-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathft_1D_layout.f
2189 lines (1813 loc) · 76.7 KB
/
ft_1D_layout.f
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
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
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
!-------------------------------------------------------------------------!
! !
! N A S P A R A L L E L B E N C H M A R K S 3.3 !
! !
! F T !
! !
!-------------------------------------------------------------------------!
! !
! This benchmark is part of the NAS Parallel Benchmark 3.3 suite. !
! It is described in NAS Technical Reports 95-020 and 02-007 !
! !
! Permission to use, copy, distribute and modify this software !
! for any purpose with or without fee is hereby granted. We !
! request, however, that all derived work reference the NAS !
! Parallel Benchmarks 3.3. This software is provided "as is" !
! without express or implied warranty. !
! !
! Information on NPB 3.3, including the technical report, the !
! original specifications, source code, results and information !
! on how to submit new results, is available at: !
! !
! http://www.nas.nasa.gov/Software/NPB/ !
! !
! Send comments or suggestions to [email protected] !
! !
! NAS Parallel Benchmarks Group !
! NASA Ames Research Center !
! Mail Stop: T27A-1 !
! Moffett Field, CA 94035-1000 !
! !
! E-mail: [email protected] !
! Fax: (650) 604-3957 !
! !
!-------------------------------------------------------------------------!
!TO REDUCE THE AMOUNT OF MEMORY REQUIRED BY THE BENCHMARK WE NO LONGER
!STORE THE ENTIRE TIME EVOLUTION ARRAY "EX" FOR ALL TIME STEPS, BUT
!JUST FOR THE FIRST. ALSO, IT IS STORED ONLY FOR THE PART OF THE GRID
!FOR WHICH THE CALLING PROCESSOR IS RESPONSIBLE, SO THAT THE MEMORY
!USAGE BECOMES SCALABLE. THIS NEW ARRAY IS CALLED "TWIDDLE" (SEE
!NPB3.0-SER)
!TO AVOID PROBLEMS WITH VERY LARGE ARRAY SIZES THAT ARE COMPUTED BY
!MULTIPLYING GRID DIMENSIONS (CAUSING INTEGER OVERFLOW IN THE VARIABLE
!NTOTAL) AND SUBSEQUENTLY DIVIDING BY THE NUMBER OF PROCESSORS, WE
!COMPUTE THE SIZE OF ARRAY PARTITIONS MORE CONSERVATIVELY AS
!((NX*NY)/NP)*NZ, WHERE NX, NY, AND NZ ARE GRID DIMENSIONS AND NP IS
!THE NUMBER OF PROCESSORS, THE RESULT IS STORED IN "NTDIVNP". FOR THE
!PERFORMANCE CALCULATION WE STORE THE TOTAL NUMBER OF GRID POINTS IN A
!FLOATING POINT NUMBER "NTOTAL_F" INSTEAD OF AN INTEGER.
!THIS FIX WILL FAIL IF THE NUMBER OF PROCESSORS IS SMALL.
!UGLY HACK OF SUBROUTINE IPOW46: FOR VERY LARGE GRIDS THE SINGLE EXPONENT
!FROM NPB2.3 MAY NOT FIT IN A 32-BIT INTEGER. HOWEVER, WE KNOW THAT THE
!"EXPONENT" ARGUMENT OF THIS ROUTINE CAN ALWAYS BE FACTORED INTO A TERM
!DIVISIBLE BY NX (EXP_1) AND ANOTHER TERM (EXP_2). NX IS USUALLY A POWER
!OF TWO, SO WE CAN KEEP HALVING IT UNTIL THE PRODUCT OF EXP_1
!AND EXP_2 IS SMALL ENOUGH (NAMELY EXP_2 ITSELF). THIS UPDATED VERSION
!OF IPWO46, WHICH NOW TAKES THE TWO FACTORS OF "EXPONENT" AS SEPARATE
!ARGUMENTS, MAY BREAK DOWN IF EXP_1 DOES NOT CONTAIN A LARGE POWER OF TWO.
c---------------------------------------------------------------------
c
c Authors: D. Bailey
c W. Saphir
c R. F. Van der Wijngaart
c
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c FT benchmark - This Code only works for CLASS=S & CLASS=C since they're symmetric(equal in all three dimensions).
