Disabling processing when input is silent. Worth it?

[AdmiralQuality]

(This is an attempt to remove this "discussion" from TSV's thread as he certainly doesn't deserve to have his kind offer degrade into a flame war http://www.kvraudio.com/forum/viewtopic ... 63#1991563 ):

I suppose you're out of argument, that or you have a so good reason to process silence that you'd prefer to keep it secret.

Because the test for silence isn't worth the bit of extra CPU that it costs. What good is getting a bit of CPU back during silent portions anyway? "Oh look, processor dropped 5%, I can play an extra note now! Oops, the effect started working again, I better play fewer notes!" Is THAT how anybody actually works? PEAK CPU usage is what matters, giving some back during silent bits is of little value, if any, in my opinion.

Sometimes I even think it might be better if every plug-in used a constant load, rather than having the load bounce up and down as I play more notes (or inputs go silent, whatever...) Hard to tell when you've asked the machine to do too much, you only find out mid-way through your performance.

So again, it's up to YOU to control which channels you want to process on my SCAMP plug-in (and it won't process unconnected channels, but as noted, many hosts, including FLStudio, don't negotiate that correctly. This is part of why I provide bypass buttons, and why there is an .ini option for limiting multichannel support.)

That's our philosophy, we don't force decisions ON our users. You of course are free to implement things however you like, I'm just explaining why WE do it that way. (And SCAMP IS self-resonant, and CAN generate its own sound without any input, if you want to use it like that. Again, not my decision to tell people how to use it, and I don't want to second guess the user and limit them in any way.)

What is your position with FLStudio/Image-Line Software anyway tony? I'm guessing you're the bratty boss' kid or something. Do tell. I have nothing bad to say about your company's product (While it's not my kettle of fish, and not the look/feel I prefer... so what? That's all subjective. Takes all kinds, and more power to you.) But wondering if your comments are the official position of Image-Line Software, or if I need to send a letter to someone there informing them of your activities, putting down and spreading misinformation about other company's products. You want to be careful about that. You're entitled to your opinion of course, but the lies, intentionally bad "tests", and misinformation are another matter.

[tony tony chopper]

Because the test for silence isn't worth the bit of extra CPU that it costs. What good is getting a bit of CPU back during silent portions anyway? "Oh look, processor dropped 5%,

Usually it's not worth much, but by usually I mean because a simple IIR doesn't take more than 0.5% of an average CPU. Still, I do it, because it's almost no CPU to check silence.

But the point was about your Scamp effect. I claimed that it eats quite a lot of CPU - 15% of my AMD 3500+, it's quite a lot. To which you replied that I had to disable channels I wasn't using. Well, to spare even half of those 15%, yes, it's definitely worth it. And it should be automatic.
But because not many plugins bother, I had to come up with a workaround ('smart disable') that stops processing effect if they haven't been fed for a couple of seconds. Still, I'd rather have plugins doing this more cleverly internally.

Is THAT how anybody actually works?

I doubt that many effects are processing a whole song, usually it's just a part of it, and then they stop being fed.

What is your position with FLStudio Image-Line Software anyway tony? I'm guessing you're the bratty boss's kid or something

I'm the lead programmer, and I wouldn't have bothered with you if in the first place you hadn't come up with lame 'fruity' jokes, and if I hadn't seen you open your big mouth again in another thread, claiming that you were the only one using RGBA bitmaps and that everyone else was stupid.

[AdmiralQuality]

Because the test for silence isn't worth the bit of extra CPU that it costs. What good is getting a bit of CPU back during silent portions anyway? "Oh look, processor dropped 5%,

Usually it's not worth much, but by usually I mean because a simple IIR doesn't take more than 0.5% of an average CPU. Still, I do it, because it's almost no CPU to check silence.

But the point was about your Scamp effect. I claimed that it eats quite a lot of CPU - 15% of my AMD 3500+, it's quite a lot. To which you replied that I had to disable channels I wasn't using. Well, to spare even half of those 15%, yes, it's definitely worth it. And it should be automatic.
But because not many plugins bother, I had to come up with a workaround ('smart disable') that stops processing effect if they haven't been fed for a couple of seconds. Still, I'd rather have plugins doing this more cleverly internally.

