提交 · dc7f7a4f416cfe5756fcf1abd0f0ed680f5dc6bc · testgroup / pytensor

01 6月, 2017 3 次提交
- GpuAdvancedIncSubtensor1_dev20 doesn't deal correctly with mixed type inputs. · dc7f7a4f
  由 Arnaud Bergeron 提交于 5月 31, 2017
  
  dc7f7a4f
- Actually test gemv, not gemm. · 87a5e2bd
  由 Arnaud Bergeron 提交于 5月 31, 2017
  
  87a5e2bd
- Remove last reference to inplace_increment. · d763c819
  由 Arnaud Bergeron 提交于 5月 31, 2017
  
  d763c819
30 5月, 2017 6 次提交
- Flake8 fix. · fd9d1e65
  由 Arnaud Bergeron 提交于 5月 29, 2017
  
  fd9d1e65
- Restore float16 priviledges for GpuAdvancedIncSubtensor1. · e3aff3f5
  由 Arnaud Bergeron 提交于 5月 29, 2017
  
  e3aff3f5
- Remove inexistant types. · b5008b0f
  由 Arnaud Bergeron 提交于 5月 29, 2017
  
  b5008b0f
- Stop skipping tests when there is no compiler. · 56b57f14
  由 Arnaud Bergeron 提交于 5月 29, 2017
  
  56b57f14
- Remove inplace_increment from cutils_ext. · 46c9f38a
  由 Arnaud Bergeron 提交于 5月 29, 2017
  
  46c9f38a
- Use numpy add ufunc instead of custom code. · c60bd3b2
  由 Arnaud Bergeron 提交于 5月 29, 2017
  
  c60bd3b2
16 5月, 2017 10 次提交
- Disable c code for some float16 GPU ops. · e04074a9
  由 Frederic Bastien 提交于 5月 10, 2017
  
  e04074a9
- Make verify grad work well with float16 · ee0051da
  由 Frederic Bastien 提交于 5月 10, 2017
  
  ee0051da
- Crash fix in GpuIncSubtensor andd make *subtensor* tests use the GPU mode more frequently. · 2ce65dc3
  由 Frederic Bastien 提交于 5月 10, 2017
  
  2ce65dc3
- flake8 · 63ef2ead
  由 Frederic Bastien 提交于 5月 07, 2017
  
  63ef2ead
- Add more float16 ops. Test to refactor. · 9067f773
  由 Frederic Bastien 提交于 5月 07, 2017
  
  9067f773
- gpu opt, don't move float16 ops or cast to/from float32 in the graph for some… · 81245414
  由 Frederic Bastien 提交于 5月 07, 2017
```
gpu opt, don't move float16 ops or cast to/from float32 in the graph for some ops that don't support float16
```
  81245414
- Use gpugemm for gemv in float16. fix gh-4341 · 8735401b
  由 Frederic Bastien 提交于 5月 07, 2017
  
  8735401b
- Fix warning and add dtype check in ops that only support float32 · 27a1361d
  由 Frederic Bastien 提交于 5月 07, 2017
  
  27a1361d
- Enable float16 ops on ops that don't care about dtype or call copy that both back-end support. · 99363898
  由 Frederic Bastien 提交于 5月 07, 2017
  
  99363898
- Do MaxAndArgmax on the GPU by casting to/from float32/float16 · a0c90645
  由 Frederic Bastien 提交于 5月 07, 2017
  
  a0c90645
06 5月, 2017 6 次提交
- Merge pull request #5921 from lamblin/fix_5919 · 71792827
  由 Frédéric Bastien 提交于 5月 06, 2017
```
Remove bug in optimization
```
  71792827
- Ignore all previous bug warnings in buildbots · bf9cf619
  由 Pascal Lamblin 提交于 5月 05, 2017
  
  bf9cf619
- Remove bug in optimization. Add warning. Update test · df3e83d1
  由 Pascal Lamblin 提交于 5月 05, 2017
  
  df3e83d1
- Merge pull request #5902 from lamblin/fix_debugmode · 26254645
  由 Frédéric Bastien 提交于 5月 05, 2017
```
[BUG, CRASH] Fixes in DebugMode for GPU
```
  26254645
- Speed up choice with replace on the GPU. · be375941
  由 Frederic Bastien 提交于 5月 05, 2017
  
  be375941
- 2 bug fixes. Make choice on the GPU with replace work well. · bb27565a
  由 Frederic Bastien 提交于 5月 05, 2017
  
  bb27565a
04 5月, 2017 12 次提交
- Install libgpuarray / pygpu in nightly DebugMode build · 3a4d1d15
  由 Pascal Lamblin 提交于 4月 27, 2017
  
  3a4d1d15
- Disable C code called from perform in float16 · 5ad68a1d
  由 Pascal Lamblin 提交于 4月 27, 2017
```
The addition gave non-sensical results.
```
  5ad68a1d
- Normalize 1 scalar instead of a full vector · 20b23e6e
  由 Pascal Lamblin 提交于 4月 27, 2017
```
It also fixes a race condition.
```
  20b23e6e
- Relax tolerance for GpuCumOp test · aa2bb423
  由 Pascal Lamblin 提交于 4月 27, 2017
  
  aa2bb423
- Do the setup only once for all tests · f6e220a3
  由 Pascal Lamblin 提交于 4月 27, 2017
  
  f6e220a3
- Disable DebugMode.check_isfinite when we know there are inf · 8cd973ad
  由 Pascal Lamblin 提交于 4月 27, 2017
  
  8cd973ad
- Avoid int16 overflow on GPU in test · e8a8b936
  由 Pascal Lamblin 提交于 4月 27, 2017
```
CPU wraps around, but GPU returns max_int16, so DebugMode was
complaining.
```
  e8a8b936
- Fix crash in GpuGemv.perform with zero-size inputs · 0582554d
  由 Pascal Lamblin 提交于 4月 27, 2017
  
  0582554d
- Fix c code using gpu_dot for gemv · 5cb343e6
  由 Pascal Lamblin 提交于 4月 27, 2017
```
Pass the right variable when not working inplace.
```
  5cb343e6
- Skip other GPU check in DebugMode · bf9ce78c
  由 Pascal Lamblin 提交于 4月 27, 2017
  
  bf9ce78c
- Remove check on GPU in DebugMode · 582210c7
  由 Pascal Lamblin 提交于 4月 27, 2017
```
The corresponding CPU test does the same
```
  582210c7
- Fix crash of set_value when the destination isn't contiguous (#5913) · fbb066c4
  由 Frédéric Bastien 提交于 5月 03, 2017
  
  fbb066c4
03 5月, 2017 1 次提交
- Merge pull request #5903 from lamblin/fix_5898 · ddafc3e2
  由 Pascal Lamblin 提交于 5月 02, 2017
```
[BUG,CRASH] Fix crash in GPU advanced indexing with no arrays
```
  ddafc3e2
02 5月, 2017 2 次提交
- Merge pull request #5909 from jeffdonahue/fix-numpy-name · d6dbce90
  由 Frédéric Bastien 提交于 5月 02, 2017
```
minor fix to test_nnet.py
```
  d6dbce90
- fix test_nnet.py NameError caused by use of old `numpy` instead of `np` · a88614e0
  由 Jeff Donahue 提交于 5月 01, 2017
  
  a88614e0