configure.ac 26.6 KB
Newer Older
1
AC_PREREQ([2.69])
2

3
AC_INIT([elpa],[2015.11.001], [elpa-library@mpcdf.mpg.de])
4
AC_SUBST([PACKAGE_VERSION])
5

6 7 8
AC_CONFIG_SRCDIR([src/elpa1.F90])

AM_INIT_AUTOMAKE([foreign -Wall subdir-objects])
9 10 11 12 13 14 15 16 17

# Without this, automake tries to be smart and rebuilt
# the autoconf generated files such as configure, aclocal.m4, etc.,
# in case the timestamps of files such as configure.ac are newer
#
# This only makes trouble for end users with out-of-date autoconf versions
# that cannot produce these files
AM_MAINTAINER_MODE([disable])

18 19 20 21
AC_CONFIG_MACRO_DIR([m4])
AC_CONFIG_HEADERS([config.h])
AM_SILENT_RULES([yes])

22 23
rm -rf config.h config-f90.h

24 25 26 27 28 29 30 31 32 33 34 35 36
# Set the libtool library version, see LIBRARY_INTERFACE
#
# See http://www.gnu.org/software/libtool/manual/html_node/Updating-version-info.html
#
# [c:r:a]
#
# c: The currently implement interface
# r: The revision number of the current interface, that is the number
#    of released source-code changes for the current interface
# a: The "age" is the number of interfaces prior to c that are also supported
#    by the current interface, as they are ABI compatible (e.g. only new symbols
#    were added by the new interface)
#
37
AC_SUBST([ELPA_SO_VERSION], [5:0:1])
38 39
#

40

41 42 43 44 45 46
AX_CHECK_GNU_MAKE()
if test x$_cv_gnu_make_command = x ; then
        AC_MSG_ERROR([Need GNU Make])
fi

AC_CHECK_PROG(CPP_FOUND,cpp,yes,no)
47
if test x"${CPP_FOUND}" = xno; then
48 49 50 51 52 53 54
  AC_MSG_ERROR([no cpp found])
fi

# gnu-make fortran module dependencies
m4_include([fdep/fortran_dependencies.m4])
FDEP_F90_GNU_MAKE_DEPS

55 56
###

57 58
m4_include([m4/ax_elpa_openmp.m4])

59 60
AC_MSG_CHECKING(whether --enable-openmp is specified)
AC_ARG_ENABLE([openmp],
61
		AS_HELP_STRING([--enable-openmp],
62 63 64 65 66 67 68 69 70
			       [use OpenMP threading, default no.]),
	      [],
	      [enable_openmp=no])
AC_MSG_RESULT([${enable_openmp}])
AM_CONDITIONAL([WITH_OPENMP],[test x"$enable_openmp" = x"yes"])
if test x"${enable_openmp}" = x"yes"; then
	AC_DEFINE([WITH_OPENMP], [1], [use OpenMP threading])
fi

71 72 73 74 75 76 77 78 79 80 81 82
AC_MSG_CHECKING(whether --enable-shared-memory-only is specified)
AC_ARG_ENABLE([shared-memory-only],
		AS_HELP_STRING([--enable-shared-memory-only],
			       [do not use MPI; ELPA will be build for one node shared-memory runs only]),
	      [],
	      [enable_shared_memory_only=no])
AC_MSG_RESULT([${enable_shared_memory_only}])
AM_CONDITIONAL([WITH_MPI],[test x"$enable_shared_memory_only" = x"no"])
if test x"${enable_shared_memory_only}" = x"no"; then
	AC_DEFINE([WITH_MPI], [1], [use MPI])
fi

Andreas Marek's avatar
Andreas Marek committed
83 84 85
dnl check whether mpi compilers are available;
dnl if not abort since it is mandatory

86

Andreas Marek's avatar
Andreas Marek committed
87 88
# C
AC_LANG([C])
89 90
AX_PROG_CC_MPI([test x"$enable_shared_memory_only" = xno],[use_mpi=yes],[use_mpi=no])

91
if test x"${enable_openmp}" = x"yes"; then
92 93 94 95 96
  AX_ELPA_OPENMP
  if test "$ac_cv_prog_cc_openmp" = unsupported; then
    AC_MSG_ERROR([Could not compile a C program with OpenMP, adjust CFLAGS])
  fi
  CFLAGS="$OPENMP_CFLAGS $CFLAGS"
97
fi
98

99
AC_PROG_INSTALL
100 101
AM_PROG_AR
AM_PROG_AS
Andreas Marek's avatar
Andreas Marek committed
102 103

# Fortran
104 105 106 107 108 109 110 111

dnl check whether single precision is requested
AC_MSG_CHECKING(whether ELPA library should contain also single precision functions)
AC_ARG_ENABLE(single-precision,[AS_HELP_STRING([--enable-single-precision],
	                                   [build with single precision])],
              want_single_precision="yes", want_single_precision="no")
AC_MSG_RESULT([${want_single_precision}])

