configure.ac 73.6 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
AC_CONFIG_MACRO_DIR([m4])
AC_CONFIG_HEADERS([config.h])
AM_SILENT_RULES([yes])
14
15
# ABI version
#
16
17
18
19
20
21
22
23
24
25
26
27
28
# 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)
#
29
AC_SUBST([ELPA_SO_VERSION], [15:2:0])
30

31
32
33
34
35
36
# AC_DEFINE_SUBST(NAME, VALUE, DESCRIPTION)
# -----------------------------------------
AC_DEFUN([AC_DEFINE_SUBST], [
AC_DEFINE([$1], [$2], [$3])
AC_SUBST([$1], ['$2'])
])
37

Andreas Marek's avatar
Andreas Marek committed
38
39
AX_BUILD_DATE_EPOCH(ELPA_BUILDTIME)

40
41
# API Version
AC_DEFINE([EARLIEST_API_VERSION], [20170403], [Earliest supported ELPA API version])
42

43
AC_DEFINE_SUBST(CURRENT_API_VERSION, 20200417, "Current ELPA API version")
Andreas Marek's avatar
Andreas Marek committed
44
45
# Autotune Version
AC_DEFINE([EARLIEST_AUTOTUNE_VERSION], [20171201], [Earliest ELPA API version, which supports autotuning])
46
47
AC_DEFINE([CURRENT_AUTOTUNE_VERSION], [20200417], [Current ELPA autotune version])
AC_DEFINE_SUBST(CURRENT_AUTOTUNE_VERSION, 20200417, "Current ELPA autotune version")
Andreas Marek's avatar
Andreas Marek committed
48
AC_DEFINE_UNQUOTED([ELPA_BUILDTIME], [$ELPA_BUILDTIME], ["Time of build"])
Andreas Marek's avatar
Andreas Marek committed
49
AX_COMPARE_VERSION([$ELPA_BUILDTIME], [gt], [1623715200],[old_elpa_version=yes],[old_elpa_version=no])
50

51
52
53
54
55
AX_CHECK_GNU_MAKE()
if test x$_cv_gnu_make_command = x ; then
        AC_MSG_ERROR([Need GNU Make])
fi

56
enable_legacy=no
Andreas Marek's avatar
Andreas Marek committed
57
58
59
60
61
62
63
64
65
66
67
68
69
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}])
Andreas Marek's avatar
Andreas Marek committed
70
AM_CONDITIONAL([OPTIONAL_C_ERROR_ARGUMENT],[test x"$optional_c_error_argument" = x"yes"])
Andreas Marek's avatar
Andreas Marek committed
71
72
73
74
if test x"${optional_c_error_argument}" = x"yes"; then
        AC_DEFINE([OPTIONAL_C_ERROR_ARGUMENT], [1], [enable error argument in C-API to be optional])
fi

75
76
77
78
79

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

80

81
dnl OpenMP
82
m4_include([m4/ax_elpa_openmp.m4])
83
84
AC_MSG_CHECKING(whether --enable-openmp is specified)
AC_ARG_ENABLE([openmp],
85
86
              AS_HELP_STRING([--enable-openmp],
                             [use OpenMP threading, default no.]),
Andreas Marek's avatar
Andreas Marek committed
87
88
89
90
91
92
93
              [
	       if test x"$enableval" = x"yes"; then
	         enable_openmp=yes
	       else
	         enable_openmp=no
	       fi
	       ],
94
              [enable_openmp=no])
95
AC_MSG_RESULT([${enable_openmp}])
96
AM_CONDITIONAL([WITH_OPENMP_TRADITIONAL],[test x"$enable_openmp" = x"yes"])
97
if test x"${enable_openmp}" = x"yes"; then
98
        AC_DEFINE([WITH_OPENMP_TRADITIONAL], [1], [use OpenMP threading])
99
100
fi

101

102
103
dnl mpi
AC_ARG_WITH(mpi, [AS_HELP_STRING([--with-mpi=[[yes|no]]], [compile with MPI. Default: yes])],,[with_mpi=yes])
104
AM_CONDITIONAL([WITH_MPI],[test x"$with_mpi" = x"yes"])
105
if test x"${with_mpi}" = x"yes"; then
106
  AC_DEFINE([WITH_MPI], [1], [use MPI])
107
108
fi

109
dnl C
110
AC_LANG_PUSH([C])
111

112

113
114
115
116
117
118
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
119

120
if test x"${enable_openmp}" = x"yes"; then
121
122
123
124
125
  AX_ELPA_OPENMP
  if test "$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"
126
fi
127

Andreas Marek's avatar
Andreas Marek committed
128
129
130
c11_standard=no
AX_CHECK_COMPILE_FLAG([-std=gnu11], [
  c11_standard=yes
131
], [
Andreas Marek's avatar
Andreas Marek committed
132
133
  echo "C compiler cannot compile -std=gnu11 code"
  echo "testing -std=c11.."
134
])
Andreas Marek's avatar
Andreas Marek committed
135
if test x"$c11_standard" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
136
137
138
  CFLAGS+=" -std=gnu11"
fi

Andreas Marek's avatar
Andreas Marek committed
139
if test x"$c11_standard" = x"no"; then
Andreas Marek's avatar
Andreas Marek committed
140
141
142
  AX_CHECK_COMPILE_FLAG([-std=c11], [
    c11_standard=yes
  ], [
143
144
145
146
147
148
149
150
151
152
153
154
155
    echo "C compiler cannot compile -std=c11 code"
    echo "testing -c11.."
  ])
  if test x"$c11_standard" = x"yes"; then
    CFLAGS+=" -std=c11"
  fi
fi

if test x"$c11_standard" = x"no"; then
  AX_CHECK_COMPILE_FLAG([-c11], [
    c11_standard=yes
  ], [
    echo "C compiler cannot compile -c11 code"
Andreas Marek's avatar
Andreas Marek committed
156
157
158
    echo "C compiler cannot compile C11 code"
    exit -1
  ])
Andreas Marek's avatar
Andreas Marek committed
159
  if test x"$c11_standard" = x"yes"; then
160
    CFLAGS+=" -c11"
Andreas Marek's avatar
Andreas Marek committed
161
162
  fi
fi
163

