Add a detailed flag to the lab start_tracing.bat

When doing lab tracing it is useful to have two different tracing modes.
In one mode a very lightweight trace is recorded in order to get CPU
and GPU usage, and little else. This mode has minimal effect on
performance.

The second mode is a detailed mode that records call stacks on CPU
samples and context switches. This distorts results but gives an ETL
file that can be used to investigate regressions.

If 'detailed' is passed to start_tracing.bat it starts a detailed
trace. Stopping and summarizing the trace is unchanged. The trace
files created will be larger - perhaps 5-10x larger?

LabScripts/GeneralTracing/set_vars.bat

@@ -26,12 +26,6 @@ set DX_Flags=DX:0x2F
 
 :Windows8
 
-rem These are some typical flags used by UIforETW:
-rem Kernel flags: Latency+POWER+DISPATCHER+DISK_IO_INIT+FILE_IO+FILE_IO_INIT+VIRT_ALLOC+MEMINFO
-rem Stackwalk flags: Profile+CSwitch+ReadyThread+VirtualAlloc+VirtualFree
-rem Buffer size flags: -buffersize 1024 -minbuffers 1200 -maxbuffers 1200 
-rem User session flags: Microsoft-Windows-Win32k:0xfdffffffefffffff+Multi-MAIN+Multi-FrameRate+Multi-Input+Multi-Worker+Microsoft-Windows-Kernel-Memory:0xE0+Microsoft-Windows-Kernel-Power+chrome:0x8000000000000200
-
 rem %temp% should be a good location for temporary traces.
 rem Make sure this is a fast drive, preferably an SSD.
 set xperftemptracedir=%temp%
@@ -41,5 +35,14 @@ set userfile=%xperftemptracedir%\user.etl
 set SessionName=usersession
 set FileName=trace.etl
 set FileAndCompressFlags="%FileName%" -compress
+
+rem PROC_THREAD+LOADER are required in order to know what binaries are loaded
+rem and what threads are running.
+rem CSWITCH records context switch data so that precise CPU usage and context
+rem switch counts can be recorded. So, these three flags are our minimum set
+rem of kernel providers.
 set KernelProviders=PROC_THREAD+LOADER+CSWITCH
+
+rem These user providers record GPU usage, window-in-focus, eventemitter events
+rem (if emitted), and process working-set samples.
 set UserProviders=%DX_Flags%+Microsoft-Windows-Win32k:0xfdffffffefffffff+Multi-MAIN+Multi-FrameRate+Multi-Input+Multi-Worker+Microsoft-Windows-Kernel-Memory:0xE0

LabScripts/GeneralTracing/start_tracing.bat

@@ -30,12 +30,20 @@ rem running. Otherwise the start command will fail with incredibly cryptic
 rem errors. Ignore all warnings because in most cases they are expected.
 xperf -stop %SessionName% -stop 2>nul
 
-rem PROC_THREAD+LOADER is required in order to know what binaries are loaded
-rem what threads are running.
-rem CSWITCH records context switch data so that precise CPU usage and contxt
-rem switch counts can be recorded.
-
+if "%1" == "detailed" goto detailed
 xperf.exe -start %logger% -on %KernelProviders% -buffersize 1024 -minbuffers 60 -maxbuffers 60 -f "%kernelfile%" -start %SessionName% -on %UserProviders% -f "%userfile%
+goto skipdetailed
+:detailed
+rem These are some typical flags used by UIforETW to get more detailed traces,
+rem which we optionally enable for lab tracing.
+rem This extra information will distort timings. This is particularly true
+rem of the CSwitch and ReadyThread stackwalk flags.
+set KernelProviders=%KernelProviders%+CSWITCH+DISK_IO+HARD_FAULTS+DPC+INTERRUPT+PROFILE
+set KernelProviders=%KernelProviders%+POWER+DISPATCHER+DISK_IO_INIT+FILE_IO+FILE_IO_INIT+VIRT_ALLOC+MEMINFO
+set StackWalkFlags=Profile+CSwitch+ReadyThread
+xperf.exe -start %logger% -on %KernelProviders% -stackwalk %StackWalkFlags% -buffersize 1024 -minbuffers 160 -maxbuffers 160 -f "%kernelfile%" -start %SessionName% -on %UserProviders% -f "%userfile%
+:skipdetailed
+
 rem You have to invoke -capturestate because otherwise the ETW gnomes will not
 rem reliably record the user-mode data to your trace. Do not anger the gnomes.
 xperf.exe -capturestate %SessionName% %UserProviders%