- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi all,
I use Amplifier XE to profile a larger code project. I started with a hotspot analysis and a general exploration for Nehalem CPUs. What puzzles me, both experiments show resuts/hotspots (also bad CPI rates) in unexpected/weird code lines:
1) I see a hotspot (i.e. a large portion of CPU time spent) in a line, where a boolean variable is declared
bool myBool = FALSE;
2) I see another hotspot on a line that only contains the opening { of a function...
I don't really know what to optimize in these lines ;)
Can anybody explain that or point me to a resource where I can do further reading? I checked the forum but didn't find anything... and it's hard to know what to search for.
Thanks a lot!
- tim
I use Amplifier XE to profile a larger code project. I started with a hotspot analysis and a general exploration for Nehalem CPUs. What puzzles me, both experiments show resuts/hotspots (also bad CPI rates) in unexpected/weird code lines:
1) I see a hotspot (i.e. a large portion of CPU time spent) in a line, where a boolean variable is declared
bool myBool = FALSE;
2) I see another hotspot on a line that only contains the opening { of a function...
I don't really know what to optimize in these lines ;)
Can anybody explain that or point me to a resource where I can do further reading? I checked the forum but didn't find anything... and it's hard to know what to search for.
Thanks a lot!
- tim
Link Copied
4 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Tim,
This may occur due to "event skid" inhardware-based collections:
http://software.intel.com/sites/products/documentation/hpc/amplifierxe/en-us/win/ug_docs/olh/common/event_skid.html
In this case you can look for real hotspot at several lines near the "weird" place.
For morecertain line informationtake a look at precise events:
http://software.intel.com/sites/products/documentation/hpc/amplifierxe/en-us/win/ug_docs/olh/common/precise_events.html
Regards,
Kirill
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Kirill,
you mention "hardware-based collection". I am wondering, whether the same event skid occurs with the standard hotspot analysis which is - as I understand it - not using hardware performance counters.
I havent't been able to collect precise events yet, but I am working on it.
Best,
- tim
you mention "hardware-based collection". I am wondering, whether the same event skid occurs with the standard hotspot analysis which is - as I understand it - not using hardware performance counters.
I havent't been able to collect precise events yet, but I am working on it.
Best,
- tim
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Yes, standard Hotspots is user-more sampling analysis, as well as Locks and Waits and concurrency. Only hardware-based collections, also called EBS (Event Based Sampling) use events from Performance Monitoring Unit of CPU. So there is no event skid in Hotspots analysis (but event skid may occur in Lightweight Hotspots, that is EBS).
Regards,
Kirill
Regards,
Kirill
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
If you compile with in-lining (even with normal defaults), association of hot spots with source code may be problematic.
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