configure.ac 53 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

7
AC_CONFIG_SRCDIR([src/elpa.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
# ABI version
#
17
18
19
20
21
22
23
24
25
26
27
28
29
# 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)
#
30
AC_SUBST([ELPA_SO_VERSION], [14:0:0])
31

32
33
34
35
36
37
38
# AC_DEFINE_SUBST(NAME, VALUE, DESCRIPTION)
# -----------------------------------------
AC_DEFUN([AC_DEFINE_SUBST], [
AC_DEFINE([$1], [$2], [$3])
AC_SUBST([$1], ['$2'])
])

39
40
# API Version
AC_DEFINE([EARLIEST_API_VERSION], [20170403], [Earliest supported ELPA API version])
41

42
AC_DEFINE_SUBST(CURRENT_API_VERSION, 20190524, "Current ELPA API version")
Andreas Marek's avatar
Andreas Marek committed
43
44
# Autotune Version
AC_DEFINE([EARLIEST_AUTOTUNE_VERSION], [20171201], [Earliest ELPA API version, which supports autotuning])
45
46
AC_DEFINE([CURRENT_AUTOTUNE_VERSION], [20190524], [Current ELPA autotune version])
AC_DEFINE_SUBST(CURRENT_AUTOTUNE_VERSION, 20190524, "Current ELPA autotune version")
47

48
49
50
51
52
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
53
54
AC_MSG_CHECKING(whether legacy interface should be provided)
AC_ARG_ENABLE([legacy-interface],
55
56
              AS_HELP_STRING([--enable-legacy-interface],
                             [build legacy API, default no]),
Andreas Marek's avatar
Andreas Marek committed
57
58
59
60
61
62
63
              [
	       if test x"$enableval" = x"yes"; then
	         enable_legayc=yes
	       else
	         enable_legacy=no
	       fi
	       ],
64
              [enable_legacy=no])
Andreas Marek's avatar
Andreas Marek committed
65
AC_MSG_RESULT([${enable_legacy}])
66
AM_CONDITIONAL([ENABLE_LEGACY],[test x"$enable_legacy" = x"yes"])
Andreas Marek's avatar
Andreas Marek committed
67
68
69
70
if test x"${enable_legacy}" = x"yes"; then
        AC_DEFINE([ENABLE_LEGACY], [1], [enable legacy interface])
fi

Andreas Marek's avatar
Andreas Marek committed
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
AC_MSG_CHECKING(whether in C interface the error argument should be optional)
AC_ARG_ENABLE([optional-argument-in-C-API],
              AS_HELP_STRING([--enable-optional-argument-in-C-API],
                             [do not build C API with error argument as optional, default no]),
              [
	       if test x"$enableval" = x"yes"; then
	         optional_c_error_argument=yes
	       else
	         optional_c_error_argument=no
	       fi
	       ],
              [optional_c_error_argument=no])
AC_MSG_RESULT([${optional_c_error_argument}])
AM_CONDITIONAL([OPTIONAL_C_ERROR_ARGUMENT],[test x"$enable_legacy" = x"yes"])
if test x"${optional_c_error_argument}" = x"yes"; then
        AC_DEFINE([OPTIONAL_C_ERROR_ARGUMENT], [1], [enable error argument in C-API to be optional])
fi

89
90
91
92
93

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

94

95
dnl OpenMP
96
m4_include([m4/ax_elpa_openmp.m4])
97
98
AC_MSG_CHECKING(whether --enable-openmp is specified)
AC_ARG_ENABLE([openmp],
99
100
              AS_HELP_STRING([--enable-openmp],
                             [use OpenMP threading, default no.]),
Andreas Marek's avatar
Andreas Marek committed
101
102
103
104
105
106
107
              [
	       if test x"$enableval" = x"yes"; then
	         enable_openmp=yes
	       else
	         enable_openmp=no
	       fi
	       ],
108
              [enable_openmp=no])
109
110
111
AC_MSG_RESULT([${enable_openmp}])
AM_CONDITIONAL([WITH_OPENMP],[test x"$enable_openmp" = x"yes"])
if test x"${enable_openmp}" = x"yes"; then
112
        AC_DEFINE([WITH_OPENMP], [1], [use OpenMP threading])
113
114
fi

115

116
117
dnl mpi
AC_ARG_WITH(mpi, [AS_HELP_STRING([--with-mpi=[[yes|no]]], [compile with MPI. Default: yes])],,[with_mpi=yes])
118
AM_CONDITIONAL([WITH_MPI],[test x"$with_mpi" = x"yes"])
119
if test x"${with_mpi}" = x"yes"; then
120
  AC_DEFINE([WITH_MPI], [1], [use MPI])
121
122
fi

123
124
125
126
127
128

dnl Scalapack tests
AC_MSG_CHECKING(whether --enable-scalapack-tests is specified)
AC_ARG_ENABLE([scalapack-tests],
              AS_HELP_STRING([--enable-scalapack-tests],
                             [build SCALAPACK test cases for performance comparison, needs MPI, default no.]),
Andreas Marek's avatar
Andreas Marek committed
129
130
131
132
133
134
135
              [
	       if test x"$enableval" = x"yes"; then
	         enable_scalapack_tests=yes
	       else
	         enable_scalapack_tests=no
	       fi
	       ],
136
137
138
139
140
141
142
143
144
145
146
147
              [enable_scalapack_tests="no"])
AC_MSG_RESULT([$enable_scalapack_tests])
if test x"${enable_scalapack_tests}" = x"yes"; then
        if test x"$with_mpi" = x"no"; then
                AC_MSG_ERROR([You cannot build the SCALAPCK test cases without MPI])
        fi
        AC_DEFINE([WITH_SCALAPACK_TESTS], [1], [build SCALAPACK test cases])
fi
AM_CONDITIONAL([WITH_SCALAPACK_TESTS], [test x"$enable_scalapack_tests" = x"yes"])


dnl C
148
AC_LANG_PUSH([C])
149

150
151
152
153
154
155
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
156

