microbenchmark_config.cpp

Go to the documentation of this file.

#include "jb/testing/microbenchmark_config.hpp"

namespace jb {
namespace testing {
/**
 * Define the default values for a microbenchmark config.
 *
 * This is an experiment in how to write highly-configurable classes
 * for JayBeams.  The intention is that any configuration parameter
 * can be set using these rules:
 *
 * -# if the configuration is set in the command-line, use the
 *    command-line value, otherwise
 * -# if the configuration is set in the configuration file, use the
 *    value in the configuration file, otherwise
 * -# if the program was compiled with a special setting for the
 *    default value, use that setting, otherwise
 * -# use whatever default was chosen by the developer.
 *
 * The configuration files themselves are found using a series of
 * rules (TBD).
 *
 * What I would want is that each component can define its own config
 * class, and the main program *choses* which ones become command-line
 * arguments.
 */
namespace defaults {

#ifndef JB_DEFAULTS_microbenchmark_warmup_iterations
#define JB_DEFAULTS_microbenchmark_warmup_iterations 100
#endif

#ifndef JB_DEFAULTS_microbenchmark_iterations
#define JB_DEFAULTS_microbenchmark_iterations 1000
#endif

#ifndef JB_DEFAULTS_microbenchmark_size
#define JB_DEFAULTS_microbenchmark_size 0
#endif

#ifndef JB_DEFAULTS_microbenchmark_verbose
#define JB_DEFAULTS_microbenchmark_verbose false
#endif

#ifndef JB_DEFAULTS_microbenchmark_reconfigure_thread
#define JB_DEFAULTS_microbenchmark_reconfigure_thread true
#endif

int warmup_iterations = JB_DEFAULTS_microbenchmark_warmup_iterations;
int iterations = JB_DEFAULTS_microbenchmark_iterations;
int size = JB_DEFAULTS_microbenchmark_size;
bool verbose = JB_DEFAULTS_microbenchmark_verbose;
bool reconfigure_thread = JB_DEFAULTS_microbenchmark_reconfigure_thread;

} // namespace defaults

microbenchmark_config::microbenchmark_config()
    : warmup_iterations(
          desc("warmup-iterations")
              .help("The number of warmup iterations in the benchmark."),
          this, defaults::warmup_iterations)
    , iterations(
          desc("iterations")
              .help("Number of iterations to run for a fixed size."),
          this, defaults::iterations)
    , size(
          desc("size").help(
              "If set (and not zero) control the size of the test."),
          this, defaults::size)
    , verbose(
          desc("verbose").help(
              "If true, dump the results of every test to stdout for"
              " statistical analysis."),
          this, defaults::verbose)
    , test_case(
          desc("test-case")
              .help("Some microbenchmarks test completely different "
                    "configurations"
                    ", settings, or even different algorithms for the "
                    "same problem."
                    "  Use this option to configure such benchmarks"
                    ", most microbenchmarks will ignore it."),
          this)
    , reconfigure_thread(
          desc("reconfigure-thread")
              .help("If set reconfigure the main thread scheduling parameters"
                    "before running the benchmark. "
                    "The actual scheduling parameters are configured via the "
                    "--thread option.  Unsetting this flag is useful when "
                    "testing "
                    "with external scheduling parameter settings, e.g. chrt(1) "
                    "and taskset(1)."),
          this, defaults::reconfigure_thread)
    , thread(
          desc("thread", "thread")
              .help("Configure how the main thread scheduling parameters are "
                    "set before running the benchmark. "
                    "By default we attempt to run the test in the FIFO "
                    "scheduling class, at the maximum allowable priority."),
          this, thread_config().scheduler("FIFO").priority("MAX"))
    , prefix(
          desc("prefix").help(
              "Define the prefix for the detailed results output."
              "  Often microbenchmark results are further analyzed with"
              " tools such as R or Python scripts, in such cases prefixing"
              " the output with the attributes of the test (size, test-case,"
              " system configuration, etc) can make it easier to mix the"
              " output from multiple microbenchmarks into a single result."
              " It is common, but not required, for these prefixes to be"
              " a comma-separated list of values."),
          this) {
}

} // namespace testing
} // namespace jb