-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathChangeLog
More file actions
1123 lines (799 loc) · 40.5 KB
/
ChangeLog
File metadata and controls
1123 lines (799 loc) · 40.5 KB
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
ChangeLog 'Dimemas package'
Structure
Date - user - Tool(Simulator | GUI | prv2dim | trf2dim | Other)
FEATURES: explanation of new features
BUGFIX: explanation of bug solved
CHANGES: explanation of codification changes
2019 / 10 / 23 - Chetan - All
BUGFIX: MPI_Alltoallw was mismatched which leads to change of message size during the
simulation is now fixed.
2019 / 08 / 29 - Chetan - All
CHANGE: prv2dim were not able to put burst time for the threads which are started with
time 0.0.
BUGFIX: Unknown events with type 9,10,11.. 14 are appearing in the simulated trace file
which are made optional unless they are present in the original trace they don't
need to appear on the simulated one.
2019 / 06 / 27 - Chetan - All
CHANGE: New MPi call MPI_Alltoallw is added.
BUGFIX: Multi node mapping defined in multiple line wasn't recognized during the
simulation which allow us to define distinct node configuration in same machine.
During --help, --v options and when we are not generating the simulated trace
file the status given by the system was 1 which is also fixed.
2019 / 03 / 29 - Chetan - Simulator
CHANGE: The default option to generate the simulated trace file -p is changed to
optional from obligatory.
--dim flag also made optional if there exist a trace file description in the
configuration file.
2019 / 03 / 1 - Chetan - All
CHANGE: 5.4.1 version released.
2019 / 02 / 21 - Chetan - All
BUGFIX: Identifiers for the cpu and node during the generation of the row file is
modified in order to get same numbering like in original trace file.
Ideal configuration files (.cfg) is added in dimemas packet for user convenient.
2019 / 01 / 25 - Chetan - All
CHANGE: MPI_(Mrecv/Imrecv/Mprove/Improve) new mpi calls is added in the simulation
2018 / 08 / 31 - Chetan - Simulator
BUGFIX: Distinguish between cpu and gpu during the generation of the row file and
also the machine name is added to Distinguish CPUs and nodes.
2018 / 08 / 29 - Chetan - All
CHANGE: 5.4.0 version released.
2018 / 08 / 08 - jmartine - Simulator
CHANGE: Now the paraver ROW file is generated taking in account the hierarchy
and mapping of tasks.
2018 / 08 / 06 - jmartine - Simulator
BUGFIX: Dedicated connection entity was fixed and added debug information for
links management and some connection messages management are fixed.
2018 / 08 / 03 - jmartine - Simulator
BUGFIX: Dedicated connection record is fixed and task mapping when manual
mapping is provided is fixed.
2018 / 06 / 13 - jmartine - All
CHANGE: Speedup the task initialization by traverse trace file just once instead
of for every task in order to get the first record time and removed
unnecesary warnings.
2018 / 06 / 12 - jmartine - Simulator
CHANGE: Injected special events to dimemas trace that informs dimemas about idle
blocks eliminated. Simulated trace is not translating idle state as running
and null pointers are checkd before free current_state.
2018 / 06 / 12 - ckc - prv2dim
BUGFIX: Fix of flag -n during the conversion of trace file from .prv to .dim
prv2dim which Will eliminate the first ideal state.
2018 / 05 / 30 - gllort - All
CHANGE: Sets default compilers to GNU toolchain
2017 / 11 / 13 - jmartine - All
CHANGE: 5.3.4 version released.
2017 / 11 / 10 - jmartine - Simulator
BUGFIX: Fixed bug in resources free when working with more than one
machine.
2017 / 10 / 30 - jmartine - Simulator
FEATURE: Now additional thread can be waked up for reading input trace.
It can achieve better speedups with long traces. There are two
flags available for govern this feature. --asynch-read for
activate it and --asynch-max-buffer that sets the maximum size
of the read buffer (10MB by default).
2017 / 10 / 19 - jmartine - prv2dim
BUGFIX: File descriptors management was broken, so an error arise when
dealing with >1024 tasks.
2017 / 08 / 22 - ckc - Simulator
BUGFIX: kernel (GPU) process mapping has been fixed. There were
situations where kernel processes were progressing in
conventional CPUs. Now pure GPUs objects has been created and
kernel processes can not be preempted from there.
2017 / 08 / 14 - Other
FEATURE: Update_CFG and trf2dim has been removed from the package.
