Intel® Performance Tuning Utility

cacheline contention issue

Inviato da lovejc il

Hi all


In cacheline view pane,


1) Could you tell me definition of cacheline contention in PTU?
for example, If thread-1 access(read? or write?) a cacheline and thread-2 also access the cacheline,
a counter for contention increase a count. right?


Is it equivalent thread contention?


2) I've read your published paper "Parallelization Made Easier with Intel Performance-Tuning Utility"
I don't know how to collect contention. What CPU event relate to collection of cacheline contention?


Thanks in advance.

Can I use vtssrun command in a paused mode?

Inviato da lovejc il

Hi all,


I'd like to get collecting data for specific code regions.


So I did execute the collector(vtssrun) in a paused mode, but it didn't work.


-------------------------------------------------------------------------


$ ./vtssrun ./exp -p -- ./a.out


Perform Statistical Call Graph data collection on the specified application


-------------------------------------------------------------------------


What is the specified application?

PTU Basic Data Profiling - tb5 file doesn't contain PEBS records

Inviato da yrr il

I'm trying out the Profile As -> Basic Data Profiling on a RHEL 4u6 x86_64 system
Processors: 2 x Xeon 2.80GHz 800MHz FSB (HT enabled, 4 threads) - Nocona E0, 64-bit, 90nm, L2: 1MB

It seems to run correctly (the 1st- and 2nd-level cache load misses and loads retired events are sampled okay), but there are several "Preprocessing failed: The passed tb5 file doesn't contain PEBS records" messages in the console, and finally, "No data profiling information to display".

PTU -- Heap profiler in Windows?

Inviato da xiaobin_w il

Hi,

I downloaded PTU today and I am very impressed!

I have a question though. It is a bit silly... I was not able to find the heap profiling command (vthprun.exe) in the bin directory. I looked at both 32bit version and 64bit version for windows and could not find it in the packages. It is in the Linux packages though, but my app doesn't run on Linux. From the user manual, the heap profiler looks quite cool, so I'd love to try it out.

Thanks in adanvance,
Xiaobin

Event sampling fails on Linux?

On RHEL4 from the 2.0 GUI, "Basic Statistical Callgraph" works fine, but provides limited diagnostic information. But "Basic Sampling" and "Basic Data Profiling" fail with the following messages in the Console:

Duration (sec): 0

Failed to perform sampling collection!


Custom configurations with Event-based sampling fail in the same manner.

Please help a new user get started. I assume event-based sampling is supported on Linux...?

Thanks,
Ron

PTU - Conversion failed

I am just starting with this tool. I ran a simple vtssrun experiment, and got the .vtss files.


I cannot view any of it, though. No matter what options, I always get the "Conversion failed" error.



> vtssview d: empprof
un2 --flat-profile


Converting trace files failed.
Conversion failed: Error calling converter


Arethere any verbose/diagnotic flags?


Sampling runs have no problems. This is on Windows XP 64.


PTU - Use with MPI?

Just wanted to see if anyone had tried using PTU with an MPI application? I'm particularly interested in whether I can get MPI to launch PTU on each node of a cluster and then get PTU to execute my application. I realize I can login to each host for a run and profile what's going on with the host, but I'm curious about getting the profile done in a complete "batch" mode in order to allow profile jobs to be submitted to a cluster being managed by a queueing system.


I will try this, but just wanted to see if I anyone might already have some experience with this.

PTU - Initial Impressions

I just started using PTU for profiling of a commercial simulation (finite element) code. I am testing on 64-bit Windows (XP) and Linux 64-bit x86.We have struggled with VTune for a number of years, and so far PTU is making a very good impression.


Overall I think this is a very good effort, and I'm very happy to see Intel making this tool available.


Things that are particularly appealing.


(1) Sampling-base call graph profiling. The efficiency is very nice.

Pagine

Iscriversi a Intel® Performance Tuning Utility