[GSLL-devel] FFT speed, tests

[Liam Healy]

OK, thanks Sumant.  I'm really surprised at the lack of significant
speedup.  On some real array tests, there were some significant speedups --
factors of 10s if not more.  I'm sure there are more possible improvements
however, but it's been a while since I looked at them.  However, do not
overlook the possibility that the declaration optimization are defective
somehow or are not "taking" in your tests.  Basically, the optimizations
should cause the compiler to macroexpand into direct access (via cffi calls)
rather than dispatching on a generic function.   As I recall however, it's
difficult to prove that it's doing so (the standard allows any and all
declarations as well as compiler macros to be ignored).

Thanks for your continued help on this, it is much appreciated!  I look
forward to whatever you come up with.

Liam

On Sun, Apr 3, 2011 at 6:09 PM, Sumant Oemrawsingh <soemraws at xs4all.nl>wrote:

> Hi Liam,
>
> Today, I had some time to take a look at the (slow) speed of the FFT
> tests. I found that having the tests as is and adding declarations
> only gives a speed-up of (barely) 10%. I think I'll need to thoroughly
> rewrite the tests to find/solve the problem. I need to do this anyway,
> since a small number of tests still keeps on failing and I haven't
> been able to track down _that_ problem yet.
>
> As such, I'll begin "from scratch", rewriting each test. This might
> take some time, and due to a shortage of that due to work-related
> things, this might take even longer than anticipated. This situation
> is quite annoying since, naturally, the FFT tests should be enabled in
> the gsll-tests package, and not only for completeness. I'll do my best
> allocating time to this.
>
> Regards,
> Sumant
>
> --
> Sumant Oemrawsingh
> Sikander on #lisp
>
> _______________________________________________
> GSLL-devel mailing list
> GSLL-devel at common-lisp.net
> http://common-lisp.net/cgi-bin/mailman/listinfo/gsll-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.common-lisp.net/pipermail/gsll-devel/attachments/20110403/7a6c072f/attachment.html>