configure.ac 33.1 KB
Newer Older
1
AC_PREREQ([2.69])
2

3 4
# The version is set in elpa.spec, to have a single point of reference
AC_INIT([elpa],m4_esyscmd_s([awk '/^ *Version:/ {print $2;}' elpa.spec]), [elpa-library@mpcdf.mpg.de])
5
AC_SUBST([PACKAGE_VERSION])
6

Andreas Marek's avatar
Andreas Marek committed
7
AC_CONFIG_SRCDIR([src/elpa_t.F90])
8 9

AM_INIT_AUTOMAKE([foreign -Wall subdir-objects])
10

11 12 13 14
AC_CONFIG_MACRO_DIR([m4])
AC_CONFIG_HEADERS([config.h])
AM_SILENT_RULES([yes])

15 16 17 18 19 20 21 22 23 24 25 26 27
# 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)
#
28
AC_SUBST([ELPA_SO_VERSION], [8:0:0])
29 30
#

31

32 33 34 35 36
AX_CHECK_GNU_MAKE()
if test x$_cv_gnu_make_command = x ; then
        AC_MSG_ERROR([Need GNU Make])
fi

Andreas Marek's avatar
Andreas Marek committed
37 38 39 40
#AC_CHECK_PROG(CPP_FOUND,cpp,yes,no)
#if test x"${CPP_FOUND}" = xno; then
#  AC_MSG_ERROR([no cpp found])
#fi
41 42 43 44 45

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

46 47
###

48 49
m4_include([m4/ax_elpa_openmp.m4])

50 51
AC_MSG_CHECKING(whether --enable-openmp is specified)
AC_ARG_ENABLE([openmp],
52 53 54 55
              AS_HELP_STRING([--enable-openmp],
                             [use OpenMP threading, default no.]),
              [],
              [enable_openmp=no])
56 57 58
AC_MSG_RESULT([${enable_openmp}])
AM_CONDITIONAL([WITH_OPENMP],[test x"$enable_openmp" = x"yes"])
if test x"${enable_openmp}" = x"yes"; then
59
        AC_DEFINE([WITH_OPENMP], [1], [use OpenMP threading])
60 61
fi

62 63
dnl mpi
AC_ARG_WITH(mpi, [AS_HELP_STRING([--with-mpi=[[yes|no]]], [compile with MPI. Default: yes])],,[with_mpi=yes])
64
AM_CONDITIONAL([WITH_MPI],[test x"$with_mpi" = x"yes"])
65
if test x"${with_mpi}" = x"yes"; then
66
  AC_DEFINE([WITH_MPI], [1], [use MPI])
67 68
fi

Andreas Marek's avatar
Andreas Marek committed
69
# C
70
AC_LANG_PUSH([C])
71 72 73 74 75 76
AX_PROG_CC_MPI([test x"$with_mpi" = x"yes"],[found_mpi_c=yes],[found_mpi_c=no])
if test x"$with_mpi" = x"yes"; then
  if test x"$found_mpi_c" = x"no"; then
    AC_MSG_ERROR([Could not compile an MPI C program])
  fi
fi
77

78
if test x"${enable_openmp}" = x"yes"; then
79 80 81 82 83
  AX_ELPA_OPENMP
  if test "$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"
84
fi
85

86
AC_PROG_INSTALL
87 88
AM_PROG_AR
AM_PROG_AS
89
AC_PROG_CC_C99
90
AM_PROG_CC_C_O
91
AC_LANG_POP([C])
Andreas Marek's avatar
Andreas Marek committed
92 93

# Fortran
94
AC_LANG_PUSH([Fortran])
Andreas Marek's avatar
Andreas Marek committed
95
m4_include([m4/ax_prog_fc_mpi.m4])
96 97 98 99 100 101
AX_PROG_FC_MPI([test x"$with_mpi" = x"yes"],[found_mpi_f=yes],[found_mpi_f=no])
if test x"$with_mpi" = x"yes"; then
  if test x"$found_mpi_f" = x"no"; then
    AC_MSG_ERROR([Could not compile an MPI Fortran program])
  fi
fi
102

103
AC_FC_SRCEXT([F90])
104 105 106 107 108
AC_FC_FREEFORM
AC_FC_MODULE_FLAG
AC_FC_MODULE_OUTPUT_FLAG
AC_FC_LIBRARY_LDFLAGS

109

110
if test x"${enable_openmp}" = x"yes"; then
111 112 113 114 115
  AX_ELPA_OPENMP
  if test "$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"
116 117
fi

118 119
dnl check which MPI binray invokes a MPI job
if test x"$with_mpi" = x"yes"; then
120
  AC_CHECK_PROGS([MPI_BINARY], [srun mpiexec.hydra mpiexec mpirun poe runjob], [no])
121
  if test x"$MPI_BINARY" = x"no"; then
122
    AC_MSG_ERROR([Could not find either of the MPI binaries: mpiexec.hydra, mpiexec, mpirun, poe, runjob, srun])
123 124 125
  fi
fi

126 127 128
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)
129
AC_ARG_ENABLE([redirect],
130 131 132 133
              [AS_HELP_STRING([--enable-redirect],
                              [for test programs, allow redirection of stdout/stderr per MPI taks in a file (useful for timing), default no.])],
              [],
              [enable_redirect=no])
134 135
AC_MSG_RESULT([${enable_redirect}])

136 137 138 139 140 141 142 143 144 145
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="$enableval"],
              [want_single_precision="no"])
AC_MSG_RESULT([${want_single_precision}])


