configure.ac 44.4 KB
Newer Older
1
AC_PREREQ([2.69])
2

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

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

AM_INIT_AUTOMAKE([foreign -Wall subdir-objects])
10
11
12
13
14
15
16
17
18

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# This only makes trouble for end users with out-of-date autoconf versions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

19
20
21
22
AC_CONFIG_MACRO_DIR([m4])
AC_CONFIG_HEADERS([config.h])
AM_SILENT_RULES([yes])

23
24
25
26
27
28
29
30
31
32
33
34
35
# 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)
#
36
AC_SUBST([ELPA_SO_VERSION], [8:0:0])
37
38
#

39

40
41
42
43
44
AX_CHECK_GNU_MAKE()
if test x$_cv_gnu_make_command = x ; then
        AC_MSG_ERROR([Need GNU Make])
fi

Andreas Marek's avatar
Andreas Marek committed
45
46
47
48
#AC_CHECK_PROG(CPP_FOUND,cpp,yes,no)
#if test x"${CPP_FOUND}" = xno; then
#  AC_MSG_ERROR([no cpp found])
#fi
49
50
51
52
53

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

54
55
###

56
57
m4_include([m4/ax_elpa_openmp.m4])

58
59
AC_MSG_CHECKING(whether --enable-openmp is specified)
AC_ARG_ENABLE([openmp],
60
		AS_HELP_STRING([--enable-openmp],
61
62
63
64
65
66
67
68
69
			       [use OpenMP threading, default no.]),
	      [],
	      [enable_openmp=no])
AC_MSG_RESULT([${enable_openmp}])
AM_CONDITIONAL([WITH_OPENMP],[test x"$enable_openmp" = x"yes"])
if test x"${enable_openmp}" = x"yes"; then
	AC_DEFINE([WITH_OPENMP], [1], [use OpenMP threading])
fi

70
71
dnl mpi
AC_ARG_WITH(mpi, [AS_HELP_STRING([--with-mpi=[[yes|no]]], [compile with MPI. Default: yes])],,[with_mpi=yes])
72
AM_CONDITIONAL([WITH_MPI],[test x"$with_mpi" = x"yes"])
73
74
if test x"${with_mpi}" = x"yes"; then
       AC_DEFINE([WITH_MPI], [1], [use MPI])
75
76
fi

Andreas Marek's avatar
Andreas Marek committed
77
78
# C
AC_LANG([C])
79
80
81
82
83
84
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
85

86
if test x"${enable_openmp}" = x"yes"; then
87
88
89
90
91
  AX_ELPA_OPENMP
  if test "$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"
92
fi
93

94
AC_PROG_INSTALL
95
96
AM_PROG_AR
AM_PROG_AS
97
AC_PROG_CC_C99
98
AM_PROG_CC_C_O
Andreas Marek's avatar
Andreas Marek committed
99
100

# Fortran
101
102
103
104
105
106
107
108

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

Andreas Marek's avatar
Andreas Marek committed
109
110
AC_LANG([Fortran])
m4_include([m4/ax_prog_fc_mpi.m4])
111
112
113
114
115
116
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
117

118
119
AC_FC_SRCEXT([F90])

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

128
129
dnl check which MPI binray invokes a MPI job
if test x"$with_mpi" = x"yes"; then
130
  AC_CHECK_PROGS([MPI_BINARY], [srun mpiexec.hydra mpiexec mpirun poe runjob], [no])
131
  if test x"$MPI_BINARY" = x"no"; then
132
    AC_MSG_ERROR([Could not find either of the MPI binaries: mpiexec.hydra, mpiexec, mpirun, poe, runjob, srun])
133
134
135
  fi
fi

Andreas Marek's avatar
Andreas Marek committed
136
137
138
139
140
141
142
143
144
145
146
## C++
#AC_LANG([C++])
#AC_PROG_CXX
#
#if test x"${enable_openmp}" = x"yes"; then
#  AX_ELPA_OPENMP
#  if test "$ac_cv_prog_cxx_openmp" = unsupported; then
#    AC_MSG_ERROR([Could not compile a C++ program with OpenMP, adjust CXXFLAGS])
#  fi
#  CXXFLAGS="$OPENMP_CXXFLAGS $CXXFLAGS"
#fi
Andreas Marek's avatar
Andreas Marek committed
147
148
149



Andreas Marek's avatar
Andreas Marek committed
150
151
dnl variables needed for the tests

Andreas Marek's avatar
Andreas Marek committed
152
153
154
155
156
157
158
159
dnl do NOT remove any variables here, until
dnl 1. you know 100% what you are doing
dnl 2. you tested ALL configure functionality afterwards
dnl Otherwise, you most likely break some functionality

dnl as default always define the generic kernels to be build
dnl this is only unset if gpu_support_only is defined, or
dnl other specific real/complex kernels are wanted
Andreas Marek's avatar
Andreas Marek committed
160
161
162
163
164
165

install_real_generic=yes
install_real_generic_simple=yes

install_complex_generic=yes
install_complex_generic_simple=yes
166

167
168
169
#want_avx=yes
#want_avx2=yes
#want_sse=yes
170
#want_avx512=yes
171

Andreas Marek's avatar
Andreas Marek committed
172
173
AC_LANG([C])

174
175
176
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)
177
178
179
180
181
182
183
184
185
AC_ARG_ENABLE([redirect],
		AS_HELP_STRING([--enable-redirect],
			       [for test programs, allow redirection of stdout/stderr per MPI taks in a file (useful for timing), default no.]),
	      [],
	      [enable_redirect=no])
AC_MSG_RESULT([${enable_redirect}])

dnl redirect
if test x"${enable_redirect}" = x"yes"; then
186
187
  AC_DEFINE([HAVE_REDIRECT], [1], [Redirect stdout and stderr of test programs per MPI tasks to a file])
fi
188
189
190
191
AM_CONDITIONAL([HAVE_REDIRECT],[test x"$enable_redirect" = x"yes"])

dnl build with ftimings support
AC_MSG_CHECKING(whether ELPA should be build with more detailed timing support)
Andreas Marek's avatar
Andreas Marek committed
192
AC_ARG_ENABLE([timings],
Andreas Marek's avatar
Andreas Marek committed
193
		AS_HELP_STRING([--enable-timings],
194
			       [more detailed timing, default no.]),
195
	      [enable_timings=yes],
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
	      [enable_timings=no])
AC_MSG_RESULT([${enable_timings}])

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

dnl PAPI for ftimings
AC_ARG_WITH([papi],
		[AS_HELP_STRING([--with-papi],[Use PAPI to also measure flop count in the detailed timing (--enable-timing), disabled by default])],
		[],
		[with_papi="no"])
if test x"${enable_timings}" = x"yes"; then
  if test x"$with_papi" = x"yes" ; then
    AC_SEARCH_LIBS([PAPI_library_init],[papi],
                  [papi_found=yes],[papi_found=no])
    if test x"$papi_found" = x"no" ; then
      AC_MSG_ERROR(["Could not find usable PAPI installation, please install or adjust CFLAGS, LDFLAGS"])
215
216
217
218
219
    fi
    AC_DEFINE([HAVE_LIBPAPI], [1], [Use the PAPI library])
  fi
fi

220
AC_MSG_CHECKING(whether double-precision SSE assembly kernel can be compiled)
Andreas Marek's avatar
Andreas Marek committed
221

