configure.ac 78.3 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
110
111
112
113
dnl CPP
AC_LANG_PUSH([C++])
AC_PROG_CXX
AC_LANG_POP([C++])

114
dnl C
115
AC_LANG_PUSH([C])
116

117

118
119
120
121
122
123
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
124

125
if test x"${enable_openmp}" = x"yes"; then
126
127
128
129
130
  AX_ELPA_OPENMP
  if test "$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"
131
fi
132

Andreas Marek's avatar
Andreas Marek committed
133
134
135
c11_standard=no
AX_CHECK_COMPILE_FLAG([-std=gnu11], [
  c11_standard=yes
136
], [
Andreas Marek's avatar
Andreas Marek committed
137
138
  echo "C compiler cannot compile -std=gnu11 code"
  echo "testing -std=c11.."
139
])
Andreas Marek's avatar
Andreas Marek committed
140
if test x"$c11_standard" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
141
142
143
  CFLAGS+=" -std=gnu11"
fi

Andreas Marek's avatar
Andreas Marek committed
144
if test x"$c11_standard" = x"no"; then
Andreas Marek's avatar
Andreas Marek committed
145
146
147
  AX_CHECK_COMPILE_FLAG([-std=c11], [
    c11_standard=yes
  ], [
148
149
150
151
152
153
154
155
156
157
158
159
160
    echo "C compiler cannot compile -std=c11 code"
    echo "testing -c11.."
  ])
  if test x"$c11_standard" = x"yes"; then
    CFLAGS+=" -std=c11"
  fi
fi

if test x"$c11_standard" = x"no"; then
  AX_CHECK_COMPILE_FLAG([-c11], [
    c11_standard=yes
  ], [
    echo "C compiler cannot compile -c11 code"
Andreas Marek's avatar
Andreas Marek committed
161
162
163
    echo "C compiler cannot compile C11 code"
    exit -1
  ])
Andreas Marek's avatar
Andreas Marek committed
164
  if test x"$c11_standard" = x"yes"; then
165
    CFLAGS+=" -c11"
Andreas Marek's avatar
Andreas Marek committed
166
167
  fi
fi
168

169
170
AX_EXT

Andreas Marek's avatar
Andreas Marek committed
171
172
173
174
175
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!]),
176
177
              [
	       if test x"$enableval" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
178
	         enable_heterogenous_cluster_support=yes
179
	       else
Andreas Marek's avatar
Andreas Marek committed
180
	         enable_heterogenous_cluster_support=no
181
182
	       fi
	       ],
Andreas Marek's avatar
Andreas Marek committed
183
184
185
186
              [enable_heterogenous_cluster_support="no"])
AC_MSG_RESULT([$enable_heterogenous_cluster_support])
if test x"${enable_heterogenous_cluster_support}" = x"yes"; then
  AC_DEFINE([HAVE_HETEROGENOUS_CLUSTER_SUPPORT], [1], [automatically support clusters with different Intel CPUs])
187
fi
Andreas Marek's avatar
Andreas Marek committed
188
AM_CONDITIONAL([HAVE_HETEROGENOUS_CLUSTER_SUPPORT],[test x"$enable_heterogenous_cluster_support" = x"yes"])
189

Andreas Marek's avatar
Andreas Marek committed
190
191
192
193
194
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}"

195
dnl then 64bit blas
Andreas Marek's avatar
WIP    
Andreas Marek committed
196
AC_MSG_CHECKING(whether 64bit integers should be used for math libraries (BLAS/LAPACK/SCALAPACK))
197
198
AC_ARG_ENABLE([64bit-integer-math-support],
              AS_HELP_STRING([--64bit-integer-math-support],
Andreas Marek's avatar
WIP    
Andreas Marek committed
199
200
201
                             [allows to link against the 64bit integer versions of the math libraries BLAS, LAPACK, and SCALAPACK]),
              [
	       if test x"$enableval" = x"yes"; then
202
	         enable_64bit_integer_math_support=yes
Andreas Marek's avatar
WIP    
Andreas Marek committed
203
	       else
204
	         enable_64bit_integer_math_support=no
Andreas Marek's avatar
WIP    
Andreas Marek committed
205
206
	       fi
	       ],
207
208
209
              [enable_64bit_integer_math_support="no"])
AC_MSG_RESULT([$enable_64bit_integer_math_support])
if test x"${enable_64bit_integer_math_support}" = x"yes"; then
210
211
212
213
  dnl at least INTEL MPI does _NOT_ support 64BIT integer mode  for C thus disable C tests in this Case
  if test x"${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
214
215
216
217
218
219
220
  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