157
if test x"${enable_openmp}" = x"yes"; then
158
159
160
161
162
  AX_ELPA_OPENMP
  if test "$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"
163
fi
164

165
166
167
168
169
170
171
172

AX_CHECK_COMPILE_FLAG([-std=c11], [
  CFLAGS+=" -std=c11"
], [
  echo "C compiler cannot compile C11 code"
  exit -1
])

173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
AC_MSG_CHECKING(whether C compiler can use _Generic )
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
int main(int argc, char **argv) {
#define elpa_set(e, name, value, error) _Generic((value), \
                int: \
                  elpa_set_integer, \
                \
                double: \
                  elpa_set_double \
        )(e, name, value, error)

  return 0;
}
    ])],
    [can_compile_generic=yes],
    [can_compile_generic=no]
  )
AC_MSG_RESULT([${can_compile_generic}])
if test x"$can_compile_generic" != x"yes"; then
  AC_MSG_ERROR([C compiler cannot handle _Generic statement! Upgrade or change C compiler])
fi
AC_DEFINE([HAVE_VSX_SSE],[1],[Altivec VSX intrinsics are supported on this CPU])

196
AC_PROG_INSTALL
197
198
AM_PROG_AR
AM_PROG_AS
199
AC_PROG_CC_C99
200
AM_PROG_CC_C_O
201
AC_LANG_POP([C])
Andreas Marek's avatar
Andreas Marek committed
202
203

# Fortran
204
AC_LANG_PUSH([Fortran])
Andreas Marek's avatar
Andreas Marek committed
205
m4_include([m4/ax_prog_fc_mpi.m4])
206
207
208
209
210
211
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
212

213
AC_FC_SRCEXT([F90])
214
215
216
217
218
AC_FC_FREEFORM
AC_FC_MODULE_FLAG
AC_FC_MODULE_OUTPUT_FLAG
AC_FC_LIBRARY_LDFLAGS

219

220
if test x"${enable_openmp}" = x"yes"; then
221
222
223
224
225
  AX_ELPA_OPENMP
  if test "$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"
226
227
fi

228
229
dnl check which MPI binray invokes a MPI job
if test x"$with_mpi" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
230
  AC_CHECK_PROGS([MPI_BINARY], [mpiexec.hydra mpiexec mpirun poe runjob srun aprun], [no])
231
  if test x"$MPI_BINARY" = x"no"; then
Andreas Marek's avatar
Andreas Marek committed
232
    AC_MSG_ERROR([Could not find either of the MPI binaries: mpiexec.hydra, mpiexec, mpirun, poe, runjob, srun, aprun])
233
234
235
  fi
fi

236
237
238
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)
239
AC_ARG_ENABLE([redirect],
240
241
              [AS_HELP_STRING([--enable-redirect],
                              [for test programs, allow redirection of stdout/stderr per MPI taks in a file (useful for timing), default no.])],
Andreas Marek's avatar
Andreas Marek committed
242
243
244
245
246
247
248
              [
               if test x"$enableval" = x"yes"; then
                 enable_redirect=yes
               else
                 enable_redirect=no
               fi
	       ],
249
              [enable_redirect=no])
250
251
AC_MSG_RESULT([${enable_redirect}])

252
253
254
255
256
257
258
259
260
261
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}])


262
263
dnl redirect
if test x"${enable_redirect}" = x"yes"; then
264
265
  AC_DEFINE([HAVE_REDIRECT], [1], [Redirect stdout and stderr of test programs per MPI tasks to a file])
fi
266
267
268
AM_CONDITIONAL([HAVE_REDIRECT],[test x"$enable_redirect" = x"yes"])

dnl build with ftimings support
269

Andreas Marek's avatar
Andreas Marek committed
270
AC_ARG_ENABLE([timings],
271
272
              [AS_HELP_STRING([--disable-timings],
                              [more detailed timing, default yes])],
Andreas Marek's avatar
Andreas Marek committed
273
274
275
276
277
278
279
              [
               if test x"$enableval" = x"yes"; then
                 enable_timings=yes
               else
                 enable_timings=no
               fi
	       ],
280
              [enable_timings=yes])
281
282
283
284

if test x"${enable_timings}" = x"yes"; then
  AC_DEFINE([HAVE_DETAILED_TIMINGS], [1], [Enable more timing])
fi
285
AM_CONDITIONAL([HAVE_DETAILED_TIMINGS], [test x"$enable_timings" = x"yes"])
286
287

dnl PAPI for ftimings
288
AC_LANG_PUSH([C])
289
AC_ARG_WITH([papi],
290
291
            [AS_HELP_STRING([--with-papi],
                            [Use PAPI to also measure flop count in the detailed timing (--enable-timing), disabled by default])],
Andreas Marek's avatar
Andreas Marek committed
292
293
294
295
296
297
298
            [
               if test x"$enableval" = x"yes"; then
                 with_papi=yes
               else
                 with_papi=no
               fi
	     ],
299
            [with_papi="no"])
300
301
if test x"${enable_timings}" = x"yes"; then
  if test x"$with_papi" = x"yes" ; then
302
    AC_SEARCH_LIBS([PAPI_library_init], [papi], [papi_found=yes], [papi_found=no])
303
304
    if test x"$papi_found" = x"no" ; then
      AC_MSG_ERROR(["Could not find usable PAPI installation, please install or adjust CFLAGS, LDFLAGS"])
305
306
307
308
    fi
    AC_DEFINE([HAVE_LIBPAPI], [1], [Use the PAPI library])
  fi
fi
309
AC_LANG_POP([C])
310