222
$CC -c $srcdir/src/elpa2_kernels/elpa2_kernels_asm_x86_64_double_precision.s -o test.o 2>/dev/null
223
if test "$?" == 0; then
224
225
226
227
228
229
230
231
232
233
234
235
  can_compile_sse_assembly=yes
  install_real_sse_assembly=yes
  install_complex_sse_assembly=yes
else
  can_compile_sse_assembly=no
  install_real_sse_assembly=no
  install_complex_sse_assembly=no
fi

rm -f ./test.o
AC_MSG_RESULT([${can_compile_sse_assembly}])

236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
if test x"${want_single_precision}" = x"yes" ; then
  AC_MSG_CHECKING(whether single-precision SSE assembly kernel can be compiled)

  $CC -c $srcdir/src/elpa2_kernels/elpa2_kernels_asm_x86_64_single_precision.s -o test.o 2>/dev/null
  if test "$?" == 0; then
    can_compile_sse_assembly=yes
    install_real_sse_assembly=yes
    install_complex_sse_assembly=yes
  else
    can_compile_sse_assembly=no
    install_real_sse_assembly=no
    install_complex_sse_assembly=no
  fi
  rm -f ./test.o
  AC_MSG_RESULT([${can_compile_sse_assembly}])
  if test x"${can_compile_sse_assembly}" = x"no" ; then
    AC_MSG_WARN([Cannot compile single-precision SSE assembly kernel: disabling SSE assembly kernels alltogether])
  fi
fi

256
dnl check whether on can compile with sse-gcc intrinsics
257
AC_MSG_CHECKING(whether we can compile SSE3 with gcc intrinsics in C)
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
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_intrinsics=yes],
 [can_compile_sse_intrinsics=no]
)
AC_MSG_RESULT([${can_compile_sse_intrinsics}])

if test "${can_compile_sse_intrinsics}" = "yes"; then
  install_real_sse_intrinsics=yes
273
274
275
276
  install_real_sse_block2=yes
  install_real_sse_block4=yes
  install_real_sse_block6=yes

277
  install_complex_sse_intrinsics=yes
278
279
  install_complex_sse_block1=yes
  install_complex_sse_block2=yes
280
else
281
  install_real_sse_intrinsics=no
282
283
284
285
  install_real_sse_block2=no
  install_real_sse_block4=no
  install_real_sse_block6=no

286
  install_complex_sse_intrinsics=no
287
288
  install_complex_sse_block1=no
  install_complex_sse_block2=no
289
fi
290

Andreas Marek's avatar
Andreas Marek committed
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
AC_MSG_CHECKING(whether --enable-avx is specified)
AC_ARG_ENABLE([avx],
		AS_HELP_STRING([--enable-avx],
			       [check whether AVX kernels can be build, default yes]),
	      [check_avx=no],
	      [check_avx=yes])
AC_MSG_RESULT([$check_avx])

if test "${check_avx}" = "yes"; then
  dnl check whether one can compile with avx - gcc intrinsics
  dnl first pass: try with specified CFLAGS and CXXFLAGS
  AC_MSG_CHECKING([whether we can compile AVX gcc intrinsics in C])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <x86intrin.h>
   int main(int argc, char **argv){
   double* q;
   __m256d a1_1 = _mm256_load_pd(q);
   return 0;
   }
   ])],
   [can_compile_avx=yes],
   [can_compile_avx=no]
  )
  AC_MSG_RESULT([${can_compile_avx}])
else
  can_compile_avx=no
fi
318

Andreas Marek's avatar
Andreas Marek committed
319
#if test "${can_compile_avx}" = "yes" ; then
320
#  AC_MSG_CHECKING([whether we can compile AVX gcc intrinsics in C++])
Andreas Marek's avatar
Andreas Marek committed
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
#  AC_LANG_PUSH([C++])
#  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
#   #include <x86intrin.h>
#   int main(int argc, char **argv){
#   double* q;
#   __m256d a1_1 = _mm256_load_pd(q);
#   return 0;
#   }
#   ])],
#   [can_compile_avx=yes],
#   [can_compile_avx=no]
#  )
#  AC_LANG_POP([C++])
#  AC_MSG_RESULT([${can_compile_avx}])
#  if test "${can_compile_avx}" = "no" ; then
#    AC_MSG_WARN([Cannot compile C++ with AVX: disabling AVX alltogether])
#  fi
#fi
339
340

dnl first pass: try with specified CFLAGS and CXXFLAGS
Andreas Marek's avatar
Andreas Marek committed
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366

AC_MSG_CHECKING(whether --enable-avx is specified)
AC_ARG_ENABLE([avx],
		AS_HELP_STRING([--enable-avx],
			       [check whether AVX kernels can be build, default yes]),
	      [check_avx=no],
	      [check_avx=yes])
AC_MSG_RESULT([$check_avx])

if test "${check_avx}" = "yes"; then
  AC_MSG_CHECKING([whether we can compile AVX intrinsics in C])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <x86intrin.h>
   int main(int argc, char **argv){
   double* q;
   __m256d a1_1 = _mm256_load_pd(q);
   return 0;
   }
   ])],
   [can_compile_avx=yes],
   [can_compile_avx=no]
  )
  AC_MSG_RESULT([${can_compile_avx}])
else
  can_compile_avx=no
fi
367

368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
dnl if test "${can_compile_avx}" = "yes" ; then
dnl   AC_MSG_CHECKING([whether we can compile AVX intrinsics in C++])
dnl   AC_LANG_PUSH([C++])
dnl   AC_COMPILE_IFELSE([AC_LANG_SOURCE([
dnl    #include <x86intrin.h>
dnl    int main(int argc, char **argv){
dnl    double* q;
dnl    __m256d a1_1 = _mm256_load_pd(q);
dnl    return 0;
dnl    }
dnl    ])],
dnl    [can_compile_avx=yes],
dnl    [can_compile_avx=no]
dnl   )
dnl   AC_LANG_POP([C++])
dnl   AC_MSG_RESULT([${can_compile_avx}])
dnl   if test "${can_compile_avx}" = "no" ; then
dnl     AC_MSG_WARN([Cannot compile C++ with AVX: disabling AVX alltogether])
dnl   fi
dnl fi
388

Andreas Marek's avatar
Andreas Marek committed
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
AC_MSG_CHECKING(whether --enable-avx2 is specified)
AC_ARG_ENABLE([avx2],
		AS_HELP_STRING([--enable-avx2],
			       [check whether AVX2 kernels can be build, default yes]),
	      [check_avx2=no],
	      [check_avx2=yes])
AC_MSG_RESULT([$check_avx2])

if test "${check_avx2}" = "yes"; then
  AC_MSG_CHECKING([whether we can compile AVX2 gcc intrinsics in C])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <x86intrin.h>
   int main(int argc, char **argv){
   double* q;
   __m256d q1 = _mm256_load_pd(q);
   __m256d y1 = _mm256_fmadd_pd(q1, q1, q1);
   return 0;
   }
   ])],
   [can_compile_avx2=yes],
   [can_compile_avx2=no]
  )
  AC_MSG_RESULT([${can_compile_avx2}])
else
  can_compile_avx2=no
fi

