configure.ac 43.9 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# This only makes trouble for end users with out-of-date autoconf versions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
if test x"${enable_openmp}" = x"yes"; then
119
120
121
122
123
  AX_ELPA_OPENMP
  if test "$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"
124
125
fi

126
127
dnl check which MPI binray invokes a MPI job
if test x"$with_mpi" = x"yes"; then
128
  AC_CHECK_PROGS([MPI_BINARY], [srun mpiexec.hydra mpiexec mpirun poe runjob], [no])
129
  if test x"$MPI_BINARY" = x"no"; then
130
    AC_MSG_ERROR([Could not find either of the MPI binaries: mpiexec.hydra, mpiexec, mpirun, poe, runjob, srun])
131
132
133
  fi
fi

Andreas Marek's avatar
Andreas Marek committed
134
135
136
137
138
139
140
141
142
143
144
## 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
145
146
147



Andreas Marek's avatar
Andreas Marek committed
148
149
dnl variables needed for the tests

Andreas Marek's avatar
Andreas Marek committed
150
151
152
153
154
155
156
157
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
158
159
160
161
162
163

install_real_generic=yes
install_real_generic_simple=yes

install_complex_generic=yes
install_complex_generic_simple=yes
164

165
166
167
#want_avx=yes
#want_avx2=yes
#want_sse=yes
168
#want_avx512=yes
169

Andreas Marek's avatar
Andreas Marek committed
170
171
AC_LANG([C])

172
173
174
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)
175
176
177
178
179
180
181
182
183
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"${enable_redirect}" = x"yes"; then
184
185
  AC_DEFINE([HAVE_REDIRECT], [1], [Redirect stdout and stderr of test programs per MPI tasks to a file])
fi
186
187
188
189
AM_CONDITIONAL([HAVE_REDIRECT],[test x"$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
190
AC_ARG_ENABLE([timings],
Andreas Marek's avatar
Andreas Marek committed
191
		AS_HELP_STRING([--enable-timings],
192
			       [more detailed timing, default no.]),
193
	      [enable_timings=yes],
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
	      [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"])
213
214
215
216
217
    fi
    AC_DEFINE([HAVE_LIBPAPI], [1], [Use the PAPI library])
  fi
fi

218
AC_MSG_CHECKING(whether double-precision SSE assembly kernel can be compiled)
Andreas Marek's avatar
Andreas Marek committed
219

220
$CC -c $srcdir/src/elpa2_kernels/elpa2_kernels_asm_x86_64_double_precision.s -o test.o 2>/dev/null
221
if test "$?" == 0; then
222
223
224
225
226
227
228
229
230
231
232
233
  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}])

234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
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

254
dnl check whether on can compile with sse-gcc intrinsics
255
AC_MSG_CHECKING(whether we can compile SSE3 with gcc intrinsics in C)
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
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
271
272
273
274
  install_real_sse_block2=yes
  install_real_sse_block4=yes
  install_real_sse_block6=yes

275
  install_complex_sse_intrinsics=yes
276
277
  install_complex_sse_block1=yes
  install_complex_sse_block2=yes
278
else
279
  install_real_sse_intrinsics=no
280
281
282
283
  install_real_sse_block2=no
  install_real_sse_block4=no
  install_real_sse_block6=no

284
  install_complex_sse_intrinsics=no
285
286
  install_complex_sse_block1=no
  install_complex_sse_block2=no
287
fi
288

Andreas Marek's avatar
Andreas Marek committed
289
290
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
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
316

Andreas Marek's avatar
Andreas Marek committed
317
#if test "${can_compile_avx}" = "yes" ; then
318
#  AC_MSG_CHECKING([whether we can compile AVX gcc intrinsics in C++])
Andreas Marek's avatar
Andreas Marek committed
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
#  AC_LANG_PUSH([C++])
#  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
#   #include <x86intrin.h>
#   int main(int argc, char **argv){
#   double* q;
#   __m256d a1_1 = _mm256_load_pd(q);
#   return 0;
#   }
#   ])],
#   [can_compile_avx=yes],
#   [can_compile_avx=no]
#  )
#  AC_LANG_POP([C++])
#  AC_MSG_RESULT([${can_compile_avx}])
#  if test "${can_compile_avx}" = "no" ; then
#    AC_MSG_WARN([Cannot compile C++ with AVX: disabling AVX alltogether])
#  fi
#fi
337
338

dnl first pass: try with specified CFLAGS and CXXFLAGS
Andreas Marek's avatar
Andreas Marek committed
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364

