configure.ac 43.4 KB
Newer Older
1
AC_PREREQ([2.69])
2
AC_INIT([elpa],[2014.06.000], elpa-library@rzg.mpg.de)
3
4
5
6
7
8
9
AC_CONFIG_SRCDIR([src/elpa1.F90])

AM_INIT_AUTOMAKE([foreign -Wall subdir-objects])
AC_CONFIG_MACRO_DIR([m4])
AC_CONFIG_HEADERS([config.h])
AM_SILENT_RULES([yes])

10
11
12
AC_SUBST([ELPA_LIB_VERSION], [2014.06.000])
# this is the version of the API, should be changed in the major revision
# if and only if the actual API changes
13
14
# see http://www.gnu.org/software/libtool/manual/html_node/Updating-version-info.html
AC_SUBST([ELPA_SO_VERSION], [1:0:0])
15

16
17
18
19
20
21
AX_CHECK_GNU_MAKE()
if test x$_cv_gnu_make_command = x ; then
        AC_MSG_ERROR([Need GNU Make])
fi

AC_CHECK_PROG(CPP_FOUND,cpp,yes,no)
22
if test x"${CPP_FOUND}" = xno; then
23
24
25
26
27
28
29
30
31
32
33
34
35
36
  AC_MSG_ERROR([no cpp found])
fi

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

AC_PROG_INSTALL
AM_PROG_CC_C_O
AM_PROG_AR
AM_PROG_AS
AC_PROG_CXX


Andreas Marek's avatar
Andreas Marek committed
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
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
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
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
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
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
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
dnl variables needed for the tests

dnl these test will cause an abort of configure if not
dnl successful. However, if MKL is found then the blas, blacs,
dnl lapack, and scalapack test can fail and vice versa
have_blas=no
have_blacs=no
have_mkl=no
have_mpi=no
have_lapack=no
have_scalapack=no
	
dnl these tests will decide which kernels can be build
dnl the usual case is all except the BlueGene (bg) kernels

can_compile_sse=no
can_compile_avx=no
can_compile_bgp=no
can_compile_bqq=no

fortran_can_check_environment=no
use_specific_real_kernel=no
use_specific_complex_kernel=no

install_real_generic=yes
install_real_generic_simple=yes
install_real_sse=no
install_real_bgp=no
install_real_bgq=no
install_real_avx_block2=no
install_real_avx_block4=no
install_real_avx_block6=no

install_complex_generic=yes
install_complex_generic_simple=yes
install_complex_sse=no
install_complex_bgp=no
install_complex_bgq=no
install_complex_avx_block1=no
install_complex_avx_block2=no
AC_MSG_CHECKING(whether SSE assembler kernel can be compiled)

echo "  .globl double_hh_trafo_
        .globl single_hh_trafo_complex_
        .text

         .macro hh_trafo_real nrows

        movq      %rdi, %r10   # Copy address of q
        movq      %rsi, %r11   # Copy address of hh

        movaps      (%r10), %xmm6       # y1 = q(1,1)
        movaps    16(%r10), %xmm7       # y2 = q(2,1)
        .if \nrows>=8
        movaps    32(%r10), %xmm8
        movaps    48(%r10), %xmm9
        .if \nrows==12
        movaps    64(%r10), %xmm10
        movaps    80(%r10), %xmm11
        .endif
        .endif

        addq      %r8, %r10             # %r10 => q(.,2)
        movddup   8(%r11,%r9), %xmm15   #  hh(2,2)

        .macro mac_pre_loop1 qoff, X, Y
        movaps    \qoff(%r10), \X       # xn = q(n,2)
        movaps    \X, %xmm12
        mulpd     %xmm15, %xmm12
        addpd     %xmm12, \Y            # yn = yn + xn*h(2,2)
        .endm

        mac_pre_loop1  0, %xmm0, %xmm6
        mac_pre_loop1 16, %xmm1, %xmm7
        .if \nrows>=8
        mac_pre_loop1 32, %xmm2, %xmm8
        mac_pre_loop1 48, %xmm3, %xmm9
        .if \nrows==12
        mac_pre_loop1 64, %xmm4, %xmm10
        mac_pre_loop1 80, %xmm5, %xmm11
        .endif
        .endif
        .purgem   mac_pre_loop1

        addq      \$8, %r11
        .align 16
1:
        cmpq %rax, %r11                 # Jump out of the loop if %r11 >= %rax
        jge       2f

        addq      %r8, %r10             # %r10 => q(.,i)

        movddup   (%r11), %xmm14        # hh(i-1,1)
        movddup   8(%r11,%r9), %xmm15   # hh(i,2)

        .macro mac_loop1 qoff, X, Y
        movaps    \qoff(%r10), %xmm13   # q(.,i)
        movaps    %xmm13, %xmm12
        mulpd     %xmm14, %xmm13
        addpd     %xmm13, \X            # xn = xn + q(.,i)*h1
        mulpd     %xmm15, %xmm12
        addpd     %xmm12, \Y            # yn = yn + q(.,i)*h2
        .endm

        mac_loop1  0, %xmm0, %xmm6
        mac_loop1 16, %xmm1, %xmm7
        .if \nrows>=8
        mac_loop1 32, %xmm2, %xmm8
        mac_loop1 48, %xmm3, %xmm9
        .if \nrows==12
        mac_loop1 64, %xmm4, %xmm10
        mac_loop1 80, %xmm5, %xmm11
        .endif
        .endif
        .purgem   mac_loop1

        addq      \$8, %r11
        jmp       1b
2:

        addq      %r8, %r10             # %r10 => q(.,nb+1)
        movddup   (%r11), %xmm14

        .macro mac_post_loop1 qoff, X
        movaps    \qoff(%r10), %xmm13   # q(.,nb+1)
        mulpd     %xmm14, %xmm13
        addpd     %xmm13, \X
        .endm

        mac_post_loop1  0, %xmm0
        mac_post_loop1 16, %xmm1
        .if \nrows>=8
        mac_post_loop1 32, %xmm2
        mac_post_loop1 48, %xmm3
        .if \nrows==12
        mac_post_loop1 64, %xmm4
        mac_post_loop1 80, %xmm5
        .endif
        .endif
        .purgem   mac_post_loop1

        movq      %rsi, %r11    # restore %r11 (hh(1,1))

        movddup (%r11), %xmm12 # hh(1,1)
        xorps   %xmm14, %xmm14
        subpd   %xmm12, %xmm14 # %xmm14 = -hh(1,1)

        mulpd   %xmm14, %xmm0
        mulpd   %xmm14, %xmm1
        .if \nrows>=8
        mulpd   %xmm14, %xmm2
        mulpd   %xmm14, %xmm3
        .if \nrows==12
        mulpd   %xmm14, %xmm4
        mulpd   %xmm14, %xmm5
        .endif
        .endif

        movddup (%r11,%r9), %xmm12  # hh(1,2)
        xorps   %xmm15, %xmm15
        subpd   %xmm12, %xmm15 # %xmm15 = -hh(1,2) = h1
        movaps  %xmm15, %xmm14
        movddup (%rsp), %xmm12 # Get s from top of stack
        mulpd   %xmm12, %xmm14 # %xmm14 = h2

        .macro mac_xform_y X, Y
        mulpd   %xmm15, \Y  # y1 = y1*h1
        movaps  \X, %xmm12
        mulpd   %xmm14, %xmm12
        addpd   %xmm12, \Y
        .endm

        mac_xform_y %xmm0, %xmm6
        mac_xform_y %xmm1, %xmm7
        .if \nrows>=8
        mac_xform_y %xmm2, %xmm8
        mac_xform_y %xmm3, %xmm9
        .if \nrows==12
        mac_xform_y %xmm4, %xmm10
        mac_xform_y %xmm5, %xmm11
        .endif
        .endif
        .purgem   mac_xform_y

        movq   %rdi, %r10   # restore original Q

        .macro mac_pre_loop2_1 qoff, Y
        movaps    \qoff(%r10), %xmm13   # q(.,1)
        addpd     \Y, %xmm13
        movaps    %xmm13, \qoff(%r10)
        .endm

        mac_pre_loop2_1  0, %xmm6
        mac_pre_loop2_1 16, %xmm7
        .if \nrows>=8
        mac_pre_loop2_1 32, %xmm8
        mac_pre_loop2_1 48, %xmm9
        .if \nrows==12
        mac_pre_loop2_1 64, %xmm10
        mac_pre_loop2_1 80, %xmm11
        .endif
        .endif
        .purgem   mac_pre_loop2_1

        addq      %r8, %r10             # %r10 => q(.,2)

        movddup   8(%r11,%r9), %xmm15   # hh(2,2)

        .macro mac_pre_loop2_2 qoff, X, Y
        movaps    \X, %xmm13
        movaps    \Y, %xmm12
        mulpd     %xmm15, %xmm12
        addpd     %xmm12, %xmm13
        addpd     \qoff(%r10), %xmm13
        movaps    %xmm13, \qoff(%r10)
        .endm

        mac_pre_loop2_2  0, %xmm0, %xmm6
        mac_pre_loop2_2 16, %xmm1, %xmm7
        .if \nrows>=8
        mac_pre_loop2_2 32, %xmm2, %xmm8
        mac_pre_loop2_2 48, %xmm3, %xmm9
        .if \nrows==12
        mac_pre_loop2_2 64, %xmm4, %xmm10
        mac_pre_loop2_2 80, %xmm5, %xmm11
        .endif
        .endif
        .purgem   mac_pre_loop2_2

        addq      \$8, %r11
        .align 16
1:
        cmpq %rax, %r11                 # Jump out of the loop if %r11 >= %rax
        jge       2f

        addq      %r8, %r10             # %r10 => q(.,i)

        movddup   (%r11), %xmm14        # hh(i-1,1)
        movddup   8(%r11,%r9), %xmm15   # hh(i,2)

        .macro mac_loop2 qoff, X, Y
        movaps    \X, %xmm13
        mulpd     %xmm14, %xmm13
        movaps    \Y, %xmm12
        mulpd     %xmm15, %xmm12
        addpd     %xmm12, %xmm13
        addpd     \qoff(%r10), %xmm13
        movaps    %xmm13, \qoff(%r10)
        .endm

        mac_loop2  0, %xmm0, %xmm6
        mac_loop2 16, %xmm1, %xmm7
        .if \nrows>=8
        mac_loop2 32, %xmm2, %xmm8
        mac_loop2 48, %xmm3, %xmm9
        .if \nrows==12
        mac_loop2 64, %xmm4, %xmm10
        mac_loop2 80, %xmm5, %xmm11
        .endif
        .endif
        .purgem   mac_loop2

        addq      \$8, %r11
        jmp       1b
2:

        addq      %r8, %r10             # %r10 => q(.,nb+1)
        movddup   (%r11), %xmm14

        .macro mac_post_loop2 qoff, X
        movaps    \qoff(%r10), %xmm13   # q(.,nb+1)
        mulpd     %xmm14, \X
        addpd     \X, %xmm13
        movaps    %xmm13, \qoff(%r10)
        .endm

        mac_post_loop2  0, %xmm0
        mac_post_loop2 16, %xmm1
        .if \nrows>=8
        mac_post_loop2 32, %xmm2
        mac_post_loop2 48, %xmm3
        .if \nrows==12
        mac_post_loop2 64, %xmm4
        mac_post_loop2 80, %xmm5
        .endif
        .endif
        .purgem   mac_post_loop2

        .endm

        .align    16,0x90
double_hh_trafo_:

        movslq    (%rdx), %rdx # nb
        movslq    (%rcx), %rcx # nq
        movslq    (%r8),  %r8  # ldq
        movslq    (%r9),  %r9  # ldh

        # Get ldq in bytes
        addq      %r8, %r8
        addq      %r8, %r8
        addq      %r8, %r8 # 8*ldq, i.e. ldq in bytes

        # Get ldh in bytes
        addq      %r9, %r9
        addq      %r9, %r9
        addq      %r9, %r9 # 8*ldh, i.e. ldh in bytes

        movq %rdx, %rax
        addq %rax, %rax
        addq %rax, %rax
        addq %rax, %rax
        addq %rsi, %rax
        subq \$8, %rax

        subq \$8, %rsp


        movq      %rsi, %r11   # Copy address of hh

        movsd     8(%r11,%r9), %xmm0 #  hh(2,2)
        addq      \$8, %r11
1:
        cmpq %rax, %r11
        jge       2f
        movsd   (%r11), %xmm14       # hh(i-1,1)
        movsd   8(%r11,%r9), %xmm15  # hh(i,2)
        mulsd   %xmm14, %xmm15
        addsd   %xmm15, %xmm0
        addq      \$8, %r11
        jmp       1b
2:
        movsd   %xmm0, (%rsp)   # put s on top of stack
#-----------------------------------------------------------

rloop_s:
        cmpq      \$8, %rcx   # if %rcx <= 8 jump out of loop
        jle       rloop_e
        hh_trafo_real 12 # transform 12 rows
        addq      \$96, %rdi  # increment q start adress by 96 bytes (6 rows)
        subq      \$12, %rcx  # decrement nq
        jmp       rloop_s
rloop_e:

        cmpq      \$4, %rcx   # if %rcx <= 4 jump to test_2
        jle       test_4
        hh_trafo_real 8 # transform 8 rows
        jmp       return1

test_4:
        cmpq      \$0, %rcx   # if %rcx <= 0 jump to return
        jle       return1
        hh_trafo_real 4 # transform 4 rows

return1:
        addq      \$8, %rsp   # reset stack pointer
        ret

        .align    16,0x90

#-------------------------------------------------------------------------------
#-------------------------------------------------------------------------------

        .macro hh_trafo_complex nrows


        movq      %rdi, %r10   # Copy address of q
        movq      %rsi, %r11   # Copy address of hh

        # set %rax to the address of hh at the end of the loops,
        # i.e. if %rdx >= %rax we must jump out of the loop.
        # please note: %rax = 16*%rdx + %rsi
        movq %rdx, %rax
        addq %rax, %rax
        addq %rax, %rax
        addq %rax, %rax
        addq %rax, %rax
        addq %rsi, %rax

#   x1 = q(1,1); y1 = 0
#   x2 = q(2,1); y2 = 0
#   ...

        movaps      (%r10), %xmm0
        movaps    16(%r10), %xmm1
        xorps     %xmm6, %xmm6
        xorps     %xmm7, %xmm7
        .if \nrows>=4
        movaps    32(%r10), %xmm2
        movaps    48(%r10), %xmm3
        xorps     %xmm8, %xmm8
        xorps     %xmm9, %xmm9
        .if \nrows==6
        movaps    64(%r10), %xmm4
        movaps    80(%r10), %xmm5
        xorps     %xmm10, %xmm10
        xorps     %xmm11, %xmm11
        .endif
        .endif


        addq      \$16, %r11  # %r11 => hh(2)
        .align 16
1:
        cmpq      %rax, %r11      # Jump out of the loop if %r11 >= %rax
        jge 2f

        addq      %r8, %r10       # %r10 => q(.,i)

        movddup    (%r11), %xmm14 # real(hh(i))
        movddup   8(%r11), %xmm15 # imag(hh(i))

        .macro mac_loop1 qoff, X, Y
        movaps    \qoff(%r10), %xmm13     # q(.,i)
        movaps    %xmm13, %xmm12
        mulpd     %xmm14, %xmm13          # q(.,i)*real(hh(i))
        addpd     %xmm13, \X              # x1 = x1 + q(.,i)*real(hh(i))
        mulpd     %xmm15, %xmm12          # q(.,i)*imag(hh(i))
        addsubpd  %xmm12, \Y              # y1 = y1 -/+ q(.,i)*imag(hh(i))
        .endm

        mac_loop1   0, %xmm0, %xmm6
        mac_loop1  16, %xmm1, %xmm7
        .if \nrows>=4
        mac_loop1  32, %xmm2, %xmm8
        mac_loop1  48, %xmm3, %xmm9
        .if \nrows==6
        mac_loop1  64, %xmm4, %xmm10
        mac_loop1  80, %xmm5, %xmm11
        .endif
        .endif

        .purgem   mac_loop1

        addq      \$16, %r11                # %r11 => hh(i+1)
        jmp       1b
2:

        # Now the content of the yn has to be swapped and added to xn
        .macro mac_post_loop_1 X, Y
        shufpd \$1, \Y, \Y
        addpd  \Y, \X
        .endm

        mac_post_loop_1  %xmm0, %xmm6
        mac_post_loop_1  %xmm1, %xmm7
        .if \nrows>=4
        mac_post_loop_1  %xmm2, %xmm8
        mac_post_loop_1  %xmm3, %xmm9
        .if \nrows==6
        mac_post_loop_1  %xmm4, %xmm10
        mac_post_loop_1  %xmm5, %xmm11
        .endif
        .endif
        .purgem   mac_post_loop_1

#   tau1 = hh(1)
#
#   h1 = -tau1
#   x1 = x1*h1; y1 = x1 with halfes exchanged
#   x2 = x2*h1; y2 = x2 with halfes exchanged
#   ...

        movq      %rsi, %r11      # restore address of hh

        xorps     %xmm14, %xmm14
        movddup    (%r11), %xmm12 # real(hh(1))
        subpd     %xmm12, %xmm14  #-real(hh(1))
        xorps     %xmm15, %xmm15
        movddup   8(%r11), %xmm12 # imag(hh(1))
        subpd     %xmm12, %xmm15  #-imag(hh(1))

        .macro mac_xform X, Y
        movaps    \X, %xmm12
        shufpd    \$1, \X, %xmm12
        mulpd     %xmm15, %xmm12
        mulpd     %xmm14, \X
        addsubpd  %xmm12, \X
        movaps    \X, \Y          # copy to y
        shufpd    \$1, \X, \Y      # exchange halfes
        .endm

        mac_xform %xmm0, %xmm6
        mac_xform %xmm1, %xmm7
        .if \nrows>=4
        mac_xform %xmm2, %xmm8
        mac_xform %xmm3, %xmm9
        .if \nrows==6
        mac_xform %xmm4, %xmm10
        mac_xform %xmm5, %xmm11
        .endif
        .endif
        .purgem mac_xform

#   q(1,1) = q(1,1) + x1
#   q(2,1) = q(2,1) + x2
#   ...

        movq      %rdi, %r10      # restore address of q
        .macro mac_pre_loop2 qoff, X
        movaps    \qoff(%r10), %xmm13     # q(.,1)
        addpd     \X, %xmm13
        movaps    %xmm13, \qoff(%r10)
        .endm

        mac_pre_loop2   0, %xmm0
        mac_pre_loop2  16, %xmm1
        .if \nrows>=4
        mac_pre_loop2  32, %xmm2
        mac_pre_loop2  48, %xmm3
        .if \nrows==6
        mac_pre_loop2  64, %xmm4
        mac_pre_loop2  80, %xmm5
        .endif
        .endif
        .purgem mac_pre_loop2

#   do i=2,nb
#      h1 = hh(i)
#      q(1,i) = q(1,i) + x1*h1
#      q(2,i) = q(2,i) + x2*h1
#      ...
#   enddo

        addq      \$16, %r11
        .align 16
1:
        cmpq      %rax, %r11      # Jump out of the loop if %r11 >= %rax
        jge 2f

        addq      %r8, %r10       # %r10 => q(.,i)

        movddup    (%r11), %xmm14 # real(hh(i))
        movddup   8(%r11), %xmm15 # imag(hh(i))

        .macro mac_loop2 qoff, X, Y
        movaps    \X, %xmm13
        mulpd     %xmm14, %xmm13
        movaps    \Y, %xmm12
        mulpd     %xmm15, %xmm12
        addsubpd  %xmm12, %xmm13
        addpd     \qoff(%r10), %xmm13
        movaps    %xmm13, \qoff(%r10)
        .endm

        mac_loop2   0, %xmm0, %xmm6
        mac_loop2  16, %xmm1, %xmm7
        .if \nrows>=4
        mac_loop2  32, %xmm2, %xmm8
        mac_loop2  48, %xmm3, %xmm9
        .if \nrows==6
        mac_loop2  64, %xmm4, %xmm10
        mac_loop2  80, %xmm5, %xmm11
        .endif
        .endif
        .purgem   mac_loop2

        addq      \$16, %r11
        jmp       1b
2:
        .endm

        .align    16,0x90
single_hh_trafo_complex_:

        # Get integer parameters into corresponding registers

        movslq    (%rdx), %rdx # nb
        movslq    (%rcx), %rcx # nq
        movslq    (%r8),  %r8  # ldq

        # Get ldq in bytes
        addq      %r8, %r8
        addq      %r8, %r8
        addq      %r8, %r8
        addq      %r8, %r8 # 16*ldq, i.e. ldq in bytes

cloop_s:
        cmpq      \$4, %rcx   # if %rcx <= 4 jump out of loop
        jle       cloop_e
        hh_trafo_complex 6 # transform 6 rows
        addq      \$96, %rdi  # increment q start adress by 96 bytes (6 rows)
        subq      \$6,  %rcx  # decrement nq
        jmp       cloop_s
cloop_e:

        cmpq      \$2, %rcx   # if %rcx <= 2 jump to test_2
        jle       test_2
        hh_trafo_complex 4 # transform 4 rows
        jmp       return2

test_2:
        cmpq      \$0, %rcx   # if %rcx <= 0 jump to return
        jle       return2
        hh_trafo_complex 2 # transform 2 rows

return2:
        ret

        .align    16,0x90" > ./test.s

 $CC -c ./test.s
 if test "$?" == 0; then
  can_compile_sse=yes
  install_real_sse=yes
  install_complex_sse=yes

 else
  can_compile_sse=no
  install_real_sse=no
  install_complex_sse=no
 fi
649

Andreas Marek's avatar
Andreas Marek committed
650
651
 rm -f ./test.s ./test.o
AC_MSG_RESULT([${can_compile_sse}])
652

Andreas Marek's avatar
Andreas Marek committed
653
dnl check whether one can compile with avx - gcc intrinsics
654
655
656
657

 AC_MSG_CHECKING([whether we can compile a gcc intrinsic AVX program])

 dnl first pass: try with specified CFLAGS and CXXFLAGS
Andreas Marek's avatar
Andreas Marek committed
658
659
 AC_COMPILE_IFELSE([AC_LANG_SOURCE([     
  #include <x86intrin.h> 
660
661
662
663
664
  void main(){
  double* q;
  __m256d a1_1 = _mm256_load_pd(q);
  }
  ])],
Andreas Marek's avatar
Andreas Marek committed
665
666
  [can_compile_avx=yes],
  [can_compile_avx=no]
667
 )
Andreas Marek's avatar
Andreas Marek committed
668
dnl first test failed: try again after updating  CFLAGS and CXXFLAGS with -mavx
669
 if test x"${can_compile_avx}" = x"no"; then
670
671
 CFLAGS="$CFLAGS -mavx"
 CXXFLAGS="$CXXFLAGS -mavx"
Andreas Marek's avatar
Andreas Marek committed
672
673
 AC_COMPILE_IFELSE([AC_LANG_SOURCE([     
  #include <x86intrin.h> 
674
675
676
677
678
  void main(){
  double* q;
  __m256d a1_1 = _mm256_load_pd(q);
  }
  ])],
Andreas Marek's avatar
Andreas Marek committed
679
680
  [can_compile_avx=yes],
  [can_compile_avx=no]
681
682
 )
 fi
Andreas Marek's avatar
Andreas Marek committed
683
AC_MSG_RESULT([${can_compile_avx}])
684

685
if test x"${can_compile_avx}" = x"yes"; then
Andreas Marek's avatar
Andreas Marek committed
686
687
688
689
690
691
  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
692
693
694
695
696
697
698
699
700
701
702

fi


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
703
AC_MSG_RESULT([${with_avx_optimization}])
704
if test x"${with_avx_optimization}" = xyes; then
Andreas Marek's avatar
Andreas Marek committed
705
706
707
 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
708

Andreas Marek's avatar
Andreas Marek committed
709
AC_LANG(Fortran)
710
711
712
713
m4_include([m4/ax_prog_fc_mpi.m4])
dnl check whether an mpi compiler is available;
dnl if not abort since it is mandatory
AX_PROG_FC_MPI([],[have_mpi=yes],[have_mpi=no
714
		 if test x"${have_mpi}" = xno; then
715
716
717
718
719
720
721
722
723
724
725
726
727
728
  		  AC_MSG_ERROR([no mpi found])
		  fi])

AC_FC_FREEFORM
AC_FC_MODULE_FLAG
AC_FC_MODULE_OUTPUT_FLAG

AC_MSG_CHECKING(whether OpenMP usage is specified)
AC_ARG_WITH([openmp],
		AS_HELP_STRING([--with-openmp],
			       [use OpenMP threading, default no.]),
	      [with_openmp=yes],
	      [with_openmp=no])
  AC_MSG_RESULT([${with_openmp}])
729
  if test x"${enable_openmp}" = xyes; then
730
731
732
733
734
     with_openmp=yes
     AC_MSG_CHECKING(whether --enable-openmp is specified)
     AC_MSG_RESULT([${enable_openmp}])
  fi
  AM_CONDITIONAL([WITH_OPENMP],[test x"$with_openmp" = x"yes"])
735
  if test x"${with_openmp}" = xyes; then
736
737
738
739
740
741
742
743
744
745
746
747
	AC_DEFINE([WITH_OPENMP], [1], [use OpenMP threading])
        AX_ELPA_OPENMP
  fi

FCFLAGS="$FCFLAGS $OPENMP_FCFLAGS $OPENMP_FFFLAGS"
#LDFLAGS="$LDFLAGS $OPENMP_FCFLAGS $OPENMP_FFFLAGS"

save_FCFLAGS=$FCFLAGS
save_LDFLAGS=$LDFLAGS

FCFLAGS="$FCFLAGS $BLACS_FCFLAGS"
LDFLAGS="$LDFLAGS $BLACS_LDFLAGS"
748
749
750
751
752
753
754
755
756
757
758
759
760
761

dnl check wheter fortran error_unit is defined
AC_MSG_CHECKING([whether Fortran module iso_fortran_env is available])
AC_COMPILE_IFELSE([AC_LANG_SOURCE([     
  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]
)
762
AC_MSG_RESULT([${can_use_iso_fortran_env}])
763

764
765
766
767
dnl check whether one can link with specified MKL (desired method)
AC_MSG_CHECKING([whether we can compile a Fortran program using MKL])


Andreas Marek's avatar
Andreas Marek committed
768
AC_COMPILE_IFELSE([AC_LANG_SOURCE([     
769
  program test_mkl
Andreas Marek's avatar
Andreas Marek committed
770
    use mkl_service	
771
772
773
774
775
776
777
778
779
780
781
782
    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"$can_compile_with_mkl" = x"yes" ; then
  AC_MSG_CHECKING([whether we can link a Fortran program with MKL])
Andreas Marek's avatar
Andreas Marek committed
783
  AC_LINK_IFELSE([AC_LANG_SOURCE([     
784
    program test_mkl
Andreas Marek's avatar
Andreas Marek committed
785
      use mkl_service	
786
787
788
789
790
      character*198 :: string
      call mkl_get_version_string(string)
      write(*,'(a)') string
    end program
  ])],
Andreas Marek's avatar
Andreas Marek committed
791
792
    [have_mkl=yes],
    [have_mkl=no]
793
  )
Andreas Marek's avatar
Andreas Marek committed
794
  AC_MSG_RESULT([${have_mkl}])
795
796
797
fi

dnl if not mkl, check all the necessary individually
798
if test x"${have_mkl}" = x"yes" ; then
799
800
801
802
  WITH_MKL=1
else

  dnl first check blas
Andreas Marek's avatar
Andreas Marek committed
803
804
805
  AC_SEARCH_LIBS([dgemm],[blas],[have_blas=yes],[have_blas=no])
  AC_MSG_CHECKING([whether we can link a program with a blas lib])
  AC_MSG_RESULT([${have_blas}])
806

807
  if test x"${have_blas}" = x"no" ; then
Andreas Marek's avatar
Andreas Marek committed
808
809
    AC_MSG_ERROR([could not link with blas: specify path]) 
  fi	
810
  dnl now lapack
Andreas Marek's avatar
Andreas Marek committed
811
812
813
  AC_SEARCH_LIBS([dlarrv],[lapack],[have_lapack=yes],[have_lapack=no])
  AC_MSG_CHECKING([whether we can link a program with a lapack lib])
  AC_MSG_RESULT([${have_lapack}])
814

815
  if test x"${have_lapack}" = x"no" ; then
Andreas Marek's avatar
Andreas Marek committed
816
    AC_MSG_ERROR([could not link with lapack: specify path]) 
817
  fi
Andreas Marek's avatar
Andreas Marek committed
818
	
819
  dnl now blacs
Andreas Marek's avatar
Andreas Marek committed
820
821
822
823
  AC_SEARCH_LIBS([blacs_gridinit],[mpiblacs],[have_blacs=yes],[have_blacs=no])
  AC_MSG_CHECKING([whether we can link a program with a blacs lib])
  AC_MSG_RESULT([${have_blacs}])

824

Andreas Marek's avatar
Andreas Marek committed
825

826
  if test x"${have_blacs}" = x"no" ; then
Andreas Marek's avatar
Andreas Marek committed
827
828
829
    AC_MSG_ERROR([could not link with blacs: specify path]) 
  fi
	
830
  dnl now scalapack
Andreas Marek's avatar
Andreas Marek committed
831
832
833
834
  AC_SEARCH_LIBS([pdtran],[mpiscalapack],[have_scalapack=yes],[have_scalapack=no])
  AC_MSG_CHECKING([whether we can link a program with a scalapack lib])
  AC_MSG_RESULT([${have_scalapack}])

835
  if test x"${have_scalapack}" = x"no" ; then
Andreas Marek's avatar
Andreas Marek committed
836
    AC_MSG_ERROR([could not link with scalapack: specify path]) 
837
838
839
840
841
  fi

  dnl check whether we can link alltogehter

  AC_MSG_CHECKING([whether we can link a Fortran program with all blacs/scalapack])
Andreas Marek's avatar
Andreas Marek committed
842
  AC_LINK_IFELSE([AC_LANG_SOURCE([     
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
    program dgemm_test

      integer , parameter:: n_cols=3,l_cols=3
      real :: hvm(n_cols,l_cols)

      call dgemm('T','N',n_cols,n_cols,l_cols,1.,hvm,ubound(hvm,1), &
      hvm(1,n_cols+1),ubound(hvm,1),0.,hvm,ubound(hvm,1))


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

859
   if test x"${can_link_with_blacs_scalapack}" = x"yes" ; then
860
861
862
863
864
865
     WITH_BLACS=1
   else
   AC_MSG_ERROR([We can neither link with MKL or another Scalpack. Please specify BLACS_LDFLAGS and BLACS_FCFLAGS!])
   fi
fi

Andreas Marek's avatar
Andreas Marek committed
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
dnl check for intrinsic fortran function of 2003 standard
AC_MSG_CHECKING([whether we can use the intrinsic Fortran function "get_environment_variable"])


AC_COMPILE_IFELSE([AC_LANG_SOURCE([     
  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}])


883
884
885
886
887
dnl important: reset them again!
FCFLAGS=$save_FCFLAGS
LDFLAGS=$save_LDFLAGS


Andreas Marek's avatar
Andreas Marek committed
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
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])


AC_LINK_IFELSE([AC_LANG_SOURCE([     
  program test_bgp
    complex*16 :: y3,q3,h2
    y3 = fxcpmadd(y3,q3,h2)
  
  end program
])],
  [can_compile_bgp=yes],
  [can_compile_bgp=no]
)
AC_MSG_RESULT([${can_compile_bgp}])

909
if test x"${can_compile_bgp}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
  install_real_bgp=yes
  install_complex_bgp=yes
fi

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

AC_LINK_IFELSE([AC_LANG_SOURCE([     
  program test_bgq
    VECTOR(REAL(8))::QPX_x1
    real*8 :: hh(10,2))
    QPX_h2 = VEC_SPLATS(hh(2,2))
  
  end program
])],
  [can_compile_bgq=yes],
  [can_compile_bgq=no]
)
AC_MSG_RESULT([${can_compile_bgq}])

930
if test x"${can_compile_bgq}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
931
932
933
934
935
936
  install_real_bgq=yes
  install_complex_bgq=yes
fi


dnl environment variable setting of kernel
937
if test x"${fortran_can_check_environment}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
938
939
940
941
942
943
944
945
 AC_DEFINE([HAVE_ENVIRONMENT_CHECKING],[1],[Fortran can querry environment variables])
fi

dnl last check whether user wants to compile ony a specific kernel
dnl
dnl real kernels
dnl
dnl generic kernel
946
947
AC_ARG_WITH([-only-real-generic-kernel],
 [AS_HELP_STRING([--with-real-generic-kernel],[only compile generic-kernel for real case])],[],[with_real_generic_kernel=no])
Andreas Marek's avatar
Andreas Marek committed
948

949
950
951
if test x"${with_real_generic_kernel}" = x"yes" ; then
 if test x"${use_specific_real_kernel}" = x"no" ; then
#  if test x"${can_compile_sse}" = x"yes" ; then
Andreas Marek's avatar
Andreas Marek committed
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973

    dnl make sure that all the other kernels are unset
    install_real_generic=yes
    install_real_generic_simple=no
    install_real_sse=no
    install_real_bgp=no	
    install_real_bgq=no
    install_real_avx_block2=no
    install_real_avx_block4=no
    install_real_avx_block6=no

    use_specific_real_kernel=yes
    AC_MSG_NOTICE([real-generic-kernel will be the only compiled kernel for real case])
#  else
#  AC_MSG_FAILURE([--with-sse-generic-kernel failed; SSE kernel cannot be compiled on the system])
#  fi 
 else
  AC_MSG_FAILURE([--with-real-generic-kernel failed; A specific kernel for real case has already been defined before!])
 fi
fi

dnl generic-simple kernel
974
975
AC_ARG_WITH([-only-real-generic-simple-kernel],
 [AS_HELP_STRING([--with-only-real-generic-simple-kernel],[only compile generic-simple-kernel for real case])],[],[with_real_generic_simple_kernel=no])
Andreas Marek's avatar
Andreas Marek committed
976

977
978
if test x"${with_real_generic_simple_kernel}" = x"yes" ; then
 if test x"${use_specific_real_kernel}" = x"no" ; then
Andreas Marek's avatar
Andreas Marek committed
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998

    dnl make sure that all the other kernels are unset
    install_real_generic=no
    install_real_generic_simple=yes
    install_real_sse=no
    install_real_bgp=no	
    install_real_bgq=no
    install_real_avx_block2=no
    install_real_avx_block4=no
    install_real_avx_block6=no

    use_specific_real_kernel=yes
    AC_MSG_NOTICE([real-generic-simple-kernel will be the only compiled kernel for real case])

 else
  AC_MSG_FAILURE([--with-real-generic-simple-kernel failed; A specific kernel for real case has already been defined before!])
 fi
fi

dnl sse kernel
999
1000
AC_ARG_WITH([-only-real-sse-kernel],
 [AS_HELP_STRING([--with-only-real-sse-kernel],[only compile sse-kernel for real case])],[],[with_real_sse_kernel=no])