221
  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
222
fi
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
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"])

244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
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])

267
AC_PROG_INSTALL
268
269
AM_PROG_AR
AM_PROG_AS
270
AC_PROG_CC_C99
271
AM_PROG_CC_C_O
272
AC_LANG_POP([C])
Andreas Marek's avatar
Andreas Marek committed
273
274

# Fortran
275
AC_LANG_PUSH([Fortran])
276

Andreas Marek's avatar
Andreas Marek committed
277
m4_include([m4/ax_prog_fc_mpi.m4])
278
279
280
281
282
283
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
284

285
AC_FC_SRCEXT([F90])
286
287
288
289
290
AC_FC_FREEFORM
AC_FC_MODULE_FLAG
AC_FC_MODULE_OUTPUT_FLAG
AC_FC_LIBRARY_LDFLAGS

291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
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"${FC_does_infer_interfaces}" = x"yes"; then
309
310
311
312
313
314
315
  # two problems
  # 1. some MPIs (at least Intel) do not provide all
  #    interfaces, which will fail
  #    in case of MPI and not all interfaces -> switch of
  #    argument checking or fail
  # 2. no MPI case: switch of PACK_REAL_TO_COMPLEX
  if test x"${with_mpi}" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
316
    AC_MSG_CHECKING(whether MPI module defines all interfaces )
317
318
319
320
321
322
323
324
    # 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
325
326
      real*8  :: a(2)
      complex*16 :: b(2)
327
328
329
330
331
332
333
334

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

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

335
336
337
      
      !! this is OK
      !call MPI_Bcast(buf, 10, MPI_INT, 0, MPI_COMM_WORLD, ierr)
338

339
340
      !! Oops, wrong order here:
      !call MPI_Bcast(10, buf, MPI_INT, 0, MPI_COMM_WORLD, ierr)
341

342
343
344
345
346
      ! 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)
347
348
349
      call MPI_Finalize(ierr)
    end program
      ])],
350
351
      [MPImodule_has_interfaces=yes],
      [MPImodule_has_interfaces=no]
352
    )
353
    AC_MSG_RESULT([${MPImodule_has_interfaces}])
354
    if test x"${MPImodule_has_interfaces}" = x"no"; then
355
      AC_MSG_CHECKING(whether we can cure missing interfaces by allowing argument mismatch )
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
      #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

371
      ])],
372
373
      [FC_infer_interfaces_cured=yes],
      [FC_infer_interfaces_cured=no]
374
      )
375
376
      AC_MSG_RESULT([${FC_infer_interfaces_cured}])
      if test x"${FC_infer_interfaces_cured}" = x"no"; then
377
378
379
380
381
382
383
        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
384
385
386
387
388
  fi
else
  AC_DEFINE([PACK_REAL_TO_COMPLEX], [1], [In some kernels pack real to complex])
fi

389
if test x"${enable_openmp}" = x"yes"; then
390
391
392
393
394
  AX_ELPA_OPENMP
  if test "$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"
395
396
fi

397
398
dnl check which MPI binray invokes a MPI job
if test x"$with_mpi" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
399
  AC_CHECK_PROGS([MPI_BINARY], [mpiexec.hydra mpiexec mpirun poe runjob srun aprun], [no])
400
  if test x"$MPI_BINARY" = x"no"; then
Andreas Marek's avatar
Andreas Marek committed
401
    AC_MSG_ERROR([Could not find either of the MPI binaries: mpiexec.hydra, mpiexec, mpirun, poe, runjob, srun, aprun])
402
403
404
  fi
fi

