[SciPy-user] scipy.basic to numpy

Arnd Baecker arnd.baecker at web.de
Fri Jan 6 01:36:16 CST 2006


Hi Pearu,

On Thu, 5 Jan 2006, Pearu Peterson wrote:

> On Thu, 5 Jan 2006, Rob Managan wrote:
>
> > Sorry for being unclear. I was referencing the numpy.dft.fft function
> > to show that it works but the fftpack version in scipy does not.
> >
> > The problem is in scipy not numpy.
>
> I tried building scipy.fftpack against fftw-2.1.3, fftw-3.0.1, and fortran
> fftpack. In all cases scipy.fftpack.test() finishes without failures on
> debian box, both on 32 and 64-bit boxes.

Do you also observe a very poor performance of fftw-3.0.1 for
(in particular for complex Arrays)?

Best, Arnd

(some) Details:

In [11]: numpy.__version__
Out[11]: '0.9.3.1837'
In [12]: scipy.__version__
Out[12]: '0.4.4.1526'

This is on an Opteron, but we also see similar results on
other machines ...


                 Fast Fourier Transform
=================================================
      |    real input     |   complex input
-------------------------------------------------
 size |  scipy  | Numeric |  scipy  | Numeric
-------------------------------------------------
  100 |    0.05 |    0.06 |    0.88 |    0.05  (secs for 7000 calls)
 1000 |    0.04 |    0.08 |    0.51 |    0.08  (secs for 2000 calls)
  256 |    0.11 |    0.10 |    1.43 |    0.11  (secs for 10000 calls)
  512 |    0.17 |    0.19 |    1.68 |    0.19  (secs for 10000 calls)
 1024 |    0.02 |    0.04 |    0.23 |    0.03  (secs for 1000 calls)
 2048 |    0.04 |    0.07 |    0.34 |    0.07  (secs for 1000 calls)
 4096 |    0.05 |    0.11 |    0.29 |    0.11  (secs for 500 calls)
 8192 |    0.11 |    0.48 |    0.65 |    0.48  (secs for 500 calls)
....
    Multi-dimensional Fast Fourier Transform
===================================================
          |    real input     |   complex input
---------------------------------------------------
   size   |  scipy  | Numeric |  scipy  |  Numeric
---------------------------------------------------
  100x100 |    0.06 |    0.06 |    0.05 |    0.07  (secs for 100 calls)
 1000x100 |    0.05 |    0.11 |    0.06 |    0.10  (secs for 7 calls)
  256x256 |    0.11 |    0.10 |    0.12 |    0.11  (secs for 10 calls)
  512x512 |    0.34 |    0.20 |    0.32 |    0.20  (secs for 3 calls)
.....
       Inverse Fast Fourier Transform
===============================================
      |     real input    |    complex input
-----------------------------------------------
 size |  scipy  | Numeric |  scipy  | Numeric
-----------------------------------------------
  100 |    0.05 |    0.15 |    0.92 |    0.14  (secs for 7000 calls)
 1000 |    0.06 |    0.17 |    0.54 |    0.18  (secs for 2000 calls)
  256 |    0.11 |    0.27 |    1.49 |    0.28  (secs for 10000 calls)
  512 |    0.17 |    0.43 |    1.76 |    0.45  (secs for 10000 calls)
 1024 |    0.02 |    0.07 |    0.24 |    0.08  (secs for 1000 calls)
 2048 |    0.05 |    0.14 |    0.35 |    0.14  (secs for 1000 calls)
 4096 |    0.05 |    0.18 |    0.30 |    0.20  (secs for 500 calls)
 8192 |    0.10 |    0.70 |    0.67 |    0.73  (secs for 500 calls)



 ! ldd
/home/abaecker/BUILDS3/BuildDir/inst_numpy/lib/python2.4/site-packages/scipy/fftpack/_fftpack.so
        libfftw3.so.3 => /scr/python/lib/libfftw3.so.3
(0x00002aaaaabb6000)
        libg2c.so.0 => /scr/python/lib64/libg2c.so.0 (0x00002aaaaad66000)
        libm.so.6 => /lib64/tls/libm.so.6 (0x00002aaaaaebc000)
        libgcc_s.so.1 => /scr/python/lib64/libgcc_s.so.1
(0x00002aaaab014000)
        libc.so.6 => /lib64/tls/libc.so.6 (0x00002aaaab11f000)
        /lib64/ld-linux-x86-64.so.2 (0x0000555555554000)





More information about the SciPy-user mailing list