Quantcast
Channel: Intel® VTune™ Profiler (Intel® VTune™ Amplifier)
Viewing all 1347 articles
Browse latest View live

First questions regarding low cpu time and code execution in timeline

$
0
0

I can see a low cpu time in some regions of the timeline [vtune cpu time low image][1].

I know how to select the region and do "filter in by selection", so the "function / call stack" lines above are updated to the region (and only the first "run_params" and "process_params") are shown.

The code is simple: run_params has a loop of 999999999 iterations that calls process_params with different params every 2 iterations. Small part:

 

   void
    run_params(int times){
      // ...

      for (size_t i=0; i<times; i++){
        if (i % 2 == 0){
          process_params(&integer1, &integer2, &fp_single1, &fp_single2, &fp_double1, &fp_double2, &ptr_a1, &ptr_a2, &ptr_b1, &ptr_b2, i % 3);
        }else{
          process_params(&integer2, &integer1, &fp_single2, &fp_single1, &fp_double2, &fp_double1, &ptr_a2, &ptr_a1, &ptr_b2, &ptr_b1, i % 3);
        }
      }

      // ...
    }

Questions:

 1. How can I see why the cpu time is so low in those regions?

 2. How can I see which source code lines are executed in a specific time (or region in timeline)?

 3. I have only one function "process_params" (although it is called from two different source code lines inside the "run_params" func). Why I get two different functions/call stack called "process_params"?

 4. I have a loop of 999999999 iterations inside the run_params function, that calls process_params that number of times. Why I get weird iteration counts? (never seen the 999999999 number in the VTune). What should I do to see the number of times process_params is called? Also, it is shown in a column of the run_params function (why?), but no in process_params.[iterations image][2]

  [1]: vtune cpu time low image


  [2]: iterations image


Function names not displayed in vtune result

$
0
0

Hello,

I am using vtune from Parallel studio 2017 update 4, with visuals studio 2017 under Windows 7. When I am doing a basic hotspot analysis, the result do not show any function names, see below

Top Hotspots
Function Module CPU Time
func@0x100ee854 Deltares.Wanda.dll 0.706s
func@0x100eeaf0 Deltares.Wanda.dll 0.676s
func@0x100ef010 Deltares.Wanda.dll 0.369s
func@0x100f0550 Deltares.Wanda.dll 0.289s
func@0x100ef720 Deltares.Wanda.dll 0.269s
[Others]  13.326s

I did not get an error that the PDB of Deltares.Wanda.dll could not be found, I have checked and the PDB file is available and being build. Debugging symbols from several standard windows dll could not be found, but this is not a problem.

How can I get the function names displayed?

Hope someone can give me help

Many thanks

Sam

Thread Topic: 

Help Me

VTune won't work on host when KVM guest is started

$
0
0

I am running VTune Amplifier XE 2017 Update 1 on a Broadwell system (Intel(R) Xeon(R) CPU E5-2698 v4).

I am able to collect hardware counters using:

amplxe-cl -collect-with runsa -knob event-config=$(cat counters.csv) -data-limit=2500 -duration=420 -target-duration-type=medium -r $outdir

where counters.csv is a list of custom hardware counters.

However, when I start a KVM guest (using virsh start <domain>), the same command line stops working. Specifically, 

amplxe-cl -collect-with runsa -knob event-config=?

lists no events at all.

Once I shutdown the guest with "virsh destroy" vtune begins to work normally, as before.

Please note that I am NOT trying to run vtune inside the virtual machine - I am still running it on the host. My intention is to collect system wide hardware counters when workloads are running inside the guest.

Could someone please suggest a solution to this issue?

Thanks

Vidya

Zone: 

Thread Topic: 

Help Me

vtune - unexpected AUX VEC type 26

$
0
0

when i'm tring to colect data with vtune

amplxe: Error: [Instrumentation Engine]: Source/pin/injector_nonmac/auxvector.cpp: CopyAux: 291: unexpected AUX VEC type 26

os: arch linux

kernel vs: 4.11.3-1

vtune_amplifier_xe_2017

 

Thread Topic: 

Help Me

BSOD when using Advanced hotspots, VTune Amplifier 2017 Update 3

$
0
0

Hello. I'm profiling a c++ desktop application.

