The FMM Procedure (Experimental)

PERFORMANCE Statement

PERFORMANCE <performance-options> ;

The PERFORMANCE statement enables you to control the performance characteristics of the FMM procedure (for example, the number of CPUs, the number of threads for multithreading, and so on). By default, the FMM procedure performs many analyses in multiple threads, and the number of threads equals the number of CPUs. Certain system and configuration options also can control the number of CPUs available to a SAS session or whether multithreaded computations are permissible. For example, you can set the number of available processors to two with

options cpucount=2;

The FMM procedure then acts as though two processors were available, regardless of the number of physically available processors.

The FMM procedure applies multithreading to the following analytical tasks:

Starting values:: all starting value computations that require a pass through the data.
Optimization:: all evaluations of objective function, gradient, and Hessian; computation of covariance matrix.
Bayesian analysis:: all sample passes through the data, formation of cross-product matrices, sampling of latent variables, and posterior diagnostics.
Scoring and ODS Graphics:: computation of all output statistics and statistics for the construction of graphics that require passes through the data.

You can specify the following performance-options:

CPUCOUNT=n
CPUCOUNT=ACTUAL: specifies the number of processors available to the FMM procedure; the number n must be between 1 and 1024. CPUCOUNT=ACTUAL sets the number of available processors equal to the number of physical processors.
DETAILS: requests a table with timing detail for the tasks performed by the FMM procedure.
NOTHREADS: disables multithreaded computations.
THREADS=YES
THREADS=NO: enables or disables multithreaded processing. The number of threads used by the FMM procedure is displayed in the "Bayes Information" or "Optimization Information" table. It typically equals the number of available CPUs, which can be different from the number of physical CPUs, and can be modified with the global CPUCOUNT SAS option or with the CPUCOUNT= option in the PERFORMANCE statement.