Juan 05/06/2024:Rules.LXnvhpc2202.mk, add correct Rules to not compile source with "acc kernel" with "stdpar=gpu" ( could inhibe hand made parallelisation )

Juan 03/06/2024:Makefile , add "all:" target at beginning to avoid side effect of print/debug targets

Juan: 3/06/2024: Merge branch 'MNH-56X-dev-OPENACC-FFT-unlessPHYEX' after turb_ver_sv with shuman integration  into MNH-56X-dev-OPENACC-FFT-unlessPHYEX-STDPAR-before-merge

Juan 31/05/2024:rain_ice_nucleation.f90/modeln.f90 , For GPU , problem with scalair pointer variable XTT/VSIGQSAT -> replace with local variable

Juan 30/05/2024:rain_ice_nucleation.f90 , add !$acc kernels & replace MAXVAL with "!$acc loop reduction(max:ZZWMAX)"

Juan 30/05/2024:modd_cst.f90/compute_frac_ice.func.h/mode_tiwmx.f90 , For GPU routine to work , add PURE & "!$acc routine seq" & "!$acc declare VAR"

Juan 30/05/2024:Rules.LXnvhpc2202.mk, IMPORTANT CHANGE:: Only compile PHYEX(+some other source) with -sdtpar=gpu, otherwise BIG degradation of performance for source already ported with "!$acc kernels/parallel" directives

Juan 30/05/2024: revert commit "49d83e1d - Quentin 17/05/2024" -> undo manuel conversion of integer -> real for BR_ function , now handeled by modi_bitrep.f90 with integer interface

Quentin 30/05/2024: add kernels on turb_hor_splt.F90, must be used with pyft version of 30/05/2024(+ bugfix on bf_ExpandArrays on pyft)

Quentin 29/05/2024: add kernels on turb_ver_dyn_flux and complete testing of shuman/gradients conversion function to routines with the last version of pyft (of 29/05/2024)

Quentin 24/05/2024: add kernels on turb_ver_thermo_flux.f90 (one current computation still bug, not ported)

Quentin 17/05/2024: add conversion of math function to BR_ bit-repro functions with pyft with the manual conversions from integer to real to make it work

Quentin 16/05/2024: acc kernels for turb_ver and tke_eps_sources with adaptation of automatic conversion of shuman functions to routines in tke_eps_sources

Juan 15/05/2024:Rules.LXnvhpc2202.mk, auto-parallelize mode_turb_hor_dyn* , with pyft  --addArrayParentheses --expandAllArraysPHYEXConcurrent ...

Juan 15/05/2024:PHYEX/aux/shuman_phy.f90|mode_gradient_m_phy.f90, add acc kernels/loop to parallelize DO loop

Juan 15/05/2024:PHYEX/turb/mode_tridiag*|emoist|etheta , reverse JK & JIJ Loop & use parallel_cr=kernel to auto-parallelize nested/sequential+parallel loops

ESCOBAR Juan authored 10 months ago and RODIER Quentin committed 10 months ago

Juan 24/04/2024:LXnvhpc2202.mk, Compile with CC=gcc, to avoid problem with Netcdf (like with ifort/icc)

Juan 13/05/2024:PHYEX/turb/mode_prandtl.f90, add acc loop collapse(3) on JK*JIJ loop , for better performance (on MNH55 , array syntax keeped )

Juan 13/05/2024:PHYEX/turb/mode_coefj.f90 , for "automatic" bit_repro convertion , add USE MODI_BITREP

Juan 13/05/2024:PHYEX/micro/* , for "automatic" bit_repro convertion , add USE MODI_BITREP & some missing (:) on array statement

Juan 13/05/2024:MNH_OPENACC_NV_CR.CPP, Add macro to managed replacement of BR_ function automatically

Juan 13/05/2024:PHYEX/micro/radar_rain_ice.f90, MOMG function , for GPU+OpenACC+STDPAR add PURE + "!$acc routine seq"

resolved_cloud.f90, for GPU+SDTPAR, try to have reproductiblility, add (:,..,:) to array & compute elsewhere statement

Juan 13/05/2024:gamma.f90/momg.f90 , for GPU+OpenACC+STDPAR add PURE + "!$acc routine seq" & comment DrHOOK

Juan 24/04/2024:discretisation.f90/dotprod.f90, for STDPAR ,use new "parallel_cr" macro to permit automatic parallelization