Allow creation of flame graphs from UIforETW menu

Flame graphs can be handy for summarizing CPU usage but if they are too
difficult to use they won't get used. This is a first attempt at making
them more accessible. This adds a context menu that creates a flame
graph for the busiest thread.

Future work could include specifying a time range, process of interest,
thread of interest, number of threads to report on, or threshold - so
many choices! It would be better if this was burned in to WPA, but oh
well.

This also rearranges the context menu by putting scripts in a sub-menu
and compression in another - it was getting too big.

UIforETW/Resource.h

@@ -86,13 +86,16 @@
 #define ID_PASTEOVERRIDE                32805
 #define ID_TRACES_IDENTIFYCHROMEPROCESSES 32807
 #define ID_TRACES_OPENTRACEINEXPERIENCEANALYZER 32808
+#define ID_TRACES_SCRIPTS               32809
+#define ID_SCRIPTS_CREATEFLAMEGRAPH     32810
+#define ID_TRACES_TRACECOMPRESSION      32811
 
 // Next default values for new objects
 // 
 #ifdef APSTUDIO_INVOKED
 #ifndef APSTUDIO_READONLY_SYMBOLS
 #define _APS_NEXT_RESOURCE_VALUE        136
-#define _APS_NEXT_COMMAND_VALUE         32809
+#define _APS_NEXT_COMMAND_VALUE         32812
 #define _APS_NEXT_CONTROL_VALUE         1034
 #define _APS_NEXT_SYMED_VALUE           101
 #endif

UIforETW/UIforETW.rc

@@ -244,14 +244,21 @@ BEGIN
         MENUITEM "Open trace in Media e&Xperience Analyzer", ID_TRACES_OPENTRACEINEXPERIENCEANALYZER
         MENUITEM "&Delete trace\tDelete",       ID_TRACES_DELETETRACE
         MENUITEM "&Rename trace\tF2",           ID_TRACES_RENAMETRACE
-        MENUITEM "&ETW compress trace (Win 8.0+)", ID_TRACES_COMPRESSTRACE
-        MENUITEM "&Zip compress trace",         ID_TRACES_ZIPCOMPRESSTRACE
-        MENUITEM "ETW Compress all &traces (Win 8.0+)", ID_TRACES_COMPRESSTRACES
-        MENUITEM "Zip compress all traces",     ID_TRACES_ZIPCOMPRESSALLTRACES
         MENUITEM "&Browse folder",              ID_TRACES_BROWSEFOLDER
-        MENUITEM "&Strip Chrome symbols",       ID_TRACES_STRIPCHROMESYMBOLS
-        MENUITEM "&Identify Chrome processes",  ID_TRACES_IDENTIFYCHROMEPROCESSES
         MENUITEM "Trace path to &clipboard\tCtrl+C", ID_TRACES_TRACEPATHTOCLIPBOARD
+        POPUP "Trace compression"
+        BEGIN
+            MENUITEM "&ETW compress trace (Win 8.0+)", ID_TRACES_COMPRESSTRACE
+            MENUITEM "&Zip compress trace",         ID_TRACES_ZIPCOMPRESSTRACE
+            MENUITEM "ETW Compress all &traces (Win 8.0+)", ID_TRACES_COMPRESSTRACES
+            MENUITEM "Zip compress all traces",     ID_TRACES_ZIPCOMPRESSALLTRACES
+        END
+        POPUP "&Scripts"
+        BEGIN
+            MENUITEM "&Strip Chrome symbols",       ID_TRACES_STRIPCHROMESYMBOLS
+            MENUITEM "&Identify Chrome processes",  ID_TRACES_IDENTIFYCHROMEPROCESSES
+            MENUITEM "Create &flame graph",         ID_SCRIPTS_CREATEFLAMEGRAPH
+        END
     END
 END
 

