[MPFR] Re: digits of Pi
Torbjorn Granlund
tg at gmplib.org
Wed Jan 13 13:45:59 CET 2010
Paul Zimmermann <Paul.Zimmermann at loria.fr> writes:
are you sure? Here are some timings:
GMP 4.3.2:
achille% ./speed -s 20000000 mpn_mul_n
overhead 0.000000003 secs, precision 10000 units of 4.17e-10 secs, CPU freq 2400.00 MHz
mpn_mul_n
20000000 38.094209000
GMP 5.0.0:
achille% ./speed -s 20000000 mpn_mul_n
overhead 0.000000003 secs, precision 10000 units of 4.17e-10 secs, CPU freq 2400.00 MHz
mpn_mul_n
20000000 42.456546000
I think this is due to that the FFT tables are shorter for the new
release. I will address that in 5.0.1.
We no longer use the mpn_mul_fft_full function, but instead the new
faster mpn_fft_mul. So as long as the tables for mpn_mul_fft (note that
mpn_fft_mul != mpn_mul_fft) are not exhausted, things should run perhaps
10% faster.
--
Torbjörn
More information about the gmp-discuss
mailing list