After updating VTune from 2016 U4 to 2017 U3 and switching to Visual Studio 2017 every time I launch Advanced Hotspots Analysis after 4-5 seconds I get OS freeze and BSOD 0x101 CLOCK_WATCHDOG_TIMEOUT.  I've tried multiple combinations of settings (basic, with call stacks, with call counts, different sampling intervals) but all cause BSOD.

Basic Hotspots working fine.

My OS is Windows 7 x64 with all updates and Visual Studio 2017 Community. CPU is Intel Core i5 3570K

0: kd> !analyze -v
*******************************************************************************
*                                                                             *
*                        Bugcheck Analysis                                    *
*                                                                             *
*******************************************************************************

CLOCK_WATCHDOG_TIMEOUT (101)
An expected clock interrupt was not received on a secondary processor in an
MP system within the allocated interval. This indicates that the specified
processor is hung and not processing interrupts.
Arguments:
Arg1: 0000000000000031, Clock interrupt time out interval in nominal clock ticks.
Arg2: 0000000000000000, 0.
Arg3: fffff88003565180, The PRCB address of the hung processor.
Arg4: 0000000000000002, 0.

Debugging Details:
------------------


BUGCHECK_STR:  CLOCK_WATCHDOG_TIMEOUT_4_PROC

CUSTOMER_CRASH_COUNT:  1

DEFAULT_BUCKET_ID:  WIN7_DRIVER_FAULT

PROCESS_NAME:  svchost.exe

CURRENT_IRQL:  d

ANALYSIS_VERSION: 6.3.9600.16384 (debuggers(dbg).130821-1623) amd64fre

STACK_TEXT:
fffff880`0b30a048 fffff800`030d9bf9 : 00000000`00000101 00000000`00000031 00000000`00000000 fffff880`03565180 : nt!KeBugCheckEx
fffff880`0b30a050 fffff800`03089d27 : 00000000`00000000 fffff800`00000002 00000000`00002710 fffff8a0`273a6ddc : nt! ?? ::FNODOBFM::`string'+0x4e3e
fffff880`0b30a0e0 fffff800`03600895 : fffff800`03626460 fffff880`0b30a290 fffff800`03626460 00000000`00000000 : nt!KeUpdateSystemTime+0x377
fffff880`0b30a1e0 fffff800`0307c6d3 : 00000000`69bf6f1a fffff800`031fee80 fffff800`031fee80 00000000`00000000 : hal!HalpHpetClockInterrupt+0x8d
fffff880`0b30a210 fffff800`03085020 : fffff800`031fee80 fffff8a0`00000001 00000000`00000000 fffff880`0b30a4a8 : nt!KiInterruptDispatchNoLock+0x163
fffff880`0b30a3a0 fffff800`0309fe71 : 00000000`00000000 00000000`00000008 00000000`00000001 fffff800`00000001 : nt!KeFlushMultipleRangeTb+0x250
fffff880`0b30a470 fffff800`030a28e8 : 00000000`00000008 fffff880`0b30a5c0 fffff8a0`1f48b000 00000000`00000001 : nt!MiFlushTbAsNeeded+0x1d1
fffff880`0b30a580 fffff800`031b4f86 : 00000000`00008000 fffffa80`0c6d3000 00000000`00000009 00000000`00000435 : nt!MiAllocatePagedPoolPages+0x4cc
fffff880`0b30a6a0 fffff800`030a05f0 : 00000000`00008000 fffffa80`0c6d3000 00000000`00000009 20206553`03075d6d : nt!MiAllocatePoolPages+0x906
fffff880`0b30a7e0 fffff800`031b844e : 00000000`00000000 ffffffff`11bfd700 00000000`00000000 00000000`00008000 : nt!ExpAllocateBigPool+0xb0
fffff880`0b30a8d0 fffff800`03093585 : 00000000`00000000 00000000`00000009 fffff8a0`015b5060 fffff800`0336eedf : nt!ExAllocatePoolWithTag+0x82e
fffff880`0b30a9c0 fffff800`032e883e : 00000000`00000000 00000000`00008000 00000000`00000000 fffff8a0`22fb02e0 : nt!ExAllocatePoolWithQuotaTag+0x55
fffff880`0b30aa10 fffff800`03344d30 : fffff8a0`22fb02e0 fffff800`00008000 fffff880`0b30ab01 fffff800`03554de0 : nt!PiControlGetInterfaceDeviceList+0x92
fffff880`0b30aa90 fffff800`0307f413 : fffffa80`12a2a060 00000000`01c3e840 fffff880`0b30ab60 00000000`01c3e8c8 : nt!NtPlugPlayControl+0x100
fffff880`0b30aae0 00000000`7718cb8a : 00000000`00000000 00000000`00000000 00000000`00000000 00000000`00000000 : nt!KiSystemServiceCopyEnd+0x13
00000000`01c3e808 00000000`00000000 : 00000000`00000000 00000000`00000000 00000000`00000000 00000000`00000000 : 0x7718cb8a


