Skip to content

GitLab

Commit ce80e103 authored by Andreas Marek's avatar Andreas Marek
Browse files

Merge branch 'master' of gitlab.mpcdf.mpg.de:elpa/elpa

parents 9519ede0 8f5c16b8
Hide whitespace changes
Inline Side-by-side
Showing with 64 additions and 53 deletions
Makefile.am
View file @ ce80e103
......@@ -2,43 +2,55 @@
ACLOCAL_AMFLAGS = ${ACLOCAL_FLAGS} -I m4
AM_FCFLAGS = $(SCALAPACK_FCFLAGS) @FC_MODINC@modules @FC_MODOUT@modules
AM_FCFLAGS = $(SCALAPACK_FCFLAGS) @FC_MODINC@modules @FC_MODINC@private_modules
AM_LDFLAGS = $(SCALAPACK_LDFLAGS)
# libelpa
lib_LTLIBRARIES = libelpa@SUFFIX@.la
libelpa@SUFFIX@_la_LINK = $(FCLINK) $(AM_LDFLAGS) -version-info $(ELPA_SO_VERSION) -lstdc++
libelpa@SUFFIX@_la_SOURCES = \
# parts with public interface
noinst_LTLIBRARIES = libelpa@SUFFIX@_public.la
libelpa@SUFFIX@_public_la_FCFLAGS = $(AM_FCFLAGS) @FC_MODOUT@modules
libelpa@SUFFIX@_public_la_SOURCES = \
src/elpa1.F90 \
src/elpa2.F90 \
src/elpa2_utilities.F90 \
src/elpa_utilities.F90
# internal parts
noinst_LTLIBRARIES += libelpa@SUFFIX@_private.la
libelpa@SUFFIX@_private_la_FCFLAGS = $(AM_FCFLAGS) @FC_MODOUT@private_modules
libelpa@SUFFIX@_private_la_SOURCES = \
src/mod_precision.f90 \
src/mod_mpi.F90 \
src/mod_mpi_stubs.F90 \
src/elpa2_kernels/mod_fortran_interfaces.F90 \
src/elpa_utilities.F90 \
src/elpa1_compute.F90 \
src/elpa1.F90 \
src/elpa2_utilities.F90 \
src/mod_pack_unpack_real.F90 \
src/elpa2_kernels/mod_single_hh_trafo_real.F90 \
src/mod_compute_hh_trafo_real.F90 \
src/mod_compute_hh_trafo_complex.F90 \
src/mod_pack_unpack_complex.F90 \
src/aligned_mem.F90 \
src/elpa1_compute.F90 \
src/elpa2_compute.F90 \
src/elpa2.F90 \
src/elpa2_kernels/mod_fortran_interfaces.F90 \
src/elpa2_kernels/mod_single_hh_trafo_real.F90 \
src/elpa_c_interface.F90 \
src/elpa_qr/qr_utils.F90 \
src/elpa_qr/elpa_qrkernels.f90 \
src/elpa_qr/elpa_pdlarfb.F90 \
src/elpa_qr/elpa_pdgeqrf.F90
EXTRA_libelpa@SUFFIX@_la_DEPENDENCIES = \
EXTRA_libelpa@SUFFIX@_private_la_DEPENDENCIES = \
src/elpa_reduce_add_vectors.X90 \
src/elpa_transpose_vectors.X90 \
src/redist_band.X90
lib_LTLIBRARIES = libelpa@SUFFIX@.la
libelpa@SUFFIX@_la_LINK = $(FCLINK) $(AM_LDFLAGS) -version-info $(ELPA_SO_VERSION) -lstdc++
libelpa@SUFFIX@_la_LIBADD = libelpa@SUFFIX@_public.la libelpa@SUFFIX@_private.la
libelpa@SUFFIX@_la_SOURCES =
if HAVE_DETAILED_TIMINGS
libelpa@SUFFIX@_la_SOURCES += \
libelpa@SUFFIX@_private_la_SOURCES += \
src/timer.F90 \
src/ftimings/ftimings.F90 \
src/ftimings/ftimings_type.F90 \
......@@ -51,101 +63,101 @@ if HAVE_DETAILED_TIMINGS
endif
if !WITH_MPI
libelpa@SUFFIX@_la_SOURCES += src/mod_time_c.F90
libelpa@SUFFIX@_private_la_SOURCES += src/mod_time_c.F90
if !HAVE_DETAILED_TIMINGS
libelpa@SUFFIX@_la_SOURCES += src/ftimings/time.c
libelpa@SUFFIX@_private_la_SOURCES += src/ftimings/time.c
endif
endif
if WITH_REAL_GENERIC_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real.F90
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real.F90
endif
if WITH_COMPLEX_GENERIC_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex.F90
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex.F90
endif
if WITH_REAL_GENERIC_SIMPLE_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_simple.F90
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_simple.F90
endif
if WITH_COMPLEX_GENERIC_SIMPLE_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_simple.F90
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_simple.F90
endif
if WITH_REAL_BGP_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_bgp.f90
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_bgp.f90
endif
if WITH_REAL_BGQ_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_bgq.f90
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_bgq.f90
endif
if WITH_REAL_SSE_ASSEMBLY_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_asm_x86_64.s
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_asm_x86_64.s
else
if WITH_COMPLEX_SSE_ASSEMBLY_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_asm_x86_64.s
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_asm_x86_64.s
endif
endif
if WITH_REAL_SSE_BLOCK2_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_sse_2hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_sse_2hv.c
endif
if WITH_REAL_AVX_BLOCK2_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_2hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_2hv.c
else
if WITH_REAL_AVX2_BLOCK2_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_2hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_2hv.c