405
406
407
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)
408
AC_ARG_ENABLE([redirect],
409
410
              [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
411
412
413
414
415
416
417
              [
               if test x"$enableval" = x"yes"; then
                 enable_redirect=yes
               else
                 enable_redirect=no
               fi
	       ],
418
              [enable_redirect=no])
419
420
AC_MSG_RESULT([${enable_redirect}])

421
422
423
424
425
426
427
428
429
430
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}])


431
432
dnl redirect
if test x"${enable_redirect}" = x"yes"; then
433
434
  AC_DEFINE([HAVE_REDIRECT], [1], [Redirect stdout and stderr of test programs per MPI tasks to a file])
fi
435
436
437
AM_CONDITIONAL([HAVE_REDIRECT],[test x"$enable_redirect" = x"yes"])

dnl build with ftimings support
438

Andreas Marek's avatar
Andreas Marek committed
439
AC_ARG_ENABLE([timings],
440
441
              [AS_HELP_STRING([--disable-timings],
                              [more detailed timing, default yes])],
Andreas Marek's avatar
Andreas Marek committed
442
443
444
445
446
447
448
              [
               if test x"$enableval" = x"yes"; then
                 enable_timings=yes
               else
                 enable_timings=no
               fi
	       ],
449
              [enable_timings=yes])
450
451
452
453

if test x"${enable_timings}" = x"yes"; then
  AC_DEFINE([HAVE_DETAILED_TIMINGS], [1], [Enable more timing])
fi
454
AM_CONDITIONAL([HAVE_DETAILED_TIMINGS], [test x"$enable_timings" = x"yes"])
455
456

dnl PAPI for ftimings
457
AC_LANG_PUSH([C])
458
AC_ARG_WITH([papi],
459
460
            [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
461
            [
Wenzhe Yu's avatar
Wenzhe Yu committed
462
               if test x"$withval" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
463
464
465
466
467
                 with_papi=yes
               else
                 with_papi=no
               fi
	     ],
468
            [with_papi="no"])
469
470
if test x"${enable_timings}" = x"yes"; then
  if test x"$with_papi" = x"yes" ; then
471
    AC_SEARCH_LIBS([PAPI_library_init], [papi], [papi_found=yes], [papi_found=no])
472
473
    if test x"$papi_found" = x"no" ; then
      AC_MSG_ERROR(["Could not find usable PAPI installation, please install or adjust CFLAGS, LDFLAGS"])
474
475
476
477
    fi
    AC_DEFINE([HAVE_LIBPAPI], [1], [Use the PAPI library])
  fi
fi
478
AC_LANG_POP([C])
479

Pavel Kus's avatar
Pavel Kus committed
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
520
521
522
523
524
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])

525
526
527
save_FCFLAGS=$FCFLAGS
save_LDFLAGS=$LDFLAGS

528
529
530
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])

531
532
FCFLAGS="$FCFLAGS $SCALAPACK_FCFLAGS"
LDFLAGS="$LDFLAGS $SCALAPACK_LDFLAGS"
533

534
dnl check whether fortran error_unit is defined
535
AC_MSG_CHECKING([whether Fortran module iso_fortran_env is available])
536
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
537
  program test_error_unit
538
    use iso_fortran_env, only : error_unit
539
540
541
542
543
544
545
546
    implicit none

    write(error_unit,*) "error_unit is defined"
  end program
])],
  [can_use_iso_fortran_env=yes],
  [can_use_iso_fortran_env=no]
)
547
AC_MSG_RESULT([${can_use_iso_fortran_env}])
548
549
550
551
if test x"${can_use_iso_fortran_env}" = x"yes" ; then
 AC_DEFINE([HAVE_ISO_FORTRAN_ENV],[1],[can use module iso_fortran_env])
fi

