''' Created on May 16, 2011 @author: bungeman ''' import sys import getopt import re import os import bench_util import json import xml.sax.saxutils # We throw out any measurement outside this range, and log a warning. MIN_REASONABLE_TIME = 0 MAX_REASONABLE_TIME = 99999 def usage(): """Prints simple usage information.""" print '-b the bench to show.' print '-c the config to show (GPU, 8888, 565, etc).' print '-d a directory containing bench_r_ files.' print '-f [:] the revisions to use for fitting.' print ' Negative is taken as offset from most recent revision.' print '-l title to use for the output graph' print '-o <path> path to which to write output; writes to stdout if not specified' print '-r <revision>[:<revision>] the revisions to show.' print ' Negative <revision> is taken as offset from most recent revision.' print '-s <setting>[=<value>] a setting to show (alpha, scalar, etc).' print '-t <time> the time to show (w, c, g, etc).' print '-x <int> the desired width of the svg.' print '-y <int> the desired height of the svg.' print '--default-setting <setting>[=<value>] setting for those without.' class Label: """The information in a label. (str, str, str, str, {str:str})""" def __init__(self, bench, config, time_type, settings): self.bench = bench self.config = config self.time_type = time_type self.settings = settings def __repr__(self): return "Label(%s, %s, %s, %s)" % ( str(self.bench), str(self.config), str(self.time_type), str(self.settings), ) def __str__(self): return "%s_%s_%s_%s" % ( str(self.bench), str(self.config), str(self.time_type), str(self.settings), ) def __eq__(self, other): return (self.bench == other.bench and self.config == other.config and self.time_type == other.time_type and self.settings == other.settings) def __hash__(self): return (hash(self.bench) ^ hash(self.config) ^ hash(self.time_type) ^ hash(frozenset(self.settings.iteritems()))) def get_latest_revision(directory): """Returns the latest revision number found within this directory. """ latest_revision_found = -1 for bench_file in os.listdir(directory): file_name_match = re.match('bench_r(\d+)_(\S+)', bench_file) if (file_name_match is None): continue revision = int(file_name_match.group(1)) if revision > latest_revision_found: latest_revision_found = revision if latest_revision_found < 0: return None else: return latest_revision_found def parse_dir(directory, default_settings, oldest_revision, newest_revision): """Parses bench data from files like bench_r<revision>_<scalar>. (str, {str, str}, Number, Number) -> {int:[BenchDataPoints]}""" revision_data_points = {} # {revision : [BenchDataPoints]} for bench_file in os.listdir(directory): file_name_match = re.match('bench_r(\d+)_(\S+)', bench_file) if (file_name_match is None): continue revision = int(file_name_match.group(1)) scalar_type = file_name_match.group(2) if (revision < oldest_revision or revision > newest_revision): continue file_handle = open(directory + '/' + bench_file, 'r') if (revision not in revision_data_points): revision_data_points[revision] = [] default_settings['scalar'] = scalar_type revision_data_points[revision].extend( bench_util.parse(default_settings, file_handle)) file_handle.close() return revision_data_points def add_to_revision_data_points(new_point, revision, revision_data_points): """Add new_point to set of revision_data_points we are building up. """ if (revision not in revision_data_points): revision_data_points[revision] = [] revision_data_points[revision].append(new_point) def filter_data_points(unfiltered_revision_data_points): """Filter out any data points that are utterly bogus. Returns (allowed_revision_data_points, ignored_revision_data_points): allowed_revision_data_points: points that survived the filter ignored_revision_data_points: points that did NOT survive the filter """ allowed_revision_data_points = {} # {revision : [BenchDataPoints]} ignored_revision_data_points = {} # {revision : [BenchDataPoints]} revisions = unfiltered_revision_data_points.keys() revisions.sort() for revision in revisions: for point in unfiltered_revision_data_points[revision]: if point.time < MIN_REASONABLE_TIME or point.time > MAX_REASONABLE_TIME: add_to_revision_data_points(point, revision, ignored_revision_data_points) else: add_to_revision_data_points(point, revision, allowed_revision_data_points) return (allowed_revision_data_points, ignored_revision_data_points) def get_abs_path(relative_path): """My own implementation of os.path.abspath() that better handles paths which approach Window's 260-character limit. See https://code.google.com/p/skia/issues/detail?id=674 This implementation adds path components one at a time, resolving the absolute path each time, to take