from lib.hachoir_core.tools import humanDurationNanosec from lib.hachoir_core.i18n import _ from math import floor from time import time class BenchmarkError(Exception): """ Error during benchmark, use str(err) to format it as string. """ def __init__(self, message): Exception.__init__(self, "Benchmark internal error: %s" % message) class BenchmarkStat: """ Benchmark statistics. This class automatically computes minimum value, maximum value and sum of all values. Methods: - append(value): append a value - getMin(): minimum value - getMax(): maximum value - getSum(): sum of all values - __len__(): get number of elements - __nonzero__(): isn't empty? """ def __init__(self): self._values = [] def append(self, value): self._values.append(value) try: self._min = min(self._min, value) self._max = max(self._max, value) self._sum += value except AttributeError: self._min = value self._max = value self._sum = value def __len__(self): return len(self._values) def __nonzero__(self): return bool(self._values) def getMin(self): return self._min def getMax(self): return self._max def getSum(self): return self._sum class Benchmark: def __init__(self, max_time=5.0, min_count=5, max_count=None, progress_time=1.0): """ Constructor: - max_time: Maximum wanted duration of the whole benchmark (default: 5 seconds, minimum: 1 second). - min_count: Minimum number of function calls to get good statistics (defaut: 5, minimum: 1). - progress_time: Time between each "progress" message (default: 1 second, minimum: 250 ms). - max_count: Maximum number of function calls (default: no limit). - verbose: Is verbose? (default: False) - disable_gc: Disable garbage collector? (default: False) """ self.max_time = max(max_time, 1.0) self.min_count = max(min_count, 1) self.max_count = max_count self.progress_time = max(progress_time, 0.25) self.verbose = False self.disable_gc = False def formatTime(self, value): """ Format a time delta to string: use humanDurationNanosec() """ return humanDurationNanosec(value * 1000000000) def displayStat(self, stat): """ Display statistics to stdout: - best time (minimum) - average time (arithmetic average) - worst time (maximum) - total time (sum) Use arithmetic avertage instead of geometric average because geometric fails if any value is zero (returns zero) and also because floating point multiplication lose precision with many values. """ average = stat.getSum() / len(stat) values = (stat.getMin(), average, stat.getMax(), stat.getSum()) values = tuple(self.formatTime(value) for value in values) print _("Benchmark: best=%s average=%s worst=%s total=%s") \ % values def _runOnce(self, func, args, kw): before = time() func(*args, **kw) after = time() return after - before def _run(self, func, args, kw): """ Call func(*args, **kw) as many times as needed to get good statistics. Algorithm: - call the function once - compute needed number of calls - and then call function N times To compute number of calls, parameters are: - time of first function call - minimum number of calls (min_count attribute) - maximum test time (max_time attribute) Notice: The function will approximate number of calls. """ # First call of the benchmark stat = BenchmarkStat() diff = self._runOnce(func, args, kw) best = diff stat.append(diff) total_time = diff # Compute needed number of calls count = int(floor(self.max_time / diff)) count = max(count, self.min_count) if self.max_count: count = min(count, self.max_count) # Not other call? Just exit if count == 1: return stat estimate = diff * count if self.verbose: print _("Run benchmark: %s calls (estimate: %s)") \ % (count, self.formatTime(estimate)) display_progress = self.verbose and (1.0 <= estimate) total_count = 1 while total_count < count: # Run benchmark and display each result if display_progress: print _("Result %s/%s: %s (best: %s)") % \ (total_count, count, self.formatTime(diff), self.formatTime(best)) part = count - total_count # Will takes more than one second? average = total_time / total_count if self.progress_time < part * average: part = max( int(self.progress_time / average), 1) for index in xrange(part): diff = self._runOnce(func, args, kw) stat.append(diff) total_time += diff best = min(diff, best) total_count += part if display_progress: print _("Result %s/%s: %s (best: %s)") % \ (count, count, self.formatTime(diff), self.formatTime(best)) return stat def validateStat(self, stat): """ Check statistics and raise a BenchmarkError if they are invalid. Example of tests: reject empty stat, reject stat with only nul values. """ if not stat: raise BenchmarkError("empty statistics") if not stat.getSum(): raise BenchmarkError("nul statistics") def run(self, func, *args, **kw): """ Run function func(*args, **kw), validate statistics, and display the result on stdout. Disable garbage collector if asked too. """ # Disable garbarge collector is needed and if it does exist # (Jython 2.2 don't have it for example) if self.disable_gc: try: import gc except ImportError: self.disable_gc = False if self.disable_gc: gc_enabled = gc.isenabled() gc.disable() else: gc_enabled = False # Run the benchmark stat = self._run(func, args, kw) if gc_enabled: gc.enable() # Validate and display stats self.validateStat(stat) self.displayStat(stat)