2017 / 08 / 14 - jmartine - prv2dim | Simulator
FEATURE: New parameters parsing. It maintains the same parameters like
before but now positional arguments can also be used in the
standard manner (-x < >, --x-xxx=< >). prv2dim -b argument has
been changed by -b for type and -f for factor.
2017 / 01 / 31 - jmartine - prv2dim | Simulator
FEATURE: Added to the dimemas model the Non-Blocking Collectives
introduced by MPI-3. Now prv2dim can deal with this new calls
so the records for the collectives operations has been
changed.
The model applied to these collectives is the same model that
is applied for the normal collectives but the difference is
that this new ones will be performed in background. Since
there is no enough information for the moment, the collective
- Wait matching is doing by mean of FIFO queues.
2016 / 11 / 23 - jmartine - prv2dim
FEATURE: Added -t --test-miss-rate argument to prv2dim in order to deal
with those regions with high density of MPI_Test miss.
The mechanism is the like with --iproble-miss-rate.
2016 / 08 / 13 - gmoren1 - All - 5.2 .13
FEATURES: GUI now can set, load and save accelerator parameters in the
node configuration panel.prv2dim now can translate CUDA and
OpenCL events, comms and barriers.If gpu information is found
in.row,more than one thread(host and kernel) are going to be
translated in the output trace.
Also, accelerator tasks information is written at the end of
the header of the Dimemas trace to offer it to Dimemas for
the mapping. GPU_burst is a new record defined to simulate
the GPU execution burst with the GPU speed ratio of the accelerator
node. Accelerator barriers are defined with the global_op_id
parameter set in negative values:-1 is with a synch with all
threads, -2 with the thread 0, -3 with the thread 1... and so.
Dimemas now can simulate accelerator threads(CUDA and OpenCL)
and do the mapping between the defined accelerator
nodes(in the cfg with a new record "accelerator node information")
and the accelerator tasks(defined in the header of Dimemas trace).
This mapping can be defined in the cfg or randomly assigned,
always ensuring accelerator tasks in accelerator nodes.
2016/07/01 - jmartine - GUI
CHANGES: Now, the DIMEMAS_HOME path is set automatically.
2016/06/10 - jmartine - Simulator
FEATURES: New feature that deal with the deadlocks that can be
generated because a bad cut of a trace. The
deadlocks can be detected and deleted by mean of an
exaustive analysis of the communications. This
feature can be activated with the new flag
--clean-deadlocks <until-trace-progression>.
2015/03/03 - jgonzale - All - 5.2.12
CHANGES: 5.2.12 Release!!
2015/02/17 - jgonzale - Dimemas
CHANGES: Setting input and output links to 0 (memory and network)
will be simulated as having infinite links
2015/02/09 - jgonzale - Other
CHANGES: Added a check in 'configure.ac.template' to ensure that
C/C++ compilers are installed in the system
2015/02/06 - jgonzale - Simulador
CHANGES: Added missing 'main.cpp' file
CHANGES: Added all missing files from 'common_utilities' directory
2015/01/20 - jgonzale - prv2dim - 5.2.11
BUGFIX: Solved bug in the translation 'InBlockComparison' that
caused different translated traces due to a random comparison
results. Now, when most fields of two Paraver Records are the
same, the comparison takes into account the trace order.
2015/01/15 - jgonzale - GUI
BUGFIX: Solved a bug that created multiple mapping warning
messages when doing a bulk modificacion of 'Target Machine'
nodes ('Do all the same' button on the 'NodeWindow.java').
Now, when pressing this button, the correctness of the application
mapping is checked when all modifications have taken place.
2015/01/08 - jgonzale - GUI
BUGFIX: Major compilation problems solved. Previous distribution
failed ('jar' file did not include the 'es/bsc/...' directory).
Now, the compilation system also generates the 'Version.java'
file and the 'updateAndCheck...' from the 'Makefile.am' of
the GUI works properly.
CHANGES: New About Window in the GUI
2015/01/07 - jgonzale - All - 5.2.10
CHANGES: 5.2.10 Release!
2015/01/07 - jgonzale - GUI
CHANGES: Added support to get a configuration file name as a command
line parameter and also to print on the standard output the
configuration file name saved.
This changes are useful to ease the Paraver integration, as the
communication between the two interfaces get clarified.
NOTE: in some cases, the compilation still files due to the
classpath (where is mixed the Apache Commons IO JAR and the
pure source of the GUI).
2014/12/30 - jgonzale - GUI
CHANGES: Complete restructuration of GUI directory. Now, it includes
a NetBeans project definition and it follows the classic code
distribution of Java applications (es.bsc.cepbatools.dimemas
packages in the corresponding directories.
