[R-SIG-Mac]more speeding

[Jan de Leeuw]

Yesterday I found that that R-1.3.1-Darwin is about 40% faster on the 
eigen/hilbert
benchmark than R-1.3.1-Carbon. This may be due  to the fact that
the Darwin version uses ATLAS. but now let's compare with R-1.4.0 
(devel)  under
Darwin.

n=500    carbon  20    darwin-1.3.1   14     darwin-1.4.0    6
n=1000 carbon 173   darwin-1.3.1 126     darwin 1.4.0  69

Thus we see that the development version of R is about twice as fast
as the 1.3.1 version. The two main differences from the compilation
point of view are that (a) R-devel uses g77 while R-patched uses
f2c, and (b) R-devel uses two-level-namespaces, while R-patched
uses flat-namespaces. Unless there are optimizations in R itself
that explain this huge difference, it must be (a). Both versions use
the same ATLAS, etc.

To compare this with a benchmark that does not use ATLAS or
FORTRAN code (I think), we use system.time(sort(rnorm(n)))
where n is either 10^6 or 10^7. For 10^6 Carbon, Darwin-1.3.1,
and Darwin-1.4.0 use 16, 14, and 15 seconds. For 10^7 this is
241, 210, and 228 seconds. The Darwin versions are 10-15%
faster, with 1.3.1 slightly faster than 1.4.0. With the MrC compiler
under MPW presumably the Carbon version will catch up here,
although it still does not have ATLAS and it still uses f2c, so
it will not do well on the linear algebra.
===
Jan de Leeuw; Professor and Chair, UCLA Department of Statistics;
US mail: 9432 Boelter Hall, Box 951554, Los Angeles, CA 90095-1554
phone (310)-825-9550;  fax (310)-206-5658;  email: deleeuw@stat.ucla.edu
homepage: http://www.stat.ucla.edu/~deleeuw
========================================================
           No matter where you go, there you are. --- Buckaroo Banzai
                    http://www.stat.ucla.edu/~deleeuw/sounds/nomatter.au
========================================================