Andreas Marek's avatar
Andreas Marek committed
112 113
AC_LANG([Fortran])
m4_include([m4/ax_prog_fc_mpi.m4])
114
AX_PROG_FC_MPI([test x"$enable_shared_memory_only" = xno],[use_mpi=yes],[use_mpi=no])
115
if test x"${enable_openmp}" = x"yes"; then
116 117 118 119 120
  AX_ELPA_OPENMP
  if test "$ac_cv_prog_fc_openmp" = unsupported; then
    AC_MSG_ERROR([Could not compile a Fortran program with OpenMP, adjust FCFLAGS])
  fi
  FCFLAGS="$OPENMP_FCFLAGS $FCFLAGS"
121 122
fi

Andreas Marek's avatar
Andreas Marek committed
123 124
# C++
AC_LANG([C++])
125
AC_PROG_CXX
126

127
if test x"${enable_openmp}" = x"yes"; then
128 129 130 131 132
  AX_ELPA_OPENMP
  if test "$ac_cv_prog_cxx_openmp" = unsupported; then
    AC_MSG_ERROR([Could not compile a C++ program with OpenMP, adjust CXXFLAGS])
  fi
  CXXFLAGS="$OPENMP_CXXFLAGS $CXXFLAGS"
133
fi
Andreas Marek's avatar
Andreas Marek committed
134 135 136



Andreas Marek's avatar
Andreas Marek committed
137 138
dnl variables needed for the tests

Andreas Marek's avatar
Andreas Marek committed
139 140 141 142 143 144 145 146
dnl do NOT remove any variables here, until
dnl 1. you know 100% what you are doing
dnl 2. you tested ALL configure functionality afterwards
dnl Otherwise, you most likely break some functionality

dnl as default always define the generic kernels to be build
dnl this is only unset if gpu_support_only is defined, or
dnl other specific real/complex kernels are wanted
Andreas Marek's avatar
Andreas Marek committed
147 148 149 150 151 152

install_real_generic=yes
install_real_generic_simple=yes

install_complex_generic=yes
install_complex_generic_simple=yes
153

Andreas Marek's avatar
Andreas Marek committed
154 155
AC_LANG([C])

156 157 158 159 160 161 162 163 164
dnl build with ftimings support
AC_MSG_CHECKING(whether ELPA should be build with ftimings support)
AC_ARG_WITH([ftimings],
		AS_HELP_STRING([--with-ftimings],
			       [detailed timings, default no.]),
	      [with_ftimings=yes],
	      [with_ftimings=no])
AC_MSG_RESULT([${with_ftimings}])

165 166 167 168 169 170 171 172 173 174 175 176 177 178
dnl build with the possibilty to redirect stdout and stderr
dnl per MPI task in a file
AC_MSG_CHECKING(whether stdout/stderr file redirect should be enabled)
AC_ARG_WITH([redirect],
		AS_HELP_STRING([--with-redirect],
			       [for test programs, allow redirection of stdout/stderr per MPI taks in a file (useful for ftimings), default no.]),
	      [with_redirect=yes],
	      [with_redirect=no])
AC_MSG_RESULT([${with_redirect}])

if test x"${with_redirect}" = x"yes"; then
  AC_DEFINE([HAVE_REDIRECT], [1], [Redirect stdout and stderr of test programs per MPI tasks to a file])
fi
AM_CONDITIONAL([HAVE_REDIRECT],[test x"$with_redirect" = x"yes"])
179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200

if test x"${with_ftimings}" = x"yes"; then
  AC_DEFINE([HAVE_DETAILED_TIMINGS], [1], [Enable more timings])
  AC_ARG_ENABLE([papi],
	[AS_HELP_STRING([--disable-papi],[Do not use PAPI to also measure flop count, autodetected by default])],
	[want_papi=$enableval],[want_papi="auto"])
  papi_found=unknown
  if test x"$want_papi" != x"no" ; then
    AC_CHECK_LIB([papi],[PAPI_library_init],[papi_found="yes"],[papi_found="no"])
    if test x"$want_papi" = x"yes" ; then
      if test x"$papi_found" = x"no" ; then
        AC_MSG_ERROR(["Could not find usable PAPI installation, please adjust CFLAGS, LDFLAGS"])
      fi
    fi
  fi
  if test x"$papi_found" = x"yes"; then
    AC_DEFINE([HAVE_LIBPAPI], [1], [Use the PAPI library])
    LIBS="-lpapi $LIBS"
  fi
fi
AM_CONDITIONAL([HAVE_DETAILED_TIMINGS],[test x"$with_ftimings" = x"yes"])

201
AC_MSG_CHECKING(whether double-precision SSE assembler kernel can be compiled)
Andreas Marek's avatar
Andreas Marek committed
202

203
$CC -c $srcdir/src/elpa2_kernels/elpa2_kernels_asm_x86_64_double_precision.s -o test.o 2>/dev/null
204
if test "$?" == 0; then
Andreas Marek's avatar
Andreas Marek committed
205 206 207
  can_compile_sse=yes
  install_real_sse=yes
  install_complex_sse=yes
208
else
Andreas Marek's avatar
Andreas Marek committed
209 210 211
  can_compile_sse=no
  install_real_sse=no
  install_complex_sse=no
212 213
fi
rm -f ./test.o
Andreas Marek's avatar
Andreas Marek committed
214
AC_MSG_RESULT([${can_compile_sse}])
215