advantage of any chdirs into outer directories that will shorten the total path length. TODO: share a single implementation with upload_to_bucket.py, instead of pasting this same code into both files.""" if os.path.isabs(relative_path): return relative_path path_parts = relative_path.split(os.sep) abs_path = os.path.abspath('.') for path_part in path_parts: abs_path = os.path.abspath(os.path.join(abs_path, path_part)) return abs_path def redirect_stdout(output_path): """Redirect all following stdout to a file. You may be asking yourself, why redirect stdout within Python rather than redirecting the script's output in the calling shell? The answer lies in https://code.google.com/p/skia/issues/detail?id=674 ('buildbot: windows GenerateBenchGraphs step fails due to filename length'): On Windows, we need to generate the absolute path within Python to avoid the operating system's 260-character pathname limit, including chdirs.""" abs_path = get_abs_path(output_path) sys.stdout = open(abs_path, 'w') def create_lines(revision_data_points, settings , bench_of_interest, config_of_interest, time_of_interest): """Convert revision data into sorted line data. ({int:[BenchDataPoints]}, {str:str}, str?, str?, str?) -> {Label:[(x,y)] | [n].x <= [n+1].x}""" revisions = revision_data_points.keys() revisions.sort() lines = {} # {Label:[(x,y)] | x[n] <= x[n+1]} for revision in revisions: for point in revision_data_points[revision]: if (bench_of_interest is not None and not bench_of_interest == point.bench): continue if (config_of_interest is not None and not config_of_interest == point.config): continue if (time_of_interest is not None and not time_of_interest == point.time_type): continue skip = False for key, value in settings.items(): if key in point.settings and point.settings[key] != value: skip = True break if skip: continue line_name = Label(point.bench , point.config , point.time_type , point.settings) if line_name not in lines: lines[line_name] = [] lines[line_name].append((revision, point.time)) return lines def bounds(lines): """Finds the bounding rectangle for the lines. {Label:[(x,y)]} -> ((min_x, min_y),(max_x,max_y))""" min_x = bench_util.Max min_y = bench_util.Max max_x = bench_util.Min max_y = bench_util.Min for line in lines.itervalues(): for x, y in line: min_x = min(min_x, x) min_y = min(min_y, y) max_x = max(max_x, x) max_y = max(max_y, y) return ((min_x, min_y), (max_x, max_y)) def create_regressions(lines, start_x, end_x): """Creates regression data from line segments. ({Label:[(x,y)] | [n].x <= [n+1].x}, Number, Number) -> {Label:LinearRegression}""" regressions = {} # {Label : LinearRegression} for label, line in lines.iteritems(): regression_line = [p for p in line if start_x <= p[0] <= end_x] if (len(regression_line) < 2): continue regression = bench_util.LinearRegression(regression_line) regressions[label] = regression return regressions def bounds_slope(regressions): """Finds the extreme up and down slopes of a set of linear regressions. ({Label:LinearRegression}) -> (max_up_slope, min_down_slope)""" max_up_slope = 0 min_down_slope = 0 for regression in regressions.itervalues(): min_slope = regression.find_min_slope() max_up_slope = max(max_up_slope, min_slope) min_down_slope = min(min_down_slope, min_slope) return (max_up_slope, min_down_slope) def main(): """Parses command line and writes output.""" try: opts, _ = getopt.getopt(sys.argv[1:] , "b:c:d:f:l:o:r:s:t:x:y:" , "default-setting=") except getopt.GetoptError, err: print str(err) usage() sys.exit(2) directory = None config_of_interest = None bench_of_interest = None time_of_interest = None revision_range = '0:' regression_range = '0:' latest_revision = None requested_height = None requested_width = None title = 'Bench graph' settings = {} default_settings = {} def parse_range(range): """Takes '<old>[:<new>]' as a string and returns (old, new). Any revision numbers that are dependent on the latest revision number will be filled in based on latest_revision. """ old, _, new = range.partition(":") old = int(old) if old < 0: old += latest_revision; if not new: new = latest_revision; new = int(new) if new < 0: new += latest_revision; return (old, new) def add_setting(settings, setting): """Takes <key>[=<value>] adds {key:value} or {key:True} to settings.""" name, _, value = setting.partition('=') if not value: settings[name] = True else: settings[name] = value try: for option, value in opts: if option == "-b": bench_of_interest = value elif option == "-c": config_of_interest = value elif option == "-d": directory = value elif option == "-f": regression_range = value elif option == "-l": title = value elif option == "-o": redirect_stdout(value) elif option == "-r": revision_range = value elif option == "-s": add_setting(settings, value) elif option == "-t": time_of_interest = value elif option == "-x": requested_width = int(value) elif option == "-y": requested_height = int(value) elif option == "--default-setting": add_setting(default_settings, value) else: usage() assert False, "unhandled option" except ValueError: usage() sys.exit(2) if directory is None: usage() sys.exit(2) latest_revision = get_latest_revision(directory) oldest_revision, newest_revision = parse_range(revision_range) oldest_regression, newest_regression = parse_range(regression_range) unfiltered_revision_data_points = parse_dir(directory , default_settings , oldest_revision , newest_revision) # Filter out any data points that are utterly bogus... make sure to report # that we did so later! (allowed_revision_data_points, ignored_revision_data_points) = filter_data_points( unfiltered_revision_data_points) # Update oldest_revision and newest_revision based on the data we could find all_revision_numbers = allowed_revision_data_points.keys() oldest_revision = min(all_revision_numbers) newest_revision = max(all_revision_numbers) lines = create_lines(allowed_revision_data_points , settings , bench_of_interest , config_of_interest , time_of_interest) regressions = create_regressions(lines , oldest_regression , newest_regression) output_xhtml(lines, oldest_revision, newest_revision, ignored_revision_data_points, regressions, requested_width, requested_height, title) def qa(out): """Stringify input and quote as an xml attribute.""" return xml.sax.saxutils.quoteattr(str(out)) def qe(out): """Stringify input and escape as xml data.""" return xml.sax.saxutils.escape(str(out)) def create_select(qualifier, lines, select_id=None): """Output select with options showing lines which qualifier maps to it. ((Label) -> str, {Label:_}, str?) -> _""" options = {} #{ option : [Label]} for label in lines.keys(): option = qualifier(label) if (option not in options): options[option] = [] options[option].append(label) option_list = list(options.keys()) option_list.sort() print '<select class="lines"', if select_id is not None: print 'id=%s' % qa(select_id) print 'multiple="true" size="10" onchange="updateSvg();">' for option in option_list: print '<option value=' + qa('[' + reduce(lambda x,y:x+json.dumps(str(y))+',',options[option],"")[0:-1] + ']') + '>'+qe(option)+'</option>' print '</select>' def output_ignored_data_points_warning(ignored_revision_data_points): """Write description of ignored_revision_data_points to stdout as xhtml. """ num_ignored_points = 0 description = '' revisions = ignored_revision_data_points.keys() if revisions: revisions.sort() revisions.reverse() for revision in revisions: num_ignored_points += len(ignored_revision_data_points[revision]) points_at_this_revision = [] for point in ignored_revision_data_points[revision]: points_at_this_revision.append(point.bench) points_at_this_revision.sort() description += 'r%d: %s\n' % (revision, points_at_this_revision) if num_ignored_points == 0: print 'Did not discard any data points; all were within the range [%d-%d]' % ( MIN_REASONABLE_TIME, MAX_REASONABLE_TIME) else: print '<table width="100%" bgcolor="ff0000"><tr><td align="center">' print 'Discarded %d data points outside of range [%d-%d]' % ( num_ignored_points, MIN_REASONABLE_TIME, MAX_REASONABLE_TIME) print '</td></tr><tr><td width="100%" align="center">' print ('<textarea rows="4" style="width:97%" readonly="true" wrap="off">' + qe(description) + '</textarea>') print '</td></tr></table>' def output_xhtml(lines, oldest_revision, newest_revision, ignored_revision_data_points, regressions, requested_width, requested_height, title): """Outputs an svg/xhtml view of the data.""" print '<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN"', print '"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">' print '<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en">' print '<head>' print '<title>%s' % qe(title) print '' print '' output_svg(lines, regressions, requested_width, requested_height) #output the manipulation controls print """ """ print '' % requested_width print """ ' print '
""" print '' print """
Bench Type Bitmap Config Timer Type (Cpu/Gpu/wall)
' create_select(lambda l: l.bench, lines, 'benchSelect') print '' create_select(lambda l: l.config, lines) print '' create_select(lambda l: l.time_type, lines) all_settings = {} variant_settings = set() for label in lines.keys(): for key, value in label.settings.items(): if key not in all_settings: all_settings[key] = value elif all_settings[key] != value: variant_settings.add(key) for k in variant_settings: create_select(lambda l: l.settings[k], lines) print '' print '' print '