552

553
554
dnl check whether one can link with specified MKL (desired method)
AC_MSG_CHECKING([whether we can compile a Fortran program using MKL])
555
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
556
  program test_mkl
557
    use mkl_service
558
559
560
561
562
563
564
565
566
567
568
569
    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"$can_compile_with_mkl" = x"yes" ; then
  AC_MSG_CHECKING([whether we can link a Fortran program with MKL])
570
  AC_LINK_IFELSE([AC_LANG_SOURCE([
571
    program test_mkl
572
      use mkl_service
573
574
575
576
577
      character*198 :: string
      call mkl_get_version_string(string)
      write(*,'(a)') string
    end program
  ])],
Andreas Marek's avatar
Andreas Marek committed
578
579
    [have_mkl=yes],
    [have_mkl=no]
580
  )
Andreas Marek's avatar
Andreas Marek committed
581
  AC_MSG_RESULT([${have_mkl}])
582
583
584
fi

dnl if not mkl, check all the necessary individually
585
if test x"${have_mkl}" = x"yes" ; then
586
587
588
589
  WITH_MKL=1
else

  dnl first check blas
590
  AC_SEARCH_LIBS([dgemm],[flexiblas openblas satlas blas],[have_blas=yes],[have_blas=no])
Andreas Marek's avatar
Andreas Marek committed
591
592
  AC_MSG_CHECKING([whether we can link a program with a blas lib])
  AC_MSG_RESULT([${have_blas}])
593

594
  if test x"${have_blas}" = x"no" ; then
595
    AC_MSG_ERROR([could not link with blas: specify path])
596
  fi
597
  dnl now lapack
Andreas Marek's avatar
Andreas Marek committed
598
599
600
  AC_SEARCH_LIBS([dlarrv],[lapack],[have_lapack=yes],[have_lapack=no])
  AC_MSG_CHECKING([whether we can link a program with a lapack lib])
  AC_MSG_RESULT([${have_lapack}])
601

602
  if test x"${have_lapack}" = x"no" ; then
603
    AC_MSG_ERROR([could not link with lapack: specify path])
604
  fi
Andreas Marek's avatar
Andreas Marek committed
605

606
  if test x"${with_mpi}" = x"yes"; then
607
    dnl test whether scalapack already contains blacs
608
    scalapack_libs="mpiscalapack scalapack scalapack-openmpi"
609
610
611
612
613
    old_LIBS="$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
614
	       
615
616
617
618
619
      AC_MSG_RESULT([${blacs_in_scalapack}])
      if test x"${blacs_in_scalapack}" = x"yes"; then
        break
      fi
    done
620

621
622
    if test x"${blacs_in_scalapack}" = x"no"; then
      LIBS="${old_LIBS}"
623

624
625
      dnl Test for stand-alone blacs
      AC_SEARCH_LIBS([bi_f77_init],[mpiblacsF77init],[],[],[-lmpiblacs])
626
627
628
629
630
631

      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
632
      AC_SEARCH_LIBS([blacs_gridinit],[mpiblacs blacs],[have_blacs=yes],[have_blacs=no])
Andreas Marek's avatar
Andreas Marek committed
633

634
635
636
      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
637
    fi
638

639
    AC_SEARCH_LIBS([pdtran],[$scalapack_libs],[have_scalapack=yes],[have_scalapack=no])
Andreas Marek's avatar
Andreas Marek committed
640

641
642
643
    if test x"${have_scalapack}" = x"no" ; then
      AC_MSG_ERROR([could not link with scalapack: specify path])
    fi
644
645
646
647
  fi

  dnl check whether we can link alltogehter
  AC_MSG_CHECKING([whether we can link a Fortran program with all blacs/scalapack])
648
  AC_LINK_IFELSE([AC_LANG_SOURCE([
649
650
    program dgemm_test

651
652
      integer , parameter:: M = 4, N = 3, K = 2
      real :: A(M,K), B(K,N), C(M,N)
653

654
      call dgemm('N','N',M,N,K,1.0,A,M,B,K,0.0,C,M)
655
656
657
658
659
660
661
662

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

663
   if test x"${can_link_with_blacs_scalapack}" = x"yes" ; then
664
665
     WITH_BLACS=1
   else
666
   AC_MSG_ERROR([We can neither link with MKL or another Scalpack. Please specify SCALAPACK_LDFLAGS and SCALAPACK_FCFLAGS!])
667
668
669
   fi
fi

670
671
672
673
dnl important: reset them again!
FCFLAGS=$save_FCFLAGS
LDFLAGS=$save_LDFLAGS