216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235
if test x"${want_single_precision}" = x"yes" ; then
  AC_MSG_CHECKING(whether single-precision SSE assembler kernel can be compiled)

  $CC -c $srcdir/src/elpa2_kernels/elpa2_kernels_asm_x86_64_double_precision.s -o test.o 2>/dev/null
  if test "$?" == 0; then
    can_compile_sse=yes
    install_real_sse=yes
    install_complex_sse=yes
  else
    can_compile_sse=no
    install_real_sse=no
    install_complex_sse=no
  fi
  rm -f ./test.o
  AC_MSG_RESULT([${can_compile_sse}])
  if test x"${can_compile_sse}" = x"no" ; then
    AC_MSG_WARN([Cannot compile single-precision SSE kernel: disabling SSE kernels alltogether])
  fi
fi

Andreas Marek's avatar
Andreas Marek committed
236
dnl check whether one can compile with avx - gcc intrinsics
237

238
dnl first pass: try with specified CFLAGS and CXXFLAGS
239
AC_MSG_CHECKING([whether we can compile AVX intrinsics in C])
240 241
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
 #include <x86intrin.h>
242
 int main(int argc, char **argv){
243 244
 double* q;
 __m256d a1_1 = _mm256_load_pd(q);
245
 return 0;
246 247 248 249 250
 }
 ])],
 [can_compile_avx=yes],
 [can_compile_avx=no]
)
Andreas Marek's avatar
Andreas Marek committed
251
AC_MSG_RESULT([${can_compile_avx}])
252 253 254
if test "${can_compile_avx}" = "yes" ; then
  AC_MSG_CHECKING([whether we can compile AVX intrinsics in C++])
  AC_LANG_PUSH([C++])
255 256
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <x86intrin.h>
257
   int main(int argc, char **argv){
258 259
   double* q;
   __m256d a1_1 = _mm256_load_pd(q);
260
   return 0;
261 262 263 264 265
   }
   ])],
   [can_compile_avx=yes],
   [can_compile_avx=no]
  )
266
  AC_LANG_POP([C++])
267
  AC_MSG_RESULT([${can_compile_avx}])
268 269
  if test "${can_compile_avx}" = "no" ; then
    AC_MSG_WARN([Cannot compile C++ with AVX: disabling AVX alltogether])
270 271 272
  fi
fi

273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309
AC_MSG_CHECKING([whether we can compile AVX2 intrinsics in C])
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
 #include <x86intrin.h>
 int main(int argc, char **argv){
 double* q;
 __m256d q1 = _mm256_load_pd(q);
 __m256d y1 = _mm256_fmadd_pd(q1, q1, q1);
 return 0;
 }
 ])],
 [can_compile_avx2=yes],
 [can_compile_avx2=no]
)
AC_MSG_RESULT([${can_compile_avx2}])
if test "${can_compile_avx2}" = "yes" ; then
  AC_MSG_CHECKING([whether we can compile AVX2 intrinsics in C++])
  AC_LANG_PUSH([C++])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <x86intrin.h>
   int main(int argc, char **argv){
   double* q;
   __m256d q1 = _mm256_load_pd(q);
   __m256d y1 = _mm256_fmadd_pd(q1, q1, q1);
   return 0;
   }
   ])],
   [can_compile_avx2=yes],
   [can_compile_avx2=no]
  )
  AC_LANG_POP([C++])
  AC_MSG_RESULT([${can_compile_avx2}])
  if test "${can_compile_avx2}" = "no" ; then
    AC_MSG_WARN([Cannot compile C++ with AVX2!])
  fi
fi


310
if test "${can_compile_avx}" = "yes" ; then
Andreas Marek's avatar
Andreas Marek committed
311 312 313 314 315 316
  install_real_avx_block2=yes
  install_real_avx_block4=yes
  install_real_avx_block6=yes

  install_complex_avx_block1=yes
  install_complex_avx_block2=yes
Andreas Marek's avatar
Andreas Marek committed
317 318 319 320 321 322 323 324 325 326 327

  want_avx=yes
else
  install_real_avx_block2=no
  install_real_avx_block4=no
  install_real_avx_block6=no

  install_complex_avx_block1=no
  install_complex_avx_block2=no

  want_avx=yes
328 329 330 331 332 333 334 335 336
fi

dnl set the AVX optimization flags if this option is specified
AC_MSG_CHECKING(whether AVX optimization flags should be set automatically)
AC_ARG_WITH([avx-optimization],
		AS_HELP_STRING([--with-avx-optimization],
			       [use AVX optimization, default no.]),
	      [with_avx_optimization=yes],
	      [with_avx_optimization=no])
Andreas Marek's avatar
Andreas Marek committed
337
AC_MSG_RESULT([${with_avx_optimization}])
338
if test x"${with_avx_optimization}" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
339 340 341
 CFLAGS="$CFLAGS -funsafe-loop-optimizations -funsafe-math-optimizations -ftree-vect-loop-version -ftree-vectorize"
 CXXFLAGS="$CXXFLAGS -funsafe-loop-optimizations -funsafe-math-optimizations -ftree-vect-loop-version -ftree-vectorize"
fi
342