146 147
dnl redirect
if test x"${enable_redirect}" = x"yes"; then
148 149
  AC_DEFINE([HAVE_REDIRECT], [1], [Redirect stdout and stderr of test programs per MPI tasks to a file])
fi
150 151 152 153
AM_CONDITIONAL([HAVE_REDIRECT],[test x"$enable_redirect" = x"yes"])

dnl build with ftimings support
AC_MSG_CHECKING(whether ELPA should be build with more detailed timing support)
Andreas Marek's avatar
Andreas Marek committed
154
AC_ARG_ENABLE([timings],
155 156 157 158
              [AS_HELP_STRING([--enable-timings],
                              [more detailed timing, default no.])],
              [enable_timings="$enableval"],
              [enable_timings=no])
159 160 161 162 163
AC_MSG_RESULT([${enable_timings}])

if test x"${enable_timings}" = x"yes"; then
  AC_DEFINE([HAVE_DETAILED_TIMINGS], [1], [Enable more timing])
fi
164
AM_CONDITIONAL([HAVE_DETAILED_TIMINGS], [test x"$enable_timings" = x"yes"])
165 166 167

dnl PAPI for ftimings
AC_ARG_WITH([papi],
168 169 170 171
            [AS_HELP_STRING([--with-papi],
                            [Use PAPI to also measure flop count in the detailed timing (--enable-timing), disabled by default])],
            [],
            [with_papi="no"])
172 173
if test x"${enable_timings}" = x"yes"; then
  if test x"$with_papi" = x"yes" ; then
174
    AC_SEARCH_LIBS([PAPI_library_init], [papi], [papi_found=yes], [papi_found=no])
175 176
    if test x"$papi_found" = x"no" ; then
      AC_MSG_ERROR(["Could not find usable PAPI installation, please install or adjust CFLAGS, LDFLAGS"])
177 178 179 180 181
    fi
    AC_DEFINE([HAVE_LIBPAPI], [1], [Use the PAPI library])
  fi
fi

182 183 184
save_FCFLAGS=$FCFLAGS
save_LDFLAGS=$LDFLAGS

185 186 187
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])

188 189
FCFLAGS="$FCFLAGS $SCALAPACK_FCFLAGS"
LDFLAGS="$LDFLAGS $SCALAPACK_LDFLAGS"
190

191
dnl check whether fortran error_unit is defined
192
AC_MSG_CHECKING([whether Fortran module iso_fortran_env is available])
193
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
194
  program test_error_unit
195
    use iso_fortran_env, only : error_unit
196 197 198 199 200 201 202 203
    implicit none

    write(error_unit,*) "error_unit is defined"
  end program
])],
  [can_use_iso_fortran_env=yes],
  [can_use_iso_fortran_env=no]
)
204
AC_MSG_RESULT([${can_use_iso_fortran_env}])
205 206 207 208
if test x"${can_use_iso_fortran_env}" = x"yes" ; then
 AC_DEFINE([HAVE_ISO_FORTRAN_ENV],[1],[can use module iso_fortran_env])
fi

209

210 211
dnl check whether one can link with specified MKL (desired method)
AC_MSG_CHECKING([whether we can compile a Fortran program using MKL])
212
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
213
  program test_mkl
214
    use mkl_service
215 216 217 218 219 220 221 222 223 224 225 226
    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"$can_compile_with_mkl" = x"yes" ; then
  AC_MSG_CHECKING([whether we can link a Fortran program with MKL])
227
  AC_LINK_IFELSE([AC_LANG_SOURCE([
228
    program test_mkl
229
      use mkl_service
230 231 232 233 234
      character*198 :: string
      call mkl_get_version_string(string)
      write(*,'(a)') string
    end program
  ])],
Andreas Marek's avatar
Andreas Marek committed
235 236
    [have_mkl=yes],
    [have_mkl=no]
237
  )
Andreas Marek's avatar
Andreas Marek committed
238
  AC_MSG_RESULT([${have_mkl}])
239 240 241
fi

dnl if not mkl, check all the necessary individually
242
if test x"${have_mkl}" = x"yes" ; then
243 244 245 246
  WITH_MKL=1
else

  dnl first check blas
247
  AC_SEARCH_LIBS([dgemm],[openblas satlas blas],[have_blas=yes],[have_blas=no])
Andreas Marek's avatar
Andreas Marek committed
248 249
  AC_MSG_CHECKING([whether we can link a program with a blas lib])
  AC_MSG_RESULT([${have_blas}])
250

251
  if test x"${have_blas}" = x"no" ; then
252
    AC_MSG_ERROR([could not link with blas: specify path])
253
  fi
254
  dnl now lapack
Andreas Marek's avatar
Andreas Marek committed
255 256 257
  AC_SEARCH_LIBS([dlarrv],[lapack],[have_lapack=yes],[have_lapack=no])
  AC_MSG_CHECKING([whether we can link a program with a lapack lib])
  AC_MSG_RESULT([${have_lapack}])
258

259
  if test x"${have_lapack}" = x"no" ; then
260
    AC_MSG_ERROR([could not link with lapack: specify path])
261
  fi
Andreas Marek's avatar
Andreas Marek committed
262

263
  if test x"${with_mpi}" = x"yes"; then
264
    dnl test whether scalapack already contains blacs
265
    scalapack_libs="mpiscalapack scalapack scalapack-openmpi"
266 267 268 269 270 271 272 273 274 275
    old_LIBS="$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
276

277 278
    if test x"${blacs_in_scalapack}" = x"no"; then
      LIBS="${old_LIBS}"
279

