forked from abinit/abinit
-
Notifications
You must be signed in to change notification settings - Fork 0
/
KNOWN_PROBLEMS
823 lines (643 loc) · 30 KB
/
KNOWN_PROBLEMS
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
Known problems
This file contains the list of known problems
at the time a version of ABINIT is released.
Note that bugs discovered later than the
release are NOT described here...
The present release ABINITv8.10.1 is for production
0. List of platforms
1. Batteries of tests
2. List of known problems
================================================================
0. Platforms
For ABINITv8.10.1, list closed on 2018 10 15
after installation and successful tests for the standard set of platforms,
as mentioned on
https://bbportal.abinit.org/#/status
and
https://wiki.abinit.org/doku.php?id=bb:builder
Corresponding configuration files are to be found in doc/build/config-examples
Supported compilers : gfortran, intel fortran, NAG, IBM XLF.
The following compilers are either obsolete, or considered too unstable/buggy for ABINIT :
g95, pathscale, open64, PGI . This does not mean that ABINIT will not work,
but there is no check on the ABINIT testfarm that ABINIT should run correctly.
Status, for the different versions of compilers that are supported (or might become soon).
GNU Fortran compiler (presently tested : v4.9, v5.3, v5.4, v6.3, v6.4, v7.2, v8.1)
All are mostly OK
gnu 4.9 and 5.4: libxc#82 is disactivated on tikal
gnu 5.3 : bigdft#14 is disactivated on ubu_gnu_5.3_openmpi, and abiref_gnu_5.3_XX
gnu 8.1+openmpi : the reading of WFK files using MPIIO is not correct for test mpiio#26
Intel ifort compiler (presently tested : 13.1, 14.0, 15.0, 16.0, 17.0, 18.0)
Ifort 13.1, mpiio#25, mpiio#26, paral#25, paral#29i, paral#93, paral#94 are disactivated (MPI IO, KGB, or calculations with exact exchange)
Ifort 14.0 is mostly OK, except for paral#93 and paral#94.
Ifort 15.0 is mostly OK, except for one test paral#30 (on ubu). Paral KGB.
Ifort 16.0 is mostly OK, except for one test paral#30 (on ubu). Paral KGB.
Ifort 17.0 is mostly OK, except for bigdft#06, bigdft#07.
Ifort 18.0 is a bit problematic.
- In "real" conditions (mrgscr and mrgdv), the ADVANCE='NO' specification for the write instruction does not work, but
it works in simple programs. Thus all tests (and their chain of tests) that rely on mrgscr and mrgdv have been disabled for this compiler.
Namely, v3#87-91, vv67mbpt#37-39, v7#86-88, v8#41-44, v8#63
- Also, the reading of WFK files using MPIIO is not correct, for tests mpiio#26 and 62.
NAG Compilo
Nag 6.2, mostly OK, except for test etsf_io#4, mpiio#28, v7#80, paral#25, paral#30
IBM
IBM 15.1 is mostly OK, except for the computation of U in cRPA (tests v7#23,24,25).
=============================================================
1. Batteries of tests
Compilation of Netcdf and corresponding testing were tried.
NetCDF is OK for production.
When NetCDF was OK, compilation of ETSFIO and corresponding testing were tried.
ETSF_IO is OK for production.
Compilation of BigDFT, and corresponding testing were tried.
Bigdft is OK for production.
Compilation of Wannier90, and corresponding testing were tried.
Wannier90 is OK for production.
Compilation of libXC, and corresponding testing were tried.
LibXC is OK for production.
Compilation of ATOMPAW and corresponding testing were tried.
ATOMPAW is OK for production.
Compilation of PSML, and corresponding testing were tried.
PSML is OK for production.
Use of MPIIO and corresponding testing were tried.
MPIIO is OK for production, although it does not work on all platforms, but this is not a ABINIT problem.
Use of OpenMP automatically has been tested, and is ready for production.
Use of Scalapack automatically was not tested, and should NOT be attempted by non-expert users with this version.
However, link to scalapack is tested.