Andreas Marek's avatar
Andreas Marek committed
416
#if test "${can_compile_avx2}" = "yes" ; then
417
#  AC_MSG_CHECKING([whether we can compile AVX2 gcc intrinsics in C++])
Andreas Marek's avatar
Andreas Marek committed
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
#  AC_LANG_PUSH([C++])
#  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
#   #include <x86intrin.h>
#   int main(int argc, char **argv){
#   double* q;
#   __m256d q1 = _mm256_load_pd(q);
#   __m256d y1 = _mm256_fmadd_pd(q1, q1, q1);
#   return 0;
#   }
#   ])],
#   [can_compile_avx2=yes],
#   [can_compile_avx2=no]
#  )
#  AC_LANG_POP([C++])
#  AC_MSG_RESULT([${can_compile_avx2}])
#  if test "${can_compile_avx2}" = "no" ; then
#    AC_MSG_WARN([Cannot compile C++ with AVX2!])
#  fi
#fi
437

Andreas Marek's avatar
Andreas Marek committed
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
AC_MSG_CHECKING(whether --enable-avx512 is specified)
AC_ARG_ENABLE([avx512],
		AS_HELP_STRING([--enable-avx512],
			       [check whether AVX512 kernels can be build, default yes]),
	      [check_avx512=no],
	      [check_avx512=yes])
AC_MSG_RESULT([$check_avx512])

if test "${check_avx512}" = "yes"; then
  AC_MSG_CHECKING([whether we can compile AVX512 gcc intrinsics in C])
  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
   #include <x86intrin.h>
   int main(int argc, char **argv){
   double* q;
   __m512d q1 = _mm512_load_pd(q);
   __m512d y1 = _mm512_fmadd_pd(q1, q1, q1);
   return 0;
   }
   ])],
   [can_compile_avx512=yes],
   [can_compile_avx512=no]
  )
  AC_MSG_RESULT([${can_compile_avx512}])
else
  can_compile_avx512=no
fi
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486

#if test "${can_compile_avx512}" = "yes" ; then
#  AC_MSG_CHECKING([whether we can compile AVX512 gcc intrinsics in C++])
#  AC_LANG_PUSH([C++])
#  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
#   #include <x86intrin.h>
#   int main(int argc, char **argv){
#   double* q;
#   __m512d q1 = _mm512_load_pd(q);
#   __m512d y1 = _mm512_fmadd_pd(q1, q1, q1);
#   return 0;
#   }
#   ])],
#   [can_compile_avx512=yes],
#   [can_compile_avx512=no]
#  )
#  AC_LANG_POP([C++])
#  AC_MSG_RESULT([${can_compile_avx512}])
#  if test "${can_compile_avx512}" = "no" ; then
#    AC_MSG_WARN([Cannot compile C++ with AVX512!])
#  fi
#fi

487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
dnl if test "${can_compile_avx2}" = "yes" ; then
dnl   AC_MSG_CHECKING([whether we can compile AVX2 intrinsics in C++])
dnl   AC_LANG_PUSH([C++])
dnl   AC_COMPILE_IFELSE([AC_LANG_SOURCE([
dnl    #include <x86intrin.h>
dnl    int main(int argc, char **argv){
dnl    double* q;
dnl    __m256d q1 = _mm256_load_pd(q);
dnl    __m256d y1 = _mm256_fmadd_pd(q1, q1, q1);
dnl    return 0;
dnl    }
dnl    ])],
dnl    [can_compile_avx2=yes],
dnl    [can_compile_avx2=no]
dnl   )
dnl   AC_LANG_POP([C++])
dnl   AC_MSG_RESULT([${can_compile_avx2}])
dnl   if test "${can_compile_avx2}" = "no" ; then
dnl     AC_MSG_WARN([Cannot compile C++ with AVX2!])
dnl   fi
dnl fi
508

509
if test "${can_compile_avx}" = "yes" ; then
Andreas Marek's avatar
Andreas Marek committed
510
511
512
513
514
515
  install_real_avx_block2=yes
  install_real_avx_block4=yes
  install_real_avx_block6=yes

  install_complex_avx_block1=yes
  install_complex_avx_block2=yes
Andreas Marek's avatar
Andreas Marek committed
516
517
518
519
520
521
522
else
  install_real_avx_block2=no
  install_real_avx_block4=no
  install_real_avx_block6=no

  install_complex_avx_block1=no
  install_complex_avx_block2=no
523
524
fi

525
526
527
528
if test "${can_compile_avx2}" = "yes" ; then
  install_real_avx2_block2=yes
  install_real_avx2_block4=yes
  install_real_avx2_block6=yes
Andreas Marek's avatar
Andreas Marek committed
529

530
531
532
533
534
535
536
537
538
539
  install_complex_avx2_block1=yes
  install_complex_avx2_block2=yes
else
  install_real_avx2_block2=no
  install_real_avx2_block4=no
  install_real_avx2_block6=no

  install_complex_avx2_block1=no
  install_complex_avx2_block2=no
fi
540

541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
if test "${can_compile_avx512}" = "yes" ; then
  install_real_avx512_block2=yes
  install_real_avx512_block4=yes
  install_real_avx512_block6=yes

  install_complex_avx512_block1=yes
  install_complex_avx512_block2=yes
else
  install_real_avx512_block2=no
  install_real_avx512_block4=no
  install_real_avx512_block6=no

  install_complex_avx512_block1=no
  install_complex_avx512_block2=no
fi


558
559
560
AM_CONDITIONAL([HAVE_SSE_ASSEMBLY],[test x"$can_compile_sse_assembly" = x"yes"])
if test x"${can_compile_sse_assembly}" = x"yes" ; then
  AC_DEFINE([HAVE_SSE_ASSEMBLY],[1],[assembly SSE is supported on this CPU])
561
fi
562
563
564
AM_CONDITIONAL([HAVE_SSE_INTRINSICS],[test x"$can_compile_sse_intrinsics" = x"yes"])
if test x"${can_compile_sse_intrinsics}" = x"yes" ; then
  AC_DEFINE([HAVE_SSE_INTRINSICS],[1],[gcc intrinsics SSE is supported on this CPU])
565
fi
566

567
568
569
570
571
572
573
AM_CONDITIONAL([HAVE_AVX],[test x"$can_compile_avx" = x"yes"])
if test x"${can_compile_avx}" = x"yes" ; then
  AC_DEFINE([HAVE_AVX],[1],[AVX is supported on this CPU])
fi
AM_CONDITIONAL([HAVE_AVX2],[test x"$can_compile_avx2" = x"yes"])
if test x"${can_compile_avx2}" = x"yes" ; then
  AC_DEFINE([HAVE_AVX2],[1],[AVX2 is supported on this CPU])
574
fi
575
576
577
578
AM_CONDITIONAL([HAVE_AVX512],[test x"$can_compile_avx512" = x"yes"])
if test x"${can_compile_avx512}" = x"yes" ; then
  AC_DEFINE([HAVE_AVX512],[1],[AVX512 is supported on this CPU])
fi
579
580
581
582
583
584
585
586

dnl set the AVX optimization flags if this option is specified
AC_MSG_CHECKING(whether AVX optimization flags should be set automatically)
AC_ARG_WITH([avx-optimization],
		AS_HELP_STRING([--with-avx-optimization],
			       [use AVX optimization, default no.]),
	      [with_avx_optimization=yes],
	      [with_avx_optimization=no])
Andreas Marek's avatar
Andreas Marek committed
587
AC_MSG_RESULT([${with_avx_optimization}])
588
if test x"${with_avx_optimization}" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
589
590
591
 CFLAGS="$CFLAGS -funsafe-loop-optimizations -funsafe-math-optimizations -ftree-vect-loop-version -ftree-vectorize"
 CXXFLAGS="$CXXFLAGS -funsafe-loop-optimizations -funsafe-math-optimizations -ftree-vect-loop-version -ftree-vectorize"
fi
592

