Skip to content

Commit ee129f3

Browse files
rbuchmatthiasdienerminitu
authored
Add 6.11.0 release notes (#3106)
Co-authored-by: Matthias Diener <[email protected]> Co-authored-by: Jaemin Choi <[email protected]>
1 parent d87c2c1 commit ee129f3

File tree

1 file changed

+108
-0
lines changed

1 file changed

+108
-0
lines changed

CHANGES

Lines changed: 108 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,114 @@
11
This file describes the most significant changes. For more detail, use
22
'git log' on a clone of the charm repository.
33

4+
================================================================================
5+
What's new in Charm++ 6.11.0
6+
================================================================================
7+
8+
This is a feature release, with the following major changes:
9+
10+
Highlights:
11+
12+
- A significant overhaul of the load balancing infrastructure and the
13+
addition of TreeLB, a new, more flexible and extensible base LB,
14+
intended to replace the previous CentralLB and HybridBaseLB designs
15+
(load balancers using these previous designs are still
16+
supported). Some included strategies have been rewritten to use
17+
TreeLB, unused and extraneous strategies have been removed. Please
18+
see
19+
https://charm.readthedocs.io/en/latest/charm++/manual.html#load-balancing
20+
for more information.
21+
22+
- Experimental support for intra-node GPU messages, allowing data to be
23+
sent to/from GPU memory without going through host memory. Please
24+
see
25+
https://charm.readthedocs.io/en/latest/charm++/manual.html#gpu-support for
26+
more information.
27+
28+
Misc:
29+
30+
- Charm++ now builds with CMake by default when using the `./build`
31+
script (requires CMake version 3.4 or higher). The old build system is
32+
still available by using `./buildold`. Please see
33+
https://charm.readthedocs.io/en/latest/charm++/manual.html#installation-with-cmake
34+
for more information.
35+
36+
- We have adopted GitHub discussions
37+
(https://github.com/UIUC-PPL/charm/discussions) as the preferred
38+
venue for Charm++ questions and discussions instead of the
39+
[email protected] mailing list.
40+
41+
- The BigSim emulator has been removed from Charm++.
42+
43+
Charm++ Features & Fixes:
44+
45+
- Support for variable sized messages in TRAM.
46+
47+
- Added a pup_buffer API with zero copy functionality.
48+
49+
- Array broadcasts are now node aware, avoiding unnecessary
50+
duplication of messages. Expedited nokeep array broadcasts are also
51+
allowed now.
52+
53+
- Added CmiNodeReduce API for Converse node level reductions.
54+
55+
- Builds with CMK_OPTIMIZE=1 will only segfault to aid in debugging in
56+
CmiAbort when ++debug or +truecrash are provided.
57+
58+
- Callbacks used in liveViz have been made ASLR safe.
59+
60+
- Updated implementation of atomics, locks, and fences to use C++11/C11
61+
versions where suitable.
62+
63+
- Fixed bugs in HAPI and updated implementation to use new CUDA APIs.
64+
65+
- Added whenidle attribute to indicate entry method to be called when
66+
a PE is idle.
67+
68+
- Improved performance, support, and fixes for UCX. IBM Power is now
69+
supported with UCX.
70+
71+
- The CmiSyncSend family now tries to avoid copies for nokeep
72+
messages.
73+
74+
- Fixed bug with element IDs with CMK_GLOBAL_LOCATION_UPDATE.
75+
76+
- Fixed bug in BlockMap in array creation.
77+
78+
- Added CcdPROCESSOR_LONG_IDLE to run a function during long periods
79+
of idleness and CcdSCHEDLOOP to run a function during every
80+
scheduler loop.
81+
82+
- Added isCheckpoint() and isMigration() methods for users to
83+
condition logic inside PUP based on its purpose.
84+
85+
- Added execution metadata to Projections logs.
86+
87+
- Added several new benchmarks and tests.
88+
89+
- Bounded number of hops messages can take when chares frequently
90+
migrate.
91+
92+
Adaptive MPI:
93+
94+
- Isomalloc sync is now enabled by default and the implementation no
95+
longer uses the filesystem to pass data. Isomalloc sync can be
96+
disabled with +no_isomalloc_sync.
97+
98+
- Added AMPI-only build target, AMPI-only. This build optimizes AMPI
99+
by disabling features of Charm++ that AMPI doesn't use.
100+
101+
- Fixed tlsglobals migration on macOS.
102+
103+
- Fixed bugs in activation of migration callbacks.
104+
105+
- Fixed a bug in MPI_Waitsome.
106+
107+
- Accept arguments from AMPI_BUILD_FLAGS environment variable for
108+
ampiCC.
109+
110+
- Improved portability of fsglobals and pipglobals.
111+
4112
================================================================================
5113
What's new in Charm++ 6.10.2
6114
================================================================================

0 commit comments

Comments
 (0)