-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathcpe-23.09-sles15-sp4-ReleaseNotes.txt
1222 lines (926 loc) · 42 KB
/
cpe-23.09-sles15-sp4-ReleaseNotes.txt
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
HPE Cray Programming Environment 23.09 Release Notes
====================================================
HPE Cray Compiling Environment (CCE) 16.0.1
===========================================
Release Date:
-------------
September 2023
Purpose:
--------
CCE 16.0.1 provides Fortran, C, and C++ compilers for
HPE Cray supercomputer and HPE Apollo 2000 Gen10Plus systems.
Key Changes and Support with CCE 16.0.1:
----------------------------------------
CCE 16.0.1:
-----------
CCE-16.0.1
- Bug fixes
- binutils updated to version 2.40
------------
CCE-16.0.0
- LLVM 16 base
- RHEL gcc-toolset-10 support/integration
- On RHEL, CCE uses headers and libraries from RHEL gcc-toolset-10 instead of
cray-gcc-10.3.0; other operating systems are unchanged
- OpenMP 5.0 full support, including the following new features:
- taskloop cancellation (Fortran)
- conditional lastprivate (Fortran)
- "if" clause on "simd" constructs (Fortran)
- iterators in "depend" clauses (Fortran)
- depend objects (Fortran)
- task reductions (Fortran)
- non-rectangular loop collapse (Fortran) - functional
- inclusive and exclusive scan operations (Fortran) - functional
- "declare mapper" (Fortran) - limited
- "loop" construct (C/C++) - limited
- Optimized OpenMP GPU parallelism mapping policy
- "omp teams" still maps to GPU coarse-grained parallelism (threadblocks or work groups)
- "omp parallel" now maps to GPU fine-grained parallelism (threads or work items)
- "omp simd" can still map to fine-grained parallelism (Fortran only)
- TCMalloc is removed in Fortran
- CCE no longer provides TCMalloc as an option replacement for the GLIBC
allocator. The related Fortran options (-htcmalloc/-hsystem_alloc) have been
removed.
- Tuned Support for AMD Genoa
- Grace SVE Support
- DWARF5 Enhancements
- Support for split DWARF
- Support for compressed DWARF
- Support for inline functions to depth 1
- Beta support for variables in OpenMP regions that are offloaded to AMD GPU
User and Application Impact:
----------------------------
CCE 16.0.1:
-----------
CCE-16.0.1
------------
CCE-16.0.0
On RHEL, CCE 16 uses headers and libraries from RHEL gcc-toolset-10. Prior
versions of CCE use headers and libraries from cray-gcc-10.3.0. Binaries built
with prior versions of CCE may need to be rebuilt. Users who are not using the
RHEL operating system are unaffected.
While the optional tcmalloc allocator provided a significant performance boost to some
multi-threaded applications, many third-party libraries and tools expect applications
to use the glibc allocator and do not work or work well with it so CCE is no longer
providing it. Users who still wish to use it will need to build it themselves.
