Permalink

Comparing changes

Choose two branches to see what’s changed or to start a new pull request. If you need to, you can also .

Open a pull request

Create a new pull request by comparing changes across two branches. If you need to, you can also .
Choose a Base Repository
BVLC/caffe
01org/caffe
123gpg321/caffe
71squared/caffe
AMDComputeLibraries/caffe
AdamStelmaszczyk/caffe
Adnan1011/NR-IQA-CNN
AkiChen/Correlative-Filters-Caffe
AlfredXiangWu/caffe
Austriker/caffe
BlGene/caffe
CUHK-MMLAB/caffe
ChWick/caffe
ChengduoZhao/caffe
Coderx7/caffe_win
Cysu/caffe
DavyVan/caffe-with-CRF
Eniac-Xie/caffe
Fang-Haoshu/RMPE
FlorisGaisser/caffe
HUJI-Deep/caffe-simnets
JanZivcak/caffe_player
LihangLiu/apollocaffe
Maratyszcza/caffe
MhLiao/TextBoxes
MichalBusta/Ristretto-caffe
Microsoft/caffe
NVIDIA/caffe
Nanne/caffe
OpenHero/caffe
Paradigm4/caffe
Pastromhaug/caffe-stochastic-depth
RadekSimkanic/caffe-for-cudnn-v2.5.48
Russell91/apollocaffe
ShaoqingRen/caffe
ShiqiYu/caffe
Tongcheng/caffe
Xiaomi2008/caffe_nd_sense_segmentation
Yangqing/caffe
YutingZhang/caffe
acarabott/caffe
aiilab/caffe
ajtulloch/caffe
akirafukui/caffe
alemagnani/caffe
aleph7/caffe
amiralush/caffe
amoussawi/caffe
anandthakker/caffe
aravindhm/caffe
ashrafk/caffe
awabot-dev/caffe
beijbom/caffe
beniz/caffe
benjibc/caffe-rpi
borisgin/caffe
brodyh/caffe
buptwangfei/caffe
caffe2/caffe
caozhengquan/caffe
cbfinn/caffe
cdoersch/caffe
chengyangfu/caffe
chenxiang204/caffe
conner99/caffe
ctensmeyer/caffe
ctuning/caffe
cwd0610/caffe
cypof/caffe
ddtm/caffe
deercoder/DeepFood
developmentseed/caffe
dineshj1/caffe
dmlc/caffe
dosovits/caffe-fr-chairs
dpaiton/caffe
drnikolaev/caffe
dtmoodie/caffe
ducha-aiki/caffe
elezar/caffe
eli-osherovich/caffe
emcmanus/caffe
erictzeng/caffe
flickr/caffe
flx42/caffe
flynnhe/caffe
forresti/caffe
freesouls/caffe
frogner/caffe
gabrielPeart/caffe
gedikli/caffe
gidariss/caffe_LocNet
guker/caffe
gustavla/caffe
hardegg/caffe
hli2020/caffe
initialneil/caffe
intelcaffe/caffe
ivankreso/caffe-xeon-phi
jackculpepper/caffe
jacobandreas/apollocaffe
jasonustc/caffe
jeffdonahue/caffe
jetpacapp/caffe
jhauswald/caffe
jiaxiang-wu/caffe-QCNN
jjhartmann/caffe
kashefy/caffe
kmatzen/caffe
kod3r/caffe
kpzhang93/caffe-face
kuprel/caffe
lihuibng/caffe
linhj184169280/caffe
lokistone/caffe
longjon/caffe
lunochod/caffe
malinna/caffe-pose_network
matthieudelaro/caffeBVLCplus
medivhna/neural-aggregation-network
memo/caffe
mengbiping/caffe
mfs6174/caffe
mjmarin/caffe
mmoghimi/BoostCNN
mohamed-ezz/caffe
mtamburrano/caffe
muupan/caffe
mydude/caffe-triplet
myfavouritekk/caffe
n-zhang/caffe
naeluh/caffe
naibaf7/caffe
nicklhy/caffe-dev
nipengadmaster/caffe
niuzhiheng/caffe
nyamnyam7/private-caffe
pannous/caffe
pmgysel/caffe
psioncoder/caffe
pulkitag/caffe
qingswu/caffe
raingo/caffe
rakesh-mohanta/caffe
rbgirshick/caffe
redknightlois/caffe
rksltnl/Caffe-Deep-Metric-Learning-CVPR16
ronghanghu/caffe
sanghoon/caffe
scanlime/caffe
semihyagcioglu/caffe
sergeyk/caffe
sguada/caffe-public
sh1r0/caffe
shaibagon/caffe
shelhamer/caffe
shi-yan/caffe
shihenw/caffe
shiquanwang/caffe
slayton58/caffe
smajida/caffe
sp2823/caffe
stanford-futuredata/ssd-benchmarking
starimpact/caffe
stephenyan1231/caffe-private
stokasto/caffe
strin/caffe-opencl
sukritshankar/caffe
syhw/caffe
szcom/caffe
thuml/hash-caffe
tiangolo/caffe
tidsp/caffe-jacinto
tjevgerres/caffe
tnarihi/caffe
torrvision/caffe
tsingjinyun/caffe
vsubhashini/caffe
waderly/caffe
wangyida/caffe
weiliu89/caffe
willyd/caffe
woozzu/caffe
xdshang/caffe-multilabel
xianjiec/caffe
xllau/caffe
xuepo/caffe
xuzhenqi/caffe
xyy19920105/caffe
yanii/caffe
ydwen/caffe-face
yjxiong/caffe
yocox/caffe
yosinski/caffe
zhangkom/caffe
zhangliliang/caffe
zhongzhuoyao/caffe
zjchuyp/caffe
zlmzju/caffe
znah/caffe
Nothing to show
...