Memory leaks are tested automatically.
Speed was not tested.
Memory was not tested.
===============================================
2. List of known problems
All the problems are listed cumulatively.
Those specific to version 8.0 are found at the end of this file
The type of the problem is identified as follows :
NUMERICAL ERROR : the physical result is incorrect,
but the job does not crash. Moreover,
the use of the input variables is legitimate.
This is a very dangerous type of problem.
There are two subclasses :
- INTERNALLY INCOHERENT
(the bug can be seen because two sets of
input variables should give the same results
and do not)
- INTRINSIC
(the bug does not cause incoherency, it can
only be seen when comparing results with
those of another code or with experimental results)
The latter one is the most dangerous, of course ...
CRASH : the code does not produce the result, because
it stops unduly.
SLOWING DOWN : the physical result is correct,
but the code does not go as fast as it should.
For example, the reading of a restart file might
not be correct, or the SCF convergence
might not be as fast as it should.
OTHER : some information (not numerical results)
might be erroneous ...
In each class of problem, one can identify
MACHINE-DEPENDENT problems (on some platform, this
problem does not occur, not on others) as well as
GENERAL problems.
Some of the problems listed below
have been reported by user, but have not yet been
examined for confirmation, in which case they are
labelled "TO BE CONFIRMED".
=================================================
P2. (XG, 001223 + later)
OTHER (TO BE DONE)
GENERAL
The memory needs are not yet evaluated correctly
for the spin-orbit case and non-collinear magnetism
case.
P8. (GMR, November 2000)
SLOWING DOWN
GENERAL
restartxf for optcell/=0 might have a bug,
despite Test_v1#80
Identified in March 2001 by GMR : only happens when
restarting from a different, non-zero, optcell.
Should be corrected, but documented, now.
P15. (XG, 010103)
OTHER
GENERAL
Should enforce coding rule related to length of file
P117. (Laurent Pizzagalli, 010622)
CRASH
Observed on PC AMD, atomic relaxation (ionmov==7),
no cell optimisation, stops in xfpack complaining
about ndim. See mail 010622.
P118. (XG, 010710)
OTHER
GENERAL (d/dk by finite differences)
Test_v3 #5 . Sensitivity to phase choice.
P121. (GMR, 010801)
NUMERICAL RESULT
MACHINE_DEPENDENT (Compaq ES40, Turing)
The use of fourdp , with fftalg 200 or 1xx
gives complex conjugate results. This was seen
in the process of merging the GW code with ABINIT.
To be examined in detail.
P204. (XG, 011020)
OTHER (TO BE DONE)
MACHINE_DEPENDENT
Test_v3#14 : the output file is machine-dependent,
because the spin-phase of spin-degenerate wavefunctions
has not been fixed...
Or, is there something more fundamental ??
P302. (XG, 020305; DDKlug, 020313)
NUMERICAL ERROR
MACHINE_DEPENDENT (IBM 44P; SGI)
RF GGA : Test_v3#8,16,18,60,61,62,66
Problem with Response Function GGA tests.
The accuracy of the RF GGA Test_v3#16
is very bad on these machines : the acoustic
modes have large frequencies...
P350. (XG030225)
OTHER
GENERAL
psp*cc.f the fifth derivative is not computed.
P404. (XG030103)
NUMERICAL ERROR
LIKELY GENERAL
fftalg 401 works, but gives a slightly
different answer than usual for Tfast#10.
To test this, replace the default value in indefo.f
P405. (XG030106)
CONVERGENCE PROBLEM
LIKELY GENERAL
Test_v3#15 : with very well non-SCF converged wavefunction (nnsclo=2),
the total energy at intermediate steps is lower than the
final energy !
P406. (XG030106)
CONVERGENCE PROBLEM
LIKELY GENERAL
Test_v3#85 : set iscf to 5 in the RF, one sees a very bad convergence...
P416. (XG030522)
CRASH
Seen on Intel/PC with PGI compiler, in parallel
Test_v1#75,80 do not work
Problem to read XFhistory part of a wavefunction file.
