Commit f76b21e
Torch input validation (#149)
* add more kernel dims
* simple test for torch-level shape checks
* add TORCH_CHECK to torch interface
* adds size checking, removes getters and replaces with const attributes.
* adds kernel dims to jit via dict
* replace get_L3_dim with L3_dim
* remove parenthesis
* add a central location for type enums
* add location and type checking. expose LX_dim correctly. Pass types as enums
* pass additional info
* better enum and proper transpose_perm size
* better enum
* better enum
* factor out executables
* rename to shape_test
* add lazy fixtures so that I can easily parameterize tuples. lazy-fixture is un maintained. I can remove in preferred, but it is convenient.
* add enum to torch.dtype
* factor these into their own file for reuse
* added conftest for fixture reuse
* factor out executables
* delete utils.py
* content moved elsewhere
* moved stream test to torch_features level
* wip commit
* retain only core changes
* just forward direction testing, added pytest-subtests to to make testing all the different buffers and buffer dimensions easier.
* return double_backward_rawptrs which got removed accidentally
* add other convs and add permutation where expected
* Some cleanup.
* Cleanup.
* Removed some fixtures that are only used once / reproducibility on generator.
* Light renaming.
* Linted.
---------
Co-authored-by: Austin Glover <austin_glover@berkeley.com>
Co-authored-by: Vivek Bharadwaj <vivek_bharadwaj@berkeley.edu>1 parent 7105978 commit f76b21e
6 files changed
Lines changed: 399 additions & 53 deletions
File tree
- openequivariance
- extension
- implementations
- convolution
- tests
0 commit comments