Re: log2 on MinGW

[Michael Goffioul]

On Mon, Nov 4, 2013 at 1:54 AM, John W. Eaton <address@hidden> wrote:

On 11/03/2013 08:39 PM, Michael Goffioul wrote:

gnulib provides an implementation of frexp, however I'm wondering
whether gnulib's implementation can be considered sub-optimal, as it has
to be generic, compared to the internal Win32 implementation, which can
use assembly. I truly don't know, I never tried to compare performance.

I checked in the following change:

  http://hg.savannah.gnu.org/hgweb/octave/rev/573f865aa8c8

If someone shows that the gnulib function is slow or doesn't
solve the problem that we have, then I think we should try to improve
the gnulib function.

I've tried to attached test bench to compare gnulib performance against system frexp. Test results are different whether you're comparing against Win32 or against glibc.

On my Linux x86_64 with glibc: gnulib's implementation is about 6.5x slower than glibc's implementation

On my WinXP 32bits: gnulib's implementation is about 1.15x slower than MSVC runtime's implementation

Time is measured using "time" in a shell.

So all things considered, using gnulib::frexp is not bad under Windows, because Win32 implementation doesn't seem good anyway. However, if gnulib's version is picked up on some other UNIX-like platforms, where glibc would be normally used, the performance penalty is significant.

Michael.

tfrexp2.c
Description: Text Data