Move the PME GPU clearing to the end of the step
commit6543a8dd50f5ecff939af3d816d0e381ad7577de
authorSzilárd Páll <pall.szilard@gmail.com>
Wed, 18 Apr 2018 15:59:39 +0000 (18 17:59 +0200)
committerSzilárd Páll <pall.szilard@gmail.com>
Thu, 19 Apr 2018 09:37:51 +0000 (19 11:37 +0200)
tree2040a7f3bd47b15a7aad298fcbae52db859fea57
parent13fed60d423bb86e0d3e6eab3820041ad9883871
Move the PME GPU clearing to the end of the step

This places the call to the end of the force-compute next to the nbnxn
output clearing also fixing the cycle counting that previously
incorrectly assigned the cost of launching the clearing to the wait for
gather cycle counter.

Change-Id: I63b27a91d88a54d701831edda592b5d6277e3fb9
src/gromacs/ewald/pme-gpu-internal.cpp
src/gromacs/ewald/pme-gpu.cpp
src/gromacs/ewald/pme-only.cpp
src/gromacs/ewald/pme.h
src/gromacs/mdlib/sim_util.cpp