Choose a Head Repository
BVLC/caffe
01org/caffe
123gpg321/caffe
71squared/caffe
AMDComputeLibraries/caffe
AdamStelmaszczyk/caffe
Adnan1011/NR-IQA-CNN
AkiChen/Correlative-Filters-Caffe
AlfredXiangWu/caffe
Austriker/caffe
BlGene/caffe
CUHK-MMLAB/caffe
ChWick/caffe
ChengduoZhao/caffe
Coderx7/caffe_win
Cysu/caffe
DavyVan/caffe-with-CRF
Eniac-Xie/caffe
Fang-Haoshu/RMPE
FlorisGaisser/caffe
HUJI-Deep/caffe-simnets
JanZivcak/caffe_player
LihangLiu/apollocaffe
Maratyszcza/caffe
MhLiao/TextBoxes
MichalBusta/Ristretto-caffe
Microsoft/caffe
NVIDIA/caffe
Nanne/caffe
OpenHero/caffe
Paradigm4/caffe
Pastromhaug/caffe-stochastic-depth
RadekSimkanic/caffe-for-cudnn-v2.5.48
Russell91/apollocaffe
ShaoqingRen/caffe
ShiqiYu/caffe
Tongcheng/caffe
Xiaomi2008/caffe_nd_sense_segmentation
Yangqing/caffe
YutingZhang/caffe
acarabott/caffe
aiilab/caffe
ajtulloch/caffe
akirafukui/caffe
alemagnani/caffe
aleph7/caffe
amiralush/caffe
amoussawi/caffe
anandthakker/caffe
aravindhm/caffe
ashrafk/caffe
awabot-dev/caffe
beijbom/caffe
beniz/caffe
benjibc/caffe-rpi
borisgin/caffe
brodyh/caffe
buptwangfei/caffe
caffe2/caffe
caozhengquan/caffe
cbfinn/caffe
cdoersch/caffe
chengyangfu/caffe
chenxiang204/caffe
conner99/caffe
ctensmeyer/caffe
ctuning/caffe
cwd0610/caffe
cypof/caffe
ddtm/caffe
deercoder/DeepFood
developmentseed/caffe
dineshj1/caffe
dmlc/caffe
dosovits/caffe-fr-chairs
dpaiton/caffe
drnikolaev/caffe
dtmoodie/caffe
ducha-aiki/caffe
elezar/caffe
eli-osherovich/caffe
emcmanus/caffe
erictzeng/caffe
flickr/caffe
flx42/caffe
flynnhe/caffe
forresti/caffe
freesouls/caffe
frogner/caffe
gabrielPeart/caffe
gedikli/caffe
gidariss/caffe_LocNet
guker/caffe
gustavla/caffe
hardegg/caffe
hli2020/caffe
initialneil/caffe
intelcaffe/caffe
ivankreso/caffe-xeon-phi
jackculpepper/caffe
jacobandreas/apollocaffe
jasonustc/caffe
jeffdonahue/caffe
jetpacapp/caffe
jhauswald/caffe
jiaxiang-wu/caffe-QCNN
jjhartmann/caffe
kashefy/caffe
kmatzen/caffe
kod3r/caffe
kpzhang93/caffe-face
kuprel/caffe
lihuibng/caffe
linhj184169280/caffe
lokistone/caffe
longjon/caffe
lunochod/caffe
malinna/caffe-pose_network
matthieudelaro/caffeBVLCplus
medivhna/neural-aggregation-network
memo/caffe
mengbiping/caffe
mfs6174/caffe
mjmarin/caffe
mmoghimi/BoostCNN
mohamed-ezz/caffe
mtamburrano/caffe
muupan/caffe
mydude/caffe-triplet
myfavouritekk/caffe
n-zhang/caffe
naeluh/caffe
naibaf7/caffe
nicklhy/caffe-dev
nipengadmaster/caffe
niuzhiheng/caffe
nyamnyam7/private-caffe
pannous/caffe
pmgysel/caffe
psioncoder/caffe
pulkitag/caffe
qingswu/caffe
raingo/caffe
rakesh-mohanta/caffe
rbgirshick/caffe
redknightlois/caffe
rksltnl/Caffe-Deep-Metric-Learning-CVPR16
ronghanghu/caffe
sanghoon/caffe
scanlime/caffe
semihyagcioglu/caffe
sergeyk/caffe
sguada/caffe-public
sh1r0/caffe
shaibagon/caffe
shelhamer/caffe
shi-yan/caffe
shihenw/caffe
shiquanwang/caffe
slayton58/caffe
smajida/caffe
sp2823/caffe
stanford-futuredata/ssd-benchmarking
starimpact/caffe
stephenyan1231/caffe-private
stokasto/caffe
strin/caffe-opencl
sukritshankar/caffe
syhw/caffe
szcom/caffe
thuml/hash-caffe
tiangolo/caffe
tidsp/caffe-jacinto
tjevgerres/caffe
tnarihi/caffe
torrvision/caffe
tsingjinyun/caffe
vsubhashini/caffe
waderly/caffe
wangyida/caffe
weiliu89/caffe
willyd/caffe
woozzu/caffe
xdshang/caffe-multilabel
xianjiec/caffe
xllau/caffe
xuepo/caffe
xuzhenqi/caffe
xyy19920105/caffe
yanii/caffe
ydwen/caffe-face
yjxiong/caffe
yocox/caffe
yosinski/caffe
zhangkom/caffe
zhangliliang/caffe
zhongzhuoyao/caffe
zjchuyp/caffe
zlmzju/caffe
znah/caffe
Nothing to show
Checking mergeability… Don’t worry, you can still create the pull request.
