Improvements to powerpc32 asm code
Tue, 03 Jun 2003 10:55:33 +1000
Mark Rodenkirch <email@example.com> writes:
> mpn_add_n mpn_add_n_new
> 1 21.1270 #10.0658
> 2 13.0883 #7.0432
> 3 10.0633 #5.7040
I take it this is -C. -CD is probably best for looking at cycles/limb
loop speeds, since it cancels out fixed overheads. A plain -c can be
used to look at total times on small sizes.
Torbjorn Granlund <firstname.lastname@example.org> writes:
> All-in-all, the 7400 is much more similar to 750 (aka G3) than
> to 7450.
For our purposes so far I think 7400 merely a 750 with two extra
completion queue entries. :-)
> (Note that testing potentially buggy assembly routines can cause
> problems with the random number generators of the test programs,
> since they rely on mpn_addmul_1.
tune/many.pl is a good way to run speed.c and try.c on new code. It's
a bit rough around the edges, but hopefully the comments are enough to