pytorch/xla

Pull Requests Commits

Cleanup GetXlaData by using CurrentXlaData.

dlibenzi committed 6 years ago

9dcb14cc

Get back lost performance and add introspection methods to a few IR nodes.

dlibenzi committed 6 years ago

2ff9b863

[XLATensor] Add max_pool2d

asuhan committed 6 years ago

3e60d74b

Clean up some code in Conv2d IR node

asuhan committed 6 years ago

1b6a0f96

Fix indentation in test/test_operations.py

asuhan committed 6 years ago

213dc7fe

Account for ATEN tensor data to be the most recent data available.

dlibenzi committed 6 years ago

f6c54120

Do not force device data conversion if we already have most recent tensor data.

dlibenzi committed 6 years ago

1bf21d14

The c10::Tensor is already a fast copy refcounted thing, so not need to wrap it in a shapred_ptr<>.

dlibenzi committed 6 years ago

735eb4f2

Update the PyTorch commit id to latest

asuhan committed 6 years ago

7b2df468

[XLATensor] Add conv2d

asuhan committed 6 years ago

cd7cbcad

[XLATensor] Add shape inference helper and make relu use it

asuhan committed 6 years ago

be4409d2

Start relaxing the fact that a tensor is either device data or IR node.

dlibenzi committed 6 years ago

6a076fee

Move the client-get into the computation client.

dlibenzi committed 6 years ago

0b0cae56

Added c10 linkage.

dlibenzi committed 6 years ago

f547929b

Fix bolted in path.

dlibenzi committed 6 years ago

243b4c74

Do not assume library folder.

dlibenzi committed 6 years ago

e14b9b91

Few more tweaks and added run shell script.

dlibenzi committed 6 years ago

989f99ee

Seems to be working now.

dlibenzi committed 6 years ago

4e7bfb3b

Added simple test.

dlibenzi committed 6 years ago

96997a62

Trying to get c++ unittests working.

dlibenzi committed 6 years ago

acd33eac

Remove unused variable in BuildConvolutionBias

asuhan committed 6 years ago

83fdf972

Rename toTensor API.

dlibenzi committed 6 years ago

8e5efefa

The ops/ folder does not need to be listed.

dlibenzi committed 6 years ago

714379a2

[XLATensor] Add relu

asuhan committed 6 years ago

4484d001

Moved arithmetic IR ops.

dlibenzi committed 6 years ago

11efb7a5

Enable XLA tensor data to persist.

dlibenzi committed 6 years ago

cc20fdf0

dlibenzi committed 6 years ago

74703986

Allow easier creation of arithmetic ops.

dlibenzi committed 6 years ago

1baee19d

Added utility APIs to create IR Node instead of replicating make_shared.

dlibenzi committed 6 years ago

43f0c84c

Fix review comments.

dlibenzi committed 6 years ago

04032a51

Newer Older