Pavel Kus's avatar
Pavel Kus committed
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
dnl Likwid
AC_LANG_PUSH([Fortran])
AC_ARG_WITH([likwid],
            [AS_HELP_STRING([--with-likwid=[[yes|no(default)|PATH]]],
                            [Use Likwid to measure performance in some parts of the library])],
            [with_likwid="$withval"],
            [with_likwid="no"])

if test x"$with_likwid" != x"no" ; then
  if test -d $with_likwid/lib ; then
    LDFLAGS="-L$with_likwid/lib $LDFLAGS"
  fi
  if test -d $with_likwid/lib64 ; then
    LDFLAGS="-L$with_likwid/lib64 $LDFLAGS"
  fi
  if test -d $with_likwid/include ; then
    FCFLAGS="-I$with_likwid/include $FCFLAGS"
  fi
  AC_SEARCH_LIBS([likwid_markerInit], [likwid], [liblikwid_found="yes"], [liblikwid_found="no"])
  if test x"$liblikwid_found" = x"no" ; then
    AC_MSG_ERROR([Could not find a usable likwid library, please adjust LDFLAGS])
  fi
  AC_MSG_CHECKING([whether we can use the likwid module in a Fortran program])
  AC_COMPILE_IFELSE([
       program foo
       use likwid

       implicit none

       call likwid_markerInit()
       call likwid_markerThreadInit()

       call likwid_markerStartRegion("foobar")
       call likwid_markerStopRegion("foobar")

       call likwid_markerClose()
       end
    ],
    [AC_MSG_RESULT([yes])],
    [AC_MSG_RESULT([no])
     AC_MSG_ERROR([Could not compile a Fortran program using the likwid module, adjust FCFLAGS])])
  AC_DEFINE([HAVE_LIKWID], [1], [Use likwid])
fi
AC_LANG_POP([Fortran])

356
357
358
save_FCFLAGS=$FCFLAGS
save_LDFLAGS=$LDFLAGS

359
360
361
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])

362
363
FCFLAGS="$FCFLAGS $SCALAPACK_FCFLAGS"
LDFLAGS="$LDFLAGS $SCALAPACK_LDFLAGS"
364

365
dnl check whether fortran error_unit is defined
366
AC_MSG_CHECKING([whether Fortran module iso_fortran_env is available])
367
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
368
  program test_error_unit
369
    use iso_fortran_env, only : error_unit
370
371
372
373
374
375
376
377
    implicit none

    write(error_unit,*) "error_unit is defined"
  end program
])],
  [can_use_iso_fortran_env=yes],
  [can_use_iso_fortran_env=no]
)
378
AC_MSG_RESULT([${can_use_iso_fortran_env}])
379
380
381
382
if test x"${can_use_iso_fortran_env}" = x"yes" ; then
 AC_DEFINE([HAVE_ISO_FORTRAN_ENV],[1],[can use module iso_fortran_env])
fi

383

384
385
dnl check whether one can link with specified MKL (desired method)
AC_MSG_CHECKING([whether we can compile a Fortran program using MKL])
386
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
387
  program test_mkl
388
    use mkl_service
389
390
391
392
393
394
395
396
397
398
399
400
    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"$can_compile_with_mkl" = x"yes" ; then
  AC_MSG_CHECKING([whether we can link a Fortran program with MKL])
401
  AC_LINK_IFELSE([AC_LANG_SOURCE([
402
    program test_mkl
403
      use mkl_service
404
405
406
407
408
      character*198 :: string
      call mkl_get_version_string(string)
      write(*,'(a)') string
    end program
  ])],
Andreas Marek's avatar
Andreas Marek committed
409
410
    [have_mkl=yes],
    [have_mkl=no]
411
  )
Andreas Marek's avatar
Andreas Marek committed
412
  AC_MSG_RESULT([${have_mkl}])
413
414
415
fi

dnl if not mkl, check all the necessary individually
416
if test x"${have_mkl}" = x"yes" ; then
417
418
419
420
  WITH_MKL=1
else

  dnl first check blas
421
  AC_SEARCH_LIBS([dgemm],[openblas satlas blas],[have_blas=yes],[have_blas=no])
Andreas Marek's avatar
Andreas Marek committed
422
423
  AC_MSG_CHECKING([whether we can link a program with a blas lib])
  AC_MSG_RESULT([${have_blas}])
424

425
  if test x"${have_blas}" = x"no" ; then
426
    AC_MSG_ERROR([could not link with blas: specify path])
427
  fi
428
  dnl now lapack
Andreas Marek's avatar
Andreas Marek committed
429
430
431
  AC_SEARCH_LIBS([dlarrv],[lapack],[have_lapack=yes],[have_lapack=no])
  AC_MSG_CHECKING([whether we can link a program with a lapack lib])
  AC_MSG_RESULT([${have_lapack}])
432

433
  if test x"${have_lapack}" = x"no" ; then
434
    AC_MSG_ERROR([could not link with lapack: specify path])
435
  fi
Andreas Marek's avatar
Andreas Marek committed
436

437
  if test x"${with_mpi}" = x"yes"; then
438
    dnl test whether scalapack already contains blacs
439
    scalapack_libs="mpiscalapack scalapack scalapack-openmpi"
440
441
442
443
444
    old_LIBS="$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])
Andreas Marek's avatar
Andreas Marek committed
445
	       
446
447
448
449
450
      AC_MSG_RESULT([${blacs_in_scalapack}])
      if test x"${blacs_in_scalapack}" = x"yes"; then
        break
      fi
    done
451

452
453
    if test x"${blacs_in_scalapack}" = x"no"; then
      LIBS="${old_LIBS}"
454

455
456
      dnl Test for stand-alone blacs
      AC_SEARCH_LIBS([bi_f77_init],[mpiblacsF77init],[],[],[-lmpiblacs])
457
458
459
460
461
462

      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
463
      AC_SEARCH_LIBS([blacs_gridinit],[mpiblacs blacs],[have_blacs=yes],[have_blacs=no])
Andreas Marek's avatar
Andreas Marek committed
464

465
466
467
      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
468
    fi
469

470
    AC_SEARCH_LIBS([pdtran],[$scalapack_libs],[have_scalapack=yes],[have_scalapack=no])
Andreas Marek's avatar
Andreas Marek committed
471

