Re: IVF 10.0 Win/VS 2005 Profiling
- From: Tim Prince <timothyprince@xxxxxxxxxxxxx>
- Date: Mon, 26 Nov 2007 19:06:15 -0800
John S wrote:
As you've seen, the purpose of these profiling options is to get data onI can find all sorts of information on using Profile Guided Optimization
using /Qprof-gen and then /Qprof-use. But this is for optimizing during
compile. What I need is to see line times for an older version of a
project versus a newer version.
Intel's documentation does not seem to be very lucid on this.
Some progress: Need to use /Qprof-genx; compile with this flag; Run code;
Use profmerge.exe to turn *.dyn into *.dpi file; Use the codecov.exe to
create coverage information in html format.
But would like to get actual time spent by line or subroutine call. All I
can find in the codecov.exe options are a -counts option which report counts
by line. Need something like this, but with actual time.
relative frequency of branches etc. Running in an un-optimized mode, it
is not useful for actual timings.
As Steve already mentioned, Intel's primary offering for time and event
profiling is VTune; secondary offering is PTU (see
softwarecommunity.intel.com/forums). Other possibilities would be
oprofile based profilers, or a compiler which supports gprof (as Intel
linux compilers do),...
.
- Follow-Ups:
- Re: IVF 10.0 Win/VS 2005 Profiling
- From: Steve Lionel
- Re: IVF 10.0 Win/VS 2005 Profiling
- References:
- IVF 10.0 Win/VS 2005 Profiling
- From: John S
- Re: IVF 10.0 Win/VS 2005 Profiling
- From: John S
- IVF 10.0 Win/VS 2005 Profiling
- Prev by Date: Re: Execution times running multiple instances of an application
- Next by Date: Re: Execution times running multiple instances of an application
- Previous by thread: Re: IVF 10.0 Win/VS 2005 Profiling
- Next by thread: Re: IVF 10.0 Win/VS 2005 Profiling
- Index(es):
Relevant Pages
|