This comparison is big! We’re only showing the most recent 250 commits
Commits on Dec 01, 2016
@willyd willyd Merge pull request #5042 from willyd/windows
Merge branch 'master' into windows
ea72266
@willyd willyd Added artifacts to appveyor builds
4ef2c3e
Commits on Dec 02, 2016
@willyd willyd Merge pull request #5049 from willyd/artifacts
Added artifacts to appveyor builds
3d1f43d
@naibaf7 naibaf7 ND-pooling regression issue fixes, LibDNN pooling improvements (corne…
…r cases), comparative large-scale pooling testing.
5ff019c
@naibaf7 naibaf7 Int_tp fix. f310a5b
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe 98b6580
Commits on Dec 04, 2016
@naibaf7 naibaf7 Pooling corner case fixes and average pooling improvements. 6ab003a
@naibaf7 naibaf7 Cleanup. f60da57
@naibaf7 naibaf7 Flush memory before backpropagating pooling in LibDNN. 2f4c6b6
Commits on Dec 05, 2016
@naibaf7 naibaf7 LibDNN Reshape -> Recompile functionality. c3a9f27
@naibaf7 naibaf7 Cleanup. 019c3d5
Commits on Dec 11, 2016
@naibaf7 naibaf7 Cleanup. c3ac3ea
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe 1ec0d1a
Commits on Dec 12, 2016
@naibaf7 naibaf7 Constant replacing. 0c7d18d
@naibaf7 naibaf7 Cleanup for OpenCL-Windows build. 185cd4c
Commits on Dec 13, 2016
@naibaf7 naibaf7 OpenCL string splitting for MSVC. 81e9b2e
@naibaf7 naibaf7 Windows OpenCL build artefacts. 7fb88d9
@naibaf7 naibaf7 Flattened OpenCL strings, MSVC memory alignment. e36d4a3
Commits on Dec 14, 2016
@naibaf7 naibaf7 Aligned free (MSVC) 8251da6
@naibaf7 naibaf7 Working Windows build for OpenCL. 0a73fc7
@naibaf7 naibaf7 Merge branch 'master' of github.com:naibaf7/caffe 48783cd
Commits on Dec 15, 2016
@naibaf7 naibaf7 LibDNN changes for AMD OpenCL 2.0 Windows compability. 19afe25
@naibaf7 naibaf7 AMD Windows OpenCL 2.0 bug, fix 2. d6091c8
@naibaf7 naibaf7 Windows OpenCL Caffe release commit.
60d551f
@naibaf7 naibaf7 int_tp type fix for classification.cpp c96a8f3
Commits on Dec 21, 2016
@naibaf7 naibaf7 Removed unnecessary LibDNN definition. 7baf096
Commits on Dec 22, 2016
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
b499dda
Commits on Dec 31, 2016
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
4ddb436
Commits on Jan 06, 2017
@gongzg gongzg avoid using printf in dead code.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
1426df4
@gongzg gongzg Refine weights swizzling logic.
Use TEST phase to determine whether do we need to swizzle
weights every time.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
7107b28
@wzw-intel @gongzg wzw-intel + gongzg Across channels LRN optimization. a3a5d29
@wzw-intel @gongzg wzw-intel + gongzg BatchNormLayer optimization.
This optimization is for TEST phase. So only the path with
use_global_stats == true optimized.
7c9cb82
@gongzg gongzg Optimize softmax for intel platform.
Combine all kernels into one and use sub group and share local
memory to reduce the memory traffic and some computation for
max and sum. It reduce the batch 4 googlenet v2's softmax layer's
time from 0.58ms to 0.1 ms, get nearly 6x performance improvement.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
23f6667
@gongzg gongzg Optimize softmax loss layer for Intel platform.
Reduce the googlenetv2's loss layer time from 0.8ms to 0.5ms,
and remove one synchronization point as we don't need to read
back the sum to cpu buffer.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
da589e4
@gongzg gongzg Refine direct gemm like convolution kernels.
Use flexible version rather than fixed block size version.
Add one SIMD16 variant kernel which is more efficient than
the original one for small kernel width convolution.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
2c662b6
@gongzg gongzg Slightly optimization for 1x1 kernel.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
c47f5ff
@gongzg gongzg Added SIMD8 support for direct convolution kernel.
When the output channel is less than 16 or the image size is very small,
SIMD8 may be more efficient than SIMD16.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
25ff13a
@gongzg gongzg fix bug when loading old auto-tuning cache record for spatial convolu…
…tion engine.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
633a737
@gongzg gongzg Check actual compute unit for SIMD8 direct spatial conv kernel.
Also check the sub group extension rather the Intel vendor. As some
versions of the driver don't support sub group extension.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
388a7f2
@gongzg gongzg Fix a bug in spatial convolution engine.
When tile x is larger than 4 * simd_x the kernel is broken, we
just simply ignore this configuration. Disable non-multiple of 32
output channels for the gemm like direct convolution kernel due
to some known bugs.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
1d6160d
@gongzg gongzg Enable the gemm like kernel for some non multiple of 32 output channels.
Also fixed a typo bug in the simd16 gemm like kernel.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
b4dfef7
@gongzg gongzg Lint fix.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
08cdbda
@gongzg gongzg Fixed some kernel compatibility issue.
According to OCL spec, we should not define __local buffer in
non kernel function. And don't compile those intel sub group
related kernels if there is no intel sub group extension support.
d02e447
Commits on Jan 08, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
78d4992
@naibaf7 naibaf7 Extended build possibilities for windows.
a85b838
@naibaf7 naibaf7 Replace tab by spaces.
d38720f
Commits on Jan 09, 2017
@willyd willyd Added missing pyyaml in appveyor build script
3649015
@willyd willyd Merge pull request #5161 from willyd/pyyaml
Install missing pyyaml package on appveyor
a9ace02
@willyd willyd Fixed appveyor build status badge and added prebuilt binaries downloa…
…d link
38862a5
@willyd willyd Merge pull request #5162 from willyd/cibadge
Fixed appveyor build status badge
2da94c4
Commits on Jan 11, 2017
@naibaf7 naibaf7 Merge branch 'opencl' of https://github.com/01org/caffe 69b7bbd
@naibaf7 naibaf7 Intel spatial kernel update, conditional fixes, lint fixes.
8eb76e0
Commits on Jan 13, 2017
@willyd willyd Merge branch 'master' into windows
e4f3702
@lunzueta @willyd lunzueta + willyd Update CMake files to build Matcaffe on Windows
58563ab
Commits on Jan 14, 2017
@willyd willyd Merge pull request #5180 from willyd/matlab-support
Update CMake files to build Matcaffe on Windows
8bb7cbc
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe a31fadc
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe
72edcdc
Commits on Jan 16, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
45bec1b
@gongzg gongzg Fix one incorrect constant value type.
We should always use Dtype constant value by default. Otherwise
it may cause compilation error for the platform doesn't support
double type. And even if the platform does support double, when
we build float type kernel, it will promote some of the routines
to double type which is not efficient.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
7d88092
Commits on Jan 17, 2017
@willyd willyd Merge branch 'master' into windows
86425b1
@naibaf7 naibaf7 Merge pull request #5190 from 01org/opencl
OpenCL: fix one incorrect constant value type.
6920930
@willyd willyd Added support for python 3.5
82c087c
Commits on Jan 18, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
1f046f4
@willyd willyd Merge pull request #5088 from willyd/python35
Adding support for python 3.5 in windows
e483c7b
@willyd willyd Fixed AppVeyor artifacts download links
f4db952
@willyd willyd Added CI with CUDA on Windows and disabled some Debug builds on AppVeyor
372e920
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
3120d52
Commits on Jan 19, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
ac24fa7
@naibaf7 naibaf7 Fix LibDNN memory flush bug.
59e986f
@naibaf7 naibaf7 ND pooling return vs. continue bugfix.
4aeffd3
@naibaf7 naibaf7 Merge. 13fd828
Commits on Jan 20, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
83b9ac8
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
948aa8e
@willyd willyd Merge pull request #4895 from willyd/cuda-ci
Added CUDA build on appveyor
655be9f
@forderud forderud Fix the following error when running the script without defining "WIT…
…H_CUDA":

"( was unexpected at this time."
a3f1cce
@willyd willyd Merge pull request #5204 from forderud/cuda-define
Fix broken build on windows when not defining WITH_CUDA
382ae79
@willyd willyd Fix broken download links for CUDA enabled AppVeyor builds
8a49d45
@forderud forderud Make it possible to change build configuration on Windows without hav…
…ing to edit build_win.cmd.
46b3d46
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe bee6df8
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe
817efe9
@naibaf7 naibaf7 Fix windows OpenCL build.