Issues or Bugs Resolved:
------------------------
CCE 16.0.1:
-----------
CCE-16.0.1
CAST-30592 Memory Sanitizer
CAST-30921 (ELCAP-169) CCE 14.0.2 fails compiling Trillinos STK test with out of registers with no backtrace - Elevation of Case 5367010990
CAST-32040 Cray Compiler OpenACC - Hundreds of .acc.s and .acc.o files left after compiling
CAST-32937 crayftn error parsing type(double precision)
CAST-33123 ftn compile error for print in omp target loop if -R b
CAST-33380 FVCOM raises CCE internal compiler error
CAST-33381 ONETEP raises a CCE internal compiler error (segfault)
CAST-33445 UK Met Office climate benchmark crashes with segfault with CPE 23.05 (CCE 16.0)
CAST-33511 Logical transfers within an OMP region in cce16.0.0
------------
CCE-16.0.0
CAST-29703 CCE - internal compiler error
CAST-30549 LAMMPS CCE build failure when perftools or perftools-lite are loaded
CAST-30570 (CESM) fails to execute with CCE 14
CAST-30965 RFE: crayftn should respect LIBRARY_PATH
CAST-31156 Several problems with Cray CCE compiler version 14.0.3 and GFS weather code
CAST-31322 setonix - MultiGPU management with OpenMP multiple threads in a single task fails for Cray-clang compiler
CAST-31444 OpenACC unsupported call to variadic function
CAST-31567 Cray Fortran compiler error when compiling function with BIND
CAST-31568 Spurious CAUTION messages from Cray Fortran with O1 or higher optimization
CAST-31719 cce 15.0.0 ftn routine call in target offload loop gets unsupported call to variadic function - Elevation of Case 5369535423
CAST-31723 Function in target region linking error in CCE 15.0.0.3
CAST-31726 Crayftn bug: zero-sized array assignment
CAST-31809 Cray compiler optimization bug at -O2 and above
CAST-31813 ftn internal compiler error
CAST-31877 Compiler abort with simple OpenMP TARGET usage
CAST-31953 cce/15.0.0 internal compiler error
CAST-31998 Parameters and fixed values as optional arguments
CAST-32107 Cray Fortran Problem
CAST-32355 Build failure when using cce/15.0.0 and rocm/5.4.0
CAST-32360 Incorrect OpenMP Fortran compiler error
CAST-32415 CCE Assertion error when compiling rng.cc - Elevation of case 5371904261
CAST-32450 Pointer variables and class dummy variables fails with ACC: find_in_present_table failed for 'XXX (dope vector)'
CAST-32458 The DART application gives wrong answers when compiled with CCE
CAST-32485 Tioga - CCE 15.0.1 fails with Assertion `DwarfReg > = 0 && "Invalid dwarf register number"' failed compiling E3SM HIP source with -O0 -g - Elevation of Case 5372060962
CAST-32612 crayftn reports an error for recursive elemental procedures.
CAST-32040 Cray Compiler OpenACC - Hundreds of .acc.s and .acc.o files left after compiling
Product and OS Dependencies:
----------------------------
This CCE release is supported on
- HPE Cray supercomputer systems running CSM with COS 2.4.X or COS 2.5.X (SLES15 SP4)
- HPE Cray supercomputer systems running HPCM with SLES15 SP4
- HPE Cray supercomputer systems running HPCM with RHEL 8.6
- HPE Apollo 2000 Gen10Plus systems running RHEL 8.6
- AMD GPU offloading support requires at least ROCm 5.5.1
Notes and Limitations:
----------------------
Documentation:
--------------
- S-5212 Cray Compiling Environment Release Overview (16.0)
- S-3901 Cray Fortran Reference Manual
- S-2179 Cray C and C++ Quick Reference
- Basic man pages: crayftn(1), craycc(1), crayCC(1), intro_openmp()
- Please see https://clang.llvm.org/docs/UsersManual.html or use the -help command line option for
more information on using Clang
Modulefile:
-----------
The following will load the modules necessary to use CCE:
module load PrgEnv-cray
The following will switch to x.y.z version of CCE:
module swap cce cce/x.y.z
Installation instructions:
--------------------------
rpm -ivh cce-16.0.1-202308051334.68fa09927f4eb-1.sles15sp5.x86_64.rpm.x86_64.rpm
The following script will set CCE version 16.0.1 default:
/opt/cray/pe/admin-pe/set_default_files/set_default_cce_16.0.1
License:
--------
Except for the third party components and software licensed by HPE through
proprietary agreements, components, files or programs contained within this
package or product are Copyright -2023 Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/cce/16.0.1/ATTRIBUTIONS_16.0.txt
CrayPE 2.7.23
==============
Release Date:
--------------
April 2022
Purpose:
--------
Add support for mixed Programming Environments PrgEnv-cray-amd and
PrgEnv-gnu-amd, to support using the AMD ROCm C/C++ compiler with
the Cray CCE and GNU Fortran compilers, respectively.