472
473
474
    if test x"${have_scalapack}" = x"no" ; then
      AC_MSG_ERROR([could not link with scalapack: specify path])
    fi
475
476
477
478
  fi

  dnl check whether we can link alltogehter
  AC_MSG_CHECKING([whether we can link a Fortran program with all blacs/scalapack])
479
  AC_LINK_IFELSE([AC_LANG_SOURCE([
480
481
    program dgemm_test

482
483
      integer , parameter:: M = 4, N = 3, K = 2
      real :: A(M,K), B(K,N), C(M,N)
484

485
      call dgemm('N','N',M,N,K,1.0,A,M,B,K,0.0,C,M)
486
487
488
489
490
491
492
493

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

494
   if test x"${can_link_with_blacs_scalapack}" = x"yes" ; then
495
496
     WITH_BLACS=1
   else
497
   AC_MSG_ERROR([We can neither link with MKL or another Scalpack. Please specify SCALAPACK_LDFLAGS and SCALAPACK_FCFLAGS!])
498
499
500
   fi
fi

501
502
503
504
dnl important: reset them again!
FCFLAGS=$save_FCFLAGS
LDFLAGS=$save_LDFLAGS

505

Andreas Marek's avatar
Andreas Marek committed
506
507
508
509
dnl check for intrinsic fortran function of 2003 standard
AC_MSG_CHECKING([whether we can use the intrinsic Fortran function "get_environment_variable"])


510
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
511
512
513
514
515
516
517
518
519
  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([${fortran_can_check_environment}])
520
521
522
if test x"${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
523

524

525
526
dnl check whether BAND_TO_FULL_BLOCKING is set
AC_MSG_CHECKING(whether BAND_TO_FLULL_BLOCKING is requested)
527
AC_ARG_ENABLE(band-to-full-blocking,[AS_HELP_STRING([--disable-band-to-full-blocking],
528
                                                    [build ELPA2 with blocking in band_to_full (default: enabled)])],
Andreas Marek's avatar
Andreas Marek committed
529
530
531
532
533
534
535
              [
               if test x"$enableval" = x"yes"; then
                 enable_band_to_full_blocking=yes
               else
                 enable_band_to_full_blocking=no
               fi
	       ],
536
537
	      [enable_band_to_full_blocking="yes"])
AC_MSG_RESULT([${enable_band_to_full_blocking}])
538

539
540
AM_CONDITIONAL([BAND_TO_FULL_BLOCKING],[test x"$enable_band_to_full_blocking" = x"yes"])
if test x"${enable_band_to_full_blocking}" = x"yes"; then
541
       AC_DEFINE([BAND_TO_FULL_BLOCKING], [1], [use blocking in trans_ev_band_to_full])
542
543
fi

544
545
546
547
548
549
550

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])

551

552
dnl setup nvcc flags and use them in later tests
553
554
555
user_sets_gpu_compute_capability="no"
AC_MSG_CHECKING(whether a GPU compute capability is specified)
AC_ARG_WITH([GPU-compute-capability],
556
557
558
            [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"])
559
560
AC_MSG_RESULT([${user_sets_gpu_compute_capability}])

561

562
563
564
565
566
567
568
569
570
571
572
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
573

574
575
dnl Test possibility of 'use mpi', if requested
if test x"${with_mpi}" = x"yes" ; then
576
577
  AC_ARG_ENABLE([mpi-module],
                AS_HELP_STRING([--disable-mpi-module],
Pavel Kus's avatar
Pavel Kus committed
578
                               [do not use the Fortran MPI module, get interfaces by 'include "mpif.h']),
Andreas Marek's avatar
Andreas Marek committed
579
580
581
582
583
584
585
                [
                 if test x"$enableval" = x"yes"; then
                   enable_mpi_module=yes
                 else
                   enable_mpi_module=no
                 fi
		 ],
586
587
                [enable_mpi_module=yes])
  if test x"${enable_mpi_module}" = x"yes" ; then
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
    AC_MSG_CHECKING(whether Fortran mpi module can be used)
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
      program test_mpi_module
        use mpi
        real :: time
        time = MPI_WTime()
      end program
    ])],
      [can_use_fortran_mpi_module=yes],
      [can_use_fortran_mpi_module=no]
    )
    AC_MSG_RESULT([${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
603
      AC_MSG_ERROR([Could not compile a Fortran program with an 'use mpi' statement. You can try again with --disable-mpi-module])
604
    fi
605
  fi
Andreas Marek's avatar
Andreas Marek committed
606
fi
607
AC_LANG_POP([Fortran])
Andreas Marek's avatar
Andreas Marek committed
608

609
610
dnl Assemble the list of kernels to build
m4_pattern_forbid([elpa_m4])
Andreas Marek's avatar
Andreas Marek committed
611

612
613
614
m4_define(elpa_m4_generic_kernels, [
        real_generic
        real_generic_simple
Andreas Marek's avatar
Andreas Marek committed
615
        real_generic_simple_block4
616
617
618
        complex_generic
        complex_generic_simple
])
619

620
621
622
623
m4_define(elpa_m4_sse_assembly_kernels, [
        real_sse_assembly
        complex_sse_assembly
])
624

625
626
627
628
629
630
631
m4_define(elpa_m4_sse_kernels, [
        real_sse_block2
        real_sse_block4
        real_sse_block6
        complex_sse_block1
        complex_sse_block2
])
632

633
634
635
636
637
638
m4_define(elpa_m4_sparc64_kernels, [
        real_sparc64_block2
        real_sparc64_block4
        real_sparc64_block6
])

639
640
641
642
643
644
m4_define(elpa_m4_neon_arch64_kernels, [
        real_neon_arch64_block2
        real_neon_arch64_block4
        real_neon_arch64_block6
])

645
646
647
648
649
650
m4_define(elpa_m4_vsx_kernels, [
        real_vsx_block2
        real_vsx_block4
        real_vsx_block6
])

651
652
653
654
655
656
657
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
658

659
660
661
662
663
664
665
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
666

667
668
669
670
671
672
673
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
674

