Incomplete GMP documentation...

... where's the email address for sending praise?  

Last night I installed GMP-4.2  on my Mac G5.  It was
impressively quick compared
to what I'd been using before (probably 4.1.4).  Well done!

I believe mpz_root can be speeded up;  I might look into it
if I get some spare time.

Would you be interested in a function which takes a real
interval (two mpf_t values)
and returns the simplest rational in that interval?

Thanks,
John Abbott.
_______________________________________________
gmp-bugs mailing list
gmp-bugsswox.com
http://swox
.com/mailman/listinfo/gmp-bugs

abbott <abbottdima.unige.it> writes:

  ... where's the email address for sending praise?  



  Last night I installed GMP-4.2 on my Mac G5.  It was
impressively quick
  compared to what I'd been using before (probably 4.1.4). 
Well done!

Kevin Ryde's improved configure helps for many systems. 
(But Mac
users tend to run into many compiler bugs, creating sticky
problems.)

  I believe mpz_root can be speeded up; I might look into it
if I get
  some spare time.

Yes, http://swox.com/gmp
/projects.html suggests a division-free
iteration:

                    2
                   x   = x  (3  A x )/2
                    i+1   i        i

This should be faster than the current code, since division
is hard to
make fast (and since current GMP division is a log(n) factor
worse
than multiplication).

That multiplication-only iteration could make use of short
products
(mpn_mullow, mpn_mulhigh, mpn_mulmid) for a great speedup
for operands
of up to around 100 limbs.

  Would you be interested in a function which takes a real
interval
  (two mpf_t values) and returns the simplest rational in
that interval?

That might be useful for mpq.

-- 
Torbjörn
_______________________________________________
gmp-bugs mailing list
gmp-bugsswox.com
http://swox
.com/mailman/listinfo/gmp-bugs

Looking through the GMP mailing list archives, I found an
odd
reply from myself:

Torbjorn Granlund <tegeswox.com> writes:

    I believe mpz_root can be speeded up; I might look into
it if I get
    some spare time.
  
  Yes, http://swox.com/gmp
/projects.html suggests a division-free
  iteration:
  
                      2
                     x   = x  (3  A x )/2
                      i+1   i        i
  
  This should be faster than the current code, since
division is hard to
  make fast (and since current GMP division is a log(n)
factor worse
  than multiplication).
  
I replied as if you talked about mpz_sqrt/mpn_sqrt, not
mpz_root.

I try again: Please look at http://swox.com/gmp/devel/
 for vastly
improved nth root code.  Compared to the code in GMP 4.2,
there
is a 20x to 30x speedup.

-- 
Torbjörn
_______________________________________________
gmp-bugs mailing list
gmp-bugsswox.com
http://swox
.com/mailman/listinfo/gmp-bugs