674

Andreas Marek's avatar
Andreas Marek committed
675
676
677
678
dnl check for intrinsic fortran function of 2003 standard
AC_MSG_CHECKING([whether we can use the intrinsic Fortran function "get_environment_variable"])


679
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
680
681
682
683
684
685
686
687
688
  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([${fortran_can_check_environment}])
689
690
691
if test x"${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
692

693

694
695
dnl check whether BAND_TO_FULL_BLOCKING is set
AC_MSG_CHECKING(whether BAND_TO_FLULL_BLOCKING is requested)
696
AC_ARG_ENABLE(band-to-full-blocking,[AS_HELP_STRING([--disable-band-to-full-blocking],
697
                                                    [build ELPA2 with blocking in band_to_full (default: enabled)])],
Andreas Marek's avatar
Andreas Marek committed
698
699
700
701
702
703
704
              [
               if test x"$enableval" = x"yes"; then
                 enable_band_to_full_blocking=yes
               else
                 enable_band_to_full_blocking=no
               fi
	       ],
705
706
	      [enable_band_to_full_blocking="yes"])
AC_MSG_RESULT([${enable_band_to_full_blocking}])
707

708
709
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
710
       AC_DEFINE([BAND_TO_FULL_BLOCKING], [1], [use blocking in trans_ev_band_to_full])
711
712
fi

713
714
715
716
717
718
719

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])

720

721
dnl setup nvcc flags and use them in later tests
Andreas Marek's avatar
Andreas Marek committed
722
723
724
725
726
727
728
user_sets_nvidia_gpu_compute_capability="no"
AC_MSG_CHECKING(whether a Nvidia GPU compute capability is specified)
AC_ARG_WITH([NVIDIA-GPU-compute-capability],
            [AS_HELP_STRING([--with-NVIDIA-GPU-compute-capability=VALUE],
                            [use compute capability VALUE for NVIDIA GPU version, default: "sm_35"])],
            [user_sets_nvidia_gpu_compute_capability="yes"],[cuda_compute_capability="sm_35"])
AC_MSG_RESULT([${user_sets_nvidia_gpu_compute_capability}])
729

730

731
dnl sanity check whether compute capability setting by user is reasonable
Andreas Marek's avatar
Andreas Marek committed
732
if test x"${user_sets_nvidia_gpu_compute_capability}" = x"yes" ; then
733
734
735
736
737
  dnl the user must set a value which starts with "sm_"
  value=$(echo $withval | cut -c1-3)
  if test x"${value}" = x"sm_" ; then
    cuda_compute_capability=$withval
  else
Andreas Marek's avatar
Andreas Marek committed
738
    AC_MSG_ERROR([Unknown Nvidia GPU compute capability set: ${withval}])
739
740
741
  fi
fi

Andreas Marek's avatar
Andreas Marek committed
742

743
744
dnl Test possibility of 'use mpi', if requested
if test x"${with_mpi}" = x"yes" ; then
745
746
  AC_ARG_ENABLE([mpi-module],
                AS_HELP_STRING([--disable-mpi-module],
Pavel Kus's avatar
Pavel Kus committed
747
                               [do not use the Fortran MPI module, get interfaces by 'include "mpif.h']),
Andreas Marek's avatar
Andreas Marek committed
748
749
750
751
752
753
754
                [
                 if test x"$enableval" = x"yes"; then
                   enable_mpi_module=yes
                 else
                   enable_mpi_module=no
                 fi
		 ],
755
756
                [enable_mpi_module=yes])
  if test x"${enable_mpi_module}" = x"yes" ; then
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
    AC_MSG_CHECKING(whether Fortran mpi module can be used)
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
      program test_mpi_module
        use mpi
        real :: time
        time = MPI_WTime()
      end program
    ])],
      [can_use_fortran_mpi_module=yes],
      [can_use_fortran_mpi_module=no]
    )
    AC_MSG_RESULT([${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
772
      AC_MSG_ERROR([Could not compile a Fortran program with an 'use mpi' statement. You can try again with --disable-mpi-module])
773
    fi
774
  fi
Andreas Marek's avatar
Andreas Marek committed
775
fi
776
AC_LANG_POP([Fortran])
Andreas Marek's avatar
Andreas Marek committed
777