In addition, the manifest file to create the 'jar' is created
dynamically, and the build is more robust (avoid the fata 'sleep
5' from previous implementation.
2014/12/30 - jgonzale - prv2dim
BUGFIX: Solved a bug when translating Paraver traces with MPI_Iprobe
events and cluster events that caused a bad generation of the
CPU bursts duration. Solved by changing the 'OutBlockComparison'
not to preserve the original events sorting (the last revsion),
but pushing the MPI_Iprobe related events (duration and number
of probes) before the cluster events.
BUGFIX: Probably the trace use to solved the previous problem (a
Linpack execution in ARM from Nikola) was clustered using a
previous version of the clustering tool, because it generates an
excess of clusters' burst closing events. The translator did
not translate correctly these events, as it manages the cluster
events to detect the old blocks. NOTE: This block management should
be changed (erased!).
2014/12/19 - jgonzale - prv2dim
CHANGES: Added support for 'cyclic simulations', i.e. simulations of
Paraver traces that were previously simulated. So far, we
use a fixed events (from 9 to 14, SHOULD BE CHANGED!) to indicate
the different communication endpoints in the first translation
(original Paraver trace). Following translations will use this
events to generate the endpoints, instead of regular communications.
TO CHECK: in following translations, the detection of 'this is
a Paraver trace generated by Dimemas' is done when we detect the
first 'special event'. That shouldn't be done like this (a regular
communication may appear before an event, and crash the translation).
A good approach will be using the 'TraceMetadata', and indicate in
the Paraver trace that it was generated using Dimemas!!
2014/11/04 - jgonzale - Simulator
BUGFIX: Solved a bug when checking the inter-machine flight times.
Changed all references to 'machine->id' to 'machine->id-1'.
CHANGES: Erase the 'Vladimir' panic when recomputing the external
network bandwidth. Now it is not used, but the simulation
works.
2014/11/04 - jgonzale - prv2dim
BUGFIX: Solved an wrong error message in the translator when none
of the MPI ranks execute the MPI_Init
2014/10/16 - jgonzale - GUI
BUGFIX: Solved the bug that prevente the selection of directorios
with dots in the name. I added the Apache 'commons-io' to
easily extract file extensions.
CHANGES: Modified the packaging of the GUI, now the .jar, plus the
'commons-io' jar are installed in $DIMEMAS_HOME/lib/GUI.
CHANGES: Added the first version of the new BSC tools icons :)
2014/10/16 - jgonzale - prv2dim
BUGFIX: Fixed a regression introduced in 5.2.9 when parsing '-b'
parameter. It was located at 'bsc_utils.cpp'
2014/10/01 - jgonzale - prv2dim - 5.2.9
CHANGES: Release 5.2.9
BUGFIX: Wrong initialization order in the ParverTraceParser caused
that '-O3' optimization crashed with a segmentation fault.
BUGFIX: Added the '#include <sys/types.h>' in
'ApplicationDescription.hpp' and 'ParaverTraceParser.hpp'.
When missing, the compilation failed in latest Linux
distributions.
2014/09/10 - jgonzale - Simulator
BUGFIX: Added rounding correction when translating times from
the Dimemas core ('dimemas_timer' -> 'double') to Paraver
trace ('prv_time_t' -> 'unsiged long long').
*NOTE: This precission adjustements should be corrected all
accross the simulator core, as the PRINT_TIMER macro does
truncate the 'dimemas_timer' it receives.
May have sense to work with integer in the simulator core?*
2014/09/05 - jgonzale - Simulator | prv2dim - 5.2.8
BUGFIX: The management of parameter '--lat' was incorrect when
using the new configuration files
CHANGES: Added a barrier synchronization primitive inside each
'MPI_Init' call when translating it. Also added the
parameter '-s' in the translator to avoid the generation
of this synchronization point.
CHANGES: Added the 'ezOptionsParser' header to manage the parameters
in 'prv2dim'. It is an easy solution to deal with
positional parameters and flags, but does not check the
types at all. It is planned to add it to the rest of the
package.
CHANGES: Added the 'bsc_utils' to check if a string or char pointer
codifies a double or a long int, and the corresponding
translators. This class could be extended. Other option is
to use the C++11 'strtod' and similars.