0a0d68e
Commits on Jan 24, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
2bcb9cd
Commits on Jan 30, 2017
@willyd willyd Merge pull request #5206 from forderud/win-build-cfg
Windows build configuration changes without editing build_win.cmd
a7cc718
Commits on Feb 01, 2017
@DVEfremov DVEfremov add android cmake support over the BVLC/caffe opencl (issue #48)
- android toolchain (from github.com/sh1r0/caffe-android-lib)
- HDF5 disabling (from github.com/sh1r0/caffe-android-lib)
- some my minor fixes to get work for android for opencl version
aed934d
@DVEfremov DVEfremov fix for disabling HDF5 for android 1dc780b
@DVEfremov DVEfremov add andoid compilation support e7f4160
defremov Revert "add andoid compilation support"
This reverts commit e7f4160.
56b2336
Commits on Feb 02, 2017
defremov android compilation support 8e12a09
defremov android compilation support fda16cc
@DVEfremov DVEfremov + defremov Mali GPU does not support host unified memory in fact #53
- as solution provided compilation param
DISABLE_DEVICE_HOST_UNIFIED_MEMORY
to force disabling support host unified memory
1dd40c4
@naibaf7 naibaf7 Merge pull request #54 from DVEfremov/issues-53
Mali GPU does not support host unified memory in fact #53
65ca1ec
@naibaf7 naibaf7 Merge pull request #50 from DVEfremov/issue-48
add android cmake support over the BVLC/caffe opencl (issue #48)
cd6e679
@naibaf7 naibaf7 Merge a855a52
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe 7713d22
@naibaf7 naibaf7 Merge branch 'master' of github.com:naibaf7/caffe 6716669
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe c0861bc
@naibaf7 naibaf7 Lint fix. fa9ffff
@naibaf7 naibaf7 Index 64 fix.
7da76f3
Commits on Feb 03, 2017
@DVEfremov DVEfremov Compilation fails for caffe opencl branch in CPU_ONLY mode issue #55
- behavior in else part (CPU verision) taken from BVLC.caffe master branch
1cff25b
@DVEfremov DVEfremov OpenCl kernel compilation errors for android #51
- add ability to disable DOUBLE SUPPORT with compilation flag
  DISABLE_DOUBLE_SUPPORT
  (hardcoded DOUBLE_SUPPORT_AVAILABLE atleast doesn't work for Mali GPU at caffe time runtime)
563bba8
@DVEfremov DVEfremov Compilation fails for caffe opencl branch in CPU_ONLY mode issue #55
- review fix
b5b16a9
@naibaf7 naibaf7 Merge pull request #56 from DVEfremov/issue-55
Compilation fails for caffe opencl branch in CPU_ONLY mode issue #55
0972b6e
Commits on Feb 06, 2017
@DVEfremov DVEfremov Mali GPU does not support host unified memory in fact #53
- missed changes to CMakeLists.txt for original issue
834ae9b
@willyd willyd Removed unnecessary 1 with SHOW_PROGRESS option and fixed error messa…
…ge in WindowsDownloadPrebuiltDependencies.cmake
d14012e
@naibaf7 naibaf7 Merge pull request #58 from DVEfremov/issue-53-2
Mali GPU does not support host unified memory in fact #53
ab6ab43
Commits on Feb 07, 2017
@DVEfremov DVEfremov OpenCl kernel compilation errors for android #51
- add ability to disable DOUBLE SUPPORT with compilation to cl_kernels.sh and header.cl (fix review comment)
- add generated cl_kernels.cpp (as mentioned before for MacOS and Windows users where code generation does not work at this moment)
e1bfc88
@naibaf7 naibaf7 Merge pull request #57 from DVEfremov/issue-51-2
OpenCl kernel compilation errors for android #51
f00d3aa
@naibaf7 naibaf7 Fix OpenCL issues with zero-sized buffers and data input layers. 674339d
@naibaf7 naibaf7 Merge branch 'master' of github.com:naibaf7/caffe
1350327
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe
fdc4098
@naibaf7 naibaf7 Stilistic fix in HDF5 workaround.
05fdfd1
Commits on Feb 09, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
69f35c5
Commits on Feb 10, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
e59b8ca
Commits on Feb 17, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
9f262c6
@willyd willyd Merge branch 'master' into windows ac59d48
@willyd willyd Added nccl ExternalProject to build nccl on Windows
2106a62
@willyd willyd Merge pull request #5231 from willyd/windows-nccl
Merge latest master changes and port nccl parallelism to windows
882bc3d
Commits on Feb 18, 2017
@willyd willyd Added *.lib/*.dll/*.pyd/*.exe to .gitignore
7c1daec
@willyd willyd Added powershell scripts to mimic the .sh script to download and crea…
…te the mnist database
ebc54c0
@willyd willyd Added copy of required dependencies to examples executables
8024d67
Commits on Feb 19, 2017
@willyd willyd Merge pull request #5287 from willyd/windows-ignore
Added *.lib/*.dll/*.pyd/*.exe to .gitignore
6900c4b
@willyd willyd Merge pull request #5288 from willyd/windows-mnist
Added powershell scripts to download and extract the mnist data
cc0614d
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe
0fc9cd4
Commits on Feb 21, 2017
@naibaf7 naibaf7 Header fix. b62ea1b
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe 647f214
@naibaf7 naibaf7 native_powr() change.
