Skip to content

HPE Cray PE tools introduction

Presenters: Harvey Richardson and Alfio Lazzaro (HPE)

  • Slides on LUMI in /appl/local/training/profiling-20241009/files/01_HPE_Cray_PE_tools.pdf

  • Recordings: To make the presentations more accessible, the presentation has been split in 6 parts:

    1. Introduction and LUMI hardware (Harvey Richardson): slides 1-8: /appl/local/training/profiling-20241009/recordings/01a_HPE_Cray_PE_tools__Hardware.mp4

    2. The HPE Cray Programming Environment (Harvey Richardson): slides 9-38: /appl/local/training/profiling-20241009/recordings/01b_HPE_Cray_PE_tools__Programming_environment.mp4

    3. Job placement (Harvey Richardson): slides 39-51: /appl/local/training/profiling-20241009/recordings/01c_HPE_Cray_PE_tools__Job_placement.mp4

    4. Cray MPICH for GPUs (Harvey Richardson): slides 52-57: /appl/local/training/profiling-20241009/recordings/01d_HPE_Cray_PE_tools__MPICH_GPU.mp4

    5. CCE Fortran and C/C++ & Offload to the GPUs (Alfio Lazzaro): slides 58-75 /appl/local/training/profiling-20241009/recordings/01e_HPE_Cray_PE_tools__CCE_Fortran_and_offload.mp4

    6. Performance Analysis (Alfio Lazzaro): slides 76-113: /appl/local/training/profiling-20241009/recordings/01f_HPE_Cray_PE_tools__Performance_analysis.mp4

    The "GDB for HPC" slides (114-127) were not covered in the presentation.

HPE training materials can only be shared with other LUMI users and therefore are not available on the web.

Q&A

  1. Slide 25 may need an update. With the 24.03 edition it is better to use gcc-native-mixed to get the gcc installation specifically for this version of the PE. And then you have to use gcc-13 --version as gcc --version would give you the (ancient) system gcc.

    • (Alfio) the comment is correct, then you have to use gcc-13 instead of gcc. However, gcc-13 is always available, so don't need to load the gcc-native-mixed module. Proper linking to gcc has been fixed in PE 24.07 (not yet on LUMI).

  2. How does ROCR_VISIBLE_DEVICES interact with OpenMP calls that detect the number of devices ("omp_get_num_devices") and set/get the default target device ("omp_set_default_device" and "omp_get_default_device")?

    • It works at the lowest level of the ROCm stack and will limit what the OpenMP runtime can detect. For example you might only see one device from OpenMP (or a HIP program)