[gmpi] Re: Topic 7: Audio packaging
- From: David Olofson <david@xxxxxxxxxxx>
- To: gmpi@xxxxxxxxxxxxx
- Date: Tue, 3 Jun 2003 12:24:39 +0200
On Tuesday 03 June 2003 12.04, Vincent Burel wrote:
[...]
> > The savings are real, and are measured of orders of magnitude
> > fewer clocks.
>
> on which kind of audio system !? and which kind of task !? that's
> the problem. We are already talking about optimization regarding
> 20% max of the usage ... in the reality this 10% will become 1%
> because most of the host and plug-in developper won't get the time
> , to implement the IsZero Flag.
But they *do* have the time to figure out mathematical shortcuts and
hack one or more platform specific SIMD versions of the DSP code...
Weird how some optimizations are suddently not of any interest.
> > A P3 or better can do a floating point multiply or add on every
> > clock if your clever with your instruction scheduling, maybe
> > every other clock if you
> > are lazy. However, if you miss the L1 or L2 cache, you stall >20
> > clocks. In other words, being slopping with locality of
> > reference, and memory access, (ie, excessive memcpy or memset)
> > can slow your performance down tenfold.
>
> well, make a test with that :
> http://perso.wanadoo.fr/vb-audio/us/mm/test_spp/test_spp.htm
> measure the SPP-M on two computers, a 2 years old one and a recent
> one . And take a look on the difference. the IsZero flag won't be
> able to make 20% of this difference on an average audio montage.
> This is simply time and energy loss.
So is any form of optimization - unless you have to sell your product
*today*, rather than next year. Still, it's your choice. If you can't
be bothered to spend time saving cycles, then don't.
> > you measure it ?
> > <<<
> >
> > Yes, I measured it. I spent weeks profiling with VTune, seeing
> > hotspots in
> > memset, and optimizing. As I said in my previous email, this
> > approach made
> > a DRAMATIC improvement in performance in our audio engine.
>
> statistically speaking you get a Dramatic improvement if more than
> 75% of your audio montage contains silences. I'm sorry but i'm not
> sure it's a usual case .
Where are your statistics, and what is it based on? Seriously, I'd
like to have an idea of how things look in other people's audio
sequencers and HDRs. I know how *I* do stuff, and I could make some
calculations based on that, but it depends a whole lot on genre and
the way you're working, I think.
> > Generally speaking, a feature which maybe implemented or not
> > implemented is
> > a potential marketing and communication problem.
> > <<<
> >
> > Keep on zero-filling and your customer's 3 GHz system well
> > perform like a 300 MHz system. That would be a potential
> > marketing problem...
>
> upto now, i never got problem with speed or optimization :-) you
> should take a look on my product Ron... you will feel that the
> silence magic flag is already set ! :-))))
Well, then you won't need to use this feature anyway, so what's the
problem? ;-)
//David Olofson - Programmer, Composer, Open Source Advocate
.- The Return of Audiality! --------------------------------.
| Free/Open Source Audio Engine for use in Games or Studio. |
| RT and off-line synth. Scripting. Sample accurate timing. |
`-----------------------------------> http://audiality.org -'
--- http://olofson.net --- http://www.reologica.se ---
----------------------------------------------------------------------
Generalized Music Plugin Interface (GMPI) public discussion list
Participation in this list is contingent upon your abiding by the
following rules: Please stay on topic. You are responsible for your own
words. Please respect your fellow subscribers. Please do not
redistribute anyone else's words without their permission.
Archive: //www.freelists.org/archives/gmpi
Email gmpi-request@xxxxxxxxxxxxx w/ subject "unsubscribe" to unsubscribe
Other related posts:
