Intel #5721
Open
Commits
Show all changes
1467 commits
Select commit
Hold shift + click to select a range
527347f
task-ssd-detection-output-speedup2: _OPENMP flag corrected Merge remo…
sfraczek 7b7dc37
task-ssd-softmax-speedup2: cannot improve it further
sfraczek 575c20a
Merge branch 'master' of repos.igk.intel.com:intelcaffe into task-lay…
michalkuligowski a827d6e
Merge branch 'master' of repos.igk.intel.com:intelcaffe into task-lay…
michalkuligowski 1572fe3
UT fix
michalkuligowski 2cb6929
MKLDNN backward pooling enabled
tpatejko dda1c4f
- changes of MKLDNN LRN BW pass integrated
jczaja 1c904db
Changes after review
tpatejko e8f728d
task-mkldnn-lrn-backward-enabled: - changes of MKLDNN LRN BW pass int…
jczaja ee7a807
task-caffe-mkldnn-integration-pooling: MKLDNN backward pooling enabled
tpatejko 1dbc048
Update of Resnet-50 model
pawelszewc c4dabb9
Merge branch 'master' of repos.igk.intel.com:intelcaffe into task-res…
pawelszewc 31fdd18
Softmax was actually improved! Revert the revert
sfraczek eb95af9
fixed problem in SPPLayer
AdrShv 5c0d016
Multinode optimized models
lwencel fefab72
task-perfmon-temp-strings: Temporary strings in caffe time 2 fixed
Sand3r- b25828f
task-ssd-softmax-speedup3: Softmax was actually improved! Revert the …
sfraczek 187b912
integrated backward pass for MKLDNN Concat Layer
AdrShv 3cd14d1
task-multinode_optimized_models: Multinode optimized models
lwencel bbdc468
integrated MKLDNN BatchNorm backward
AdrShv e04136c
fixed bug in LayerSetUp, hide statistics from solver
AdrShv 213072b
task-resnet50-topology-update: Update of Resnet-50 model
pawelszewc 2d41403
task-mkldnn_batchnorm_backward3: fixed bug in LayerSetUp, hide statis…
AdrShv 79cbf4c
task-mkldnn-concat-backward: integrated backward pass for MKLDNN Conc…
AdrShv 0ead2cd
task-layerDataCompareToolSimplified: Upgraded compare tool Upgraded c…
michalkuligowski 52845ef
Added more switches to compare tool
michalkuligowski 09f8615
Added more switches to compare tool
michalkuligowski ac707f6
Added more switches to compare tool
michalkuligowski 05373db
Initial empty repository
452cc9b
Merge branch 'master from intel-caffe' into tf
opencici2006 c42a5d1
minor review changes
michalkuligowski bce926c
Fix mkl convolution net reshape
tsocha f68b69f
mbox
sfraczek 1af7124
fixed memory leaks detected by LeakSanitizer
AdrShv 9263663
changed mbox perf measurement to static
sfraczek 81e88e1
task-compareToolSwitchesAdded: Added more switches to compare tool
michalkuligowski 05a3521
task-fixed_memory_leaks_detected_by_LeakSanitizer: fixed memory leaks…
AdrShv 1d95f0b
task-mkl-reshape-fix: Fix mkl convolution net reshape
tsocha 56ce805
- Template for MKLDNN split layer added
jczaja 6d7f32f
Fix for Klocwork error
michalkuligowski 7fa7dc4
Improve Caffe time 2.0
hshen14 94a8936
task-mkldnn-split-backward: - Template for MKLDNN split layer added
jczaja 902fa4b
task-fixKlocwork1620: Fix for Klocwork error
michalkuligowski 855c86a
Layer factory enabled eltwise and split layer
sfraczek ebfafec
ELtwise disabled for now
sfraczek 0da45fd
changed === to == and fixed test
sfraczek 5c63b12
Merge repos.igk.intel.com:intelcaffe
jgong5 4f54ea3
task-mkldnn-layer-factory2: changed === to == and fixed test
sfraczek 8f44a4b
MKLDNN version update
lwencel e02de8c
task-caffe-time-2.0-0310: Improve Caffe time 2.0
hshen14 5b4a58a
Optimization for annotated data layer with OpenMP with fixes: empty q…
tpatejko e04a6eb
SSD annotated data layer: changes in data_transformer.cpp missing
tpatejko 43a553e
SSD annotated data layer optimization: removing redundant code
tpatejko 83946ed
Implement of MKLDNN-eltwise_layer and related unit tests.
328d5af
Implement of MKLDNN-eltwise_layer and related unit tests(Changed acco…
f8d0261
task-ssd-annotated-data-layer-optimization3: SSD annotated data layer…
tpatejko 73aae23
task-MKLDNN-eltwise_layer-Chong: Implement of MKLDNN-eltwise_layer an…
b13f5b1
- Fix to Split layer
jczaja ae34178
- Enabled MKLDNN Average pooling in layer factory
jczaja 991638c
task-mkldnn-googlenet-training-fixes: - Enabled MKLDNN Average poolin…
jczaja 8033f1f
Check the setting ot coefficients in the Backward.
f5ee516
task-performance-ssd: changed mbox perf measurement to static mbox
sfraczek 55116c5
task-MKLDNN-eltwise_layer-Backward_coefficient_Check-Chong: Check the…
ce8ca11
CaffeTime2.0 support for nested events improved
Sand3r- fa535ed
Merge branch 'master' of repos.igk.intel.com:intelcaffe into task-per…
Sand3r- bb2d4fc
Fix a custom layer engine from prototxt
tsocha b34f8ab
task-custom-layer-engine: Fix a custom layer engine from prototxt
tsocha 7ac495e
task-mkldnn-version-update2: MKLDNN version update
lwencel e67a594
Fixing MKLDNN part in Makefile. Updating git hash to the newest one
tpatejko 4841ff0
Handle ccache because earlier versions of CMake have problems handlin…
tpatejko 4c00af6
Correcting issue with missing separator
tpatejko 1d211ad
Plus added so MKLDNN_CMAKE_FLAGS could be set from command line
tpatejko b48a874
Corrected C compiler passed to MKLDNN
tpatejko a2a14f8
Prevent building when there's mkldnn already built
tpatejko 63df400
Remove Debian-specific stuff from Makefile
tpatejko f0cd66d
Downloading MKLDNN is not executed when test are run
tpatejko b9f5c48
Disabled Test1x1Convolution tests because of problems on IvyBridge
tpatejko bf617d1
Enabling mkldnn eltwise layer for failing unit test
tpatejko ad2e525
Both cmake and makefile use the same mkldnn commit hash
tpatejko 2c52ca6
Disabling the rest of MKLDNN convolution tests because of issues with…
tpatejko 4f69180
task-caffe-mkldnn-selfcontained-with-makefile3: Selfcontained IntelCa…
tpatejko c7a057f
fix bug in loading trained model, for models which contain BatchNorm …
749dbab
Merge branch 'master' into task-fix_bug_loading_trained_model_with_Ba…
bee02c4
task-fix_bug_loading_trained_model_with_BatchNorm: fix bug in loading…
246209c
- Bumped MKL-DNN version to get benefits of pooling optimizations
jczaja 61cca5c
task-bump-mkl-dnn-version: - Bumped MKL-DNN version to get benefits o…
jczaja d26146a
task-perfmon-nested-fix: CaffeTime2.0 support for nested events improved
Sand3r- 0d7833e
added deleting for split_top_vec elements
AdrShv 674cc53
Merge branch 'master' into prv-memory_leaks_fix
AdrShv 8b2e76f
task-fix_memory_leask_in_test_spp_layer: fixed memory leaks in SPPLay…
AdrShv 8a73e9f
Fix a MKL reshape
tsocha 0dc44cc
task-mkl-total-reshape-fix: Fix a MKL reshape
tsocha ceae29d
Fix make regression
hshen14 177d5cf
task-make-regression: Fix make regression
hshen14 c4cddd5
Fix a bug in relu and change solvers from GPU to CPU
tsocha 137ef7f
Merge branch 'master' into task-mkl-total-reshape-hotfix
tsocha b7b601b
task-mkl-total-reshape-hotfix: Fix a bug in relu and change solvers f…
tsocha 67d1232
Fix the compiler warnings in the source code.
3a08e63
Fixed released memory access at InnerProduct-TestForwardTranspose
Sand3r- 4c59615
Fix the compiler warnings in the source code.
05f3fce
task-complier-warning-removal-Chong: Fix the compiler warnings in the…
25a684c
In the previous, IntelCaffe had to zero diff for the bottom blob in t…
f5d0329
Fixed memory leak at TestForwardTranspose
Sand3r- 6d14cf2
Merge branch 'master' of repos.igk.intel.com:intelcaffe into task-use…
Sand3r- 0d3c880
task-use-after-free-fc-tests-fix: Fixed released memory access at Inn…
Sand3r- 0c2cedb
Remove unnecessary memory allocation for mkl convolution reshape
hshen14 cd37c5a
Merge branch 'master' of repos.igk.intel.com:intelcaffe into task-rem…
hshen14 887631f
Split reshape to two parts
hshen14 3e045db
Correcting annotated data layer after asan reported heap-use-after-fr…
tpatejko 91acd98
test
daisyden c1ddcff
task-remove-necessary-memory-allocation-for-mkl-conv: Remove unnecess…
hshen14 3616235
Revert: Fix a bug in relu
tsocha fbd850f
Disable unit test for mkl reshape
tsocha f32a7b8
task-correct-annotated-data-layer-asan-error: Correcting annotated da…
tpatejko 90a5d0e
task-avoid-zeroing-out-MKL-pooling-buffer: In the previous, IntelCaff…
bb3c2a9
task-disable-mkl-reshape: Disable unit test for mkl reshape
tsocha b4f889c
moved simply from examples/ssd to models/intel_optimized_models/ssd
sfraczek eb32d59
modified paths in SSD's solvers.prototxt to point to new train/val pr…
sfraczek c1fd4e2
Revert "In the previous, IntelCaffe had to zero diff for the bottom b…
tsocha 9a40841
task-Fix-mkl-googlenet-NaN: Revert "In the previous, IntelCaffe had t…
tsocha 0d13707
- Fix to MKLDNN split when tops diffs comes at diffrent formats (first
jczaja 3dff92d
task-mkldnn-split-fix: - Fix to MKLDNN split when tops diffs comes at…
jczaja f7b0c90
task-ssd-moved-models: modified paths in SSD's solvers.prototxt to po…
sfraczek 32eb7d6
Adds CT2.0 support for MKLDNN
Sand3r- 606b32d
Fix lint issues
Sand3r- 132c30a
Merge branch 'intelcaffe'
hshen14 4908f1b
task-ct2.0-mkldnn: Add CT2.0 support for MKLDNN
Sand3r- afb4b5b
test
daisyden f817dd5
Update the support of CompilationRuleOne with MKLDNN.
4d9ae6c
Merge branch 'intelcaffe'
hshen14 2ed2106
Update the code with Jacek's comment.
d1c2851
task-extend-CompilationRuleOne-for-MKLDNN-v2: Update the support of C…
34ad973
cleanup multi-node directories and files
apwojcik 9633d72
remove MPI dependencies
apwojcik 58fe50f
remove DISTR_WEIGHT_UPDATE
apwojcik fd1e271
upgrade to MLSL 2017 Preview release
apwojcik 4fdaf34
remove model parallelism experimental code
apwojcik c117572
update for CentOS compiler
apwojcik 0795782
style cleanup for preprocesor directives
apwojcik ee2656f
remove empty operations
apwojcik c021b77
added MLSL statistics
apwojcik 9a755bb
allow MLSL operations only for TRAIN
apwojcik 1e2f27f
add MLSL support for MKLDNN
apwojcik f96b759
task-upgrade-mlsl-2017-preview: add MLSL support for MKLDNN
apwojcik a0a6fd8
Fix the bug that the stride, pad, kernel width and height should reve…
ba8614d
Merge branch 'intelcaffe'
hshen14 9d93a03
Fix the reverse of w and h in MKLDNN convolution layer's stride, kern…
e46d591
task-fix-reverse-stride-pad-kernel-in-MKLDNN-convolution: Fix the bug…
fb54d02
task-fix-reverse-stride-pad-kernel-in-MKLDNN-convolution-Add-unittest…
c52cbd7
Compare-collect tool does not check/report data used in in-place comp…
tpatejko 614c605
task-collect-compare-inplace-computations: Compare-collect tool does …
tpatejko 5ce744b
Fix memory leaks in hdf5, ip and spp tests
Sand3r- 161b2fa
Remove whitespace at eol at spp test
Sand3r- 102bf13
fix issue ICL-64
20299f2
Fix the threshod of failed unit test.
2f63616
Enable googlenet-v2, fix the bias in the innerproduct, fix the batchn…
6d6ee48
Disable the average pooling unit test.
79eace3
Just update the mkldnn to the version to involve the fix of inner pro…
0b043e4
Add LICENSE file for resnet
daisyden abac898
task-fix-test-mem-leaks: Fix memory leaks in hdf5, ip and spp tests
Sand3r- 34c4459
task-enable-googlenet-v2-with-innerproduct-and-batchnorm-fix: Just up…
8ee9fad
Add the unit tests for MKLDNN inner product layer to cover the cases …
0ba59d4
Enable the ResNet50.
36ea405
Eanble ResNet-50 in shanghai local repository.
409699c
Add a empty line to push the patch for trying to trigger the PRE-CI.
d6574e4
Merge branch 'intelcaffe'
hshen14 5ff1e2f
Fix the customer issue of Concat layer. Remove the unnecessary reshap…
3e255f5
Fix the customer issue of Concat layer. Remove the unnecessary reshap…
8b1c616
Merge changes I409699c3,I36ea405b
bcf15e7
Merge "Fix the customer issue of Concat layer. Remove the unnecessary…
dca65f1
task-add-unittests-mkldnn-innerproduct-layer-without-bias: Add the un…
8000c6a
task-enable-ResNet50: Enable the ResNet50.
ff1c871
Enabling mkldnn convolution tests after qb was updated
tpatejko 9122ba9
task-mkldnn-convolution-tests-enabled: Enabling mkldnn convolution te…
tpatejko 219ebcd
task-fix-concat-layer-for-customer-issue: Fix the customer issue of C…
364af6f
Merge DCG C1
hshen14 1c69b49
Merge branch 'intelcaffe'
hshen14 db951c0
Fix the issue IntelCaffe builds MKL-DNN without MKL GEMM
caf5c42
Fix Eltwise Layer: Incorrect top data memory format passed when creat…
aa92506
Upgrade MKLDNN commit id
hshen14 b0ef323
Upgrade MKL-DNN to version ecf1883a94239a19d442356d32e1076a15b88e7a
hshen14 57c0277
Merge "Fix Eltwise Layer: Incorrect top data memory format passed whe…
dc10f0b
Upgrade MKL to 20170425
hshen14 5e87a13
Merge branch 'master' into dcg_c1
hshen14 9f6924b
Fix Extra and Incorrect Reorder Primitives executed in FWD/BWD Convol…
03101d3
Comment average pooling unit test for further debugging.
badec0f
Merge branch 'master' into dcg_c1
hshen14 172ade8
Merge branch 'dcg_c1' of ssh://git-ccr-1.devtools.intel.com:29418/dl_…
hshen14 0c52087
Merge branch 'master' into dcg_c1
hshen14 348359f
fix issue ICL-64
4a0c837
Fix the threshod of failed unit test.
a02fdc9
Add LICENSE file for resnet
daisyden 9697e63
Change the wordings in the comments.
0b9431c
Merge "Change the wordings in the comments."
24e64f4
Use MKL 0425 link in Caffe github
hshen14 122557c
Fix the issue that Bwd LRN desc is created with incorrect src memory …
0ca7a6c
Warning removal
hshen14 283fae1
test gerrit
61d99ed
Revert "test gerrit"
5bc1e99
Relu optimization related to ICL-84.
cef51e9
Merge "Relu optimization related to ICL-84."
c7941f0
Fix Pooling bwd executed using non-optimized layout for diff data.
211915e
Further improve the Relu performance when prv top diff format is diff…
845985e
Optimize the LRN layer bwd using optimized layout for bottom diff data.
e2e7332
Fix Googlenet v3 Caffe time test aborted with MKLDNN engine.
b8cb4f5
Merge branch 'release_1.0.0_rc2' of ssh://git-ccr-1.devtools.intel.co…
daisyden 9e6c1ba
Align MKL-DNN external latest commit id: a7e17b753c622906f8bdc78f8510…
hshen14 93a9e4d
fix icl-40 Test phase at the end of training was removed. Remove goog…
daisyden 439a1c3
fix a bug of mkldnn integration: no need to call prv descriptor when …
jgong5 d5b6220
1. Merge the fix of a bug of mkldnn integration: no need to call prv …
37c1261
enable model parallelism and sync bn statistics and scale/shift
jgong5 8bd0ca2
simplify ByPass logic
jgong5 23cbe39
Support customized batch size and engine for cpp_classification.
8e32af1
1. Update the mkldnn version to fix the ResNet regression and improve…
ee1cfe5
Merge "enable model parallelism and sync bn statistics and scale/shift"
jgong5 487ee9e
Merge "simplify ByPass logic"
jgong5 a09179f
Merge "use MLSL_DEFAULT_COLOR to replace -1"
jgong5 401413e
Merge "disable model parallelism for cross-channel LRN"
jgong5 f636ba6
Merge "1. Update the mkldnn version to fix the ResNet regression and …
98622e2
use MLSL_DEFAULT_COLOR to replace -1
jgong5 ade99a1
disable model parallelism for cross-channel LRN
jgong5 178f530
Add the shell script to support the automatic downloading and install…
ad7ace9
1. Update the prepare_mlsl script for return value. 2. Enable the mls…
35ddee9
Update the prepare_mlsl shell script to handle the case that MLSLROOT…
70aaf4b
Update the prepare_mlsl.sh for more complex situations of customized …
77cddab
Fix the asum_data() logic of SYNCED_PRV, fix ICL-74.
9ea6387
Fix batch normalization global status checker; add optimization for b…
hshen14 e58f238
1. Change the tab format. 2. Change the MLSLROOT to MLSL_ROOT.
6a61456
Add option for average pooling include/exclude padding, and default i…
31ef95e
Fix the problem that VERSION_LINE may not be an integer if the MLSL_R…
5bb522e
fix BN bug of multinode
jgong5 88b9626
Merge "Fix the asum_data() logic of SYNCED_PRV, fix ICL-74."
07c6fc9
Merge changes I31ef95ee,Ie58f2388
3cd80b4
Merge "Fix batch normalization global status checker; add optimizatio…
hshen14 498286a
Merge changes Ib9bd590d,I5bb522e4
jgong5 cfa6761
Fix (ICL-128) GoogleNet-V2 inference resut with MKLDNN engine will be…
385e770
Update Dockfile
hshen14 d4117e4
Add Winograd support
hshen14 b9bd590
Merge branch 'master' of ssh://git-ccr-1.devtools.intel.com:29418/dl_…
jgong5 da63fb3
Add mkl_deconvolution layer and corresponding unitest
c33d743
Merge "Add mkl_deconvolution layer and corresponding unitest"
fde50b1
Add RemoveBNScale feature
7a01fb1
Fix bug in RemoveBNScale feature
28cf6b8
Fix Winograd fallback to direct convolution
hshen14 792b0d1
Fix bug in RemoveBNScale feature with GoogleNetv2
881163a
Fix bug in RemoveBNScale
97cd241
Fix bug in RemoveBNScale
9f0108f
Merge branch 'release_1.0.1_rc1' of ssh://git-ccr-1.devtools.intel.co…
daisyden b0d1651
Update and refine the shell script to download the MLSL release and h…
5b13124
Merge branch 'release_1.0.1_rc1' of ssh://git-ccr-1.devtools.intel.co…
daisyden
The table of contents is too big for display.
Oops, something went wrong.