343
AC_LANG([Fortran])
344 345 346
AC_FC_FREEFORM
AC_FC_MODULE_FLAG
AC_FC_MODULE_OUTPUT_FLAG
347
AC_FC_LIBRARY_LDFLAGS
348 349 350 351

save_FCFLAGS=$FCFLAGS
save_LDFLAGS=$LDFLAGS

352 353 354
AC_ARG_VAR([SCALAPACK_LDFLAGS],[Extra LDFLAGS necessary to link a program with Scalapack])
AC_ARG_VAR([SCALAPACK_FCFLAGS],[Extra FCFLAGS necessary to compile a Fortran program with Scalapack])

355 356
FCFLAGS="$FCFLAGS $SCALAPACK_FCFLAGS"
LDFLAGS="$LDFLAGS $SCALAPACK_LDFLAGS"
357

358
dnl check whether fortran error_unit is defined
359
AC_MSG_CHECKING([whether Fortran module iso_fortran_env is available])
360
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
361 362 363 364 365 366 367 368 369 370
  program test_error_unit
    use ISO_FORTRAN_ENV, only : error_unit
    implicit none

    write(error_unit,*) "error_unit is defined"
  end program
])],
  [can_use_iso_fortran_env=yes],
  [can_use_iso_fortran_env=no]
)
371
AC_MSG_RESULT([${can_use_iso_fortran_env}])
372

373 374 375 376
dnl check whether one can link with specified MKL (desired method)
AC_MSG_CHECKING([whether we can compile a Fortran program using MKL])


377
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
378
  program test_mkl
379
    use mkl_service
380 381 382 383 384 385 386 387 388 389 390 391
    character*198 :: string
    call mkl_get_version_string(string)
    write(*,'(a)') string
  end program
])],
  [can_compile_with_mkl=yes],
  [can_compile_with_mkl=no]
)
AC_MSG_RESULT([${can_compile_with_mkl}])

if test x"$can_compile_with_mkl" = x"yes" ; then
  AC_MSG_CHECKING([whether we can link a Fortran program with MKL])
392
  AC_LINK_IFELSE([AC_LANG_SOURCE([
393
    program test_mkl
394
      use mkl_service
395 396 397 398 399
      character*198 :: string
      call mkl_get_version_string(string)
      write(*,'(a)') string
    end program
  ])],
Andreas Marek's avatar
Andreas Marek committed
400 401
    [have_mkl=yes],
    [have_mkl=no]
402
  )
Andreas Marek's avatar
Andreas Marek committed
403
  AC_MSG_RESULT([${have_mkl}])
404 405 406
fi

dnl if not mkl, check all the necessary individually
407
if test x"${have_mkl}" = x"yes" ; then
408 409 410 411
  WITH_MKL=1
else

  dnl first check blas
Andreas Marek's avatar
Andreas Marek committed
412 413 414
  AC_SEARCH_LIBS([dgemm],[blas],[have_blas=yes],[have_blas=no])
  AC_MSG_CHECKING([whether we can link a program with a blas lib])
  AC_MSG_RESULT([${have_blas}])
415

416
  if test x"${have_blas}" = x"no" ; then
417
    AC_MSG_ERROR([could not link with blas: specify path])
418
  fi
419
  dnl now lapack
Andreas Marek's avatar
Andreas Marek committed
420 421 422
  AC_SEARCH_LIBS([dlarrv],[lapack],[have_lapack=yes],[have_lapack=no])
  AC_MSG_CHECKING([whether we can link a program with a lapack lib])
  AC_MSG_RESULT([${have_lapack}])
423

424
  if test x"${have_lapack}" = x"no" ; then
425
    AC_MSG_ERROR([could not link with lapack: specify path])
426
  fi
Andreas Marek's avatar
Andreas Marek committed
427

428 429 430 431 432 433 434 435 436 437 438 439 440
  if test x"${enable_shared_memory_only}" = x"no"; then
    dnl test whether scalapack already contains blacs
    scalapack_libs="mpiscalapack scalapack"
    old_LIBS="$LIBS"
    for lib in ${scalapack_libs}; do
      LIBS="-l${lib} ${old_LIBS}"
      AC_MSG_CHECKING([whether -l${lib} already contains a BLACS implementation])
      AC_LINK_IFELSE([AC_LANG_FUNC_LINK_TRY([blacs_gridinit])],[blacs_in_scalapack=yes],[blacs_in_scalapack=no])
      AC_MSG_RESULT([${blacs_in_scalapack}])
      if test x"${blacs_in_scalapack}" = x"yes"; then
        break
      fi
    done
441

442 443
    if test x"${blacs_in_scalapack}" = x"no"; then
      LIBS="${old_LIBS}"
444

445 446 447
      dnl Test for stand-alone blacs
      AC_SEARCH_LIBS([bi_f77_init],[mpiblacsF77init],[],[],[-lmpiblacs])
      AC_SEARCH_LIBS([blacs_gridinit],[mpiblacs blacs],[have_blacs=yes],[have_blacs=no])
Andreas Marek's avatar
Andreas Marek committed
448

449 450 451
      if test x"${have_blacs}" = x"no"; then
        AC_MSG_ERROR([No usable BLACS found. If installed in a non-standard place, please specify suitable LDFLAGS and FCFLAGS as arguments to configure])
      fi
452
    fi
453