164
165
AX_EXT

Andreas Marek's avatar
Andreas Marek committed
166
167
168
169
170
dnl heterogenous-cluster-support
AC_MSG_CHECKING(whether heterogenous-cluster-support should be enabled)
AC_ARG_ENABLE([heterogenous-cluster-support],
              AS_HELP_STRING([--heterogenous-cluster-support],
                             [allow ELPA to automatically switch to a kernel supported on all CPUs of a cluster (only works for Intel CPUs at the moment), default no. Activate only if necessary has a performance penalty! This feature is exerpimental!]),
171
172
              [
	       if test x"$enableval" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
173
	         enable_heterogenous_cluster_support=yes
174
	       else
Andreas Marek's avatar
Andreas Marek committed
175
	         enable_heterogenous_cluster_support=no
176
177
	       fi
	       ],
Andreas Marek's avatar
Andreas Marek committed
178
179
180
181
              [enable_heterogenous_cluster_support="no"])
AC_MSG_RESULT([$enable_heterogenous_cluster_support])
if test x"${enable_heterogenous_cluster_support}" = x"yes"; then
  AC_DEFINE([HAVE_HETEROGENOUS_CLUSTER_SUPPORT], [1], [automatically support clusters with different Intel CPUs])
182
fi
Andreas Marek's avatar
Andreas Marek committed
183
AM_CONDITIONAL([HAVE_HETEROGENOUS_CLUSTER_SUPPORT],[test x"$enable_heterogenous_cluster_support" = x"yes"])
184

Andreas Marek's avatar
Andreas Marek committed
185
186
187
188
189
dnl 64bit integer support for BLACS/LAPACK/SCALAPACK support
dnl first long int
AC_CHECK_SIZEOF([long int])
size_of_long_int="${ac_cv_sizeof_long_int}"

190
dnl then 64bit blas
Andreas Marek's avatar
WIP    
Andreas Marek committed
191
AC_MSG_CHECKING(whether 64bit integers should be used for math libraries (BLAS/LAPACK/SCALAPACK))
192
193
AC_ARG_ENABLE([64bit-integer-math-support],
              AS_HELP_STRING([--64bit-integer-math-support],
Andreas Marek's avatar
WIP    
Andreas Marek committed
194
195
196
                             [allows to link against the 64bit integer versions of the math libraries BLAS, LAPACK, and SCALAPACK]),
              [
	       if test x"$enableval" = x"yes"; then
197
	         enable_64bit_integer_math_support=yes
Andreas Marek's avatar
WIP    
Andreas Marek committed
198
	       else
199
	         enable_64bit_integer_math_support=no
Andreas Marek's avatar
WIP    
Andreas Marek committed
200
201
	       fi
	       ],
202
203
204
              [enable_64bit_integer_math_support="no"])
AC_MSG_RESULT([$enable_64bit_integer_math_support])
if test x"${enable_64bit_integer_math_support}" = x"yes"; then
205
206
207
208
  dnl at least INTEL MPI does _NOT_ support 64BIT integer mode  for C thus disable C tests in this Case
  if test x"${enable_c_tests}" = x"yes"; then
     AC_MSG_ERROR([You cannot both define 64bit integer support and C tests. Reconfigure!])
  fi
Andreas Marek's avatar
Andreas Marek committed
209
210
211
212
213
214
215
  dnl check whether long int is the correct data-type in C
  if test x"${size_of_long_int}" = x"8"; then
    echo "Found C data-type \"long int\" with 8 bytes"
  else
    AC_MSG_ERROR([The C data-type "long int" is only ${size_of_long_int} bytes; Needed is 8 bytes])
  fi

216
  AC_DEFINE([HAVE_64BIT_INTEGER_MATH_SUPPORT], [1], [allow to link against the 64bit integer versions of math libraries])
Andreas Marek's avatar
WIP    
Andreas Marek committed
217
fi
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
AM_CONDITIONAL([HAVE_64BIT_INTEGER_MATH_SUPPORT],[test x"$enable_64bit_integer_math_support" = x"yes"])

dnl then 64bit blas
AC_MSG_CHECKING(whether 64bit integers should be used for the MPI library)
AC_ARG_ENABLE([64bit-integer-mpi-support],
              AS_HELP_STRING([--64bit-integer-mpi-support],
                             [allows to link against the 64bit integer versions of the MPI library]),
              [
	       if test x"$enableval" = x"yes"; then
	         enable_64bit_integer_mpi_support=yes
	       else
	         enable_64bit_integer_mpi_support=no
	       fi
	       ],
              [enable_64bit_integer_mpi_support="no"])
AC_MSG_RESULT([$enable_64bit_integer_mpi_support])
if test x"${enable_64bit_integer_mpi_support}" = x"yes"; then
  AC_DEFINE([HAVE_64BIT_INTEGER_MPI_SUPPORT], [1], [allow to link against the 64bit integer versions of the MPI library])
fi
AM_CONDITIONAL([HAVE_64BIT_INTEGER_MPI_SUPPORT],[test x"$enable_64bit_integer_mpi_support" = x"yes"])

239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
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])

