提交 · dev_io_standby_buf · Oneflow-Inc / oneflow

23 5月, 2018 1 次提交
- W
  two buf in one stream · 6529a7c5
  由 willzhang4a58 提交于 5月 23, 2018
```
Former-commit-id: dcf1b183c3b86ac547f9ac4581e2d05de0feb853
```
  6529a7c5
21 5月, 2018 5 次提交
- W
  1_machine_4_gpu.benchmark · 51065dc5
  由 willzhang4a58 提交于 5月 21, 2018
```
Former-commit-id: 83fad21e
```
  51065dc5
- W
  alexnet report · fa972888
  由 willzhang4a58 提交于 5月 21, 2018
```
Former-commit-id: 7b0ad6f0
```
  fa972888
- J
  align alexnet to tf_benchmark (#893) · 4105fed1
  由 Jinhui Yuan 提交于 5月 21, 2018
```
* align alexnet to tf_benchmark

* batch size 480 for 2gpu

* fix resize preprocess


Former-commit-id: 7e57b20f
```
  4105fed1
- W
  impl cudaCallBack by cuda_event · cac50d7a
  由 willzhang4a58 提交于 5月 21, 2018
```
Former-commit-id: 33a36d9b
```
  cac50d7a
- W
  CudaWorkType · c61a7e05
  由 willzhang4a58 提交于 5月 21, 2018
```
Former-commit-id: 33b2e550
```
  c61a7e05
19 5月, 2018 3 次提交

W
4 thread for 1 gpu · 185682f0
由 willzhang4a58 提交于 5月 19, 2018
```
Former-commit-id: e9b4bded
```
185682f0

由 Jinhui Yuan 提交于 5月 19, 2018

* use cudnn 2d pooling when possible

* fix set pooling 2d desc

* another way to set cudnn pooling tensor desc

* remove useless code

* solve 1D pooling with 3D cudnn api

* unify 2d and 3d cudnn polling config

* fix size_t in decreasing loop index

* refine pooling


Former-commit-id: 891a51a9

90f197de

J
update prototxt of benchmark (#884) · 046476d2
由 Jinhui Yuan 提交于 5月 19, 2018
```
Former-commit-id: 36de5ce1
```
046476d2

18 5月, 2018 6 次提交
- L
  fast concat (#883) · 21dc7340
  由 Li Xinqi 提交于 5月 18, 2018
```
Former-commit-id: 31b5df8f
```
  21dc7340
- C
  remove fw_buf and bw_buf (#882) · 2bd57e63
  由 chengtbf 提交于 5月 18, 2018
```
Former-commit-id: 992c9ade
```
  2bd57e63
- W
  of_submit · 903022f8
  由 willzhang4a58 提交于 5月 18, 2018
```
Former-commit-id: 8f2e09b5
```
  903022f8
- L
  fast softmax kernel (#871) · 57113d6e
  由 Li Xinqi 提交于 5月 18, 2018
```
Former-commit-id: bb1c1ec2
```
  57113d6e
- L
  Rename model tmp to const model (#878) · 2a4c765b
  由 leaves-zwx 提交于 5月 18, 2018
```
Former-commit-id: 12b4e61b
```
  2a4c765b
- W
  -novirtual is_write_ready · b5836447
  由 willzhang4a58 提交于 5月 18, 2018
```
Former-commit-id: 9ce67155
```
  b5836447
17 5月, 2018 10 次提交
- W
  let cudnn_buf use device_context::buf · 510e3fe0
  由 willzhang4a58 提交于 5月 17, 2018
```
Former-commit-id: 1570b052
```
  510e3fe0
- W
  add buf_size in device_context · a5fbf8d2
  由 willzhang4a58 提交于 5月 17, 2018
```
Former-commit-id: bc7726be
```
  a5fbf8d2
- C
  id_manager_test (#880) · 1ef91667
  由 chengtbf 提交于 5月 17, 2018
```
Former-commit-id: f4c827a5
```
  1ef91667
- W
  direct send msg instead of add_cuda_callback when on same work stream · 7c0183a9
  由 willzhang4a58 提交于 5月 17, 2018
```
Former-commit-id: 16b68fb4
```
  7c0183a9
- W
  remove work_stream_id in device_context · d7e8a66f
  由 willzhang4a58 提交于 5月 17, 2018
```
Former-commit-id: b247082e
```
  d7e8a66f
- C
  global/local work stream id (#879) · fa4af660
  由 chengtbf 提交于 5月 17, 2018
```
Former-commit-id: 55579820
```
  fa4af660
- W
  add buf to device_context · 6b3cf7a0
  由 willzhang4a58 提交于 5月 17, 2018
```
Former-commit-id: 19cd02b4
```
  6b3cf7a0
- W
  refine device context · e8cf4b55
  由 willzhang4a58 提交于 5月 17, 2018
```
Former-commit-id: 8ea2b4ee
```
  e8cf4b55
- W
  feat: compile buf size · ca00c630
  由 willzhang4a58 提交于 5月 17, 2018
```
Former-commit-id: ad5182e9
```
  ca00c630
- L
  Dev decrease writeable regst desc num (#876) · 5f417b59
  由 Li Xinqi 提交于 5月 17, 2018
```
* decrease actual writeable produced regst desc num

* refine type: uint64_t => int64_t


Former-commit-id: d28e8887
```
  5f417b59
16 5月, 2018 5 次提交
- L
  const buf (#874) · d2977a1a
  由 Li Xinqi 提交于 5月 16, 2018
```
Former-commit-id: 1b7c94e7
```
  d2977a1a
- W
  fix: mutli-machine use same addr · 0cd96312
  由 willzhang4a58 提交于 5月 16, 2018
```
Former-commit-id: 34b9b26d
```
  0cd96312
- W
  refine comm_net · 0e236153
  由 willzhang4a58 提交于 5月 16, 2018
```
Former-commit-id: fa235709
```
  0e236153
- W
  fix: deadlock because of poll_exit_flag · be18202b
  由 willzhang4a58 提交于 5月 16, 2018
```
Former-commit-id: 9ef8f304
```
  be18202b
- W
  refine ibverb comm_network (#875) · 295363f6
  由 Will Zhang 提交于 5月 16, 2018
```
Former-commit-id: b58f251c
```
  295363f6
15 5月, 2018 1 次提交
- D
  fix example dir (#872) · b3fb811f
  由 dssgsra 提交于 5月 15, 2018
```
* fix example dir

* update


Former-commit-id: 6c66c7c4
```
  b3fb811f
14 5月, 2018 2 次提交

