This is to support snappy compression/decompression on new version of snapshot dataset, as the dataset uses input streams instead.
PiperOrigin-RevId: 317435680
Change-Id: Ie57b43e73b6b7911398d883c3c5a0de72973288e
This change leaves all existing targets in "tensorflow/core/BUILD" in place, with some becoming aliases. In future, we will remove aliases and point to the new locations.
PiperOrigin-RevId: 311194740
Change-Id: Id413277651b260641c1c2e06cb54d16629e6e662
This forwarding header was moved to "tensorflow/core/common_runtime/graph_constructor.h" and all known uses have been rewritten to use the version in "common_runtime".
PiperOrigin-RevId: 309767804
Change-Id: I92cc94f95ff02615a142bf19532f8677b3be36bc
This changes moves the remaining modules in "core/graph/..." that depend on modules in "core/common_runtime/..." to "core/common_runtime/", and creates separate fine-grained targets for each of them:
* "gradients.{h.cc}"
* "graph_def_builder_util.{h,cc}"
* "mkl_{layout,tfconversion}_pass.{h,cc}"
* "quantize_training.{h,cc}"
PiperOrigin-RevId: 308841156
Change-Id: I7774e2498f465290e7addd331625bd91d90acbac
None of the tests in core/lib/ depend on libraries in core/common_runtime/.
PiperOrigin-RevId: 306710822
Change-Id: I55f18cee26fdd17cb35916bc47324d7733aebc80
Also create subsidiary BUILD files for "common_runtime/{gpu,sycl}".
This change leaves all existing targets in "tensorflow/core/BUILD" in place, with some becoming aliases. In future, we will split up the targets in "tensorflow/core/common_runtime/BUILD" to be more granular.
PiperOrigin-RevId: 304476929
Change-Id: Ida5348ce41b4d98337dd8a128bfcc44e20be9c3d
Abstract interface objects such as AbstractContextInterface,
AbstractOperationInterface & AbstractTensorHandleInterface were added to
decouple client code from the underlying implementation and allow for
easy switching to TFRT. Unfortunately the interface implementations had
to live in their own objects due to circular dependencies. This change
hopes to address most of them with the following changes & clean-ups:
* Move abstract interface definitions into separate build targets
* Make EagerContext implement AbstractContextInterface.
* Make EagerOperation implement AbstractOperationInterface.
* Make TensorHandle implement AbstractTensorHandleInterface.
* Use tensorflow::DataType instead of TF_DataType in interfaces.
* Replace unique_ptr with raw pointer and expose Release() function in
abstract interfaces while making the destructor protected. This allows
interface implementations to control their life-cycle, enabling
EagerContext and TensorHandle which are reference counted to implement
the interfaces directly.
* Make TensorHandle & EagerContext destructors private since the objects
are reference counted.
* Move out interface methods from c_api.cc
* Use absl::Span instead of absl::FixedArray in operation interface.
* Make CustomDeviceAPI use TFE_DeleteTensorHandle instead of directly
deleting the object.
* Remove unused TensorHandle::OnHostCPU method.
* Move remaining circular dependencies into core.cc.
PiperOrigin-RevId: 304269288
Change-Id: I2cced4a21ece708f413182b43f47e4891654901e
This change is part of an ongoing refactoring to simplify "executor.cc" and enable the substitution of more efficient implementations of `PropagateOutputs()`.
PiperOrigin-RevId: 304262448
Change-Id: I46a2d7fcdde89a71c502d272f35adfd34b0c4cab
In an ongoing effort to make "executor.cc" more readable, this change creates a new `ImmutableExecutorState` type containing most `ExecutorImpl` in "third_party/tensorflow/core/common_runtime/immutable_executor_state.h". It removes friend access from `ExecutorState` to the fields of `ExecutorImpl` and replaces it with public accessor methods for to the relevant fields.
PiperOrigin-RevId: 303886451
Change-Id: Ibea4e0d0b4fe88d10ca38705fbb2e6edc83f07ca
In an ongoing effort to make "executor.cc" more readable, this change moves the following types to "third_party/tensorflow/core/common_runtime/graph_view.h":
* `ControlEdgeInfo`
* `EdgeInfo`
* `GraphView`
* `NodeItem`
PiperOrigin-RevId: 303438916
Change-Id: Ib49a18a70b5caa9338a18d179a6ceaad0dafe262
This change also removes methods on `TensorReference` and `NodeExecStatsInterface` that are no longer used by any code; and fix some #includes and BUILD files that depended on indirect inclusion of "tensor_reference.h".
PiperOrigin-RevId: 300648897
Change-Id: I1f734b1de096f42158f7cd70ab9e27ad832db3f6
This commit fixes regressions introduced in the ROCm CSB by th following commit :
df00d7ebbf
The above commit classifies puts "cudnn_rnn_kernels" under "if_cuda" in the tensorflow/core/BUILD file. That target is mis-named (should really be gpu_rnn_kernels) and is required by ROCm too. This commit simply changes the condition from "if_cuda" to "if_cuda_or_rocm"
* Shape functions under shape_inference:: namespace stay in common_shape_fns.cc.
* Other functions are moved to kernel_shape_util.h.
Also, I removed common_shape_fns.h include in ops_util.h since it is not used in ops_util and only by files that include ops_util.h.
PiperOrigin-RevId: 298692539
Change-Id: I79df418a377d4eaed67aeec6f3736acb9d5b111d
And add yet another target to core/BUILD until we fully roll out
cc_shared_library :(
PiperOrigin-RevId: 297002883
Change-Id: Ieae6e732acfe6fffbbf0f7ff20189f1655661e9f
Fixes#33758
Downstream projects depending on TensorFlow: If bazel complains, please substitute `@zlib_archive` with `@zlib`, and `@grpc` with `@com_github_grpc_grpc` in WORKPLACE.
PiperOrigin-RevId: 295824868
Change-Id: If2259d59e9d82543369e5670916b1398374c9889