280 281
      dnl Test for stand-alone blacs
      AC_SEARCH_LIBS([bi_f77_init],[mpiblacsF77init],[],[],[-lmpiblacs])
282 283 284 285 286 287

      dnl for debian
      AC_SEARCH_LIBS([blacs_gridinit],[blacs-openmpi],[have_blacs=yes],[have_blacs=no],[-lblacsCinit-openmpi -lscalapack-openmpi])
      if test x"${have_blacs}" = x"no"; then
        unset ac_cv_search_blacs_gridinit
      fi
288
      AC_SEARCH_LIBS([blacs_gridinit],[mpiblacs blacs],[have_blacs=yes],[have_blacs=no])
Andreas Marek's avatar
Andreas Marek committed
289

290 291 292
      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
293
    fi
294

295
    AC_SEARCH_LIBS([pdtran],[$scalapack_libs],[have_scalapack=yes],[have_scalapack=no])
Andreas Marek's avatar
Andreas Marek committed
296

297 298 299
    if test x"${have_scalapack}" = x"no" ; then
      AC_MSG_ERROR([could not link with scalapack: specify path])
    fi
300 301 302 303
  fi

  dnl check whether we can link alltogehter
  AC_MSG_CHECKING([whether we can link a Fortran program with all blacs/scalapack])
304
  AC_LINK_IFELSE([AC_LANG_SOURCE([
305 306
    program dgemm_test

307 308
      integer , parameter:: M = 4, N = 3, K = 2
      real :: A(M,K), B(K,N), C(M,N)
309

310
      call dgemm('N','N',M,N,K,1.0,A,M,B,K,0.0,C,M)
311 312 313 314 315 316 317 318

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

319
   if test x"${can_link_with_blacs_scalapack}" = x"yes" ; then
320 321
     WITH_BLACS=1
   else
322
   AC_MSG_ERROR([We can neither link with MKL or another Scalpack. Please specify SCALAPACK_LDFLAGS and SCALAPACK_FCFLAGS!])
323 324 325
   fi
fi

326 327 328 329
dnl important: reset them again!
FCFLAGS=$save_FCFLAGS
LDFLAGS=$save_LDFLAGS

330

Andreas Marek's avatar
Andreas Marek committed
331 332 333 334
dnl check for intrinsic fortran function of 2003 standard
AC_MSG_CHECKING([whether we can use the intrinsic Fortran function "get_environment_variable"])


335
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
336 337 338 339 340 341 342 343 344
  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([${fortran_can_check_environment}])
345 346 347
if test x"${fortran_can_check_environment}" = x"yes" ; then
 AC_DEFINE([HAVE_ENVIRONMENT_CHECKING],[1],[Fortran can query environment variables])
fi
Andreas Marek's avatar
Andreas Marek committed
348

349

350
dnl check whether BAND_TO_FULL_BLOCKING is set
351
use_band_to_full_blocking=yes
352 353
AC_MSG_CHECKING(whether BAND_TO_FLULL_BLOCKING is requested)
AC_ARG_ENABLE(band-to-full-blocking,[AS_HELP_STRING([--enable-band-to-full-blocking],
354 355
                                                    [build ELPA2 with blocking in band_to_full (default: enabled)])],
              want_band_to_full_blocking="yes", want_to_full_blocking="no")
356
AC_MSG_RESULT([${want_band_to_full_blocking}])
357 358 359 360 361 362 363 364

if test x"${enableval}" = x"no" ; then
  use_band_to_full_blocking=no
fi

AM_CONDITIONAL([BAND_TO_FULL_BLOCKING],[test x"$use_band_to_full_blocking" = x"yes"])
if test x"${use_band_to_full_blocking}" = x"yes"; then
       AC_DEFINE([BAND_TO_FULL_BLOCKING], [1], [use blocking in trans_ev_band_to_full])
365 366
fi

367 368 369 370 371 372 373

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])

374

375
dnl setup nvcc flags and use them in later tests
376 377 378
user_sets_gpu_compute_capability="no"
AC_MSG_CHECKING(whether a GPU compute capability is specified)
AC_ARG_WITH([GPU-compute-capability],
379 380 381
            [AS_HELP_STRING([--with-GPU-compute-capability=VALUE],
                            [use compute capability VALUE for GPU version, default: "sm_35"])],
            [user_sets_gpu_compute_capability="yes"],[cuda_compute_capability="sm_35"])
382 383
AC_MSG_RESULT([${user_sets_gpu_compute_capability}])

384

385 386 387 388 389 390 391 392 393 394 395
dnl sanity check whether compute capability setting by user is reasonable
if test x"${user_sets_gpu_compute_capability}" = x"yes" ; then
  dnl the user must set a value which starts with "sm_"
  value=$(echo $withval | cut -c1-3)
  if test x"${value}" = x"sm_" ; then
    cuda_compute_capability=$withval
  else
    AC_MSG_ERROR([Unknown GPU compute capability set: ${withval}])
  fi
fi

