[Gsll-devel] Possible use of user-supplied vector sizes and offsets for FFT

[Liam Healy]

Sumant,

In writing GSLL, I've tried to adhere to the philosophy of providing a
good Lisp interface to GSL as it exists in the latest release.  When
I've had ideas on better ways to do GSL, I've restrained myself from
putting them in GSLL.  So, there may well be ways they could have made
a better, more versatile, etc. library, but I have not tried to
implement them.  I've chosen this policy in order to limit the amount
of work required for GSLL, which has been substantial as it is.

Specifically with regard to FFTs, the 2+ dimensions case may be very
useful, but I don't see that GSL has done it, so I will avoid it in
GSLL.  The stride is there in GSLL because it's in GSL; offset is not.

So where does that leave your good ideas?  A lot of what you're
talking about I think can be helped along by having a mechanism for
array creation, slicing, concatenation, etc., and perhaps another
chunk by a port of fftw that is compatible with GSLL arrays.  As you
have seen and we have discussed, there has been a lot of discussion on
array utilities and they are a highly requested feature.  I have given
this a lot of thought recently and decided there needs to be a
separate system to do this, in part because I was uneasy in adding
things to GSLL that weren't part of GSL and in part because I realize
the extent of what needs to be done really justifies a separate system
(or more likely, systems).  I have not had a chance to create and test
much code, but I hope to soon.

I'll continue to work on the FFT port and let you work it over before
I pull it into master.  Then I'm going to try to pull together the
beginnings of the array system; it will take a while to get that
settled but I think it will be worth making this as versatile as
possible.

Liam