[MPFR] multiplications vs. subtractions: which one is faster?

[Patrick Pelissier]

Hi,

Here are some benchmarks of MPFR 2.2.1 on Pentium M with prec=3000
(Min/average/max):

 mpfr_sub:         553 /   607.50 /   818
 mpfr_mul:       26572 / 26906.87 / 27175
 mpfr_add:         551 /   594.00 /   799

Subtraction version:  6400*607=3,884,800 ticks
Multiplication version: 138*594+138*26906 = 3,795,000 ticks

Taking into account the cache and the branch prediction, I think both
versions are equal.
If you reduce the precision, the multiplication version should be
faster, whereas the subtraction version should be faster if you
increase it.

--
 Sincerely,
  Patrick Pelissier


On 7/15/07, Michael <comtech.usa at gmail.com> wrote:
> Hi all,
>
> In my project I am seeking ultimate speed. In one small part of my
> program using GMP/MPFR, I am facing the following two design choices:
>
> The operands/variables are of 1000 digits precision in MPFR mpfr_t
> format(floating point numbers).
>
> 1. Using two for loops with around 80*80=6400 substractions;
> 2. Using one for loop 7+15+21+32+63=138 additions and 138 multiplications,
>
> Which approach will be faster? Please shed some lights on me from your
> experiences.
>
> Since the change in design is complicated (the MPFR programming is
> much like arranging registers in assembly language programming), I
> want to first do some thought experimenting and comparison.
>
> Moreover, I would appreciate any further
> suggestions/advice/comments/tips on how to improve speed using MPFR
> library.
>
> Thanks a lot!
>