262
AC_PROG_INSTALL
263
264
AM_PROG_AR
AM_PROG_AS
265
AC_PROG_CC_C99
266
AM_PROG_CC_C_O
267
AC_LANG_POP([C])
Andreas Marek's avatar
Andreas Marek committed
268
269

# Fortran
270
AC_LANG_PUSH([Fortran])
271

Andreas Marek's avatar
Andreas Marek committed
272
m4_include([m4/ax_prog_fc_mpi.m4])
273
274
275
276
277
278
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
279

280
AC_FC_SRCEXT([F90])
281
282
283
284
285
AC_FC_FREEFORM
AC_FC_MODULE_FLAG
AC_FC_MODULE_OUTPUT_FLAG
AC_FC_LIBRARY_LDFLAGS

286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
AC_MSG_CHECKING(whether Fortran compiler infers interfaces )
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
program test
implicit none
integer :: a
real*8  :: b

  call foo(a)
  call foo(b)

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

if test x"${FC_does_infer_interfaces}" = x"yes"; then
304
305
306
307
308
309
310
  # two problems
  # 1. some MPIs (at least Intel) do not provide all
  #    interfaces, which will fail
  #    in case of MPI and not all interfaces -> switch of
  #    argument checking or fail
  # 2. no MPI case: switch of PACK_REAL_TO_COMPLEX
  if test x"${with_mpi}" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
311
    AC_MSG_CHECKING(whether MPI module defines all interfaces )
312
313
314
315
316
317
318
319
    # check whether MPI module defines all interfaces; not the case for Intel MPI!
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
    program test_mpi_interfaces
      use mpi
      implicit none
      integer :: rank
      integer :: buf(10)
      integer :: ierr
320
321
      real*8  :: a(2)
      complex*16 :: b(2)
322
323
324
325
326
327
328
329

      call MPI_Init(ierr)
      call MPI_Comm_rank(MPI_COMM_WORLD, rank, ierr)

      if (rank == 0) then
         buf(:) = 42;
      end if

330
331
332
      
      !! this is OK
      !call MPI_Bcast(buf, 10, MPI_INT, 0, MPI_COMM_WORLD, ierr)
333

334
335
      !! Oops, wrong order here:
      !call MPI_Bcast(10, buf, MPI_INT, 0, MPI_COMM_WORLD, ierr)
336

337
338
339
340
341
      ! if the correct interfaces exists in the MPI module
      ! this must work. If not and the compiler infers interfaces
      ! this will fail
      call MPI_Bcast(a, 2, MPI_REAL8, 0, MPI_COMM_WORLD, ierr)
      call MPI_Bcast(b, 2, MPI_DOUBLE_COMPLEX, 0, MPI_COMM_WORLD, ierr)
342
343
344
      call MPI_Finalize(ierr)
    end program
      ])],
345
346
      [MPImodule_has_interfaces=yes],
      [MPImodule_has_interfaces=no]
347
    )
348
    AC_MSG_RESULT([${MPImodule_has_interfaces}])
349
    if test x"${MPImodule_has_interfaces}" = x"no"; then
350
      AC_MSG_CHECKING(whether we can cure missing interfaces by allowing argument mismatch )
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
      #check whether we can cure this by disabling argument checking
      FCFLAGS_SAVE2="$FCFLAGS"
      FCFLAGS="$FCFLAGS -fallow-argument-mismatch"
      # test again
      AC_COMPILE_IFELSE([AC_LANG_SOURCE([
      program test
      implicit none
      integer :: a
      real*8  :: b
      
        call foo(a)
        call foo(b)
      
      end program

366
      ])],
367
368
      [FC_infer_interfaces_cured=yes],
      [FC_infer_interfaces_cured=no]
369
      )
370
371
      AC_MSG_RESULT([${FC_infer_interfaces_cured}])
      if test x"${FC_infer_interfaces_cured}" = x"no"; then
372
373
374
375
376
377
378
        FCFLAGS = "$FCFLAGS_SAVE2"
        AC_MSG_ERROR([Fortran compiler infers interfaces; but MPI module does not supply all of them])
      fi
    fi
  #else
  #  # no MPI case switch of PACK_REAL_TO_COMPLEX
  #  # i.e. do _not_ define it
379
380
381
382
383
  fi
else
  AC_DEFINE([PACK_REAL_TO_COMPLEX], [1], [In some kernels pack real to complex])
fi

384
if test x"${enable_openmp}" = x"yes"; then
385
386
387
388
389
  AX_ELPA_OPENMP
  if test "$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"
390
391
fi

392
393
dnl check which MPI binray invokes a MPI job
if test x"$with_mpi" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
394
  AC_CHECK_PROGS([MPI_BINARY], [mpiexec.hydra mpiexec mpirun poe runjob srun aprun], [no])
395
  if test x"$MPI_BINARY" = x"no"; then
Andreas Marek's avatar
Andreas Marek committed
396
    AC_MSG_ERROR([Could not find either of the MPI binaries: mpiexec.hydra, mpiexec, mpirun, poe, runjob, srun, aprun])
397
398
399
  fi
fi

400
401
402
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)
403
AC_ARG_ENABLE([redirect],
404
405
              [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
406
407
408
409
410
411
412
              [
               if test x"$enableval" = x"yes"; then
                 enable_redirect=yes
               else
                 enable_redirect=no
               fi
	       ],
413
              [enable_redirect=no])
414
415
AC_MSG_RESULT([${enable_redirect}])

416
417
418
419
420
421
422
423
424
425
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}])


