Feature Request: support 64 bit float

[tremblap]

Is your feature request related to a problem? Please describe the problem.

Pd is moving forward with 64 bit float

Describe the solution you'd like to see.

we should provide both as per the table here:
https://msp.ucsd.edu/Pd_documentation/resources/chapter4.htm#s4.1.2
we should maybe also change the name of the binaries to acknowledge we distribute FAT ones

Describe alternatives you've considered

users compile themselves, but that will start to be messy soon

Additional context

I presume it is a cmake job

[tremblap]

maybe it is a 5 minute job for a magician like @AlexHarker or @weefuzzy - if not I'll tackle it

[AlexHarker]

I suspect this is a cmake job

[weefuzzy]

I need more information. 64 bit float where? Signal chains, buffers, messages? Is it a matter of rebuilding with a define set or something?

[tremblap]

Sorry!!! 64-bit float in the signal chain:

https://lists.puredata.info/pipermail/pd-list/2023-12/132768.html

[AlexHarker]

-DPD_FLOATSIZE=64 for a 64 bit float build and then presumably things will likely break. I can probably fix those things, but I am less confident in poking at cmake build systems (particularly to make a new target). CI would also need adapting.

[AlexHarker]

It's not just 64 bit float in the signal chain - it is everywhere.

You probably want to make these changes which avoid explicitly using float and replace with aliases (t_float / t_sample in use already):

https://github.com/AlexHarker/flucoma-pd/tree/feature/64bit-float

Those are totally untested and there are lots of type templates in play so it might be totally broken, but it does compile with the flag set - I'd have to pass to someone else for build system changes.

[weefuzzy]

Ok cheers. I wouldn't be inclined to touch the cake if this is to be an option, but to pass it in to the build invocation instead. So the main changes would be to CI.

Have buffers also changed? That would be a much bigger set of source changes...

[AlexHarker]

Yes - that would be a possible option - just to pass this as a build option on CI, although you'd need to get a differently named thing out the end (or rename on CI), if you want to build both and be able to install on one computer.

I hadn't checked buffers, but arrays use words, which use t_floats, so yes. I've now updated the branch above with what I think might be the necessary changes there (and checked that it builds). It's not so many places, but again untested, so no guarantee this all propagates nicely and does all the right things - @tremblap could test it though - I just set the compiler flag on my Xcode project (so it would apply to all targets) after running cmake. Alternatively it would probably work to cal the cmake with -DPD_FLOATSIZE=64.

BTW - I always advise touching cake, but only so one can eat it.

CI sounds like a job for @jamesb93 (cough cough) - should he be nice enough to want to do it

[AlexHarker]

Wait @weefuzzy - how do I pass the flag to cmake so it will send it to the right place without altering the cmake?

[AlexHarker]

I've tried export CXXFLAGS="-DPD_FLOATSIZE=64" prior to running the cmake but that doesn't work...

[weefuzzy]

I've tried export CXXFLAGS="-DPD_FLOATSIZE=64" prior to running the cmake but that doesn't work...

That looks sensible. I'll have a look when I can

[tremblap]

Thanks all for your thoughts. I have not played with Pd64 yet, but indeed it seems to have 64float everywhere, which is new and frightening/exciting for all our eigen vector views under the hood... so I am terrified to try now.

@weefuzzy before you try to read/think anything, do you want me to try @AlexHarker 's branch and see if I read and write to a buffer via the simplest objects (I think of dataset i/o and bufcompose to start with, literally just copying in and out of fluidland to arrays) or is that not a useful first test?

[AlexHarker]

To build you will need to manually set the flag @tremblap - I can't do that on the branch, because cmake makes the projects ephemeral. I would set it at the project level as in this screenshot

[AlexHarker]

I am cautiously optimistic that it will work. I will bet one cake that it will (at least broadly speaking) work