internal speed of core cells

hi,

for an ensemble I will need a matrix of 32x32 , 32 audio signals multiplied each by a coefficent and 32 output.

I've buid a core cell with a double iteration inside that just do 1024 operation.

because i don't want to patch 1024 modules.

each 1024operation has to happen in less than 1 sample.

the problem is that it works if I start the iteration with a single event but doesn't work if I plug the SRC at the start. I have tried to plug an A TO E perm and I found a frequency of less than 47.646Hz for the cycle to complete.

what is it that I don't understand?

the core cell

the double iteration, the loop macro is just a compare module.

the function macrois just (sin (i)*sin(j))

I've tried to set the function macro to non solid just in case because I don't understand the solid/non solid parameter but it doesn't change anything

Find more posts tagged with

Comments

totole robot

after further analysis even for event loop the loop happen at sample rate whicis not what I want.

I decided to use brute force by cascading +1 and merging but it looks like only the last value goes out of the macro.

this one is for one row.

this one os for 32x32 values

K is for column and L is for row.

and this is where the calculation and writing happens

what is it that I don't understand?

colB

https://community.native-instruments.com/discussion/comment/248596#Comment_248596

what is it that I don't understand?

How core works ;-)

In core, there is 'logical simultaneity', so operations that have the same event source are considered to be simultaneous. That means when they are merged, all but one must be discarded. In that case, the lowest input to the merge will override all others that arrive at the same logical time.

All this is in the manuals. Understanding the event model is crucial to success in core. It is not the same as primary. It's not particularly difficult, unless you don't get the basics nailed down, in that case, its a frustrating nightmare supposedly.

There is no iteration (we've been asking for core iteration for 20 years!)

There are different ways to 'unroll' loops, once you get good, it can be relatively painless, although maintenance is frustrating. The main problem is cpu efficieny can take a hit due to all the redundant code. Unrolled loops are bad for optimal cache performance.

Learn to create a single iteration cell, then double that a few times using copy/paste, and then five edits later you have 32 iterations.

I have a few example with unrolled loops. FFT convolution, velvet reverb…

In the factory library, the grain sampler unrolls the grain processing section to handle 32 grains, worth disecting that…

The other common way to have multiple similar processes is to use a polyphony hack. Sometimes unrolled loops fits better, sometimes polyphony does.

totole robot

OK, thank you very much. I was thinking of using polyphony.

I've read the manual but didn't understand that part it seems.

It's a pity they didn't translate into other languages.

I will also look at your work in the library. 🙂

Quick links

Unanswered questions