[XLA:PJRT] Sets the number of retries for initializing the TPU backend to 0, since we always link in the TPU client to the Python extension even when it's not being run on TPU. PiperOrigin-RevId: 337190434 Change-Id: I81e2f93d1603c796f630aafa2cd3ae5106e8c6af
281 lines
9.8 KiB
Python
281 lines
9.8 KiB
Python
load("//tensorflow/core/platform:rules_cc.bzl", "cc_library")
|
|
load("//tensorflow:tensorflow.bzl", "tf_cc_test")
|
|
load("@local_config_cuda//cuda:build_defs.bzl", "if_cuda")
|
|
|
|
package(
|
|
default_visibility = ["//tensorflow:internal"],
|
|
licenses = ["notice"], # Apache 2.0
|
|
)
|
|
|
|
cc_library(
|
|
name = "worker_thread",
|
|
srcs = ["worker_thread.cc"],
|
|
hdrs = ["worker_thread.h"],
|
|
deps = [
|
|
"//tensorflow/core:lib",
|
|
"@com_google_absl//absl/synchronization",
|
|
],
|
|
)
|
|
|
|
cc_library(
|
|
name = "event_pool",
|
|
srcs = ["event_pool.cc"],
|
|
hdrs = ["event_pool.h"],
|
|
deps = [
|
|
"//tensorflow/compiler/xla:status_macros",
|
|
"//tensorflow/compiler/xla:statusor",
|
|
"//tensorflow/compiler/xla:types",
|
|
"//tensorflow/core:lib",
|
|
"//tensorflow/core/platform:stream_executor",
|
|
"@com_google_absl//absl/memory",
|
|
"@com_google_absl//absl/synchronization",
|
|
],
|
|
)
|
|
|
|
cc_library(
|
|
name = "semaphore",
|
|
srcs = ["semaphore.cc"],
|
|
hdrs = ["semaphore.h"],
|
|
deps = [
|
|
"//tensorflow/compiler/xla:types",
|
|
"//tensorflow/core:lib",
|
|
"@com_google_absl//absl/base:core_headers",
|
|
"@com_google_absl//absl/synchronization",
|
|
],
|
|
)
|
|
|
|
tf_cc_test(
|
|
name = "semaphore_test",
|
|
srcs = ["semaphore_test.cc"],
|
|
deps = [
|
|
":semaphore",
|
|
"//tensorflow/compiler/xla:test",
|
|
"//tensorflow/core:lib",
|
|
"//tensorflow/core:test_main",
|
|
"@com_google_absl//absl/synchronization",
|
|
],
|
|
)
|
|
|
|
cc_library(
|
|
name = "tracked_device_buffer",
|
|
srcs = ["tracked_device_buffer.cc"],
|
|
hdrs = ["tracked_device_buffer.h"],
|
|
visibility = [
|
|
"//learning/pathways/data_parallel:__pkg__",
|
|
"//tensorflow:internal",
|
|
],
|
|
deps = [
|
|
":event_pool",
|
|
":local_device_state",
|
|
"//tensorflow/compiler/xla:shape_util",
|
|
"//tensorflow/compiler/xla:types",
|
|
"//tensorflow/compiler/xla/service:shaped_buffer",
|
|
"//tensorflow/compiler/xla/service:transfer_manager",
|
|
"//tensorflow/core:lib",
|
|
"//tensorflow/stream_executor:device_memory",
|
|
"//tensorflow/stream_executor:device_memory_allocator",
|
|
"//tensorflow/stream_executor:event",
|
|
"@com_google_absl//absl/container:flat_hash_set",
|
|
"@com_google_absl//absl/synchronization",
|
|
],
|
|
)
|
|
|
|
tf_cc_test(
|
|
name = "tracked_device_buffer_test",
|
|
srcs = ["tracked_device_buffer_test.cc"],
|
|
deps = [
|
|
":tracked_device_buffer",
|
|
"//tensorflow/compiler/xla:literal_util",
|
|
"//tensorflow/compiler/xla:shape_util",
|
|
"//tensorflow/compiler/xla:status_macros",
|
|
"//tensorflow/compiler/xla:test",
|
|
"//tensorflow/compiler/xla/client:client_library",
|
|
"//tensorflow/compiler/xla/service:cpu_plugin",
|
|
"//tensorflow/core:test_main",
|
|
"//tensorflow/stream_executor:device_memory",
|
|
"//tensorflow/stream_executor:device_memory_allocator",
|
|
],
|
|
)
|
|
|
|
cc_library(
|
|
name = "local_device_state",
|
|
srcs = ["local_device_state.cc"],
|
|
hdrs = ["local_device_state.h"],
|
|
deps = [
|
|
":event_pool",
|
|
":semaphore",
|
|
":worker_thread",
|
|
"//tensorflow/compiler/xla:status",
|
|
"//tensorflow/compiler/xla:util",
|
|
"//tensorflow/compiler/xla/client:local_client",
|
|
"//tensorflow/core:lib",
|
|
"//tensorflow/core/platform:stream_executor",
|
|
"//tensorflow/stream_executor:event",
|
|
"@com_google_absl//absl/memory",
|
|
"@com_google_absl//absl/synchronization",
|
|
],
|
|
)
|
|
|
|
cc_library(
|
|
name = "pjrt_client",
|
|
srcs = ["pjrt_client.cc"],
|
|
hdrs = ["pjrt_client.h"],
|
|
visibility = ["//tensorflow/compiler/xla:friends"],
|
|
deps = [
|
|
":event_pool",
|
|
":local_device_state",
|
|
":tracked_device_buffer",
|
|
"//tensorflow/compiler/xla:cpu_function_runtime",
|
|
"//tensorflow/compiler/xla:executable_run_options",
|
|
"//tensorflow/compiler/xla:literal",
|
|
"//tensorflow/compiler/xla:literal_util",
|
|
"//tensorflow/compiler/xla:shape_util",
|
|
"//tensorflow/compiler/xla:status",
|
|
"//tensorflow/compiler/xla:statusor",
|
|
"//tensorflow/compiler/xla:util",
|
|
"//tensorflow/compiler/xla:xla_data_proto_cc",
|
|
"//tensorflow/compiler/xla/client:executable_build_options",
|
|
"//tensorflow/compiler/xla/client:local_client",
|
|
"//tensorflow/compiler/xla/client:xla_computation",
|
|
"//tensorflow/compiler/xla/pjrt/distributed:protocol_proto_cc",
|
|
"//tensorflow/compiler/xla/service:computation_placer",
|
|
"//tensorflow/compiler/xla/service:executable",
|
|
"//tensorflow/compiler/xla/service:hlo",
|
|
"//tensorflow/compiler/xla/service:hlo_cost_analysis",
|
|
"//tensorflow/compiler/xla/service:maybe_owning_device_memory",
|
|
"//tensorflow/compiler/xla/service:shaped_buffer",
|
|
"//tensorflow/compiler/xla/service/gpu:gpu_executable_run_options",
|
|
"//tensorflow/core:lib",
|
|
"//tensorflow/core/framework:allocator",
|
|
"//tensorflow/core/profiler/lib:connected_traceme",
|
|
"//tensorflow/core/profiler/lib:traceme",
|
|
"//tensorflow/core/profiler/lib:traceme_encode",
|
|
"//tensorflow/stream_executor:event",
|
|
"//tensorflow/stream_executor:stream",
|
|
"//tensorflow/stream_executor/host:host_platform_id",
|
|
"//tensorflow/stream_executor/lib",
|
|
"@com_google_absl//absl/base",
|
|
"@com_google_absl//absl/container:flat_hash_map",
|
|
"@com_google_absl//absl/container:flat_hash_set",
|
|
"@com_google_absl//absl/container:inlined_vector",
|
|
"@com_google_absl//absl/memory",
|
|
"@com_google_absl//absl/strings",
|
|
"@com_google_absl//absl/strings:str_format",
|
|
"@com_google_absl//absl/synchronization",
|
|
"@com_google_absl//absl/time",
|
|
"@com_google_absl//absl/types:optional",
|
|
"@com_google_absl//absl/types:span",
|
|
],
|
|
)
|
|
|
|
cc_library(
|
|
name = "tpu_client",
|
|
srcs = ["tpu_client.cc"],
|
|
hdrs = ["tpu_client.h"],
|
|
visibility = [
|
|
"//learning/brain/research/jax:__subpackages__",
|
|
"//learning/deepmind/tensorflow/tensorfn:__subpackages__",
|
|
"//learning/pathways:__subpackages__",
|
|
"//tensorflow/compiler/xla:friends",
|
|
],
|
|
deps = [
|
|
":local_device_state",
|
|
":pjrt_client",
|
|
":tracked_device_buffer",
|
|
"//tensorflow/compiler/xla:shape_util",
|
|
"//tensorflow/compiler/xla:status",
|
|
"//tensorflow/compiler/xla:statusor",
|
|
"//tensorflow/compiler/xla:util",
|
|