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c---------------------------------------------------------------------
program ft
c---------------------------------------------------------------------
c---------------------------------------------------------------------
implicit none
include 'mpif.h'
include 'global.h'
integer i, ierr
c---------------------------------------------------------------------
c u0, u1, u2 are the main arrays in the problem.
c Depending on the decomposition, these arrays will have different
c dimensions. To accomodate all possibilities, we allocate them as
c one-dimensional arrays and pass them to subroutines for different
c views
c - u0 contains the initial (transformed) initial condition
c - u1 and u2 are working arrays
c---------------------------------------------------------------------
double complex u0(ntdivnp),
> u1(ntdivnp),
> u2(ntdivnp)
double precision twiddle(ntdivnp)
c---------------------------------------------------------------------
c Large arrays are in common so that they are allocated on the
c heap rather than the stack. This common block is not
c referenced directly anywhere else. Padding is to avoid accidental
c cache problems, since all array sizes are powers of two.
c---------------------------------------------------------------------
double complex pad1(3), pad2(3), pad3(3)
common /bigarrays/ u0, pad1, u1, pad2, u2, pad3, twiddle
integer iter
double precision total_time, mflops
logical verified
character class
call MPI_Init(ierr)
c---------------------------------------------------------------------
c Run the entire problem once to make sure all data is touched.
c This reduces variable startup costs, which is important for such a
c short benchmark. The other NPB 2 implementations are similar.
c---------------------------------------------------------------------
do i = 1, t_max
call timer_clear(i)
end do
call timer_start(T_init)
call setup()
call compute_indexmap(twiddle, dims(1,3), dims(2,3), dims(3,3))
call compute_initial_conditions(u1, dims(1,1), dims(2,1),
> dims(3,1))
call fft_init (dims(1,1))
call fft(1, u1, u0)
call timer_stop(T_init)
if (me .eq. 0) then
print *,'Initialization time =',timer_read(T_init)
endif
c---------------------------------------------------------------------
c Start over from the beginning. Note that all operations must
c be timed, in contrast to other benchmarks.
c---------------------------------------------------------------------
do i = 1, t_max
call timer_clear(i)
end do
call MPI_Barrier(MPI_COMM_WORLD, ierr)
call timer_start(T_total)
if (timers_enabled) call timer_start(T_setup)
call compute_indexmap(twiddle, dims(1,3), dims(2,3), dims(3,3))
call compute_initial_conditions(u1, dims(1,1), dims(2,1),
> dims(3,1))
call fft_init (dims(1,1))
! if (timers_enabled) call synchup()
if (timers_enabled) call timer_stop(T_setup)
if (timers_enabled) call timer_start(T_fft)
call fft(1, u1, u0)
if (timers_enabled) call timer_stop(T_fft)
do iter = 1, niter
if (timers_enabled) call timer_start(T_evolve)
call evolve(u0, u1, twiddle, dims(1,1), dims(2,1), dims(3,1))
if (timers_enabled) call timer_stop(T_evolve)
if (timers_enabled) call timer_start(T_fft)
call fft(-1, u1, u2)
if (timers_enabled) call timer_stop(T_fft)
! if (timers_enabled) call synchup()
if (timers_enabled) call timer_start(T_checksum)
call checksum(iter, u2, dims(1,1), dims(2,1), dims(3,1))
if (timers_enabled) call timer_stop(T_checksum)
end do
call verify(nx, ny, nz, niter, verified, class)
call timer_stop(t_total)
if (np .ne. np_min) verified = .false.