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
365

366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
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
386

Andreas Marek's avatar
Andreas Marek committed
387
388
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
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
414
#if test "${can_compile_avx2}" = "yes" ; then
415
#  AC_MSG_CHECKING([whether we can compile AVX2 gcc intrinsics in C++])
Andreas Marek's avatar
Andreas Marek committed
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
#  AC_LANG_PUSH([C++])
#  AC_COMPILE_IFELSE([AC_LANG_SOURCE([
#   #include <x86intrin.h>
#   int main(int argc, char **argv){
#   double* q;
#   __m256d q1 = _mm256_load_pd(q);
#   __m256d y1 = _mm256_fmadd_pd(q1, q1, q1);
#   return 0;
#   }
#   ])],
#   [can_compile_avx2=yes],
#   [can_compile_avx2=no]
#  )
#  AC_LANG_POP([C++])
#  AC_MSG_RESULT([${can_compile_avx2}])
#  if test "${can_compile_avx2}" = "no" ; then
#    AC_MSG_WARN([Cannot compile C++ with AVX2!])
#  fi
#fi
435

Andreas Marek's avatar
Andreas Marek committed
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
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
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484

#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

485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
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
506

507
if test "${can_compile_avx}" = "yes" ; then
Andreas Marek's avatar
Andreas Marek committed
508
509
510
511
512
513
  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
514
515
516
517
518
519
520
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
521
522
fi

523
524
525
526
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
527

528
529
530
531
532
533
534
535
536
537
  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
538

539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
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


556
557
558
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])
559
fi
560
561
562
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])
563
fi
564

565
566
567
568
569
570
571
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])
572
fi
573
574
575
576
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
577
578
579
580
581
582
583
584

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
585
AC_MSG_RESULT([${with_avx_optimization}])
586
if test x"${with_avx_optimization}" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
587
588
589
 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
590

591
AC_LANG([Fortran])
592
593
594
AC_FC_FREEFORM
AC_FC_MODULE_FLAG
AC_FC_MODULE_OUTPUT_FLAG
595
AC_FC_LIBRARY_LDFLAGS
596
597
598
599

save_FCFLAGS=$FCFLAGS
save_LDFLAGS=$LDFLAGS

600
601
602
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])

603
604
FCFLAGS="$FCFLAGS $SCALAPACK_FCFLAGS"
LDFLAGS="$LDFLAGS $SCALAPACK_LDFLAGS"
605

606
dnl check whether fortran error_unit is defined
607
AC_MSG_CHECKING([whether Fortran module iso_fortran_env is available])
608
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
609
610
611
612
613
614
615
616
617
618
  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]
)
619
AC_MSG_RESULT([${can_use_iso_fortran_env}])
620

621
622
623
624
dnl check whether one can link with specified MKL (desired method)
AC_MSG_CHECKING([whether we can compile a Fortran program using MKL])


625
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
626
  program test_mkl
627
    use mkl_service
628
629
630
631
632
633
634
635
636
637
638
639
    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"$can_compile_with_mkl" = x"yes" ; then
  AC_MSG_CHECKING([whether we can link a Fortran program with MKL])
640
  AC_LINK_IFELSE([AC_LANG_SOURCE([
641
    program test_mkl
642
      use mkl_service
643
644
645
646
647
      character*198 :: string
      call mkl_get_version_string(string)
      write(*,'(a)') string
    end program
  ])],
Andreas Marek's avatar
Andreas Marek committed
648
649
    [have_mkl=yes],
    [have_mkl=no]
650
  )
Andreas Marek's avatar
Andreas Marek committed
651
  AC_MSG_RESULT([${have_mkl}])
652
653
654
fi

dnl if not mkl, check all the necessary individually
655
if test x"${have_mkl}" = x"yes" ; then
656
657
658
659
  WITH_MKL=1
else

  dnl first check blas
660
  AC_SEARCH_LIBS([dgemm],[openblas satlas blas],[have_blas=yes],[have_blas=no])
Andreas Marek's avatar
Andreas Marek committed
661
662
  AC_MSG_CHECKING([whether we can link a program with a blas lib])
  AC_MSG_RESULT([${have_blas}])
663

