
98 Cray T3E User’s Guide
9.4.2 The appview command
In addition to MPP Apprentice, the appview command that a quick sum-
mary of the profiling data. Its output is similar to the output from the
conventional Unix profiler prof.Theappview command was developed
at CSC and it relies on a few scripts that extract and sort information
from the textual report produced by the command apprentice -r.
The following example illustrates the usage. The command line is
appview app.rif | more
The default RIF file name is app.rif and can be omitted from the com-
mand line. The output looks like this:
###
# # ##### ##### # # # ###### # #
####### # ## ##
# # # # # # # # # ##### # #
####### ##### ##### # # # # # ## #
# # # # # # # # ## ##
# # # # # # ###### # #
Fri Feb 21 13:32:07 EET 1997
Total time 24.273 seconds.
No. of routines 8
Routine Exclusive (%) Inclusive in_Called Parallel
======= ========= === ========= ========= ========
COLL2 14.968 ( 61.67) 14.968 0.000 14.946
_HSIN 8.216 ( 33.85) <not instrumented>
INIT 1.062 ( 4.38) 9.277 8.216 1.062
_FWF 0.022 ( 0.09) <not instrumented>
f$init 0.005 ( 0.02) <not instrumented>
COLL2_TEST 0.000 ( 0.00) 24.250 24.250 0.000
BARRIER 0.000 ( 0.00) <not instrumented>
$END 0.000 ( 0.00) <not instrumented>
The columns Exclusive and Inclusive show the execution time with-
out and with the time spent in called subprograms, respectively. The
column in_Called show the time for called subprograms.
9.4.3 PAT
The Performance Analysis Tool (PAT) provides a low-overhead method
for profiling programs and obtaining timing and performance counter
information. PAT can also be used for determining load balance across
processing elements, generating and viewing trace files, performing
event traces, etc. These advanced features are not available in the MPP
Apprentice. On the other hand, PAT lacks the graphical interface of the
Apprentice. PAT is used in evaluating the performance and scaling new
T3E projects at CSC.
PAT periodically samples the program counter to generate an execution-
time profile and uses the processor performance counters to gather
Kommentare zu diesen Handbüchern