778
779
dnl Assemble the list of kernels to build
m4_pattern_forbid([elpa_m4])
Andreas Marek's avatar
Andreas Marek committed
780

781
782
783
m4_define(elpa_m4_generic_kernels, [
        real_generic
        real_generic_simple
Andreas Marek's avatar
Andreas Marek committed
784
        real_generic_simple_block4
785
        real_generic_simple_block6
786
787
788
        complex_generic
        complex_generic_simple
])
789

790
791
792
793
m4_define(elpa_m4_sse_assembly_kernels, [
        real_sse_assembly
        complex_sse_assembly
])
794

795
796
797
798
799
800
801
m4_define(elpa_m4_sse_kernels, [
        real_sse_block2
        real_sse_block4
        real_sse_block6
        complex_sse_block1
        complex_sse_block2
])
802

803
804
805
806
807
808
m4_define(elpa_m4_sparc64_kernels, [
        real_sparc64_block2
        real_sparc64_block4
        real_sparc64_block6
])

809
810
811
812
m4_define(elpa_m4_neon_arch64_kernels, [
        real_neon_arch64_block2
        real_neon_arch64_block4
        real_neon_arch64_block6
813
814
        complex_neon_arch64_block1
        complex_neon_arch64_block2
815
816
])

817
818
819
820
821
822
m4_define(elpa_m4_vsx_kernels, [
        real_vsx_block2
        real_vsx_block4
        real_vsx_block6
])

823
824
825
826
827
828
829
830
m4_define(elpa_m4_sve128_kernels, [
        real_sve128_block2
        real_sve128_block4
        real_sve128_block6
        complex_sve128_block1
        complex_sve128_block2
])

831
832
833
834
835
836
837
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
838

839
840
841
842
843
844
845
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
846

847
848
849
850
851
852
853
854
m4_define(elpa_m4_sve256_kernels, [
        real_sve256_block2
        real_sve256_block4
        real_sve256_block6
        complex_sve256_block1
        complex_sve256_block2
])

855
856
857
858
859
860
861
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
862

863
864
865
866
867
868
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
869
870
])

871
872
873
874
m4_define(elpa_m4_bgp_kernels, [
        real_bgp
        complex_bgp
])
Andreas Marek's avatar
Andreas Marek committed
875

876
877
878
879
m4_define(elpa_m4_bgq_kernels, [
        real_bgq
        complex_bgq
])
Andreas Marek's avatar
Andreas Marek committed
880

Andreas Marek's avatar
Andreas Marek committed
881
882
883
884
m4_define(elpa_m4_gpu_kernels, [
        real_gpu
        complex_gpu
])
885

886
m4_define(elpa_m4_kernel_types, [generic sparc64 neon_arch64 vsx sse sse_assembly sve128 avx avx2 sve256 avx512 sve512 bgp bgq gpu])
887

888
889
890
891
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
892

893
894
895
896
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
897

898
899
900
901
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) ]))
902

903
904
905
906
907
908
909
910
911
912
913
914
915
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]
        ])
])
916

917
918
dnl Modify list of kernels with configure arguments
ELPA_SELECT_KERNELS([generic],[enable])
919
ELPA_SELECT_KERNELS([sparc64],[disable])
920
ELPA_SELECT_KERNELS([neon_arch64],[disable])
921
ELPA_SELECT_KERNELS([vsx],[disable])
922
923
924
925
ELPA_SELECT_KERNELS([sse],[enable])
ELPA_SELECT_KERNELS([sse_assembly],[enable])
ELPA_SELECT_KERNELS([avx],[enable])
ELPA_SELECT_KERNELS([avx2],[enable])
926
ELPA_SELECT_KERNELS([avx512],[enable])
927
928
ELPA_SELECT_KERNELS([sve128],[disable])
ELPA_SELECT_KERNELS([sve256],[disable])
929
ELPA_SELECT_KERNELS([sve512],[disable])
Andreas Marek's avatar
Andreas Marek committed
930
ELPA_SELECT_KERNELS([gpu],[disable])
931
932
933
934
935
936
ELPA_SELECT_KERNELS([bgp],[disable])
ELPA_SELECT_KERNELS([bgq],[disable])

m4_foreach_w([elpa_m4_kind],[real complex],[
        default_[]elpa_m4_kind[]_kernel=""
])
937