2014/07/31 - jgonzale - All - 5.2.7
CHANGES: Release 5.2.7
2014/07/31 - jgonzale - GUI
CHANGES: Cosmetic changes in 'records_definition.txt' file, that
contains the header of the configuration files generated
by the GUI
2014/07/31 - jgonzale - prv2dim
BUGFIX: Changed the order of the ApplicationList parsing, to
correctly capture the number of communicators
2014/07/31 - jgonzale - Simulator
BUGFIX: Check the existence of the nodes when creating and
allocating tasks
CHANGES: Added a 'printf' wrapper named 'info' that checks the
verbosity level of the simulator when printing the
information
CHANGES: Changed the different numbers on 'exit' calls to
'EXIT_FAILURE'. NOTE: in the 'venusclient.cc' it used a
exit value of 100, this change may cause a interconnection
issue.
CHANGES: The 'thread_state' structure and source files and references
in different structures has been eliminated. This struct
was never used. Will be deprecated in future cleanings
2014/06/03 - jgonzale - Simulator
BUGFIX: Bug in the allocation of the Machine Flight times (single
malloc changed to loop of mallocs in file 'simulator.c')
2014/05/28 - jgonzale - All - 5.2.6
CHANGES: Release 5.2.6
2014/05/26 - jgonzale - Simulator
CHANGES: When a thread finishes with pending messages to send or
receive, the IDLE state is only written if it is an
actual thread ('original_thread')
CHANGES: Now, if there is an existing ROW file, it is copied with
the same output trace prefix
2014/05/12 - jgonzale - prv2dim
CHANGES: If first cluster is imbalanced, no error is shown
CHANGES: Added macro 'SHOW_PERCENTAGE_PROGRESS'
2014/03/27 - jgonzale - All - 5.2.5
CHANGES: Release 5.2.5
2014/03/20 - jgonzale - Simulator
BUGFIX: Solved an incorrect location of 'logical_receives' when
multiple IRecv operations were executed. Now, a queue
named 'irecv_executed' (in t_task/t_thread structure) is
used to indicate the Wait operations the time the IRecvs
took place.
2014/03/20 - jgonzale - prv2dim
FEATURES: MPI_Start/MPI_Startall primitives translate their contents
as IRecv or Send requests.
2014/03/18 - jgonzale - All
CHANGES: Clean of most C/C++ compilation warnings. Not in the GUI
2014/03/17 - jgonzale - prv2dim
BUGFIX: Record comparison inside 'TranslatorRecord.hpp' has been
leveraged when comparing two Events. Now, the trace sort is
preserved. Solved bug #67. NOTE: its has been checked with 4
traces producing appart from the bug example, but it could
cause a regression.
2014/03/10 - jgonzale - Simulator
FEATURES: First addition of Karthikeyan's EEE model. Not tested yet, but
no regressions observed
BUGFIX: Node speed factor is now applied to balanced modules duration
2014/02/26 - jgonzale - Simulator
BUGFIX: Bug when detecting the number of opened files. Changed 'fstat'
to 'fcntl'
BUGFIX: Bug when re-positioning the input trace, on 32-bit linux
2014/02/26 - jgonzale - prv2dim
BUGFIX: Changes in the way the communications are sorted
2014/02/17 - jgonzale - Simulator | prv2dim - 5.2.4
BUGFIX: Bug when generating Paraver trace, communicators numbering
do not need +1 offset
FEATURES: When translating Paraver traces generated by Dimemas, the
Wait synchronizations may appear inside Send primitives
2014/02/10 - jgonzale - Simulator - 5.2.3
BUGFIX: Error when reading external dimemas collectives file
2014/02/05 - jgonzale - Simulator | GUI - 5.2.2
BUGFIX: Modules balancing and ratio where wrongly captured in the
new configuration parsing
BUGFIX: Modules definition in the GUI where sorted randomly
2014/02/04 - jgonzale - GUI - 5.2.1
BUGFIX: Bug when saving a 'n tasks per node' mapping without
pressing ENTER in the text field
BUGFIX: Bug when distributing GUI resource files in the 'Makefile.am'
2014/01/28 - jgonzale - Simulator - 5.2.0 (corrected)
BUGFIX: Bug when parsing a 'predefined map info' that includes a
trace
2014/01/27 - jgonzale - Simulator | GUI - 5.2.0 (corrected)
BUGFIX: Added 'trims' when reading configuration files in the GUI
to solve bug #30
BUGFIX: Correct defition of the 'IO_fseek' in the dimemas header
BUGFIX: Correct creation of output PCF name when output Paraver
trace name does not have any extension, or the extension
is different from '.prv'.
CHANGES: Changed 'INCLUDES' per 'AM_CPPFLAGS' in the different
'Makefile.am' files
BUGFIX: Latency times were not adjusted when reading new
configuration records (multiply the written value per 1e9).