675
676
677
678
m4_define(elpa_m4_bgp_kernels, [
        real_bgp
        complex_bgp
])
Andreas Marek's avatar
Andreas Marek committed
679

680
681
682
683
m4_define(elpa_m4_bgq_kernels, [
        real_bgq
        complex_bgq
])
Andreas Marek's avatar
Andreas Marek committed
684

685
m4_define(elpa_m4_gpu_kernels, [
686
687
        real_gpu
        complex_gpu
688
])
689

690
m4_define(elpa_m4_kernel_types, [generic sparc64 neon_arch64 vsx sse sse_assembly avx avx2 avx512 bgp bgq gpu])
691

692
693
694
695
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
696

697
698
699
700
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
701

702
703
704
705
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) ]))
706

707
708
709
710
711
712
713
714
715
716
717
718
719
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]
        ])
])
720

721
722
dnl Modify list of kernels with configure arguments
ELPA_SELECT_KERNELS([generic],[enable])
723
ELPA_SELECT_KERNELS([sparc64],[disable])
724
ELPA_SELECT_KERNELS([neon_arch64],[disable])
725
ELPA_SELECT_KERNELS([vsx],[disable])
726
727
728
729
730
731
732
733
734
735
736
737
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=""
])
738

739
if test x"${enable_bgp}" = x"yes" -o x"$enable_bgq" = x"yes"; then
740
        m4_foreach_w([elpa_m4_kernel], elpa_m4_sparc64_kernels elpa_m4_neon_arch64_kernels elpa_m4_vsx_kernels elpa_m4_sse_kernels elpa_m4_avx_kernels elpa_m4_avx2_kernels elpa_m4_avx512_kernels, [
741
742
743
744
745
746
747
748
749
750
751
752
                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:]
753
754
                 m4_foreach_w([elpa_m4_kernel],m4_expand(elpa_m4_[]elpa_m4_kind[]_kernels),[m4_bpatsubst(elpa_m4_kernel,elpa_m4_kind[]_,[]) ]))]),
                 [fixed_]elpa_m4_kind[_kernel="]elpa_m4_kind[_$withval"],[fixed_]elpa_m4_kind[_kernel=""])
755
        if test -n "$fixed_[]elpa_m4_kind[]_kernel" ; then
756
                found="no"
757
758
759
                m4_foreach_w([elpa_m4_otherkernel],m4_expand(elpa_m4_[]elpa_m4_kind[]_kernels),[
                        if test "$fixed_]elpa_m4_kind[_kernel" = "]elpa_m4_otherkernel[" ; then
                                use_[]elpa_m4_otherkernel[]=yes
760
                                found="yes"
761
762
763
764
                        else
                                use_[]elpa_m4_otherkernel[]=no
                        fi
                ])
765
766
767
                if test x"$found" = x"no" ; then
                        AC_MSG_ERROR([Invalid kernel "$fixed_]elpa_m4_kind[_kernel" specified for --with-fixed-]elpa_m4_kind[-kernel])
                fi
768
769
770
771
                default_[]elpa_m4_kind[]_kernel="$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
])
772

773
774
775
776
777
778
779
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
        ])
780
781
        use_real_gpu=yes
        use_complex_gpu=yes
782
fi
783

784

785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
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
])
800
m4_foreach_w([elpa_m4_arch],[sparc64 neon_arch64 vsx sse avx avx2 avx512],[
801
802
803
804
        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])
])
805

806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
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
])
826

827

828
829
830
dnl the list of kernels is now assembled
dnl choosing a default kernel

831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
m4_foreach_w([elpa_m4_kind],[real complex],[
        AC_ARG_WITH([default-]elpa_m4_kind[-kernel], m4_expand([AS_HELP_STRING([--with-default-]elpa_m4_kind[-kernel]=KERNEL,
                [set a specific ]elpa_m4_kind[ kernel as default kernel. Available kernels are:]
                 m4_foreach_w([elpa_m4_kernel],m4_expand(elpa_m4_[]elpa_m4_kind[]_kernels),[m4_bpatsubst(elpa_m4_kernel,elpa_m4_kind[]_,[]) ]))]),
                 [default_]elpa_m4_kind[_kernel="]elpa_m4_kind[_$withval"],[default_]elpa_m4_kind[_kernel=""])
        #if test -n "$default_[]elpa_m4_kind[]_kernel" ; then
        #        found="no"
        #        m4_foreach_w([elpa_m4_otherkernel],m4_expand(elpa_m4_[]elpa_m4_kind[]_kernels),[
        #                if test "$default_]elpa_m4_kind[_kernel" = "]elpa_m4_otherkernel[" ; then
        #                        use_[]elpa_m4_otherkernel[]=yes
        #                        found="yes"
        #                else
        #                        use_[]elpa_m4_otherkernel[]=no
        #                fi
        #        ])
        #        if test x"$found" = x"no" ; then
        #                AC_MSG_ERROR([Invalid kernel "$default_]elpa_m4_kind[_kernel" specified for --with-default-]elpa_m4_kind[-kernel])
        #        fi
        #        AC_DEFINE([WITH_DEFAULT_]m4_toupper(elpa_m4_kind)[_KERNEL],[1],[use specific ]elpa_m4_kind[ default kernel (set at compile time)])
        #fi
])



855
856
857
m4_foreach_w([elpa_m4_kind],[real complex],[
        m4_foreach_w([elpa_m4_kernel],
                     m4_foreach_w([elpa_m4_cand_kernel],
858
                                  elpa_m4_avx512_kernels elpa_m4_avx2_kernels elpa_m4_avx_kernels elpa_m4_sse_kernels elpa_m4_sse_assembly_kernels elpa_m4_sparc64_kernels elpa_m4_neon_arch64_kernels elpa_m4_vsx_kernels elpa_m4_generic_kernels,
859
860
861
862
863
864
865
866
867
868
869
870
871
                                  [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 | \
872
                                                                perl -pe 's/^[[^,]]*, *//; s/,.*//;'`
873
874
        AC_SUBST([ELPA_2STAGE_]m4_toupper(elpa_m4_kind)[_DEFAULT])
])
Andreas Marek's avatar
Andreas Marek committed
875

