fill data array only once at gen~ init?

Roughly analogous, this is what I usually do. I use a history, so that I can trigger it via a message from outside gen~ if I want to:

History reset(1);
if (reset) {
    // do stuff that only happens once
    // e.g.
    for (i=0; i<dim(dat); i+=1) {
        dat.poke(i*i, i);
    }
}    

Thanks for the reply. I found an alternative: loading a buffer with stored data from an audio file, then sending gen~ a message to store the buffer in a data object, as illustrated in one of the gen~ help file tabs. I'll do some analysis on it.

Only thing to watch out for there is that the data object won't be resized to match the buffer size. But so long as your data and buffer are the same size, that works.

Yes, also Buffer and Data objects in functions must have size and channels declared, or Max doesn't compile. And if the function is called by another function, the Buffer and Data declarations must also be at the top level, or you get anomalous errors on compile attempts.

But I am running into a memory problem. Max crashes if I try to load 32 voices at launch. I have to load with one voice, then change it to 32 voices after load. Memory then increases to 5 gigabytes. So it won't work in 32-bit version at all.

If I have large buffer and data objects, is there anything I can do to reduce the memory usage? That would be very helpful.

Buffer doesn't need size/channels, only Data -- buffer inherits the size/channels from the buffer~ it refers to. Actually even Data shouldn't need them either, it defaults to 512 frames of 1 channels.

By the way you can also pass buffer/data into functions, rather than using global names. For example:

getfirstsample(src) {
    return src.peek(0);
}

Data d;
d.poke(74, 0);
out1 = getfirstsample(d);

I'm not sure what you mean about 32 voices. If you mean your gen~ inside a poly~ with 32 voices, and they are all using the same buffer~ data, then I strongly recommend you use Buffer rather than Data in your gen patcher. Data means a local copy for each gen~ instance, whereas Buffer uses the same memory as the underlying buffer~ object.

The only reason to use Data is if you want 64-bit data generated in gen~. Since you're loading from a buffer~, the data is 32 bit anyway.

Aha! That's just what I wanted to know about passing data/buffer in! Thank you!

My first tests of a single instance of gen~ indicate that data arrays, when declared inside a function, load another copy. I was able to get memory down from 465MB to 439MB after 4 hours unrolling functions. But for the work and and the difficulty of maintenance, it would be alot easier if there were another way to reduce memory, and it doesn't seem as significant as other factors. The memory goes up to 565MB if I close the patch and open it again. And the memory usage doesn't change if I am open the patch on 32-bit or 64-bit Max on Windows.

Would it make a difference if I somehow forced gen~ to work in runtime mode? Or are there global settings that make a significant difference to memory usage?

And yes, I am trying to embed gen~ in a poly~ again.

No idea about run-time mode (not likely), but absolutely use Buffer instead of Data.

Buffer is just a reference wrapper, should have negligible effect on memory.
Data makes a copy (and converts to 64-bit), definitely will eat up memory in a poly~.

Well as far as gen~ goes, my tests do indicate that each declaration still increases memory, whether buffer or data.

Can one resize buffer() or delay()after declaration in gen~? Some of my delays are very large to support 192khz samplerate, and people with slower machines don't need such a big delay. It would be nice to resize buffers and delays and things depending on sample rate

EDIT: maybe they are double buffered?

Alas, passing pointers to data and buffers with multiple channels doesn't work though. I'll file a bug.