Control rate

[UrbanCyborg]

I said timers were the easiest way, not the safest way.

Reid

[ColinP]

I said timers were the easiest way, not the safest way.

Reid

<Chuckles>

Multi-threading is probably the most difficult aspect of programming.

Even in the old days of direct-to-the-metal programming where one had total control of a processor and there was no invisible code it was still a nightmare.

One colorful anecdote that might amuse...

Back in the mists of time I was presenting the so-called final version of a game called Eco that simulated aspects of evolution to the head of the big publisher. It had run without any problems for "an eternity" but crashed half way through the presentation. He was very understanding but I was mortified.

It took about three or four days of intense effort before I discovered that the bug was caused by the 3D graphics pipeline interrupting another thread during a tiny critical segment of code that was simulating two creatures having sex.

[utdgrant]

It took about three or four days of intense effort before I discovered that the bug was caused by the 3D graphics pipeline interrupting another thread during a tiny critical segment of code that was simulating two creatures having sex.

Coitus interruptus?

[UrbanCyborg]

Ah, games. I recall one I was on as sound coder/designer. Another team member kept complaining that there was something wrong with my sound code. Took a few days for me (of course) to find out where the bug actually was. His code was walking on whatever happened to be active at the time, which, naturally enough, was frequently the sound code. Didn't even get an acknowledgement from the guy for making me waste the better part of a week tracking his obscure bug.

Reid

[utdgrant]

I put together a module for playing about with dividing the ProcessSample() method by an integer amount.

The 'audio' code segment is executed on every pass.

The 'control' code segment is only executed once every N passes. A panel switch allows you to select different values of N in real time.

I placed a chuck of CPU-intensive operations within the 'control' segment such that it would overload my Core i7 processor at N=1.
When you select '1' on the divider switch, audible distorion occurs. Even changing the divider switch back to '4' (on my machine, at least) gives the processor enough time to spread the load of the intensive operations over time, bringing the CPU load back under 25% and restoring a clean audio signal.

ControlRateDividerModule.png (95.44 KiB) Viewed 5381 times

YouTube video of module in operation

Source code is available on my Google Drive.

[ColinP]

Reid, very bad form from that colleague. Of course such sh*t happens but it shouldn't.

Grant, coitus interruptus were exactly the words that came to mind but this old atheist was taught Latin while attending a Catholic school. What's your excuse?

In your little experiment did you look at different buffer sizes? I'd expect things would get skewed if you were dividing by anything where the buffer size was larger than the divisor. If not that would be very curious.

[utdgrant]

Grant, coitus interruptus were exactly the words that came to mind but this old atheist was taught Latin while attending a Catholic school. What's your excuse?

This old atheist was brought up as a Glaswegian Protestant, but one of my greatest joys in life is puns.

That's my excuse, and I'm sticking to it!

[ColinP]

I genuinely misread puns as nuns.

Uncontrollable laughter ensued.

[UrbanCyborg]

Okay, nun of that, now!

Latin humor, eh? Way back in Bruce Reich's Third-year Music Theory class we were given the assignment to write a motet in period style. Casting about for an appropriate text, I settled on translating the opening chant of the Doors' The Soft Parade into Latin. Actually spent more time on that than on writing the motet itself.

[utdgrant]

In your little experiment did you look at different buffer sizes? I'd expect things would get skewed if you were dividing by anything where the buffer size was larger than the divisor. If not that would be very curious.

Ah, reading back, I may very well have misled you. My bad.


The "audio phase" of ProcessSample() is executed every sample period.


The "control phase" of ProcessSample() is executed every N sample periods, where N is the divider setting.


I put my "BATSH*T CRAZY BONKERS (tm)" lump of heavy-duty code (1,000 double-precision log(x) and 1,000 double-precision exp(x)) within the "control phase".


This means that where N is 1, you max out the processor because you're executing the BCB code 100% of the time.


With N set to 4, you are only executing the BCB code 1 out of every 4 cycles, i.e. 25% of the time. If you set N to 256 and your buffer size was only 64 samples, then you would get 1 execution of the BCB code segment every four BUFFERS.