gnupg compilation problems on Solaris 10 64 bit

[David Shaw]

On Fri, Nov 21, 2008 at 06:01:19PM -0500, Robert J. Hansen wrote:
> David Shaw wrote:
> > GnuPG has some assembler code for doing math on large numbers.  Every
> > now and then a new platform has trouble with it until we tweak things.
> > --disable-asm uses portable C code instead of the assembly.
> 
> What's the engineering reason for the ASM code as opposed to just
> sticking with the C code?  It seems that for the vast majority of users,
> there's no difference in performance between the C code and the
> ASM-tuned code.  My guess (prejudice?) is that this would really only
> make a big difference for high volume operations.

There is a nonzero benefit to the assembly code, but it is not large.
On the 3ghz 32-bit Linux box I'm currently sitting in front of, 100
encrypts of a 2-byte file to a 1024DSA/4096ElG key takes 11.3 seconds
on average for the assembly version, and 13.0 seconds on average for
the portable C version.

Unless you're going a lot of public key operations in a row, any
difference is going to be dwarfed by other factors.

David