Repository navigation
Expand file tree
/
Copy pathChangeLog
More file actions
160 lines (105 loc) · 4.69 KB
/
Copy pathChangeLog
File metadata and controls
160 lines (105 loc) · 4.69 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
2026-07-07 Kim Walisch <kim.walisch@gmail.com>
Version 4.2
* `libpopcnt.h`: Improve tail bytes perf.
* `libpopcnt.h`: Up to 3x faster popcount for medium arrays (96 bytes to 1 KB)
using a simple AVX2 popcnt256 loop instead of scalar POPCNT or Harley-Seal.
* `benchmark.yml`: Add automated benchmark CI.
2026-07-01 Kim Walisch <kim.walisch@gmail.com>
Version 4.1
* `libpopcnt.h`: Improve ARM NEON performance by up to 2x.
* `libpopcnt.h`: Improve ARM SVE performance by up to 2x.
* `libpopcnt.h`: Up to 10% faster AVX512 algorithm using Clang compiler.
2026-06-27 Kim Walisch <kim.walisch@gmail.com>
Version 4.0
* `libpopcnt.h`: Add ARM SVE runtime dispatching for Linux and Windows.
* `libpopcnt.h`: Enable AVX2/AVX512 runtime dispatching for clang-cl >= 22 on Windows.
* `libpopcnt.h`: Fix undefined behavior caused by incorrect casts.
* `libpopcnt.h`: Fix potential threading data race in CPUID initialization code.
* `libpopcnt.h`: Fix inaccurate AVX2 array size comment (#20).
* `libpopcnt.h`: Fix MSVC C4057 warning in CPUID initialization code.
* `benchmark.cpp`: Add ARM SVE runtime detection.
* `CMakeLists.txt`: Increase minimum CMake version to 3.9.
* `ci.yml`: Fix hanging test.
* `ci.yml`: Add FreeBSD CI test.
2024-06-29 Kim Walisch <kim.walisch@gmail.com>
Version 3.1
* Improve AVX512 algorithm for trailing 64 bytes.
* AVX512 algorithm does not require AVX512-BITALG extension anymore.
2024-06-27 Kim Walisch <kim.walisch@gmail.com>
Version 3.0
* Add ARM SVE algorithm.
* Replace AVX512BW algorithm by faster AVX512 VPOPCNTDQ algorithm.
* Add MSVC support for ARM NEON.
* Improve preprocessor checks using __has_include() macro.
* Port tests from AppVeyor to GitHub actions.
* Get rid of unaligned uint64_t memory acceses, this fixes
test failures when using GCC compiler sanitizers.
* Prefix all libpopcnt macros using LIBPOPCNT_ to avoid any naming collisions.
2020-08-10 Kim Walisch <kim.walisch@gmail.com>
* Use unaligned memory accesses to improve performance.
Aligning memory causes branch mispredictions which
can significantly deteriorate performance especially
for small array sizes.
* On x86/x64 runtime checks are now removed if the user
compiles his code with e.g. -mpopcnt, -mavx2,
-march=native, ...
2020-07-19 Kim Walisch <kim.walisch@gmail.com>
* Enable AVX2, AVX512 by default for MSVC.
Thanks to KOConchobhair for the pull request #15.
2019-12-31 Kim Walisch <kim.walisch@gmail.com>
Version 2.3 released.
* Up to 10% speedup on ARM NEON.
* Fix unaligned memory access on ARM.
2018-01-13 Kim Walisch <kim.walisch@gmail.com>
Version 2.2 released.
* Up to 6x faster on old x86 CPUs without POPCNT.
2017-11-10 Kim Walisch <kim.walisch@gmail.com>
Version 2.1 released.
* Up to 20% ARM NEON speedup.
* test/test1.cpp: Add only 1 bits test case.
* test/test2.c: Add only 1 bits test case.
2017-10-27 Kim Walisch <kim.walisch@gmail.com>
Version 2.0 released.
* Add AVX512 support.
* Support AVX2 and AVX512 for the MSVC compiler.
* Support CPUID for MSVC and the C programming language.
2017-09-09 Kim Walisch <kim.walisch@gmail.com>
Version 1.9 released.
* libpopcnt.h: Fix MSVC POPCNT detection.
2017-04-24 Kim Walisch <kim.walisch@gmail.com>
Version 1.8 released.
* Fixed "illegal instruction" runtime crash with GCC 4.4.7.
* benchmark.cpp: Print statistics and algorithm name.
* Reduce CMake minimum version to 2.8.
2017-04-08 Kim Walisch <kim.walisch@gmail.com>
Version 1.7 released.
* Refactor ARM NEON popcount algorithm.
2017-04-04 Kim Walisch <kim.walisch@gmail.com>
Version 1.6 released.
* Add ARM NEON popcount algorithm.
* Fix x86 segmentation fault for CPUs without POPCNT.
2017-04-02 Kim Walisch <kim.walisch@gmail.com>
Version 1.5 released.
* libpopcnt.h now supports C and C++!
* Use CMake build system.
* test2.c: Add C test.
* Fix clang-cl bug on Windows.
2017-04-01 Kim Walisch <kim.walisch@gmail.com>
Version 1.4 released.
* libpopcnt.h: Fix compiler warning.
2017-03-30 Kim Walisch <kim.walisch@gmail.com>
Version 1.3 released.
libpopcnt.h is now C++ only (previously C/C++), the reason being
that the cpuid check cannot be made thread-safe using plain C,
whereas in C++ its trivial (Meyers Singleton).
* Add benchmark.cpp.
* Use AVX2 for arrays >= 512 bytes (previously 1024).
* Improve Makefile.
2017-03-26 Kim Walisch <kim.walisch@gmail.com>
Version 1.2 released.
* Add cpuid check for x86 CPUs.
* Compiles without -mpopcnt, -mavx2 flags.
* Successfully tested on IBM POWER8 (generates popcntd, GCC 5.4).
* Successfully tested using clang-cl (Windows).
* Add ChangeLog.
* Update README.md.