total_time = timer_read(t_total)
if( total_time .ne. 0. ) then
mflops = 1.0d-6*ntotal_f *
> (14.8157+7.19641*log(ntotal_f)
> + (5.23518+7.21113*log(ntotal_f))*niter)
> /total_time
else
mflops = 0.0
endif
if (me .eq. 0) then
call print_results('FT', class, nx, ny, nz, niter, np_min, np,
> total_time, mflops, ' floating point', verified,
> npbversion, compiletime, cs1, cs2, cs3, cs4, cs5, cs6, cs7)
endif
if (timers_enabled) call print_timers()
call MPI_Finalize(ierr)
end
c---------------------------------------------------------------------
c---------------------------------------------------------------------
subroutine evolve(u0, u1, twiddle, d1, d2, d3)
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c evolve u0 -> u1 (t time steps) in fourier space
c---------------------------------------------------------------------
implicit none
include 'global.h'
integer d1, d2, d3
double precision exi
double complex u0(d1,d2,d3)
double complex u1(d1,d2,d3)
double precision twiddle(d1,d2,d3)
integer i, j, k
do k = 1, d3
do j = 1, d2
do i = 1, d1
u0(i,j,k) = u0(i,j,k)*(twiddle(i,j,k))
u1(i,j,k) = u0(i,j,k)
end do
end do
end do
return
end
c---------------------------------------------------------------------
c---------------------------------------------------------------------
subroutine compute_initial_conditions(u0, d1, d2, d3)
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c Fill in array u0 with initial conditions from
c random number generator
c---------------------------------------------------------------------
implicit none
include 'global.h'
integer d1, d2, d3
double complex u0(d1, d2, d3)
integer k
double precision x0, start, an, dummy
c---------------------------------------------------------------------
c 0-D and 1-D layouts are easy because each processor gets a contiguous
c chunk of the array, in the Fortran ordering sense.
c For a 2-D layout, it's a bit more complicated. We always
c have entire x-lines (contiguous) in processor.
c We can do ny/np1 of them at a time since we have
c ny/np1 contiguous in y-direction. But then we jump
c by z-planes (nz/np2 of them, total).
c For the 0-D and 1-D layouts we could do larger chunks, but
c this turns out to have no measurable impact on performance.
c---------------------------------------------------------------------
start = seed
c---------------------------------------------------------------------
c Jump to the starting element for our first plane.
c---------------------------------------------------------------------
call ipow46(a, 2*nx, (zstart(1)-1)*ny + (ystart(1)-1), an)
dummy = randlc(start, an)
call ipow46(a, 2*nx, ny, an)
c---------------------------------------------------------------------
c Go through by z planes filling in one square at a time.
c---------------------------------------------------------------------
do k = 1, dims(3, 1) ! nz/np2
x0 = start
call vranlc(2*nx*dims(2, 1), x0, a, u0(1, 1, k))
if (k .ne. dims(3, 1)) dummy = randlc(start, an)
end do
return
end
c---------------------------------------------------------------------
c---------------------------------------------------------------------
subroutine ipow46(a, exp_1, exp_2, result)
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c compute a^exponent mod 2^46
c---------------------------------------------------------------------
implicit none
double precision a, result, dummy, q, r
integer exp_1, exp_2, n, n2, ierr
external randlc
double precision randlc
logical two_pow
c---------------------------------------------------------------------
c Use
c a^n = a^(n/2)*a^(n/2) if n even else
c a^n = a*a^(n-1) if n odd
c---------------------------------------------------------------------
result = 1
if (exp_2 .eq. 0 .or. exp_1 .eq. 0) return
q = a
r = 1
n = exp_1
two_pow = .true.
do while (two_pow)
n2 = n/2
if (n2 * 2 .eq. n) then
dummy = randlc(q, q)
n = n2
else
n = n * exp_2
two_pow = .false.
endif
end do
do while (n .gt. 1)
n2 = n/2
if (n2 * 2 .eq. n) then
dummy = randlc(q, q)
n = n2
else
dummy = randlc(r, q)
n = n-1
endif
end do
dummy = randlc(r, q)
result = r
return
end
c---------------------------------------------------------------------
c---------------------------------------------------------------------
subroutine setup
c---------------------------------------------------------------------
c---------------------------------------------------------------------
implicit none
include 'mpinpb.h'
include 'global.h'
integer ierr, i, j, fstatus
debug = .FALSE.
call MPI_Comm_size(MPI_COMM_WORLD, np, ierr)
call MPI_Comm_rank(MPI_COMM_WORLD, me, ierr)
if (.not. convertdouble) then
dc_type = MPI_DOUBLE_COMPLEX
else
dc_type = MPI_COMPLEX
endif
if (me .eq. 0) then
write(*, 1000)
open (unit=2,file='timer.flag',status='old',iostat=fstatus)
timers_enabled = .false.