Andreas Marek's avatar
Andreas Marek committed
396

397 398
dnl Test possibility of 'use mpi', if requested
if test x"${with_mpi}" = x"yes" ; then
399 400
  AC_ARG_ENABLE([mpi-module],
                AS_HELP_STRING([--disable-mpi-module],
401
                               [do not use the Fortran MPI module, get interfaces by 'include "mpif.h')]),
402
                [],
403 404
                [enable_mpi_module=yes])
  if test x"${enable_mpi_module}" = x"yes" ; then
405 406 407 408 409 410 411 412 413 414 415 416 417 418 419
    AC_MSG_CHECKING(whether Fortran mpi module can be used)
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
      program test_mpi_module
        use mpi
        real :: time
        time = MPI_WTime()
      end program
    ])],
      [can_use_fortran_mpi_module=yes],
      [can_use_fortran_mpi_module=no]
    )
    AC_MSG_RESULT([${can_use_fortran_mpi_module}])
    if test x"${can_use_fortran_mpi_module}" = x"yes" ; then
      AC_DEFINE([HAVE_MPI_MODULE],[1],[can use the Fortran mpi module])
    else
420
      AC_MSG_ERROR([Could not compile a Fortran program with an 'use mpi' statement. You can try again with --disable-mpi-module])
421
    fi
422
  fi
Andreas Marek's avatar
Andreas Marek committed
423
fi
424
AC_LANG_POP([Fortran])
Andreas Marek's avatar
Andreas Marek committed
425

426 427
dnl Assemble the list of kernels to build
m4_pattern_forbid([elpa_m4])
Andreas Marek's avatar
Andreas Marek committed
428

429 430 431 432 433 434
m4_define(elpa_m4_generic_kernels, [
        real_generic
        real_generic_simple
        complex_generic
        complex_generic_simple
])
435

436 437 438 439
m4_define(elpa_m4_sse_assembly_kernels, [
        real_sse_assembly
        complex_sse_assembly
])
440

441 442 443 444 445 446 447
m4_define(elpa_m4_sse_kernels, [
        real_sse_block2
        real_sse_block4
        real_sse_block6
        complex_sse_block1
        complex_sse_block2
])
448

449 450 451 452 453 454 455
m4_define(elpa_m4_avx_kernels, [
        real_avx_block2
        real_avx_block4
        real_avx_block6
        complex_avx_block1
        complex_avx_block2
])
Andreas Marek's avatar
Andreas Marek committed
456

457 458 459 460 461 462 463
m4_define(elpa_m4_avx2_kernels, [
        real_avx2_block2
        real_avx2_block4
        real_avx2_block6
        complex_avx2_block1
        complex_avx2_block2
])
Andreas Marek's avatar
Andreas Marek committed
464

465 466 467 468 469 470 471
m4_define(elpa_m4_avx512_kernels, [
        real_avx512_block2
        real_avx512_block4
        real_avx512_block6
        complex_avx512_block1
        complex_avx512_block2
])
Andreas Marek's avatar
Andreas Marek committed
472

473 474 475 476
m4_define(elpa_m4_bgp_kernels, [
        real_bgp
        complex_bgp
])
Andreas Marek's avatar
Andreas Marek committed
477

478 479 480 481
m4_define(elpa_m4_bgq_kernels, [
        real_bgq
        complex_bgq
])
Andreas Marek's avatar
Andreas Marek committed
482

483 484 485
m4_define(elpa_m4_gpu_kernels, [
        gpu
])
486

487
m4_define(elpa_m4_kernel_types, [generic sse sse_assembly avx avx2 avx512 bgp bgq gpu])
488

489 490 491 492
m4_define(elpa_m4_all_kernels,
          m4_foreach_w([elpa_m4_type],
                       elpa_m4_kernel_types,
                       [m4_foreach_w([elpa_m4_kernel],[elpa_m4_]elpa_m4_type[_kernels],elpa_m4_kernel )]))
Andreas Marek's avatar
Andreas Marek committed
493

494 495 496 497
m4_define(elpa_m4_real_kernels,
          m4_foreach_w(elpa_m4_kernel,
                       elpa_m4_all_kernels,
                       [m4_bmatch(elpa_m4_kernel,real,elpa_m4_kernel) ]))
Andreas Marek's avatar
Andreas Marek committed
498

499 500 501 502
m4_define(elpa_m4_complex_kernels,
          m4_foreach_w(elpa_m4_kernel,
                       elpa_m4_all_kernels,
                       [m4_bmatch(elpa_m4_kernel,complex,elpa_m4_kernel) ]))
503

504 505 506 507 508 509 510 511 512 513 514 515 516
dnl
dnl ELPA_SELECT_KERNELS([flagname],[default])
dnl
dnl  default should be one of `enable' or `disable'
dnl
AC_DEFUN([ELPA_SELECT_KERNELS], [
        AC_ARG_ENABLE(m4_bpatsubst($1,[_],[-]),[AS_HELP_STRING([--]m4_case([$2],[enable],[disable],[disable],[enable])[-]m4_bpatsubst($1,[_],[-]),
                m4_case([$2],[enable],[do not build],[disable],[build])[ ]m4_toupper($1)[ kernels, default: $2d])],
                      [],[enable_$1=]m4_case([$2],[enable],[yes],[disable],[no]))
        m4_foreach_w(elpa_m4_kernel,elpa_m4_$1_kernels,[
                use_[]elpa_m4_kernel[=$enable_$1]
        ])
])
517

518 519 520 521 522 523 524 525 526 527 528 529 530 531
dnl Modify list of kernels with configure arguments
ELPA_SELECT_KERNELS([generic],[enable])
ELPA_SELECT_KERNELS([sse],[enable])
ELPA_SELECT_KERNELS([sse_assembly],[enable])
ELPA_SELECT_KERNELS([avx],[enable])
ELPA_SELECT_KERNELS([avx2],[enable])
ELPA_SELECT_KERNELS([avx512],[disable])
ELPA_SELECT_KERNELS([gpu],[disable])
ELPA_SELECT_KERNELS([bgp],[disable])
ELPA_SELECT_KERNELS([bgq],[disable])

m4_foreach_w([elpa_m4_kind],[real complex],[
        default_[]elpa_m4_kind[]_kernel=""
])
532