593
AC_LANG([Fortran])
594
595
596
AC_FC_FREEFORM
AC_FC_MODULE_FLAG
AC_FC_MODULE_OUTPUT_FLAG
597
AC_FC_LIBRARY_LDFLAGS
598
599
600
601

save_FCFLAGS=$FCFLAGS
save_LDFLAGS=$LDFLAGS

602
603
604
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])

605
606
FCFLAGS="$FCFLAGS $SCALAPACK_FCFLAGS"
LDFLAGS="$LDFLAGS $SCALAPACK_LDFLAGS"
607

608
dnl check whether fortran error_unit is defined
609
AC_MSG_CHECKING([whether Fortran module iso_fortran_env is available])
610
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
611
612
613
614
615
616
617
618
619
620
  program test_error_unit
    use ISO_FORTRAN_ENV, only : error_unit
    implicit none

    write(error_unit,*) "error_unit is defined"
  end program
])],
  [can_use_iso_fortran_env=yes],
  [can_use_iso_fortran_env=no]
)
621
AC_MSG_RESULT([${can_use_iso_fortran_env}])
622

623
624
625
626
dnl check whether one can link with specified MKL (desired method)
AC_MSG_CHECKING([whether we can compile a Fortran program using MKL])


627
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
628
  program test_mkl
629
    use mkl_service
630
631
632
633
634
635
636
637
638
639
640
641
    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"$can_compile_with_mkl" = x"yes" ; then
  AC_MSG_CHECKING([whether we can link a Fortran program with MKL])
642
  AC_LINK_IFELSE([AC_LANG_SOURCE([
643
    program test_mkl
644
      use mkl_service
645
646
647
648
649
      character*198 :: string
      call mkl_get_version_string(string)
      write(*,'(a)') string
    end program
  ])],
Andreas Marek's avatar
Andreas Marek committed
650
651
    [have_mkl=yes],
    [have_mkl=no]
652
  )
Andreas Marek's avatar
Andreas Marek committed
653
  AC_MSG_RESULT([${have_mkl}])
654
655
656
fi

dnl if not mkl, check all the necessary individually
657
if test x"${have_mkl}" = x"yes" ; then
658
659
660
661
  WITH_MKL=1
else

  dnl first check blas
662
  AC_SEARCH_LIBS([dgemm],[openblas satlas blas],[have_blas=yes],[have_blas=no])
Andreas Marek's avatar
Andreas Marek committed
663
664
  AC_MSG_CHECKING([whether we can link a program with a blas lib])
  AC_MSG_RESULT([${have_blas}])
665

666
  if test x"${have_blas}" = x"no" ; then
667
    AC_MSG_ERROR([could not link with blas: specify path])
668
  fi
669
  dnl now lapack
Andreas Marek's avatar
Andreas Marek committed
670
671
672
  AC_SEARCH_LIBS([dlarrv],[lapack],[have_lapack=yes],[have_lapack=no])
  AC_MSG_CHECKING([whether we can link a program with a lapack lib])
  AC_MSG_RESULT([${have_lapack}])
673

674
  if test x"${have_lapack}" = x"no" ; then
675
    AC_MSG_ERROR([could not link with lapack: specify path])
676
  fi
Andreas Marek's avatar
Andreas Marek committed
677

678
  if test x"${with_mpi}" = x"yes"; then
679
    dnl test whether scalapack already contains blacs
680
    scalapack_libs="mpiscalapack scalapack scalapack-openmpi"
681
682
683
684
685
686
687
688
689
690
    old_LIBS="$LIBS"
    for lib in ${scalapack_libs}; do
      LIBS="-l${lib} ${old_LIBS}"
      AC_MSG_CHECKING([whether -l${lib} already contains a BLACS implementation])
      AC_LINK_IFELSE([AC_LANG_FUNC_LINK_TRY([blacs_gridinit])],[blacs_in_scalapack=yes],[blacs_in_scalapack=no])
      AC_MSG_RESULT([${blacs_in_scalapack}])
      if test x"${blacs_in_scalapack}" = x"yes"; then
        break
      fi
    done
691

692
693
    if test x"${blacs_in_scalapack}" = x"no"; then
      LIBS="${old_LIBS}"
694

695
696
      dnl Test for stand-alone blacs
      AC_SEARCH_LIBS([bi_f77_init],[mpiblacsF77init],[],[],[-lmpiblacs])
697
698
699
700
701
702

      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
703
      AC_SEARCH_LIBS([blacs_gridinit],[mpiblacs blacs],[have_blacs=yes],[have_blacs=no])
Andreas Marek's avatar
Andreas Marek committed
704

705
706
707
      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
708
    fi
709

710
    AC_SEARCH_LIBS([pdtran],[$scalapack_libs],[have_scalapack=yes],[have_scalapack=no])
Andreas Marek's avatar
Andreas Marek committed
711

712
713
714
    if test x"${have_scalapack}" = x"no" ; then
      AC_MSG_ERROR([could not link with scalapack: specify path])
    fi
715
716
717
718
  fi

  dnl check whether we can link alltogehter
  AC_MSG_CHECKING([whether we can link a Fortran program with all blacs/scalapack])
719
  AC_LINK_IFELSE([AC_LANG_SOURCE([
720
721
    program dgemm_test

722
723
      integer , parameter:: M = 4, N = 3, K = 2
      real :: A(M,K), B(K,N), C(M,N)
724

725
      call dgemm('N','N',M,N,K,1.0,A,M,B,K,0.0,C,M)
726
727
728
729
730
731
732
733

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

734
   if test x"${can_link_with_blacs_scalapack}" = x"yes" ; then
735
736
     WITH_BLACS=1
   else
737
   AC_MSG_ERROR([We can neither link with MKL or another Scalpack. Please specify SCALAPACK_LDFLAGS and SCALAPACK_FCFLAGS!])
738
739
740
   fi
fi

741
742
743
744
dnl important: reset them again!
FCFLAGS=$save_FCFLAGS
LDFLAGS=$save_LDFLAGS

Andreas Marek's avatar
Andreas Marek committed
745
746
747
748
dnl check for intrinsic fortran function of 2003 standard
AC_MSG_CHECKING([whether we can use the intrinsic Fortran function "get_environment_variable"])


749
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
750
751
752
753
754
755
756
757
758
  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([${fortran_can_check_environment}])
759
760
761
if test x"${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
762

763
dnl check whether BAND_TO_FULL_BLOCKING is set
764
use_band_to_full_blocking=yes
765
766
AC_MSG_CHECKING(whether BAND_TO_FLULL_BLOCKING is requested)
AC_ARG_ENABLE(band-to-full-blocking,[AS_HELP_STRING([--enable-band-to-full-blocking],
767
	                                            [build ELPA2 with blocking in band_to_full (default: enabled)])],
768
769
	      want_band_to_full_blocking="yes", want_to_full_blocking="no")
AC_MSG_RESULT([${want_band_to_full_blocking}])
770
771
772
773
774
775
776
777

