- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
I'm carrying out some research on parallel libraries and would like some feedback if possible on the characteristics of the profile in use in this environment, is it event based or sampling based. With regards to timers, how intrusive are they? Would they be considered coarse or fine? Any info would be much appreciated.
Thanks
Dave
I'm carrying out some research on parallel libraries and would like some feedback if possible on the characteristics of the profile in use in this environment, is it event based or sampling based. With regards to timers, how intrusive are they? Would they be considered coarse or fine? Any info would be much appreciated.
Thanks
Dave
Link Copied
2 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello,
Some basic information I can provide is:
IntelParallel Amplifier uses a time based statistical sampling.It isdesigned to be lightweight and unobtrusive. With respect to coarse or fine grained, could you clarify how exactly you are defining these terms?
I see that you are looking to do some research on parallel libraries, will you be using profiling tools on these libraries?
Does this information help? Is there any thing else I can provide?
Thanks.
Some basic information I can provide is:
IntelParallel Amplifier uses a time based statistical sampling.It isdesigned to be lightweight and unobtrusive. With respect to coarse or fine grained, could you clarify how exactly you are defining these terms?
I see that you are looking to do some research on parallel libraries, will you be using profiling tools on these libraries?
Does this information help? Is there any thing else I can provide?
Thanks.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
The easiest way to think about the parallel amplifier (the profiler tool in parallel studio) is that it records the stack each time interval. This is a statistical approach which does not need any instrumentation of the binaries and therefor it is a very lightweight. You application does not slow down. This comes with a cost however. It does not show for example the numbers of calls to a particular function because. It can only give you an fairly accurate approximation of the time spend in each function.

Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page