P41.9. (XG030719)
PORTABILITY PROBLEM
GENERAL
Test_v4#37 is not portable : the spin decomposition of the
spinor wavefunctions is not gauge invariant. This
test has been disabled.
P42.7. (DH030929)
OTHER
Should examine the d2sym3.f routine, giving extra
zeros in the output file. See mail
from DHamman, on 2003-9-29.
P44.2. (DHamann20040527)
NUMERICAL INACCURACY
For the treatment of pseudopotentials with pspcod=6 (FHI),
the non-linear XC core charge is not treated consistently
in psp6cc.f and in the rest of the code.
See Don's mail to the developers of 20040527
See also input variable optnlxccc .
P45.5. (XG050220)
OTHER : PARALLEL TESTING
Test_v4#90 (CASINO output) crash in parallel (test_paral on grumpy).
P45.6. (XG050225)
NUMERICAL PROBLEM (with ionmov=12)
Test_v4#97 has one strange output
P45.17 (B. Andriyevsky 050405)
Problem of normalisation of the epsilon'' from
the "conducti" code ?!
A factor of pi (3.1416) might be missing.
See his mail of 5 April 2005 to the forum.
See also other mails exchanged on the forum.
P45.18 (D Hamann 050510)
NUMERICAL ERROR
With finite-difference ddk calculation
See his mail of 10 May, 2005.
(Excerpt : ...
The problem is that 4.4.4 and 4.5.2 WILL run drf.in and produce incorrect
results, all related to the dataset 3 finite-differend ddk calculation.
The results are only somewhat incorrect, not obvious garbage. Here is a
small sample, Born charges from "2nd-order matrix" in the .out files:
Correct (from erf.out):
1 1 1 4 -4.1367712586 0.0000000000
1 1 2 4 -0.0863319602 0.0000000000
1 1 3 4 0.0706059161 0.0000000000
Incorrect (from drf.out):
1 1 1 4 -3.4782061716 0.0000000000
1 1 2 4 -0.0832639192 0.0000000000
1 1 3 4 0.0560776887 0.0000000000
Putting in some istwfk printing, I found that setting berryopt/=0 now forces
istwfk=1 for dataset 3. The problem apparently is that dataset 2 has some
istwfk>1, and apparently dataset 3 is getting incorrect WFK's from dataset 2
and using them. Why it doesn't re-converge them first isn't clear.
Setting istwfk=1 in dataset 2 (erf.in) fixes the problem and gets correct
results. (My hack also had the effect of setting istwfk=1 in dataset 2 as
well as 3.)
Curiously, removing the dataset 2 calculation altogether and feeding the
kptopt1=1 WFK's from dataset 1 directly to the finite-difference ddk
calculation (nrf.in) also fixes the problem, despite the fact that dataset 1
has some istwfk>1.
Clearly something bad is going on in WFK readin, but only in certain cases.
I've never looked at this part of the code, and I'm afraid I can't divert
any more time now towards tracking this down.
P46.1. (XG050701)
PORTABILITY PROBLEM
Test tutorial Finite electric field
alas_relaxdion.in (the last one)
does not work on my PC : it stops before the end.
P50.16 (XG061215)
DOC
Many documentation files should be updated or removed ...
Although most of the files put on the Web (.html) are OK.
P51.22 (MGiantomassi060627)
The new value of rprimd is not stored in _WFK file .
So, restart cannot benefit from modifications of acell.
See his mail of June 27, 2006. See also P8.
P53.11 (XG070121)
Still have to get rid off old iscf value for tutorial and tutorespfn automatic
testing, as well as updating tutorial html files.
Later, get rid off old iscf value in other automatic tests.
P53.22 (XG070302)
PARALLELISM : EFFICIENCY
outkss.F90 actually works k point by k point, sequentially ...
In the k point and spin loop, one should write locally
vkb,vkbd, en and wfg , without communication, then
transmit them in a separate spin and k point loop.
The k point loop only starts at 1020 , the write start at line 1506.
