Summary:
Pull Request resolved: pytorch/pytorch#11205

Our short term plan for supporting out of tree complex development requires an
external library to add a custom subclass of Type without access to the
code generation facilities in ATen.  This commit reorganizes Type so
as to minimize the amount of boilerplate you have to write when making
a subclass of Type.

In particular, it:
- Creates a new CPUTypeDefault/CUDATypeDefault class, which you are
  intended to inherit from, which provides default implementations
  of CPU/CUDA that is layout/dtype agnostic.
- Adds new getCPUAllocator() and getCUDAAllocator() functions, as
  a more public API to get your hands on Allocator
- Adds allocator() and getDeviceFromPtr(), abstracting the device
  specific parts of storage() methods; these methods are now
  implemented in base TypeDefault.
- Delete the static typeString() method, which is now dead.
- Move is_cuda/is_sparse/is_distributed to TypeDefault.

Reviewed By: SsnL

Differential Revision: D9631619

fbshipit-source-id: 40b600d99691230e36e03eb56434c351cbc2aa3a

resolve conflict in data parallel model
* master: (201 commits)
  Add cost inference to ConvGradient and WeightedSum operators (pytorch#10744)
  Move collapse dims into a single place (pytorch#11272)
  Fix some more warnings (pytorch#11257)
  Fix the batchnorm onnx exporting when affine=False
  Improve error message to include return types too (pytorch#11245)
  Check doxygen output in travis (pytorch#11124)
  Accept more numpy scalars as doubles (pytorch#9659)
  Fixed log message (pytorch#10874)
  Fix to distribution.__repr__ with lazy attributes (pytorch#11263)
  Add import export step to end to end tests
  Add complex hooks for out of tree complex implementation. (pytorch#11216)
  Unify opt flag for cmake codegen (pytorch#11227)
  nomnigraph - fix memory error in NN subgraph matchOp (pytorch#11127)
  Port PackedSequences functions to C++ (pytorch#11224)
  Treat numerical differences as warnings instead of errors when tracing (pytorch#11246)
  add a Float16UniformFill (pytorch#11123)
  Implement torch.tensordot (pytorch#10025)
  keep net type info when generating model complete net (pytorch#11032)
  Get rid of some uses of type() (pytorch#11215)
  Reorganize methods in Type, add CPUTypeDefault/CUDATypeDefault (pytorch#11205)
  ...

…ch#11205)

Summary:
Pull Request resolved: pytorch#11205

Our short term plan for supporting out of tree complex development requires an
external library to add a custom subclass of Type without access to the
code generation facilities in ATen.  This commit reorganizes Type so
as to minimize the amount of boilerplate you have to write when making
a subclass of Type.

In particular, it:
- Creates a new CPUTypeDefault/CUDATypeDefault class, which you are
  intended to inherit from, which provides default implementations
  of CPU/CUDA that is layout/dtype agnostic.
- Adds new getCPUAllocator() and getCUDAAllocator() functions, as
  a more public API to get your hands on Allocator
- Adds allocator() and getDeviceFromPtr(), abstracting the device
  specific parts of storage() methods; these methods are now
  implemented in base TypeDefault.
- Delete the static typeString() method, which is now dead.
- Move is_cuda/is_sparse/is_distributed to TypeDefault.

Reviewed By: SsnL

Differential Revision: D9631619

fbshipit-source-id: 40b600d99691230e36e03eb56434c351cbc2aa3a