533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560
if test x"${enable_bgp}" = x"yes" -o x"$enable_bgq" = x"yes"; then
        m4_foreach_w([elpa_m4_kernel], elpa_m4_sse_kernels elpa_m4_avx_kernels elpa_m4_avx2_kernels elpa_m4_avx512_kernels, [
                if x"$use_[]elpa_m4_kernel[]" = x"yes" ; then
                        echo "Disabling elpa_m4_kernel due to BGP/BGQ option"
                fi
                use_[]elpa_m4_kernel[]=no
        ])
fi


dnl Deal with --with-fixed-[real|complex]-kernel arguments
m4_foreach_w([elpa_m4_kind],[real complex],[
        AC_ARG_WITH([fixed-]elpa_m4_kind[-kernel], m4_expand([AS_HELP_STRING([--with-fixed-]elpa_m4_kind[-kernel]=KERNEL,
                [compile with only a single specific ]elpa_m4_kind[ kernel. Available kernels are:]
                 m4_expand(elpa_m4_[]elpa_m4_kind[]_kernels))]),
                 [fixed_]elpa_m4_kind[_kernel="$withval"],[fixed_]elpa_m4_kind[_kernel=""])
        if test -n "$fixed_[]elpa_m4_kind[]_kernel" ; then
                m4_foreach_w([elpa_m4_otherkernel],m4_expand(elpa_m4_[]elpa_m4_kind[]_kernels),[
                        if test "$fixed_]elpa_m4_kind[_kernel" = "]elpa_m4_otherkernel[" ; then
                                use_[]elpa_m4_otherkernel[]=yes
                        else
                                use_[]elpa_m4_otherkernel[]=no
                        fi
                ])
                default_[]elpa_m4_kind[]_kernel="$fixed_[]elpa_m4_kind[]_kernel"
                AC_DEFINE([WITH_FIXED_]m4_toupper(elpa_m4_kind)[_KERNEL],[1],[use only one specific ]elpa_m4_kind[ kernel (set at compile time)])
        fi
])
561

562 563 564 565 566 567 568 569 570
AC_ARG_WITH(gpu-support-only, [AS_HELP_STRING([--with-gpu-support-only],
            [Compile and always use the GPU version])],
            [],[with_gpu_support_only=no])
if test x"$with_gpu_support_only" = x"yes" ; then
        m4_foreach_w([elpa_m4_kernel],elpa_m4_all_kernels,[
                      use_[]elpa_m4_kernel[]=no
        ])
        use_gpu=yes
fi
571

572

573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592
dnl
dnl  ELPA_KERNEL_DEPENDS([kernel],[other kernels])
dnl
dnl   Switch on each of the other kernels if the given kernel is selected
dnl
AC_DEFUN([ELPA_KERNEL_DEPENDS],[
          if test x"$use_$1" = x"yes"; then
                  m4_foreach_w([elpa_m4_requiredkernel],[$2],[
                               if test x"$use_[]elpa_m4_requiredkernel[]" = x"no" ; then
                                       echo "Enabling elpa_m4_requiredkernel kernel, is a prerequisite for $1"
                               fi
                               use_[]elpa_m4_requiredkernel[]=yes
                  ])
          fi
])
m4_foreach_w([elpa_m4_arch],[sse avx avx2 avx512],[
        ELPA_KERNEL_DEPENDS([real_]elpa_m4_arch[_block6], [real_]elpa_m4_arch[_block4 real_]elpa_m4_arch[_block2])
        ELPA_KERNEL_DEPENDS([real_]elpa_m4_arch[_block4], [real_]elpa_m4_arch[_block2])
        ELPA_KERNEL_DEPENDS([complex_]elpa_m4_arch[_block2], [complex_]elpa_m4_arch[_block1])
])
593

594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613
m4_foreach_w([elpa_m4_type],elpa_m4_kernel_types,[
        need_[]elpa_m4_type=no
        need_[]elpa_m4_type[]_kernels=""
        m4_foreach_w([elpa_m4_kernel],m4_expand([elpa_m4_]elpa_m4_type[_kernels]),[
                if test x"$use_[]elpa_m4_kernel" = x"yes" ; then
                        need_[]elpa_m4_type=yes
                        need_[]elpa_m4_type[]_kernels="$need_[]elpa_m4_type[]_kernels elpa_m4_kernel"
                fi
        ])
])
m4_foreach_w([elpa_m4_type],elpa_m4_kernel_types,[
        if test x"$need_[]elpa_m4_type[]" = x"yes" ; then
               echo "Using m4_toupper(elpa_m4_type) for kernels$need_[]elpa_m4_type[]_kernels"
        fi
])
m4_foreach_w([elpa_m4_type],elpa_m4_kernel_types,[
        if test x"$need_[]elpa_m4_type[]" != x"yes" ; then
               echo "Not using m4_toupper(elpa_m4_type) as no selected kernel needs it"
        fi
])
614

615

616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638
dnl the list of kernels is now assembled
dnl choosing a default kernel

m4_foreach_w([elpa_m4_kind],[real complex],[
        m4_foreach_w([elpa_m4_kernel],
                     m4_foreach_w([elpa_m4_cand_kernel],
                                  elpa_m4_avx512_kernels elpa_m4_avx2_kernels elpa_m4_avx_kernels elpa_m4_sse_kernels elpa_m4_sse_assembly_kernels elpa_m4_generic_kernels,
                                  [m4_bmatch(elpa_m4_cand_kernel,elpa_m4_kind,elpa_m4_cand_kernel)] ),
                     [
                     if test -z "$default_[]elpa_m4_kind[]_kernel"; then
                             if test x"$use_[]elpa_m4_kernel" = x"yes"; then
                                     default_[]elpa_m4_kind[]_kernel="elpa_m4_kernel"
                             fi
                     fi
                     ])
        if test -z "$default_[]elpa_m4_kind[]_kernel"; then
                AC_MSG_ERROR([Internal error, could not determine a default kernel])
        fi
        # find the number of this kernel
        ELPA_2STAGE_[]m4_toupper(elpa_m4_kind)[]_DEFAULT=`grep -i '^ *X(ELPA_2STAGE_'$default_[]elpa_m4_kind[]_kernel'\>' $srcdir/elpa/elpa_constants.h.in | \
                                                                awk '// { gsub(/.*, */, ""); gsub(/).*/, ""); print;}'`
        AC_SUBST([ELPA_2STAGE_]m4_toupper(elpa_m4_kind)[_DEFAULT])
])
Andreas Marek's avatar
Andreas Marek committed
639

640
AC_LANG_PUSH([C])
Andreas Marek's avatar
Andreas Marek committed
641

642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659
if test x"${need_sse}" = x"yes"; then
  AC_MSG_CHECKING(whether we can compile SSE3 with gcc intrinsics in C)
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
#include <x86intrin.h>
int main(int argc, char **argv) {
  double* q;
  __m128d h1 = _mm_loaddup_pd(q);
  return 0;
}
    ])],
    [can_compile_sse=yes],
    [can_compile_sse=no]
  )
  AC_MSG_RESULT([${can_compile_sse}])
  if test x"$can_compile_sse" != x"yes"; then
    AC_MSG_ERROR([Could not compile test program, try with --disable-sse, or adjust the C compiler or CFLAGS])
  fi
  AC_DEFINE([HAVE_SSE_INTRINSICS],[1],[gcc intrinsics SSE is supported on this CPU])
Andreas Marek's avatar
Andreas Marek committed
660 661 662
fi


663 664
if test x"${need_sse_assembly}" = x"yes"; then
  AC_MSG_CHECKING(whether double-precision SSE assembly kernels can be compiled)
Andreas Marek's avatar
Andreas Marek committed
665

666
  $CC $CFLAGS -c $srcdir/src/elpa2/kernels/elpa2_kernels_asm_x86_64_double_precision.s -o conftest.o 2>&5
Andreas Marek's avatar
Andreas Marek committed
667

668 669 670 671 672 673 674 675 676 677
  if test "$?" == 0; then
    can_compile_sse_asm_double=yes
  else
    can_compile_sse_asm_double=no
  fi
  rm -f ./conftest.o
  AC_MSG_RESULT([${can_compile_sse_asm_double}])
  if test x"$can_compile_sse_asm_double" != x"yes"; then
    AC_MSG_ERROR([Could not compile test program, try with --disable-sse-assembly, or adjust the C compiler or CFLAGS])
  fi
678

679 680
  if test x"${want_single_precision}" = x"yes" ; then
    AC_MSG_CHECKING(whether single-precision SSE assembly kernels can be compiled)
681

682
    $CC $CFLAGS -c $srcdir/src/elpa2/kernels/elpa2_kernels_asm_x86_64_single_precision.s -o conftest.o 2>&5
683

684 685 686 687 688 689 690 691 692 693 694
    if test "$?" == 0; then
      can_compile_sse_asm_single=yes
    else
      can_compile_sse_asm_single=no
    fi
    rm -f ./conftest.o
    AC_MSG_RESULT([${can_compile_sse_asm_single}])
    if test x"$can_compile_sse_asm_single" != x"yes"; then
      AC_MSG_ERROR([Could not compile test program, try with --disable-sse-assembly, or adjust the C compiler or CFLAGS])
    fi
  fi
Andreas Marek's avatar
Andreas Marek committed
695 696 697 698
fi



699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717
if test x"${need_avx}" = x"yes"; then
  dnl check whether one can compile AVX gcc intrinsics
  AC_MSG_CHECKING([whether we can compile AVX gcc intrinsics in C])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <x86intrin.h>
   int main(int argc, char **argv){
   double* q;
   __m256d a1_1 = _mm256_load_pd(q);
   return 0;
   }
   ])],
   [can_compile_avx=yes],
   [can_compile_avx=no]
  )
  AC_MSG_RESULT([${can_compile_avx}])
  if test x"$can_compile_avx" != x"yes"; then
    AC_MSG_ERROR([Could not compile a test program with AVX, try with --disable-avx, or adjust the C compiler or CFLAGS])
  fi
  AC_DEFINE([HAVE_AVX],[1],[AVX is supported on this CPU])
718 719 720
fi


721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739
if test x"${need_avx2}" = x"yes"; then
  AC_MSG_CHECKING([whether we can compile AVX2 gcc intrinsics in C])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <x86intrin.h>
   int main(int argc, char **argv){
   double* q;
   __m256d q1 = _mm256_load_pd(q);
   __m256d y1 = _mm256_fmadd_pd(q1, q1, q1);
   return 0;
   }
   ])],
   [can_compile_avx2=yes],
   [can_compile_avx2=no]
  )
  AC_MSG_RESULT([${can_compile_avx2}])
  if test x"$can_compile_avx2" != x"yes"; then
    AC_MSG_ERROR([Could not compile a test program with AVX2, try with --disable-avx2, or adjust the C compiler or CFLAGS])
  fi
  AC_DEFINE([HAVE_AVX2],[1],[AVX2 is supported on this CPU])
740 741
fi

