Summary:
This does 6 things:
- add c10/util/Registry.h as the unified registry util
- cleaned up some APIs such as export condition
- fully remove aten/core/registry.h
- fully remove caffe2/core/registry.h
- remove a bogus aten/registry.h
- unifying all macros
- set up registry testing in c10
Also, an important note that we used to mark the templated Registry class as EXPORT - this should not happen, because one should almost never export a template class. This PR fixes that.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/12077
Reviewed By: ezyang
Differential Revision: D10050771
Pulled By: Yangqing
fbshipit-source-id: 417b249b49fed6a67956e7c6b6d22374bcee24cf
Summary:
Some more `ATEN_API` additions for hidden visibility.
Running CI tests to see what fails to link.
cc Yangqing mingzhe09088 ezyang
Pull Request resolved: https://github.com/pytorch/pytorch/pull/10624
Reviewed By: mingzhe09088
Differential Revision: D9392728
Pulled By: orionr
fbshipit-source-id: e0f0861496b12c9a4e40c10b6e0c9e0df18e8726
Summary:
Implemented ApplyTransformIfFaster
Determine if a transform is faster, then return whichever net is better.
Reviewed By: bwasti
Differential Revision: D5534535
fbshipit-source-id: 509943205b0c454bf30fb01343ac4e88d1441c39
Summary:
Pattern match currently only supports one type of pattern matching: connected components.
It will be useful to sometimes use different algorithms to pattern match, either a subset of the operators in order, or general non-connected subgraphs. While generalized pattern matching can match for all types, it is inefficient to use it when sorted order or connected component suffice.
You can can set the PatternMatchType to be one of the three options (it is connected by default), and Transform will use the associated algorithm.
We will need this for common subexpression elimination - specifically, sorted order matching.
Reviewed By: bwasti
Differential Revision: D5629321
fbshipit-source-id: 2104f2d4384fe4aba06a386881a08ca324f290a6
Summary:
In order to pybind, we need transform in core.
It's a basically finished product, with a big test suite. It's safe.
We can begin hooking up observers after this, and I have a diff coming up that pybinds some apply_transform function.
Reviewed By: bwasti
Differential Revision: D5522200
fbshipit-source-id: dea6aa606fc689af84e2533569d1ef348cb5f3f2
