提交 · 4d530ed01448447308d563cdccf32d09df44290d · testgroup / pytensor · GitLab

31 10月, 2013 33 次提交
- Add a Gpu version of the reduction op (only supports add and multiply for now). · 4d530ed0
  由 Arnaud Bergeron 提交于 9月 18, 2013
  
  4d530ed0
- Add c_code() to GpuAlloc(). · 6e74a08b
  由 Arnaud Bergeron 提交于 9月 17, 2013
  
  6e74a08b
- Use faster version of the GPU -> host copy. · 4c2e0344
  由 Arnaud Bergeron 提交于 9月 17, 2013
  
  4c2e0344
- Remove traces of the wrapper from GpuArrayType. · 1c762aa8
  由 Arnaud Bergeron 提交于 9月 16, 2013
  
  1c762aa8
- After discussion, remove the wrapper thing since this is contrary to the direction we want. · 39be617c
  由 Arnaud Bergeron 提交于 9月 16, 2013
```
This reverts commit 326b84115cc08ff9a2539e7ffd15d8ec840677ca.
```
  39be617c
- Rework op_lifter to cover more cases and add some new op optimizations. · 95e906d8
  由 Arnaud Bergeron 提交于 9月 15, 2013
  
  95e906d8
- In response to pygpu change: · 631960c6
  由 Arnaud Bergeron 提交于 9月 15, 2013
```
GpuArray_default_context() -> pygpu_default_context()
```
  631960c6
- Fix DeepCopyOp C code. · 2cfc4c11
  由 Arnaud Bergeron 提交于 9月 15, 2013
  
  2cfc4c11
- Add a pretty print for GpuArrayType. · e1c763e1
  由 Arnaud Bergeron 提交于 9月 15, 2013
  
  e1c763e1
- Add a config variable to force ops to work synchronously. · aeac604e
  由 Arnaud Bergeron 提交于 9月 15, 2013
  
  aeac604e
- Make sure to sync a GpuArray before passing it on as a CudaNdarray. · d2d012e2
  由 Arnaud Bergeron 提交于 9月 15, 2013
  
  d2d012e2
- Replace Alloc by a shell around the tensor one. · 95725506
  由 Arnaud Bergeron 提交于 9月 15, 2013
  
  95725506
- Make HideC actually work. · 3f9dc6c6
  由 Arnaud Bergeron 提交于 9月 15, 2013
  
  3f9dc6c6
- Make local_dimshuffle_lift work with subclasses for DimShuffle. · 69e5ee16
  由 Arnaud Bergeron 提交于 9月 13, 2013
  
  69e5ee16
- Make sure to properly hide the tensro C implementations. · dd91c021
  由 Arnaud Bergeron 提交于 9月 13, 2013
  
  dd91c021
- Add optimisations for GpuDimShuffle and SpecifyShape. · c12e1612
  由 Arnaud Bergeron 提交于 9月 13, 2013
  
  c12e1612
- Update to the new signature for new_GpuArray(). · a1bbaede
  由 Arnaud Bergeron 提交于 9月 13, 2013
  
  a1bbaede
- Make sure to hide the c_code for GpuElemwise. · 7459819e
  由 Arnaud Bergeron 提交于 9月 12, 2013
  
  7459819e
- Add GpuDimShuffle (both the C and python versions). · 800bf55c
  由 Arnaud Bergeron 提交于 9月 12, 2013
  
  800bf55c
- New algortihm for the perform that avoids one of the reshapes. · 68ab8d91
  由 Arnaud Bergeron 提交于 9月 12, 2013
  
  68ab8d91
- Add tests for GpuElemwise (inplace and non-inplace). · 0f1c32b5
  由 Arnaud Bergeron 提交于 9月 12, 2013
  
  0f1c32b5
- Rename optimization to gpua_ to avoid conflicts with the cuda module. · 3c63af3e
  由 Arnaud Bergeron 提交于 9月 11, 2013
  
  3c63af3e
- Enable support for inplace Elemwise (and optimizations to use it). · a8fe1069
  由 Arnaud Bergeron 提交于 9月 11, 2013
  
  a8fe1069
- Rewrite GpuElemwise to be more like the tensor version and much cleaner. · 644dbb6c
  由 Arnaud Bergeron 提交于 9月 11, 2013
  
  644dbb6c
- Use x.zeros_like(). · d5bdbb4a
  由 Arnaud Bergeron 提交于 9月 11, 2013
  
  d5bdbb4a
- Move the test for DeepCopyOp to a new test_type.py file since its code is defined in type.py. · 186360ce
  由 Arnaud Bergeron 提交于 9月 11, 2013
  
  186360ce
- Mirror tensor and move the subtensor op in its own file. · 5e1935ba
  由 Arnaud Bergeron 提交于 9月 11, 2013
  
  5e1935ba
- Clean up unneeded stuff in _operators and use the new wrapping facility. · d4d7141a
  由 Arnaud Bergeron 提交于 9月 11, 2013
  
  d4d7141a
- Allow subclasses of _tensor_py_operators to wrap the results in their own type. · 8f4208f6
  由 Arnaud Bergeron 提交于 9月 11, 2013
  
  8f4208f6
- Enable fusion of GpuElemwise ops. · fcbd4a34
  由 Arnaud Bergeron 提交于 8月 22, 2013
  
  fcbd4a34
- Enable the caching to GpuSubtensor. · 2bb9365e
  由 Arnaud Bergeron 提交于 8月 22, 2013
  
  2bb9365e
- Fix the DeepCopyOp code so that it doesn't use generic names. · 19a7fa85
  由 Arnaud Bergeron 提交于 8月 22, 2013
  
  19a7fa85
- Add GpuSubtensor. · 1584e29d
  由 Arnaud Bergeron 提交于 8月 22, 2013
  
  1584e29d
26 10月, 2013 3 次提交
- Merge pull request #1571 from lamblin/test_bug_cho · 382d2ed1
  由 Frédéric Bastien 提交于 10月 25, 2013
```
Add test for gh-1549
```
  382d2ed1
- Skip test if cuda not available · 8911a6b7
  由 Pascal Lamblin 提交于 10月 25, 2013
  
  8911a6b7
- Merge pull request #1578 from lamblin/fix_scan_merge_inouts · e6633d20
  由 Razvan Pascanu 提交于 10月 25, 2013
```
Fix scan merge inouts
```
  e6633d20
25 10月, 2013 4 次提交
- Add test case · 0a40491c
  由 Pascal Lamblin 提交于 10月 24, 2013
  
  0a40491c
- Fix case when test_values do not have shapes · 72060cc5
  由 Pascal Lamblin 提交于 10月 24, 2013
  
  72060cc5
- Fix circular import · 9defa5a2
  由 Pascal Lamblin 提交于 10月 24, 2013
  
  9defa5a2
- Add new checks in optimization, to prevent incorrect replacement. · af8f19b6
  由 Pascal Lamblin 提交于 10月 24, 2013
```
We want to avoid merging different outputs of a scan node, that compute
the same thing, but keep a different number of steps.
```
  af8f19b6