Bugs fixed in this release:
---------------------------
- CAST-29420 - Cray Wrappers optimized for Zen2 and not Zen3
Dependencies:
---------------------------------------------
The CrayPE 2.7.23 release is dependent on .pc files in the following
software products:
ATP 1.6.3 or later
FFTW 3.3.0.4 or later
FFTW 2.1.5.6 or later
Global Arrays 5.1.0.2 or later
HDF5 1.8.11 or later
iobuf 2.0.5 or later
LibSci 12.1.01 or later
MPT 6.0.2 or later
NetCDF 4.3.0 or later
Parallel-NetCDF 1.3.1.1 or later
PMI 4.0.1 or later
PETSc 3.4.2.0 or later
Trilinos 11.4.1.0 or later
TPSL 1.3.04 or later
TotalView 8.12-totalview-support-1.1.5 or later
Documentation:
---------------
See manpages for cc, CC, ftn, intro_craype-api, intro_hugepages and pkg-config
See section 2.6 Using Targeting Modules of the Cray Programming Environment
User's Guide (S-2529-116)
See http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Installation instructions:
--------------------------
rpm -ivh craype-2.7.23-202308030531.56cb54942b4bd-0.xc.sles15sp4.x86_64.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_craype_2.7.23
Product description:
--------------------
CrayPE contains drivers, cc, CC, and ftn to compile for the CCE, PGI, GNU,
and Intel Programming Environments.
The craype-* targeting modules are also provided in this product.
Certain components, files or programs contained within this package or product
are Copyright 2013-2022 Hewlett Packard Enterprise Development LP.
Craypkg-gen 1.3.30
==============
Release date
------------
August 2023
Purpose
-------
Bug fix release.
Bugs fixed in this release
--------------------------
- Implement a setuptools build and install for craypkg-gen
- Remove rpmbuild functionality of craypkg-gen
- Switch front end cli from deprecated optparse to ArgumentParser
- Fix PREFIX in craypkg-gen lua modulefile
- Implement better methods for discovering version strings
- Add better support for generating GNU compiler modulefiles
- Add better support for determining compat versions for lmod hierarchy
Dependencies
------------
The craypkg-gen 1.3.30 release is supported on the following HPE
Cray systems:
- Cray EX systems SLE 15.0 or later and RHEL 8.0 or later.
Driver support for integrating Third Party C, C++, and Fortran libraries
through .pc files using pkg-config is used in CrayPE 2.x and later.
Limitation
----------
- Library dependencies for static libraries are not added to the .pc files
for keywords Requires.private and Libs.private. A warning is issued by
the craypkg-gen tool to advise users to add this information to the
libraries .pc files.
- RPM limits packages to 4GB
Documentation
-------------
Man pages for craypkg-gen are found by executing `module load craypkg-gen`
and then `man craypkg-gen`. See
http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Examples for creating modulefiles for Intel, PGI and Python are included in
the craypkg-gen ‘doc’ directory:
/opt/cray/craypkg-gen/1.3.30/doc/intel_example.txt
/opt/cray/craypkg-gen/1.3.30/doc/pgi_example.txt
/opt/cray/craypkg-gen/1.3.30/doc/python_example.txt
Example:
- As an example, the Intel 16.0.3.210 compiler was recently released.
After installing the compiler the administrator creates a modulefile for
this release by executing the following commands:
# module load craypkg-gen
# craypkg-gen -m /opt/intel/compilers_and_libraries_2016.3.210
This version of the Intel compiler is made default by executing the
command:
# /opt/admin-pe/set_default_craypkg/set_default_intel_16.0.3.210
Installation instructions
-------------------------
rpm -ivh craypkg-gen-1.3.30-1.3.30^18^g9d02289-1.sles15sp4.noarch.rpm
To change the product version to default after installation:
/opt/cray/pe/admin-pe/set_default_files/set_default_craypkg-gen_1.3.30
Product description
-------------------
The craypkg-gen 1.3.30 utility provides the system administrator
a tool to integrate third party software with the Cray software stack.
Craypkg-gen assists with integration by creating .pc files for C, C++, and
Fortran libraries, pkg-config enabled modulefiles and RPMs.
The workflow for using craypkg-gen is
1) Build the open source software
2) Create .pc files for libraries
3) Create pkg-config enabled modulefiles
4) Customize pkg-config and modulefiles if needed
Certain components, files or programs contained within this package or product
are Copyright 2013-2023 Hewlett Packard Enterprise Development LP.