In addition, 'inter_node' latencies where wrongly used when
initializing the nodes
FEATURES: If the latency is smaller than 1ns, it is discarded when
executing a communication. This avoids potential problem of
overlapped states in Paraver
2014/01/24 - jgonzale - Simulator | GUI - 5.2.0
FEATURES: New parameters: '--dim <input_trace>' set a external trace
'--ppn <tasks_per_node>' predefined mapping
'--fill' predefined mapping
'--interlvd' predefined mapping
'--bw <bandwidth>' inter-node bandwidth
'--lat <latency>' inter-node latency
FEATURES: Support for new configuration files with two new record types
'multine node information', to collapse the definition of
multiple nodes with same configuration, and 'predefined map
information', to map tasks to nodes without expressing
directly the task mapping array. In addition, traces are now
not required to be written in the configuration file, so
the simulator execution is more flexible (focusing a better
integration with Paraver)
2013/10/18 - jgonzale - Simulator - 5.1.2
BUGFIX: Bug #38 related to offsets location was not completely
solved. Final solution
2013/10/16 - jgonzale - Simulator - 5.1.1
BUGFIX: Solved a bug when locating the offsets in a trace without
offset information (#38)
2013/10/02 - jgonzale - prv2dim | trf2dim | Other
CHANGES: Merged the trf2dim and trf2trf from Vladimir. Including some
'common_utilities' files
2013/04/30 - jgonzale - prv2dim - 5.1.0
FEATURES: The translator also copies the '.row' file if available.
2013/04/29 - jgonzale - Simulator - 5.1.0
FEATURES: Added back support for v4 CFG files. Simulator sets
mem_buses = # processors and 1 half duplex intra-node link
when using v4 configuration files.
CHANGES: When copying the output PCF, all information befor the event
types are embedded, using the Dimemas default states
2013/03/28 - jgonzale - GUI - 5.0.0
BUGFIX Wrong order in 'Node information' labels
2013/03/28 - jgonzale - Simulator
BUGFIX: Correction in scape characters of 'DimemasUpdateCFG'
2013/03/28 - jgonzale - Simulator | GUI | prv2dim - 5.0.0
FEATURES: Added intra-node contention simulation. This contention
is modelled in the same way as the intra-node contention,
using # of buses, # of input links and # of output links.
It requires adding these three parameters to the configuration
file ('node information' has changed), the GUI and also
the DimemasUpdateCFG, to translate old CFG files no the
newer version.
Due to the introduction of these new feature and the
modificacion of the configuration file, the major version
has been increased.
BUGFIX: 'prv2dim' did not translate the MPI_Rsend primitives. Now
they are translated as regular MPI_Send.
2013/03/07 - jgonzale - Simulator
CHANGES: New way to compute the collective FANs, taking into
account the number of partners at different stages:
- Number of tasks per node
- Number of nodes per machine
- Number of machines in WAN
CHANGES: LIN model now applies P = num_partners-1 (as the total
number of messages interchanged)
BUGFIX: Solved a bug when loading the collectives fine tuning
using an external file. It loaded the values mixing the
internal network and the WAN.
2013/02/26 - jgonzale - prv2dim - 4.4.1pre
BUGFIX: Solved a bug when reading blank lines of a trace with
CRLF line terminators
FEATURES: Added a warning when wrong records (TaskIds out of bounds
found in the trace
2013/02/21 - jgonzale - Simulator
BUGFIX: Add missing include of 'list.h' to 'simulator.c' file that
caused a segmentation fault in linux-ia64 machines.
2013/02/21 - jgonzale - Simulator
BUGFIX: Add missing information about collective communications ids
to the 'extern_comm_model.h' header file
2013/02/20 - jgonzale - Simulator - 4.4.0
FEATURES: Backport of the 'lib_extern_model' support to compute some
communications parameters externally to the simulator core.
2013/01/31 - jgonzale - GUI - 4.3.7
CHANGES: Added execution permission for all users to the 'DimemasGUI'
script
2013/01/31 - jgonzale - prv2dim - 4.3.7
CHANGES: Changed the message of the translator when detecting
non-deterministic MPI primitives in the translation
2013/01/30 - jgonzale - GUI
BUGFIX: Solved a problem when using 'chunked mapping'. Now each
node is saturated in terms of its number of processors and
the processes that not fall in any node, are mapped to
the last node
2013/01/03 - jgonzale - prv2dim - 4.3.6
BUGFIX: MPI_Test[*] primitives are translated as communication
wait records
2013/01/02 - jgonzale - Simulator | GUI
BUGFIX: Dimemas[GUI] not robust against ':' in trace name
2012/12/17 - jgonzale - Simulator - 4.3.5
BUGFIX: Collective operations compute the inter-node 'fans' when
just one node involved
BUGFIX: Event type/values are now managed using 'unsigned long long'
variables
FEATURES: Added 'TODO' file
2012/12/13 - jgonzale - Simulator - 4.3.4
CHANGES: Added the PID to Paraver temporal files to allow multiple
simultaneos executions.