454
    AC_SEARCH_LIBS([pdtran],[$scalapack_libs],[have_scalapack=yes],[have_scalapack=no])
Andreas Marek's avatar
Andreas Marek committed
455

456 457 458
    if test x"${have_scalapack}" = x"no" ; then
      AC_MSG_ERROR([could not link with scalapack: specify path])
    fi
459 460 461 462
  fi

  dnl check whether we can link alltogehter
  AC_MSG_CHECKING([whether we can link a Fortran program with all blacs/scalapack])
463
  AC_LINK_IFELSE([AC_LANG_SOURCE([
464 465
    program dgemm_test

466 467
      integer , parameter:: M = 4, N = 3, K = 2
      real :: A(M,K), B(K,N), C(M,N)
468

469
      call dgemm('N','N',M,N,K,1.0,A,M,B,K,0.0,C,M)
470 471 472 473 474 475 476 477

     end program dgemm_test
     ])],
     [can_link_with_blacs_scalapack=yes],
     [can_link_with_blacs_scalapack=no]
   )
   AC_MSG_RESULT([${can_link_with_blacs_scalapack}])

478
   if test x"${can_link_with_blacs_scalapack}" = x"yes" ; then
479 480
     WITH_BLACS=1
   else
481
   AC_MSG_ERROR([We can neither link with MKL or another Scalpack. Please specify SCALAPACK_LDFLAGS and SCALAPACK_FCFLAGS!])
482 483 484
   fi
fi

485 486 487 488
dnl important: reset them again!
FCFLAGS=$save_FCFLAGS
LDFLAGS=$save_LDFLAGS

Andreas Marek's avatar
Andreas Marek committed
489 490 491 492
dnl check for intrinsic fortran function of 2003 standard
AC_MSG_CHECKING([whether we can use the intrinsic Fortran function "get_environment_variable"])


493
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
494 495 496 497 498 499 500 501 502 503 504
  program test_get_environment

    character(len=256) :: homedir
    call get_environment_variable("HOME",homedir)
  end program
])],
  [fortran_can_check_environment=yes],
  [fortran_can_check_environment=no]
)
AC_MSG_RESULT([${fortran_can_check_environment}])

505 506
dnl check whether GPU version is requested

507
#CUDA_INSTALL_PATH="/usr/local/cuda/"
508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525
#CUDA_SDK_INSTALL_PATH="/usr/local/NVIDIA_GPU_Computing_SDK"

AC_MSG_CHECKING(whether GPU support is requested)
AC_ARG_ENABLE(gpu-support,[AS_HELP_STRING([--enable-gpu-support],
	                                   [build ELPA2 with GPU-support])],
              want_gpu="yes", want_gpu="no")
AC_MSG_RESULT([${want_gpu}])


AC_ARG_WITH([cuda-path],[AS_HELP_STRING([--with-cuda-path=PATH],[prefix where CUDA is installed @<:@default=auto@:>@])],
            [CUDA_INSTALL_PATH=$withval], [with_cuda=auto])

AC_ARG_WITH([cuda-sdk-path],[AS_HELP_STRING([--with-cuda-sdk-path=PATH],[prefix where CUDA SDK is installed @<:@default=auto@:>@])],
            [CUDA_SDK_INSTALL_PATH=$withval],[with_cuda_sdk=auto])

dnl setup nvcc flags and use them in later tests
if test x"${want_gpu}" = x"yes" ; then
  AC_LANG_PUSH([C])
526
  CUDA_CFLAGS="$CUDA_CFLAGS -arch sm_35 -O2 -I$CUDA_INSTALL_PATH/include"
527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549
  LDFLAGS="$LDFLAGS -L$CUDA_INSTALL_PATH/lib64"
  NVCCFLAGS="$NVCCFLAGS $CUDA_CFLAGS $CUDA_LDFLAGS"
  NVCC="nvcc"
  AC_SUBST(NVCC)
  AC_SUBST(NVCCFLAGS)

  dnl check whether nvcc compiler is found
  AC_CHECK_PROG(nvcc_found,nvcc,yes,no)
  if test x"${nvcc_found}" = x"no" ; then
    AC_MSG_ERROR([nvcc not found; try to set the cuda-path or disable GPU support])
  fi

  dnl check whether we find cublas
  AC_SEARCH_LIBS([cublasDgemm],[cublas],[have_cublas=yes],[have_cublas=no])
  if test x"${have_cublas}" = x"no"; then
    AC_MSG_ERROR([Could not link cublas; try to set the cuda-path or disable GPU support])
  fi
  AC_SEARCH_LIBS([cudaMemcpy],[cudart],[have_cudart=yes],[have_cudart=no])
  if test x"${have_cudart}" = x"no"; then
    AC_MSG_ERROR([Could not link cudart; try to set the cuda-path or disable GPU support])
  fi
  AC_LANG_POP([C])
  install_gpu=yes
Andreas Marek's avatar
Andreas Marek committed
550
  can_compile_gpu=yes
551 552
fi

Andreas Marek's avatar
Andreas Marek committed
553 554 555 556 557 558 559 560
dnl now check which kernels can be compiled

dnl the checks for SSE were already done before
dnl the checks for AVX were already done before

dnl check BGP kernel
AC_MSG_CHECKING([whether we can compile with BGP intrinsics])

