forked from dividiti/ck-caffe
-
Notifications
You must be signed in to change notification settings - Fork 0
/
CHANGES
155 lines (94 loc) · 6.69 KB
/
CHANGES
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
2018.06.11 - [FGG] fixed LMDB package v0.9.21 on Windows.
See https://github.com/BVLC/caffe/issues/5631
2018.06.07 - [FGG] added CYGWIN path for BOOST include on Windows
(update all repos and run "ck refresh env" before using)
2018.05.25 - [FGG] added experimental Caffe build via spack - currently doesn't work
2018.05.14 - [FGG] moved imagenet soft/packages for imagenet (aux,val) to ck-env
to be able to reuse across different AI frameworks (TF, Caffe, Caffe2, MXNet, CNTK
2018.03.13 - [LG] using new update_tags_by_target_os_uos{} feature from ck-env for MacOSX support
2018.03.09 - added support for latest protobuf, boost 1.65.x
- improved unification (started adding custom.py to some Caffe packages)
- improved build on Windows with latest visual studio
- added "ck virtual env" for Python-based Caffe built via CK
- improved cross-compilation for CK CUDA clients
2018.03.06 - [LG] make LevelDB only a Linux+MacOSX dependency (MacOSX support)
2018.03.05 - [LG] adding the explicit dependency of Caffe on LevelDB library (MacOSX support)
2018.02.21 - [LG] make sure we pass CK_RANLIB_PATH_FOR_CMAKE to cmake (to make Intel Caffe compilable with ICC)
2017.11.29 - fixing Caffe compliation of time client on Windows (#121)
2017.11.20 - improved Caffe OpenCL/CUDA installation on Windows
2017.11.11 - improved Caffe compilation with CLBlast 1.2.0 and Lib DNN across all platforms
2017.10.20 - added Caffe ARMCL branch (not yet completely working)
2017.10.14 - added stable Caffe version (CPU) with all dependencies compiled by the CK
- fixed compilation on Windows and Android
- added channel-stable tag
2017.10.13 - started adding stable Caffe packages (snapshots and fixed CK dependencies)
2017.10.05 - added check on OpenCL device to remind user to disable unified memory
(useful for embedded GPU)
2017.10.04 - moved non-universal packages to ck-old-packages (you can get it via ck pull repo:ck-old-packages)
- forcing Boost to be installed from package (with version >= 1.64.0)
2017.06.11 - moved outdated ck-caffe:package:lib-caffe-nvidia-fp16-cuda & ck-caffe:package:lib-caffe-nvidia-fp16-cudnn to ck-old-packages
2017.06.01 - fixed Caffe soft description on Windows (now detects both proto.lib or recently changed caffeproto.lib)
2017.05.22 - improved Caffe library detection on non-standard platforms (lib64 vs lib)
2017.05.18 - added customizable autotuning example (script:explore-batch-size-unified-and-customized)
2017.04.28 - added support to create binary package distro (particularly for Windows crowd-benchmarking)
2017.04.25 - added support for Python (compile universal package with --env.CAFFE_BUILD_PYTHON=ON
and then use it via "ck xset env tags=lib,caffe && . ./tmp-ck-env.bat && ipython2")
2017.04.23 - Caffe crowd-benchmarking now support Android mobiles connected via adb
2017.04.16 - removed outdated experiment.bench.caffe.mobile module (now using generic experiment.bench.dnn.mobile)
- fixed caffe desktop crowd-tuning (still requires major update)
2017.04.05 - improving OpenCV detection
2017.03.25 - fixes for Raspberry Pi 3
- updating deps in all universal packages (cmake + compiler)
2017.03.18 - fixes for Android install scripts (to support latest 3.7.2 cmake)
2017.03.14 - moved package:dnn-* to ck-old-packages repo
2017.03.12 - added 2 stripped IMAGENET packages with 500 images (to avoid download 8GB just for testing purposed):
* ck install package:imagenet-2012-val-min
* ck install package:imagenet-2012-val-min-stripped
2017.03.02 - updated generation of DNN Android scenarios
2017.03.01 - moved Caffe CPU branch for Android to official OpenCL branch with CPU_ONLY
(the reason is that we added all our latest Android changes there)
2017.02.28 - fixed Caffe package with CUDA/CUDNN deps for Windows and Linux
2017.02.01 - added --env.CK_VIENNACL_DEBUG=ON flag to turn on debugging of ViennaCL
2017.01.20 - unifying compilation of Caffe OpenCL for Android on Windows and Linux
2017.01.19 - cleaned up various packages
2017.01.18 - added Caffe OpenCL package for windows (lib-caffe-bvlc-opencl-viennacl-win)
2017.01.10 - added caffe-time-cuda and caffe-classification-cuda
2017.01.08 - major update to support Windows + Android
2017.01.02 - fixed parsing of output for Linux/Android/Windows
- skip downloading and installing glog,gflags,protobuf if already installed
2016.12.30 - cleaning up classification and caffe-time for Android
2016.12.29 - major update to compile Caffe for Android via CK
- proper support for various ARM features (fpsoft/fphard/neon)
2016.12.26 - added support to compile Caffe lib for Windows (can be improved)
- added support to compile Caffe lib for Android (official benchmarking)
2016.12.21 - major update (added new run-time SW deps handling)
2016.12.20 - major update of pruning command lines by deps and vice versa ...
2016.12.05 - improved Caffe and Classification meta (including version)
- added automatic generation and recording of libcaffe.so and cliassification via CK
2016.11.30 - added 'no_compile' to caffe meta
- fixed total execution time in CK
2016.11.11 - added Caffe CPU build for Android via CK
- force to use packages when compiling for Android
(i.e. not to search for already installed soft
for Android not to mix up with Linux versions)
- fixed classification for Android and added OpenME
2016.11.08 - major update to caffe-classification with the support
for CK jpeg datasets;
- added standard DNN demo datasets (cat, fish-bike) to ctuning-datasets-min repo
- can use different models in caffe-classification
2016.11.04 - making os.getcwd() safer
2016.10.24 - added support for Caffe mobile crowd-tuning
- added caffe-classification program
2016.10.21 - started preparing mobile Caffe crowd-tuning
2016.10.18 - first variant of dashboard for crowd-benchmarking and crowd-tuning
2016.10.17 - added crowd-benchmarking (ck crowdbench caffe)
2016.09.09 - updated DVDT OpenCL profiler soft package
2016.07.18 * added batch size autotuning and graphs
2016.07.12 * many updates in packaging, customization, model fixes, etc
* added env CK_CAFFE_BATCH_SIZE and CK_CAFFE_ITERATIONS for DNN autotuning and crowdtuning
2016.07.03 * fixing auto-detection of CAFFE + do not install package if already installed
2016.06.22 * major update (packages + program + dataset)
* opencl-based caffe is working now
2016.06.21 * moving some math libs to ck-math repo
2016.06.18 * fixing ViennaCL for Linux
2016.06.06 * first CK commit