P53.27 (XG070401, modified XG070601)
NUMERICAL ERROR
GW parallelism over bands
Using two or four procs does not give the same macroscopic
dielectric constant as with one proc., at the level of the fourth digit .
See also test paral N .
P54.1 (XG070407)
NUMERICAL ERROR
With jellium slab, forces differed between PC + PGI4.0 or IFC
See test v5#56
However, in atmdata.F90, the znucl 0 has been defined,
and everything is now the same between the two platforms.
Still, should investigate what is the correct answer ...
P54.7 (XG070601 - actually, much earlier than that, but not documented)
CRASH : finite electric field with atomic displacements
Test ffield_7 does not work on my PC, and has been disabled .
The corresponding section of the ffield tutorial has been commented ...
P54.18 (XG070825)
Other
The computation of born Effective charges and dielectric tensor
under finite electric field should not be activated
with rfuser , but by the usual rfelfd=2, used with berryopt=4 .
See Tv5#23 .
Also, content of the doc file (now doc/input_variables/README_EFIELD_BEC_DIEL)
should be mentioned in the description of the variable rfelfd=2 .
P54.19 (XG070830)
Other
The decomposition of the total energy is not printed out in the case of RF under
electric field. Cases v4#80 and v5#23
P55.22 (XG071109)
Numerical errors, finite electric field
for phonons
Test v4 # 80
Non variational and variational expressions give different numbers ...
See line 1215 of t80.out, compare with line 1342 and line 1344.
Also, lines 2944-2946 should not be printed, as well as
lines 2954-2956, 2964-2966, etc ...
P55.26 (XG071123)
Numerical error
Automatic test v5#24 (Electric field response with spin-orbit)
On PC + g95 , large numerical error (tolerance has been set to a large value)
P55.34 (XG080301)
Numerical problems
Parallel over band, FFT and k points
Test R : there are significant differences
between tR0.out and tR4.out .
(no problem for physical results, e.g. the energy difference
during the convergence is less than 0.1mHa, but not much smaller).
P56.03 (XG080802)
PAW with lobpcg algorithm (and band/FFT parallelism), and istwf 0.
Test W0 : interestingly, it runs correctly in parallel...
The use of iprcel 45 for dataset 4 has been disabled.
dielmt : 15 largest eigenvalues of the hermitian RPA dielectric matrix
1-5 : NaN NaN NaN NaN NaN
6-10 : NaN NaN NaN NaN NaN
11-15: NaN NaN NaN NaN NaN
P56.08 (XG080828)
PAW with lobpcg algorithm (and band/FFT parallelism), and istwf 0.
Test W : numerical results for dataset 5, 6 and 7 are wrong, it seems...
They should be identical to test R and are not.
Also, test W is not portable.
It was disabled for v5.6.0
P56.17 (XG080913)
Wannier90 test procedure is not yet complete :
only the .out are tested.
P56.22 (XG081202)
# (Perhaps solved, should be checked)
# PAW + Spin-orbit is still problematic.
# In the non-spin-polarized case (Kramers degeneracy) nspden=1 with spin-orbit gives different
# results than nspden=4. Also, the forces are not correct.
# See test v5#17
Apparently solved ! (MT feb. 09)
Forces obtained with nspden=1 are 1.d-20 because of symmtries (nsym=6);
With nspden=4, they are 1.d-6, but nsym=1 (due to spinat value).
If better convergence is achieved with nspden=4, forces decrease.
If nsym=1 with nspden=1, forces become 1.d-6.
WARNING : XG090823, the portability of the computation of forces is still
very low.
P56.23 (XG081203)
The Etsf_io support of density files from PAW is not yet working.
Thes test etsf_io #10 has been disabled.
P57.06 (XG090110)
the bin_seq and bin_seqpar tar.gz files contain Makefiles that do not allow to perform the
seq tests. Only tests_speed ...
P57.08 (JZwanzig090112)
On Opteron+ifort10.1 : transferability problem with v3#97 and v4#68
(Interestingly, no problem with chum, that is also an Opteron+ifort10.1).