FFTW 3.3.10.5
============
Release Date:
-------------
September 2023
Purpose:
--------
This Cray FFTW 3.3.10.5 release is supported on Cray EX (formerly
Shasta) systems. FFTW is supported on the host CPU but not on the
accelerator of Cray systems.
The Cray FFTW 3.3.10.5 release provides the following:
- Compiler and dependency updates
See the Product and OS Dependencies section for details.
Product and OS Dependencies:
----------------------------
The Cray FFTW 3.3.10.5 release is supported on the following Cray systems:
Cray EX Systems with SLES 15 SP3 or later based OS
The FFTW 3.3.10.5 release requires the following software products:
craype 2.7.5 or later
MPT 8.0 or later
One or more of the following major compiler versions:
CCE 15, 16
GCC 10, 11, 12
AOCC 3, 4
Intel 2021 or later
Notes and Limitations:
----------------------
Starting with cray-fftw/3.3.6.1 the fftw module has been renamed from fftw
to cray-fftw.
Documentation:
--------------
http://www.fftw.org/index.html#documentation
See the intro_fftw3 manual page for additional information.
Modulefile:
-----------
module load cray-fftw
Installation
------------
To install the rpm, execute:
rpm -ivh cray-fftw-3.3.10.5-202308072035.1f10cbf899fcf-1.shasta.sles15sp4.x86_64.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_fftw_3.3.10.5
Certain components, files or programs contained within this package or
product are Copyright 2011-2023 Hewlett Packard Enterprise Development LP
Cray LibSci 23.09.1.1
=====================
Release Date:
-------------
September 2023
Purpose:
--------
Cray LibSci 23.09.1.1 provides scientific libraries for Cray
HPC systems. Cray LibSci is supported on the host CPU but
not on the accelerator of these systems.
The Cray LibSci 23.09.1.1 release provides the following:
* misc packaging updates
Cray LibSci 23.09.1.1 includes the following versions of publicly
available libraries:
* LAPACK 3.10.1 - For further information, see
http://www.netlib.org/lapack
* ScaLAPACK 2.2.0 - (Scalable LAPACK) For further information, see
http://www.netlib.org/scalapack.
* QDWH 2.0.0 KSVD 1.0.0 - Polar decomposition and SVD packages, see
https://github.com/ecrc/qdwh
https://github.com/ecrc/ksvd
Product and OS Dependencies:
----------------------------
The Cray LibSci 23.09.1.1 release requires the following:
SLE 15 or RHEL 8
craype/2.7.3 or later
cray-mpich/8.1.0 or later
One or more of the following compiler major versions:
CCE 15.x, 16.x
GCC 10.x, 11.x, 12.x (SLES)
gcc-toolset-10.3.1-1 (RHEL)
AOCC 3.x, 4.x
AMD ROCm 5.x
Intel 2023.x
Nvidia 22.x
Notes and Limitations:
----------------------
OMP threaded versions:
The CrayPE 2.1.2 and later releases add support for link line generation
for the multi-threaded versions of the libsci library based on the OpenMP
flags the user specifies for each compiler:
* CCE by default links to the OpenMP LibSci library. CrayPE will link in
the serial version of LibSci when the CCE flag -hnoomp is used.
* GNU by default links serial LibSci library. CrayPE will link in the
OpenMP version of LibSci when the GNU flag -fopenmp is used.
* INTEL by default links serial LibSci library. CrayPE will link in the
OpenMP version of LibSci when the INTEL flag -qopenmp is used.
Stack size limit:
The 'ulimit -s unlimited' system stack size setting is required for
cray-libsci on all supported Cray platforms. This is typically set as a
system default by admin but may otherwise need to be set at runtime.
QDWH and KSVD:
These packages are currently included for CCE, GNU, and Intel compilers.
Non-default dynamic linking:
When using a non-default version of cray-libsci with dynamic linking
users should set at runtime and at linktime after loading the desired
cray-libsci module--or set the equivalent for their linux shell:
export LD_LIBRARY_PATH=$CRAY_LD_LIBRARY_PATH:$LD_LIBRARY_PATH
Documentation:
--------------
See the intro_libsci man page for additional information.
