MGGA energy functionals with GPU give incorrect energies
Here's the test log and octopus output for the functionals/18-mgga.02-br89_oep.inp test. The differences seem to be too large in comparison with the set tolerances. Can you please check the results?
Test environment
- CPU: Intel® Xeon® Platinum 8360Y
- GPU: NVIDIA A100-SXM4-40GB
- Libraries: gcc/10 gsl/2.4 libxc/5.1.2 cgal/5.2 impi/2021.2 hdf5-serial/1.8.21 metis/5.1 boost/1.74 mkl/2021.2 netcdf-serial/4.4.1 parmetis/4.0 cuda/11.2
- Configuration options : maxdim3 openmp mpi cuda sse2 avx libxc5
- Optional libraries : cgal metis mpi2 netcdf parmetis scalapack
- Architecture : x86_64
- C compiler : mpicc (gcc)
- C compiler flags : -O3 -march=native -g
- C++ compiler : mpicxx (g++)
- C++ compiler flags : -O3 -march=native -g -std=c++14
- Fortran compiler : mpif90 (gfortran) (GCC version 10.2.0)
- Fortran compiler flags : -O3 -march=native -g -Wall -Wno-maybe-uninitialized -Wno-unused-dummy-argument -Wno-c-binding-type -fallow-argument-mismatch -fallow-invalid-boz
Edited by Micael Oliveira