Changeset 1601 for LMDZ4/branches/LMDZ4_AR5/arch

Timestamp:

Dec 7, 2011, 8:05:43 AM (13 years ago)

Author:

Ehouarn Millour

Message:

Updates and upgrades of filter:

• add minor corrections (r1591 of trunk) in filtreg_mod.F90: some arrays were oversized and the computation of the indexes from which the filter is applied could go wrong in extreme cases.
• adapt filtreg_p.F (r1597 of trunk) so that BLAS routine DGEMM is only used if 'BLAS' preprocessing flag is set.
• update FFT filter routines to match current 'trunk' versions (mostly cosmetic changes, exept for the implementation of use of FFTW in mod_fft_fftw.F90).
• update "arch-PW6_VARGAS.fcm" to enable use of FFTW

NB: implemented FFTs assume working precision to be double precision ; trying to use them when working precision is single precision will clearly end in tragedy.
EM

File:

• LMDZ4/branches/LMDZ4_AR5/arch/arch-PW6_VARGAS.fcm (modified) (2 diffs)

LMDZ4/branches/LMDZ4_AR5/arch/arch-PW6_VARGAS.fcm

@@ -3 +3 @@
 %AR                  ar
 %MAKE                gmake
-%FPP_FLAGS           -P
-%FPP_DEF             NC_DOUBLE BLAS SGEMV=DGEMV SGEMM=DGEMM
-%BASE_FFLAGS         -qautodbl=dbl4 -qxlf90=autodealloc -qextname=flush
+%FPP_FLAGS           -P -I/usr/local/pub/FFTW/3.2/include
+%FPP_DEF             NC_DOUBLE BLAS SGEMV=DGEMV SGEMM=DGEMM FFT_FFTW
+%BASE_FFLAGS         -qautodbl=dbl4 -qxlf90=autodealloc -qmaxmem=-1 -qzerosize -I/usr/local/pub/FFTW/3.2/include
 %PROD_FFLAGS         -O3
 %DEV_FFLAGS          -O2 -qfullpath -qinitauto=7FBFFFFF -qfloat=nans -qflttrap=overflow:zerodivide:invalid:enable -qsigtrap
@@ -11 +11 @@
 %MPI_FFLAGS          -I/usr/lpp/ppe.poe/include/thread64
 %OMP_FFLAGS          -qsmp=omp
-%BASE_LD             -lessl
+%BASE_LD             -lessl -L/usr/local/pub/FFTW/3.2/lib -lfftw3
 %MPI_LD              
 %OMP_LD              -qsmp=omp