Y
fix reduce add kernel bug (#873) · 35afe4e6
由 Yi Zhu 提交于 5月 14, 2018
```
Former-commit-id: 0779c01c
```
35afe4e6

Refine update (#869) · a300d6ed

由 Jinhui Yuan 提交于 5月 14, 2018

* initial version for new model update logic

* remove l2 regularization api in kernel base class

* add naive_model_update_kernel.cu

* update: momentum and rms

* refine rms comment

* refine momentum

* fix model update formula

* minor refine

* refactor model_update


Former-commit-id: d4d36eeb

a300d6ed

11 5月, 2018 1 次提交

Updata softmax kernel and sparse softmax xent loss kernel test (#865) · d6593372

由 leaves-zwx 提交于 5月 11, 2018

* Updata softmax kernel and sparse softmax xent loss kernel test

* Move set_piece_size to OpKernelTestCase constructor
Move set_device_type to OpKernelTestCase::InferBlobDesc
Update sparse_cross_entropy_loss_kernel_test
Update relu_kernel_test
Replace __LOC__ macro with FILE_LINE_STR


Former-commit-id: cc67a58b

d6593372

10 5月, 2018 3 次提交
- Y
  Refine RDMA implementation (#866) · 1ebddfe6
  由 Yi Zhu 提交于 5月 10, 2018
```
* save work

* refine code


Former-commit-id: f5d3b072
```
  1ebddfe6
- W
  fix: reduce loss with no axis · bec29847
  由 willzhang4a58 提交于 5月 10, 2018
```
Former-commit-id: 34db5d96
```
  bec29847
- C
  fw_buf_blob and bw_buf_blob , conv_op 2 cudnn buf (#863) · a39b6f09
  由 chengtbf 提交于 5月 10, 2018
```
* fw_buf_blob and bw_buf_blob , conv_op 2 cudnn buf

* ConvOpCtx


Former-commit-id: 509027c0
```
  a39b6f09
09 5月, 2018 2 次提交
- N
  fix: MdDiffAccLogicalNode should connect to MdUpdt one-to-onely without... · efb8d2e6
  由 Niu Chong 提交于 5月 09, 2018
```
fix: MdDiffAccLogicalNode should connect to MdUpdt one-to-onely without ReduceStruct when ModelParallel (#864)



Former-commit-id: c6654971
```
  efb8d2e6
- W
  set default piece_exp=19*batch · 3d175902
  由 willzhang4a58 提交于 5月 09, 2018
```
Former-commit-id: 62858b04
```
  3d175902
08 5月, 2018 1 次提交
- Y
  Fix lrn nchw cpu bug (#862) · f67da814
  由 Yi Zhu 提交于 5月 08, 2018
```
* fix lrn nchw cpu bug

* refine code


Former-commit-id: 7b4cc966
```
  f67da814

Oneflow-Inc / oneflow 上一次同步 2 年多

Oneflow-Inc / oneflow
上一次同步 2 年多