P59.01 (MG090604)
On fock(IBM PW5) + XLF9.1, GW tests were failing due to call to deep_copy_ch1d in module m_copy.F90
It seems that XLF doesn't handle correctly pointers to string arrays neither when the length is assumed
nor when it's explicitly passed as argument. For the time being, deep_copy_ch1d is declared as private procedure.
P59.03 (XG090713)
For Tv4#69 and Tv2#36 , the .cnf file has to use opt=-easy for chpit+intel11.0 .
Perhaps this will be corrected with a better version of intel 11 ?
Otherwise, should examine why chpit+intel11.0 is so sensitive for these tests, and set back opt=-medium
P59.06 (XG090719)
Original Tetsf_io#08 crashed with chpit+intel11.1 and test+gcc44.0 , while it works
on many other platforms. This happened at the time of the switch between abinis and abinit (with MPI).
It was observed that with ndtset 2 (instead of ndtset 4), no crash occur. This is the current status
of the test, but one should find the problem with ndtset 4.
P59.07 (XG090720)
On bigmac + g95, several problems :
- v3#83 hang
- all libxc fail
- v1#07
- v5#11
P59.09 (XG090827)
Although the Berry phase calculation for norm-conserving case is behaving properly
in parallel, the finite-electric field calculation is not entirely correct :
the eigenenergies differ in the sequential and parallel cases.
See test paral#L
P59.10 (XG090905)
Test v5#40 has low portability. Had to enforce -ridiculous in tests/v5/tests.cnf .
Identified a potential source of low portability : when eigenenergies are degenerate,
but the local DOS contribution is different for degenerate states, there is
a large arbitrariness of the result.
In ABINITv6.0.0, tried to average over the local DOS contributions of degenerate wavefunctions,
writing a new routine dos_degeneratewfs.F90 . Unfortunately, the portability was not improved ?!
See Tv5#40 .
(MG160823) dos_fractions_average_m has been removed to reduce memory requirements.
P59.11 (XG090907)
See mail of BBertrand
L'algorithme de Broyden pour déterminer un acell optimal ne converge pas
dans ce réseau rhombohedral
P59.13 (XG090909)
The input variables bmass and vmass should be merged.
Moreover, the doc is confusing : pointing both to bmass ...
P59.14 (XG090912)
vprtrb does not behave properly, see Tv6#02
P59.15 (MG090914)
GW test procedure is not yet complete :
only the .out are tested. Other external files (_GW, _SIG, _SGR) are not tested.
Similar problem in t532 in which the output and the reference _BXSF files should be compared.
P59.16 (SLeroux091015)
CRASH
With gfortran compiler,
if the first dataset has kptopt1 = 0 and the second dataset has kptopt2 = 1,
during the deallocation of variables before changing the dataset,
abinit crash during deallocation of the variable kg in the routine gstate.F90 (line 1434)
(see tests/v6/t06.in, the commented dataset, for an example of input leading to crash).
There's no problem with ifort compiler.
P59.17 (XG091024)
Not sure pawprtwf works correctly, see the normalisation of wavefunctions
in Tetsf_io#09 (file t09.log)
norm = 1.2344680606460083
norm = 0.97410639792764531
norm = 0.98106849665753615
norm = 0.97966490153644847
norm = 1.0448355540211380
norm = 0.98043908884436204
Close to one, but why bigger than one ?
P61.1 (MG100312)
Check on group closure in v4/t26 and v4/t27 fails if all possible
product of symmetries is considered (new version of chkgrp).
The rotational part belong to the input set of symmetries but
the magnetic part differs. To be investigated in more detail.
For the time being the check on the magnetic part has been removed.
P61.2 (XG100315)
The tests v6#72-76 have big problems of portability, see the enormous tolerances ...
P61.3 (MG100312)
The values of dtsets% printed at the end of the run are not consistent
with the values used during the different runs.
driver.F90 indeed passes a local dtset datatype to its children and
this local datatype is not copied to dtsets(:) before returning.