Is THAT how anybody actually works?

I doubt that many effects are processing a whole song, usually it's just a part of it, and then they stop being fed.

What is your position with FLStudio Image-Line Software anyway tony? I'm guessing you're the bratty boss's kid or something

I'm the lead programmer, and I wouldn't have bothered with you if in the first place you hadn't come up with lame 'fruity' jokes, and if I hadn't seen you open your big mouth again in another thread, claiming that you were the only one using RGBA bitmaps and that everyone else was stupid.

I never said that everyone else was stupid. I WAS trying to encourage more developers to use these techniques because, believe me, I did a survey and not many are.

Remember tony, it was you in this thread: http://www.kvraudio.com/forum/viewtopic.php?p=1988765 who initiated any discussion with such comments as:

IMHO, if the [filter modulating] oscillator is at normal [audio rate] speed, it'll just be useless noise

Where the entire history of analog synthesis, and the MANY examples of classic synths that offer this feature, as well as the other KVR users who posted about how important this feature was for them (including GOOD sounding samples) would show you're missing something. But no, you're smarter than Bob Moog, Tom Oberheim, etc. And all you have to say about the other users are things like, in response to Totc's great example files ( http://www.kvraudio.com/forum/viewtopic ... 75#1988175 )

http://www.kvraudio.com/forum/viewtopic ... 61#1988361

I maintain that modulating filter freq by fast oscillators is useless, like the mp3's posted here demonstrate.

So, you're just smarter than EVERYBODY, aren't you?

And the fruity comments started after you published misinformation and intentionally bad sounding test samples of my product. From the FLstudio website! CAREFUL!

For the record.

[AdmiralQuality]

And regarding...

I doubt that many effects are processing a whole song, usually it's just a part of it, and then they stop being fed.

So automate the bypass button for the parts where you don't want the effect used. Again, I see this as the user's decision to make.

[tony tony chopper]

Or after you intentionally didn't post any example of your own product. So I just fed it a plain, normal sawtooth (since you were talking about -oscillators- modulating filters). If it's bad sounding you know who to blame.

[AdmiralQuality]

If it's bad sounding, I blame YOU for not even trying to adjust the patch to suit your input signal. This is what I mean by intentionally bad tests and misinformation.

And I didn't post a sample from SCAMP because it wasn't ABOUT my product, twit, it was about a feature that MANY products used. I regret even mentioning my product as a way to get it (indeed, I mentioned some of the competition's products too, just to show I wasn't the only one who knows about this.)

When someone is as stubborn in their misguided opinions as you, it's easy to see how you arrived at and maintain them. I was just trying to HELP you, and was being very friendly about it and trying to walk you through it, until you started making things intentionally difficult.

So, there's the argument you wanted and were taunting me for. Please everyone chime in with your vote on the issue (disabling processing during silence) so tony can maybe learn something. (Or maybe I'll learn something and start adding this feature. I'm open to it.)

[tony tony chopper]

Well, I don't see what users could have AGAINST a transparent auto processing skipping. For them it's just the same thing for less CPU.

We're not splitting hairs about 0.2% CPU here, but 15% (of today's good CPU's), and this is the whole point, IMHO 15% is quite a lot.

[AdmiralQuality]

And another thing

But the point was about your Scamp effect. I claimed that it eats quite a lot of CPU - 15% of my AMD 3500+, it's quite a lot. To which you replied that I had to disable channels I wasn't using. Well, to spare even half of those 15%, yes, it's definitely worth it. And it should be automatic.

Your math skills are showing. Disabling 4 of 6 channels would infact spare you 2/3 of the processing power. More than "even half". But did you try this? No. And again here, you re-post an unfair, intentionally biased observation of the processing load of my product.

SCAMP was a bit of an experiment into multichannel processing. I haven't had any need for the multichannel stuff myself, but I know there's someone out there who's very happy that they can select the channels in a surround mix that they'd like the effect to process.

