WMP11 Beta 2? Did WMV9 AP change in it?

[mpgxsvcd]

With the WMP11 Beta 2 that was recently released did WMV9 AP change in it? Are there any new settings for AP? Will there ever be more settings added?

[zambelli]

With the WMP11 Beta 2 that was recently released did WMV9 AP change in it? Are there any new settings for AP? Will there ever be more settings added?
Yes. No. No. :)

[check]

Is there a public changelog for MS codec changes?

[zambelli]

Is there a public changelog for MS codec changes?
Unfortunately, no. It's something that personally irks me quite a bit at Microsoft. :(

There were about a dozen bugs fixed between Beta1 and Beta2. The most important ones were related to interlaced encoding support. It was actually rather broken in Beta1 (i.e. only field interlaced encoding was used), but is fixed in Beta2. The encoder will now respond correctly to VideoType setting values.

We're hoping to get some last minute encoder perf improvements into the final WMP11 release, but aside from that - the Beta2 codec is more or less what will ship in final WMP11.

[check]

Thanks for the info :D

[Sharktooth]

You should concentrate on the DECODER perf improvements.
Decoding speed on non SSE2 systems (athlon XPs) is quite crippled...

[zambelli]

You should concentrate on the DECODER perf improvements.
Decoding speed on non SSE2 systems (athlon XPs) is quite crippled...
Don't you think it's time to upgrade? ;)
Seriously though, it'd be hard to justify spending time on making SSE or MMX optimizations if SSE2 optimizations are already in place. I'm not saying it's completely invalid - I'm just saying there are probably more important work items.

[Wilbert]

First it was a XP, now everyone is supposed to have a processor with SSE2 optimizations?

[zambelli]

First it was a XP, now everyone is supposed to have a processor with SSE2 optimizations?
Nearly every computer sold in stores today has SSE2 optimizations. Why is it so unreasonable to optimize code for today's and tomorrow's technology?
If Adobe went and made x64 optimizations in Premiere, would you object and say that they're forcing you to upgrade to 64-bit processors?

[bond]

libavcodec uses no sse2 at all for decoding h.264 with very good performance

[Wilbert]

Nearly every computer sold in stores today has SSE2 optimizations. Why is it so unreasonable to optimize code for today's and tomorrow's technology?
If Adobe went and made x64 optimizations in Premiere, would you object and say that they're forcing you to upgrade to 64-bit processors?
C'mon zambelli, is it so hard to be a bit reasonable?

Nobody is saying it is wrong or bad to add sse2 optimizations, 64 bit optimizations or developing stuff for XP. It's wrong and bad to *not* add mmx/sse optimizations at the same time, or make sure stuff is also working on a W2K.

But i guess you find this unreasonable (meaning economically not justified with respect to the costs)? I guess we should buy a new PC every six months?

[zambelli]

Nobody is saying it is wrong or bad to add sse2 optimizations, 64 bit optimizations or developing stuff for XP. It's wrong and bad to *not* add mmx/sse optimizations at the same time, or make sure stuff is also working on a W2K.
Nobody said SSE or MMX optimizations don't get made. If there are sections of the code that are better optimized for SSE2, that's a completely different story. Having each line of code separately optimized for MMX, SSE, SSE2, SSE3, x64 and who knows what else - isn't the most efficient way to use a developer's time.

But i guess you find this unreasonable (meaning economically not justified with respect to the costs)? I guess we should buy a new PC every six months?
SSE2 has been around since 2001. That's not exactly 6 months.

Anyway, before this turns into a bickering match... If Sharktooth thinks SSE decoding is waaay slower than SSE2 decoding (on a similar system), I'd like to see some data. What are we talking about here? 5%? 50%?

[foxyshadis]

akupenguin is undoubtedly one of the very few people on the forum who can authoritatively state the benefits of different instruction sets for different types of functions, and a codec as a whole. Unless someone wants to comb the svn for his various "20% mvpred speedup, .8% total speedup" comments and add them up.

I hate to bring it up but Microsoft does have a strong interest in selling more new machines, at least, much more so than open-source. So if that dovetails with the policy, who's going to complain?

[zambelli]

I hate to bring it up but Microsoft does have a strong interest in selling more new machines, at least, much more so than open-source. So if that dovetails with the policy, who's going to complain?
Because that's not how decisions are made at that level. If a developer is planning optimization work on some code, "selling more new machines" is not going to be on top of his/her list of considerations.

[Sharktooth]

Well, as you may already know, Athlon CPUs are VERY fast with MMX stuff, while Intel ones are faster with SSE.
So, adding MMX optimizations (or a dual code path) will help ALL AMD users even the ones with the very latest CPUs.
Such optimizations would be also very useful in the Visual Studio compiler... but that's a different story...

[benwaggoner]

Nearly every computer sold in stores today has SSE2 optimizations. Why is it so unreasonable to optimize code for today's and tomorrow's technology?
If Adobe went and made x64 optimizations in Premiere, would you object and say that they're forcing you to upgrade to 64-bit processors?
Speaking of which, Adobe dropped support for pre-SSE2 systems entirely with Premiere Pro 2.0.

[zambelli]

Well, as you may already know, Athlon CPUs are VERY fast with MMX stuff, while Intel ones are faster with SSE.
Athlon64's and newer AMD CPUs support SSE2. So it's not even an AMD vs Intel thing... It's just that when a call has to be made which instruction set to optimize for in a limited amount of time: it makes most sense to optimize for today's technology and not yesterday's technology.

[bond]

what parts of the m$ vc-1 decoder have a benefit in being tuned for sse2 instead of mmx/sse?

[Wilbert]

Athlon64's and newer AMD CPUs support SSE2. So it's not even an AMD vs Intel thing... It's just that when a call has to be made which instruction set to optimize for in a limited amount of time: it makes most sense to optimize for today's technology and not yesterday's technology.
Didn't you understand Sharktooth's post?

[Sharktooth]

I'll try to make it more clear... it's not a matter of today and yesterday... it's a matter of efficiency.
MMX is faster even in the newest Athlons 64 X2 and FX...

[zambelli]

I'll try to make it more clear... it's not a matter of today and yesterday... it's a matter of efficiency.
MMX is faster even in the newest Athlons 64 X2 and FX...
Are you saying MMX is faster than SSE2 on an Athlon64 CPU?

[IgorC]

Are you saying MMX is faster than SSE2 on an Athlon64 CPU?
No.
Disabling mmx (isse) optimization in H.264 encoders slow down 2x times. (at least in Nero/Ateme encoders)
Disabling sse - a few percents of slow down
Disabling sse2 - only 1-2% of slow down.

[squid_80]

Are you saying MMX is faster than SSE2 on an Athlon64 CPU?

Definitely in some cases. Especially for things like SAD calculations where it's virtually impossible to guarantee alignment. Most of the time SSE2 gives equal performance (hence no gain) and if you get lucky you can see a 10% gain.

[Sharktooth]

Ehrr... thanx for the replies :)

[DeathTheSheep]

Wow, I'm liking the progress of the encoder. What about the audio encoder? Is that going to go through any more huge revamps before the final WMP11 release (the Pro10 encoder, specifically)?

Thanks!