742

743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761
if test x"${need_avx512}" = x"yes"; then
  AC_MSG_CHECKING([whether we can compile AVX512 gcc intrinsics in C])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <x86intrin.h>
   int main(int argc, char **argv){
   double* q;
   __m512d q1 = _mm512_load_pd(q);
   __m512d y1 = _mm512_fmadd_pd(q1, q1, q1);
   return 0;
   }
   ])],
   [can_compile_avx512=yes],
   [can_compile_avx512=no]
  )
  AC_MSG_RESULT([${can_compile_avx512}])
  if test x"$can_compile_avx512" != x"yes"; then
    AC_MSG_ERROR([Could not compile a test program with AVX512, adjust the C compiler or CFLAGS])
  fi
  AC_DEFINE([HAVE_AVX512],[1],[AVX512 is supported on this CPU])
762
fi
763
AC_LANG_POP([C])
764 765


766 767 768 769 770 771 772
AC_LANG_PUSH([Fortran])
if test x"${need_bgp}" = x"yes"; then
  AC_MSG_CHECKING([whether we can compile with BGP intrinsics])
  AC_LINK_IFELSE([AC_LANG_SOURCE([
    program test_bgp
      complex*16 :: y3,q3,h2
      y3 = fxcpmadd(y3,q3,h2)
773

774 775 776 777 778 779 780 781 782
    end program
  ])],
    [can_compile_bgp=yes],
    [can_compile_bgp=no]
  )
  AC_MSG_RESULT([${can_compile_bgp}])
  if test x"$can_compile_bgp" != x"yes"; then
    AC_MSG_ERROR([Could not compile a test program with BGP intrinsics, adjust the FC compiler or FCFLAGS])
  fi
783 784
fi

Andreas Marek's avatar
Andreas Marek committed
785

786 787 788 789 790 791 792
if test x"${need_bgq}" = x"yes"; then
  AC_MSG_CHECKING([whether we can compile with BGQ intrinsics])
  AC_LINK_IFELSE([AC_LANG_SOURCE([
    program test_bgq
      VECTOR(REAL(8))::QPX_h2
      real*8         :: hh(10,2)
      QPX_h2 = VEC_SPLATS(hh(2,2))
Andreas Marek's avatar
Andreas Marek committed
793

794 795 796 797 798 799 800 801 802
    end program
  ])],
    [can_compile_bgq=yes],
    [can_compile_bgq=no]
  )
  AC_MSG_RESULT([${can_compile_bgq}])
  if test x"$can_compile_bgq" != x"yes"; then
    AC_MSG_ERROR([Could not compile a test program with BGQ intrinsics, adjust the FC compiler or FCFLAGS])
  fi
803
fi
804
AC_LANG_POP([Fortran])
805

806 807 808 809 810 811 812 813
if test x"${need_gpu}" = x"yes" ; then
  AC_LANG_PUSH([C])
  CUDA_CFLAGS="$CUDA_CFLAGS -arch $cuda_compute_capability -O2 -I$CUDA_INSTALL_PATH/include"
  LDFLAGS="$LDFLAGS -L$CUDA_INSTALL_PATH/lib64"
  NVCCFLAGS="$NVCCFLAGS $CUDA_CFLAGS $CUDA_LDFLAGS"
  NVCC="nvcc"
  AC_SUBST(NVCC)
  AC_SUBST(NVCCFLAGS)
814

815 816 817 818 819
  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
820

821 822 823 824 825 826 827 828 829 830
  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])
831 832
fi

Andreas Marek's avatar
Andreas Marek committed
833

834 835 836 837 838 839
m4_foreach_w([elpa_m4_kernel],elpa_m4_all_kernels,[
  AM_CONDITIONAL([WITH_]m4_toupper(elpa_m4_kernel)[_KERNEL],[test x"$use_[]elpa_m4_kernel[]" = x"yes"])
  if test x"$use_[]elpa_m4_kernel[]" = x"yes" ; then
    AC_DEFINE([WITH_]m4_toupper(elpa_m4_kernel)[_KERNEL],[1],[Build elpa_m4_kernel kernel])
  fi
])
Andreas Marek's avatar
Andreas Marek committed
840 841


842 843 844
AM_CONDITIONAL([WITH_GPU_VERSION],[test x"$use_gpu" = x"yes"])
if test x"${use_gpu}" = x"yes" ; then
  AC_DEFINE([WITH_GPU_VERSION],[1],[enable GPU support])
Andreas Marek's avatar
Andreas Marek committed
845 846
fi

847 848
LT_INIT

849 850 851 852 853 854
DX_PDF_FEATURE(OFF)
DX_PS_FEATURE(OFF)
DX_MAN_FEATURE(ON)
DX_HTML_FEATURE(ON)
DX_INIT_DOXYGEN([ELPA], [Doxyfile], [docs])

855
USE_ASSUMED_SIZE=yes
Andreas Marek's avatar
Andreas Marek committed
856
AC_MSG_CHECKING(whether assumed size Fortran arrays should be used)
857
AC_ARG_ENABLE([assumed-size],
858
                AS_HELP_STRING([--disable-assumed-size],
859 860
                               [do NOT use assumed-size Fortran arrays]),
                               [if test x"$enableval" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
861
                                  USE_ASSUMED_SIZE=yes
862 863 864
                                else
                                  USE_ASSUMED_SIZE=no
                                fi],
865
              [USE_ASSUMED_SIZE=yes])
866 867 868
AC_MSG_RESULT([${USE_ASSUMED_SIZE}])
AM_CONDITIONAL([WITH_USE_ASSUMED_SIZE],[test x"$USE_ASSUMED_SIZE" = x"yes"])
if test x"${USE_ASSUMED_SIZE}" = x"yes" ; then
869
  AC_DEFINE([USE_ASSUMED_SIZE],[1],[for performance reasons use assumed size Fortran arrays, even if not debuggable])
870 871
fi