if (fstatus .eq. 0) then
timers_enabled = .true.
close(2)
endif
open (unit=2,file='inputft.data',status='old', iostat=fstatus)
if (fstatus .eq. 0) then
write(*,233)
233 format(' Reading from input file inputft.data')
read (2,*) niter
read (2,*) layout_type
read (2,*) np1, np2
close(2)
c---------------------------------------------------------------------
c check to make sure input data is consistent
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c 1. product of processor grid dims must equal number of processors
c---------------------------------------------------------------------
if (np1 * np2 .ne. np) then
write(*, 238)
238 format(' np1 and np2 given in input file are not valid.')
write(*, 239) np1*np2, np
239 format(' Product is ', i5, ' and should be ', i5)
call MPI_Abort(MPI_COMM_WORLD, 1, ierr)
endif
c---------------------------------------------------------------------
c 2. layout type must be valid
c---------------------------------------------------------------------
if (layout_type .ne. layout_0D .and.
> layout_type .ne. layout_1D .and.
> layout_type .ne. layout_2D) then
write(*, 240)
240 format(' Layout type specified in inputft.data is
> invalid ')
call MPI_Abort(MPI_COMM_WORLD, 1, ierr)
endif
c---------------------------------------------------------------------
c 3. 0D layout must be 1x1 grid
c---------------------------------------------------------------------
if (layout_type .eq. layout_0D .and.
> (np1 .ne.1 .or. np2 .ne. 1)) then
write(*, 241)
241 format(' For 0D layout, both np1 and np2 must be 1 ')
call MPI_Abort(MPI_COMM_WORLD, 1, ierr)
endif
c---------------------------------------------------------------------
c 4. 1D layout must be 1xN grid
c---------------------------------------------------------------------
if (layout_type .eq. layout_1D .and. np1 .ne. 1) then
write(*, 242)
242 format(' For 1D layout, np1 must be 1 ')
call MPI_Abort(MPI_COMM_WORLD, 1, ierr)
endif
else
write(*,234)
niter = niter_default
if (np .eq. 1) then
np1 = 1
np2 = 1
layout_type = layout_0D
else if (np .le. nz) then
np1 = 1
np2 = np
layout_type = layout_1D
else
np1 = nz
np2 = np/nz
layout_type = layout_2D
endif
endif
if (np .lt. np_min) then
write(*, 10) np_min
10 format(' Error: Compiled for ', I5, ' processors. ')
write(*, 11) np
11 format(' Only ', i5, ' processors found ')
call MPI_Abort(MPI_COMM_WORLD, 1, ierr)
endif
234 format(' No input file inputft.data. Using compiled defaults')
write(*, 1001) nx, ny, nz
write(*, 1002) niter
write(*, 1004) np
write(*, 1005) np1, np2
if (np .ne. np_min) write(*, 1006) np_min
if (layout_type .eq. layout_0D) then
write(*, 1010) '0D'
else if (layout_type .eq. layout_1D) then
write(*, 1010) '1D'
else
write(*, 1010) '2D'
endif
1000 format(//,' NAS Parallel Benchmarks 3.3 -- FT Benchmark',/)
1001 format(' Size : ', i4, 'x', i4, 'x', i4)
1002 format(' Iterations : ', 7x, i7)
1004 format(' Number of processes : ', 7x, i7)
1005 format(' Processor array : ', 5x, i4, 'x', i4)
1006 format(' WARNING: compiled for ', i5, ' processes. ',
> ' Will not verify. ')
1010 format(' Layout type : ', 9x, A5)
endif
c---------------------------------------------------------------------
c Broadcast parameters
c---------------------------------------------------------------------
call MPI_BCAST(np1, 1, MPI_INTEGER, 0, MPI_COMM_WORLD, ierr)
call MPI_BCAST(np2, 1, MPI_INTEGER, 0, MPI_COMM_WORLD, ierr)
call MPI_BCAST(layout_type, 1, MPI_INTEGER, 0, MPI_COMM_WORLD,
& ierr)
call MPI_BCAST(niter, 1, MPI_INTEGER, 0, MPI_COMM_WORLD, ierr)
call MPI_BCAST(timers_enabled, 1, MPI_LOGICAL, 0, MPI_COMM_WORLD,
& ierr)
if (np1 .eq. 1 .and. np2 .eq. 1) then
layout_type = layout_0D
else if (np1 .eq. 1) then
layout_type = layout_1D
else
layout_type = layout_2D
endif
if (layout_type .eq. layout_0D) then
do i = 1, 3
dims(1, i) = nx
dims(2, i) = ny
dims(3, i) = nz
end do
else if (layout_type .eq. layout_1D) then
dims(1, 1) = nx
dims(2, 1) = ny
dims(3, 1) = nz
dims(1, 2) = nx
dims(2, 2) = ny
dims(3, 2) = nz
dims(1, 3) = nz
dims(2, 3) = nx
dims(3, 3) = ny
else if (layout_type .eq. layout_2D) then
dims(1, 1) = nx
dims(2, 1) = ny
dims(3, 1) = nz
dims(1, 2) = ny
dims(2, 2) = nx
dims(3, 2) = nz
dims(1, 3) = nz
dims(2, 3) = nx
dims(3, 3) = ny
endif
do i = 1, 3
dims(2, i) = dims(2, i) / np1
dims(3, i) = dims(3, i) / np2
end do
c---------------------------------------------------------------------
c Determine processor coordinates of this processor
c Processor grid is np1xnp2.
c Arrays are always (n1, n2/np1, n3/np2)
c Processor coords are zero-based.
c---------------------------------------------------------------------
me2 = mod(me, np2) ! goes from 0...np2-1
me1 = me/np2 ! goes from 0...np1-1
c---------------------------------------------------------------------
c Communicators for rows/columns of processor grid.
c commslice1 is communicator of all procs with same me1, ranked as me2
c commslice2 is communicator of all procs with same me2, ranked as me1
c mpi_comm_split(comm, color, key, ...)
c---------------------------------------------------------------------
call MPI_Comm_split(MPI_COMM_WORLD, me1, me2, commslice1, ierr)
call MPI_Comm_split(MPI_COMM_WORLD, me2, me1, commslice2, ierr)
! if (timers_enabled) call synchup()
if (debug) print *, 'proc coords: ', me, me1, me2
c---------------------------------------------------------------------
c Determine which section of the grid is owned by this
c processor.
c---------------------------------------------------------------------
if (layout_type .eq. layout_0d) then
do i = 1, 3
xstart(i) = 1
xend(i) = nx
ystart(i) = 1
yend(i) = ny
zstart(i) = 1
zend(i) = nz
end do
else if (layout_type .eq. layout_1d) then
xstart(1) = 1
xend(1) = nx
ystart(1) = 1
yend(1) = ny
zstart(1) = 1 + me2 * nz/np2
zend(1) = (me2+1) * nz/np2
xstart(2) = 1
xend(2) = nx
ystart(2) = 1
yend(2) = ny
zstart(2) = 1 + me2 * nz/np2
zend(2) = (me2+1) * nz/np2
xstart(3) = 1
xend(3) = nx
ystart(3) = 1 + me2 * ny/np2
yend(3) = (me2+1) * ny/np2
zstart(3) = 1
zend(3) = nz
else if (layout_type .eq. layout_2d) then
xstart(1) = 1
xend(1) = nx
ystart(1) = 1 + me1 * ny/np1
yend(1) = (me1+1) * ny/np1
zstart(1) = 1 + me2 * nz/np2
zend(1) = (me2+1) * nz/np2
xstart(2) = 1 + me1 * nx/np1
xend(2) = (me1+1)*nx/np1
ystart(2) = 1
yend(2) = ny
zstart(2) = zstart(1)
zend(2) = zend(1)
xstart(3) = xstart(2)
xend(3) = xend(2)
ystart(3) = 1 + me2 *ny/np2
yend(3) = (me2+1)*ny/np2
zstart(3) = 1
zend(3) = nz
endif
c---------------------------------------------------------------------
c Set up info for blocking of ffts and transposes. This improves
c performance on cache-based systems. Blocking involves
c working on a chunk of the problem at a time, taking chunks
c along the first, second, or third dimension.