See the csmlversion man page for information to display version information
on the currently loaded scientific libraries.
Modulefile:
-----------
module load cray-libsci
Installation instructions:
--------------------------
LibSci is now packaged into separate compiler specific RPMs to allow
rpmbuild to correctly include compiler dependencies.
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs
contained within this package or product are Copyright 2001-2023
Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in this
package are detailed in the file:
/opt/cray/libsci/23.09.1.1/ATTRIBUTIONS_libsci23.09.1.1.txt
LibSci_ACC 23.09.1.1
====================
Release Date:
-------------
September 2023
Purpose and Misc:
-----------------
The Cray LibSci_ACC 23.09.1.1 release provides accelerated versions of
scientific libraries for Cray systems with AMD MI100 or MI200 targets.
Product and OS Dependencies:
----------------------------
Cray LibSci_ACC 23.09.1.1 requires the following platform:
HPE Cray EX system with AMD MI100 or MI200 support
Cray LibSci_ACC 23.09.1.1 also requires:
For SLES rpms:
CCE 14.0 or later
AMD ROCm 5.0 or later
Cray MPICH 8.1.15 or later
Cray LIBSCI 22.08 or later
For RHEL rpms:
CCE 16.0 or later
AMD ROCm 5.0 or later
Cray MPICH 8.1.26 or later
Cray LIBSCI 23.05 or later
Notes and Limitations:
----------------------
Cray LibSci_ACC 23.09.1.1 supports AMD MI100 and MI200 targets.
LAPACK workspace calculations from other implementations or hard-coded
values may not be compatible with libsci_acc. Workspaces reported from
a query to the corresponding libsci_acc function should be used.
The current supported usecase is a one-rank to one-gpu mapping for apps.
Performance improvements can be achieved in programs calling libsci_acc
subroutines by using pinned memory. See the intro_libsci_acc man page for
details.
Use of the aprun option "-cc none" or the srun option "--cpu_bind=none"
to disable core affinity is strongly suggested to maintain performance.
When using non-default cray-libsci_acc versions users should set equivalent
at runtime and linktime after loading the desired cray-libsci_acc module:
export LD_LIBRARY_PATH=$CRAY_LD_LIBRARY_PATH:$LD_LIBRARY_PATH
Documentation:
--------------------------------
See the intro_libsci_acc man page for additional information.
Modulefile:
-----------
module load cray-libsci_acc
Sample Installation instructions:
---------------------------------
Rpm install command:
rpm -ivh cray-libsci-acc-23.09.1.1-*.x86_64.rpm
The "*" in the install command represents compiler version combinations.
To make this the default version:
/opt/cray/pe/admin-pe/set_default_files/set_default_libsci_acc_23.09.1.1
License:
--------
Except for the third party modules and software licensed by HPE
through proprietary agreements, components, files or programs
contained within this package or product are Copyright 2022 Hewlett
Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in
this package are detailed in the file:
/opt/cray/pe/libsci_acc/23.09.1.1/ATTRIBUTIONS_libsci_acc*.txt
Cray MPICH 8.1.27:
=======================================
Release Date:
-------------
August 11, 2023
Purpose:
--------
Cray MPICH 8.1.27 is based upon ANL MPICH 3.4a2 with support for libfabric
and is optimized for the Cray Programming Environment.
Major Differences Cray MPICH 8.1.27 from the XC Cray MPICH include:
- Uses the new ANL MPICH CH4 code path and libfabric for network
support.
- Does not support -default64 mode for Fortran
- Does not support C++ language bindings
New Cray MPICH features for HPE Cray EX and Apollo systems:
- Starting from the 8.1.26 release, Cray MPICH supports the Intel Sapphire Rapids CPU HBM
processor architecture.
- On systems with AMD GPUs, Cray MPICH 8.1.26 supports all ROCm
versions starting from ROCm 5.0, including the latest ROCm 5.5.0
release.
The Cray MPICH 8.1.25 release and prior versions of
Cray MPICH are only compatible with ROCm versions up to (and
including) the ROCm 5.4.0 release.