And again, if your host (that you programmed) negotiated the channels correctly, and told my plug-in that there were only 2 channels being passed to it, it wouldn't LET you process channels 4 to 6. And AGAIN, there's an .ini file option for disabling this behavior and FORCING it to 2 channel, so there you go, exactly what you claim to want. But you post misinformation to the contrary.

Seriously, watch it.

[AdmiralQuality]

Well, I don't see what users could have AGAINST a transparent auto processing skipping. For them it's just the same thing for less CPU.

We're not splitting hairs about 0.2% CPU here, but 15% (of today's good CPU's), and this is the whole point, IMHO 15% is quite a lot.

And here's the lie AGAIN!

[stefancrs]

FWIW, and I don't care about your quarrel with each other, most people often use different tracks / instruments in different parts of their songs. Some channels might produce audio just during the chorus, and others just in the bridge or in the verse. The "constant CPU" would not hold when it comes to synths for instance (like having all the polyphonic voices active and processing even when they're not playing anything).

Just my 2c, now you can resume the quarrelling

edit: and, oh, oscillators modulating filters (and other things) are fun

[tony tony chopper]

'even half' means even just half. And unfair?? It's actually quite fair, since you say yourself you could skip 5/6 of your inputs. That makes a 80% (actually more, but let's be 'unfair') CPU usage increase, and I'm *unfair* for saying 50%?

Really if you can spare 80% of the CPU usage in your plugin by not processing silence, you must have a very good reason not to.

And here's the lie AGAIN!

cool. But don't bother explaining which lie, ok?



Anyway I can prove my point, I've done it. For those who want to test, the EQUO effect, which is a bank of 31 stereo IIR bandpasses, skips silence. It eats way under 1% when fed with silence (after a couple of ms), and 4% (up to several times more, depends of 3DNow/SSE2 and if oversampling is used) when fed with valid signal.

Which is why I also often advise not to add noise to avoid denormalization, but use 'denormals are zero', or force tiny numbers down to zero.

[AdmiralQuality]

it wouldn't LET you process channels 4 to 6.

I meant channels 3 to 6 (which is 4 out of 6) Just to clarify.

The next version of SCAMP is going to default to stereo, just so people like you who are too lazy to look at what's actually going on before they pass judgement stop knee-jerking at it.

Then I'll get to find out how many people actually use it on surround tracks, cuz THEY'LL start complaining.

[caoxiang]

Calm down guys.

[AdmiralQuality]

Calm down guys.

[AdmiralQuality]

Which is why I also often advise not to add noise to avoid denormalization, but use 'denormals are zero', or force tiny numbers down to zero.

Now the guy who thinks stuff should run on Win9x (see TSV's thread) wants everyone to have SSE.

The lie is that it takes 15% of your CPU to process stereo (and yes, if it was a mono sawtooth, then half that again.) If you bothered to USE IT CORRECTLY, it would take hardly any.

Still, SCAMP is relatively expensive compared to many other IIRs. INTENTIONALLY -- compare the sound of the resonance in SCAMP, particularly at the higher end of the cutoff, to any other digital filter, and you might understand why SCAMP is working harder.

(It's subective so don't bother tony, you claim to not hear it and will just post more B.S. I get several emails a week from happy users who tell me they DO get it though.)

And as you pointed out in the X-mod thread, when the envelope follower is active, its also modding the cutoff coeffs for EVERY sample. Once again, our philosophy, is that if it's worth doing, it's worth doing RIGHT.

How would you like it if I dug into a critique of FLStudio and posted results from intentionally misusing it and ignoring features that would hacve helped me? How would your EMPLOYER like it? Do they like that you're doing exactly that to other companies? Do you think you're losing FLStudio sales to SCAMP customers??? Why would you even START a fight like this (well, it was to senselessly defend your misguided position on osc->filter modulation, which even after so many people have posted in favour of it, you insist is a "useless feature" that just makes "noise".)

Like I said, you're an idiot because you're so stubborn. Learning to admit when you're wrong would serve you (and your employer) better in the long run.