STACK_COMMAND:  kb

SYMBOL_NAME:  ANALYSIS_INCONCLUSIVE

FOLLOWUP_NAME:  MachineOwner

MODULE_NAME: Unknown_Module

IMAGE_NAME:  Unknown_Image

DEBUG_FLR_IMAGE_TIMESTAMP:  0

IMAGE_VERSION:

FAILURE_BUCKET_ID:  X64_CLOCK_WATCHDOG_TIMEOUT_4_PROC_ANALYSIS_INCONCLUSIVE

BUCKET_ID:  X64_CLOCK_WATCHDOG_TIMEOUT_4_PROC_ANALYSIS_INCONCLUSIVE

ANALYSIS_SOURCE:  KM

FAILURE_ID_HASH_STRING:  km:x64_clock_watchdog_timeout_4_proc_analysis_inconclusive

FAILURE_ID_HASH:  {d61e255a-5a6a-8ac9-10f0-973c0c42bda6}

Followup: MachineOwner
---------

 

Zone: 

Thread Topic: 

Bug Report

Collection Failed when running XE2017 for C++ hotspots analysis

$
0
0

Failed to run the target in both GUI and command line.

Errors I got:

amplxe: Warning: Cannot stop posix timer: __NR_timer_settime() system call returned -1.
amplxe: Warning: Cannot stop posix timer: __NR_timer_settime() system call returned -1.
amplxe: Error: [Instrumentation Engine]: Source/pin/elfio/img_elf.cpp: ProcessSectionHeaders: 601: assertion failed: SEC_vaddr_i(sec) &gt;= IMG_seg_text_vaddr_i(img) &SEC_vaddr_i(sec) &lt; IMG_seg_data_vaddr_i(img) 
amplxe: Collection failed.
amplxe: Internal Error

My program is a C++ QT Application running on Linux. 

Can anyone help me. Thank you.

Cannot enable advanced capabilities for Hardware Event-based Sampling: problem with the driver (vtss/vtspp)

$
0
0

I'm trying to use vTune Amplifier XE 2017 Update 3 on RHEL 7.1 to collect advanced-hotspots with stack-and-callcount.

The installation has been successful, and I've followed all the troubleshooting steps in "Building and Managing the Sampling Driver". 

Here's the command I try to run:

-bash-4.2$ cat /etc/redhat-release
Red Hat Enterprise Linux Server release 7.1 (Maipo)
-bash-4.2$ amplxe-cl -collect advanced-hotspots -knob collection-detail=stack-and-callcount -analyze-system -d 10
amplxe: Error: Cannot enable advanced capabilities for Hardware Event-based Sampling: problem with the driver (vtss/vtsspp). Check that the driver is running and the driver group is in the current user group list. See "Building and Managing the Sampling Driver" help topic for further details.
amplxe: Warning: Stack size option will be changed to 1024 during data collection. If you need a different size please specify it explicity in the analysis type.

