Adastra’s software stack

• CINES software stack
  • Introduction
    • A note on dependencies
  • Basic tools and libraries
  • Extensible software stacks
  • Using the CINES software stack
  • Advanced module mixing
  • Using CINES’ pre-configured (PreConf) Spack installation
    • Getting started
    • If a Spack install fails
    • Limitations
    • Recommendation for shared Spack installation
    • Recommendations on correct usage
    • Further reading

• CINES Spack module catalog

• Tools
  • CINES Spack modules
  • Building software
    • CMake
    • Ninja
  • Conda
  • Debugging
    • Cray tools
    • GDB
    • rocGDB
    • Valgrind
    • Sanitizers
    • ldd and nm
    • gstack
  • Julia
  • Jupyter
  • Linker, assembly, ABI
    • objdump
    • ld-linux.so
  • Monitoring
    • htop
    • rocmsmi
  • ParaView & VTK
  • Profiling
    • Cray-PAT’s perftools
    • HPCToolkit
    • Intel VTune & Advisor
    • rocProfiler Compute
    • rocProfiler Systems
    • perf
    • rocPROF
    • rocPROFv2
    • rocPROFv3
    • PyTorch
    • Scalene
    • /usr/bin/time
  • Python
    • Virtual environment
  • R (project)
    • Installing packages
  • Trivia
    • ClangFormat
    • ClangTidy
    • lfs find
    • numactl
    • VSCode (Visual Studio Code)
    • Current bash source directory
    • source
    • xeyes
    • watch
    • wget

• Libraries & toolchains
  • CINES Spack modules
  • Example script for CINES Spack modules libraries
  • Apex
  • BLAS, LAPACK & ScaLAPACK
    • LibSci
    • OpenBLAS
  • C++ standard parallelization (STDPAR)
    • AMD HIP support
  • Deepspeed
    • JITed kernels
    • Prebuilt kernels
    • Some environment variables
  • Flash attention (v1/v2)
    • Flash attention v1 for PyTorch <= 2.3.1+rocm5.7
    • Flash attention v2 for PyTorch > 2.3.1+rocm6
    • Benchmarks
  • HDF5
    • h5py
  • Low level primitives
  • MIOpen
  • MPI
    • Cray MPICH
    • Using MPI in Fortran a code
    • Detecting if MPI is GPU aware
    • mpi4py
  • OpenMP
    • CPU
    • GPU offload
    • Some environment variables
  • PyTorch
    • Installing PyTorch for AMD GPUs
    • Building dependencies
    • Torch elastic
    • IDRIS’ accelerate wrapper
    • Distributed training
    • Some environment variables
    • Debugging parallel loops
    • Miscellaneous documents
  • RCCL
    • Debugging networking issues
    • Throughput tuning
  • ROCm
    • Modules and paths
    • Build integration
    • Relocatable Device Code (RDC)
    • C preprocessor definitions
    • Some environment variables
    • HIP compiler flag details
    • Offload architecture
    • Miscellaneous documents
  • Sycl
    • Intel LLVM
    • AdaptiveCPP
  • The transformers & datasets libraries
    • Some environment variables
  • vLLM
    • For ROCm 6.1.2 and up:
    • Usage
    • Some environment variables
    • Miscellaneous documents

• Containers
  • Available containers
  • Using a container
    • Running as a SLURM batch job
    • Running as a SLURM interactive session

• Simulation software scripts and tips
  • CINES software stack
    • Amber
    • CP2K
    • DFTB+
    • GROMACS
    • LAMMPS
    • NAMD
    • Quantum espresso
  • Other simulation softwares
    • Boltz
    • VASP