aboutsummaryrefslogtreecommitdiffhomepage
path: root/tensorflow/core/framework/memory_types.cc
Commit message (Collapse)AuthorAge
* Make functions defined with tfe.defun respect devices when executing.Gravatar Akshay Agrawal2018-07-03
| | | | | | | | | | | | | | | | | | | Modifies GraphModeFunction to emit PartitionedCall ops instead of Call ops so that the created functions can execute across devices. This should strictly increase the set of functions that tfe.defun can faithfully execute. Previous to this change, functions executed through tfe.defun would ignore device annotations and only run on a single device. It is not yet possible to execute a function across multiple processes. Specifically, this CL: (1) Adds a stateful version of PartitionedCall, (2) Modifies `defun` to emit PartitionedCall or StatefulPartitionedCall by default, (3) Makes `tf.gradients` aware of the existence of `(Stateful)PartitionedCall`, (4) Fixes bugs in PartitionedCallOp related to the placement of resource-touching ops / which args and retvals are always on host memory, and also removes the requirement for args/retvals to be passed through the host. PiperOrigin-RevId: 203164388
* [TF] Mark DT_STRING and DT_RESOURCE types as always sitting on host memory.Gravatar Eugene Brevdo2017-12-12
| | | | | | | | | | This is important when these arguments may appear in op input lists or output lists, where the signature may not be able to declare them as sitting on host. For DT_RESOURCE types, just the handles are marked as sitting on host memory; the actual data may reside on GPU. PiperOrigin-RevId: 178837213
* Prepare to not include node_def.proto.h in node_def_util.hGravatar Geoffrey Irving2017-06-23
| | | | | | | | | | The goal is to make kernels mostly independent of proto headers, which will let us lock down our .so imports. This CL makes a bunch of .cc files either include node_def.proto.h themselves or not need the definition of NodeDef; a second CL will make node_def_util.h not include node_def.proto.h. RELNOTES: n/a PiperOrigin-RevId: 159982117
* Remove unnecessary copies of value parameters.Gravatar Peter Hawkins2017-05-10
| | | | PiperOrigin-RevId: 155511618
* Internal cleanup.Gravatar A. Unique TensorFlower2016-11-03
| | | | Change: 138135876
* Remove special handling of host-memory/device-memory for int32 arguments in ↵Gravatar A. Unique TensorFlower2016-08-12
| | | | | | | type lists. The current special case behavior is preserved only for functions and their gradients. Change: 130100547
* Improve memory type deduction for operators with type list arguments.Gravatar A. Unique TensorFlower2016-07-28
| | | | | | | | | Currently, if an operator has any type list arguments, HostMemory type annotations are ignored and the memory types for all arguments (type lists or not) are deduced from the data types. This CL makes two changes: * If an operator has a type list argument, only deduce memory types from data types for the type lists themselves. Use the normal rules for other arguments. * In addition, make it possible to force type list arguments to be in HostMemory, overriding the default memory types. Change: 128717962
* In C++ shape inference, support registering a shape inference function in OPGravatar A. Unique TensorFlower2016-06-14
| | | | | | | | | registration. Change op registration signature to return Status and return the registration data as an out parameter. Add a shape inference function and test for AddN. Support in function library needs to be expanded in a future change. Change: 124871850
* Update copyright for 3p/tf/core.Gravatar A. Unique TensorFlower2016-06-02
| | | | Change: 123900938
* Change selective registration of op kernels to work off of the class nameGravatar A. Unique TensorFlower2016-03-28
| | | | | | | | instead of the filename. Change FindKernelDef to also return the class name, to help a tool that generates ops_to_register.h to find the set of class names. Change: 118381472
* Hardening the error checking related to memory types.Gravatar A. Unique TensorFlower2016-02-11
| | | | Change: 114484861
* Moves MemoryType inference code out of OpKernel so that it can reused.Gravatar A. Unique TensorFlower2016-02-11
Change: 114448861