QMCPACK Release v3.7.0 - 2019-03-29

Download QMCPACK v3.7.0

Notes

This release includes GPU support for the AFQMC implementation, Quantum Espresso v6.4 support, and in the real-space code makes the structure-of-arrays (SoA) code path the default. A large number of feature refinements, bugfixes, testing improvements and source code cleanup have been performed.

  • The improved structures of arrays (SoA) build is now the default. This is generally significantly faster and uses less memory than the AoS build due to better algorithms, but does not yet have the full range of functionality. The older AoS build can be selected with -DENABLE_SOA=0.
  • AFQMC code fully supports GPU acceleration via NVIDIA CUDA. Use -DENABLE_CUDA=1.
  • Quantum Espresso v6.4 is supported. #1457
  • Better error handling e.g. #1423
  • Workarounds for MPI support on Summit. #1479
  • ppconvert should be more reliable. #891
  • Delayed update implementation on GPUs. #1279
  • Continued improvements to the testing system and test coverage. While still under development, a new set of deterministic tests is intended to rapidly and reliably test the code, with good coverage. Tests pass for real and complex, but not yet mixed-precision or GPU builds.
  • Source code has been formatted with clang-format for consistency throughout.

Known Bugs

See list of open bugs.

  • There is a bug that could result in an incorrect local electron-ion pseudopotential energy with CUDA v9.1 and Kepler GPUs. This is still being investigated. #1440

  • QMCPACK will not build with OpenMPI v4 due to use of deprecated functions. This will be addressed when the new MPI wrappers are fully adopted. Older OpenMPI libraries are fully capable.

NEXUS