CHANGES: Increased the bison stack size.
2012/12/03 - jgonzale - Simulator - 4.3.3
BUGFIX: TASK_end / PARAVER_end order reversed. On file_data_access
total number of file pointers reserved minus 2, so as to
guarantee the availability to the Paraver trace generator.
2012/12/03 - jgonzale - Simulator - 4.3.2
BUGFIX: TASK_end was called after PARAVER_end causing a resource lock
regarding the total number of files opened.
2012/11/23 - jgonzale - Other
CHANGES: Version update to '4.3.1'
CHANGES: Modifications to adapt to the build farm
2012/10/19 - jgonzale - Other
CHANGES: Version update to '4.3.0'
CHANGES: Minor cosmetic changes in the output
2012/10/19 - jgonzale - Other
BUGFIX: Solved a bug in the compilation system of the 'common_utilities'
that prevent the Dimemas trace to create nanosecond traces.
2012/10/19 - jgonzale - Simulator
CHANGES: Using negative values in the 'modules information' field
of the CFG file now indicated the simulator that all
CPU bursts inside this module will change its duration to
absoulte value of the number indicated. Required to
perform 'forced balancing' simulations.
2012/10/10 - jgonzale - prv2dim
CHANGES: Removed a wrong information message in the
'ParaverTraceParser'.
CHANGES: Version updated to '4.2.2'.
2012/10/10 - jgonzale - Other
BUGFIX: Solved the missing definition 'DIMEMAS_NEW_TRACE' on
'common_utilities' commodity library.
CHANGES: Version updated to '4.2.1'.
CHANGES: Updated README file.
2012/10/09 - jgonzale - Simulator
BUGFIX: Solved a massive leak in 'file_data_access.c' when
reading actions. 'op_fields' variable was never free'd
causing a total memory waste (each trace line stay in
memory during the whole execution)
CHANGES: 'MALLOC_*' (in 'mallocame.h|c') never used anymore.
All memory allocation now uses the pure standard
'malloc/free' calls.
CHANGES: All 'common-files' dependencies are now inside the
source code package. This is the first step in to rebuild
an actual 'common-files' in the Performance Tools team.
BUGFIX: Almost all compilation 'warnings' cleaned.
2012/09/12 - jgonzale - GUI
BUGFIX: Corrected the binary path to call the simulator
(Dimemas3->Dimemas).
2012/09/03 - vladimir - Simulator
CHANGES: Updates on the StarSs scheduler.
2012/09/03 - jgonzale - Other
CHANGES: Added a missing header (basic_types.h) and changed the
building numbering system.
2012/07/31 - jgonzale - Other
CHANGES: Small change in the building system to guarantee the
big file flags inclusion.
2012/07/12 - jgonzale - GUI
CHANGES: GUI script is now called 'DimemasGUI'. It is installation
agnostic, and performs better checks for user-environment
variables and JRE installation.
2012/07/11 - jgonzale - Other
CHANGES: 'configure.ac.' moved to 'configure.ac.template' to keep
track of the versioning easily.
2012/07/11 - jgonzale - trf2dim
BUGFIX: Offsets generation adapted to new trace format.
2012/07/11 - jgonzale - Other
CHANGES: Full integration of the simulator, GUI and translators in
a single package to ease the tool distribution and maintenance.
CHANGES: Commit tagged as Version 4 of the tool, considering stability
of the stability of HOPSA integration.
2012/07/05 - jgonzale - GUI
CHANGES: Deep modification of 'DimemasGUI' to correctly generate new
CFGs files (using new block codification). It is still
under validation.
2012/03/27 - jgonzale - prv2dim
FEATURES: changed 'iprobe' translation to indicate the iprobe threshold
in a miliseconds 'rate' instead of absolute values
2012/03/23 - jgonzale - prv2dim
BUGFIX: solved a wrong header creation when less file descriptors than
tasks to translate were available.
CHANGES: better distinction of wrong communication primitives inside
wrong block (FEATURE of 2012/03/22).
2012/03/22 - grodrigu - Simulator
BUGFIX: Solved a bug on parsing: when reading NOOPs offset location
was not updated properly.
