o ϯi @sddlZddlZddlmZmZmZmZddlZ dde de deegefgegee e ffffddZ ddd d egefd e d eede de ddf d dZ dS)N)AnyCallableDictList num_iters warmup_itersreturncs4dtgtfdtgtttffffdd }|S)a This is intened to be used as a decorator to time any function. Args: num_iters (int): number of iterations used to compute the average time (sec) required to run the function. If negative, the number of iterations is determined dynamically by running the function a few times to make sure the estimate is stable. warmup_iters (int): number of iterations used to warm up the function. This is useful for functions that exhibit poor performance during the first few times they run (due to caches, autotuning, etc). Returns: Dict[str, float]: dictionary of the aggregated timing estimates. "iterations": number of iterations used to compute the estimated time. "mean": averate time (sec) used to run the function. "median": median time (sec) used to run the function. "min": minimal time (sec) used to run the function. "max": maximal time (sec) used to run the function. "stddev": standard deviation of the time (sec) used to run the function. funcr cs*dtdtdtttfffdd }|S)Nargskwargsr c stD] }|i|qd}d}g}dkrntj}t|D]*}t}|i|t|}||||7}|d7}dkrK|dkrKnq!|t|ksTJi} || d<||| d<t|| d<t|| d<t || d<t || d <| S) Nrg? iterationsmeanmedianminmaxstddev) rangesysmaxsizetimeappendlennprrrstd) r r _ total_timecount run_times max_num_iters start_timerun_timeret)r rrK/home/ubuntu/.local/lib/python3.10/site-packages/fvcore/common/benchmark.py decorated%s2     z,timeit..decorator..decorated)rrstrfloat)r r&rr)r r% decorator$s& ztimeit..decorator)rrrr'r()rrr*r$r)r%timeit s0#r+r)r bm_name kwargs_listc Cs~tdg}|D]A}|di|}t||d|}|} |} |r0| dddd|D7} || t| ddt| ddt| d gqt|}tt |d d d ftd } tt |d d d ftd } tt |d d dftd } tt |d d dftd }d}td d| d| d| d|t||D]}td |d | t |d t |dt |dqt|d S)u Benchmark the input function and print out the results. Args: func (callable): a closure that returns a function for benchmarking, where initialization can be done before the function to benchmark. bm_name (str): name of the benchmark to print out, e.g. "BM_UPDATE". kwargs_list (list): a list of argument dict to pass to the function. The intput function will be timed separately for each argument dict. num_iters (int): number of iterations to run. Defaults to run until 0.5s. warmup_iters (int): number of iterations used to warm up the function. Outputs: For each argument dict, print out the time (in microseconds) required to run the function along with the number of iterations used to get the timing estimate. Example output: Benchmark Avg Time(μs) Peak Time(μs) Iterations ------------------------------------------------------------------- BM_UPDATE_100 820 914 610 BM_UPDATE_1000 7655 8709 66 BM_UPDATE_10000 78062 81748 7 ------------------------------------------------------------------- r)rcss|] \}}t|VqdS)N)r').0kvr$r$r% tszbenchmark..ri@BrrNr)keyr zP--------------------------------------------------------------------------------z{:{}s} {:>{}s} {:>{}s} {:>{}s} Benchmarku Avg Time(μs)uPeak Time(μs) Iterationsz{:{}s} {:15.0f} {:15.0f} {:14d}r$) printr+joinitemsrr'rarrayrrformatr(int)r r,r-rroutputsr func_bm time_funcr#namec1c2c3c4dashoutputr$r$r% benchmarkJsF!   & rH)rr)rrtypingrrrrnumpyrr=r'r(r+rHr$r$r$r%s6$ E