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

AM_CONDITIONAL([BAND_TO_FULL_BLOCKING],[test x"$use_band_to_full_blocking" = x"yes"])
if test x"${use_band_to_full_blocking}" = x"yes"; then
       AC_DEFINE([BAND_TO_FULL_BLOCKING], [1], [use blocking in trans_ev_band_to_full])
778
779
fi

780
781
dnl check whether GPU version is requested

782
#CUDA_INSTALL_PATH="/usr/local/cuda/"
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
#CUDA_SDK_INSTALL_PATH="/usr/local/NVIDIA_GPU_Computing_SDK"

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


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

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

dnl setup nvcc flags and use them in later tests
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
user_sets_gpu_compute_capability="no"
AC_MSG_CHECKING(whether a GPU compute capability is specified)
AC_ARG_WITH([GPU-compute-capability],
            [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"])
AC_MSG_RESULT([${user_sets_gpu_compute_capability}])

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

818
819
if test x"${want_gpu}" = x"yes" ; then
  AC_LANG_PUSH([C])
820
  CUDA_CFLAGS="$CUDA_CFLAGS -arch $cuda_compute_capability -O2 -I$CUDA_INSTALL_PATH/include"
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
  LDFLAGS="$LDFLAGS -L$CUDA_INSTALL_PATH/lib64"
  NVCCFLAGS="$NVCCFLAGS $CUDA_CFLAGS $CUDA_LDFLAGS"
  NVCC="nvcc"
  AC_SUBST(NVCC)
  AC_SUBST(NVCCFLAGS)

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

  dnl check whether we find cublas
  AC_SEARCH_LIBS([cublasDgemm],[cublas],[have_cublas=yes],[have_cublas=no])
  if test x"${have_cublas}" = x"no"; then
    AC_MSG_ERROR([Could not link cublas; try to set the cuda-path or disable GPU support])
  fi
  AC_SEARCH_LIBS([cudaMemcpy],[cudart],[have_cudart=yes],[have_cudart=no])
  if test x"${have_cudart}" = x"no"; then
    AC_MSG_ERROR([Could not link cudart; try to set the cuda-path or disable GPU support])
  fi
  AC_LANG_POP([C])
  install_gpu=yes
Andreas Marek's avatar
Andreas Marek committed
844
  can_compile_gpu=yes
845
846
fi

Andreas Marek's avatar
Andreas Marek committed
847
848
849
850
851
852
853
854
dnl now check which kernels can be compiled

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

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

855
AC_LINK_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
856
857
858
  program test_bgp
    complex*16 :: y3,q3,h2
    y3 = fxcpmadd(y3,q3,h2)
859

Andreas Marek's avatar
Andreas Marek committed
860
861
862
863
864
865
866
  end program
])],
  [can_compile_bgp=yes],
  [can_compile_bgp=no]
)
AC_MSG_RESULT([${can_compile_bgp}])

867
if test x"${can_compile_bgp}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
868
869
  install_real_bgp=yes
  install_complex_bgp=yes
Andreas Marek's avatar
Andreas Marek committed
870
871
872
else
  install_real_bgp=no
  install_complex_bgp=no
Andreas Marek's avatar
Andreas Marek committed
873
874
875
876
877
fi

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

878
AC_LINK_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
879
  program test_bgq
Andreas Marek's avatar
Andreas Marek committed
880
881
    VECTOR(REAL(8))::QPX_h2
    real*8         :: hh(10,2)
Andreas Marek's avatar
Andreas Marek committed
882
    QPX_h2 = VEC_SPLATS(hh(2,2))
883

Andreas Marek's avatar
Andreas Marek committed
884
885
886
887
888
889
890
  end program
])],
  [can_compile_bgq=yes],
  [can_compile_bgq=no]
)
AC_MSG_RESULT([${can_compile_bgq}])

891
if test x"${can_compile_bgq}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
892
893
  install_real_bgq=yes
  install_complex_bgq=yes
Andreas Marek's avatar
Andreas Marek committed
894
895
896
else
  install_real_bgq=no
  install_complex_bgq=no
Andreas Marek's avatar
Andreas Marek committed
897
898
fi

899
900
dnl Test possibility of 'use mpi', if requested
if test x"${with_mpi}" = x"yes" ; then
901
902
903
  AC_ARG_ENABLE([mpi-module],
                AS_HELP_STRING([--disable-mpi-module],
                               [Do not use the Fortran MPI module, get interfaces by 'include "mpif.h')]),
904
                [],
905
906
                [enable_mpi_module=yes])
  if test x"${enable_mpi_module}" = x"yes" ; then
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
    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
922
      AC_MSG_ERROR([Could not compile a Fortran program with an 'use mpi' statement. You can try again with --disable-mpi-module])
923
    fi
924
  fi
Andreas Marek's avatar
Andreas Marek committed
925
926
fi

927
928
929
930
dnl default value
use_specific_real_kernel=no
use_specific_complex_kernel=no

931
932
933
dnl GPU version only
m4_include([m4/ax_elpa_gpu_version_only.m4])
DEFINE_OPTION_GPU_SUPPORT_ONLY([gpu-version-only],[gpu-support],[install_gpu])
Andreas Marek's avatar
Andreas Marek committed
934

935
936
dnl last check whether user wants to compile only a specific kernel
dnl
937
m4_include([m4/ax_elpa_specific_kernels.m4])
Andreas Marek's avatar
Andreas Marek committed
938

939
dnl real kernels
940
dnl do not remove this variable it is needed in the macros
941

942
943
dnl generic kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-generic-kernel-only],[generic-kernel],[install_real_generic])
Andreas Marek's avatar
Andreas Marek committed
944

945
946
dnl generic-simple kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-generic-simple-kernel-only],[generic-simple-kernel],[install_real_generic_simple])
Andreas Marek's avatar
Andreas Marek committed
947

948
dnl sse kernel
949
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-sse-assembly-kernel-only],[sse-assembly-kernel],[install_real_sse_assembly])
Andreas Marek's avatar
Andreas Marek committed
950

951
952
dnl bgp kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-bgp-kernel-only],[bgp-kernel],[install_real_bgp])
Andreas Marek's avatar
Andreas Marek committed
953

954
955
dnl bgq kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-bgq-kernel-only],[bgq-kernel],[install_real_bgq])
Andreas Marek's avatar
Andreas Marek committed
956

957
958
dnl real-sse-block2 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-sse-block2-kernel-only],[real-sse-block2-kernel],[install_real_sse_block2])
959

960
dnl real-sse-block4 kernel
Andreas Marek's avatar
Andreas Marek committed
961
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-sse-block4-kernel-only],[real-sse-block4-kernel],[install_real_sse_block4])
962

963
964
dnl real-sse-block6 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-sse-block6-kernel-only],[real-sse-block6-kernel],[install_real_sse_block6])
965

966
967
dnl real-avx-block2 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx-block2-kernel-only],[real-avx-block2-kernel],[install_real_avx_block2])
Andreas Marek's avatar
Andreas Marek committed
968

969
dnl real-avx-block4 kernel
Andreas Marek's avatar
Andreas Marek committed
970
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx-block4-kernel-only],[real-avx-block4-kernel],[install_real_avx_block4])
971