426
427
dnl redirect
if test x"${enable_redirect}" = x"yes"; then
428
429
  AC_DEFINE([HAVE_REDIRECT], [1], [Redirect stdout and stderr of test programs per MPI tasks to a file])
fi
430
431
432
AM_CONDITIONAL([HAVE_REDIRECT],[test x"$enable_redirect" = x"yes"])

dnl build with ftimings support
433

Andreas Marek's avatar
Andreas Marek committed
434
AC_ARG_ENABLE([timings],
435
436
              [AS_HELP_STRING([--disable-timings],
                              [more detailed timing, default yes])],
Andreas Marek's avatar
Andreas Marek committed
437
438
439
440
441
442
443
              [
               if test x"$enableval" = x"yes"; then
                 enable_timings=yes
               else
                 enable_timings=no
               fi
	       ],
444
              [enable_timings=yes])
445
446
447
448

if test x"${enable_timings}" = x"yes"; then
  AC_DEFINE([HAVE_DETAILED_TIMINGS], [1], [Enable more timing])
fi
449
AM_CONDITIONAL([HAVE_DETAILED_TIMINGS], [test x"$enable_timings" = x"yes"])
450
451

dnl PAPI for ftimings
452
AC_LANG_PUSH([C])
453
AC_ARG_WITH([papi],
454
455
            [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
456
            [
Wenzhe Yu's avatar
Wenzhe Yu committed
457
               if test x"$withval" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
458
459
460
461
462
                 with_papi=yes
               else
                 with_papi=no
               fi
	     ],
463
            [with_papi="no"])
464
465
if test x"${enable_timings}" = x"yes"; then
  if test x"$with_papi" = x"yes" ; then
466
    AC_SEARCH_LIBS([PAPI_library_init], [papi], [papi_found=yes], [papi_found=no])
467
468
    if test x"$papi_found" = x"no" ; then
      AC_MSG_ERROR(["Could not find usable PAPI installation, please install or adjust CFLAGS, LDFLAGS"])
469
470
471
472
    fi
    AC_DEFINE([HAVE_LIBPAPI], [1], [Use the PAPI library])
  fi
fi
473
AC_LANG_POP([C])
474

Pavel Kus's avatar
Pavel Kus committed
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
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])

520
521
522
save_FCFLAGS=$FCFLAGS
save_LDFLAGS=$LDFLAGS

523
524
525
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])

526
527
FCFLAGS="$FCFLAGS $SCALAPACK_FCFLAGS"
LDFLAGS="$LDFLAGS $SCALAPACK_LDFLAGS"
528

529
dnl check whether fortran error_unit is defined
530
AC_MSG_CHECKING([whether Fortran module iso_fortran_env is available])
531
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
532
  program test_error_unit
533
    use iso_fortran_env, only : error_unit
534
535
536
537
538
539
540
541
    implicit none

    write(error_unit,*) "error_unit is defined"
  end program
])],
  [can_use_iso_fortran_env=yes],
  [can_use_iso_fortran_env=no]
)
542
AC_MSG_RESULT([${can_use_iso_fortran_env}])
543
544
545
546
if test x"${can_use_iso_fortran_env}" = x"yes" ; then
 AC_DEFINE([HAVE_ISO_FORTRAN_ENV],[1],[can use module iso_fortran_env])
fi

547

548
549
dnl check whether one can link with specified MKL (desired method)
AC_MSG_CHECKING([whether we can compile a Fortran program using MKL])
550
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
551
  program test_mkl
552
    use mkl_service
553
554
555
556
557
558
559
560
561
562
563
564
    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"$can_compile_with_mkl" = x"yes" ; then
  AC_MSG_CHECKING([whether we can link a Fortran program with MKL])
565
  AC_LINK_IFELSE([AC_LANG_SOURCE([
566
    program test_mkl
567
      use mkl_service
568
569
570
571
572
      character*198 :: string
      call mkl_get_version_string(string)
      write(*,'(a)') string
    end program
  ])],
Andreas Marek's avatar
Andreas Marek committed
573
574
    [have_mkl=yes],
    [have_mkl=no]
575
  )
Andreas Marek's avatar
Andreas Marek committed
576
  AC_MSG_RESULT([${have_mkl}])
577
578
579
fi

dnl if not mkl, check all the necessary individually
580
if test x"${have_mkl}" = x"yes" ; then
581
582
583
584
  WITH_MKL=1
else

  dnl first check blas
585
  AC_SEARCH_LIBS([dgemm],[flexiblas openblas satlas blas],[have_blas=yes],[have_blas=no])
Andreas Marek's avatar
Andreas Marek committed
586
587
  AC_MSG_CHECKING([whether we can link a program with a blas lib])
  AC_MSG_RESULT([${have_blas}])
588

589
  if test x"${have_blas}" = x"no" ; then
590
    AC_MSG_ERROR([could not link with blas: specify path])
591
  fi
592
  dnl now lapack
Andreas Marek's avatar
Andreas Marek committed
593
594
595
  AC_SEARCH_LIBS([dlarrv],[lapack],[have_lapack=yes],[have_lapack=no])
  AC_MSG_CHECKING([whether we can link a program with a lapack lib])
  AC_MSG_RESULT([${have_lapack}])
596

597
  if test x"${have_lapack}" = x"no" ; then
598
    AC_MSG_ERROR([could not link with lapack: specify path])
599
  fi
Andreas Marek's avatar
Andreas Marek committed
600

601
  if test x"${with_mpi}" = x"yes"; then
602
    dnl test whether scalapack already contains blacs
603
    scalapack_libs="mpiscalapack scalapack scalapack-openmpi"
604
605
606
607
608
    old_LIBS="$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
609
	       
610
611
612
613
614
      AC_MSG_RESULT([${blacs_in_scalapack}])
      if test x"${blacs_in_scalapack}" = x"yes"; then
        break
      fi
    done
615

616
617
    if test x"${blacs_in_scalapack}" = x"no"; then
      LIBS="${old_LIBS}"
618