UIforETW/UIforETWDlg.cpp

@@ -1614,6 +1614,7 @@ void CUIforETWDlg::OnContextMenu(CWnd* pWnd, CPoint point)
 				ID_TRACES_STRIPCHROMESYMBOLS,
 				ID_TRACES_IDENTIFYCHROMEPROCESSES,
 				ID_TRACES_TRACEPATHTOCLIPBOARD,
+				ID_SCRIPTS_CREATEFLAMEGRAPH,
 			};
 
 			for (const auto& id : disableList)
@@ -1677,6 +1678,9 @@ void CUIforETWDlg::OnContextMenu(CWnd* pWnd, CPoint point)
 				IdentifyChromeProcesses(tracePath);
 				UpdateNotesState();
 				break;
+			case ID_SCRIPTS_CREATEFLAMEGRAPH:
+				CreateFlameGraph(tracePath);
+				break;
 			case ID_TRACES_TRACEPATHTOCLIPBOARD:
 				// Comma delimited for easy pasting into DOS commands.
 				SetClipboardText(L"\"" + tracePath + L"\"");
@@ -1878,6 +1882,23 @@ void CUIforETWDlg::IdentifyChromeProcesses(const std::wstring& traceFilename)
 }
 
 
+void CUIforETWDlg::CreateFlameGraph(const std::wstring& traceFilename)
+{
+	outputPrintf(L"\nCreating CPU Usage (Sampled) flame graph of busiest process ((requires python, perl and flamegraph.pl)...\n");
+	std::wstring pythonPath = FindPython();
+	if (!pythonPath.empty())
+	{
+		ChildProcess child(pythonPath);
+		std::wstring args = L" -u \"" + GetExeDir() + L"xperf_to_collapsedstacks.py\" \"" + traceFilename + L"\"";
+		child.Run(bShowCommands_, GetFilePart(pythonPath) + args);
+	}
+	else
+	{
+		outputPrintf(L"Couldn't find python.\n");
+	}
+}
+
+
 void CUIforETWDlg::PreprocessTrace(const std::wstring& traceFilename)
 {
 	if (bChromeDeveloper_)

UIforETW/UIforETWDlg.h

@@ -219,6 +219,7 @@ class CUIforETWDlg : public CDialogEx
 	void StripChromeSymbols(const std::wstring& traceFilename);
 	void IdentifyChromeProcesses(const std::wstring& traceFilename);
 	void PreprocessTrace(const std::wstring& traceFilename);
+	void CreateFlameGraph(const std::wstring& traceFilename);
 	void LaunchTraceViewer(const std::wstring traceFilename, const std::wstring viewerPath);
 	void SaveNotesIfNeeded();
 	void ShutdownTasks();

bin/.gitignore

@@ -7,3 +7,6 @@ symsrv.dll
 # setting, even though it seems like it should.
 *.iobj
 *.ipdb
+
+# Exclude flamegraph.pl so that users can install it
+flamegraph.pl

bin/xperf_to_collapsedstacks.py

@@ -57,28 +57,38 @@
 import os
 
 # How many threads to create collapsed stacks for.
-numToShow = 5
+numToShow = 1
 
-if len(sys.argv) < 4:
-	print "Usage: %s trace.etl begin end"
+scriptPath = os.path.abspath(sys.argv[0])
+scriptDir = os.path.split(scriptPath)[0]
+flameGraphPath = os.path.join(scriptDir, "flamegraph.pl")
+if not os.path.exists(flameGraphPath):
+	print "Couldn't find %s. Download it from https://github.com/brendangregg/FlameGraph/blob/master/flamegraph.pl" % flameGraphPath
+
+if len(sys.argv) < 2:
+	print "Usage: %s trace.etl begin end" % sys.argv[0]
 	print "Begin and end specify the time range to be processed, in seconds."
 	sys.exit(0)
 