938
if test x"${enable_bgp}" = x"yes" -o x"$enable_bgq" = x"yes"; then
939
        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, [
940
941
942
943
944
945
946
947
948
949
950
951
                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:]
952
953
                 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[_$withval"],[fixed_]elpa_m4_kind[_kernel=""])
954
        if test -n "$fixed_[]elpa_m4_kind[]_kernel" ; then
955
                found="no"
956
957
958
                m4_foreach_w([elpa_m4_otherkernel],m4_expand(elpa_m4_[]elpa_m4_kind[]_kernels),[
                        if test "$fixed_]elpa_m4_kind[_kernel" = "]elpa_m4_otherkernel[" ; then
                                use_[]elpa_m4_otherkernel[]=yes
959
                                found="yes"
960
961
962
963
                        else
                                use_[]elpa_m4_otherkernel[]=no
                        fi
                ])
964
965
966
                if test x"$found" = x"no" ; then
                        AC_MSG_ERROR([Invalid kernel "$fixed_]elpa_m4_kind[_kernel" specified for --with-fixed-]elpa_m4_kind[-kernel])
                fi
967
968
969
970
                default_[]elpa_m4_kind[]_kernel="$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
])
971

Andreas Marek's avatar
Andreas Marek committed
972
973
974
975
976
977
978
979
980
981
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
982

983

984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
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
])
999
m4_foreach_w([elpa_m4_arch],[sparc64 neon_arch64 vsx sse avx avx2 avx512 sve128 sve256 sve512],[
1000
1001
1002
1003
        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])
])
1004

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

1026

1027
1028
1029
dnl the list of kernels is now assembled
dnl choosing a default kernel

1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
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
])



1054
1055
1056
m4_foreach_w([elpa_m4_kind],[real complex],[
        m4_foreach_w([elpa_m4_kernel],
                     m4_foreach_w([elpa_m4_cand_kernel],
1057
                                  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,
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
                                  [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 | \
1071
                                                                perl -pe 's/^[[^,]]*, *//; s/,.*//;'`
1072
1073
        AC_SUBST([ELPA_2STAGE_]m4_toupper(elpa_m4_kind)[_DEFAULT])
])
Andreas Marek's avatar
Andreas Marek committed
1074

1075
1076
1077
1078
1079
1080
1081
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 }
1082
AC_LANG_PUSH([C])
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
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
  
1102
AC_MSG_CHECKING(whether we can link C programs against Fortran without _ )
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
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([${can_link_without_}])
1116
if test x"$can_link_without_" = x"yes"; then
1117
1118
1119
  AC_DEFINE([NEED_NO_UNDERSCORE_TO_LINK_AGAINST_FORTRAN],[1],[need not to append an underscore])
fi

1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
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

1140
1141
1142
if test x"${need_sparc64}" = x"yes"; then
  AC_MSG_CHECKING(whether we can compile SPARC64 with intrinsics in C)
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
1143
1144
#include <fjmfunc.h>
#include <emmintrin.h>
1145
int main(int argc, char **argv) {
Andreas Marek's avatar
Andreas Marek committed
1146
1147
  __m128d  tau1;
  __m128d h1 = _fjsp_neg_v2r8(tau1);
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
  return 0;
}
    ])],
    [can_compile_sparc64=yes],
    [can_compile_sparc64=no]
  )
  AC_MSG_RESULT([${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
1160

1161
1162
1163
1164
1165
1166
if test x"${need_neon_arch64}" = x"yes"; then
  AC_MSG_CHECKING(whether we can compile NEON ARCH64 with intrinsics in C)
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
#include <arm_neon.h>
int main(int argc, char **argv) {
  __Float64x2_t  x1, x2, x3, x4;
Andreas Marek's avatar
Andreas Marek committed
1167
  x4 = vfmaq_f64(x1, x2, x3);
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
  return 0;
}
    ])],
    [can_compile_neon_arch64=yes],
    [can_compile_neon_arch64=no]
  )
  AC_MSG_RESULT([${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