- Cray MPICH uses the libfabric "verbs;ofi_rxm" provider by default.
This is the supported and optimized OFI libfabric provider for
Slingshot-10 and Apollo systems.
- Cray MPICH offers support for multiple NICs per node. Starting with
version 8.0.8, by default Cray MPICH will use all available NICs on
a node. Several rank-to-NIC assignment policies are supported. For
details on choosing a policy for assigning ranks to NICS, or for
selecting a subset of available NICs, please see the following
environment variables documented in the mpi man page.
MPICH_OFI_NIC_VERBOSE
MPICH_OFI_NIC_POLICY
MPICH_OFI_NIC_MAPPING
MPICH_OFI_NUM_NICS
- Enhancements to the MPICH_OFI_NIC_POLICY NUMA mode have been added.
Starting with version 8.0.14, if the user selects the NUMA policy,
the NIC closest to the rank is selected. A NIC no longer needs to
reside in the same numa node as the rank. If multiple NICs are
assigned to the same numa node, the local ranks will round-robin
between them. Numa distances are analyzed to select the closest NIC.
- Cray MPICH supports creating a full connection grid during MPI_Init.
By default, OFI connections between ranks are set up on demand. This
allows for optimal performance while minimizing memory requirements.
However, for jobs requiring an all-to-all communication pattern, it
may be beneficial to create all OFI connections in a coordinated
manner at startup. See the MPICH_OFI_STARTUP_CONNECT description in
the mpi man page.
- Cray MPICH supports runtime switching to the UCX netmod starting
with version 8.0.14. To do this load the craype-network-ucx module
and module swap between Cray-MPICH and Cray-MPICH-UCX modules. For
more information including relevant environment variables reference
the intro_mpi man page with the Cray-MPICH-UCX module loaded.
- Lmod support for HPE Cray EX starting with Cray MPICH 8.0.16.
Key Changes and Bugs Closed:
----------------------------
Changes in Cray MPICH 8.1.27
- CAST-29466 - Bugfix for illegal Barrier and Bcast interaction
- CAST-32622 - New ROMIO hint overstriping_factor to support Lustre over striping
- CAST-32717 - Retain rank to mr_key mapping for MR_SCALABLE
- CAST-32814 - Support checking of global config file
- CAST-32861 - Correct MANPATH in cray-mpich-abi module
- CAST-33213 - Fix incorrect comm usage with MPICH_MPIIO_STATS
- CAST-33226 - Fix for Iprobe to allow FI_EAGAIN return value with the CXI provider
- CAST-33323 - Fix allreduce correctness with large payloads
- PE-43686 - Correct PE_PERFTOOLS_MPICH_LIBDIR in cray-mpich-ucx module
- PE-48333 - Add info about singleton execution to intro_mpi man page
- PE-48931 - Use MPICHALLTOALLV_THROTTLE in MPI_alltoallw, MPI_Ialltoallv and MPI_Ialltoallw algorithms
- PEEF-2327 - Fix MPICH_NO_LOCAL to correctly honor MPICH_OFI_NUM_NIC selection
Product and OS Dependencies:
----------------------------
The Cray MPICH 8.1.27 release is supported on the following HPE systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Product and OS Dependencies by network type:
--------------------------------------------------+
| Shasta |
----------------------------+---------------------+
craype | >= 2.7.6 |
----------------------------+---------------------+
cray-pals | >= 1.0.6 |
----------------------------+---------------------+
cray-pmi | >= 6.0.1 |
----------------------------+---------------------+
libfabric | >= 1.9.0 |
----------------------------+---------------------+
One or more compilers:
* AMD ROCM 5.0 or later
* AOCC 3.0 or later
* CCE 14.0 or later
* GNU 9.1 or later
* Intel 2022.1 or later
* Nvidia 20.7 or later
Notes and Limitations:
----------------------
Limitations in Cray MPICH 8.1.27:
- Cray MPICH 8.1.27 can support only ~2040 simultaneous MPI
communicators. This limit is less the XC Cray MPICH limit of
~4090 simultaneous communicators. Cray intends to raise the
limit in a future release of Cray MPICH for Shasta to at least
the XC limit.
