Philippe + Maxime MOGE 13/09/2016: imported + adapted gradient_v.f90 and gradient_w.f90 from Maxime's work

Juan 9/08/2016: For ppc64le pb , in gribapi/net tar.gz replace config.guess & config.sub with new version

Juan 9/08/2016: for Pgi conf problem , add CPP=ccp flags in Makefile & Rules.LXpgi.mk via CDF_CONF/GRIBAPI_CONF

* added updates between a call to GET_HALO_D. This is necessary to get correct results with OpenACC in parallel (reason not clear, maybe a bug in PGI 16.4)
* bug correction in get_halo_d (missing case if HDIR=S0_X or S0_Y)
* minor changes

Philippe 29/07/2016: force disabling of vectorisation for the loop to compute the sum in SUM_1DFIELD_ll (necessary to get bit reproductibility with PGI 16.4)

* added option to CHECK_MPPDB3D: MPPDB_CHECK_LB_CORNERS to check or not the corners
* check that the message is the same on all processes to detect some mistakes
* some little cleaning

Philippe 04/07/2016: compute compute SUM by hand instead of intrinsic Fortran in SUM_1DFIELD_ll to allow bit reproductibility with PGI compiler (16.4)

Philippe 04/07/2016: OpenACC: added _FAKEOPENACC key to compile with _OPENACC with a non OpenACCC compiler

Philippe 01/07/2016: -Mvect=nosimd forced to allow bit reproductibility (was causing problem with the SUM intrinsic in SUM_1DFIELD_ll) + added target OPENACCDEFONLY + some cleaning

Philippe 28/06/2016: bug correction: release of ZT3D block was incorrect. Due to that, sometimes the same block could be attributed several times.

Philippe 27/06/2016: 2 gradient_m versions: with or without linearized loops (useful for bound checking)

Philippe 27/06/2016: 2 shuman versions: with or without linearized loops + use of shuman instead of macros for OpenACC in ppm.f90

Philippe 23/06/2016: OpenACC: split shuman.f90 into 2 files: 1 with and 1 without the *_DEVICE versions

Philippe 14/06/2016: OpenACC: added a data zone to work around a bug visible with PGI compiler (at least to version 16.4)