I've already changed driver.F90 but this modification will affect
several output files. Therefore it should be done by Xavier during the
final merge.
P62.1 (XG100624)
In test libxc#08 , the first values of ETOT differ from platform to platform, by a few part
per million, while the final result converges to the same value. However, in the mean time,
the potential residual differs by a few percent, for absolute values that are very large.
So, the tolerances have to be enormous, while the test is actually portable
at the level of the final result. It should not be very difficult to find
what's happening for the initialisation ... Well, one day or so ...
P63.1 (XG100829)
In test tutorespfn#ffield_6, the first value of ETOT differs in the MPI and serial versions
on testf. As they are reference platforms, I had to use opt=-ridiculous.
However, the test is actually portable
at the level of the final result. It should not be very difficult to find
what's happening for the initialisation. See also P62.1 .
P63.2 (XG100906)
make abirules
is problematic ... Problems should be solved for v6.4.0 in production ... !
P66.1 (XG110115)
On chum/pathscale, the test mpiio/Q4 fails
-P-0000 rwwf.F90:372:BUG
-P-0000 Reading option of rwwf. One should have npw=npw1
-P-0000 However, npw=169, and npw1=0.
-P-0000
-P-0000 leave_new : decision taken to exit ...
So, mpiio testing has been desactivated on chum/pathscale .
P66.2 (XG110205)
Several input variables are not documented or tested, see tests/abirules/Refs/t01.out
P66.3 (XG110206)
The tutorial PAW3 should be automatically tested,
especially now that AtomPAW is a plugin of ABINIT.
A new test directory has been created, to contain
the tutorials in which a plugin is used !
P66.4 (MG110312)
fftalg=412 produces wrong results when fourwf is called with option=2, istwfk=1, cplex=2.
Calling fourwf with fftalg=412 and option=3 (R-->G) causes a sigfault error.
For the time being, the corresponding automatic tests in fftprof have been disabled.
P66.5 (XG110401)
Enabling the GPU for littlebuda has a strange side effect, of making Tv6#66 crash with a segfault.
So, enable_gpu = no was set in build-example.conf
[bb_intel-gfortran4.5_littlebuda]
...
enable_gpu = no
P67.1 (XG110307)
The FFTW3 algorithm cannot be used with DMFT. The default
setting of fftalg=312 in case the FFTW3 library is present has been
disactivated in indefo.F90 when usedmft/=0 . Should be reactivated
when DMFT and FFTW3 are compatible ...
P67.2 (XG110309)
There is a lack of portability for the output of Tv5#07, between
the reference platform and the FFTW3 calculation.
T5#07 being related to self-consistent electron-positron calculation,
with iterative self-consistency control that is less flexible than the
usual one, it has not been possible to adjust the test such as to obtain a better
portability. However, the final results are OK.
P67.3 (XG110315)
There is a lack of portability of Tv5#15 .
The problem lies in the fact that when the density
becomes negative, due to Bragg oscillations, the xc kernel can be extremely large, and when
use for the prediction of the residual of the forces, the differences between machines
can be extremely large. In order to see this effect, put ecut 7 and npulayit 7 (its default value).
P67.4 (XG110331)
The test Tv5#76 gives wrong numerical answer on littlebuda_gcc45_gpu.
Likely due to numerical precision on the GPU card ?
P67.5 (XG110511)
Switching to optforces=2 makes Tparal#T4 crash.
Also leads to some small modifications, hard to understand,
in Tparal #O4, #U4, AB4.
P68.1 (XG110528)
Tests nlo7-10 should be out of tutorespfn ...
They test ird1den and ird1wf, but are not used in the tutorial !
P68.2 (XG110604)
mGGAs from libxc raise several problems ...
(1) Tran-Blaha is hard to make converge. See Tlibxc#11 and 12
(2) TPSS and VSxc from Van Voorhis and Scuseria lead to divergence
when the gradient is zero, while the kinetic energy density is not zero.
