Building on https://mcmap.net/q/1339540/-monitoring-the-fps-of-a-direct-x-application:
I had more success not using the stopwatch as the event triggers seems to be asynchronous with the actual frames. I kept getting batches of 20-50 frames all at once, making the estimated FPS fluctuate between 50 and 250% of the actual value.
Instead i used TimeStampRelativeMSec
//handle event
m_EtwSession.Source.AllEvents += data =>
{
//filter out frame presentation events
if((int) data.ID == EventID_DxgiPresentStart && data.ProviderGuid == DXGI_provider)
{
int pid = data.ProcessID;
long t;
t = watch.ElapsedMilliseconds;
//if process is not yet in Dictionary, add it
if (!frames.ContainsKey(pid))
{
frames[pid] = new TimestampCollection();
string name = "";
var proc = Process.GetProcessById(pid);
if (proc != null)
{
using (proc)
{
name = proc.ProcessName;
}
}
else name = pid.ToString();
frames[pid].Name = name;
}
frames[pid].Add((long)data.TimeStampRelativeMSec);
}
};
property from the TraceEvent class, and calculate FPS by rounding the average time between an arbitrary number of past entries:
public double GetFrameTime(int count)
{
double returnValue = 0;
int listCount = timestamps.Count;
if(listCount > count)
{
for(int i = 1; i <= count; i++)
{
returnValue += timestamps[listCount - i] - timestamps[listCount - (i + 1)];
}
returnValue /= count;
}
return returnValue;
}
This method gave me far more accurate (Compared to, as available, in-game counters) of several different games i've tried.