664
  if test x"${have_blas}" = x"no" ; then
665
    AC_MSG_ERROR([could not link with blas: specify path])
666
  fi
667
  dnl now lapack
Andreas Marek's avatar
Andreas Marek committed
668
669
670
  AC_SEARCH_LIBS([dlarrv],[lapack],[have_lapack=yes],[have_lapack=no])
  AC_MSG_CHECKING([whether we can link a program with a lapack lib])
  AC_MSG_RESULT([${have_lapack}])
671

672
  if test x"${have_lapack}" = x"no" ; then
673
    AC_MSG_ERROR([could not link with lapack: specify path])
674
  fi
Andreas Marek's avatar
Andreas Marek committed
675

676
  if test x"${with_mpi}" = x"yes"; then
677
    dnl test whether scalapack already contains blacs
678
    scalapack_libs="mpiscalapack scalapack scalapack-openmpi"
679
680
681
682
683
684
685
686
687
688
    old_LIBS="$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
689

690
691
    if test x"${blacs_in_scalapack}" = x"no"; then
      LIBS="${old_LIBS}"
692

693
694
      dnl Test for stand-alone blacs
      AC_SEARCH_LIBS([bi_f77_init],[mpiblacsF77init],[],[],[-lmpiblacs])
695
696
697
698
699
700

      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
701
      AC_SEARCH_LIBS([blacs_gridinit],[mpiblacs blacs],[have_blacs=yes],[have_blacs=no])
Andreas Marek's avatar
Andreas Marek committed
702

703
704
705
      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
706
    fi
707

708
    AC_SEARCH_LIBS([pdtran],[$scalapack_libs],[have_scalapack=yes],[have_scalapack=no])
Andreas Marek's avatar
Andreas Marek committed
709

710
711
712
    if test x"${have_scalapack}" = x"no" ; then
      AC_MSG_ERROR([could not link with scalapack: specify path])
    fi
713
714
715
716
  fi

  dnl check whether we can link alltogehter
  AC_MSG_CHECKING([whether we can link a Fortran program with all blacs/scalapack])
717
  AC_LINK_IFELSE([AC_LANG_SOURCE([
718
719
    program dgemm_test

720
721
      integer , parameter:: M = 4, N = 3, K = 2
      real :: A(M,K), B(K,N), C(M,N)
722

723
      call dgemm('N','N',M,N,K,1.0,A,M,B,K,0.0,C,M)
724
725
726
727
728
729
730
731

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

732
   if test x"${can_link_with_blacs_scalapack}" = x"yes" ; then
733
734
     WITH_BLACS=1
   else
735
   AC_MSG_ERROR([We can neither link with MKL or another Scalpack. Please specify SCALAPACK_LDFLAGS and SCALAPACK_FCFLAGS!])
736
737
738
   fi
fi

739
740
741
742
dnl important: reset them again!
FCFLAGS=$save_FCFLAGS
LDFLAGS=$save_LDFLAGS

Andreas Marek's avatar
Andreas Marek committed
743
744
745
746
dnl check for intrinsic fortran function of 2003 standard
AC_MSG_CHECKING([whether we can use the intrinsic Fortran function "get_environment_variable"])


747
AC_COMPILE_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
748
749
750
751
752
753
754
755
756
  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([${fortran_can_check_environment}])
757
758
759
if test x"${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
760

761
dnl check whether BAND_TO_FULL_BLOCKING is set
762
use_band_to_full_blocking=yes
763
764
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],
765
	                                            [build ELPA2 with blocking in band_to_full (default: enabled)])],
766
767
	      want_band_to_full_blocking="yes", want_to_full_blocking="no")
AC_MSG_RESULT([${want_band_to_full_blocking}])
768
769
770
771
772
773
774
775

