Re: packaging static lib oriented software
On 21 Jan 1998 bruce@va.debian.org wrote:
> I also have a lot of trouble believing a 30% slow-down for PIC code.
> Do you have good benchmarks for this?
>
When I took over povray, I found that the debian version was performing
significantly worse at the standard benchmark (skyvase.pov) than expected.
8 minutes on a pentium/166 vs 5-6 minutes from the povray benchmark site.
The upstream source ran in 6 minutes. The difference was that povray had
been split into a set of small frontends (for X11, SVGALIB, and no
display), calling a shared library to do all the rendering. This shared
library ran 30% slower than the equivalent code. It was probably a
worst-case scenario, many main code/library transitions with heavy
computation in the shared library (where one cpu register was
unavailable).
Perhaps compiling the povray library non-PIC would reduce this penalty
(I didn't try), but it would eliminate all but the diskspace advantages of
shared libraries (no interprocess code sharing).
Scientific libraries (lapack,...) are huge and full of rarely used code.
They run faster, and can be installed without (huge) libraries if linked
staticly. Scientists are lazy programmers, and aren't going to use
many of the functions in the libraries anyway, so there's no point
installing them. :)
I wonder if X and Mesa would benefit from static linking against libm?
-Drake
--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
debian-devel-request@lists.debian.org .
Trouble? e-mail to templin@bucknell.edu .
Reply to: