/external/tensorflow/tensorflow/core/common_runtime/ |
D | hierarchical_tree_broadcaster.cc | 77 CollectiveParams* col_params) { in InitializeCollectiveParams() argument 78 CHECK_EQ(col_params->instance.type, BROADCAST_COLLECTIVE); in InitializeCollectiveParams() 79 CHECK_EQ(col_params->instance.impl_details.collective_name, in InitializeCollectiveParams() 82 col_params->instance.device_names[col_params->default_rank]; in InitializeCollectiveParams() 87 << str_util::Join(col_params->instance.task_names, ", "); in InitializeCollectiveParams() 89 const string* prior_task_name = &col_params->instance.task_names[0]; in InitializeCollectiveParams() 91 for (int di = 1; di < col_params->group.group_size; ++di) { in InitializeCollectiveParams() 92 if (col_params->instance.task_names[di] != *prior_task_name) { in InitializeCollectiveParams() 95 prior_task_name = &col_params->instance.task_names[di]; in InitializeCollectiveParams() 101 CHECK_EQ(col_params->group.num_tasks, dev_per_task.size()); in InitializeCollectiveParams() [all …]
|
D | ring_alg.cc | 108 Status GenerateSubdivsInCollectiveParams(CollectiveParams* col_params) { in GenerateSubdivsInCollectiveParams() argument 109 if (col_params->instance.shape.num_elements() == 0) { in GenerateSubdivsInCollectiveParams() 113 col_params->group.group_size / col_params->group.num_tasks; in GenerateSubdivsInCollectiveParams() 118 col_params->instance.impl_details.collective_name); in GenerateSubdivsInCollectiveParams() 125 const size_t tensor_size = col_params->instance.shape.num_elements() * in GenerateSubdivsInCollectiveParams() 126 DataTypeSize(col_params->instance.data_type); in GenerateSubdivsInCollectiveParams() 130 int num_chunks = col_params->group.group_size * num_subdivs; in GenerateSubdivsInCollectiveParams() 137 col_params->instance.impl_details.collective_name); in GenerateSubdivsInCollectiveParams() 142 col_params->instance.impl_details.subdiv_offsets.reserve(num_subdivs); in GenerateSubdivsInCollectiveParams() 146 col_params->instance.impl_details.subdiv_offsets.push_back(subdiv_offset); in GenerateSubdivsInCollectiveParams() [all …]
|
D | base_collective_executor.cc | 219 const CollectiveParams& col_params, in ExecuteAsync() argument 240 const Tensor* input = (col_params.instance.type == REDUCTION_COLLECTIVE || in ExecuteAsync() 241 col_params.instance.type == GATHER_COLLECTIVE || in ExecuteAsync() 242 (col_params.instance.type == BROADCAST_COLLECTIVE && in ExecuteAsync() 243 col_params.is_source)) in ExecuteAsync() 247 Status status = CreateCollective(col_params, &col_impl); in ExecuteAsync() 254 new CollectiveContext(this, dev_mgr_, ctx, CtxParams(ctx), col_params, in ExecuteAsync() 284 const CollectiveParams& col_params, in CreateCollective() argument 288 switch (col_params.instance.data_type) { in CreateCollective() 290 if (col_params.group.device_type == DEVICE_GPU) { in CreateCollective() [all …]
|
D | collective_util.cc | 54 string SubdivPermDebugString(const CollectiveParams& col_params) { in SubdivPermDebugString() argument 56 col_params.instance.impl_details.subdiv_permutations; in SubdivPermDebugString() 63 CHECK_GT(col_params.instance.device_names.size(), idx); in SubdivPermDebugString() 64 strings::StrAppend(&buf, col_params.instance.device_names[idx], "\n"); in SubdivPermDebugString() 68 for (auto o : col_params.instance.impl_details.subdiv_offsets) in SubdivPermDebugString() 71 for (auto d : col_params.subdiv_rank) strings::StrAppend(&buf, d, " "); in SubdivPermDebugString() 72 if (col_params.instance.type == BROADCAST_COLLECTIVE) { in SubdivPermDebugString() 74 for (auto src : col_params.instance.impl_details.subdiv_source_rank) in SubdivPermDebugString()
|
