Minimizing Overdraw
With the Intel® GPA Frame Analyzer, you can discover which ergs influence specific pixels to determine redundant or unimportant ergs. For example, lighting effects are typically produced by multiple rendering passes; use this option to determine whether you can reduce the number of iterations to produce a similar visual effect with higher performance.
To detect the sequence of ergs that led to the pixel’s final color value:
-
Select the render targets you want to inspect.
To inspect all render targets within the scene, check the Entire Frame check box in the Erg Table with Metric Data List. To inspect specific render targets, select at least one erg from each render target that you are interested in analyzing. -
Click a render target in the Render Target Thumbnail List to select it.
-
From the View drop-down list select Overdraw to see which pixels were written to the most times, per render target.
-
Check the Pixel History check box; this will change the cursor when the mouse is positioned over a render target.
-
Within the Render Target View, select a pixel of interest with the mouse. Darker pixel values mean that those pixels were written to fewer times than lighter pixel values. The Pixel History tab opens.
-
In the Pixel History Thumbnail List, analyze ergs that have affected the pixel. To select any single erg, right-click on any thumbnail in the Pixel History Thumbnail List and select the appropriate context menu option. To select all ergs in the list, click the Select Ergs button in the Pixel History tab.
-
Determine what can be done to get the same or similar results at less cost:
-
If you see that two ergs affecting the pixel are of the same type, you may improve performance by removing one of them. To see how much performance you can gain by removing the erg, select it, and check the Disable Erg(s) check box in the Experiments tab.
-
Detect ergs drawn by mistake. Remove such ergs by checking the Disable Erg(s) check box in the Experiments tab to see the potential performance improvement.
-
See Also