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])
776
777
fi

778
779
dnl check whether GPU version is requested

780
#CUDA_INSTALL_PATH="/usr/local/cuda/"
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
#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
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
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

816
817
if test x"${want_gpu}" = x"yes" ; then
  AC_LANG_PUSH([C])
818
  CUDA_CFLAGS="$CUDA_CFLAGS -arch $cuda_compute_capability -O2 -I$CUDA_INSTALL_PATH/include"
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
  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
842
  can_compile_gpu=yes
843
844
fi

Andreas Marek's avatar
Andreas Marek committed
845
846
847
848
849
850
851
852
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])

853
AC_LINK_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
854
855
856
  program test_bgp
    complex*16 :: y3,q3,h2
    y3 = fxcpmadd(y3,q3,h2)
857

Andreas Marek's avatar
Andreas Marek committed
858
859
860
861
862
863
864
  end program
])],
  [can_compile_bgp=yes],
  [can_compile_bgp=no]
)
AC_MSG_RESULT([${can_compile_bgp}])

865
if test x"${can_compile_bgp}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
866
867
  install_real_bgp=yes
  install_complex_bgp=yes
Andreas Marek's avatar
Andreas Marek committed
868
869
870
else
  install_real_bgp=no
  install_complex_bgp=no
Andreas Marek's avatar
Andreas Marek committed
871
872
873
874
875
fi

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

876
AC_LINK_IFELSE([AC_LANG_SOURCE([
Andreas Marek's avatar
Andreas Marek committed
877
  program test_bgq
Andreas Marek's avatar
Andreas Marek committed
878
879
    VECTOR(REAL(8))::QPX_h2
    real*8         :: hh(10,2)
Andreas Marek's avatar
Andreas Marek committed
880
    QPX_h2 = VEC_SPLATS(hh(2,2))
881

Andreas Marek's avatar
Andreas Marek committed
882
883
884
885
886
887
888
  end program
])],
  [can_compile_bgq=yes],
  [can_compile_bgq=no]
)
AC_MSG_RESULT([${can_compile_bgq}])

889
if test x"${can_compile_bgq}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
890
891
  install_real_bgq=yes
  install_complex_bgq=yes
Andreas Marek's avatar
Andreas Marek committed
892
893
894
else
  install_real_bgq=no
  install_complex_bgq=no
Andreas Marek's avatar
Andreas Marek committed
895
896
fi

897
898
dnl Test possibility of 'use mpi', if requested
if test x"${with_mpi}" = x"yes" ; then
899
900
901
  AC_ARG_ENABLE([mpi-module],
                AS_HELP_STRING([--disable-mpi-module],
                               [Do not use the Fortran MPI module, get interfaces by 'include "mpif.h')]),
902
                [],
903
904
                [enable_mpi_module=yes])
  if test x"${enable_mpi_module}" = x"yes" ; then
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
    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
920
      AC_MSG_ERROR([Could not compile a Fortran program with an 'use mpi' statement. You can try again with --disable-mpi-module])
921
    fi
922
  fi
Andreas Marek's avatar
Andreas Marek committed
923
924
fi

925
926
927
928
dnl default value
use_specific_real_kernel=no
use_specific_complex_kernel=no

929
930
931
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
932

933
934
dnl last check whether user wants to compile only a specific kernel
dnl
935
m4_include([m4/ax_elpa_specific_kernels.m4])
Andreas Marek's avatar
Andreas Marek committed
936

937
dnl real kernels
938
dnl do not remove this variable it is needed in the macros
939

940
941
dnl generic kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-generic-kernel-only],[generic-kernel],[install_real_generic])
Andreas Marek's avatar
Andreas Marek committed
942

943
944
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
945

946
dnl sse kernel
947
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-sse-assembly-kernel-only],[sse-assembly-kernel],[install_real_sse_assembly])
Andreas Marek's avatar
Andreas Marek committed
948

949
950
dnl bgp kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-bgp-kernel-only],[bgp-kernel],[install_real_bgp])
Andreas Marek's avatar
Andreas Marek committed
951

