FOR005 cpu-time-for-wall-clock

warning enabled by default

cpu_time measures time in code that runs in parallel (OpenMP, OpenACC, MPI).

Why

cpu_time returns processor time of the calling process or thread, not elapsed time. With threads it adds up (or not, depending on the compiler) the time of all threads; with a GPU it reports almost nothing, because the host thread just waits. Measure elapsed time with system_clock (count and count_rate), omp_get_wtime() or MPI_Wtime(), after synchronising asynchronous work (!$acc wait).

Reported

call cpu_time(t0)

Preferred

call system_clock(c0, rate)
...
call system_clock(c1)
elapsed = real(c1 - c0) / rate

Suppressing

For one occurrence, with the reason on the same line (or alone on the line before the statement):

! fobis: ignore[FOR005] <why this occurrence is fine>

For the whole project, ignore = FOR005 in the fobos [lint] section.

References

Sections of the shipped knowledge packs behind this rule:

• hpc-practice:gpu-practice#gpu-benchmark-timing (fobis kb show hpc-practice:gpu-practice#gpu-benchmark-timing)
• openmp:patterns#portable-timing (fobis kb show openmp:patterns#portable-timing)