c
c - In cffts1 blocking is on 2nd dimension (with fft on 1st dim)
c - In cffts2/3 blocking is on 1st dimension (with fft on 2nd and 3rd dims)
c Since 1st dim is always in processor, we'll assume it's long enough
c (default blocking factor is 16 so min size for 1st dim is 16)
c The only case we have to worry about is cffts1 in a 2d decomposition.
c so the blocking factor should not be larger than the 2nd dimension.
c---------------------------------------------------------------------
fftblock = fftblock_default
fftblockpad = fftblockpad_default
if (layout_type .eq. layout_2d) then
if (dims(2, 1) .lt. fftblock) fftblock = dims(2, 1)
if (dims(2, 2) .lt. fftblock) fftblock = dims(2, 2)
if (dims(2, 3) .lt. fftblock) fftblock = dims(2, 3)
endif
if (fftblock .ne. fftblock_default) fftblockpad = fftblock+3
return
end
c---------------------------------------------------------------------
c---------------------------------------------------------------------
subroutine compute_indexmap(twiddle, d1, d2, d3)
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c---------------------------------------------------------------------
c compute function from local (i,j,k) to ibar^2+jbar^2+kbar^2
c for time evolution exponent.
c---------------------------------------------------------------------
implicit none
include 'mpinpb.h'
include 'global.h'
integer d1, d2, d3
integer i, j, k, ii, ii2, jj, ij2, kk
double precision ap, twiddle(d1, d2, d3)
c---------------------------------------------------------------------
c this function is very different depending on whether
c we are in the 0d, 1d or 2d layout. Compute separately.
c basically we want to convert the fortran indices
c 1 2 3 4 5 6 7 8
c to
c 0 1 2 3 -4 -3 -2 -1
c The following magic formula does the trick:
c mod(i-1+n/2, n) - n/2
c---------------------------------------------------------------------
ap = - 4.d0 * alpha * pi *pi
if (layout_type .eq. layout_0d) then ! xyz layout
do i = 1, dims(1,3)
ii = mod(i+xstart(3)-2+nx/2, nx) - nx/2
ii2 = ii*ii
do j = 1, dims(2,3)
jj = mod(j+ystart(3)-2+ny/2, ny) - ny/2
ij2 = jj*jj+ii2
do k = 1, dims(3,3)
kk = mod(k+zstart(3)-2+nz/2, nz) - nz/2
twiddle(i,j,k) = dexp(ap*dfloat(kk*kk+ij2))
end do
end do
end do
else if (layout_type .eq. layout_1d) then ! zxy layout
do i = 1,dims(2,3)
ii = mod(i+xstart(3)-2+nx/2, nx) - nx/2
ii2 = ii*ii
do j = 1,dims(3,3)
jj = mod(j+ystart(3)-2+ny/2, ny) - ny/2
ij2 = jj*jj+ii2
do k = 1,dims(1,3)
kk = mod(k+zstart(3)-2+nz/2, nz) - nz/2
twiddle(k,i,j) = dexp(ap*dfloat(kk*kk+ij2))
end do
end do
end do
else if (layout_type .eq. layout_2d) then ! zxy layout
do i = 1,dims(2,3)
ii = mod(i+xstart(3)-2+nx/2, nx) - nx/2
ii2 = ii*ii
do j = 1, dims(3,3)
jj = mod(j+ystart(3)-2+ny/2, ny) - ny/2
ij2 = jj*jj+ii2
do k =1,dims(1,3)
kk = mod(k+zstart(3)-2+nz/2, nz) - nz/2
twiddle(k,i,j) = dexp(ap*dfloat(kk*kk+ij2))
end do
end do
end do
else
print *, ' Unknown layout type ', layout_type
stop
endif
return
end
c---------------------------------------------------------------------
c---------------------------------------------------------------------
subroutine print_timers()
c---------------------------------------------------------------------
c---------------------------------------------------------------------