endif
endif
if WITH_REAL_SSE_BLOCK4_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_sse_4hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_sse_4hv.c
endif
if WITH_REAL_AVX_BLOCK4_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_4hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_4hv.c
else
if WITH_REAL_AVX2_BLOCK4_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_4hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_4hv.c
endif
endif
if WITH_REAL_SSE_BLOCK6_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_sse_6hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_sse_6hv.c
endif
if WITH_REAL_AVX_BLOCK6_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_6hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_6hv.c
else
if WITH_REAL_AVX2_BLOCK6_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_6hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx-avx2_6hv.c
endif
endif
if WITH_COMPLEX_SSE_BLOCK1_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_sse_1hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_sse_1hv.c
endif
if WITH_COMPLEX_AVX_BLOCK1_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx-avx2_1hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx-avx2_1hv.c
else
if WITH_COMPLEX_AVX2_BLOCK1_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx-avx2_1hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx-avx2_1hv.c
endif
endif
if WITH_COMPLEX_SSE_BLOCK2_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_sse_2hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_sse_2hv.c
endif
if WITH_COMPLEX_AVX_BLOCK2_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx-avx2_2hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx-avx2_2hv.c
else
if WITH_COMPLEX_AVX2_BLOCK2_KERNEL
libelpa@SUFFIX@_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx-avx2_2hv.c
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx-avx2_2hv.c
endif
endif
......
configure.ac
View file @ ce80e103
......@@ -855,11 +855,10 @@ AC_SUBST([FC_MODOUT])
AC_SUBST([OPENMP_CFLAGS])
AC_SUBST([OPENMP_FCFLAGS])
AC_SUBST([OPENMP_LDFLAGS])
#AC_SUBST(OPT_FCFLAGS)
AC_SUBST([DOXYGEN_OUTPUT_DIR], [docs])
rm -rf modules/ .fortran_dependencies/
mkdir modules
rm -rf modules/ private_modules/ .fortran_dependencies/
mkdir modules private_modules
#gl_VISIBILITY
#AH_BOTTOM([#if HAVE_VISIBILITY
......
generated_headers.am
View file @ ce80e103
......@@ -4,10 +4,10 @@ define extract_interface
endef
elpa test:
mkdir $@
@mkdir $@
test/shared_sources: | test
mkdir $@
@mkdir $@
config-f90.h: config.h
@echo "Generating $@...";
......
src/elpa1.F90
View file @ ce80e103
......@@ -83,13 +83,7 @@
module ELPA1
use precision
use elpa_utilities
use elpa1_compute
#ifdef HAVE_DETAILED_TIMINGS
use timings
#endif
use elpa_mpi
implicit none
PRIVATE ! By default, all routines contained are private
......@@ -295,6 +289,8 @@ function solve_evp_real_1stage(na, nev, a, lda, ev, q, ldq, nblk, matrixCols, mp
#ifdef HAVE_DETAILED_TIMINGS
use timings
#endif
use elpa_mpi
use elpa1_compute
implicit none
integer(kind=ik), intent(in) :: na, nev, lda, ldq, nblk, matrixCols, mpi_comm_rows, mpi_comm_cols
......@@ -397,6 +393,8 @@ function solve_evp_complex_1stage(na, nev, a, lda, ev, q, ldq, nblk, matrixCols,
use timings
#endif
use precision
use elpa_mpi
use elpa1_compute
implicit none
integer(kind=ik), intent(in) :: na, nev, lda, ldq, nblk, matrixCols, mpi_comm_rows, mpi_comm_cols
......
src/elpa2.F90
View file @ ce80e103
......@@ -67,13 +67,9 @@ module ELPA2
! Version 1.1.2, 2011-02-21
use elpa_utilities
use elpa1_compute
use elpa1, only : elpa_print_times, time_evp_back, time_evp_fwd, time_evp_solve
use elpa2_utilities
use elpa2_compute
use elpa_pdgeqrf
use elpa_mpi
implicit none
PRIVATE ! By default, all routines contained are private
......@@ -135,6 +131,9 @@ function solve_evp_real_2stage(na, nev, a, lda, ev, q, ldq, nblk, &
#ifdef HAVE_DETAILED_TIMINGS
use timings
#endif
use elpa1_compute
use elpa2_compute
use elpa_mpi
use precision
implicit none
logical, intent(in), optional :: useQR
......@@ -375,6 +374,9 @@ function solve_evp_complex_2stage(na, nev, a, lda, ev, q, ldq, nblk, &
#ifdef HAVE_DETAILED_TIMINGS
use timings
#endif
use elpa1_compute
use elpa2_compute
use elpa_mpi
use precision
implicit none
integer(kind=ik), intent(in), optional :: THIS_COMPLEX_ELPA_KERNEL_API
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment