| .. |
|
BUILD
|
Open source the XRTTpuDeviceAccessor and register XRTStateOps on TPU on the open source XRT side if it is a 1vm use case. 2vm code remain unchanged.
|
2021-01-13 14:00:49 -08:00 |
|
c_api_conversions.cc
|
[libtpu] Plumb HloModule::debug_options_ through the C API.
|
2021-01-11 12:15:28 -08:00 |
|
c_api_conversions.h
|
[libtpu] Reimplement XLA_Shape in C API for performance.
|
2020-10-19 14:17:22 -07:00 |
|
c_api_decl.h
|
Add TPU runtime version.
|
2021-01-14 13:29:47 -08:00 |
|
c_api_defn.h
|
Create dummy topology types for libtpu C API.
|
2020-08-10 17:13:50 -07:00 |
|
noncopyable_buffer.h
|
[TPU] Make NoncopyableBuffer::Clone take an optional alignment requirement.
|
2020-12-05 17:21:39 -08:00 |
|
proto_helper.cc
|
|
|
|
proto_helper.h
|
Implement Serialization/Deserialization for remote compilation cache.
|
2020-09-01 12:51:30 -07:00 |
|
status_helper.h
|
Change all SE_Status to the common TF_Status for better compatibility
|
2020-10-28 22:08:59 -07:00 |
|
tpu_computation_placer.cc
|
Put a bunch of TPU classes in the tensorflow::tpu namespace.
|
2020-09-21 13:59:05 -07:00 |
|
tpu_computation_placer.h
|
Add TpuComputationPlacer::AssignLocalDevices() to TPU API.
|
2020-09-03 11:49:36 -07:00 |
|
tpu_event.h
|
Put a bunch of TPU classes in the tensorflow::tpu namespace.
|
2020-09-21 13:59:05 -07:00 |
|
tpu_executable_interface.cc
|
[TPU] Don't pass host_shapes across the TPU API boundary. They can be computed from device shapes.
|
2020-10-07 18:48:10 -07:00 |
|
tpu_executable_interface.h
|
[XLA] Add a xla::Executable constructor without HLO profile related arguments, to prepare for removing those later.
|
2020-11-02 13:25:23 -08:00 |
|
tpu_executable.cc
|
Distinguish between not having a cross-program prefetch address and having a nullptr cross-program prefetch address.
|
2020-12-02 16:19:42 -08:00 |
|
tpu_executable.h
|
[TPU 1VM] Consolidate all TPU ops related APIs into a single file
|
2020-10-23 18:09:32 -07:00 |
|
tpu_executor_c_api.h
|
Add TPU runtime version.
|
2021-01-14 13:29:47 -08:00 |
|
tpu_executor_interface.h
|
Add TpuExecutorInterface::GetCoreLocationExternal() to TPU API.
|
2020-08-17 13:15:15 -07:00 |
|
tpu_executor.cc
|
Change all SE_Status to the common TF_Status for better compatibility
|
2020-10-28 22:08:59 -07:00 |
|
tpu_executor.h
|
Put a bunch of TPU classes in the tensorflow::tpu namespace.
|
2020-09-21 13:59:05 -07:00 |
|
tpu_node_context.cc
|
[TPU 1VM] Consolidate all TPU ops related APIs into a single file
|
2020-10-23 18:09:32 -07:00 |
|
tpu_node_context.h
|
[TPU 1VM] Consolidate all TPU ops related APIs into a single file
|
2020-10-23 18:09:32 -07:00 |
|
tpu_platform_interface.cc
|
Refactor TpuPlatformInterface::GetRegisteredPlatform() to make tries_left configurable.
|
2020-09-21 19:53:06 -07:00 |
|
tpu_platform_interface.h
|
Add TPU runtime version.
|
2021-01-14 13:29:47 -08:00 |
|
tpu_platform_registration.cc
|
Put a bunch of TPU classes in the tensorflow::tpu namespace.
|
2020-09-21 13:59:05 -07:00 |
|
tpu_platform.cc
|
Add TPU runtime version.
|
2021-01-14 13:29:47 -08:00 |
|
tpu_platform.h
|
Add TPU runtime version.
|
2021-01-14 13:29:47 -08:00 |
|
tpu_stream_interface.h
|
Clean up build files.
|
2020-06-25 17:32:38 -07:00 |
|
tpu_stream.h
|
Add TpuStream::EnqueueTransferHostToDevice and TpuStream::EnqueueTransferDeviceToHost
|
2020-09-08 16:22:00 -07:00 |
|
tpu_timer.h
|
Rename tpu_load_library to tpu_api_dlsym_initializer.
|
2020-06-22 17:45:58 -07:00 |
|
tpu_topology.cc
|
Add HostCount() / ChipsPerHost() API to TPUTopologyExternal.
|
2020-10-07 15:39:48 -07:00 |
|
tpu_topology.h
|
Add HostCount() / ChipsPerHost() API to TPUTopologyExternal.
|
2020-10-07 15:39:48 -07:00 |
|
tpu_transfer_manager_interface.cc
|
Introduce additional TPU infeed and outfeed ops
|
2020-08-07 17:57:57 -07:00 |
|
tpu_transfer_manager_interface.h
|
Introduce additional TPU infeed and outfeed ops
|
2020-08-07 17:57:57 -07:00 |
|
tpu_transfer_manager_registration.cc
|
Put a bunch of TPU classes in the tensorflow::tpu namespace.
|
2020-09-21 13:59:05 -07:00 |
|
tpu_transfer_manager.cc
|
Make tensorflow::tpu::NoncopyableBuffer allocate uint8 pointer rather than uint32 to allow unaligned U8/S8/U16/S16/BF16 arrays.
|
2020-11-06 10:22:44 -08:00 |
|
tpu_transfer_manager.h
|
Put a bunch of TPU classes in the tensorflow::tpu namespace.
|
2020-09-21 13:59:05 -07:00 |
