| /external/tensorflow/tensorflow/core/common_runtime/ |
| D | permuter.cc | 40 : col_ctx_(nullptr), col_params_(nullptr), done_(nullptr), counter_(0) {} in Permuter()
56 col_ctx_ = col_ctx; in InitializeCollectiveContext()
71 col_ctx_->input, CheckCounterAndCallDone()); in Run()
74 DispatchRecv(i, col_params_->instance.permutation[i], col_ctx_->output, in Run()
83 strings::StrCat(col_ctx_->exec_key, src_rank, target_rank); in DispatchSend()
85 << col_ctx_->device_name << " to_device " in DispatchSend()
88 col_ctx_->col_exec->remote_access()->PostToPeer( in DispatchSend()
91 col_ctx_->device, col_ctx_->op_ctx->op_device_context(), in DispatchSend()
92 col_ctx_->op_ctx->output_alloc_attr(0), tensor, col_ctx_->device_locality, in DispatchSend()
93 col_ctx_->op_ctx->cancellation_manager(), done); in DispatchSend()
[all …]
|
| D | ring_reducer.cc | 56 CHECK(col_ctx_); in Run()
60 col_ctx_->col_exec->UnblockDependencies(*col_params_); in Run()
83 VLOG(1) << "RingReducer::Run for device " << col_ctx_->device_name in Run()
90 if ((col_ctx_->input != col_ctx_->output) && in Run()
91 (DMAHelper::base(col_ctx_->input) != DMAHelper::base(col_ctx_->output))) { in Run()
98 col_ctx_->op_ctx->op_device_context(), in Run()
99 col_ctx_->op_ctx->op_device_context(), col_ctx_->device, in Run()
100 col_ctx_->device, col_ctx_->op_ctx->input_alloc_attr(0), in Run()
101 col_ctx_->op_ctx->output_alloc_attr(0), col_ctx_->input, in Run()
102 col_ctx_->output, 0 /*dev_to_dev_stream_index*/, in Run()
[all …]
|
| D | hierarchical_tree_broadcaster.cc | 57 : col_ctx_(nullptr), in HierarchicalTreeBroadcaster()
190 col_ctx_ = col_ctx; in InitializeCollectiveContext()
198 CHECK(col_ctx_); in Run()
307 VLOG(1) << "Running Broadcast tree device=" << col_ctx_->device_name in RunTree()
323 DispatchRecv(si, recv_from_rank, my_rank, col_ctx_->output, in RunTree()
347 (is_source_ ? col_ctx_->input : col_ctx_->output), in RunTree()
368 << col_ctx_->device_name << " subdiv=" << si; in RunTree()
369 if (col_ctx_->input != col_ctx_->output && in RunTree()
370 (DMAHelper::base(col_ctx_->input) != in RunTree()
371 DMAHelper::base(col_ctx_->output))) { in RunTree()
[all …]
|
| D | ring_gatherer.cc | 64 DCHECK(col_ctx_); in Run()
87 VLOG(1) << "RingGatherer::Run for device " << col_ctx_->device_name in Run()
93 AllocatorAttributes attr = col_ctx_->op_ctx->output_alloc_attr(0); in Run()
94 ca_.reset(MakeCollectiveAdapter(col_ctx_->output, group_size_ * num_subdivs_, in Run()
95 col_ctx_->device->GetAllocator(attr), in Run()
107 col_ctx_->op_ctx->op_device_context(), in Run()
108 col_ctx_->op_ctx->op_device_context(), col_ctx_->device, in Run()
109 col_ctx_->device, col_ctx_->op_ctx->input_alloc_attr(0), in Run()
110 col_ctx_->op_ctx->output_alloc_attr(0), col_ctx_->input, &alias_chunk, in Run()
141 col_ctx_->device->tensorflow_gpu_device_info(); in RunAsyncParts()
[all …]
|
| D | ring_alg.cc | 102 col_ctx_(nullptr), in RingAlg()
247 col_ctx_ = col_ctx; in InitializeCollectiveContext()
256 col_ctx_->op_ctx->device()->tensorflow_gpu_device_info(); in TensorDebugString()
260 &tensor, "" /*tensor_name*/, col_ctx_->device, &cpu_tensor); in TensorDebugString()
287 if (col_ctx_->op_ctx->cancellation_manager() == nullptr || in StartAbort()
288 (!col_ctx_->op_ctx->cancellation_manager()->IsCancelled() && in StartAbort()
289 !col_ctx_->op_ctx->cancellation_manager()->IsCancelling())) { in StartAbort()
290 col_ctx_->col_exec->StartAbort(s); in StartAbort()
298 ca_->ConsumeFinalValue(col_ctx_->output); in Finish()
384 string send_buf_key = RingAlgBufKey(name_, col_ctx_->exec_key, in DispatchSend()
[all …]
|
| D | permuter.h | 67 std::shared_ptr<CollectiveContext> col_ctx_;
|
| D | hierarchical_tree_broadcaster.h | 84 std::shared_ptr<CollectiveContext> col_ctx_; variable
|
| D | ring_alg.h | 112 std::shared_ptr<CollectiveContext> col_ctx_; variable
|
| /external/tensorflow/tensorflow/core/kernels/ |
| D | collective_nccl_reducer.cc | 37 switch (col_ctx_->output->dtype()) { in Run()
60 DataTypeString(col_ctx_->output->dtype()))); in Run()
64 col_ctx_->device->GetAllocator(col_ctx_->op_ctx->input_alloc_attr(0)), in Run()
65 col_ctx_->output->dtype(), TensorShape({})); in Run()
66 DeviceContext* op_dev_ctx = col_ctx_->op_ctx->op_device_context(); in Run()
70 &group_size_val, col_ctx_->device, &group_size, in Run()
94 col_ctx_->nccl_communicator->Enqueue(col_ctx_, std::move(done_callback)); in Run()
118 col_ctx_->op_ctx, col_ctx_->op_params, col_ctx_->device, in Run()
119 col_params_->final_op, col_ctx_->output, &group_size); in Run()
|
| D | collective_nccl_broadcaster.cc | 27 col_ctx_->nccl_communicator->Enqueue(col_ctx_, std::move(done)); in Run()
|
| D | collective_nccl_gatherer.cc | 27 col_ctx_->nccl_communicator->Enqueue(col_ctx_, std::move(done)); in Run()
|
| D | collective_nccl.cc | 27 : type_(type), name_(name), col_ctx_(nullptr), col_params_(nullptr) {} in NcclBase()
63 col_ctx_ = col_ctx; in InitializeCollectiveContext()
|
| D | collective_nccl.h | 42 std::shared_ptr<CollectiveContext> col_ctx_; variable
|