When following the troubleshooting advice, I get

-bash-4.2$ /opt/intel/vtune_amplifier_xe_2017/sepdk/src/insmod-sep -q
ACPI DIMM information is not available on this machine: [Errno 13] Permission denied: '/sys/firmware/acpi/tables/SLIT'
/opt/intel/vtune_amplifier_xe_2017/sepdk/src/insmod-sep: line 180: /opt/intel/vtune_amplifier_xe_2017/sepdk/src/dmi_config_log.txt: Permission denied
ACPI information for proximity domain is not available on this machine: [Errno 13] Permission denied: '/sys/firmware/acpi/tables/SRAT'
/opt/intel/vtune_amplifier_xe_2017/sepdk/src/insmod-sep: line 189: /opt/intel/vtune_amplifier_xe_2017/sepdk/src/dmi_config_log.txt: Permission denied
pax driver is loaded and owned by group "users" with file permissions "666".
socperf2_0 driver is loaded and owned by group "users" with file permissions "666".
sep4_1 driver is loaded and owned by group "users" with file permissions "666".
vtsspp driver is loaded and owned by group "users" with file permissions "666".
-bash-4.2$

I've checked the config settings for the kernel build, and all the settings mentioned in the troubleshooting steps are correct.

I have no problem using the normal perf event sampling methods, e.g.

-bash-4.2$ amplxe-cl -collect advanced-hotspots -knob collection-detail=stack-sampling -analyze-system -d 10
amplxe: Warning: Stack size option will be changed to 1024 during data collection. If you need a different size please specify it explicity in the analysis type.
amplxe: Warning: To profile kernel modules during the session, make sure they are available in the /lib/modules/kernel_version/ location.
amplxe: Collection started. To stop the collection, either press CTRL-C or enter from another console window: amplxe-cl -r /home/evant/r003ah -command stop.

...

Is there some more information I can get amplxe-cl to print regarding the specific problems with the driver?

Thread Topic: 

Bug Report

Running executable under VTune causes hang

$
0
0

Hi,

I am running an executable under VTune (2017 Update 2 (build 499904)). When I run the executable outside of Vtune, it exits normally and returns to the prompt in Linux (uname -a output: Linux sklad2.cluster.local 3.10.0-514.10.2.el7.x86_64 #1 SMP Fri Mar 3 00:04:05 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux). However in vtune it fails to return to prompt or in the GUI it fails to stop profiling. If I force-kill the application being profiled, no data has been recorded. The application continues to run and subsequently consumes more CPU resources than is expected. What is causing this behavior? What should I look at?

Normal execution run

./ITK/build-icc-004-avx2-ipo/bin/ITKLevelSetsv4TestDriver itkMultiLevelSetMalcolmImageSubset2DTest
Domain map computed
Level set container created
LevelSet 0: CV internal term created
LevelSet 0: CV external term created
Term container 0 created
Equation container created

Stopping criterion created

 

LDD output

ldd ~/ITK/build-icc-004-avx2-ipo/bin/ITKLevelSetsv4TestDriver
	linux-vdso.so.1 =>  (0x00007ffdf6b61000)
	libm.so.6 => /lib64/libm.so.6 (0x00007f2881a6b000)
	libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f288184f000)
	libdl.so.2 => /lib64/libdl.so.2 (0x00007f288164b000)
	libstdc++.so.6 => /lib64/libstdc++.so.6 (0x00007f2881341000)
	libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007f288112b000)
	libc.so.6 => /lib64/libc.so.6 (0x00007f2880d6a000)
	/lib64/ld-linux-x86-64.so.2 (0x00007f2881d8a000)

 

Command line for running vtune which causes hang

/opt/intel/vtune_amplifier_xe_2017.2.0.499904/bin64/amplxe-cl -collect hotspots -knob enable-user-tasks=false  -r /tmp/r005hs --duration 70 -- $HOME/ITK/build-icc-004-avx2-ipo/bin/ITKLevelSetsv4TestDriver itkMultiLevelSetMalcolmImageSubset2DTest

 

Thread Topic: 

