|optimization using profiling email@example.com (1998-11-12)|
|Re: optimization using profiling firstname.lastname@example.org (1998-11-15)|
|Re: optimization using profiling email@example.com (John Williams) (1998-11-15)|
|Re: optimization using profiling firstname.lastname@example.org (Joachim Durchholz) (1998-11-15)|
|From:||Joachim Durchholz <email@example.com>|
|Date:||15 Nov 1998 13:36:01 -0500|
> How? What can be done after profiling that can't be done without it ?
For example, it can tell the optimizer whether a loop is usually
executed not at all, once, or a thousand times. This has influcence on
the decision whether to move code out of the loop body (which is
actually contraproductive if the loop is usually not executed at all).
Test coverage data is less than useful for this (the idea test coverage
will execute each path exactly once). But you have to carefully select
*typical* input to get useful profiling information, else the profiler
output will indeed misguide the optimizer.
Return to the
Search the comp.compilers archives again.