""" output_ignored_data_points_warning(ignored_revision_data_points) print '
' print '
' print '' % ( qe(title), bench_util.CreateRevisionLink(oldest_revision), bench_util.CreateRevisionLink(newest_revision)) print """
%s

revisions r%s - r%s

Brighter red indicates tests that have gotten worse; brighter green indicates tests that have gotten better.

To highlight individual tests, hold down CONTROL and mouse over graph lines.

To highlight revision numbers, hold down SHIFT and mouse over the graph area.

To only show certain tests on the graph, select any combination of tests in the selectors at left. (To show all, select all.)

Use buttons at left to mark/clear points on the lines for selected benchmarks.

""" def compute_size(requested_width, requested_height, rev_width, time_height): """Converts potentially empty requested size into a concrete size. (Number?, Number?) -> (Number, Number)""" pic_width = 0 pic_height = 0 if (requested_width is not None and requested_height is not None): pic_height = requested_height pic_width = requested_width elif (requested_width is not None): pic_width = requested_width pic_height = pic_width * (float(time_height) / rev_width) elif (requested_height is not None): pic_height = requested_height pic_width = pic_height * (float(rev_width) / time_height) else: pic_height = 800 pic_width = max(rev_width*3 , pic_height * (float(rev_width) / time_height)) return (pic_width, pic_height) def output_svg(lines, regressions, requested_width, requested_height): """Outputs an svg view of the data.""" (global_min_x, _), (global_max_x, global_max_y) = bounds(lines) max_up_slope, min_down_slope = bounds_slope(regressions) #output global_min_y = 0 x = global_min_x y = global_min_y w = global_max_x - global_min_x h = global_max_y - global_min_y font_size = 16 line_width = 2 pic_width, pic_height = compute_size(requested_width, requested_height , w, h) def cw(w1): """Converts a revision difference to display width.""" return (pic_width / float(w)) * w1 def cx(x): """Converts a revision to a horizontal display position.""" return cw(x - global_min_x) def ch(h1): """Converts a time difference to a display height.""" return -(pic_height / float(h)) * h1 def cy(y): """Converts a time to a vertical display position.""" return pic_height + ch(y - global_min_y) print '' print """ """ #output the revisions print """ """ def print_rect(x, y, w, h, revision): """Outputs a revision rectangle in display space, taking arguments in revision space.""" disp_y = cy(y) disp_h = ch(h) if disp_h < 0: disp_y += disp_h disp_h = -disp_h print '' xes = set() for line in lines.itervalues(): for point in line: xes.add(point[0]) revisions = list(xes) revisions.sort() left = x current_revision = revisions[0] for next_revision in revisions[1:]: width = (((next_revision - current_revision) / 2.0) + (current_revision - left)) print_rect(left, y, width, h, current_revision) left += width current_revision = next_revision print_rect(left, y, x+w - left, h, current_revision) #output the lines print """ """ for label, line in lines.items(): print '' % qa(label) r = 128 g = 128 b = 128 a = .10 if label in regressions: regression = regressions[label] min_slope = regression.find_min_slope() if min_slope < 0: d = max(0, (min_slope / min_down_slope)) g += int(d*128) a += d*0.9 elif min_slope > 0: d = max(0, (min_slope / max_up_slope)) r += int(d*128) a += d*0.9 slope = regression.slope intercept = regression.intercept min_x = regression.min_x max_x = regression.max_x print '' print '' print '' #output the labels print ' ' % qa(font_size) print '' print ' ' print '' if __name__ == "__main__": main()