Help Me

Error: Too many events are opened

$
0
0

my command line:

amplxe-cl -collect memory-access -knob sampling-interval=100 -target-pid=50457

Output message

amplxe: Warning: To profile kernel modules during the session, make sure they are available in the /lib/modules/kernel_version/ location.
Duplicate Event Discarded: MEM_LOAD_UOPS_L3_MISS_RETIRED.LOCAL_DRAM_PS
Duplicate Event Discarded: MEM_LOAD_UOPS_L3_MISS_RETIRED.REMOTE_DRAM_PS
amplxe: Warning: Event-based sampling analysis with sampling drivers supports collection of precise events. For driverless collection, these events will be collected as non-precise: MEM_UOPS_RETIRED.ALL_LOADS_PS,MEM_UOPS_RETIRED.ALL_STORES_PS,MEM_TRANS_RETIRED.LOAD_LATENCY_GT_4,MEM_LOAD_UOPS_L3_MISS_RETIRED.LOCAL_DRAM_PS,MEM_LOAD_UOPS_L3_MISS_RETIRED.REMOTE_DRAM_PS,MEM_LOAD_UOPS_RETIRED.L3_HIT_PS,MEM_LOAD_UOPS_RETIRED.L3_MISS_PS,MEM_LOAD_UOPS_L3_MISS_RETIRED.REMOTE_HITM_PS,MEM_LOAD_UOPS_L3_MISS_RETIRED.REMOTE_FWD_PS
amplxe: Collection started. To stop the collection, either press CTRL-C or enter from another console window: amplxe-cl -r /home/arda/log/r016macc -command stop.
amplxe: Per-node peak bandwidth measurement is enabled for this collection. Collection time may increase.
^Camplxe: Error: Error:
Too many events are opened.
Try again after reducing the number of events.

amplxe: Collection failed.
amplxe: Internal Error

Sorry, I don't know how to reduce events number, can you help me?

Filtered Out... with no filtering?

$
0
0

I just took my first profile data with this tool. The advanced hotspots view has a timeline with all the threads. However, there are multiple gaps in the timeline of 100s of milliseconds each where there is no data for any thread. Hovering over a gap gives a tool tip of "filtered out". Unfortunately, I don't know what this means and as far as I can tell, I have no filters set. Searching the forum, the web, and the help for "filtered out" didn't reveal the cause.

What is happening here?

[Instrumentation Engine]: Source/pin/elfio/img_elf.cpp: ProcessSectionHeaders: 601: assertion failed: SEC_vaddr_i(sec) >= IMG_se

$
0
0

Hello all,

I'm trying to use VTune's software analysis modes (basic hotspots, concurrency, locks&waits) but it seems like an update somewhere in the toolchain triggers a fail since recently. Now all binaries I'm trying to profile immediately abort with the following error message:

Collection failed
    Do 22 Jun 2017 15:01:40 CEST  Collection failed. The data cannot be displayed.
    [Instrumentation Engine]: Source/pin/elfio/img_elf.cpp: ProcessSectionHeaders: 601: assertion failed: SEC_vaddr_i(sec) >= IMG_seg_text_vaddr_i(img) && SEC_vaddr_i(sec) < IMG_seg_data_vaddr_i(img)

I doubt that the ELF format changed. GCC is also not the reason, I believe, since it only broke recently on Arch but also breaks on older GCC 6... I've observed this issue on various distributions now, i.e. Ubuntu 17.04 with GCC 6.3.0 20170406, OpenSuse Tumbleweed and ArchLinux with GCC 7.1.1 20170528.

Any idea what's going on here? Maybe some kind of page protection mechanism that got enabled recently?

Intel VTune Amplifier issues

$
0
0

I recently underwent training on the Intel VTune Amplifier tool and had been trying the lab exercises mentioned in the Getting Started document provided with the Linux samples provided in the training. Most of the Linux samples could not be built due to absence of specific files. Also, some of the things mentioned in the Getting Started document didn't seem available for view when using the tool. Is there any limitation of features in the eval version of this tool?

I am seeing the following issues. Any pointers on its resolution will be welcome!

