tgstat aims to optimize and expand some of R’s common
functionality. Various approaches are used to achieve this goal,
including the use of optimized functions provided by the Basic Linear
Algebra Subprograms (BLAS) library. Many optimized BLAS implementations
are available, both proprietary (e.g. Intel’s MKL, Apple’s vecLib) and
opensource (e.g. OpenBLAS, ATLAS). Unfortunately, R often uses by
default the reference BLAS implementation, which is known to have poor
performance.
tgstat runs best when R is linked with an optimized BLAS
implementation. However, having tgstat rely on the reference BLAS will
result in very poor performance and is strongly discouraged. If your R
implementation uses an optimized BLAS, set
options(tgs_use.blas=TRUE) to allow tgstat to make BLAS
calls. Otherwise, set options(tgs_use.blas=FALSE) (default)
which instructs tgstat to avoid BLAS and instead rely only on its own
optimization methods. If in doubt, it is possible to run one of tgstat’s
CPU intensive functions (e.g. tgs_cor) and compare its run
time under both options(tgs_use.blas=FALSE).
Exact instructions for linking R with an optimized BLAS library are system dependent and are out of scope of this document.
tgstat uses multitasking to speed up performance. The number of
concurrent running processes depends on the task and the number of CPU
cores available on the specific machine. It is also possible to manually
set a hard limit for the maximal number of processes used by
options(tgs_max.processes=...).