876
877
878
879
880
881
882
dnl #include <fjmfunc.h>
dnl #include <emmintrin.h>
dnl int main(int argc, char **argv) {
dnl   __m128d q;
dnl   __m128d h1 = _fjsp_neg_v2r8(q);
dnl   return 0;
dnl }
883
AC_LANG_PUSH([C])
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904

if test x"${need_vsx}" = x"yes"; then
  AC_MSG_CHECKING(whether we can compile Altivec VSX with intrinsics in C)
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
#include <altivec.h>
int main(int argc, char **argv) {
  __vector double a, b, c;
  c = vec_add(a,b);
  return 0;
}
    ])],
    [can_compile_vsx=yes],
    [can_compile_vsx=no]
  )
  AC_MSG_RESULT([${can_compile_vsx}])
  if test x"$can_compile_vsx" != x"yes"; then
    AC_MSG_ERROR([Could not compile test program, try with --disable-vsx, or adjust the C compiler or CFLAGS])
  fi
  AC_DEFINE([HAVE_VSX_SSE],[1],[Altivec VSX intrinsics are supported on this CPU])
fi

905
906
907
if test x"${need_sparc64}" = x"yes"; then
  AC_MSG_CHECKING(whether we can compile SPARC64 with intrinsics in C)
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
908
909
#include <fjmfunc.h>
#include <emmintrin.h>
910
int main(int argc, char **argv) {
Andreas Marek's avatar
Andreas Marek committed
911
912
  __m128d  tau1;
  __m128d h1 = _fjsp_neg_v2r8(tau1);
913
914
915
916
917
918
919
920
921
922
923
924
  return 0;
}
    ])],
    [can_compile_sparc64=yes],
    [can_compile_sparc64=no]
  )
  AC_MSG_RESULT([${can_compile_sparc64}])
  if test x"$can_compile_sparc64" != x"yes"; then
    AC_MSG_ERROR([Could not compile test program, try with --disable-sparc64, or adjust the C compiler or CFLAGS])
  fi
  AC_DEFINE([HAVE_SPARC64_SSE],[1],[SPARC64 intrinsics are supported on this CPU])
fi
Andreas Marek's avatar
Andreas Marek committed
925

926
927
928
929
930
931
if test x"${need_neon_arch64}" = x"yes"; then
  AC_MSG_CHECKING(whether we can compile NEON ARCH64 with intrinsics in C)
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
#include <arm_neon.h>
int main(int argc, char **argv) {
  __Float64x2_t  x1, x2, x3, x4;
Andreas Marek's avatar
Andreas Marek committed
932
  x4 = vfmaq_f64(x1, x2, x3);
933
934
935
936
937
938
939
940
941
942
943
944
945
946
  return 0;
}
    ])],
    [can_compile_neon_arch64=yes],
    [can_compile_neon_arch64=no]
  )
  AC_MSG_RESULT([${can_compile_neon_arch64}])
  if test x"$can_compile_neon_arch64" != x"yes"; then
    AC_MSG_ERROR([Could not compile test program, try with --disable-neon_arch64, or adjust the C compiler or CFLAGS])
  fi
  AC_DEFINE([HAVE_NEON_ARCH64_SSE],[1],[NEON_ARCH64 intrinsics are supported on this CPU])
fi


947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
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
965
966
967
fi


968
969
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
970

Andreas Marek's avatar
Andreas Marek committed
971
  $CC $CFLAGS -c $srcdir/src/elpa2/kernels/asm_x86_64_double_precision.s -o conftest.o 2>&5
Andreas Marek's avatar
Andreas Marek committed
972

973
974
975
976
977
978
979
980
981
982
  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
983

984
985
  if test x"${want_single_precision}" = x"yes" ; then
    AC_MSG_CHECKING(whether single-precision SSE assembly kernels can be compiled)
986

Andreas Marek's avatar
Andreas Marek committed
987
    $CC $CFLAGS -c $srcdir/src/elpa2/kernels/asm_x86_64_single_precision.s -o conftest.o 2>&5
988

989
990
991
992
993
994
995
996
997
998
999
    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
1000
1001
1002
1003
fi



1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
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])
1023
1024
1025
fi


1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
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])
1045
1046
fi

1047

1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
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])
Andreas Marek's avatar
Andreas Marek committed
1067
1068

  if test x"$can_compile_avx512" = x"yes"; then
1069
    AC_MSG_CHECKING([whether we compile for Xeon])
1070
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
1071
1072
1073
     #include <x86intrin.h>
     int main(int argc, char **argv){
     __m512d sign;
1074
     __m512d h1_real;
Andreas Marek's avatar
Andreas Marek committed
1075
1076
1077
1078
1079
1080
1081
1082

     __m512d x1 = _mm512_xor_pd(h1_real, sign);
     return 0;
     }
     ])],
     [can_compile_avx512_xeon=yes],
     [can_compile_avx512_xeon=no]
    )
1083
1084
1085
    AC_MSG_RESULT([${can_compile_avx512_xeon}])

    AC_MSG_CHECKING([whether we compile for Xeon PHI])
1086
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
1087
1088
1089
1090
     #include <x86intrin.h>
     int main(int argc, char **argv){
     __m512d sign;
     __m512d h1;
Andreas Marek's avatar
Andreas Marek committed
1091
     __m512d h2_real;
Andreas Marek's avatar
Andreas Marek committed
1092

1093
     __m512d x1 = (__m512d) _mm512_xor_epi64((__m512i) h2_real, (__m512i) sign);
Andreas Marek's avatar
Andreas Marek committed
1094
1095
1096
1097
1098
1099
     return 0;
     }
     ])],
     [can_compile_avx512_xeon_phi=yes],
     [can_compile_avx512_xeon_phi=no]
    )
1100
1101
    AC_MSG_RESULT([${can_compile_avx512_xeon_phi}])