619
620
      dnl Test for stand-alone blacs
      AC_SEARCH_LIBS([bi_f77_init],[mpiblacsF77init],[],[],[-lmpiblacs])
621
622
623
624
625
626

      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
627
      AC_SEARCH_LIBS([blacs_gridinit],[mpiblacs blacs],[have_blacs=yes],[have_blacs=no])
Andreas Marek's avatar
Andreas Marek committed
628

629
630
631
      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
632
    fi
633

634
    AC_SEARCH_LIBS([pdtran],[$scalapack_libs],[have_scalapack=yes],[have_scalapack=no])
Andreas Marek's avatar
Andreas Marek committed
635

636
637
638
    if test x"${have_scalapack}" = x"no" ; then
      AC_MSG_ERROR([could not link with scalapack: specify path])
    fi
639
640
641
642
  fi

  dnl check whether we can link alltogehter
  AC_MSG_CHECKING([whether we can link a Fortran program with all blacs/scalapack])
643
  AC_LINK_IFELSE([AC_LANG_SOURCE([
644
645
    program dgemm_test

646
647
      integer , parameter:: M = 4, N = 3, K = 2
      real :: A(M,K), B(K,N), C(M,N)
648

649
      call dgemm('N','N',M,N,K,1.0,A,M,B,K,0.0,C,M)
650
651
652
653
654
655
656
657

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

658
   if test x"${can_link_with_blacs_scalapack}" = x"yes" ; then
659
660
     WITH_BLACS=1
   else
661
   AC_MSG_ERROR([We can neither link with MKL or another Scalpack. Please specify SCALAPACK_LDFLAGS and SCALAPACK_FCFLAGS!])
662
663
664
   fi
fi

665
666
667
668
dnl important: reset them again!
FCFLAGS=$save_FCFLAGS
LDFLAGS=$save_LDFLAGS

669

Andreas Marek's avatar
Andreas Marek committed
670
671
672
673
dnl check for intrinsic fortran function of 2003 standard
AC_MSG_CHECKING([whether we can use the intrinsic Fortran function "get_environment_variable"])


674
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
675
676
677
678
679
680
681
682
683
  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([${fortran_can_check_environment}])
684
685
686
if test x"${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
687

688

689
690
dnl check whether BAND_TO_FULL_BLOCKING is set
AC_MSG_CHECKING(whether BAND_TO_FLULL_BLOCKING is requested)
691
AC_ARG_ENABLE(band-to-full-blocking,[AS_HELP_STRING([--disable-band-to-full-blocking],
692
                                                    [build ELPA2 with blocking in band_to_full (default: enabled)])],
Andreas Marek's avatar
Andreas Marek committed
693
694
695
696
697
698
699
              [
               if test x"$enableval" = x"yes"; then
                 enable_band_to_full_blocking=yes
               else
                 enable_band_to_full_blocking=no
               fi
	       ],
700
701
	      [enable_band_to_full_blocking="yes"])
AC_MSG_RESULT([${enable_band_to_full_blocking}])
702

703
704
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
705
       AC_DEFINE([BAND_TO_FULL_BLOCKING], [1], [use blocking in trans_ev_band_to_full])
706
707
fi

708
709
710
711
712
713
714

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])

715

716
dnl setup nvcc flags and use them in later tests
717
718
719
user_sets_gpu_compute_capability="no"
AC_MSG_CHECKING(whether a GPU compute capability is specified)
AC_ARG_WITH([GPU-compute-capability],
720
721
722
            [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"])
723
724
AC_MSG_RESULT([${user_sets_gpu_compute_capability}])

725

726
727
728
729
730
731
732
733
734
735
736
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
737

738
739
dnl Test possibility of 'use mpi', if requested
if test x"${with_mpi}" = x"yes" ; then
740
741
  AC_ARG_ENABLE([mpi-module],
                AS_HELP_STRING([--disable-mpi-module],
Pavel Kus's avatar
Pavel Kus committed
742
                               [do not use the Fortran MPI module, get interfaces by 'include "mpif.h']),
Andreas Marek's avatar
Andreas Marek committed
743
744
745
746
747
748
749
                [
                 if test x"$enableval" = x"yes"; then
                   enable_mpi_module=yes
                 else
                   enable_mpi_module=no
                 fi
		 ],
750
751
                [enable_mpi_module=yes])
  if test x"${enable_mpi_module}" = x"yes" ; then
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
    AC_MSG_CHECKING(whether Fortran mpi module can be used)
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
      program test_mpi_module
        use mpi
        real :: time
        time = MPI_WTime()
      end program
    ])],
      [can_use_fortran_mpi_module=yes],
      [can_use_fortran_mpi_module=no]
    )
    AC_MSG_RESULT([${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
767
      AC_MSG_ERROR([Could not compile a Fortran program with an 'use mpi' statement. You can try again with --disable-mpi-module])
768
    fi
769
  fi
Andreas Marek's avatar
Andreas Marek committed
770
fi
771
AC_LANG_POP([Fortran])
Andreas Marek's avatar
Andreas Marek committed
772

773
774
dnl Assemble the list of kernels to build
m4_pattern_forbid([elpa_m4])
Andreas Marek's avatar
Andreas Marek committed
775

776
777
778
m4_define(elpa_m4_generic_kernels, [
        real_generic
        real_generic_simple
Andreas Marek's avatar
Andreas Marek committed
779
        real_generic_simple_block4
780
        real_generic_simple_block6
781
782
783
        complex_generic
        complex_generic_simple
])
784

785
786
787
788
m4_define(elpa_m4_sse_assembly_kernels, [
        real_sse_assembly
        complex_sse_assembly
])
789

