pytorch/xla

Pull Requests Commits

Count XLA tensors.

dlibenzi committed 6 years ago

704f59ff

Bump compilation caches up and made them environment configurable.

dlibenzi committed 6 years ago

33358305

Fix selected slice shape.

dlibenzi committed 6 years ago

08f8d6f8

Bound the step for slices.

dlibenzi committed 6 years ago

0e9c6f6c

Remove TensorData and use DeviceData back as it is no more needed.

dlibenzi committed 6 years ago

a4b7006e

Added Select and Unselect operations to handle slices with strides != 1.

dlibenzi committed 6 years ago

eb316286

Enable the model comparator to be left active and unconfigured, resulting in noop.

dlibenzi committed 6 years ago

470c9444

Added aten::triangular_solve op

vishwakftw committed 6 years ago

3e5051fb

Fix XLA tensor views which are not handled by the XLA tensor core.

dlibenzi committed 6 years ago

3ff502ba

Minor cleanup of model comparator.

dlibenzi committed 6 years ago

aae5e5cf

Added model comparator Python module.

dlibenzi committed 6 years ago

1dfe3e2b

Fix aten::clamp to use lowest() instead of min().

dlibenzi committed 6 years ago

7746ff65

Replace IntList with IntArrayRef

asuhan committed 6 years ago

90849b0f

Add index_{add, copy} and their inplace versions to ATen XLA tensor

asuhan committed 6 years ago

78a09d93

No need to add CRS for one device replications.

dlibenzi committed 6 years ago

963e6fa6

Added script to normalize the IR graph text output, to allow textual diffs.

dlibenzi committed 6 years ago

5686da1d

Enable shape caching for IndexOp

asuhan committed 6 years ago

d8be8bd9

Replace Shape::dimensions_size() uses with Shape::rank()

asuhan committed 6 years ago

717db87f

Add _unsafe_view to ATen XLA tensor

asuhan committed 6 years ago

aa597be8

Pin MKL to older version as the new one which shipos with Conda breaks the PyTorch build.

dlibenzi committed 6 years ago

4455e8ef

Add index_fill to ATen XLA tensor

asuhan committed 6 years ago

d19ba5d2

Avoid trimming IR graph during sync.

dlibenzi committed 6 years ago

ea6bda64

Fix issues in https://github.com/pytorch/xla/issues/501 related to PT operations which should have been views into the buffers.

dlibenzi committed 6 years ago

079d1240

Use a global apply tensors graph cache, to boost callers other than module.cpp.

dlibenzi committed 6 years ago

fb922759

Add support to auto-generate *_out() functions.

dlibenzi committed 6 years ago

5e37ec43

Fix gcc 7.3 build

asuhan committed 6 years ago

c79f9dad

Add index_put, index_put_ to ATen XLA tensor

asuhan committed 6 years ago

f6fc28ad

Remove landed 17847 PR.

dlibenzi committed 6 years ago

55f74f0e

Fix changes in PyTorch argmax/argmin APIs.

dlibenzi committed 6 years ago

d684efe2

Make IR graph trim parameters configurable.

dlibenzi committed 6 years ago

9478f472

Older