872 873 874
if test x"${want_single_precision}" = x"yes" ; then
  AC_DEFINE([WANT_SINGLE_PRECISION_REAL],[1],[build also single-precision for real calculation])
  AC_DEFINE([WANT_SINGLE_PRECISION_COMPLEX],[1],[build also single-precision for complex calculation])
875
fi
876 877
AM_CONDITIONAL([WANT_SINGLE_PRECISION_REAL],[test x"$want_single_precision" = x"yes"])
AM_CONDITIONAL([WANT_SINGLE_PRECISION_COMPLEX],[test x"$want_single_precision" = x"yes"])
878

879
AC_SUBST([MPI_BINARY])
880 881 882 883 884
AC_SUBST([WITH_MKL])
AC_SUBST([WITH_BLACS])
AC_SUBST([with_amd_bulldozer_kernel])
AC_SUBST([FC_MODINC])
AC_SUBST([FC_MODOUT])
885 886 887
AC_SUBST([OPENMP_CFLAGS])
AC_SUBST([OPENMP_FCFLAGS])
AC_SUBST([OPENMP_LDFLAGS])
888
AC_SUBST([DOXYGEN_OUTPUT_DIR], [docs])
889

890 891
rm -rf modules/ private_modules/ .fortran_dependencies/
mkdir modules private_modules
892 893 894 895 896 897 898 899 900 901

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

902 903 904 905 906 907 908 909 910

# 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')
911

912 913
if test x"${with_mpi}" = x"yes"; then
  if test x"${enable_openmp}" = x"yes"; then
914
        SUFFIX="_openmp"
915
  else
916
        SUFFIX=""
917 918 919
  fi
else
  if test x"${enable_openmp}" = x"yes"; then
920
        SUFFIX="_onenode_openmp"
921
  else
922
        SUFFIX="_onenode"
923
  fi
924
fi
925

926 927
AC_SUBST([SUFFIX])
AC_SUBST([PKG_CONFIG_FILE],[elpa${SUFFIX}-${PACKAGE_VERSION}.pc])
928

929 930
AC_CONFIG_FILES([
  Makefile
931
  Doxyfile
932
  ${PKG_CONFIG_FILE}:elpa.pc.in
933
  elpa/elpa_constants.h
934
])
935

936 937 938 939 940 941
m4_include([m4/ax_fc_check_define.m4])
AC_MSG_CHECKING([if workaround for broken preprocessor is needed])

need_manual_cpp=no
AX_FC_CHECK_DEFINE([__INTEL_COMPILER],[need_manual_cpp=yes],[])
AX_FC_CHECK_DEFINE([__PGI],[need_manual_cpp=yes],[])
Lorenz Huedepohl's avatar
Lorenz Huedepohl committed
942 943
ACTUAL_FC="$FC"
AC_SUBST([ACTUAL_FC])
944
if test x"$need_manual_cpp" = x"yes" ; then
945 946
        AC_MSG_RESULT([yes])
        FC="\$(top_srcdir)/manual_cpp $FC"
947
else
948
        AC_MSG_RESULT([no])
949 950
fi

Lorenz Huedepohl's avatar
Lorenz Huedepohl committed
951 952 953 954 955 956 957 958 959 960 961 962 963

dnl PGI compiler uses -module to specify module output
dnl directory. This clashes with libtools -module link option
dnl => escape it for libtool with -Xcompiler -module
AC_MSG_CHECKING([whether we have to escape '-module' for libtool])
if test x"$FC_MODOUT" = x'-module ' ; then
        FC_MODOUT="-Xcompiler $FC_MODOUT -Xcompiler \$(ac_empty)"
        FC="\$(top_srcdir)/remove_xcompiler $FC"
        AC_MSG_RESULT([yes])
else
        AC_MSG_RESULT([no])
fi

964

965
AC_OUTPUT
966

967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983
echo ""
echo "The following ELPA2 kernels will be build:"
echo ""
m4_foreach_w([elpa_m4_kind],[real complex],[
        m4_foreach_w([elpa_m4_kernel],m4_expand(elpa_m4_[]elpa_m4_kind[]_kernels),[
                if test x"$use_[]elpa_m4_kernel" = x"yes" ; then
                        echo -n "  elpa_m4_kernel"
                        if test "$fixed_]elpa_m4_kind[_kernel" = "]elpa_m4_kernel[" ; then
                                echo -n " (selected as fixed kernel)"
                        fi
                        if test "$default_]elpa_m4_kind[_kernel" = "]elpa_m4_kernel[" ; then
                                echo -n " (default)"
                        fi
                        echo ""
                fi
        ])
])
984

985 986 987 988 989 990 991
echo " "
echo "***********************************************************************"
echo "*  This is a prerelease of the ELPA library 2016.11.001.pre library!  *"
echo "*  It contains an experimental API which will most likely change with *"
echo "*  the next stable release.                                           *"
echo "***********************************************************************"
echo " "
992
make -f $srcdir/generated_headers.am generated-headers top_srcdir="$srcdir"