790
791
792
793
794
795
796
m4_define(elpa_m4_sse_kernels, [
        real_sse_block2
        real_sse_block4
        real_sse_block6
        complex_sse_block1
        complex_sse_block2
])
797

798
799
800
801
802
803
m4_define(elpa_m4_sparc64_kernels, [
        real_sparc64_block2
        real_sparc64_block4
        real_sparc64_block6
])

804
805
806
807
m4_define(elpa_m4_neon_arch64_kernels, [
        real_neon_arch64_block2
        real_neon_arch64_block4
        real_neon_arch64_block6
808
809
        complex_neon_arch64_block1
        complex_neon_arch64_block2
810
811
])

812
813
814
815
816
817
m4_define(elpa_m4_vsx_kernels, [
        real_vsx_block2
        real_vsx_block4
        real_vsx_block6
])

818
819
820
821
822
823
824
825
m4_define(elpa_m4_sve128_kernels, [
        real_sve128_block2
        real_sve128_block4
        real_sve128_block6
        complex_sve128_block1
        complex_sve128_block2
])

826
827
828
829
830
831
832
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
833

834
835
836
837
838
839
840
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
841

842
843
844
845
846
847
848
849
m4_define(elpa_m4_sve256_kernels, [
        real_sve256_block2
        real_sve256_block4
        real_sve256_block6
        complex_sve256_block1
        complex_sve256_block2
])

850
851
852
853
854
855
856
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
857

858
859
860
861
862
863
m4_define(elpa_m4_sve512_kernels, [
        real_sve512_block2
        real_sve512_block4
        real_sve512_block6
        complex_sve512_block1
        complex_sve512_block2
Andreas Marek's avatar
Andreas Marek committed
864
865
])

866
867
868
869
m4_define(elpa_m4_bgp_kernels, [
        real_bgp
        complex_bgp
])
Andreas Marek's avatar
Andreas Marek committed
870

871
872
873
874
m4_define(elpa_m4_bgq_kernels, [
        real_bgq
        complex_bgq
])
Andreas Marek's avatar
Andreas Marek committed
875

Andreas Marek's avatar
Andreas Marek committed
876
877
878
879
m4_define(elpa_m4_gpu_kernels, [
        real_gpu
        complex_gpu
])
880

881
m4_define(elpa_m4_kernel_types, [generic sparc64 neon_arch64 vsx sse sse_assembly sve128 avx avx2 sve256 avx512 sve512 bgp bgq gpu])
882

883
884
885
886
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
887

888
889
890
891
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
892

893
894
895
896
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) ]))
897

898
899
900
901
902
903
904
905
906
907
908
909
910
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]
        ])
])
911

912
913
dnl Modify list of kernels with configure arguments
ELPA_SELECT_KERNELS([generic],[enable])
914
ELPA_SELECT_KERNELS([sparc64],[disable])
915
ELPA_SELECT_KERNELS([neon_arch64],[disable])
916
ELPA_SELECT_KERNELS([vsx],[disable])
917
918
919
920
ELPA_SELECT_KERNELS([sse],[enable])
ELPA_SELECT_KERNELS([sse_assembly],[enable])
ELPA_SELECT_KERNELS([avx],[enable])
ELPA_SELECT_KERNELS([avx2],[enable])
921
ELPA_SELECT_KERNELS([avx512],[enable])
922
923
ELPA_SELECT_KERNELS([sve128],[disable])
ELPA_SELECT_KERNELS([sve256],[disable])
924
ELPA_SELECT_KERNELS([sve512],[disable])
Andreas Marek's avatar
Andreas Marek committed
925
ELPA_SELECT_KERNELS([gpu],[disable])
926
927
928
929
930
931
ELPA_SELECT_KERNELS([bgp],[disable])
ELPA_SELECT_KERNELS([bgq],[disable])

m4_foreach_w([elpa_m4_kind],[real complex],[
        default_[]elpa_m4_kind[]_kernel=""
])
932

933
if test x"${enable_bgp}" = x"yes" -o x"$enable_bgq" = x"yes"; then
934
        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 elpa_m4_sve128_kernels elpa_m4_sve256_kernels elpa_m4_sve512_kernels, [
935
936
937
938
939
940
941
942
943
944
945
946
                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:]
947
948
                 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[_$withval"],[fixed_]elpa_m4_kind[_kernel=""])
949
        if test -n "$fixed_[]elpa_m4_kind[]_kernel" ; then
950
                found="no"
951
952
953
                m4_foreach_w([elpa_m4_otherkernel],m4_expand(elpa_m4_[]elpa_m4_kind[]_kernels),[
                        if test "$fixed_]elpa_m4_kind[_kernel" = "]elpa_m4_otherkernel[" ; then
                                use_[]elpa_m4_otherkernel[]=yes
954
                                found="yes"
955
956
957
958
                        else
                                use_[]elpa_m4_otherkernel[]=no
                        fi
                ])
959
960
961
                if test x"$found" = x"no" ; then
                        AC_MSG_ERROR([Invalid kernel "$fixed_]elpa_m4_kind[_kernel" specified for --with-fixed-]elpa_m4_kind[-kernel])
                fi
962
963
964
965
                default_[]elpa_m4_kind[]_kernel="$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
])
966

Andreas Marek's avatar
Andreas Marek committed
967
968
969
970
971
972
973
974
975
976
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_real_gpu=yes
        use_complex_gpu=yes
fi
977

978

979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
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
])
994
m4_foreach_w([elpa_m4_arch],[sparc64 neon_arch64 vsx sse avx avx2 avx512 sve128 sve256 sve512],[
995
996
997
998
        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])
])
999

1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
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
])
1020

1021

1022
1023
1024
dnl the list of kernels is now assembled
dnl choosing a default kernel