D | ring_gatherer.cc | 43 Status RingGatherer::InitializeCollectiveParams(CollectiveParams* col_params) { in InitializeCollectiveParams() argument 44 DCHECK_EQ(col_params->instance.type, GATHER_COLLECTIVE); in InitializeCollectiveParams() 45 DCHECK_EQ(col_params->instance.impl_details.collective_name, "RingGather"); in InitializeCollectiveParams() 49 if (!col_params->instance.impl_details.subdiv_offsets.empty() && in InitializeCollectiveParams() 50 (col_params->instance.impl_details.subdiv_offsets.size() > 1 || in InitializeCollectiveParams() 51 col_params->instance.impl_details.subdiv_offsets[0] != 0)) { in InitializeCollectiveParams() 55 if (col_params->instance.impl_details.subdiv_offsets.empty()) { in InitializeCollectiveParams() 56 col_params->instance.impl_details.subdiv_offsets.push_back(0); in InitializeCollectiveParams() 58 return RingAlg::InitializeCollectiveParams(col_params); in InitializeCollectiveParams()
|
D | base_collective_executor.h | 110 void ExecuteAsync(OpKernelContext* ctx, const CollectiveParams& col_params, 148 void WaitForDependencies(const CollectiveParams& col_params) override; 152 void Launched(const CollectiveParams& col_params) override; 166 Status CreateCollective(const CollectiveParams& col_params, 169 bool CheckDependencies(const CollectiveParams& col_params)
|
D | ring_reducer.cc | 46 Status RingReducer::InitializeCollectiveParams(CollectiveParams* col_params) { in InitializeCollectiveParams() argument 48 CHECK_EQ(col_params->instance.type, REDUCTION_COLLECTIVE); in InitializeCollectiveParams() 49 CHECK_EQ(col_params->instance.impl_details.collective_name, "RingReduce"); in InitializeCollectiveParams() 50 return RingAlg::InitializeCollectiveParams(col_params); in InitializeCollectiveParams()
|
D | ring_gatherer.h | 36 Status InitializeCollectiveParams(CollectiveParams* col_params) override;
|
D | ring_reducer.h | 42 Status InitializeCollectiveParams(CollectiveParams* col_params) override;
|
D | collective_util.h | 33 string SubdivPermDebugString(const CollectiveParams& col_params);
|
D | hierarchical_tree_broadcaster.h | 38 Status InitializeCollectiveParams(CollectiveParams* col_params) override;
|
D | ring_alg.h | 38 Status InitializeCollectiveParams(CollectiveParams* col_params) override;
|
/external/tensorflow/tensorflow/core/kernels/ |
D | collective_nccl_reducer.cc | 31 Status NcclReducer::InitializeCollectiveParams(CollectiveParams* col_params) { in InitializeCollectiveParams() argument 32 if (col_params->instance.type != REDUCTION_COLLECTIVE || in InitializeCollectiveParams() 33 col_params->instance.impl_details.collective_name != "NcclReduce") { in InitializeCollectiveParams() 35 col_params->instance.type, " expected ", in InitializeCollectiveParams() 37 col_params->instance.impl_details.collective_name, in InitializeCollectiveParams() 46 col_params_ = &col_ctx->col_params; in InitializeCollectiveContext() 53 CollectiveParams* col_params) { in InitializeInstanceBeforeGroupDiscovery() argument 54 if (col_params->default_rank == 0 && col_params->group.num_tasks > 1) { in InitializeInstanceBeforeGroupDiscovery() 55 col_params->instance.communicator_key = in InitializeInstanceBeforeGroupDiscovery()
|
D | collective_nccl_reducer.h | 29 Status InitializeCollectiveParams(CollectiveParams* col_params) override; 36 CollectiveParams* col_params) override;
|
/external/tensorflow/tensorflow/core/framework/ |
D | collective.h | 262 const CollectiveParams& col_params, in ExecuteAsync() argument 288 virtual void WaitForDependencies(const CollectiveParams& col_params) {} in WaitForDependencies() argument 292 virtual void Launched(const CollectiveParams& col_params) {} in Launched() argument 343 const CollectiveParams& col_params, const string& exec_key, 352 const CollectiveParams& col_params; variable 378 virtual Status InitializeCollectiveParams(CollectiveParams* col_params) = 0; 392 CollectiveParams* col_params) = 0;
|
D | collective.cc | 160 const CollectiveParams& col_params, in CollectiveContext() argument 167 col_params(col_params), in CollectiveContext() 173 device_name(col_params.instance.device_names[col_params.default_rank]) {} in CollectiveContext()
|