- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Thanks for this valuable report.
Did you work on advanced-hotspots data collector? To simplify the problem, try command line:
amplxe-cl -collect advanced-hotspots -no-auto-finalize -duration 60 -target-pid PID
Then, you can try:
amplxe-cl -finalize -r r00?ah ; have you seen similar problem in GUI? What is error message?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I wonder that the tool will not crash when resolving symbol info:-)
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
You must try to keep the amount of data collection within limits. The sampling rate can be adjusted by command line sample-after value or by the estimated time in the Advanced section of the GUI setup. The estimated time is self-explanatory for the case where there are perhaps up to 4 total threads running on the platform. Beyond that, you must reduce estimated time to a fraction of real time. The .tb6 file size limits were increased several updates back, but even 300MB may be excessive.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
It seems that faulting IP is located at this code point rip 0x7fb483a16448 0x7fb483a16448 <dicerresolver_2_2::internal::ism_module_bank::get_symbol(dicerresolver_2_2::internal::segment_rva const&)+264>.
Now because this function belongs to VTune library so the source code and signature will not be available.By looking at function description and if the error is signaled as a segfault I can theorize that probably the value passed by reference could point to non-accessible memory.It can be this value segment_rva const&.
As I do not have an experience with GDB so I would like to ask you to resolve with the help of GDB address which containes rsi register(rsi 0x7fb55deb53b0).
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Peter Wang (Intel) wrote:
Did you work on advanced-hotspots data collector? To simplify the problem, try command line:
amplxe-cl -collect advanced-hotspots -no-auto-finalize -duration 60 -target-pid PID
Then, you can try:
amplxe-cl -finalize -r r00?ah ; have you seen similar problem in GUI? What is error message?
Using the GUI, I've tried the Basic Hotspots, Advanced Hotspots, and General Exploration - Nehalem / Westmere analysis types. The Basic Hotspots analysis is the least flaky, but still fails for the same reason the majority of the time.
I followed your suggestions using the command line interface (amplxe-cl), and everything appears stable. The collect and finalize steps both succeeded. Next I changed into my the project folder for my application, and I reran "amplxe-cl -finalize" against each of them. The command consistently succeeded. :D
Going back to the original report about the GUI, the UI never displayed an explicit error message. It'd only trigger a SIGSEGV, which would be handled by the UI popping up a "would you like to submit this crash report to Intel?" dialog.
So, as of writing, I now have a workaround by dealing with sample collection via command line, and I'll open the results in the GUI for review. Especially considering the price attached to the package, I'd still prefer that the GUI not explode, however. ;)
Thanks for the assist!
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Tim Prince wrote:
You must try to keep the amount of data collection within limits. The sampling rate can be adjusted by command line sample-after value or by the estimated time in the Advanced section of the GUI setup. The estimated time is self-explanatory for the case where there are perhaps up to 4 total threads running on the platform. Beyond that, you must reduce estimated time to a fraction of real time. The .tb6 file size limits were increased several updates back, but even 300MB may be excessive.
That's a fair observation, however I should note that I'm sampling an application for less than a minute, and my sampled data (r?????/data.0) maxes out at around 25M for the longest sample duration. I don't think I was collecting excessive amounts of information. :) That said, I'd find it kinda ironic that a developer tool geared toward helping people write robust code crashes 'cuz it may have been fed too much data. (And it's not like some 3rd party program generated the input data!)
iliyapolak wrote:
As I do not have an experience with GDB so I would like to ask you to resolve with the help of GDB address which containes rsi register(rsi 0x7fb55deb53b0).
gdb was unable to resolve the address to anything. :\
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
>>>gdb was unable to resolve the address to anything. :\>>>
Maybe this is the culprit?Btw can explain what is the error code?Is that segfault?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
>>>Btw can explain what is the error code?Is that segfault?>>>
Please disregard this quoted sentence,because I have found the error description in your first post.
Regarding the mentioned address which holds rsi register I meant possibility to resolve the address to some memory region like process heap.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello Ryan!
When GUI handled SIGSEGV, were you able to submit crash report? That could help us to find the reason of GUI problem.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Vitaly Slobodskoy (Intel) wrote:
Hello Ryan!
When GUI handled SIGSEGV, were you able to submit crash report? That could help us to find the reason of GUI problem.
I avoided sending the crash reports, because I was concerned that doing so might leak confidential information. I'm working on another task now, but upon switching back I'll give this a shot. Thanks!
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page