972
973
dnl real-avx-block6 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx-block6-kernel-only],[real-avx-block6-kernel],[install_real_avx_block6])
Andreas Marek's avatar
Andreas Marek committed
974

975
976
dnl real-avx2-block2 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx2-block2-kernel-only],[real-avx2-block2-kernel],[install_real_avx2_block2])
977

978
dnl real-avx2-block4 kernel
Andreas Marek's avatar
Andreas Marek committed
979
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx2-block4-kernel-only],[real-avx2-block4-kernel],[install_real_avx2_block4])
980

981
982
dnl real-avx2-block6 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx2-block6-kernel-only],[real-avx2-block6-kernel],[install_real_avx2_block6])
Andreas Marek's avatar
Andreas Marek committed
983

984
985
986
987
dnl real-avx512-block2 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx512-block2-kernel-only],[real-avx512-block2-kernel],[install_real_avx512_block2])

dnl real-avx512-block4 kernel
Andreas Marek's avatar
Andreas Marek committed
988
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx512-block4-kernel-only],[real-avx512-block4-kernel],[install_real_avx512_block4])
989
990
991
992
993

dnl real-avx512-block6 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx512-block6-kernel-only],[real-avx512-block6-kernel],[install_real_avx512_block6])


Andreas Marek's avatar
Andreas Marek committed
994
dnl complex kernels
995

996
997
dnl generic kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-generic-kernel-only],[generic-kernel],[install_complex_generic])
Andreas Marek's avatar
Andreas Marek committed
998

999
1000
dnl generic-simple kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-generic-simple-kernel-only],[generic-simple-kernel],[install_complex_generic_simple])
Andreas Marek's avatar
Andreas Marek committed
1001

1002
dnl sse kernel
1003
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-sse-assembly-kernel-only],[sse-assembly-kernel],[install_complex_sse_assembly])
Andreas Marek's avatar
Andreas Marek committed
1004

1005
1006
dnl complex-bqp kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-bgp-kernel-only],[bgp-kernel],[install_complex_bgp])
Andreas Marek's avatar
Andreas Marek committed
1007

1008
1009
dnl complex-bqq kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-bgq-kernel-only],[bgq-kernel],[install_complex_bgq])
Andreas Marek's avatar
Andreas Marek committed
1010

1011
1012
dnl complex-sse-block1 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-sse-block1-kernel-only],[complex-sse-block1-kernel],[install_complex_sse_block1])
1013

1014
1015
dnl complex-sse-block2 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-sse-block2-kernel-only],[complex-sse-block2-kernel],[install_complex_sse_block2])
1016

1017
1018
dnl complex-avx-block1 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-avx-block1-kernel-only],[complex-avx-block1-kernel],[install_complex_avx_block1])
Andreas Marek's avatar
Andreas Marek committed
1019

1020
1021
dnl complex-avx-block2 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-avx-block2-kernel-only],[complex-avx-block2-kernel],[install_complex_avx_block2])
Andreas Marek's avatar
Andreas Marek committed
1022

1023
1024
dnl complex-avx2-block1 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-avx2-block1-kernel-only],[complex-avx2-block1-kernel],[install_complex_avx2_block1])
1025

1026
1027
dnl complex-avx2-block2 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-avx2-block2-kernel-only],[complex-avx2-block2-kernel],[install_complex_avx2_block2])
1028

1029
1030
1031
1032
1033
1034
1035
dnl complex-avx512-block1 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-avx512-block1-kernel-only],[complex-avx512-block1-kernel],[install_complex_avx512_block1])

dnl complex-avx512-block2 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-avx512-block2-kernel-only],[complex-avx512-block2-kernel],[install_complex_avx512_block2])


1036

1037
dnl set the conditionals according to the previous tests
1038

1039
if test x"${can_use_iso_fortran_env}" = x"yes" ; then
1040
1041
1042
 AC_DEFINE([HAVE_ISO_FORTRAN_ENV],[1],[can use module iso_fortran_env])
fi

1043
1044
1045
AM_CONDITIONAL([WITH_GPU_VERSION],[test x"$install_gpu" = x"yes"])
if test x"${install_gpu}" = x"yes" ; then
  AC_DEFINE([WITH_GPU_VERSION],[1],[enable GPU support])
1046
fi
1047

Andreas Marek's avatar
Andreas Marek committed
1048
AM_CONDITIONAL([WITH_REAL_GENERIC_KERNEL],[test x"$install_real_generic" = x"yes"])
1049
if test x"${install_real_generic}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1050
1051
1052
1053
 AC_DEFINE([WITH_REAL_GENERIC_KERNEL],[1],[can use real generic kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_GENERIC_KERNEL],[test x"$install_complex_generic" = x"yes"])
1054
if test x"${install_complex_generic}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1055
1056
1057
1058
 AC_DEFINE([WITH_COMPLEX_GENERIC_KERNEL],[1],[can use complex generic kernel])
fi

AM_CONDITIONAL([WITH_REAL_GENERIC_SIMPLE_KERNEL],[test x"$install_real_generic_simple" = x"yes"])
1059
if test x"${install_real_generic_simple}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1060
1061
1062
1063
 AC_DEFINE([WITH_REAL_GENERIC_SIMPLE_KERNEL],[1],[can use real generic-simple kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_GENERIC_SIMPLE_KERNEL],[test x"$install_complex_generic_simple" = x"yes"])
1064
if test x"${install_complex_generic_simple}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1065
1066
1067
 AC_DEFINE([WITH_COMPLEX_GENERIC_SIMPLE_KERNEL],[1],[can use complex generic-simple kernel])
fi

1068
1069
1070
AM_CONDITIONAL([WITH_REAL_SSE_ASSEMBLY_KERNEL],[test x"$install_real_sse_assembly" = x"yes"])
if test x"${install_real_sse_assembly}" = x"yes" ; then
 AC_DEFINE([WITH_REAL_SSE_ASSEMBLY_KERNEL],[1],[can use real SSE assembly kernel])
Andreas Marek's avatar
Andreas Marek committed
1071
1072
fi

1073
1074
1075
AM_CONDITIONAL([WITH_COMPLEX_SSE_ASSEMBLY_KERNEL],[test x"$install_complex_sse_assembly" = x"yes"])
if test x"${install_complex_sse_assembly}" = x"yes" ; then
 AC_DEFINE([WITH_COMPLEX_SSE_ASSEMBLY_KERNEL],[1],[can use complex SSE assembly kernel])
Andreas Marek's avatar
Andreas Marek committed
1076
1077
fi

1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
AM_CONDITIONAL([WITH_REAL_SSE_BLOCK2_KERNEL],[test x"$install_real_sse_block2" = x"yes"])
if test x"${install_real_sse_block2}" = x"yes" ; then
 AC_DEFINE([WITH_REAL_SSE_BLOCK2_KERNEL],[1],[can use real_sse_block2 kernel])
fi

AM_CONDITIONAL([WITH_REAL_SSE_BLOCK4_KERNEL],[test x"$install_real_sse_block4" = x"yes"])
if test x"${install_real_sse_block4}" = x"yes" ; then
 AC_DEFINE([WITH_REAL_SSE_BLOCK4_KERNEL],[1],[can use real_sse_block4 kernel])
fi