561
AC_LINK_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
562 563 564
  program test_bgp
    complex*16 :: y3,q3,h2
    y3 = fxcpmadd(y3,q3,h2)
565

Andreas Marek's avatar
Andreas Marek committed
566 567 568 569 570 571 572
  end program
])],
  [can_compile_bgp=yes],
  [can_compile_bgp=no]
)
AC_MSG_RESULT([${can_compile_bgp}])

573
if test x"${can_compile_bgp}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
574 575
  install_real_bgp=yes
  install_complex_bgp=yes
Andreas Marek's avatar
Andreas Marek committed
576 577 578
else
  install_real_bgp=no
  install_complex_bgp=no
Andreas Marek's avatar
Andreas Marek committed
579 580 581 582 583
fi

dnl check BGQ kernel
AC_MSG_CHECKING([whether we can compile with BGQ intrinsics])

584
AC_LINK_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
585
  program test_bgq
Andreas Marek's avatar
Andreas Marek committed
586 587
    VECTOR(REAL(8))::QPX_h2
    real*8         :: hh(10,2)
Andreas Marek's avatar
Andreas Marek committed
588
    QPX_h2 = VEC_SPLATS(hh(2,2))
589

Andreas Marek's avatar
Andreas Marek committed
590 591 592 593 594 595 596
  end program
])],
  [can_compile_bgq=yes],
  [can_compile_bgq=no]
)
AC_MSG_RESULT([${can_compile_bgq}])

597
if test x"${can_compile_bgq}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
598 599
  install_real_bgq=yes
  install_complex_bgq=yes
Andreas Marek's avatar
Andreas Marek committed
600 601 602
else
  install_real_bgq=no
  install_complex_bgq=no
Andreas Marek's avatar
Andreas Marek committed
603 604 605 606
fi


dnl environment variable setting of kernel
607
if test x"${fortran_can_check_environment}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
608 609 610
 AC_DEFINE([HAVE_ENVIRONMENT_CHECKING],[1],[Fortran can querry environment variables])
fi

611 612 613 614
dnl default value
use_specific_real_kernel=no
use_specific_complex_kernel=no

615 616 617
dnl GPU version only
m4_include([m4/ax_elpa_gpu_version_only.m4])
DEFINE_OPTION_GPU_SUPPORT_ONLY([gpu-version-only],[gpu-support],[install_gpu])
Andreas Marek's avatar
Andreas Marek committed
618

619 620
dnl last check whether user wants to compile only a specific kernel
dnl
621
m4_include([m4/ax_elpa_specific_kernels.m4])
Andreas Marek's avatar
Andreas Marek committed
622

623
dnl real kernels
624
dnl do not remove this variable it is needed in the macros
625

626 627
dnl generic kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-generic-kernel-only],[generic-kernel],[install_real_generic])
Andreas Marek's avatar
Andreas Marek committed
628

629 630
dnl generic-simple kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-generic-simple-kernel-only],[generic-simple-kernel],[install_real_generic_simple])
Andreas Marek's avatar
Andreas Marek committed
631

632 633
dnl sse kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-sse-kernel-only],[sse-kernel],[install_real_sse])
Andreas Marek's avatar
Andreas Marek committed
634

635 636
dnl bgp kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-bgp-kernel-only],[bgp-kernel],[install_real_bgp])
Andreas Marek's avatar
Andreas Marek committed
637

638 639
dnl bgq kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-bgq-kernel-only],[bgq-kernel],[install_real_bgq])
Andreas Marek's avatar
Andreas Marek committed
640

641 642
dnl real-avx-block2 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx-block2-kernel-only],[real-avx-block2-kernel],[install_real_avx_block2])
Andreas Marek's avatar
Andreas Marek committed
643

644 645
dnl real-avx-block4 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx-block4-kernel]-only,[real-avx-block4-kernel],[install_real_avx_block4])
646

647 648
dnl real-avx-block6 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx-block6-kernel-only],[real-avx-block6-kernel],[install_real_avx_block6])
Andreas Marek's avatar
Andreas Marek committed
649 650

dnl complex kernels
651

652 653
dnl generic kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-generic-kernel-only],[generic-kernel],[install_complex_generic])
Andreas Marek's avatar
Andreas Marek committed
654

655 656
dnl generic-simple kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-generic-simple-kernel-only],[generic-simple-kernel],[install_complex_generic_simple])
Andreas Marek's avatar
Andreas Marek committed
657

658 659
dnl sse kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-sse-kernel-only],[sse-kernel],[install_complex_sse])
Andreas Marek's avatar
Andreas Marek committed
660

661 662
dnl complex-bqp kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-bgp-kernel-only],[bgp-kernel],[install_complex_bgp])
Andreas Marek's avatar
Andreas Marek committed
663

664 665
dnl complex-bqq kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-bgq-kernel-only],[bgq-kernel],[install_complex_bgq])
Andreas Marek's avatar
Andreas Marek committed
666

667 668
dnl complex-avx-block1 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-avx-block1-kernel-only],[complex-avx-block1-kernel],[install_complex_avx_block1])
Andreas Marek's avatar
Andreas Marek committed
669

