dumppm
dumppm (DUMP Performance Monitor results from samppm)
====== ===== =========== ======= ======= ==== =======
dumppm is used to dump the binary file produced by a samppm run
into a human/script readable columnar format. To see how to use it:
% dumppm -h
usage: dumppm [-c] [-d] [-h] [-l n1,n2,...] [<] infile [>] outfile
-c only list events
-d print event differences and relative times
-h print help
-l print only events n1, n2, ... (event 0 is time)
Note that dumppm can operate as a filter, e.g.:
% cat file.prog.1234 | dumppm -d | more
% dumppm -d < file.prog.1234 | more
% dumppm -l 2,3 file.prog.1234 > results
% dumppm -d file.prog.1234 results
The `-c' flag is used to quickly tell one what is contained in the file,
and shows the correspondence between event and output column. For example:
% samppm -e CPU_CYCLES -e LOADS_RETIRED -r 2 -o dat myprog
<normal program output here>
% dumppm -c dat.myprog.3214
dumppm: event 1 is: CPU_CYCLES
dumppm: event 2 is: LOADS_RETIRED
%
If the `-c' is dropped, the output will consist of 3 columns:
time of day, sampled CPU_CYCLES value, and sampled LOADS_RETIRED value, e.g.:
1042470662.010721 2059042 420371
1042470662.012656 3454206 711769
1042470662.014595 4866156 1006722
...
If the `-d' flag is added, time becomes relative, and differences between
counts are output instead:
0.000000 2059042 420371
0.001935 1395164 291398
0.003874 1411950 294953
...
If a different order of, and possibly a different number of columns are
needed, the `-l' flag can be used to specify which of the original
columns to output, in which order. E.g. to output the change in
CPU_CYCLES followed by relative time, use "-d -l 1,0":
2059042 0.000000
1395164 0.001935
1411950 0.003874
...
Further filtering and plotting of the data is left to the user.