AM_CONDITIONAL([WITH_REAL_SSE_BLOCK6_KERNEL],[test x"$install_real_sse_block6" = x"yes"])
if test x"${install_real_sse_block6}" = x"yes" ; then
 AC_DEFINE([WITH_REAL_SSE_BLOCK6_KERNEL],[1],[can use real_sse_block6 kernel])
fi

Andreas Marek's avatar
Andreas Marek committed
1093
AM_CONDITIONAL([WITH_REAL_AVX_BLOCK2_KERNEL],[test x"$install_real_avx_block2" = x"yes"])
1094
if test x"${install_real_avx_block2}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1095
1096
1097
1098
 AC_DEFINE([WITH_REAL_AVX_BLOCK2_KERNEL],[1],[can use real_avx_block2 kernel])
fi

AM_CONDITIONAL([WITH_REAL_AVX_BLOCK4_KERNEL],[test x"$install_real_avx_block4" = x"yes"])
1099
if test x"${install_real_avx_block4}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1100
1101
1102
1103
 AC_DEFINE([WITH_REAL_AVX_BLOCK4_KERNEL],[1],[can use real_avx_block4 kernel])
fi

AM_CONDITIONAL([WITH_REAL_AVX_BLOCK6_KERNEL],[test x"$install_real_avx_block6" = x"yes"])
1104
if test x"${install_real_avx_block6}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1105
1106
1107
 AC_DEFINE([WITH_REAL_AVX_BLOCK6_KERNEL],[1],[can use real_avx_block6 kernel])
fi

1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
AM_CONDITIONAL([WITH_REAL_AVX2_BLOCK2_KERNEL],[test x"$install_real_avx2_block2" = x"yes"])
if test x"${install_real_avx2_block2}" = x"yes" ; then
 AC_DEFINE([WITH_REAL_AVX2_BLOCK2_KERNEL],[1],[can use real_avx2_block2 kernel])
fi

AM_CONDITIONAL([WITH_REAL_AVX2_BLOCK4_KERNEL],[test x"$install_real_avx2_block4" = x"yes"])
if test x"${install_real_avx2_block4}" = x"yes" ; then
 AC_DEFINE([WITH_REAL_AVX2_BLOCK4_KERNEL],[1],[can use real_avx2_block4 kernel])
fi

AM_CONDITIONAL([WITH_REAL_AVX2_BLOCK6_KERNEL],[test x"$install_real_avx2_block6" = x"yes"])
if test x"${install_real_avx2_block6}" = x"yes" ; then
 AC_DEFINE([WITH_REAL_AVX2_BLOCK6_KERNEL],[1],[can use real_avx2_block6 kernel])
fi

1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
AM_CONDITIONAL([WITH_REAL_AVX512_BLOCK2_KERNEL],[test x"$install_real_avx512_block2" = x"yes"])
if test x"${install_real_avx512_block2}" = x"yes" ; then
 AC_DEFINE([WITH_REAL_AVX512_BLOCK2_KERNEL],[1],[can use real_avx512_block2 kernel])
fi

AM_CONDITIONAL([WITH_REAL_AVX512_BLOCK4_KERNEL],[test x"$install_real_avx512_block4" = x"yes"])
if test x"${install_real_avx512_block4}" = x"yes" ; then
 AC_DEFINE([WITH_REAL_AVX512_BLOCK4_KERNEL],[1],[can use real_avx512_block4 kernel])
fi

AM_CONDITIONAL([WITH_REAL_AVX512_BLOCK6_KERNEL],[test x"$install_real_avx512_block6" = x"yes"])
if test x"${install_real_avx512_block6}" = x"yes" ; then
 AC_DEFINE([WITH_REAL_AVX512_BLOCK6_KERNEL],[1],[can use real_avx512_block6 kernel])
fi

1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
AM_CONDITIONAL([WITH_COMPLEX_SSE_BLOCK1_KERNEL],[test x"$install_complex_sse_block1" = x"yes"])
if test x"${install_complex_sse_block1}" = x"yes" ; then
 AC_DEFINE([WITH_COMPLEX_SSE_BLOCK1_KERNEL],[1],[can use complex_sse_block1 kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_SSE_BLOCK2_KERNEL],[test x"$install_complex_sse_block2" = x"yes"])
if test x"${install_complex_sse_block2}" = x"yes" ; then
 AC_DEFINE([WITH_COMPLEX_SSE_BLOCK2_KERNEL],[1],[can use complex_sse_block2 kernel])
fi

Andreas Marek's avatar
Andreas Marek committed
1148
AM_CONDITIONAL([WITH_COMPLEX_AVX_BLOCK1_KERNEL],[test x"$install_complex_avx_block1" = x"yes"])
1149
if test x"${install_complex_avx_block1}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1150
1151
1152
1153
 AC_DEFINE([WITH_COMPLEX_AVX_BLOCK1_KERNEL],[1],[can use complex_avx_block1 kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_AVX_BLOCK2_KERNEL],[test x"$install_complex_avx_block2" = x"yes"])
1154
if test x"${install_complex_avx_block2}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1155
1156
1157
 AC_DEFINE([WITH_COMPLEX_AVX_BLOCK2_KERNEL],[1],[can use complex_avx_block2 kernel])
fi

1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
AM_CONDITIONAL([WITH_COMPLEX_AVX2_BLOCK1_KERNEL],[test x"$install_complex_avx2_block1" = x"yes"])
if test x"${install_complex_avx2_block1}" = x"yes" ; then
 AC_DEFINE([WITH_COMPLEX_AVX2_BLOCK1_KERNEL],[1],[can use complex_avx2_block1 kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_AVX2_BLOCK2_KERNEL],[test x"$install_complex_avx2_block2" = x"yes"])
if test x"${install_complex_avx2_block2}" = x"yes" ; then
 AC_DEFINE([WITH_COMPLEX_AVX2_BLOCK2_KERNEL],[1],[can use complex_avx2_block2 kernel])
fi

1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
AM_CONDITIONAL([WITH_COMPLEX_AVX512_BLOCK1_KERNEL],[test x"$install_complex_avx512_block1" = x"yes"])
if test x"${install_complex_avx512_block1}" = x"yes" ; then
 AC_DEFINE([WITH_COMPLEX_AVX512_BLOCK1_KERNEL],[1],[can use complex_avx512_block1 kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_AVX512_BLOCK2_KERNEL],[test x"$install_complex_avx512_block2" = x"yes"])
if test x"${install_complex_avx512_block2}" = x"yes" ; then
 AC_DEFINE([WITH_COMPLEX_AVX512_BLOCK2_KERNEL],[1],[can use complex_avx512_block2 kernel])
fi

Andreas Marek's avatar
Andreas Marek committed
1178
AM_CONDITIONAL([WITH_REAL_BGP_KERNEL],[test x"$install_real_bgp" = x"yes"])
1179
if test x"${install_real_bgp}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1180
1181
1182
1183
 AC_DEFINE([WITH_REAL_BGP_KERNEL],[1],[can use real BGP kernel])
fi

AM_CONDITIONAL([WITH_REAL_BGQ_KERNEL],[test x"$install_real_bgq" = x"yes"])
1184
if test x"${install_real_bgq}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1185
1186
1187
 AC_DEFINE([WITH_REAL_BGQ_KERNEL],[1],[can use real BGQ kernel])
fi