implicit none
integer i, ierr
include 'global.h'
include 'mpinpb.h'
character*25 tstrings(T_max+2)
double precision t1(T_max+2), tsum(T_max+2),
> tming(T_max+2), tmaxg(T_max+2)
data tstrings / ' total ',
> ' setup ',
> ' fft ',
> ' evolve ',
> ' checksum ',
> ' fftlow ',
> ' fftcopy ',
> ' transpose ',
> ' transpose1_loc ',
> ' transpose1_glo ',
> ' transpose1_fin ',
> ' transpose2_loc ',
> ' transpose2_glo ',
> ' transpose2_fin ',
> ' sync ',
> ' init ',
> ' totcomp ',
> ' totcomm ' /
do i = 1, t_max
t1(i) = timer_read(i)
end do
t1(t_max+2) = t1(t_transxzglo) + t1(t_transxyglo) + t1(t_synch)
t1(t_max+1) = t1(t_total) - t1(t_max+2)
call MPI_Reduce(t1, tsum, t_max+2, MPI_DOUBLE_PRECISION,
> MPI_SUM, 0, MPI_COMM_WORLD, ierr)
call MPI_Reduce(t1, tming, t_max+2, MPI_DOUBLE_PRECISION,
> MPI_MIN, 0, MPI_COMM_WORLD, ierr)
call MPI_Reduce(t1, tmaxg, t_max+2, MPI_DOUBLE_PRECISION,
> MPI_MAX, 0, MPI_COMM_WORLD, ierr)
if (me .ne. 0) return
write(*, 800) np
do i = 1, t_max+2
if (tsum(i) .ne. 0.0d0) then
write(*, 810) i, tstrings(i), tming(i), tmaxg(i), tsum(i)/np
endif
end do
800 format(' nprocs =', i6, 19x, 'minimum', 5x, 'maximum',
> 5x, 'average')
810 format(' timer ', i2, '(', A16, ') :', 3(2X,F10.4))
return
end
c---------------------------------------------------------------------
c---------------------------------------------------------------------
subroutine fft(dir, x1, x2)
c---------------------------------------------------------------------
c---------------------------------------------------------------------
implicit none
integer dir
include 'global.h'
double complex x1(ntdivnp), x2(ntdivnp)
#ifdef __FFTW__
include 'fftw3.f'
integer*8 plan_x_z_forward,plan_x_z_backward
integer*8 plan_y_forward, plan_y_backward
integer rank,howmany,vector(1)
integer istride,idist,inembed(3),
> ostride,odist,onembed(3)
integer i,j,k,indi
#endif
double complex scratch(fftblockpad_default*maxdim*2)
c---------------------------------------------------------------------
c note: args x1, x2 must be different arrays
c note: args for cfftsx are (direction, layout, xin, xout, scratch)
c xin/xout may be the same and it can be somewhat faster
c if they are
c note: args for transpose are (layout1, layout2, xin, xout)
c xin/xout must be different
c---------------------------------------------------------------------
if (dir .eq. 1) then
if (layout_type .eq. layout_0d) then
call cffts1(1, dims(1,1), dims(2,1), dims(3,1),
> x1, x1, scratch)
call cffts2(1, dims(1,2), dims(2,2), dims(3,2),
> x1, x1, scratch)
call cffts3(1, dims(1,3), dims(2,3), dims(3,3),
> x1, x2, scratch)
else if (layout_type .eq. layout_1d) then
#ifndef __FFTW__
call cffts1(1, dims(1,1), dims(2,1), dims(3,1),
> x1, x1, scratch)
#else
rank = 1
vector = dims(1,1)
howmany = dims(2,1) * dims(3,1)
inembed(1) = dims(1,1)
inembed(2) = dims(2,1)
inembed(3) = dims(3,1)
istride = 1
idist = dims(1,1)
onembed(1) = inembed(1)
onembed(2) = inembed(2)
onembed(3) = inembed(3)
ostride = istride
odist = idist
call dfftw_plan_many_dft(plan_x_z_backward, rank, vector,
> howmany,
> x1, inembed, istride, idist,
> x1, onembed, ostride, odist,
> FFTW_BACKWARD, FFTW_ESTIMATE)
call dfftw_execute_dft(plan_x_z_backward, x1, x1)