8f320c7
@naibaf7 naibaf7 LRN change.
f3ba72c
Commits on Mar 03, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
8035fe9
Commits on Mar 04, 2017
@willyd willyd Updated prebuilt dependencies. Fixes #5348
527f97c
@naibaf7 naibaf7 Add possibility to disable host unified memory in Makefile build. 3fe9518
Commits on Mar 08, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe 27fb723
@naibaf7 naibaf7 Int types 4a6347e
@naibaf7 naibaf7 Lint fix.
b649df1
Commits on Mar 09, 2017
@naibaf7 naibaf7 CMake changes.
7b0bafd
Commits on Mar 12, 2017
@willyd willyd Merge pull request #5355 from willyd/lmdb-fix
Updated prebuilt dependencies. Fixes #5348.
d8c6e87
@willyd willyd Added train_lenet.ps1, fixed create_mnist.ps1 and update mnist readme…
….md for windows.
a35cc49
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe
fff7186
Commits on Mar 13, 2017
@gongzg gongzg Remove unecessary finish in synced memory.
For the to_gpu function with uninitialized memory case, we
do not need to finish queue, and if the HEAD is at CPU and
we support zero copy, then we also don't need to finish
the queue.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
2108299
@gongzg gongzg Refine timing mechanism for auto-tuning phase.
The caffe's timer has some overhead, and when our tunning kernel is
very tiny, the overhead may cause very unstable timing result, so
I increase the iteration count to reduce this type of overhead.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
4421e9a
@gongzg gongzg Refine softmax layer's forward code path.
If the spatial dimension is relatively large, we should use the default code
path to achieve better parallelism.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
1a77cc6
@gongzg gongzg Refine error handling for spatial convolution.
Sometimes, the sub buffer creation may fail, we need to take care of it.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
2a35d32
@wzw-intel @gongzg wzw-intel + gongzg Fix incorrectly add __BEIGNET__ macro into option.
Some features e.g. opencl_unroll_hint are not allowed for beignet
compiler, use __BEIGNET__ macro to choose whether to build with these
features.

Also add an helper func to faciliate judging beignet driver.

Signed-off-by: Zhiwen Wu <zhiwen.wu@intel.com>
8ea60a9
@gongzg gongzg Fix a bug in spatial convolution engine.
If the input image size changed during runtime, and the kernel type
change to 2 or 5, we need to swizzle the weights again.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
0ed9083
@wzw-intel @gongzg wzw-intel + gongzg spatial conv: Remove image padding
Added a new basic convolution kernel that supports input image with
no padding, so that no image padding in host code need anymore.

Signed-off-by: Zhiwen Wu<zhiwen.wu@intel.com>
14dc7fc
@gongzg gongzg Remove unecessary clFinish in spatial convolution engine.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
b5d98a8
@listenlink @gongzg listenlink + gongzg Enable conv_spatial dilation parameters
Change-Id: I392c4e73319fcfc18e628f9476b9bfdcba3cc206
c08a8a9
@gongzg gongzg Fix one kernel compilation test failure.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
035465c
@listenlink @gongzg listenlink + gongzg Enable OCL version of hdf5_data_layer to fix a race condition
If we simply use the cpu code path to copy the data, we will introduce
one race condition between the GPU queue and the CPU. The scenario is:
when we call it in an iteration loop. The data blob is in a zero-copy
blob, and the first pass may be still blocking on the GPU side. The
second pass will modify the data blob on CPU side before the data is
accessed at the first pass on GPU side.

We can simply add a synchronization point between the two iterations, but that
is not a good fix as we force the GPU queue to flush and wait it to finish.
The best way is to do the copy on the GPU side and in the same queue. Thus we
don't need to worry about this race condition any more and without any interfere
the GPU queue.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
cee4dbc
@gongzg Pan, Xiuli + gongzg Fix a constant value bug
This will cause relu gradient fail.
6ae9830
@gongzg gongzg Don't use fixed image size for GEMM like kernels.
Prepare to support varying sizes.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
4147a7d
@gongzg gongzg Use tuning size rather than actual size.
No need to tune different kernel for each different input size.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
f646c14
@gongzg gongzg Refine spatial kernel's cache mechanism.
Add the platform and driver information to change to use
system cache directory if possible. After this change, we
can reuse a offline tuned configurations.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
fc86e43
@gongzg gongzg Redirect the Intel OpenCL backend information to wiki page.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
e9fbabe
@gongzg gongzg Eliminate some OCL kernel warnings.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
8d133fd
@gongzg gongzg CMAKE_EXT should be empty for now.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
859f31d
@gongzg gongzg Lint fix.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
11665ce
@naibaf7 naibaf7 Merge pull request #5394 from 01org/opencl
Opencl some bug fixes and feature enhancement
479d3a0
Commits on Mar 14, 2017
@gongzg gongzg Fix a bug in host memory free for OpenCL backend
We need to use normal free to deallocate memory for OpenCL backend.
