Stress Tensor and MGGA functionals

[marci73]

FYI

Recent tests carried out for a water box of about 100 molecules and a cutoff of 1000 Ry show a significant slow down of MGGA calculations when the stress tensor is computed.
The MD step becomes about 5 times slower and the whole additional time goes into grid_integrate_task_list
The tests have been carried on daint-gpu

[oschuett]

Could you post the GRID STATISTICS printed at the end of the calculation? Most likely the additional derivatives needed for the stress tensor are hitting this cliff.