[CM] s7: Whether to use WITH_GMP

Kjetil Matheussen k.s.matheussen at gmail.com
Fri Apr 14 05:10:39 PDT 2017


Hmm, looking a bit at the code now, and most likely, the
scheme math operations in Radium probably takes
significantly less time than other things. Math
speed probably doesn't matter, so I'm going
to try to enable gmp for a while.



On Fri, Apr 14, 2017 at 1:54 PM, Kjetil Matheussen <k.s.matheussen at gmail.com
> wrote:

> Thank you. How about float vs. integer? Is gmp
> used only for integers, or is it used for floats as well?
>
> In Radium, I think only calculations with floating
> points are important to be fast. And only
> integer calculations needs to be accurate. Big-number
> accuracy for floating points are probably not important,
> and normal 64 bit "double" should be good enough.
>
> Would it make sense to enable gmp for s7 in Radium
> before the optimizer is rebuilt?
>
>
>
> On Fri, Apr 14, 2017 at 1:41 PM, <bil at ccrma.stanford.edu> wrote:
>
>> In the gmp version, it's probably safest to assume
>> everything is going through gmp. In cases where the
>> optimizer can ascertain ranges, s7 will use
>> faster code, but I just tore out half the optimizer,
>> and am rebuilding it.  So I don't know how the
>> in-between stuff interacts with gmp -- give me a
>> month or two!  My old timings indicated that gmp
>> could be 10 to 50 times slower than straight C (but at
>> least it gave the right answer).
>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://cm-mail.stanford.edu/mailman/private/cmdist/attachments/20170414/ebcbb2f3/attachment.html>


More information about the Cmdist mailing list