952
953
dnl bgq kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-bgq-kernel-only],[bgq-kernel],[install_real_bgq])
Andreas Marek's avatar
Andreas Marek committed
954

955
956
dnl real-sse-block2 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-sse-block2-kernel-only],[real-sse-block2-kernel],[install_real_sse_block2])
957

958
dnl real-sse-block4 kernel
Andreas Marek's avatar
Andreas Marek committed
959
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-sse-block4-kernel-only],[real-sse-block4-kernel],[install_real_sse_block4])
960

961
962
dnl real-sse-block6 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-sse-block6-kernel-only],[real-sse-block6-kernel],[install_real_sse_block6])
963

964
965
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
966

967
dnl real-avx-block4 kernel
Andreas Marek's avatar
Andreas Marek committed
968
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx-block4-kernel-only],[real-avx-block4-kernel],[install_real_avx_block4])
969

970
971
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
972

973
974
dnl real-avx2-block2 kernel
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx2-block2-kernel-only],[real-avx2-block2-kernel],[install_real_avx2_block2])
975

976
dnl real-avx2-block4 kernel
Andreas Marek's avatar
Andreas Marek committed
977
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx2-block4-kernel-only],[real-avx2-block4-kernel],[install_real_avx2_block4])
978

979
980
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
981

982
983
984
985
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
986
DEFINE_OPTION_SPECIFIC_REAL_KERNEL([real-avx512-block4-kernel-only],[real-avx512-block4-kernel],[install_real_avx512_block4])
987
988
989
990
991

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
992
dnl complex kernels
993

994
995
dnl generic kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-generic-kernel-only],[generic-kernel],[install_complex_generic])
Andreas Marek's avatar
Andreas Marek committed
996

997
998
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
999

1000
dnl sse kernel
1001
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-sse-assembly-kernel-only],[sse-assembly-kernel],[install_complex_sse_assembly])
Andreas Marek's avatar
Andreas Marek committed
1002

1003
1004
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
1005

1006
1007
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
1008

1009
1010
dnl complex-sse-block1 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-sse-block1-kernel-only],[complex-sse-block1-kernel],[install_complex_sse_block1])
1011

1012
1013
dnl complex-sse-block2 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-sse-block2-kernel-only],[complex-sse-block2-kernel],[install_complex_sse_block2])
1014

1015
1016
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
1017

1018
1019
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
1020

1021
1022
dnl complex-avx2-block1 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-avx2-block1-kernel-only],[complex-avx2-block1-kernel],[install_complex_avx2_block1])
1023

1024
1025
dnl complex-avx2-block2 kernel
DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL([complex-avx2-block2-kernel-only],[complex-avx2-block2-kernel],[install_complex_avx2_block2])
1026

1027
1028
1029
1030
1031
1032
1033
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])


1034

1035
dnl set the conditionals according to the previous tests
1036

1037
if test x"${can_use_iso_fortran_env}" = x"yes" ; then
1038
1039
1040
 AC_DEFINE([HAVE_ISO_FORTRAN_ENV],[1],[can use module iso_fortran_env])
fi

1041
1042
1043
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])
1044
fi
1045

Andreas Marek's avatar
Andreas Marek committed
1046
AM_CONDITIONAL([WITH_REAL_GENERIC_KERNEL],[test x"$install_real_generic" = x"yes"])
1047
if test x"${install_real_generic}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1048
1049
1050
1051
 AC_DEFINE([WITH_REAL_GENERIC_KERNEL],[1],[can use real generic kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_GENERIC_KERNEL],[test x"$install_complex_generic" = x"yes"])
1052
if test x"${install_complex_generic}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1053
1054
1055
1056
 AC_DEFINE([WITH_COMPLEX_GENERIC_KERNEL],[1],[can use complex generic kernel])
fi