2012/03/22 - jgonzale - prv2dim
FEATURES: SVN numbering using 'configure.ac.template' and a modification
on 'autogen.sh' that uses 'svnversion' application.
FEATURES: Added a check when communication primitives inside a MPI
block does not correspond to the expected (for example a send
primitive inside MPI_recv). A warning message is thrown.
That solves a miss-translation bug discovered in Dalton
caused by the time granularity of the Paraver trace.
2012/03/05 - jgonzale - Simulator
CHANGES: MALLOC_free_memory does not need size any more. Erased
alternative implementation of memory allocation (pool based).
2012/02/28 - jgonzale - Other
CHANGES: Minimun integration
2012/02/21 - grodrigu - Simulator
FEATURES: Added support for application reloading
BUGFIX: Solved a bug when using 'off_t', 'size_t' and 'int' to
index structures.
BUGFIX: Corrected spelling mistake in 'configure.ac'
BUGFIX: Solved a pattern matching when reading MPI communaction
primitives and some pointer errors in 'file_data_access.c'
2012/02/03 - jgonzale - Simulator
CHANGES: Complete integration of all Dimemas versions.
CHANGES: Total change of 'file_data_acces.c'. Now is uses 'getline'
and 'sscanf' as basic ASCII parsing of DIM traces. It also
performs a more sofisticated file descriptor sharing in
conjunction with the new IO library (under the 'io' directory).
Some primitives defined in the new trace description are not
supported yet (One-Sided definitions and operations).
The new IO library acts as a wrapper between the actual C
'stdio.h' and the Dimemas code. The main feature is the
ability to account the number of available file descriptors.
This accounting is crucial for optimal file descriptor sharing.
CHANGES: Also present in the 2012/01/11 commit, but more extended, is
the clear separation of simulator loading and the model
itself. It is not perfect, but the files 'configuration.c|h'
and 'comm_configuration.c|h') include the vast majority of
'target machine' initialization. The schedulers initiliazation
has not been moved to this location, due to its 'function
pointers' design.
CHANGES: Paraver trace generation has been completely redesign. Now
it produces a set of sorted temporal files for states and
communications and a single sorted file for events. The final
merge is an external merge sort (found via Wikipedia in:
http://cis.stvincent.edu/html/tutorials/swd/extsort/extsort.html)
with a fan in of 2. In this way, the required open files
is reduced just to 3 (States&Comms. File, Events File, Output
PRV tracefile). The memory usage is up to 3 times the size
of the MAX_IN_FLIGHT_RECORDS (unfortunately, not in bytes),
located in 'external_sort.h'. When producing the temporal
files it just used 1 buffer of this size. When merging it uses
3 buffers. The number of states and comms. temporal files is
also related to this value (by default: 10000).
PCF file is copied from the first trace that appears in the
configuration file. If not available, default PCF is generated.
CHANGES: Now the Dimemas core is *always in nanoseconds*. This is not
a dramatic change in terms of configuration files, input
traces, etc. (due they expressed in seconds). The output
files are afected because the granularity of the timestamps
is bigger (9 significant digits).
CHANGES: There are some (possible) regressions pending to be checked:
- Application reloads
- Performance messages (they work but need to manually add
'rt' library)
- Multi-machine environment and multi-application environment
(no traces available)
- New modules management version on DimemasGUI not updated
FEATURES: Added a simple binary called 'DimemasUpdateCFG' to CFGs with old
modules definition to new version.
2012/02/03 - jgonzale - prv2dim
CHANGES: Traces are generated *always* in nanoseconds granularity.
FEATURES: PCF file is copied from input trace. If it does not exists
a default is generated.
FEATURES: Added CodeBlocks project
2012/01/11 - jgonzale
CHANGES: First commit with the merge of the Vladimir+German+Official
versions. Now, the TRF trace is deprecated, and all improvements
should be written on the DIM trace. In further commits, all
code that manages the SDDF trace will be erased progresively
adopting the DIM (Paraver style trace) codification of
equivalent semantics.
There have been modifications by Vladimir in the original DIM
trace, basically adding offsets for all threads of all tasks,
sorting each task per thread and adding the thread field on
the P2P communication primitives. This has to be documented.
The current version supports the codification of CPU ratios
using user events present on DIM trace. Paraver and Dimemas
will use same event encoding, and Dimemas blocks are discontinued.
Now it lacks support of generating PCF file, but it can be
copied using the '-pc' parameter.
Most part of the File IO simulation is not currently supported.
It should be analyzed and ported.
FEATURES: It has been added the skeleton to implement the new scalable
IO, as part of the HOPSA project. Every IO made by Dimemas
should use the new interface to be implemented.