 etlFilename = sys.argv[1]
 textFilename = os.path.join(os.environ["temp"], os.path.split(etlFilename)[1])
-begin = int(float(sys.argv[2])*1e6)
-end = int(float(sys.argv[3])*1e6)
-textFilename = textFilename.replace(".etl", "_%d_%d.txt" % (begin, end))
+if len(sys.argv) >= 4:
+	begin = int(float(sys.argv[2])*1e6)
+	end = int(float(sys.argv[3])*1e6)
+	textFilename = textFilename.replace(".etl", "_%d_%d.txt" % (begin, end))
+	command = 'xperf -i "%s" -symbols -o "%s" -a dumper -stacktimeshifting -range %d %d' % (etlFilename, textFilename, begin, end)
+else:
+	textFilename = textFilename.replace(".etl", ".txt")
+	command = 'xperf -i "%s" -symbols -o "%s" -a dumper -stacktimeshifting' % (etlFilename, textFilename)
 
 # Optimization for when working on the script -- don't reprocess the xperf
 # trace if the parameters haven't changed.
-if True or not os.path.exists(textFilename):
-	command = 'xperf -i "%s" -symbols -o "%s" -a dumper -stacktimeshifting -range %d %d' % (etlFilename, textFilename, begin, end)
+if not os.path.exists(textFilename):
 	print command
 	for line in os.popen(command).readlines():
 		print line
 else:
-	print "Using existing intermediate file - '%s'." % textFilename
+	print "Using cached xperf results in '%s'." % textFilename
 
 # Create regular expressions for parsing the SampledProfile and Stack lines.
 # SampledProfile,  TimeStamp,     Process Name ( PID),   ThreadID,           PrgrmCtr, CPU, ThreadStartImage!Function,            Image!Function, Count, SampledProfile type
@@ -134,7 +144,10 @@
 				process = unresolvedSamples.pop(key)
 				# Create a fully specified process and thread name under which these
 				# samples will be stored.
-				processAndThread = "%s(%d)" % (process.replace(" ", "_"), ThreadID)
+				#processAndThread = "%s(%d)" % (process.replace(" ", "_"), ThreadID)
+				processAndThread = "%s_%d" % (process.replace(" ", "_"), ThreadID)
+				processAndThread = processAndThread.replace("(", "")
+				processAndThread = processAndThread.replace(")", "")
 				# Convert the array of stack lines to a single call stack string.
 				stackSummary = []
 				for entry in stack:
@@ -195,14 +208,32 @@
 
 print "Found %d samples from %d threads." % (totalSamples, len(samples))
 
+tempDir = os.environ["temp"]
+count = 0
 for numSamples, processAndThread in sortedThreads[:numToShow]:
 	threadSamples = samples[processAndThread]
-	outputName = "%s_collapse.txt" % processAndThread
-	print "Writing %d samples to %s" % (numSamples, outputName)
+	#outputName = "%s_collapse.txt" % processAndThread
+	outputName = os.path.join(tempDir, "collapsed_stacks_%d.txt" % count)
+	count += 1
+	print "Writing %d samples to temporary file %s" % (numSamples, outputName)
 	sortedStacks = []
 	for stack in threadSamples:
 		sortedStacks.append("%s %d\n" % (stack, threadSamples[stack]))
 	sortedStacks.sort()
 	out = open(outputName, "wt")
 	for stack in sortedStacks:
 		out.write(stack)
+
+
+	destPath = os.path.join(tempDir, "%s.svg" % processAndThread)
+	title = "CPU Usage flame graph of %s" % processAndThread
+	perlCommand = 'perl "%s" --title="%s" "%s" >"%s"' % (flameGraphPath, title, outputName, destPath)
+	#print perlCommand
+	for line in os.popen(perlCommand).readlines():
+		print line
+	resultSize = os.path.getsize(destPath)
+	if resultSize < 100: # Arbitrary sane minimum
+		print "Result size is %d bytes - is perl in your path?" % resultSize
+	else:
+		os.popen(destPath)
+		print "Results are in '%s' - they should be auto-opened in the default SVG viewer." % destPath