#endif
#ifndef __FFTW__
call cffts2(1, dims(1,2), dims(2,2), dims(3,2),
> x1, x1, scratch)
#else
rank = 1
vector = dims(1,2)
howmany = dims(2,2)
inembed(1) = dims(1,2)
inembed(2) = dims(2,2)
inembed(3) = dims(3,2)
istride = dims(1,2)
idist = 1
onembed(1) = inembed(1)
onembed(2) = onembed(2)
onembed(3) = onembed(3)
ostride = istride
odist = idist
call dfftw_plan_many_dft(plan_y_backward, rank, vector,
> howmany,
> x1, inembed, istride, idist,
> x1, onembed, ostride, odist,
> FFTW_BACKWARD, FFTW_ESTIMATE)
do k = 1, dims(3,2)
indi = (k - 1) * dims(1,2) * dims(2,2) + 1
call dfftw_execute_dft(plan_y_backward, x1(indi), x1(indi))
end do
call dfftw_destroy_plan(plan_y_backward)
#endif
if (timers_enabled) call timer_start(T_transpose)
call transpose_xy_z(2, 3, x1, x2)
if (timers_enabled) call timer_stop(T_transpose)
#ifndef __FFTW__
call cffts1(1, dims(1,3), dims(2,3), dims(3,3),
> x2, x2, scratch)
#else
rank = 1
vector = dims(1,3)
howmany = dims(2,3) * dims(3,3)
inembed(1) = dims(1,3)
inembed(2) = dims(2,3)
inembed(3) = dims(3,3)
istride = 1
idist = dims(1,3)
onembed(1) = inembed(1)
onembed(2) = inembed(2)
onembed(3) = inembed(3)
ostride = istride
odist = idist
call dfftw_execute_dft(plan_x_z_backward, x2, x2)
call dfftw_destroy_plan(plan_x_z_backward)
#endif
else if (layout_type .eq. layout_2d) then
call cffts1(1, dims(1,1), dims(2,1), dims(3,1),
> x1, x1, scratch)
if (timers_enabled) call timer_start(T_transpose)
call transpose_x_y(1, 2, x1, x2)
if (timers_enabled) call timer_stop(T_transpose)
call cffts1(1, dims(1,2), dims(2,2), dims(3,2),
> x2, x2, scratch)
if (timers_enabled) call timer_start(T_transpose)
call transpose_x_z(2, 3, x2, x1)
if (timers_enabled) call timer_stop(T_transpose)
call cffts1(1, dims(1,3), dims(2,3), dims(3,3),
> x1, x2, scratch)
endif
else
if (layout_type .eq. layout_0d) then
call cffts3(-1, dims(1,3), dims(2,3), dims(3,3),
> x1, x1, scratch)
call cffts2(-1, dims(1,2), dims(2,2), dims(3,2),
> x1, x1, scratch)
call cffts1(-1, dims(1,1), dims(2,1), dims(3,1),
> x1, x2, scratch)
else if (layout_type .eq. layout_1d) then
#ifndef __FFTW__
call cffts1(-1, dims(1,3), dims(2,3), dims(3,3),
> x1, x1, scratch)
#else
rank = 1
vector = dims(1,3)
howmany = dims(2,3) * dims(3,3)
inembed(1) = dims(1,3)
inembed(2) = dims(2,3)
inembed(3) = dims(3,3)
istride = 1
idist = dims(1,3)
onembed(1) = inembed(1)
onembed(2) = inembed(2)
onembed(3) = onembed(3)
ostride = istride
odist = idist
call dfftw_plan_many_dft(plan_x_z_forward, rank, vector,
> howmany,
> x1, inembed, istride, idist,
> x1, onembed, ostride, odist,
> FFTW_FORWARD, FFTW_ESTIMATE)
call dfftw_execute_dft(plan_x_z_forward, x1, x1)
#endif
if (timers_enabled) call timer_start(T_transpose)
call transpose_x_yz(3, 2, x1, x2)
if (timers_enabled) call timer_stop(T_transpose)
#ifndef __FFTW__
call cffts2(-1, dims(1,2), dims(2,2), dims(3,2),
> x2, x2, scratch)
#else
rank = 1
vector = dims(1,2)
howmany = dims(2,2)
inembed(1) = dims(1,2)
inembed(2) = dims(2,2)
inembed(3) = dims(3,2)
istride = dims(1,2)
idist = 1
onembed(1) = inembed(1)
onembed(2) = inembed(2)
onembed(3) = inembed(3)
ostride = dims(1,2)
odist = idist
call dfftw_plan_many_dft(plan_y_forward, rank, vector,
> howmany,
> x1, inembed, istride, idist,