|Loop profilers on various platforms email@example.com (Rohini) (2005-05-05)|
|Re: Loop profilers on various platforms firstname.lastname@example.org (glen herrmannsfeldt) (2005-05-05)|
|Re: Loop profilers on various platforms email@example.com (Robert H) (2005-05-13)|
|From:||glen herrmannsfeldt <firstname.lastname@example.org>|
|Date:||5 May 2005 16:35:01 -0400|
|Posted-Date:||05 May 2005 16:35:01 EDT|
> I wanted to know which all platforms have a loop profiling tool.
> Any pointers on this will hellp me a lot.
What is loop profiling?
The IBM Fortran H compiler has a number of loop optimization, such as
moving invariant computations out of a loop, strength reduction, and
probably more. To do that, it has to find loops, which it does even
if they are made up of IF and GOTO, though DO loops are easier for
people to find.
If loop optimization is on (OPT=2), it had an option to print out a
listing indented based on the nesting level. Is that loop profiling?
Another that I knew from about the same time frame was called FETE
(Fortran Execution Time Estimator). It would take an input Fortran
program, add extra statements to it that would then be fed to the
compiler. The listing of the modified program would normally not be
printed. (I never saw one.) The modified program was them compiled
and executed. At the normal end of the original program, it would
then print out a source listing where each statement included the
number of times it was execute, an approximate time for executing that
statement, and one other, I believe the number of times a logical IF
expression was .TRUE.
FETE, then, could be used to find out where the program is spending
most of its time, usually due to loops, which could be called loop
profiling. I tried to find FETE a few years ago, but it might be that
it doesn't exist anymore.
Return to the
Search the comp.compilers archives again.