2012/01/05 - jgonzale - prv2dim
CHANGES: Major changes to generate the new DIM format accepted by
the Dimemas merged from Vladimir (StarSs)/Official(MPI)
versions
Major changes in the trace file result from the P2P
communication primitives, which include the origin/destination
thread. In MPI, this field takes the value '-1', but should
be changed to keep the trace elegant.
As a comment, it manipulates the MPI/Cluster/User blocks
in the old fashion, but generates user events using the
Paraver encoding. We should evaluate dropping support
to TRF files translation.
2011/10/24 - jgonzale - prv2dim
CHANGES: Building system now builds both translators at the
same time
FEATURES: Added support to translate 'User Functions' in the same
way the Clusters are translated.
2011/07/14 - jgonzale - Other
CHANGES: MAJOR modifications in building system. Now, each big module
of core code is compiled as a facility library. That was
modified in previous update, but not added up to this one
2011/07/13 - jgonzale - Other
CHANGES: Minor cosmetic changes in building system
2011/03/22 - jgonzale - prv2dim
CHANGES: Translator only translates the first thread when
tasks with multiple threads are found
CHANGES: Warning informative message when trace is disordered
CHANGES: Trace flushing outside an MPI are not translated as
a running burst
2010/11/10 - jgonzale - prv2dim
CHANGES: Changes in version numbering (MAJOR.SVN_REVISION)
2010/10/29 - jgonzale - trf2dim
CHANGES: extensive cleaning of building system. Rename of all the
package to 'trf2dim'. First open sourced
2010/10/28 - jgonzale - prv2dim
BUGFIX: 'ParaverTraceParser.cpp +430'. Bad comparison. Thanks icpc!
CHANGES: Reestructuration of building system to guarantee the
correctness of 'make dist'. The tool can be now open
sourced
2010/10/06 - jgonzale - trf2dim
CHANGES: added standard C headers
2010/09/09 - jgonzale - prv2dim
BUGFIX: MPI_Bsend is translated as a regular asynchronous send
(THIS HAS TO BE CHECKED!)
2010/07/13 - jgonzale - Simulator
BUGFIX: Solved an error in the PCF generation, caused by the macro
change. It generated 'null' event value names
2010/07/07 - jgonzale - prv2dim
BUGFIX: changes to correctly manage the root identifiers in
collective operations.
2010/07/07 - jgonzale - Simulator
BUGFIX: Solved a *big* issue regarding to the use of communicators
when simulating collective operations. The 'global OP' records
generated by 'prv2dim' didn't express the actual root of an
collective operation, just which thread is the root. Right
now, this is taken into account in the simulation engine.
2010/05/11 - jgonzale - Simulator
CHANGES: Cluster blocks are now translated coherently when generating
Paraver trace
2010/05/04 - jgonzale - Simulator
CHANGES: Increase value of 'BUFSIZE' and 'YYMAXDEPTH' to improve
scalability. Should be revised
2010/05/04 - jgonzale - prv2dim
BUGFIX: solved an error when translating traces sharing file
descriptor and generating old trace format
CHANGES: added more progress information
2010/03/19 - jgonzale - GUI
CHANGES: Only 'javac' is checked when looking for Java compilers
BUGFIX: Solved used of 'getenv' into the Dimemas GUI
2010/03/15 - jgonzale - GUI
BUGFIX: GUI is only compiled if Java Compiler is present. Previously it
tried to compile it always.
CHANGES: Java m4 macros don't stop when no available compiler/VM/jar is
found
CHANGES: Configure script no longer offers debug/optimization flags.
User must provide this information using CFLAGS/CXXFLAGS.
CHANGES: Removed 'aclocal.m4' from the package distribution.
2010/03/10 - jgonzale - GUI
CHANGES: Minor changes on GUI
2010/03/09 - jgonzale - Simulator
CHANGES: Completed changes to integrate the Java GUI into the main
Dimemas packaging. See DimemasGUI directory
CHANGES: Added 'm4' directories to add automake macros
2010/03/03 - jgonzale - Simulator
BUGFIX: Solved a small misuse when generating the distribution
package
CHANGES: Minor cosmetic changes in 'README' file and elimination of
very old configuration options
2010/02/15 - jgonzale - Simulator
CHANGES: Cleaning to free Dimemas code
2010/01/14 - jgonzale - prv2dim
BUGFIX: solved a bug which produced a segfault when sharing file
descriptors
2009/10/14 - jgonzale - Simulator
BUGFIX: Solved a bug in 'transferencia' function, which did not
return the total transfer time correctly. Right now, resources