1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
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
])



1049
1050
1051
m4_foreach_w([elpa_m4_kind],[real complex],[
        m4_foreach_w([elpa_m4_kernel],
                     m4_foreach_w([elpa_m4_cand_kernel],
1052
                                  elpa_m4_avx512_kernels elpa_m4_avx2_kernels elpa_m4_avx_kernels elpa_m4_sse_kernels elpa_m4_sse_assembly_kernels elpa_m4_sve128_kernels elpa_m4_sve256_kernels elpa_m4_sve512_kernels elpa_m4_sparc64_kernels elpa_m4_neon_arch64_kernels elpa_m4_vsx_kernels elpa_m4_generic_kernels elpa_m4_gpu_kernels,
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
                                  [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 | \
1066
                                                                perl -pe 's/^[[^,]]*, *//; s/,.*//;'`
1067
1068
        AC_SUBST([ELPA_2STAGE_]m4_toupper(elpa_m4_kind)[_DEFAULT])
])
Andreas Marek's avatar
Andreas Marek committed
1069

1070
1071
1072
1073
1074
1075
1076
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 }
1077
AC_LANG_PUSH([C])
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
dnl check whether one can link against Fortran programs from C
AC_MSG_CHECKING(whether we need _ in C programs to link against a Fortran library)
AC_LINK_IFELSE([AC_LANG_SOURCE([
int main(int argc, char **argv) {

  int m, n, k, lda, ldb, ldc;
  double alpha, beta;
  double *a, *b, *c;
  dgemm_("N", "N", &m, &n, &k, &alpha, a, lda, b, &ldb, &beta, c, &ldc);
}
 ])],
   [can_link_with_=yes],
   [can_link_with_=no]
 )
AC_MSG_RESULT([${can_link_with_}])
if test x"$can_link_with_" = x"yes"; then
  AC_DEFINE([NEED_UNDERSCORE_TO_LINK_AGAINST_FORTRAN],[1],[need to append an underscore])
fi
  
1097
AC_MSG_CHECKING(whether we can link C programs against Fortran without _ )
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
AC_LINK_IFELSE([AC_LANG_SOURCE([
int main(int argc, char **argv) {

  int m, n, k, lda, ldb, ldc;
  double alpha, beta;
  double *a, *b, *c;
  dgemm("N", "N", &m, &n, &k, &alpha, a, lda, b, &ldb, &beta, c, &ldc);
}
 ])],
   [can_link_without_=yes],
   [can_link_without_=no]
 )
AC_MSG_RESULT([${can_link_without_}])
1111
if test x"$can_link_without_" = x"yes"; then
1112
1113
1114
  AC_DEFINE([NEED_NO_UNDERSCORE_TO_LINK_AGAINST_FORTRAN],[1],[need not to append an underscore])
fi

1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
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

1135
1136
1137
if test x"${need_sparc64}" = x"yes"; then
  AC_MSG_CHECKING(whether we can compile SPARC64 with intrinsics in C)
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
1138
1139
#include <fjmfunc.h>
#include <emmintrin.h>
1140
int main(int argc, char **argv) {
Andreas Marek's avatar
Andreas Marek committed
1141
1142
  __m128d  tau1;
  __m128d h1 = _fjsp_neg_v2r8(tau1);
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
  return 0;
}
    ])],
    [can_compile_sparc64=yes],
    [can_compile_sparc64=no]
  )
  AC_MSG_RESULT([${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
1155

1156
1157
1158
1159
1160
1161
if test x"${need_neon_arch64}" = x"yes"; then
  AC_MSG_CHECKING(whether we can compile NEON ARCH64 with intrinsics in C)
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
#include <arm_neon.h>
int main(int argc, char **argv) {
  __Float64x2_t  x1, x2, x3, x4;
Andreas Marek's avatar
Andreas Marek committed
1162
  x4 = vfmaq_f64(x1, x2, x3);
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
  return 0;
}
    ])],
    [can_compile_neon_arch64=yes],
    [can_compile_neon_arch64=no]
  )
  AC_MSG_RESULT([${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


1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
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
1195
1196
1197
fi


1198
1199
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
1200

Andreas Marek's avatar
Andreas Marek committed
1201
  $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
1202

1203
1204
1205
1206
1207
1208
1209
1210
  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
1211
    AC_MSG_ERROR([Could not compile test program, try with --disable-sse-assembly, or adjust the C compiler or CFLAGS. Possibly (some of) the flags " $SIMD_FLAGS " solve this issue])
1212
  fi
1213

1214
1215
  if test x"${want_single_precision}" = x"yes" ; then
    AC_MSG_CHECKING(whether single-precision SSE assembly kernels can be compiled)
1216

Andreas Marek's avatar
Andreas Marek committed
1217
    $CC $CFLAGS -c $srcdir/src/elpa2/kernels/asm_x86_64_single_precision.s -o conftest.o 2>&5
1218

1219
1220
1221
1222
1223
1224
1225
1226
    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
1227
      AC_MSG_ERROR([Could not compile test program, try with --disable-sse-assembly, or adjust the C compiler or CFLAGS. Possibly (some of) the flags " $SIMD_FLAGS " solve this issue])
1228
1229
    fi
  fi
Andreas Marek's avatar
Andreas Marek committed
1230
1231
1232
1233
fi



1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
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
1250
    AC_MSG_ERROR([Could not compile a test program with AVX, try with --disable-avx, or adjust the C compiler or CFLAGS. Possibly (some of) the flags " $SIMD_FLAGS " solve this issue])
1251
1252
  fi
  AC_DEFINE([HAVE_AVX],[1],[AVX is supported on this CPU])
1253
1254
1255
fi