AM_CONDITIONAL([WITH_REAL_GENERIC_SIMPLE_KERNEL],[test x"$install_real_generic_simple" = x"yes"])
1057
if test x"${install_real_generic_simple}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1058
1059
1060
1061
 AC_DEFINE([WITH_REAL_GENERIC_SIMPLE_KERNEL],[1],[can use real generic-simple kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_GENERIC_SIMPLE_KERNEL],[test x"$install_complex_generic_simple" = x"yes"])
1062
if test x"${install_complex_generic_simple}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1063
1064
1065
 AC_DEFINE([WITH_COMPLEX_GENERIC_SIMPLE_KERNEL],[1],[can use complex generic-simple kernel])
fi

1066
1067
1068
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
1069
1070
fi

1071
1072
1073
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
1074
1075
fi

1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
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
1091
AM_CONDITIONAL([WITH_REAL_AVX_BLOCK2_KERNEL],[test x"$install_real_avx_block2" = x"yes"])
1092
if test x"${install_real_avx_block2}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1093
1094
1095
1096
 AC_DEFINE([WITH_REAL_AVX_BLOCK2_KERNEL],[1],[can use real_avx_block2 kernel])
fi

AM_CONDITIONAL([WITH_REAL_AVX_BLOCK4_KERNEL],[test x"$install_real_avx_block4" = x"yes"])
1097
if test x"${install_real_avx_block4}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1098
1099
1100
1101
 AC_DEFINE([WITH_REAL_AVX_BLOCK4_KERNEL],[1],[can use real_avx_block4 kernel])
fi

AM_CONDITIONAL([WITH_REAL_AVX_BLOCK6_KERNEL],[test x"$install_real_avx_block6" = x"yes"])
1102
if test x"${install_real_avx_block6}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1103
1104
1105
 AC_DEFINE([WITH_REAL_AVX_BLOCK6_KERNEL],[1],[can use real_avx_block6 kernel])
fi

1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
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

1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
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

1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
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
1146
AM_CONDITIONAL([WITH_COMPLEX_AVX_BLOCK1_KERNEL],[test x"$install_complex_avx_block1" = x"yes"])
1147
if test x"${install_complex_avx_block1}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1148
1149
1150
1151
 AC_DEFINE([WITH_COMPLEX_AVX_BLOCK1_KERNEL],[1],[can use complex_avx_block1 kernel])
fi

AM_CONDITIONAL([WITH_COMPLEX_AVX_BLOCK2_KERNEL],[test x"$install_complex_avx_block2" = x"yes"])
1152
if test x"${install_complex_avx_block2}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1153
1154
1155
 AC_DEFINE([WITH_COMPLEX_AVX_BLOCK2_KERNEL],[1],[can use complex_avx_block2 kernel])
fi

1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
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

1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
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
1176
AM_CONDITIONAL([WITH_REAL_BGP_KERNEL],[test x"$install_real_bgp" = x"yes"])
1177
if test x"${install_real_bgp}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1178
1179
1180
1181
 AC_DEFINE([WITH_REAL_BGP_KERNEL],[1],[can use real BGP kernel])
fi

AM_CONDITIONAL([WITH_REAL_BGQ_KERNEL],[test x"$install_real_bgq" = x"yes"])
1182
if test x"${install_real_bgq}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1183
1184
1185
 AC_DEFINE([WITH_REAL_BGQ_KERNEL],[1],[can use real BGQ kernel])
fi

1186
if test x"${use_specific_complex_kernel}" = x"no" ; then
1187
  AC_DEFINE([WITH_NO_SPECIFIC_COMPLEX_KERNEL],[1],[do not use only one specific complex kernel (set at compile time)])
1188
else
1189
  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
1190
1191
fi

1192
if test x"${use_specific_real_kernel}" = x"no" ; then
1193
  AC_DEFINE([WITH_NO_SPECIFIC_REAL_KERNEL],[1],[do not use only one specific real kernel (set at compile time)])
1194
1195
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
1196
1197
fi

1198
1199
LT_INIT

1200
1201
1202
1203
1204
1205
DX_PDF_FEATURE(OFF)
DX_PS_FEATURE(OFF)
DX_MAN_FEATURE(ON)
DX_HTML_FEATURE(ON)
DX_INIT_DOXYGEN([ELPA], [Doxyfile], [docs])