"//tensorflow/compiler/xla/client:client_library",
|
|
"//tensorflow/compiler/xla/service:computation_placer",
|
|
"//tensorflow/compiler/xla/service:shaped_buffer",
|
|
"//tensorflow/core:lib",
|
|
"//tensorflow/core/tpu:tpu_executor_dlsym_initializer",
|
|
"//tensorflow/core/tpu:tpu_on_demand_compiler",
|
|
"//tensorflow/stream_executor:device_memory",
|
|
"//tensorflow/stream_executor:stream",
|
|
"//tensorflow/stream_executor/lib",
|
|
"//tensorflow/stream_executor/tpu:tpu_computation_placer",
|
|
"//tensorflow/stream_executor/tpu:tpu_executable_interface",
|
|
"//tensorflow/stream_executor/tpu:tpu_executor",
|
|
"//tensorflow/stream_executor/tpu:tpu_executor_interface",
|
|
"//tensorflow/stream_executor/tpu:tpu_platform_interface",
|
|
"//tensorflow/stream_executor/tpu:tpu_topology_external",
|
|
"//tensorflow/stream_executor/tpu:tpu_transfer_manager",
|
|
"@com_google_absl//absl/container:inlined_vector",
|
|
"@com_google_absl//absl/flags:flag",
|
|
"@com_google_absl//absl/memory",
|
|
"@com_google_absl//absl/status",
|
|
],
|
|
)
|
|
|
|
cc_library(
|
|
name = "interpreter_device",
|
|
srcs = ["interpreter_device.cc"],
|
|
hdrs = ["interpreter_device.h"],
|
|
deps = [
|
|
":pjrt_client",
|
|
"//tensorflow/compiler/xla:statusor",
|
|
"//tensorflow/compiler/xla/client:client_library",
|
|
"//tensorflow/compiler/xla/service:interpreter_plugin",
|
|
"//tensorflow/compiler/xla/service:platform_util",
|
|
"@com_google_absl//absl/strings",
|
|
],
|
|
)
|
|
|
|
cc_library(
|
|
name = "cpu_device",
|
|
srcs = ["cpu_device.cc"],
|
|
hdrs = ["cpu_device.h"],
|
|
deps = [
|
|
":pjrt_client",
|
|
"//tensorflow/compiler/xla:statusor",
|
|
"//tensorflow/compiler/xla/client:client_library",
|
|
"//tensorflow/compiler/xla/service:platform_util",
|
|
"@com_google_absl//absl/strings",
|
|
],
|
|
)
|
|
|
|
cc_library(
|
|
name = "nvidia_gpu_device",
|
|
srcs = ["nvidia_gpu_device.cc"],
|
|
hdrs = ["nvidia_gpu_device.h"],
|
|
copts = if_cuda(["-DNCCL_ENABLED=1"]),
|
|
deps = [
|
|
":pjrt_client",
|
|
"//tensorflow/compiler/xla/service/gpu:gpu_executable_run_options",
|
|
"//tensorflow/compiler/xla:statusor",
|
|
"//tensorflow/compiler/xla/client:client_library",
|
|
"//tensorflow/compiler/xla/pjrt/distributed:client",
|
|
"//tensorflow/compiler/xla/service:platform_util",
|
|
"//tensorflow/compiler/xla:util",
|
|
"//tensorflow/core/common_runtime:bfc_allocator",
|
|
"//tensorflow/core/common_runtime/gpu:gpu_mem_allocator",
|
|
"//tensorflow/core:lib_internal",
|
|
"//tensorflow/stream_executor:tf_allocator_adapter",
|
|
] + if_cuda(["@local_config_nccl//:nccl"]),
|
|
)
|
|
|
|
tf_cc_test(
|
|
name = "gpu_multistream_test",
|
|
srcs = ["gpu_multistream_test.cc"],
|
|
tags = [
|
|
# TODO(phawkins): figure out TF test infra such that this only runs under GPU.
|
|
"no_oss",
|
|
"requires-gpu-nvidia",
|
|
"notap",
|
|
],
|
|
deps = [
|
|
":nvidia_gpu_device",
|
|
":pjrt_client",
|
|
"//tensorflow/compiler/xla:test",
|
|
"//tensorflow/compiler/xla/client:executable_build_options",
|
|
"//tensorflow/compiler/xla/client:xla_builder",
|
|
"//tensorflow/compiler/xla/service:gpu_plugin",
|
|
"//tensorflow/compiler/xla/tests:literal_test_util",
|
|
"//tensorflow/core:lib",
|
|
"//tensorflow/core:test_main",
|
|
"//tensorflow/core/platform:random",
|
|
],
|
|
)
|