Developer Tools

Debuggers

On Midway3 GDB and Valgrind as modules for debugging and memory leak checking with your code.

To debug your code with gdb, you compile your code with -g and then run gdb:

g++ -g -o test test.cpp
module load gdb
gdb --args ./test param1 param2

You can also use gdb to debug your Python codes. Alternatively, use pbd.

python3 -m pdb myscript.py

To check if there is any memory leak with your code, use valgrind

g++ -g -o test test.cpp
module load valgrind
valgrind --leak-check=full --track-origins=yes ./test param1 param2

Please refer to the official documentation of GDB and Valgrind for more information.

For CUDA codes, after loading the CUDA toolkit module (e.g. module load cuda/12.2) you can use cuda-gdb and cuda-sanitizer.

Profilers

On Midway3 there are currently 3 profiling tools: TAU, NVIDIA Nsight and Intel VTune.

TAU is a tool for profiling and tracing toolkit for performance analysis of parallel programs written in Fortran, C, C++, UPC, Java, Python. You can load tau/2.31 on Midway3.

The TAU binaries and scripts in the subfolders $TAU_HOME/bin, and several Makefile.* in $TAU_HOME/lib . To use TAU, you need to rebuild your code using one of the scripts inside $TAU_HOME/bin: C codes use tau_cc.sh; C++ codes use tau_cxx.sh (see for example, the doc page at OLCF). Should a new Makefile.foo be needed, then the CC and CXX environment variables need to be exported:

export CC=tau_cc.sh CXX=tau_cxx.sh F90=tau_f90.sh

before the first cmake run, or before running make.

Run the generated binaries as usual (e.g. mpirun -np 4 ./binary args) the output will now include the files profile.* in the working directory. At this point, run paraprof to launch the GUI analysis (need module load java), or pprof for command-line output.

Different options for compile time are available (e.g. -optVerbose), options for run time (e.g. TAU_TRACK_MEMORY_LEAKS=1)

NVIDIA Nsight is a system-wide performance analysis tool designed to visualize an application’s algorithms, identify the largest opportunities to optimize, and tune to scale efficiently across any quantity or size of CPUs and GPUs.

To use NVIDIA Nsight System and Nsight Compute, load one of the available cuda modules.

module load cuda/12.2

To launch the corresponding GUI applications, use

nsys-ui

and

ncu-ui

You can also use the command-line interface nsys and ncu. Please refer to the NVIDIA Nsight documentation for further details.

Intel VTune is part of the oneAPI(C) software suite for optimizing application performance, system performance, and system configuration for HPC. To use the profilers, vtune and the GUI vtune-gui, in a ThinLinc session, you load the oneapi module

module load oneapi/2024.2
vtune-gui