Documentation:
--------------
For more information see the intro_mpi man page.
Modulefile:
-----------
module load cray-mpich/8.1.27
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright -2021 Hewlett Packard
Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/mpich/8.1.27/ATTRIBUTIONS
Copyright -2023 Hewlett Packard Enterprise Development LP
Copyright 2022-2023 Hewlett Packard Enterprise Development LP
Product Release Information
================================================================================
Product identification:
-----------------------
HPE Cray MPIxlate: version 1.0.2 (built on 2023-08-08T02:05:11Z from 96dbe47)
Purpose:
--------
HPE Cray MPIxlate enables applications compiled using an MPI library that
is not binary compatible with HPE Cray MPI, to be run without recompilation
on supported HPE platforms.
Warnings & Limitations:
-----------------------
- See NOTES in mpixlate(1)
Changes from previous release:
------------------------------
- New features:
* None
- Defects Fixed:
* CAST-33166: Translate MPI_NO_OP (and MPI_REPLACE) from Open MPI
Platform, Operating System and Product dependencies:
----------------------------------------------------
Supported HPE platforms:
- HPE Cray EX systems with HPCM or CSM
Supported Operating Systems:
- SLES 15 SP4
- COS 2.4.X/COS 2.5
Product dependencies:
-----------------+----------
cray-mpich[-ucx] | >= 8.1.27
-----------------+----------
Documentation:
--------------
mpixlate(1)
Environment Modules
-------------------
module load cray-mpixlate/1.0.2
================================================================================
cray-papi
=================
Release date:
-------------
September 2023
Package:
--------
cray-papi-7.0.1.1-202307051905.965b813db6366-2 x86_64
Purpose:
--------
New version of papi 7.0.1.1 release 2
Documentation:
--------------
Overview: https://github.com/icl-utk-edu/papi
Web Site: https://icl.utk.edu/papi
Product description:
--------------------
PAPI aims to provide the tool designer and application engineer with a
consistent interface and methodology for use of the performance counter
hardware found in most major microprocessors. PAPI enables software
engineers to see, in near real time, the relation between software
performance and processor events.
Dependencies:
-------------
For a list of software used when validating this version of
PAPI on Cray and HPE systems, see the HPE Cray Programming
Environment release announcements.
Copyright 2015-2017,2019-2023 Hewlett Packard Enterprise Development LP
Perftools 23.09.0
===============
Release Date: September, 2023
Purpose:
========
This is a feature and bugfix release for the following systems:
- HPE Cray EX and HPE Cray Supercomputer Systems with HPCM
- HPE Cray EX and HPE Cray Supercomputer Systems with CSM
- HPE Apollo 2000 Gen 10 Plus Systems (x86)
Key enhancements or changes from the previous release:
======================================================
o Upgrade to DWARF 0.7.0
o Upgrade to ELF 0.8.16
o Upgrade to PAPI 7.0.1.1
- Upstream bugfixes
- Built with ROCm 5.5.1 and CUDA 12.0
- Configured with upstream nvml and rocm_smi components
o Introduced GPU programming API major version checks
o Improved presentation of accelerator function table
o Added support for push/pop regions in sampling mode
o Update CUDA trace wrappers to support up to CUDA 12.2
o Enable PAPI GPU management components, namely nvml & rocm_smi
o Improve Sapphire Rapids counter groups and derived metrics
o Set libunwind as the default callstack resolution method
o Remove tracing support for MPI C++ bindings
o Remove tracing support for Charm++
o Remove support for Aries NIC
o Remove support for outdated CPUs, namely Intel's Sandy/Ivy Bridge, Haswell,
Broadwell, and AMD's Opteron
o Remove support for Multi-Channel DRAM
Bugs Fixed:
===========
o Program hang when using perftools-lite with libunwind
o Spurious API warnings
o Appearance of undesired internal function calls in calltrees
o Inconsistency in number of OMPT calls when using differing callstack resolution modes
o Erroneous call counts for some untraced functions
o Erroneous overhead values for some libunwind experiments
o Inconsistency in some reported HWPC values in full trace mode
o An issue using performance counter files with multiple processes
o Incorrect Cray Cassini HWPC descriptions
o App2 error when loading MPI message traffic
o App2 failure on certain systems when viewing full trace data on a
remote server
Notes:
======
o This release of Perftools only supports ROCm 5.5.1
o This release of Perftools only supports CUDA 12.0 and later
Known Issues:
=============
o Fortran applications compiled against CCE 16 may fail during instrumentation
with the following error message: "hidden symbol `<SYMBOL>' in <LIB> is
referenced by DSO". To work around this error specify the following during
the pat_build invocation '-Dlink-instr=-lgcc_s'.
