TIMPI/NEWS at master · libMesh/TIMPI · GitHub

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
0.0.0
  * Made TIMPI code independent of libMesh

0.1.0
  * Split TIMPI into a separate repository

1.0.0
  * automake fixups
  * Nonblocking vec<vec> receives + syncs
  * MPI detection round 2
  * Add Communicator::max(map)
  * export TIMPI_RUN in Travis MATRIX_EVAL
  * Add identical_sizes argument to broadcast
  * Make sure that we can test OPROF and PROF methods
  * Add map_sum() specialization for non*fixed types
  * Add back old map broadcast method
  * Parallel testing
  * Add Communicator::broadcast(unordered_map) specialization
  * Communicator::broadcast(map): Save one communication
  * Add Communicator::sum(std::unordered_map)
  * Add support for a timpi*config
  * Add Communicator::sum(std::map)
  * Unit test, compilation fixes
  * Dispatch to allgather_packed_range from allgather for dynamic size types
  * Remove configure output, temp file
  * Added .travis.yml CI tests

1.1.0
  * broadcast now autodispatches to packed_range when necessary
  * Recursive specializations for Packing<std::pair>
  * configure bugfix for when environment sets LIBS, CPPFLAGS

1.1.1
  * autoconf m4 refactoring

1.1.2
  * Fix for push_parallel_packed_range

1.1.3
  * Warnings fixes in push_parallel_packed_range
  * More container support in push_parallel_packed_range
  * push_parallel_packed_range is no longer timpi_experimental
  * More unit tests
  * Better buffer size control in more packed_range functions
  * Reuse derived type object in push_parallel_vector_data
  * Have bootstrap check for autoconf submodule initialization

1.2
  * Directly include header files in parallel_sync.h
  * Use default constructor/destructor/assignment in Status, DataType
  * Oversized pull responses bugfix
  * Clarified Communicator::duplicate() doc comment

1.2.1
  * Use "-f" when using mv to overwrite our own files

1.3
  * More dispatches to packed range; displace is_fixed_type idiom
  * Add nonblocking min, max, and sum

1.4
  * Explicitly instantiate DataType superclass for tuple
  * Add -Wcast-align to clang ignore_warnings.h
  * Ensure methods with SFINAE abling can be used in serial
  * Support build_standard_type w/ user allocators
  * Quad-precision support

1.5
  * Move Packing<string> into sources
  * Packing<tuple> specialization
  * StandardType<std::array> support
  * Packing<array> support
  * Packing<list, map, unordered_set, etc>

1.6
  * Generalize push_parallel_[packed_range/vector_data] to allow data moves

1.7
  * Packing<vector> support

1.7.1
  * Packing size fixes
  * Update autoconf submodule
  * Make DataType destructor virtual to avoid leaks

1.7.2
  * Implement copy assignment operators for StandardType

1.8.0
  * Expanded OpFunction capabilities
  * Use SemiPermanent for derived StandardType classes
  * Update compiler requirements from C++11 to C++14
  * Code coverage support

1.8.1
  * Simplify logic in parallel_sync (fixing oversized pulls)

1.8.2
  * Fix new bug in parallel_sync (possible vector<vector> corruption)

1.8.3
  * Prevent gcc spurious warnings in unit test
  * Add Intel(R) icc 21 support
  * Avoid overzealous gcc memcpy warnings, do our own checks

1.8.4
  * Avoid our *own* overzealous checks before memcpy
  * Dispatch to packed range for gather of dynamic size types

1.8.5
  * Fixes for packed range specializations
  * Avoid overwriting user-specified $LIBS
  * More parallel unit testing
  * Disable overzealous compiler warning

1.9.0
  * Packing<string> now uses unsigned int buffer_type too, for better
    compatibility in nested Packing types
  * More unit test coverage, including for packed ranges, vector<bool>
    specializations, allgather<vector<vector>>
  * Auto-dispatch from parallel push to packed_range
  * Multiset (multimap, etc) support in set_union()
  * Break out Packing declarations in their own header
  * Many assorted compiler warning and bug fixes
  * Wider compiler support
  * Better error messages
  * More assertions
  * Clarified comments
  * Autoconf submodule updates
  * Support for 64-bit counts when MPI-4 is available
  * Enabled `MPI_THREAD_SERIALIZED` and `MPI_THREAD_MULTIPLE`
  * More efficient parallel_sync with rvalue references
  * Non-Blocking eXchange algorithm improvements.
  * Options to use older parallel synch algorithms in lieu of
    Non-Blocking eXchange.
  * Default to FPE-safe flags when building