RSBENCH(1)

πŸ’– rsbench(1) β€” the user-commands baddie ✨

okay bestie, first things first: this is the manual page for rsbench, and it lives in section 1 (User Commands) of the manual. yes we are that section. slay. πŸ’…


🌸 NAME 🌸

rsbench β€” benchmark and test for librsb. that's it. that's the whole personality. it benches, it tests, it does both. periodt. πŸŽ€


πŸ¦‹ SYNOPSIS πŸ¦‹

rsbench [--bench] [OPTIONS]
rsbench [ -o OPCODE] [ -O {subprogram-code}] [ {subprogram-specific-arguments} ]

two whole ways to call her. pick your fighter. ✨


πŸ‘‘ DESCRIPTION πŸ‘‘

rsbench is a swiss army knife for testing the library functionality and performance. no cap. it's giving "one binary, so many moods" and honestly? obsessed. πŸ§°πŸ¦‹

so the way that it works is you choose your {subprogram-code}, and the menu is:

  • r β€” for the reference benchmark (will produce a machine specific file)
  • c β€” for the complete benchmark
  • e β€” for the matrix experimentation code
  • d β€” for a single matrix dumpout
  • b β€” for the (current, going to be obsoleted) benchmark
  • t β€” for some matrix construction tests
  • o β€” obsolete, will soon be removed
⚠️ okay bestie listen β€” b is the current one but it is literally marked going to be obsoleted, and o is already obsolete and will soon be removed. do not get emotionally attached to o. i'm serious. 😭

{subprogram-specific-arguments} will be available from the subprograms. every subprogram got its own lil bag of tricks, iykyk.

and now, the examples, because this is where it clicks fr fr:

rsbench -O b -h

will show the current benchmark subprogram's options.

rsbench -o a -O b -h

will show the spmv benchmark subprogram's options.

rsbench -o n -O b -h

will show the negation benchmark subprogram's options.

see how -O picks the subprogram and -o narrows it down? ✨ that's the vibe check passing.

The default {subprogram-code} is 'b'. so if you don't say anything, she's gonna pick b for you. lowkey polite of her. 🫢

With OPCODE among 'actinS'.


🌈 the OPTIONS shelf 🌈

rsbench where OPTIONS are taken from:

  • -h, --help
  • --bench
  • -o, --matrix-operation <arg>
  • -O, --subprogram-operation <arg>
  • -I, --information
  • -C, --configuration
  • -H, --hardware-counters
  • -M, --memory-benchmark
  • -e, --experiments
  • -v, --version
  • -B, --blas-testing
  • -Q, --quick-blas-testing <arg>
  • -E, --error-testing <arg>
  • -F, --fp-bench
  • -t, --transpose-test
  • --limits-testing
  • -G, --guess-blocking
  • -g, --generate-matrix
  • --plot-matrix, --matrix-ls, --matrix-ls-latex
  • -P, --matrix-print <arg>
  • --read-performance-record <arg>, --help-read-performance-record, --setenv <arg>
πŸ’€ real talk: this page does not hand out one-line descriptions for these, and i am not about to make stuff up about your flags. the names are doing the talking. and ngl, they're eating. πŸ‘‘

πŸŽ€ the autotune argument tea πŸŽ€

Arguments to --want-autotune of the format "Ss[Xx[Tt[V[V]]]]", where S is the autotuning time in seconds, X is the number of tries, T the number of starting threads, V can be either q for quiet autotuning or v for a verbose one (can be specified twice).

Valid examples: 3.0s2x4tv, 3.0s2x0tq, 3.0s, 2.0s10x.

See documentation of rsb_tune_spmm for a full explanation of these parameters role in auto-tuning. yes, the brackets nest. yes, [V[V]] really can show up twice. she's a little extra and we love that for her. ✨🌸


✍️ AUTHOR ✍️

Written by michelemartone_AT_users_DOT_sourceforge_DOT_net.

πŸ› REPORTING BUGS πŸ›

Report bugs to michelemartone_AT_users_DOT_sourceforge_DOT_net.

Copyright Β© 2008-2026 Michele Martone.


and that's the whole page, bestie β€” you ate that read. stay iconic, go bench something. πŸ’œβœ¨

this was rsbench version: 1.3.0, dated 2026 β€” bye babes, see you in the next man page πŸ«ΆπŸŽ€

πŸ“¦ where this page comes from