1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
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
1272
    AC_MSG_ERROR([Could not compile a test program with AVX2, try with --disable-avx2, or adjust the C compiler or CFLAGS.  Possibly (some of) the flags " $SIMD_FLAGS " solve this issue])
1273
1274
  fi
  AC_DEFINE([HAVE_AVX2],[1],[AVX2 is supported on this CPU])
1275
1276
fi

1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
if test x"${need_sve128}" = x"yes"; then
  AC_MSG_CHECKING([whether we can compile SVE128 gcc intrinsics in C])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <arm_sve.h>
   int main(int argc, char **argv){
   double *q;
   svfloat64_t q1 = svld1_f64(svptrue_b64(), q);
   svfloat64_t y1 = svmad_f64_z(svptrue_b64(), q1, q1, q1);
   return 0;
   }
   ])],
   [can_compile_sve128=yes],
   [can_compile_sve128=no]
  )
  AC_MSG_RESULT([${can_compile_sve128}])
  if test x"$can_compile_sve128" != x"yes"; then
    AC_MSG_ERROR([Could not compile a test program with SVE128, adjust the C compiler or CFLAGS. Possibly (some of) the flags " $SIMD_FLAGS " solve this issue])
  fi
  AC_DEFINE([HAVE_SVE128],[1],[SVE128 is supported on this CPU])
fi

if test x"${need_sve256}" = x"yes"; then
  AC_MSG_CHECKING([whether we can compile SVE256 gcc intrinsics in C])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <arm_sve.h>
   int main(int argc, char **argv){
   double *q;
   svfloat64_t q1 = svld1_f64(svptrue_b64(), q);
   svfloat64_t y1 = svmad_f64_z(svptrue_b64(), q1, q1, q1);
   return 0;
   }
   ])],
   [can_compile_sve256=yes],
   [can_compile_sve256=no]
  )
  AC_MSG_RESULT([${can_compile_sve256}])
  if test x"$can_compile_sve256" != x"yes"; then
    AC_MSG_ERROR([Could not compile a test program with SVE256, adjust the C compiler or CFLAGS. Possibly (some of) the flags " $SIMD_FLAGS " solve this issue])
  fi
  AC_DEFINE([HAVE_SVE256],[1],[SVE256 is supported on this CPU])
fi

Andreas Marek's avatar
Andreas Marek committed
1319
1320
1321
1322
1323
if test x"${need_sve512}" = x"yes"; then
  AC_MSG_CHECKING([whether we can compile SVE512 gcc intrinsics in C])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <arm_sve.h>
   int main(int argc, char **argv){
Andreas Marek's avatar
Andreas Marek committed
1324
1325
   double *q;
   svfloat64_t q1 = svld1_f64(svptrue_b64(), q);
Andreas Marek's avatar
Andreas Marek committed
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
   svfloat64_t y1 = svmad_f64_z(svptrue_b64(), q1, q1, q1);
   return 0;
   }
   ])],
   [can_compile_sve512=yes],
   [can_compile_sve512=no]
  )
  AC_MSG_RESULT([${can_compile_sve512}])
  if test x"$can_compile_sve512" != x"yes"; then
    AC_MSG_ERROR([Could not compile a test program with SVE512, adjust the C compiler or CFLAGS. Possibly (some of) the flags " $SIMD_FLAGS " solve this issue])
  fi
  AC_DEFINE([HAVE_SVE512],[1],[SVE512 is supported on this CPU])
fi

1340

1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
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
1357
    AC_MSG_ERROR([Could not compile a test program with AVX512, adjust the C compiler or CFLAGS. Possibly (some of) the flags " $SIMD_FLAGS " solve this issue])
1358
1359
  fi
  AC_DEFINE([HAVE_AVX512],[1],[AVX512 is supported on this CPU])
Andreas Marek's avatar
Andreas Marek committed
1360
1361

  if test x"$can_compile_avx512" = x"yes"; then
1362
    AC_MSG_CHECKING([whether we compile for Xeon])
1363
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
1364
1365
1366
     #include <x86intrin.h>
     int main(int argc, char **argv){
     __m512d sign;
1367
     __m512d h1_real;
Andreas Marek's avatar
Andreas Marek committed
1368
1369
1370
1371
1372
1373
1374
1375

     __m512d x1 = _mm512_xor_pd(h1_real, sign);
     return 0;
     }
     ])],
     [can_compile_avx512_xeon=yes],
     [can_compile_avx512_xeon=no]
    )
1376
1377
1378
    AC_MSG_RESULT([${can_compile_avx512_xeon}])

    AC_MSG_CHECKING([whether we compile for Xeon PHI])
1379
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
1380
1381
1382
1383
     #include <x86intrin.h>
     int main(int argc, char **argv){
     __m512d sign;
     __m512d h1;
Andreas Marek's avatar
Andreas Marek committed
1384
     __m512d h2_real;
Andreas Marek's avatar
Andreas Marek committed
1385

1386
     __m512d x1 = (__m512d) _mm512_xor_epi64((__m512i) h2_real, (__m512i) sign);
Andreas Marek's avatar
Andreas Marek committed
1387
1388
1389
1390
1391
1392
     return 0;
     }
     ])],
     [can_compile_avx512_xeon_phi=yes],
     [can_compile_avx512_xeon_phi=no]
    )
1393
1394
    AC_MSG_RESULT([${can_compile_avx512_xeon_phi}])

1395
1396
1397
1398
1399
1400
    # 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
1401
	  echo "Xeon PHI found ... disabling AVX512 Xeon"
1402
1403
1404
1405
1406
          can_compile_avx512_xeon=no
        fi
      fi
    fi

Andreas Marek's avatar
Andreas Marek committed
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
    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