Changeset 2115 for LMDZ5


Ignore:
Timestamp:
Sep 8, 2014, 8:20:44 AM (10 years ago)
Author:
Ehouarn Millour
Message:

Remove default use of BLAS calls which are only in the non-FFT filter matrix products; matmul() is nowadays well optimized (and moreover using BLAS leads, with ifort 13 on Ada, to slightly different results in seq/MPI/OpenMP bench runs).
EM

Location:
LMDZ5/trunk/arch
Files:
2 edited

Legend:

Unmodified
Added
Removed

  • LMDZ5/trunk/arch/arch-X64_ADA.fcm

    r1957 r2115  
    44%MAKE                gmake
    55%FPP_FLAGS           -P -traditional -I/smplocal/pub/FFTW/3.3.3_dyn/include/
    6 %FPP_DEF             NC_DOUBLE BLAS SGEMV=DGEMV SGEMM=DGEMM FFT_FFTW
     6%FPP_DEF             NC_DOUBLE FFT_FFTW
    77%BASE_FFLAGS         -auto -mcmodel=large -integer-size 32 -real-size 64 -align all
    88%PROD_FFLAGS         -O2 -ip -fp-model strict -axAVX,SSE4.2

  • LMDZ5/trunk/arch/arch-X64_CURIE.fcm

    r2108 r2115  
    44%MAKE                gmake
    55%FPP_FLAGS           -P -traditional
    6 %FPP_DEF             NC_DOUBLE BLAS SGEMV=DGEMV SGEMM=DGEMM FFT_MKL
     6%FPP_DEF             NC_DOUBLE FFT_MKL
    77#%BASE_FFLAGS         -xHost -i4 -r8 -auto -align all -I$(MKL_INC_DIR) -I$(MKL_INC_DIR)/intel64/lp64
    88%BASE_FFLAGS         -i4 -r8 -auto -align all -I$(MKL_INC_DIR) -I$(MKL_INC_DIR)/intel64/lp64 -fp-model strict
Note: See TracChangeset for help on using the changeset viewer.