Option

-all

Reports all three statistics: MPT/s, MOP/s, MB/s

-cal M

Sets to 2^M the problem size for calibration. The default is four, that is a problem size of 16.

Calibration attempts to determine the number of times to run a benchmark to ensure that the total measured time is within a threshold of the goal time, set by -ms.

-nocal

Do not calibrate. The default is to calibrate with a problem size of 2^4 or 16. The -cal option can change this size.

-center M

Sets the range type to centered with the center value M.

-data

Must be combined with -samples M where M >= 3. Displays many columns of data suitable for a comma-separated-values spreadsheet. The columns are:

problem size

median time

minimum time

maximum time

memory per point

operations per point

input memory per point

output memory per point

number of loops

time per loop

-diag

Run the benchmark in diagnostic mode. For many benchmarks, this displays information on what evaluators or backend libraries are used in the benchmark.

-fix_loop

Runs a rough set of executions with no calibration and one loop per problem size.

-geom

Uses a geometric progression of problem sizes. For start <= I <= stop, the problem size is 2^I. See also -linear.

-iob

Reports millions of input/output operations per second (MB/s) (by summing read and write iob_per_point).

-lat

Reports seconds per point.

-lib_config

Displays a summary of the configuration of the VSIPL++ library.

-linear M

Uses a linear progression of problem sizes. For start <= I <= stop, the problem size is M*I.

-loop_start M

Initial number of times to run the problem during calibration. The default is one. Calibration may increase or decrease M to ensure that the total measured time is within a threshold of the goal time, set by -ms.

-mem

In the table of results, the left column usually is the number of points. This option switches that value to the memory footprint of the problem. It can be useful in illustrating how performance is affected by problem size.

-mfile F

Reads the progression values from file F. Implies -start M where M is the first value read from the file.

-ms M

Sets the goal time that each measurement should take, in hundredths-of-a-second. Defaults to 25 (250 ms).

-ops

Reports millions of operations per second (MOP/s).

-p:KEY VALUE

Sets the benchmark specific parameter KEY to the value VALUE. KEY and VALUE are string values. See the individual benchmarks for valid uses.

-param M

Sets the user-specified parameter, where value is used in some instances to override a default value of a parameter, such as the number of rows or columns in an input matrix for example. The effect, if any, is dependent on each individual benchmark.

Using -param to set parameters is being phased out in favor of -p:KEY VALUE.

-pause

When using multiple processes, stop each one to allow a debugger to be attached. The process continues after the user types a single character.

-pool def

Uses the default memory pool, which allocates aligned memory with the new operator.

-pool huge

Uses the huge page memory pool, which allocates aligned memory in huge pages. Uses the default number of huge pages.

-pool huge:N

Uses the huge page memory pool, which allocates aligned memory in huge pages. Allocates N huge memory pages.

-prof

Enables profiling. An accumulate-mode profile output is generated for each problem size.

-pts

Reports millions of points per second (MPT/s). This is the default.

-riob

Reports millions of input operations per second (MB/s).

-samples M

Sets the number of samples taken to M. Defaults to one. When M > 2, the median value is reported.

-show_loop

Displays the number of times the algorithm is run for each problem size.

-show_time

Displays the total time the algorithm runs for each problem size.

-single M

Runs the benchmark once on the problem size 2^M only.

-start M

Sets the starting problem size to 2^M. Defaults to two (four points). Uses this size for calibration also.

-steady M

Runs the benchmark in a loop on the problem size 2^M only. The user must stop the program manually.

-stop M

Sets the stopping problem size to 2^M. Defaults to 21 (2097152 points).

-verbose

Sets a flag to enable printing of internal information. Used by authors during benchmark development.

-wiob

Reports millions of output operations per second (MB/s).