If MKL is enabled, current code will use mkl_free to deallocate a
memory allocated by normal posix_memalign/malloc() and will cause
segfault.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
51ef50e
@vpa1977 vpa1977 removed deprecated API call 6dd1bac
@vpa1977 vpa1977 fix formatting
77bd399
@naibaf7 naibaf7 Merge pull request #5403 from 01org/opencl
Fix a bug in host memory free for OpenCL backend
e4f2474
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe bef2cb5
@naibaf7 naibaf7 Merge branch 'opencl' of github.com:BVLC/caffe
c61d487
Commits on Mar 17, 2017
@willyd willyd Merge branch 'master' into windows
Fixed issue related to prebuilt dependencies being downloaded in build
directory. As reported in #5378.
2a35878
@willyd willyd Fixed lint issue when .pb.h and .pb.cc files do not exist
ca360a1
Commits on Mar 22, 2017
@willyd willyd Fixed wrong VS 2013 Update 5 version string. Fixes #5430
de0ee29
Commits on Mar 23, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe 69ddb2f
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe
106ca68
Commits on Mar 24, 2017
@willyd willyd Merge pull request #5441 from willyd/vs2013-up5-fix
Fixed wrong VS 2013 Update 5 version string. Fixes #5430.
5d3dced
Commits on Mar 25, 2017
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe 7538339
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
cb3ef08
Commits on Mar 27, 2017
@willyd willyd Removed deprecated VS based build. 621c5c7
@willyd willyd Merge branch 'master' into windows
751ff4d
@naibaf7 naibaf7 Merge pull request #5402 from vpa1977/opencl
Opencl - replace deprecated method call in prelu layer backwards_gpu()
3452f39
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe 0932f3c
@naibaf7 naibaf7 Revert "Opencl - replace deprecated method call in prelu layer backwa…
…rds_gpu()"
39562fe
@naibaf7 naibaf7 Merge pull request #5454 from BVLC/revert-5402-opencl
Revert "Opencl - replace deprecated method call in prelu layer backwards_gpu()"
a09b87c
@naibaf7 naibaf7 Merge branch 'opencl' of github.com:BVLC/caffe bc5f6fa
Commits on Mar 28, 2017
@cepiross cepiross Resolve cmake compilation error in *.cu files
72f4de7
@naibaf7 naibaf7 Merge pull request #5457 from cepiross/viennacl
Resolve cmake OSX compilation error in *.cu files
3e0bde1
@naibaf7 naibaf7 Merge branch 'opencl' of github.com:BVLC/caffe 555bca7
@AhmedOS AhmedOS Dependencies URLs updated for newer version
d800357
Commits on Mar 29, 2017
@willyd willyd Merge pull request #5460 from AhmedOS/patch-1
Dependencies URLs updated for newer version
88ddc95
Commits on Mar 31, 2017
@naibaf7 naibaf7 LibDNN Deconvolution implementation. 8ef65eb
Commits on Apr 01, 2017
@naibaf7 naibaf7 Memory initialization improvements for future data types, LibDNN and …
…Deconv fixes.
524de01
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe 8131572
@naibaf7 naibaf7 Deconv layer improvements. d6aa4eb
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe fd6f197
@naibaf7 naibaf7 Remove debug flag in LibDNN. 5f310c4
@naibaf7 naibaf7 Removed duplicate layer parameter in PyCaffe.
59113c0
Commits on Apr 07, 2017
@wzw-intel wzw-intel libdnn: add spatial convolution implmentation
A new class in libdnn LibDNNConvSpatial which is porting
from clcaffe's ConvolutionLayerSpatial class but with following
changes:
 - Use basic kernel as benchmark kernel
 - Add LibDNNConvConfig::phase_test to indicate using in TEST phase

Also, the kernel source from conv_layer_spatial.cl is refined to
reduce reduntance and embedded into host code, so that it has no
dependence on Caffe components.

Note that the tuner mechanism is different from LibDNNTuner in
this version. It will cache the tuned parameters on local storage.
See "Auto-tuning and kernel cache mechanism" in
https://github.com/01org/caffe/wiki/clCaffe for details.
fbdb611
@gongzg gongzg Disable viennacl cache mechanism during spatial engine's tuning phase.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
d479336
@gongzg gongzg Refine zero copy support.
As all memory with OpenCL backend are allocated with qualified size and
alignment, we can ignore the size check.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
86754bc
@gongzg Richman, Reuven + gongzg softmax layer cpu fwd - no need to max values with themselves 730ce18
@gongzg gongzg Remove unecessary queue finish in relu layer.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
a110f87
@gongzg gongzg Fix a compilation error when DEBUG enabled without CUDA.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
bc9bd7c
@gongzg gongzg Fix one varying size bug for convolution kernel and minor improvement.
The last block width/height should be varying as well.
And put some uniform size related calculations to CPU side.
Also need to re-calculate global size for different image size.

Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
5459535
@gongzg gongzg Fix ocl kernel compilation errors.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
6808534
@naibaf7 naibaf7 Update softmax_layer.cpp
f444524
@naibaf7 naibaf7 Merge pull request #5497 from 01org/opencl
Opencl - some bug fixes and cache mechanism refinement.