P610.3 (JMB111201)
Due to open64 compiler bug, the compilation of "79_seqpar_mpi/lobpcgwf.F90" crashs
The optimization is disabled for the folder 'src/79_seqpar_mpi' :
fcflags_opt_79_seqpar_mpi = -O0 -m64 -align64
P611.3 (XG111220)
Disabled Tv6#102, due to a problem when merged mverstra branch...
P613.1 (XG120423 - updated on 120526 - updated on 120702)
Switch from Bigdft 1.2 to Bigdft 1.6 : the latter does not compile on IBM6 (xlf12). Problem of xlf12 likely.
So, had to disable bigdft on this machine.
P615.1 (XG120513)
The test Tv5#98 (electron-phonon, prediction of the superconducting temperature) is particularly non-portable.
The absolute tolerance is about 4e+3 ...
P615.3 (XG120521)
The test v6#121 had to be transferred to Tseq#v6_121.
See P59.09 . This is linked to a different algorithm for SCF for the sequential and parallel version.
This is a misconception !
One should change this algorithm, and only use the one that is parallel ...
In particular, look at make_grad_berry.F90 routine. In one case, one uses the cg array,
and in the other case, one uses the cgq array.
Even when executed only in serial, the test is not very portable, with high tolerance. In particular,
the openMP results indicates likely an intrinsic problem.
The same is true for other tests seq#v6_XXX with XXX=121,122,123,124, and likely 125 and 126.
A large tolerance has been used for these tests.
P615.5 (MG121010)
[v67mbpt][t22]: crash on cronos when openmp is enabled.
P615.5 (MG121010)
fftalg 112 and ffalg 412 do not support ndat > 1 in fourwf.
This should be solved ASAP by rationalizing the interfaces of the different low-level
routines. BTW: this prevents me from testing the different FFT options in fftprof (see hack a m_fftprof:1378)
P70.3 (MG121101)
The scripts parents.pl does not recognize correctly the children of the procedure.
This problem starts to be very annoying since big modules are now full of CHILDREN sections
that are completely useless.
P71.1 (XG130220)
The abirules checks #3 and 4 are not done anymore on the directory src/65_psp (see the modification of script tests/Scripts/warningchk.py).
This should be temporary : during the time of development of pspatm_pspio.F90.
P71.2 (XG130221)
Functionals ixc -165 and -166 are not portable on IMB6 XLF14, as of LibXC 2.0.1 .
There are not very often used by ABINIT practitioners, so that their testing has been disabled from Tlibxc#06 and 07.
This should be monitored regularly, however.
Also, the TB mGGA is badly portable, also on IBM6 XLF14. Tolerances have been set to very high values in Tlibxc#08.
P71.3 (JMB130406)
Problems with gcc4.8 (yquem_gcc48), Bigdft and Fox had to be unabled :
- problème de link lorsqu'on active bigdft :
problème de "multiple definition" entre
src/11_memory_mpi/lib11_memory_mpi.a(m_profiling.o)
et
lib/libabinit.a(m_profiling.o)
- "petit" problème avec FoX :-\
------------------------------------------------------------------------
Case_01
2
< .Version 7.3.1 of ABINIT
> .Version 6.15.2 of ABINIT
3
< .(MPI version, prepared for a x86_64_linux_gnu4.8 computer)
> .(MPI version, prepared for a x86_64_linux_gnu4.4 computer)
17
< .Starting date : Thu 4 Apr 2013.
> .Starting date : Thu 9 Aug 2012.
178
< 3s 2.00 r= 1.89/3p 2.00 r= 1.89/3d 0.00 r= 1.89/4f NaN r= 1.89/
> 3s 2.00 r= 1.89/3p 2.00 r= 1.89/3d 0.00 r= 1.89/4f 0.00 r= 1.89/
P73.1 (XG130413)
Format of writing is not appropriate for large forces (well, irrealistic, but this should be treated nevertheless).
See tests/paral/Refs/tAJ4.out, line 636
P73.2 (SP130516)
CRASH.