670 671
dnl complex-avx-block2 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-avx-block2-kernel-only],[complex-avx-block2-kernel],[install_complex_avx_block2])
Andreas Marek's avatar
Andreas Marek committed
672

673
dnl set the conditionals according to the previous tests
674

675
if test x"${can_use_iso_fortran_env}" = x"yes" ; then
676 677 678
 AC_DEFINE([HAVE_ISO_FORTRAN_ENV],[1],[can use module iso_fortran_env])
fi

679 680 681
AM_CONDITIONAL([WITH_GPU_VERSION],[test x"$install_gpu" = x"yes"])
if test x"${install_gpu}" = x"yes" ; then
  AC_DEFINE([WITH_GPU_VERSION],[1],[enable GPU support])
682
fi
683

Andreas Marek's avatar
Andreas Marek committed
684
AM_CONDITIONAL([WITH_REAL_GENERIC_KERNEL],[test x"$install_real_generic" = x"yes"])
685
if test x"${install_real_generic}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
686 687 688 689
 AC_DEFINE([WITH_REAL_GENERIC_KERNEL],[1],[can use real generic kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_GENERIC_KERNEL],[test x"$install_complex_generic" = x"yes"])
690
if test x"${install_complex_generic}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
691 692 693 694
 AC_DEFINE([WITH_COMPLEX_GENERIC_KERNEL],[1],[can use complex generic kernel])
fi

AM_CONDITIONAL([WITH_REAL_GENERIC_SIMPLE_KERNEL],[test x"$install_real_generic_simple" = x"yes"])
695
if test x"${install_real_generic_simple}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
696 697 698 699
 AC_DEFINE([WITH_REAL_GENERIC_SIMPLE_KERNEL],[1],[can use real generic-simple kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_GENERIC_SIMPLE_KERNEL],[test x"$install_complex_generic_simple" = x"yes"])
700
if test x"${install_complex_generic_simple}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
701 702 703 704
 AC_DEFINE([WITH_COMPLEX_GENERIC_SIMPLE_KERNEL],[1],[can use complex generic-simple kernel])
fi

AM_CONDITIONAL([WITH_REAL_SSE_KERNEL],[test x"$install_real_sse" = x"yes"])
705
if test x"${install_real_sse}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
706 707 708 709
 AC_DEFINE([WITH_REAL_SSE_KERNEL],[1],[can use real SSE kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_SSE_KERNEL],[test x"$install_complex_sse" = x"yes"])
710
if test x"${install_complex_sse}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
711 712 713 714
 AC_DEFINE([WITH_COMPLEX_SSE_KERNEL],[1],[can use complex SSE kernel])
fi

AM_CONDITIONAL([WITH_REAL_AVX_BLOCK2_KERNEL],[test x"$install_real_avx_block2" = x"yes"])
715
if test x"${install_real_avx_block2}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
716 717 718 719
 AC_DEFINE([WITH_REAL_AVX_BLOCK2_KERNEL],[1],[can use real_avx_block2 kernel])
fi

AM_CONDITIONAL([WITH_REAL_AVX_BLOCK4_KERNEL],[test x"$install_real_avx_block4" = x"yes"])
720
if test x"${install_real_avx_block4}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
721 722 723 724
 AC_DEFINE([WITH_REAL_AVX_BLOCK4_KERNEL],[1],[can use real_avx_block4 kernel])
fi

AM_CONDITIONAL([WITH_REAL_AVX_BLOCK6_KERNEL],[test x"$install_real_avx_block6" = x"yes"])
725
if test x"${install_real_avx_block6}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
726 727 728 729
 AC_DEFINE([WITH_REAL_AVX_BLOCK6_KERNEL],[1],[can use real_avx_block6 kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_AVX_BLOCK1_KERNEL],[test x"$install_complex_avx_block1" = x"yes"])
730
if test x"${install_complex_avx_block1}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
731 732 733 734
 AC_DEFINE([WITH_COMPLEX_AVX_BLOCK1_KERNEL],[1],[can use complex_avx_block1 kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_AVX_BLOCK2_KERNEL],[test x"$install_complex_avx_block2" = x"yes"])
735
if test x"${install_complex_avx_block2}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
736 737 738 739
 AC_DEFINE([WITH_COMPLEX_AVX_BLOCK2_KERNEL],[1],[can use complex_avx_block2 kernel])
fi

AM_CONDITIONAL([WITH_REAL_BGP_KERNEL],[test x"$install_real_bgp" = x"yes"])
740
if test x"${install_real_bgp}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
741 742 743 744
 AC_DEFINE([WITH_REAL_BGP_KERNEL],[1],[can use real BGP kernel])
fi

AM_CONDITIONAL([WITH_REAL_BGQ_KERNEL],[test x"$install_real_bgq" = x"yes"])
745
if test x"${install_real_bgq}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
746 747 748
 AC_DEFINE([WITH_REAL_BGQ_KERNEL],[1],[can use real BGQ kernel])
fi

749
if test x"${use_specific_complex_kernel}" = x"no" ; then
750
	AC_DEFINE([WITH_NO_SPECIFIC_COMPLEX_KERNEL],[1],[do not use only one specific complex kernel (set at compile time)])
Andreas Marek's avatar
Andreas Marek committed
751 752
fi

