might be faster than any plain "full search" that works without SSE4, but it can be outperformed by better (more optimized) search algorithms easily. So "new" instructions are no guarantee for actual speed-ups and especially they cannot replace smart implementations!

[Konrad Klar]

@LoRd_MuldeR
Thanks for comprehensive answer.

[foxyshadis]

Those images were gigantic, please don't to that again. Copy and paste the text instead, or link to an image.

[Dark Shikari]

As of a recent revision by Pengvado/Akupenguin, ffh264 uses SSE2 and SSSE3 for motion compensation, giving a ~4% overall speed boost over pure MMX (20-30% in just those functions).

[Manao]

SSE4 is not worthless. It's impact on speed is small, but so is the impact on speed of every single instructions since SSE2 (and even SSE2 started to shine only on Core 2 anyway). But don't expect miracles either, it won't speed up the encoder by 50%. The speed up obtained by Pengvado is enormous (truly), yet, overall it's only a 4% speed up.

Likely, SSE4 would perhaps give a 1% speedup overall, provided a developper spend a fair amount of time coding for it. For the same amount of time, he might instead implement efficient threading in ffh264, effectively bringing the most out of your Core 2 / X2 / Penryn / Phenom, without caring about specificity of instruction sets (SSE4 won't exist anytime soon on Amds)

[LoRd_MuldeR]

For the same amount of time, he might instead implement efficient threading in ffh264

I'd prefer the latter :D

[_xxl]

For the same amount of time, he might instead implement efficient threading in ffh264
Parallel Scalability of H.264:
http://ce.et.tudelft.nl/publicationfiles/1474_68_h264_scalability_hipeac.pdf

[Manao]

The devs already know how to do it. They just lack time and/or motivation. The programming behind the article, though based on ffmpeg, was afaik never made into a patch (sadly). The idea behind the article isn't new, it's equivalent to x264's threading algorithm (except they did give it a catchy name : "3D wave"). It has already been debated at length on ffmpeg's mailing list.

[Dark Shikari]

If you want ffh264 threading, get together ~$15000 or so and make an offer; for a good price someone will definitely do it, perhaps even akupenguin.

[Shinigami-Sama]

If you want ffh264 threading, get together ~$15000 or so and make an offer; for a good price someone will definitely do it, perhaps even akupenguin.

how about a big shiney meathook and a picture of their home?
:eek:


but 15k?
seems stiff

[221122]

do e8400 have ssse3 ?

if not do i lose something with ffdshow ?

or wait to "amd " for more quality video ?

report how sse4 can give to video more
http://arstechnica.com/news.ars/post/20070417-intel-details-penryn-performance-sse4.html?rel

[clsid]

report how sse4 can give to video more
http://arstechnica.com/news.ars/post/20070417-intel-details-penryn-performance-sse4.html?rel
If you would have read the comments in this thread then you would have known that the DivX benchmark in that test means nothing at all. The DivX people added a NEW algorithm (that is not very useful) which happens to work very well with SSE4 instructions. It is a marketing stunt for which you and other suckers have fallen.

[Sharktooth]

@221122: SSE1/2/3/4/.... do NOT give ANY quality in decoding...
the numbers on those reviews are simply not indicative... they came from particularly chosen scenarios and they do not represent the everyday usage.
how many times should WE keep answering the same questions? After ONE PAGE full of "SSE4 are not so good" you keep posting "reports" of SSE4 presumed miracle advantages... STOP IT and :search: FFS!

[221122]

Sharktooth- again

sharktooth - if you don't want to answer don't answer !

i am not you friend .

please " Sharktooth " don't answer me again .

again

Sharktooth stop answer me !

Sharktooth you don't want take part in this concatenation
SOO DON'T !

Sharktooth i am not your doctor ! i can't give you drug .

Sharktooth - when I TELL YOU STOP BOTHER ME I MEAN THAT !

[Sharktooth]

actually, i gave you the answers...
when i mean STOP ASKING THE SAME STUFF AGAIN AND AGAIN I MEAN THAT.
you already got answers!!! dont pollute this forum always with same questions. THANX.

[221122]

edit mod: removed profanity.

[Sharktooth]

dont want answers? dont ask!
i just gave you answers to your questions and i've pointed you to the forum search function to make you reach the correct threads. read the forum rules (http://forum.doom9.org/forum-rules.htm) again.

[221122]

edit mod: removed profanity.

[Sharktooth]

actually you're the one acting like a child.
for the record, this is a public forum and if you put questions here, what can you expect? ppl try to answer...
but if you keep asking the same things over and over even if you got answers, dont read the forum rules (you violated another one with your last post...), keep evangelizing stuff it was already been analyzed months ago and reported as "marketing stunt", etc... what can you expect?

[Wilbert]

@221122,

Behave yourself!

[Dark Shikari]

My reaction after reading this thread:

http://i25.tinypic.com/vgshg3.jpg

[Shinigami-Sama]

@221122

http://shinigami.ca/stuffage/wrong18.jpg

[foxyshadis]

Thread has run its course and then some.