Add a couple of other intel_upload_blit_large variants for comparison.

benchmarks/intel_upload_blit_large.c

@@ -40,6 +40,8 @@
  * Additionally, those applications should be rendering at the screen refresh
  * rate, while this test has no limits, and so can get itself into the
  * working set larger than aperture size performance disaster.
+ *
+ * The current workload doing this path is pixmap upload for non-KMS.
  */
 
 #include <stdlib.h>