[R-SIG-Mac] auto vectorizing gcc?
simon.urbanek at math.uni-augsburg.de
Fri Nov 5 20:53:36 CET 2004
On Nov 5, 2004, at 8:49 AM, Jan de Leeuw wrote:
> Now where would the speedup be coming from ? vecLib
> has already been optimized, and vectorizing does not
> handle double precision floating point.
The only things I could imagine are some loops on R objects (such as
integer vectors), but you're right, it's very unlikely that this would
affect R code. I could only think of C code in packages under some
circumstances ... but admittedly rare. According to my tests you can
get ca. 30% speed gain alone by using gcc optimization flags, but R
won't make several basic the checks then ;). If you know what you're
doing, it may be ok (but don't rely on NAs then - they may turn into
More information about the R-SIG-Mac