1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
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
1200
1201
1202
fi


1203
1204
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
1205

Andreas Marek's avatar
Andreas Marek committed
1206
  $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
1207

1208
1209
1210
1211
1212
1213
1214
1215
  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
1216
    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])
1217
  fi
1218

1219
1220
  if test x"${want_single_precision}" = x"yes" ; then
    AC_MSG_CHECKING(whether single-precision SSE assembly kernels can be compiled)
1221

Andreas Marek's avatar
Andreas Marek committed
1222
    $CC $CFLAGS -c $srcdir/src/elpa2/kernels/asm_x86_64_single_precision.s -o conftest.o 2>&5
1223

1224
1225
1226
1227
1228
1229
1230
1231
    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
1232
      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])
1233
1234
    fi
  fi
Andreas Marek's avatar
Andreas Marek committed
1235
1236
1237
1238
fi



1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
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
1255
    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])
1256
1257
  fi
  AC_DEFINE([HAVE_AVX],[1],[AVX is supported on this CPU])
1258
1259
1260
fi


1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
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
1277
    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])
1278
1279
  fi
  AC_DEFINE([HAVE_AVX2],[1],[AVX2 is supported on this CPU])
1280
1281
fi

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
1319
1320
1321
1322
1323
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
1324
1325
1326
1327
1328
if test x"${need_sve512}" = x"yes"; then
  AC_MSG_CHECKING([whether we can compile SVE512 gcc intrinsics in C])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <arm_sve.h>
   int main(int argc, char **argv){
Andreas Marek's avatar
Andreas Marek committed
1329
1330
   double *q;
   svfloat64_t q1 = svld1_f64(svptrue_b64(), q);
Andreas Marek's avatar
Andreas Marek committed
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
   svfloat64_t y1 = svmad_f64_z(svptrue_b64(), q1, q1, q1);
   return 0;
   }
   ])],
   [can_compile_sve512=yes],
   [can_compile_sve512=no]
  )
  AC_MSG_RESULT([${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

1345

1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
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
1362
    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])
1363
1364
  fi
  AC_DEFINE([HAVE_AVX512],[1],[AVX512 is supported on this CPU])
Andreas Marek's avatar
Andreas Marek committed
1365
1366

  if test x"$can_compile_avx512" = x"yes"; then
1367
    AC_MSG_CHECKING([whether we compile for Xeon])
1368
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
1369
1370
1371
     #include <x86intrin.h>
     int main(int argc, char **argv){
     __m512d sign;
1372
     __m512d h1_real;
Andreas Marek's avatar
Andreas Marek committed
1373
1374
1375
1376
1377
1378
1379
1380

     __m512d x1 = _mm512_xor_pd(h1_real, sign);
     return 0;
     }
     ])],
     [can_compile_avx512_xeon=yes],
     [can_compile_avx512_xeon=no]
    )
1381
1382
1383
    AC_MSG_RESULT([${can_compile_avx512_xeon}])

    AC_MSG_CHECKING([whether we compile for Xeon PHI])
1384
    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
1385
1386
1387
1388
     #include <x86intrin.h>
     int main(int argc, char **argv){
     __m512d sign;
     __m512d h1;
Andreas Marek's avatar
Andreas Marek committed
1389
     __m512d h2_real;
Andreas Marek's avatar
Andreas Marek committed
1390

1391
     __m512d x1 = (__m512d) _mm512_xor_epi64((__m512i) h2_real, (__m512i) sign);
Andreas Marek's avatar
Andreas Marek committed
1392
1393
1394
1395
1396
1397
     return 0;
     }
     ])],
     [can_compile_avx512_xeon_phi=yes],
     [can_compile_avx512_xeon_phi=no]
    )
1398
1399
    AC_MSG_RESULT([${can_compile_avx512_xeon_phi}])

1400
1401
1402
1403
1404
1405
    # 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
1406
	  echo "Xeon PHI found ... disabling AVX512 Xeon"
Andreas Marek's avatar