1188
if test x"${use_specific_complex_kernel}" = x"no" ; then
1189
  AC_DEFINE([WITH_NO_SPECIFIC_COMPLEX_KERNEL],[1],[do not use only one specific complex kernel (set at compile time)])
1190
else
1191
  AC_DEFINE([WITH_ONE_SPECIFIC_COMPLEX_KERNEL],[1],[use only one specific complex kernel (set at compile time)])
Andreas Marek's avatar
Andreas Marek committed
1192
1193
fi

1194
if test x"${use_specific_real_kernel}" = x"no" ; then
1195
  AC_DEFINE([WITH_NO_SPECIFIC_REAL_KERNEL],[1],[do not use only one specific real kernel (set at compile time)])
1196
1197
else
  AC_DEFINE([WITH_ONE_SPECIFIC_REAL_KERNEL],[1],[use only one specific real kernel (set at compile time)])
Andreas Marek's avatar
Andreas Marek committed
1198
1199
fi

1200
1201
LT_INIT

1202
1203
1204
1205
1206
1207
DX_PDF_FEATURE(OFF)
DX_PS_FEATURE(OFF)
DX_MAN_FEATURE(ON)
DX_HTML_FEATURE(ON)
DX_INIT_DOXYGEN([ELPA], [Doxyfile], [docs])

1208
USE_ASSUMED_SIZE=yes
Andreas Marek's avatar
Andreas Marek committed
1209
AC_MSG_CHECKING(whether assumed size Fortran arrays should be used)
1210
AC_ARG_ENABLE([assumed-size],
1211
1212
                AS_HELP_STRING([--disable-assumed-size],
			       [do NOT use assumed-size Fortran arrays]),
Andreas Marek's avatar
Andreas Marek committed
1213
1214
1215
1216
1217
			       [if test x"$enableval" = x"yes"; then
                                  USE_ASSUMED_SIZE=yes
				else
			          USE_ASSUMED_SIZE=no
				fi],
1218
              [USE_ASSUMED_SIZE=yes])
1219
1220
1221
AC_MSG_RESULT([${USE_ASSUMED_SIZE}])
AM_CONDITIONAL([WITH_USE_ASSUMED_SIZE],[test x"$USE_ASSUMED_SIZE" = x"yes"])
if test x"${USE_ASSUMED_SIZE}" = x"yes" ; then
1222
  AC_DEFINE([USE_ASSUMED_SIZE],[1],[for performance reasons use assumed size Fortran arrays, even if not debuggable])
1223
1224
fi

1225
1226
1227
if test x"${want_single_precision}" = x"yes" ; then
  AC_DEFINE([WANT_SINGLE_PRECISION_REAL],[1],[build also single-precision for real calculation])
  AC_DEFINE([WANT_SINGLE_PRECISION_COMPLEX],[1],[build also single-precision for complex calculation])
1228
fi
1229
1230
AM_CONDITIONAL([WANT_SINGLE_PRECISION_REAL],[test x"$want_single_precision" = x"yes"])
AM_CONDITIONAL([WANT_SINGLE_PRECISION_COMPLEX],[test x"$want_single_precision" = x"yes"])
1231

1232
AC_SUBST([MPI_BINARY])
1233
1234
1235
1236
1237
AC_SUBST([WITH_MKL])
AC_SUBST([WITH_BLACS])
AC_SUBST([with_amd_bulldozer_kernel])
AC_SUBST([FC_MODINC])
AC_SUBST([FC_MODOUT])
1238
1239
1240
AC_SUBST([OPENMP_CFLAGS])
AC_SUBST([OPENMP_FCFLAGS])
AC_SUBST([OPENMP_LDFLAGS])
1241
AC_SUBST([DOXYGEN_OUTPUT_DIR], [docs])
1242

1243
1244
rm -rf modules/ private_modules/ .fortran_dependencies/
mkdir modules private_modules
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254

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

1255
1256
1257
1258
1259
1260
1261
1262
1263

# Some part of libtool is too smart and tries to parse the output of
#   gfortran -v
# and catches anything that resembles a -l library link specification.
# Unfortunately, recent versions of gfortran emit
#   -l gfortran
# with a space between -l and gfortran. The empty -l is then included
# into "postdeps_FC" and causes linking errors later on.
postdeps_FC=$(echo $postdeps_FC | sed 's/-l //g')
1264

1265
1266
if test x"${with_mpi}" = x"yes"; then
  if test x"${enable_openmp}" = x"yes"; then
1267
	SUFFIX="_openmp"
1268
  else
1269
	SUFFIX=""
1270
1271
1272
  fi
else
  if test x"${enable_openmp}" = x"yes"; then
Lorenz Huedepohl's avatar
Typo    
Lorenz Huedepohl committed
1273
	SUFFIX="_onenode_openmp"
1274
1275
1276
  else
	SUFFIX="_onenode"
  fi
1277
fi
1278

1279
1280
AC_SUBST([SUFFIX])
AC_SUBST([PKG_CONFIG_FILE],[elpa${SUFFIX}-${PACKAGE_VERSION}.pc])
1281

1282
1283
AC_CONFIG_FILES([
  Makefile
1284
  Doxyfile
1285
  ${PKG_CONFIG_FILE}:elpa.pc.in
1286
])
1287

1288
1289
1290
1291
1292
1293
m4_include([m4/ax_fc_check_define.m4])
AC_MSG_CHECKING([if workaround for broken preprocessor is needed])

need_manual_cpp=no
AX_FC_CHECK_DEFINE([__INTEL_COMPILER],[need_manual_cpp=yes],[])
AX_FC_CHECK_DEFINE([__PGI],[need_manual_cpp=yes],[])
Lorenz Huedepohl's avatar
Lorenz Huedepohl committed
1294
1295
ACTUAL_FC="$FC"
AC_SUBST([ACTUAL_FC])
1296
1297
1298
1299
1300
1301
1302
if test x"$need_manual_cpp" = x"yes" ; then
	AC_MSG_RESULT([yes])
	FC="\$(top_srcdir)/manual_cpp $FC"
else
	AC_MSG_RESULT([no])
fi

Lorenz Huedepohl's avatar
Lorenz Huedepohl committed
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316

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


1317
AC_OUTPUT
1318

1319
if test "${can_compile_avx}" = "no" ; then
1320
#  if test x"${want_avx}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1321
    AC_MSG_WARN([Could not compile AVX instructions])
1322
#  fi
1323
fi
1324
if test "${can_compile_avx2}" = "no" ; then
1325
#  if test x"${want_avx2}" = x"yes" ; then
1326
    AC_MSG_WARN([Could not compile AVX2 instructions])
1327
#  fi
1328
fi
1329
1330
1331
1332
1333
1334
if test "${can_compile_avx512}" = "no" ; then
#  if test x"${want_avx512}" = x"yes" ; then
    AC_MSG_WARN([Could not compile AVX512 instructions])
#  fi
fi

1335
1336
1337
1338

if test "${can_compile_sse}" = "no" ; then
  AC_MSG_WARN([Could not compile SSE instructions])
fi
1339
1340
1341
1342
1343
1344
1345
echo " "
echo "***********************************************************************"
echo "*  This is a prerelease of the ELPA library 2016.11.001.pre library!  *"
echo "*  It contains an experimental API which will most likely change with *"
echo "*  the next stable release.                                           *"
echo "***********************************************************************"
echo " "
1346
make -f $srcdir/generated_headers.am generated-headers top_srcdir="$srcdir"