Page tree


HPCToolkit is an integrated suite of tools for measurement and analysis of program performance on computers ranging from multicore desktop systems to the large scale supercomputers.

It provides accurate measurements of a program’s work, resource consumption, and inefficiency, correlates these metrics with the program’s source code, works with multilingual, fully optimised binaries, has very low measurement overhead, and scales to large parallel systems.

HPCToolkit's measurements provide support for analysing a program execution cost, inefficiency, and scaling characteristics both within and across nodes of a parallel system.   

More information:

How to use 

You can check the versions installed in Gadi with a module query:

 $ module avail hpctoolkit
$ module avail hpcviewer  # For visualisation

We normally recommend using the latest version available and always recommend to specify the version number with the module command:

 $ module load hpctoolkit/2021.05.15
$ module load hpcviewer/2021.05.15  # For visualisation

For more details on using modules see our software applications guide.

Collect Profile Measurements 

Measurement of application performance takes two different forms depending on whether your application is dynamically or statically linked. To monitor a dynamically linked application, simply use hpcrun to launch the application. To monitor a statically linked application, link your application using hpclink.   

  • Dynamically linked binaries 
    • To monitor a sequential or multithreaded application, use:   

       $ hpcrun [options] prog.exe [arguments]
    • To monitor an MPI application, use:   

       $ mpirun hpcrun [options] prog.exe [arguments]
  • Statically linked binaries
    • To link hpcrun's monitoring code into prog.exe, use:   

       $ hpclink <linker> -o prog.exe <linker-arguments>

If no options is specified to hpcrun, walltime will be measured for prog.exe. Otherwise, please specify PAPI events to be measured for prog.exe. A available list of PAPI events can be retrieved by running following command:   

 $ hpcrun -L prog.exe

A sample PBS job script for using hpcrun with measurements passed through environment variables is like following:   

An example PBS job submission script named is provided below. It requests 48 CPUs, 128 GiB memory, and 400 GiB local disk on a compute node on Gadi from the normal queue with exclusive access for 30 minutes against the project a00. It also requests the system to enter the working directory once the job is started.

This script should be saved in the working directory from which the analysis will be done. To change the number of CPU cores, memory, or jobfs required, simply modify the appropriate PBS resource requests at the top of the job script files according to the information available in our queue structure guide.

Note that if your application does not work in parallel, setting the number of CPU cores to 1 and changing the memory and jobfs accordingly is required to prevent the compute resource waste.

#PBS -P a00
#PBS -q normal
#PBS -l ncpus=48
#PBS -l mem=128GB
#PBS -l jobfs=400GB
#PBS -l walltime=00:30:00
#PBS -l wd
# Load modules, always specify version number.
module load openmpi/4.1.1
module load hpctoolkit/2021.05.15
# Must include `#PBS -l storage=scratch/ab12+gdata/yz98` if the job
# needs access to `/scratch/ab12/` and `/g/data/yz98/`

# Set measurements
# Run application
mpirun -np $PBS_NCPUS hpcrun prog.exe

The above PBS job script when measurements passed as an option with hpcrun is like the following:      

#PBS -P a00
#PBS -q normal
#PBS -l ncpus=48
#PBS -l mem=128GB
#PBS -l jobfs=400GB
#PBS -l walltime=00:30:00
#PBS -l wd
# Load modules, always specify version number.
module load openmpi/4.1.1
module load hpctoolkit/2021.05.15
# Must include `#PBS -l storage=scratch/ab12+gdata/yz98` if the job
# needs access to `/scratch/ab12/` and `/g/data/yz98/`
# Run application
mpirun -np $PBS_NCPUS hpcrun -e WALLCLOCK@5000 prog.exe

To run the job you would use the PBS command:

 $ qsub

Sampling Frequency and Measurements

In the above example, 5000 is a sample rate for each individual measurement. With larger number of the sample rate, the sample frequency is lower, and associate overhead of HPCToolkit is lower. In general, the overhead of HPCToolKit is around 1% to 3%.   

Some other useful measurements include:   

  • WALLCLOCK: Walltime spent on each functions, or outstanding instructions.  
  • PAPI_FP_INS: Floating point instructions (x87)  
  • PAPI_VEC_SP: Single precision vector/SIMD instructions  
  • PAPI_VEC_DP: Double precision vector/SIMD instructions  
  • PAPI_LD_INS: Load instructions  
  • PAPI_SR_INS: Store instructions  
  • PAPI_BR_INS: Branch instructions  
  • and more…, please refer to hpcrun -L prog.exe for a complete list of measurable events, or the PAPI Preset Events list. 

The available measurement events are different between different systems. Please make sure the event is available and measurable using hpcrun -L prog.exe.

To measure multiple events at once, following format of event options or environment variable can be used:  

  • -e WALLCLOCK@5000 -e PAPI_LD_INS@4000001 -e PAPI_SR_INS@4000001 

  • export HPCRUN_EVENT_LIST="WALLCLOCK@5000;PAPI_LD_INS@4000001;PAPI_SR_INS@4000001"

Profile Data Parse 

hpcrun will generate a directory named hpctoolkit-<prog.exe>-measurements-<jobid> in your job's directory.

Please follow the following sequence to parse the raw measurements in hpctoolkit-<prog.exe>-measurements-<jobid>  

Recovering Program Structure 

 $ hpcstruct prog.exe

This will generate a prog.exe.hpcstruct file which contains the code structure of prog.exe  

Parse the Raw Measurements 

For serial program:    

 $ hpcprof -S prog.exe.hpcstruct -I <source code directory>/'*' hpctoolkit-<prog.exe>-measurements-<jobid>

For parallel (MPI/OpenMP) program:   

 $ hpcprof --force-metric --metric=<metrics option> -S prog.exe.hpcstruct -I <source code directory>/'*' hpctoolkit-<prog.exe>-measurements-<jobid>

Options for --metric (or -M) includes:  

  • sum: show (only) sum over threads/processes metrics (default)
  • stats: show (only) sum, mean, standard dev, coef of var, min, and max over threads/processes metrics
  • thread: show only thread metrics

Please refer hpcprof --help for more details.   

A graphical presentable database will be generated after hpcprof executed. It is a directory with name like:    

 $ hpctoolkit-<prog.exe>-database-<jobid>

Graphical Viewer 

To visualise the HPCToolKIt profile data on Gadi, login to Gadi with X11 (X-Windows) forwarding. Add the -Y option for Linux/Mac/Unix to your SSH command to request SSH to forward the X11 connection to your local computer. For Windows, we recommend to use MobaXterm ( as it automatically uses X11 forwarding.  

For more information on MobaXterm and X-forwarding, please see our connecting to Gadi page

# Load module, always specify version number.
$ module load hpcviewer/2021.05.15
$ hpcviewer hpctoolkit-<prog.exe>-database-<jobid>

Two different metric is presented: inclusive and exclusive, denoted by "I" and "E" respectively in the metric panel of hpcviewer  

  • "I" indicates the inclusive measurement: represents the sum of all costs attributed to this call site and any of its descendants.   
  • "E" indicates the exclusive measurements: only represents the sum of all costs  attributed strictly to this call site.  
Authors: Javed Shaikh, Mohsin Ali
  • No labels