o When PAT_RT_CALLSTACK_MODE is unset or set to 'unwind', deadlocks have been
observed intermittently during startup of certain GPU-based codes. Rerunning
the program generally resolves the issue. Alternatively, set
PAT_RT_CALLSTACK_MODE to 'hybrid' (the default in previous releases) or 'frames'
if the libunwind functionality is not required.
o Presently the DWARF issued by AMD codegen for optimized code is not fully
supported. To trace user-defined functions when compiling with CCE for AMD GPU
targets, and using an optimization level higher than `-O0`, users must specify the
-h func_trace (ftn) or -finstrument-functions (C, C++) compiler options to enable
and use with pat_build -w.
o OpenACC applications may abort during runtime while performing tracing experiments.
To work around this error set PAT_RT_THREAD_JOIN_WAIT to '-1'.
Dependencies:
=============
o A PDF reader (such as evince, acroread or okular) is required to use the
app2 online help.
o For a list of software used when validating this version of
Perftools on Cray and HPE systems, see the HPE Cray Programming Environment
release announcements.
Installation instructions:
==========================
For Apollo 80 systems, please see the HPE Cray Programming Environments
Installation Guide: Apollo 80 (ARM) System (S-8013).
For Apollo 2000 systems, please see the HPE Cray Programming Environments
Installation Guide: Apollo 2000 Gen10 Plus (x86) System (S-8012).
For HPE Cray EX (Shasta Architecture) systems, refer to the Cray Asynchronous
Installer Guide (S-8003).
Installation of app2 remote client (+ server) on Mac systems:
-------------------------------------------------------------
Apprentice2Installer-23.09.0.dmg
RevealInstaller-23.09.0.dmg
The Cray Apprentice2 and Reveal installers for Mac are included in the
perftools-clients rpm, and placed in
$CRAYPAT_ROOT/share/desktop_installers/
on a Cray Shasta User Access Node (UAN) or on a Cray login node when the
perftools software is installed.
Download the Cray Apprentice2 installer onto a desktop or laptop
running Mac OS El Capitan through Mojave. Double click on installer
to begin installation. The installer will walk you through the
process for your system.
Installation of app2 remote client (+ server) on Windows 10 systems:
--------------------------------------------------------------------
Apprentice2Installer-23.09.0.exe
The Cray Apprentice2 installer for Windows is included in the
perftools-clients rpm, and placed in
$CRAYPAT_ROOT/share/desktop_installers/
on a Cray Shasta User Access Node (UAN) or on a Cray login node when the
perftools software is installed.
Download the Cray Apprentice2 installer onto a desktop or laptop
running Windows 10. Double click on installer to begin installation.
The installer will walk you through the process for your system.
Documentation:
==============
See the following man pages:
app2, grid_order, intro_craypat, pat_build, pat_help, pat_info, pat_opts,
pat_report, pat_run, pat_view, reveal
intro_papi
perftools-base, perftools-lite, perftools-preload
accpc, cray_cassini, cray_pm, hwpc, nwpc, papi_counters, uncore
PAPIlicnotices
Search for perftools on https://support.hpe.com to access documentation
on Cray Performance Measurement and Analysis Tools
License:
========
Except for the third party modules discussed below and software licensed
by HPE through proprietary agreements, components, files or programs
contained within this package or product are Copyright 2001-2023
Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in this
package are detailed in the file:
$CRAYPAT_ROOT/ATTRIBUTIONS_perftools.txt
Cray PMI 6.1.12