For those interested:
I compiled R-1.9.0beta_2004-04-02 with pgf77/gcc/g++ and g77/gcc/g++, both
_without_ fast BLAS. Both passed make check-all. Simple crossprod(m) for
fairly large `m' shows that the version compiled with pgf77 is about 15%
faster:
[andy@leo:scratch]% echo "set.seed(1); m <- matrix(runif(5e7), 1e4);
gc();
system.time(crossprod(m))" | R-1.9.0-PGI/bin/R --slave -q
used (Mb) gc trigger (Mb)
Ncells 431805 23.1 741108 39.6
Vcells 50125266 382.5 150395933 1147.5
[1] 671.14 0.30 762.35 0.00 0.00
[andy@leo:scratch]% echo "set.seed(1); m <- matrix(runif(5e7), 1e4);
gc();
system.time(crossprod(m))" | R-1.9.0-GCC/bin/R --slave -q
used (Mb) gc trigger (Mb)
Ncells 431805 23.1 741108 39.6
Vcells 50125266 382.5 150395933 1147.5
[1] 777.55 0.40 882.37 0.00 0.00
I'd welcome comments and suggestions on code for further comparison. I will
also keep trying to compile R with pgcc/pgCC and see if further gain is
possible.
Best,
Andy
--
Andy Liaw, PhD
Biometrics Research PO Box 2000, RY33-300
Merck Research Labs Rahway, NJ 07065
mailto:andy_liaw@merck.com 732-594-0820
