def calculateFrameStatistics(self, traceName):
"""
For each frame marker event, calculate the sum of the instrumentation data
for the intermediate events. This operation will help to highlight frames
that are especially resource intensive.
@param traceName: Trace for which frame data should be calculated.
"""
if not traceName in self.analyzer.traces:
self.analyzer.fail("Trace not found: %s" % traceName)
trace = self.analyzer.traces[traceName]
events = TraceOperations.calculateFrameStatistics(self.analyzer.project, trace)
self.analyzer.reportInfo("Data for %d frames collected."% len(events))
def calculateFrameStatistics(self, traceName):
"""
For each frame marker event, calculate the sum of the instrumentation data
for the intermediate events. This operation will help to highlight frames
that are especially resource intensive.
@param traceName: Trace for which frame data should be calculated.
"""
if not traceName in self.analyzer.traces:
self.analyzer.fail("Trace not found: %s" % traceName)
trace = self.analyzer.traces[traceName]
events = TraceOperations.calculateFrameStatistics(
self.analyzer.project, trace)
self.analyzer.reportInfo("Data for %d frames collected." % len(events))
def calculateStatistics(self, frameSimilarityFunc = None, timeThreshold = .25, histogramThreshold = .96, frameCountThreshold = 20,
monochromaticThreshold = .99):
# Reset the list of frames
self.frames = []
self.interestingFrames = []
# Combine frame level statistics
TraceOperations.calculateFrameStatistics(self.project, self.trace)
# Calculate some high level statistics
events = []
for event in self.trace.events:
func = self.library.functions[event.name]
events.append(event)
if func.isFrameMarker:
frame = Frame()
frame.startTime = events[0].time / 1e6
frame.endTime = (events[-1].time + events[-1].duration) / 1e6
frame.events = events
frame.renderEvents = [e for e in events if self.library.functions[e.name].isRenderCall]
frame.swapEvent = event
frame.number = len(self.frames)
# Calculate frame duration
frame.duration = frame.endTime - frame.startTime
if frame.duration < MIN_FRAME_DURATION:
# If the frame has an essentially zero duration, the following event should give a proper time reading
try:
frame.duration = max(self.trace.events[self.trace.events.index(event) + 1].time / 1e6 - frame.startTime, MIN_FRAME_DURATION)
except IndexError:
frame.duration = MIN_FRAME_DURATION
self.frames.append(frame)
events = []
# Now prune the list of frames down to frames of special interest
lastFrame = None
histograms = {}
def getFrameHistogram(frame):
if not frame.swapEvent in histograms:
try:
image = player.Instrumentation.loadBuffer(frame.swapEvent)
except:
return
if image:
# Calculate a normalized histogram
hist = image.histogram()
f = 1.0 / (image.size[0] * image.size[1] * len(image.getbands()))
hist = [h * f for h in hist]
histograms[frame.swapEvent] = hist
return histograms.get(frame.swapEvent)
task = Task.startTask("choose-frames", "Choosing interesting frames", len(self.frames))
for frame in self.frames:
task.step()
if lastFrame and len(self.frames) > frameCountThreshold:
# If this frame's duration is pretty much the same as the last one, skip it
if abs(frame.duration - lastFrame.duration) <= timeThreshold * lastFrame.duration:
continue
# Get color buffer histograms for both images
hist1 = getFrameHistogram(frame)
hist2 = getFrameHistogram(lastFrame)
if hist1 and hist2:
# Calculate the Bhattacharyya distance, i.e. the cosine of the angle
# between the two histograms
dist = sum([math.sqrt(h2 * h1) for h1, h2 in zip(hist1, hist2)])
# If the distance is close to one, i.e. the frames are nearly identical,
# skip this frame.
if dist > histogramThreshold:
continue
# If the frame is mostly just one color, skip it
if max(hist1) >= monochromaticThreshold:
continue
# If we have a similarity function, discard similar frames
if frameSimilarityFunc and frameSimilarityFunc(lastFrame, frame):
continue
# Mark the new frame as interesting
self.interestingFrames.append(frame)
lastFrame = frame
def calculateStatistics(self,
frameSimilarityFunc=None,
timeThreshold=.25,
histogramThreshold=.96,
frameCountThreshold=20,
monochromaticThreshold=.99):
# Reset the list of frames
self.frames = []
self.interestingFrames = []
# Combine frame level statistics
TraceOperations.calculateFrameStatistics(self.project, self.trace)
# Calculate some high level statistics
events = []
for event in self.trace.events:
func = self.library.functions[event.name]
events.append(event)
if func.isFrameMarker:
frame = Frame()
frame.startTime = events[0].time / 1e6
frame.endTime = (events[-1].time + events[-1].duration) / 1e6
frame.events = events
frame.renderEvents = [
e for e in events
if self.library.functions[e.name].isRenderCall
]
frame.swapEvent = event
frame.number = len(self.frames)
# Calculate frame duration
frame.duration = frame.endTime - frame.startTime
if frame.duration < MIN_FRAME_DURATION:
# If the frame has an essentially zero duration, the following event should give a proper time reading
try:
frame.duration = max(
self.trace.events[self.trace.events.index(event) +
1].time / 1e6 - frame.startTime,
MIN_FRAME_DURATION)
except IndexError:
frame.duration = MIN_FRAME_DURATION
self.frames.append(frame)
events = []
# Now prune the list of frames down to frames of special interest
lastFrame = None
histograms = {}
def getFrameHistogram(frame):
if not frame.swapEvent in histograms:
try:
image = player.Instrumentation.loadBuffer(frame.swapEvent)
except:
return
if image:
# Calculate a normalized histogram
hist = image.histogram()
f = 1.0 / (image.size[0] * image.size[1] *
len(image.getbands()))
hist = [h * f for h in hist]
histograms[frame.swapEvent] = hist
return histograms.get(frame.swapEvent)
task = Task.startTask("choose-frames", "Choosing interesting frames",
len(self.frames))
for frame in self.frames:
task.step()
if lastFrame and len(self.frames) > frameCountThreshold:
# If this frame's duration is pretty much the same as the last one, skip it
if abs(frame.duration - lastFrame.duration
) <= timeThreshold * lastFrame.duration:
continue
# Get color buffer histograms for both images
hist1 = getFrameHistogram(frame)
hist2 = getFrameHistogram(lastFrame)
if hist1 and hist2:
# Calculate the Bhattacharyya distance, i.e. the cosine of the angle
# between the two histograms
dist = sum(
[math.sqrt(h2 * h1) for h1, h2 in zip(hist1, hist2)])
# If the distance is close to one, i.e. the frames are nearly identical,
# skip this frame.
if dist > histogramThreshold:
continue
# If the frame is mostly just one color, skip it
if max(hist1) >= monochromaticThreshold:
continue
# If we have a similarity function, discard similar frames
if frameSimilarityFunc and frameSimilarityFunc(
lastFrame, frame):
continue
# Mark the new frame as interesting
self.interestingFrames.append(frame)
lastFrame = frame