1206
USE_ASSUMED_SIZE=yes
Andreas Marek's avatar
Andreas Marek committed
1207
AC_MSG_CHECKING(whether assumed size Fortran arrays should be used)
1208
AC_ARG_ENABLE([assumed-size],
1209
1210
                AS_HELP_STRING([--disable-assumed-size],
			       [do NOT use assumed-size Fortran arrays]),
Andreas Marek's avatar
Andreas Marek committed
1211
1212
1213
1214
1215
			       [if test x"$enableval" = x"yes"; then
                                  USE_ASSUMED_SIZE=yes
				else
			          USE_ASSUMED_SIZE=no
				fi],
1216
              [USE_ASSUMED_SIZE=yes])
1217
1218
1219
AC_MSG_RESULT([${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
1220
  AC_DEFINE([USE_ASSUMED_SIZE],[1],[for performance reasons use assumed size Fortran arrays, even if not debuggable])
1221
1222
fi

1223
1224
1225
if test x"${want_single_precision}" = x"yes" ; then
  AC_DEFINE([WANT_SINGLE_PRECISION_REAL],[1],[build also single-precision for real calculation])
  AC_DEFINE([WANT_SINGLE_PRECISION_COMPLEX],[1],[build also single-precision for complex calculation])
1226
fi
1227
1228
AM_CONDITIONAL([WANT_SINGLE_PRECISION_REAL],[test x"$want_single_precision" = x"yes"])
AM_CONDITIONAL([WANT_SINGLE_PRECISION_COMPLEX],[test x"$want_single_precision" = x"yes"])
1229

1230
AC_SUBST([MPI_BINARY])
1231
1232
1233
1234
1235
AC_SUBST([WITH_MKL])
AC_SUBST([WITH_BLACS])
AC_SUBST([with_amd_bulldozer_kernel])
AC_SUBST([FC_MODINC])
AC_SUBST([FC_MODOUT])
1236
1237
1238
AC_SUBST([OPENMP_CFLAGS])
AC_SUBST([OPENMP_FCFLAGS])
AC_SUBST([OPENMP_LDFLAGS])
1239
AC_SUBST([DOXYGEN_OUTPUT_DIR], [docs])
1240

1241
1242
rm -rf modules/ private_modules/ .fortran_dependencies/
mkdir modules private_modules
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252

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

1253
1254
1255
1256
1257
1258
1259
1260
1261

# 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')
1262

1263
1264
if test x"${with_mpi}" = x"yes"; then
  if test x"${enable_openmp}" = x"yes"; then
1265
	SUFFIX="_openmp"
1266
  else
1267
	SUFFIX=""
1268
1269
1270
  fi
else
  if test x"${enable_openmp}" = x"yes"; then
Lorenz Huedepohl's avatar
Typo  
Lorenz Huedepohl committed
1271
	SUFFIX="_onenode_openmp"
1272
1273
1274
  else
	SUFFIX="_onenode"
  fi
1275
fi
1276

1277
1278
AC_SUBST([SUFFIX])
AC_SUBST([PKG_CONFIG_FILE],[elpa${SUFFIX}-${PACKAGE_VERSION}.pc])
1279

1280
1281
AC_CONFIG_FILES([
  Makefile
1282
  Doxyfile
1283
  ${PKG_CONFIG_FILE}:elpa.pc.in
1284
])
1285

1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
AC_MSG_CHECKING([if workaround for Intel's broken preprocessor is needed])
if test x"$FC" = x"mpiifort" ; then
	need_manual_cpp=yes
fi
if test x"$FC" = x"ifort" ; then
	need_manual_cpp=yes
fi
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

1300
AC_OUTPUT
1301

1302
if test "${can_compile_avx}" = "no" ; then
1303
#  if test x"${want_avx}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
1304
    AC_MSG_WARN([Could not compile AVX instructions])
1305
#  fi
1306
fi
1307
if test "${can_compile_avx2}" = "no" ; then
1308
#  if test x"${want_avx2}" = x"yes" ; then
1309
    AC_MSG_WARN([Could not compile AVX2 instructions])
Andreas Marek's avatar