[XLA:Python] Remove xla_client.Buffer class.
The only remaining method was Buffer.from_pyval. Callers should use
LocalClient.buffer_from_pyval instead.
PiperOrigin-RevId: 310248281
Change-Id: I3cca4e5ea85b7632ac5ef2f40fec488e50fe0fc8
diff --git a/tensorflow/compiler/xla/python/xla_client.py b/tensorflow/compiler/xla/python/xla_client.py
index b06cba1..d9cd906 100644
--- a/tensorflow/compiler/xla/python/xla_client.py
+++ b/tensorflow/compiler/xla/python/xla_client.py
@@ -261,44 +261,6 @@
"""
-class Buffer(object):
- """Represents a handle to data owned by XLA.
-
- The referent is ready for use in executing a local, compiled
- Computation. On XLA platforms involving a device (e.g. GPU), this
- means the referent is in device memory.
- """
-
- @staticmethod
- def from_pyval(pyval, device=None, backend=None, force_copy=False):
- """Copies the `pyval` to a freshly allocated on-device buffer."""
- backend = backend or get_local_backend()
- return backend.buffer_from_pyval(pyval, device, force_copy=force_copy)
-
- # Buffer is not an instantiable type and exists only for its static methods.
- # The underlying buffer objects are C++ object with the following
- # API:
- # def shape(self) -> Shape:
- # def device(self) -> int:
- # def delete(self):
- # def is_deleted(self) -> bool:
- # def block_host_until_ready(self):
- # """Blocks the calling thread until the buffer is ready on device."""
- # def copy_to_host_async(self):
- # """Requests a copy of the buffer to the host.
- #
- # Does not block waiting for the copy. Values fetched are available via
- # `to_py()`; the purpose of `copy_to_host_async` is to prefetch values
- # for subsequent `to_py()` calls, especially when requesting many values
- # at once.
- # """
- # def to_py(self):
- # """Returns the value of the buffer as a Python tuple tree of ndarrays."""
- #
- # TODO(phawkins): remove Buffer and its static methods completely, have
- # clients call methods on Backend to create buffers.
-
-
def shape_from_pyval(pyval):
"""Returns a Shape that describes a tuple-tree of Numpy arrays."""
