[R] Does stats::cor use the underlying BLAS? 32- vs. 64-bit issue
peter.langfelder at gmail.com
Sat Dec 4 02:42:27 CET 2010
I'm curious whether the standard correlation calculation implemented
in stats::cor uses the underlying BLAS (at least in the default case
use = "all.obs").
On a 32-bit linux system using R-2.11.1 compiled with GotoBLAS,
stats::cor is as fast as matrix multiplication, so it would appear
stats::cor does use the BLAS library. On the other hand, on a 64-bit
linux R-2.11.1 compiled with ATLAS BLAS, stats::cor doesn't seem to
use the BLAS (from both speed and CPU usage - CPU usage in BLAS matrix
multiplication goes up to several hundred %, reflecting
multithreading, whereas in stats::cor calculation it remains near
Any expert views on why the difference between 32- and 64-bit systems?
More information about the R-help