1) Not displaying LLC misses in details

2) Lab 4 – How to enable prefetching

3) Lab 5 - ipp_resize_lab folder does not have any labcode folder as mentioned in Getting Started document.

4) Lab 6 - tbb_lab – build using command ./build.sh failed. 

$sudo ./build.sh

./build.sh: line 4: icc:command not found

5) Lab 7 - itt_csv not compling. Complaining about missing md5.h file.

 

Thanks

Monisha

VTune Amplifier XE 2017 (3.0.510739) on Linux no longer invoking child application

$
0
0

Hello,

I recently started evaluating VTune internally, and I had some good successes initially. However, I was working with the VTune "scripting" team on trying to diagnose a Python issue with VTune, and now VTune has just "given up". It is worth noting that, at the current time, the Python analysis only supports the "basic" hotspot (and Locks-n-Waits) collectors.

To elaborate on my issue: whatever kind of binary is passed as follows:

* amplxe-cl -c hotspots

nothing is collected, and all that is reported is:

* amplxe: Error: 0x40000025 (Inapplicable report) -- The report 'summary' is not applicable to the result

However, I fully understand this error when nothing is collected/the process is too short-lived.

The problem here is that VTune is simply not invoking/launching the child job -- the process I wish to analyse has print statements (for example), and these _do not_ appear. This is why there is no data collector, the child process isn't being launched at all.

VTune did previously work on my machine before I updated my Linux kernel (I updated from 3.10.0-514.21.1.el7.x86_64 to 3.10.0-514.21.2.el7.x86_64 - the smallest of point releases!). After I noticed that VTune was broken, I even completely uninstalled/reinstalled the product, but to no avail.

Other things I have tried:

* advanced-hotspots _do_ work, but these are incompatible with Python analyses (so they do not relevant for the purposes of this post)

* manually making sure the kernel modules are loaded using insmod-sep

but these have also made no difference!

I was extremely impressed with what VTune could do before it "broke", so this is extremely infuriating!

Any helps/tips/suggestions/things I can try to give you all more information?

Cheers,

Andrew

Thread Topic: 

Help Me

getting to view source with results

$
0
0

OK - I collected data for my application.  I can open results and see data without source - but I can not associate source with the information.

Here is what I used to collect the VTune information in my batch file.

mpirun -np 1 amplxe-cl -collect hotspot –r ./vtunehotproj ./a.out : -np 127 ./a.out.exe

I submitted my job and it ran and created a directory vtunehotproj.n025 (why n025?  I don't know but that is fine).  In this directory there is a file vtunehotproj.n025.amplxe and 4 directories:  archive, config, data.0, and sqlite-db.

Now I come back an hour later and see the job ran.  I run amplxe-gui and I choose open results and select the file vtunehotproj.n025.amplxe and it lists where most time is spent by file name.  I want to associate source  - but I see I can not add a source directory path to a "results".  So I search through the documentation and see I need to create a project, import the data into the object and then add the source path.  So I go in amplxe-gui and create a project.  Then I select the import option -but this will not let me import vtunehotproj.n025.amplxe.   So I decide to select a directory - so I select data.0 and import this.   But now I no longer get information about my application.  I associate the binary ./a.out as well as the source direcotry - but now I have even less information about the application   amplxe gui now shows me time in a.out, libmpi.so.12 libc.so.6 libpthread.so.0 and [unknown stack frame(s)].   I click on ./a.out to see details - but it gives me a chart with the same modules above and treats ./a.out as a monolithic blob with no functions.

If I select open results - I can see hotspots by function by no association with source code line numbers (yes I compiled with -g).  If I import results I lose all function information so even though I can associate a "source code directory" to the project nothing is resolved!   This is worthless.   How can I get function breakdown AND source association?    What directory or what file do I need to "import" ?

 

 

Zone: 

Thread Topic: 

Bug Report

Where is the = menu button

$
0
0

I wanted to try to refinalize my results after updating source directories.    I clicked on the ? button in the gui to get directions.   The directions say to select a particular icon on the gui menu.   I can not find that icon.  I am attaching a screenshot which shows captures the VTune view as well as the help documentation.   The background window is the gui (without the specified icon), the front window is the VTune documentation showing the icon I should be looking for (and failing to find).

