Number of threads used by SparseArray operations

Use get_SparseArray_nthread or set_SparseArray_nthread to get or set the number of threads to use by the multithreaded operations implemented in the SparseArray package.

Usage

get_SparseArray_nthread()
set_SparseArray_nthread(nthread=NULL)

Arguments

nthread

The number of threads to use by multithreaded operations implemented in the SparseArray package.

On systems where OpenMP is available, this must be NULL or an integer value >= 1. When NULL (the default), a "reasonable" value will be used that never exceeds one third of the number of logical cpus available on the machine.

On systems where OpenMP is not available, the supplied nthread is ignored and set_SparseArray_nthread() is a no-op.

Details

Multithreaded operations in the SparseArray package are implemented in C with OpenMP (https://www.openmp.org/).

Note that OpenMP is not available on all systems. On systems where it's available, get_SparseArray_nthread() is guaranteed to return a value >= 1. On systems where it's not available (e.g. macOS), get_SparseArray_nthread() returns 0 and set_SparseArray_nthread() is a no-op.

IMPORTANT: The portable way to disable multithreading is by calling set_SparseArray_nthread(1), NOT set_SparseArray_nthread(0) (the latter returns an error on systems where OpenMP is available).

Value

get_SparseArray_nthread() returns an integer value >= 1 on systems where OpenMP is available, and 0 on systems where it's not.

set_SparseArray_nthread() returns the previous nthread value, that is, the value returned by get_SparseArray_nthread() before the call to set_SparseArray_nthread(). Note that the value is returned invisibly.

See also

• SparseArray_matrixStats for SparseArray col/row summarization methods.

• SparseMatrix_mult for SparseMatrix multiplication and cross-product.

• SparseArray objects.

Examples

get_SparseArray_nthread()
#> [1] 1

if (get_SparseArray_nthread() != 0) {  # multithreading is available
    svt1 <- poissonSparseMatrix(77000L, 15000L, density=0.01)

    ## 'user' time is typically N x 'elapsed' time where N is roughly the
    ## number of threads that was effectively used:
    system.time(cv1 <- colVars(svt1))

    svt2 <- poissonSparseMatrix(77000L, 300L, density=0.3) * 0.77
    system.time(cp12 <- crossprod(svt1, svt2))

    prev_nthread <- set_SparseArray_nthread(1)  # disable multithreading
    system.time(cv1 <- colVars(svt1))
    system.time(cp12 <- crossprod(svt1, svt2))

    ## Restore previous 'nthread' value:
    set_SparseArray_nthread(prev_nthread)
}