提交 · 3e4da0dcf57db39265fc6ecca51610787de8df33 · testgroup / pytensor · GitLab

16 12月, 2015 34 次提交
- Don't restrict 'small' workspace implementation of GradW to v4 and later · 3e4da0dc
  由 carriepl 提交于 11月 09, 2015
  
  3e4da0dc
- Add missing syntax · a28cb169
  由 carriepl 提交于 11月 09, 2015
  
  a28cb169
- Fix verification of spatial sizes in cudnn conv (gpuarray backend) · 757b04c0
  由 carriepl 提交于 10月 27, 2015
  
  757b04c0
- Fix condition based on CuDNN version · fdffa58e
  由 carriepl 提交于 10月 27, 2015
  
  fdffa58e
- Call cudnnGetConvolution2dDescriptor instead of cudnnGetConvolutionNdDescriptor · 6f1dc3d8
  由 carriepl 提交于 10月 26, 2015
  
  6f1dc3d8
- Add comments and fix condition in GpuDnnConvGradW C code (gpua backend) · bf918482
  由 carriepl 提交于 10月 22, 2015
  
  bf918482
- Update GpuDnnConvGradI for CuDNN v4 (gpua backend) · 2a857e3a
  由 carriepl 提交于 10月 22, 2015
  
  2a857e3a
- Update GpuDnnConvGradW for CuDNN v4 (gpua backend) · 1e48b734
  由 carriepl 提交于 10月 22, 2015
  
  1e48b734
- Update GpuDnnConv for CuDNN V4 (gpua backend) · 64439f41
  由 carriepl 提交于 10月 22, 2015
  
  64439f41
- Define new V4 elements from V3 elements in cudnn_helper (gpua backend) · 93f6f441
  由 carriepl 提交于 10月 22, 2015
  
  93f6f441
- Add precision param to dnnConv (gpua backend) · 674cd4f1
  由 carriepl 提交于 10月 22, 2015
  
  674cd4f1
- Add precision param to GpuDnnConvDesc · 46c83387
  由 carriepl 提交于 10月 22, 2015
  
  46c83387
- Add definition for new implementation in cudnn_helper.h (cuda backend) · 5a015a8d
  由 carriepl 提交于 10月 22, 2015
  
  5a015a8d
- Retain compatibility with CuDNN v2 (cuda backend) · 330da07d
  由 carriepl 提交于 10月 22, 2015
  
  330da07d
- Add checks to make sure backward tiled-fft is not used when it shouldn't (cuda backend) · cde07713
  由 carriepl 提交于 10月 22, 2015
  
  cde07713
- Add checks to make sure tiled-fft is not used when it shouldn't (cuda backend) · b177f3a8
  由 carriepl 提交于 10月 22, 2015
  
  b177f3a8
- Limit new convolution implementations to CuDNN V4 · cf26ae5c
  由 carriepl 提交于 10月 22, 2015
  
  cf26ae5c
- Make dnn_conv and dnn_conv3d support param precision (cuda backend) · 2a84aa6b
  由 carriepl 提交于 10月 20, 2015
  
  2a84aa6b
- Add new precision parameter to GpuDnnConvDesc (cuda backend) · 0ed16db1
  由 carriepl 提交于 10月 20, 2015
  
  0ed16db1
- Add new flag to control CuDNN convolution precision · 3914bf65
  由 carriepl 提交于 10月 20, 2015
  
  3914bf65
- modify cuda/dnn.py to support new convolution implementations · c36f9a29
  由 carriepl 提交于 10月 20, 2015
  
  c36f9a29
- Deprecate algo_bwd flag in favor of algo_bwd_data and algo_bwd_filter · 13bad511
  由 carriepl 提交于 10月 20, 2015
  
  13bad511
- fix typos · fc7d7277
  由 carriepl 提交于 10月 20, 2015
  
  fc7d7277
- start new cudnn v4 convolution mode · e200ca29
  由 Frederic 提交于 10月 20, 2015
  
  e200ca29
- Remove things specific to GpuConv2d in tests · bcb9318e
  由 Pascal Lamblin 提交于 12月 15, 2015
  
  bcb9318e
- Better error message when abstract conv does not get replaced · 757f0d20
  由 Pascal Lamblin 提交于 12月 15, 2015
  
  757f0d20
- Do not try to do constant folding on abstract ops · e6d7fbb8
  由 Pascal Lamblin 提交于 12月 15, 2015
  
  e6d7fbb8
- Remove tests for logical shapes · b5472f1f
  由 Pascal Lamblin 提交于 12月 15, 2015
  
  b5472f1f
- Raise error if logical shapes are passed. · c9cdbb10
  由 Pascal Lamblin 提交于 12月 15, 2015
```
This would change the actual output, so we cannot simply ignore them.
```
  c9cdbb10
- Give more informative warning messages · 85ca50fa
  由 Pascal Lamblin 提交于 12月 15, 2015
  
  85ca50fa
- Run conv2d tests with the default interface as well · c482805d
  由 Pascal Lamblin 提交于 12月 14, 2015
  
  c482805d
- Add infer_shape to abstract convolutions · b9c51f28
  由 Pascal Lamblin 提交于 12月 14, 2015
  
  b9c51f28
- Merge pull request #3737 from lamblin/gpuarray_abstractconv · 6d4633be
  由 Frédéric Bastien 提交于 12月 15, 2015
```
Gpuarray abstractconv
```
  6d4633be
- Merge pull request #3783 from nouiz/profilemode · 58cd0ac8
  由 carriepl 提交于 12月 15, 2015
```
Fix crash with mode=ProfileMode fix gh-3782
```
  58cd0ac8
15 12月, 2015 6 次提交
- Fix crash with mode=ProfileMode fix gh-3782 · 7270f65b
  由 Frederic 提交于 12月 15, 2015
  
  7270f65b
- Merge pull request #3774 from lamblin/abstractconv_fixes · 24ace594
  由 Frédéric Bastien 提交于 12月 15, 2015
```
Fix broadcastable pattern of gradient in abstract conv
```
  24ace594
- Merge pull request #3772 from gw0/feat-pydot-fallback · 458e1594
  由 Frédéric Bastien 提交于 12月 15, 2015
```
Import pydot-ng with pydot as fallback.
```
  458e1594
- Update pydot dependency check in tests. · 2fa6f48e
  由 gw0 [http://gw.tnode.com/] 提交于 12月 15, 2015
  
  2fa6f48e
- Add test case reported · 0f5f1890
  由 Pascal Lamblin 提交于 12月 12, 2015
  
  0f5f1890
- Fix broadcastable pattern of gradient in abstract conv · 1705a123
  由 Pascal Lamblin 提交于 12月 12, 2015
  
  1705a123