1102
1103
1104
1105
1106
1107
    # this is needed for the intel compiler
    if test x"$can_compile_avx512_xeon" = x"yes" ; then
      if test x"$can_compile_avx512_xeon_phi" = x"yes" ; then
        # we want only one to be true; this is ugly but could not come up with a better way
        grep Phi /proc/cpuinfo > /dev/null
        if test x"$?" = x"0" ; then
1108
	  echo "Xeon PHI found ... disabling AVX512 Xeon"
1109
1110
1111
1112
1113
          can_compile_avx512_xeon=no
        fi
      fi
    fi

Andreas Marek's avatar
Andreas Marek committed
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
    if test x"$can_compile_avx512_xeon" = x"yes"; then
      AC_DEFINE([HAVE_AVX512_XEON],[1],[AVX512 for Xeon is supported on this CPU])
    else
      if test x"$can_compile_avx512_xeon_phi" = x"yes"; then
        AC_DEFINE([HAVE_AVX512_XEON_PHI],[1],[AVX512 for Xeon-PHI is supported on this CPU])
      else
        AC_MSG_ERROR([Oho! We can neither compile AVX512 intrinsics for Xeon nor Xeon Phi. This should not happen!])
      fi
    fi
  fi
1124
fi
1125
AC_LANG_POP([C])
1126
1127


1128
1129
1130
1131
1132
1133
1134
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)
1135

1136
1137
1138
1139
1140
1141
1142
1143
1144
    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
1145
1146
fi

Andreas Marek's avatar
Andreas Marek committed
1147

1148
1149
1150
1151
1152
1153
1154
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
1155

1156
1157
1158
1159
1160
1161
1162
1163
1164
    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
1165
fi
1166
AC_LANG_POP([Fortran])
1167

1168
1169
1170
1171
1172
1173
1174
1175
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)
1176

1177
1178
1179
1180
1181
  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
1182

1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
  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])
1193
1194
fi

Andreas Marek's avatar
Andreas Marek committed
1195

1196
1197
1198
1199
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])
1200
1201
1202
    kernel_defined=1
  else
    kernel_defined=0
1203
  fi
1204
1205
  [ELPA_2STAGE_]m4_toupper(elpa_m4_kernel)[_COMPILED]=$kernel_defined
  AC_SUBST([ELPA_2STAGE_]m4_toupper(elpa_m4_kernel)[_COMPILED])
1206
])
Andreas Marek's avatar
Andreas Marek committed
1207
1208


1209
1210
AM_CONDITIONAL([WITH_GPU_VERSION],[test x"$use_real_gpu" = x"yes" -o x"$use_complex_gpu" = x"yes"])
if test x"$use_real_gpu" = x"yes" -o x"$use_complex_gpu" = x"yes" ; then
1211
  AC_DEFINE([WITH_GPU_VERSION],[1],[enable GPU support])
1212
  AC_DEFINE([WITH_GPU_KERNEL],[1],[GPU kernel should be build])
1213
1214
1215
1216
1217
  ELPA_2STAGE_COMPLEX_GPU_COMPILED=1
  ELPA_2STAGE_REAL_GPU_COMPILED=1
else
  ELPA_2STAGE_COMPLEX_GPU_COMPILED=0
  ELPA_2STAGE_REAL_GPU_COMPILED=0
Andreas Marek's avatar
Andreas Marek committed
1218
fi
1219
1220
AC_SUBST([ELPA_2STAGE_COMPLEX_GPU_COMPILED])
AC_SUBST([ELPA_2STAGE_REAL_GPU_COMPILED])
Andreas Marek's avatar
Andreas Marek committed
1221

1222
1223
LT_INIT

1224
1225
1226
1227
1228
1229
DX_PDF_FEATURE(OFF)
DX_PS_FEATURE(OFF)
DX_MAN_FEATURE(ON)
DX_HTML_FEATURE(ON)
DX_INIT_DOXYGEN([ELPA], [Doxyfile], [docs])

Andreas Marek's avatar
Andreas Marek committed
1230
AC_MSG_CHECKING(whether assumed size Fortran arrays should be used)
1231
AC_ARG_ENABLE([assumed-size],
1232
                AS_HELP_STRING([--disable-assumed-size],
1233
1234
                               [do NOT use assumed-size Fortran arrays]),
                               [if test x"$enableval" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
1235
                                  USE_ASSUMED_SIZE=yes
1236
1237
1238
                                else
                                  USE_ASSUMED_SIZE=no
                                fi],
1239
              [USE_ASSUMED_SIZE=yes])
1240
1241
1242
AC_MSG_RESULT([${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
1243
  AC_DEFINE([USE_ASSUMED_SIZE],[1],[for performance reasons use assumed size Fortran arrays, even if not debuggable])
1244
1245
fi

Andreas Marek's avatar
Andreas Marek committed
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
AC_MSG_CHECKING(whether Fortran2008 features should be enabled)
AC_ARG_ENABLE([Fortran2008-features],
              AS_HELP_STRING([--enable-Fortran2008-features],
                             [enables some Fortran 2008 features, default yes.]),
			     [
			      if test x"$enableval" = x"yes"; then
			        enable_fortran2008_features=yes
		              else
			        enable_fortran2008_features=no
			      fi
			      ],
Andreas Marek's avatar
Andreas Marek committed
1257
              [enable_fortran2008_features=yes])
Andreas Marek's avatar
Andreas Marek committed
1258
1259
1260
1261
1262
1263
AC_MSG_RESULT([${enable_fortran2008_features}])
AM_CONDITIONAL([USE_FORTRAN2008],[test x"$enable_fortran2008_features" = x"yes"])
if test x"${enable_fortran2008_features}" = x"yes"; then
 AC_DEFINE([USE_FORTRAN2008], [1], [use some Fortran 2008 features])
fi

