[mlpack-svn] r15376 - mlpack/conf/jenkins-conf/benchmark/util

[fastlab-svn at coffeetalk-1.cc.gatech.edu]

Author: marcus
Date: Mon Jul  1 13:28:46 2013
New Revision: 15376

Log:
Add valgrind massif memory profiling function.

Added:
   mlpack/conf/jenkins-conf/benchmark/util/profiler.py

Added: mlpack/conf/jenkins-conf/benchmark/util/profiler.py
==============================================================================
--- (empty file)
+++ mlpack/conf/jenkins-conf/benchmark/util/profiler.py	Mon Jul  1 13:28:46 2013
@@ -0,0 +1,82 @@
+'''
+	@file profiler.py
+	@author Marcus Edel
+
+	Contains functions to get profiling informations.
+'''
+
+import os
+import sys
+import inspect
+
+# import the util path, this method even works if the path contains
+# symlinks to modules.
+cmd_subfolder = os.path.realpath(os.path.abspath(os.path.join(
+	os.path.split(inspect.getfile(inspect.currentframe()))[0], '')))
+if cmd_subfolder not in sys.path:
+	sys.path.insert(0, cmd_subfolder)
+
+from log import *
+
+'''
+This class implements functions the get profiling informations.
+'''
+class Profiler(object):
+
+	'''
+	Use valgrind massif to get memory profiling information and the save the ouput
+	in the specified file.
+
+	@param cmd - Method command line to profile.
+	@param output - Save the report at the output path with the specified name.
+	@param valgrind - Path to the valgrind binary.
+	@param options - Specified massif options.
+	@ return Returns False if the method was not successful, if the method was 
+	successful save the report file in the specified file. 
+	'''
+	@staticmethod
+	def MassifMemoryUsage(cmd, output, valgrind = "valgrind", options = ""):
+		import shlex, subprocess
+
+		cmd = shlex.split(("%s --tool=massif --massif-out-file=%s %s") % 
+				(valgrind, output, options)) + cmd
+		try:
+			s = subprocess.check_output(cmd, stderr=subprocess.STDOUT, shell=False)		
+		except Exception:
+			Log.Fatal("Could not execute command: " + str(cmd))
+			return -1
+
+	'''
+	Returns the memory used by a process and his children. We don't know when the 
+	process is done so we have to poll to get the memory. To avoid memory overflow
+	we use  a ringbuffer to limit the size of the memory values.
+
+	@param process - Popen instance.
+	@param Buffersize - Memory value count.
+	@return List of memory values.
+	'''
+	@staticmethod
+	def SubprocessMemoryUsage(process, Buffersize = 200):
+		import psutil, time, collections
+
+		# Create the process list with the main process and his childrens.
+		p = psutil.Process(process.pid)
+		children = list(p.get_children(recursive=True)) + [p]
+
+		memoryTable = collections.deque(maxlen = Buffersize)
+
+		# We have to poll to get the memory values.
+		while process.poll() == None:
+			try:
+				for p in children:
+					memoryTable.append(int(p.get_memory_info()[0]))
+			# Sometimes a subprocess has terminated in the time between we measure the
+			# memory. In this case, we continue.
+			except psutil.NoSuchProcess: 
+				continue				
+			except psutil.AccessDenied: 
+				continue
+
+			time.sleep(0.01)
+
+		return memoryTable