PGPROF shows Program Hot Spots

Another parallel profiler, called pgprof, comes with the PGHPF compiler. Compile and link the program for both function and line level performance data using the -Mprof=lines option. Running the program will generate a profile data file called pgprof.out. Examine the output using the graphical user interface (type "pgprof")

[dump of pgprof window]

Here we see it is important to parallelize the large loop in linkbr which calls rotmea which calls syslop.

(Back to main talk)