753
if test x"${use_specific_real_kernel}" = x"no" ; then
754
  AC_DEFINE([WITH_NO_SPECIFIC_REAL_KERNEL],[1],[do not use only one specific real kernel (set at compile time)])
Andreas Marek's avatar
Andreas Marek committed
755 756
fi

757 758
LT_INIT

759 760 761 762 763 764
DX_PDF_FEATURE(OFF)
DX_PS_FEATURE(OFF)
DX_MAN_FEATURE(ON)
DX_HTML_FEATURE(ON)
DX_INIT_DOXYGEN([ELPA], [Doxyfile], [docs])

765
DESPERATELY_WANT_ASSUMED_SIZE=0
766
if test x"${DESPERATELY_WANT_ASSUMED_SIZE}" = x"yes" ; then
767 768 769
  AC_DEFINE([DESPERATELY_WANT_ASSUMED_SIZE],[1],[use assumed size arrays, even if not debuggable])
fi

770 771 772
if test x"${want_single_precision}" = x"yes" ; then
  AC_DEFINE([WANT_SINGLE_PRECISION_REAL],[1],[build also single-precision for real calculation])
  AC_DEFINE([WANT_SINGLE_PRECISION_COMPLEX],[1],[build also single-precision for complex calculation])
773
fi
774 775
AM_CONDITIONAL([WANT_SINGLE_PRECISION_REAL],[test x"$want_single_precision" = x"yes"])
AM_CONDITIONAL([WANT_SINGLE_PRECISION_COMPLEX],[test x"$want_single_precision" = x"yes"])
776

777 778 779 780 781
AC_SUBST([WITH_MKL])
AC_SUBST([WITH_BLACS])
AC_SUBST([with_amd_bulldozer_kernel])
AC_SUBST([FC_MODINC])
AC_SUBST([FC_MODOUT])
782 783 784
AC_SUBST([OPENMP_CFLAGS])
AC_SUBST([OPENMP_FCFLAGS])
AC_SUBST([OPENMP_LDFLAGS])
785
#AC_SUBST(OPT_FCFLAGS)
786
AC_SUBST([DOXYGEN_OUTPUT_DIR], [docs])
787

788
rm -rf modules/ .fortran_dependencies/
789 790 791 792 793 794 795 796 797 798 799
mkdir modules

#gl_VISIBILITY
#AH_BOTTOM([#if HAVE_VISIBILITY
#define EXPORTED __attribute__((__visibility__("default")))
#define HIDDEN   __attribute__((__visibility__("hidden")))
#else
#define EXPORTED
#define HIDDEN
#endif])

800 801 802 803 804 805 806 807 808

# Some part of libtool is too smart and tries to parse the output of
#   gfortran -v
# and catches anything that resembles a -l library link specification.
# Unfortunately, recent versions of gfortran emit
#   -l gfortran
# with a space between -l and gfortran. The empty -l is then included
# into "postdeps_FC" and causes linking errors later on.
postdeps_FC=$(echo $postdeps_FC | sed 's/-l //g')
809

810
if test x"${enable_openmp}" = x"yes"; then
811
	SUFFIX="_openmp"
812
else
813
	SUFFIX=""
814
fi
815 816
AC_SUBST([SUFFIX])
AC_SUBST([PKG_CONFIG_FILE],[elpa${SUFFIX}-${PACKAGE_VERSION}.pc])
817

818 819
AC_CONFIG_FILES([
  Makefile
820
  Doxyfile
821
  ${PKG_CONFIG_FILE}:elpa.pc.in
822
])
823

824
AC_OUTPUT
825 826 827 828 829 830


if test -e config.h ; then
  grep "^#define" config.h > config-f90.h
else
  echo "Warning! No config.h was generated, check for errors and re-run configure!"
Andreas Marek's avatar
Andreas Marek committed
831
  exit 1
832
fi
Andreas Marek's avatar
Andreas Marek committed
833 834 835

echo "Generating elpa/elpa_generated.h..."
mkdir -p elpa
836 837 838 839 840
grep -h "^ *!c>" $srcdir/src/elpa_c_interface.F90 | sed 's/^ *!c>//;' > elpa/elpa_generated.h || exit 1

echo "Generating test/shared_sources/generated.h..."
mkdir -p test/shared_sources
grep -h "^ *!c>" $srcdir/test/shared_sources/*.F90 | sed 's/^ *!c>//;' > test/shared_sources/generated.h || exit 1
Andreas Marek's avatar
Andreas Marek committed
841

842
if test "${can_compile_avx}" = "no" ; then
Andreas Marek's avatar
Andreas Marek committed
843 844 845
  if test x"${want_avx}" = x"yes" ; then
    AC_MSG_WARN([Could not compile AVX instructions])
  fi
846
fi
847 848 849 850 851
if test "${can_compile_avx2}" = "no" ; then
  if test x"${want_avx}" = x"yes" ; then
    AC_MSG_WARN([Could not compile AVX2 instructions])
  fi
fi
852 853 854 855

if test "${can_compile_sse}" = "no" ; then
  AC_MSG_WARN([Could not compile SSE instructions])
fi