Andreas Marek's avatar
Andreas Marek committed
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
AC_MSG_CHECKING(whether autotuning functionality should be enabled)
AC_ARG_ENABLE([autotuning],
              AS_HELP_STRING([--enable-autotuning],
                             [enables autotuning functionality, default yes.]),
			     [
			      if test x"$enableval" = x"yes"; then
			        enable_autotuning=yes
		              else
			        enable_autotuning=no
			      fi
			      ],
              [enable_autotuning=yes])
AC_MSG_RESULT([${enable_autotuning}])
AM_CONDITIONAL([ENABLE_AUTOTUNING],[test x"$enable_autotuning" = x"yes"])
if test x"${enable_autotuning}" = x"yes"; then
 AC_DEFINE([ENABLE_AUTOTUNING], [1], [enable autotuning functionality])
fi

AC_MSG_CHECKING(whether C tests should be provided)
AC_ARG_ENABLE([c-tests],
              AS_HELP_STRING([--enable-c-tests],
                             [enables the C tests for elpa, default yes.]),
			     [
			      if test x"$enableval" = x"yes"; then
			        enable_c_tests=yes
		              else
			        enable_c_tests=no
			      fi
			      ],
              [enable_c_tests=yes])
AC_MSG_RESULT([${enable_c_tests}])
AM_CONDITIONAL([ENABLE_C_TESTS],[test x"$enable_c_tests" = x"yes"])
if test x"${enable_c_tests}" = x"yes"; then
 AC_DEFINE([ENABLE_C_TESTS], [1], [enable C tests])
fi

Andreas Marek's avatar
Andreas Marek committed
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
AC_MSG_CHECKING(whether we build for K-computer)
AC_ARG_ENABLE([K-computer],
              AS_HELP_STRING([--enable-K-computer],
                             [enable builds on K-Computer, default no.]),
			     [if test x"$enableval"=x"yes"; then
			        enable_kcomputer=yes
		              else
				enable_kcomputer=no
			      fi],
              [enable_kcomputer=no])
AC_MSG_RESULT([${enable_kcomputer}])
AM_CONDITIONAL([BUILD_KCOMPUTER],[test x"$enable_kcomputer" = x"yes"])
if test x"${enable_kcomputer}" = x"yes"; then
 AC_DEFINE([BUILD_KCOMPUTER], [1], [build for K-Computer])
 FC_MODINC="-I"
  if test x"${USE_ASSUMED_SIZE}" = x"yes" ; then
    AC_MSG_ERROR(on K-computer you have to switch off assumed-size arrays!)
  fi
  if test x"${enable_fortran2008_features}" = x"yes" ; then
    AC_MSG_ERROR(on K-computer you have to switch off Fortran 2008 features!)
  fi
fi

1323
1324
1325
if test x"${want_single_precision}" = x"yes" ; then
  AC_DEFINE([WANT_SINGLE_PRECISION_REAL],[1],[build also single-precision for real calculation])
  AC_DEFINE([WANT_SINGLE_PRECISION_COMPLEX],[1],[build also single-precision for complex calculation])
1326
fi
1327
1328
AM_CONDITIONAL([WANT_SINGLE_PRECISION_REAL],[test x"$want_single_precision" = x"yes"])
AM_CONDITIONAL([WANT_SINGLE_PRECISION_COMPLEX],[test x"$want_single_precision" = x"yes"])
1329

1330
AC_SUBST([MPI_BINARY])
1331
1332
1333
1334
AC_SUBST([WITH_MKL])
AC_SUBST([WITH_BLACS])
AC_SUBST([FC_MODINC])
AC_SUBST([FC_MODOUT])
1335
1336
1337
AC_SUBST([OPENMP_CFLAGS])
AC_SUBST([OPENMP_FCFLAGS])
AC_SUBST([OPENMP_LDFLAGS])
1338
AC_SUBST([DOXYGEN_OUTPUT_DIR], [docs])
1339

1340
mkdir -p modules private_modules test_modules
1341

1342
1343
1344
1345
1346
1347
1348
1349
1350
1351

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

1352
1353
1354
1355
1356
1357
1358
1359
1360

# 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')
1361

1362
1363
if test x"${with_mpi}" = x"yes"; then
  if test x"${enable_openmp}" = x"yes"; then
1364
        SUFFIX="_openmp"
1365
  else
1366
        SUFFIX=""
1367
1368
1369
  fi
else
  if test x"${enable_openmp}" = x"yes"; then
1370
        SUFFIX="_onenode_openmp"
1371
  else
1372
        SUFFIX="_onenode"
1373
  fi
1374
fi
1375

1376
1377
AC_SUBST([SUFFIX])
AC_SUBST([PKG_CONFIG_FILE],[elpa${SUFFIX}-${PACKAGE_VERSION}.pc])
1378

1379
1380
AC_CONFIG_FILES([
  Makefile
1381
  Doxyfile
1382
  ${PKG_CONFIG_FILE}:elpa.pc.in
1383
  elpa/elpa_constants.h
1384
  elpa/elpa_version.h
1385
])
1386

1387
1388
1389
1390
m4_include([m4/ax_fc_check_define.m4])
AC_MSG_CHECKING([if workaround for broken preprocessor is needed])

need_manual_cpp=no
1391
1392
AX_FC_CHECK_DEFINE([__INTEL_COMPILER],[is_intel=yes],[])
AX_FC_CHECK_DEFINE([__PGI],[is_pgi=yes],[])
Lorenz Huedepohl's avatar
Lorenz Huedepohl committed
1393
1394
ACTUAL_FC="$FC"
AC_SUBST([ACTUAL_FC])
1395
1396
1397
1398
1399
1400
1401
1402

if test x"$is_intel" = x"yes" ; then
	need_manual_cpp=yes
fi
if test x"$is_pgi" = x"yes" ; then
	need_manual_cpp=yes
fi

1403
if test x"$need_manual_cpp" = x"yes" ; then
1404
1405
        AC_MSG_RESULT([yes])
        FC="\$(top_srcdir)/manual_cpp $FC"
1406
else
1407
        AC_MSG_RESULT([no])
1408
1409
fi