The tests/tutorespfn/Input/trf2_1.in crashes when changing the k-point grid from
ngkpt 4 4 4
nshiftk 4
shiftk 0.0 0.0 0.5
0.0 0.5 0.0
0.5 0.0 0.0
0.5 0.5 0.5
to:
ngkpt 6 6 6
nshiftk 1
shiftk 0.0 0.0 0.0
CRASH:
inwffil.F90:405:ERROR
The file trf2_1o_DS1_WFK cannot be used to start the
present calculation. It was asked that the wavefunctions be accurate, but
at least one of the k points could not be generated from a symmetrical one.
dksqmax= 4.586052E-04
In v7.3.3, lots of segfaults !
See http://buildbot-od.abinit.org/builders/green_intel11/builds/34
P73.4 (MG130614)
paral/t76 gets stuck on max2 if #MPI > 2.
For the time being, this test is disabled on max2
P73.5 (XG130623)
Place 65_lotf_base and 68_lotf under control of abiauty
P77.1 (XG140527)
Test paral#27 MPI4 on vishnu had to be disabled, because erroneous echo at the end of the run.
Should be easy to fix, but no time at present.
P711.1 (XG141130)
Had to disable check-local . To be discussed with Yann.
P801.1 (XG160124)
The test v6#67 has been disabled for nag compiler on petrus, due to a segmentation fault.
This happened at the test of trunk/8.0.1-private/1228, after the merge of several branches for which,
each separately, the test was executing properly. The seg fault has been tracked in dataset 22, just after
having written the density GSR file, the code wanted wo exit fftdatar_write, to go back to the execution of outscfcv,
and the segfault happened there. I was unable to fix the problem.
P802.1 (XG160213)
During the cleaning operation, I tried to remove the line fftalg 401 in mpiio/Input/t25.in (also mpiio/Input/t42.in, paral/Input/t25.in
Although the test executed, this lead to a breakdown of the Analysis script when all tests were done.
TO be clarified.
P802.2 (XG160220)
The tolerance on test tutorial#fold2bloch_2 is VERY large. There is likely a problem of portability in fold2bloch.
P802.4 (XG160228)
In case nsppol=2, with non-zero non-default spinmagntarget, the Fermi energy for spin up and spin down are independent of each other.
Both should be printed.
P804.1 (XG160430)
On cronos, the only parallel tests that are working nightly (beyond the usual MPI on 1, 2, 4 and 10 procs) are the
tutoparal#tdfpt_03 and 04 with 24 procs. One should enable gspw, mbt and string ones ...
P804.1 (XG160501)
AIM utility is not maintained anymore, while in few cases, problems are seen. Might use
http://theory.cm.utexas.edu/henkelman/code/bader (and advice using it).
P808.1 (XG160612)
prt_cif.F90 routine calls spgdata.F90 with spgaxor=-1 and spgorig=-1 . In some cases, brvsb and intsp cannot be determined.
This should be fixed. Also one should make a more exhaustive testing of prtcif input variable !
P811.1 (XG160701)
Tests with DFT+DMFT with CT-QMC algorithm do not work yet on ubu_intel16_mpich.
P811.2 (XG160701)
Several tests for ubu_intel_13.1_openmpi do not work : paral#21,25,26,28,93,94 and mpiio#21,22,25,26,49,51
So, for the time being, keep coba2_intel13 .
P811.3 (XG160701)
With abiref_nag_6.1_openmpi, erratically, one test fail with a timeout (900 secs). Likely due to the execution with -j .
Still, this was not the case with petrus_nag. So, for the time being, keep the latter.
P871.1 (XG171106)
Tutorial positron#4 on bob with Openmp 2 threads, the tolwfr is likely not openmp-shared correcty, because the difference
with the reference for this quantity can be very large, while the difference for the total energy and other quantities
is very small.
P890.1 (XG180504)
Test v8#34 (Hybrid Monte-Carlo) from Serguei Prokhorenko gives randomly different results on nag@abiref.
P890.2 (XG180609)
Test v6_t52 (for electron-phonon).
The tolerance is extremely large. Why is this test lacking so much portability ? Why is this not important ?
At least this portability problem should be commented upon ?!