8a7b4d0
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
6c559ba
Commits on Apr 08, 2017
@willyd willyd Merge branch 'master' into windows
a166218
Commits on Apr 09, 2017
@willyd willyd Fixed issue with missing pydot and graphviz under windows
a48e81f
Commits on Apr 10, 2017
@willyd willyd Merge pull request #5501 from willyd/pydot-fix
Fixed issue with missing pydot and graphviz under windows
90e8abc
Commits on Apr 11, 2017
@gongzg gongzg Fix segfault when VIENNACL_CACHE_PATH is not set.
Signed-off-by: Zhigang Gong <zhigang.gong@intel.com>
bc1ea90
Commits on Apr 13, 2017
@naibaf7 naibaf7 Merge pull request #5510 from 01org/opencl
opencl: Fix segfault in spatial convolution engine when VIENNACL_CACHE_PATH is not set
8a80a89
Commits on Apr 14, 2017
@psyhtest psyhtest Exclude HDF5-specific code when not supported. dc07695
Commits on Apr 17, 2017
@naibaf7 naibaf7 Merge pull request #63 from ctuning/issue62
Exclude HDF5-specific code when not supported
1eec1b0
Commits on Apr 18, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe aa1c026
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe 3cd6298
@naibaf7 naibaf7 Merge branch 'opencl' of github.com:BVLC/caffe
61db960
Commits on Apr 19, 2017
@naibaf7 naibaf7 PyCaffe fixes. 70858cb
@naibaf7 naibaf7 Fix MSVC error for PyCaffe.
77c3428
@naibaf7 naibaf7 Fixed unterminated #ifndef
880111e
Commits on Apr 21, 2017
@PENGUINLIONG PENGUINLIONG Added GPU arch option.
Allow users to specify GPU arch in `build_win.cmd` for convenient cross cmopilation.
6280643
@willyd willyd Merge pull request #5553 from PENGUINLIONG/dev-penguinliong
Added GPU arch option.
37ff78a
Commits on Apr 26, 2017
@willyd willyd Merge branch 'master' into windows 1f75c51
@willyd willyd Merge branch 'master' into windows af191e2
Commits on Apr 27, 2017
@willyd willyd Added missing dependency between libcaffe and nccl
c46b910
Commits on Apr 28, 2017
@naibaf7 naibaf7 Extended/Updated FindOpenCL.cmake
09060b3
Commits on May 02, 2017
@naibaf7 naibaf7 Make MetaLayers and fix_input_dims part of PyCaffe (internalized from…
… PyGreentea).
33e975c
@naibaf7 naibaf7 Cleanup.
2c829c1
Commits on May 04, 2017
@willyd willyd Added support for resolving MATLAB prerequisites. Also fixes a bug wi…
…th CMake GetPrerequisites module (see https://gitlab.kitware.com/cmake/cmake/merge_requests/804).
ea0d92d
Commits on May 05, 2017
@willyd willyd Merge pull request #5590 from willyd/matlab-prerequisites
Added support for resolving MATLAB prerequisites
5f92dbd
@willyd willyd Merge pull request #5389 from willyd/mnist-fix
MNIST example for windows fixes
77624b5
@willyd willyd Merge branch 'master' into windows
35f3fcf
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe e78fcba
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe
c60c950
Commits on May 09, 2017
@naibaf7 naibaf7 Add option to disable host unified memory in CMake. 955bb0d
@naibaf7 naibaf7 Add option to disable host unified memory on windows. ebaa07a
@naibaf7 naibaf7 Merge branch 'master' of github.com:naibaf7/caffe
b36508e
@naibaf7 naibaf7 Fix build_win.cmd typo
84d4bba
@naibaf7 naibaf7 Changed default option.
fa02ee6
Commits on May 13, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
320d9a9
Commits on May 16, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
0f19d3e
Commits on May 17, 2017
@naibaf7 naibaf7 Test wider parameter range for LibDNN Deconvolution. b026f68
@naibaf7 naibaf7 Update caffeproto name. e0f77c3
Commits on May 23, 2017
@naibaf7 naibaf7 Merge branch 'opencl' of https://github.com/wzw-intel/caffe into wzw-…
…intel-opencl
0b22688
@naibaf7 naibaf7 Merge branch 'wzw-intel-opencl' into opencl
b78e56d
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe efabcdb
@naibaf7 naibaf7 Merge branch 'opencl' of github.com:BVLC/caffe
80be8de
Commits on May 24, 2017
@willyd willyd Merge branch 'master' into windows
6bfc5ca
Commits on May 31, 2017
@naibaf7 naibaf7 Add stable upconvolution path for fixing unstable MALIS gradient issues.
02310e1
Commits on Jun 05, 2017
@cepiross cepiross Resolve build failure in libdnn_conv_spatial
f809d18
@naibaf7 naibaf7 Merge pull request #5669 from cepiross/viennacl
Resolve build failure in libdnn_conv_spatial
f7a57a5
@naibaf7 naibaf7 Netgen min shape update.
557a2af
Commits on Jun 15, 2017
@naibaf7 naibaf7 Merge branch 'windows' of github.com:BVLC/caffe 1eb0dff
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe
483c58f
Commits on Jul 04, 2017
@naibaf7 naibaf7 Merge branch 'master' of github.com:BVLC/caffe 02821ac
@naibaf7 naibaf7 Updated Readme
73221fd