(sorry if resolution is poor - I had to adjust image to allow upload to this forum).

AttachmentSize
Downloadimage/jpegtt2.jpg183.27 KB

Thread Topic: 

Bug Report

vTune axmple-cl memory-access hangs on

$
0
0

Hi.

I use the vTune amplxe-cl -collect memory-access command to collect data.The vTune version is vtune_amplifier_xe_2016_update4.tar.gz , my linux version is CentOS7 , and my CPU is Core i3.When I use the action option with memory-access , bandwidth , and  cpugpu-concurrency ,the command will hangs on there. When I use other action option like hotspots, advance-hotspots can not hang on and run normally. My command like these :

amplxe-cl -collect memory-access -target-duration-type medium -target-pid 9692

amplxe-cl -collect memory-access -target-duration-type medium -target-pid 9692 -duration 10

amplxe-cl -collect memory-access -knob analyze-mem-objects=true -knob dram-bandwidth-limits=true  -target-pid 20753

 

All these commands will hang on when using. Can someone help me?

Thanks.

some collections break with newest RHEL/CentOS 7.3 kernel?

$
0
0

We are running VTune 17.3  ("Intel(R) VTune(TM) Amplifier XE 2017 Update 3 (build 510739)") on our Xeon Phi 7250 systems with RHEL/CentOS 7.3.

We are running VTune with perf_events support, rather than with Intel's SEP module.

Last week we upgraded to the newest kernel (3.10.0-514.21.2.el7) and noticed that a number of analyses that used to work are no longer working.  In particular, the "hotspots" analysis fails to run, while "hpc performance" and "advanced hotspots" typically do run.

The failures look like:

$ amplxe-cl -collect hotspots ./stream.runtime.avx2.10x
amplxe: Collection started. To stop the collection, either press CTRL-C or enter from another console window: amplxe-cl -r /home1/00145/mccalpin/Stampede2/STREAM/r000hs -command stop.
amplxe: Warning: Can't find 32-bit pin tool. 32-bit processes will not be profiled.
amplxe: Collection failed.
amplxe: Internal Error

 

how to get the option CPU specific analysis

$
0
0

HI

I am using Vtune for System Version 2017 update3, I notice from the help document that there is the option of  CPU specific analysis on the Analysis Type Page, as shown in the attachment 1

However, I could not find this option on my PC(as shown in attachment 2). Is it due to the Vtune Version?

Many thanks for your help!

 

 

AttachmentSize
Downloadimage/pngc.png92.12 KB
Downloadimage/pngd.png92.23 KB

Failed to get a thread handle when attaching to the process due to OS error (Access is denied. )

$
0
0

I run into the following problem when try to collect data:

Collection failed

    7/6/2017 3:28:51 PM  Collection failed. The data cannot be displayed. 
    Failed to get a thread handle when attaching to the process due to OS error (Access is denied.
). Data for the thread will not be collected.

The OS: Windows 10.

Target application: MSSQL2014 (sqlservr.exe) and it is Log On as "Network Service", Session 0.

I try to use "Attach to Process"& Basic Hotspots.

 

 

Zone: 

Application Performance Snapshot 2018 Beta Update is available for download

$
0
0

Hello,

VTune team wants to inform you that the new update is available for Application Performance Snapshot 2018 Beta for Linux on the web.

The tool is a light weight and easy to use performance report about MPI and OpenMP runtime usage efficiency and basic memory access, FPU Utilization, I/O and memory footprint metrics shown for the whole application run. 

The new update improves startup time under analysis for the case of big number of ranks per node, supports CRAY* MPI Fortran applications and has result summary info section with number of ranks and threads and HW platform info.

Please do not hesitate to share feedback, ask questions to this forum on APS.

Thanks & Regards, Dmitry

Viewing all 1347 articles
Browse latest View live


<script src="https://jsc.adskeeper.com/r/s/rssing.com.1596347.js" async> </script>