@@ -434,27 +434,10 @@ Status GraphExecutor::ExecuteGraphWithStream(GraphId graph_id, | |||
GE_CHECK_NOTNULL_EXEC(ge_root_model, return FAILED); | |||
auto model_id = ge_root_model->GetModelId(); | |||
InputData input_data; | |||
input_data.index = 0; | |||
input_data.model_id = model_id; | |||
std::vector<GeTensorDesc> input_desc; | |||
auto ret = GetExecuteData(input_tensor, input_data.blobs, input_desc); | |||
if (ret != SUCCESS) { | |||
return ret; | |||
} | |||
OutputData output_data; | |||
output_data.index = 0; | |||
output_data.model_id = model_id; | |||
std::vector<GeTensorDesc> output_desc; | |||
ret = GetExecuteData(output_tensor, output_data.blobs, output_desc); | |||
if (ret != SUCCESS) { | |||
return ret; | |||
} | |||
auto async_mode = true; | |||
auto model_manager = ge::ModelManager::GetInstance(); | |||
GE_CHECK_NOTNULL(model_manager); | |||
ret = model_manager->ExecuteModel(model_id, stream, async_mode, input_data, input_desc, output_data, output_desc); | |||
auto ret = model_manager->ExecuteModel(model_id, stream, async_mode, input_tensor, output_tensor); | |||
if (ret != SUCCESS) { | |||
return ret; | |||
} | |||
@@ -3530,6 +3530,31 @@ Status DavinciModel::CopyModelData(const InputData &input_data, OutputData &outp | |||
return SUCCESS; | |||
} | |||
void DavinciModel::BuildZeroCopyTasksLookupTable() { | |||
std::lock_guard<std::mutex> lk(lookup_table_build_lock_); | |||
if (lookup_table_built_) { | |||
return; | |||
} | |||
const auto default_label_hash = std::hash<std::string>{}(kDefaultBatchLable); | |||
for (auto &task : zero_copy_tasks_) { | |||
auto label_hash = std::hash<std::string>{}(task.GetBatchLabel()); | |||
auto addr2offsets = task.GetTaskArgsOffset(); | |||
label_hash2tasks_[label_hash].insert(&task); | |||
if (label_hash == default_label_hash) { | |||
for (auto &addr2offset : addr2offsets) { | |||
addr2default_label_tasks_[addr2offset.first].insert(&task); | |||
} | |||
} else { | |||
for (auto &addr2offset : addr2offsets) { | |||
addr2specific_label_tasks_[addr2offset.first].insert(&task); | |||
} | |||
} | |||
} | |||
lookup_table_built_ = true; | |||
} | |||
/// | |||
/// @ingroup ge | |||
/// @brief Copy Data addr to model for direct use. | |||
@@ -3551,6 +3576,8 @@ Status DavinciModel::UpdateIoTaskArgs(const std::map<uint32_t, ZeroCopyOffset> & | |||
return ACL_ERROR_GE_PARAM_INVALID; | |||
} | |||
BuildZeroCopyTasksLookupTable(); | |||
for (const auto &data : data_info) { | |||
if (data.first >= blobs.size()) { // check data index. | |||
REPORT_INNER_ERROR("E19999", "is_input:%d, data index:%u from model >= blobs.size:%zu from user, mode_id:%u" | |||
@@ -3592,21 +3619,34 @@ Status DavinciModel::UpdateIoTaskArgs(const std::map<uint32_t, ZeroCopyOffset> & | |||
continue; | |||
} | |||
const static auto kDefaultLabelHash = std::hash<std::string>{}(kDefaultBatchLable); | |||
auto batch_label_hash = std::hash<std::string>{}(batch_label); | |||
std::unordered_set<ZeroCopyTask *> same_batch_label_tasks; | |||
if (batch_label_hash != kDefaultLabelHash) { | |||
auto iter = label_hash2tasks_.find(batch_label_hash); | |||
if (iter != label_hash2tasks_.end()) { | |||
same_batch_label_tasks = iter->second; | |||
} | |||
} | |||
for (size_t count = 0; count < data.second.GetDataCount(); ++count) { | |||
void *addr = data.second.GetDataInfo().at(count).second; | |||
auto addr = reinterpret_cast<uintptr_t>(data.second.GetDataInfo().at(count).second); | |||
void *buffer_addr = reinterpret_cast<void *>(reinterpret_cast<uintptr_t>(buffer.data) + | |||
data.second.GetRelativeOffset().at(count)); | |||
GELOGI("[ZCPY] Copy %s blobs_index %u, virtual_addr: %p, size: %ld, user_data_addr: %p, batch_label: %s", | |||
GELOGI("[ZCPY] Copy %s blobs_index %u, virtual_addr: 0x%lx, size: %ld, user_data_addr: %p, batch_label: %s", | |||
is_input ? "input" : "output", data.first, addr, data.second.GetDataInfo().at(count).first, | |||
buffer_addr, batch_label.c_str()); | |||
// For input data, just copy for rts task. | |||
for (auto &task : zero_copy_tasks_) { | |||
bool not_same_batch = (task.GetBatchLabel() != kDefaultBatchLable && task.GetBatchLabel() != batch_label); | |||
if (not_same_batch) { | |||
continue; | |||
for (auto &task : addr2default_label_tasks_[addr]) { // always update default label tasks | |||
(void)task->UpdateTaskParam(addr, buffer_addr); | |||
} | |||
if (batch_label_hash != kDefaultLabelHash) { | |||
for (auto &task : addr2specific_label_tasks_[addr]) { | |||
if (same_batch_label_tasks.count(task) > 0) { | |||
(void)task->UpdateTaskParam(addr, buffer_addr); | |||
} | |||
} | |||
uintptr_t addr_val = reinterpret_cast<uintptr_t>(addr); | |||
(void)task.UpdateTaskParam(addr_val, buffer_addr); | |||
} | |||
} | |||
} | |||
@@ -917,6 +917,7 @@ class DavinciModel { | |||
Status GetGearAndRealOutSizeInfo(const ComputeGraphPtr &graph, const NodePtr &node); | |||
Status GetRealOutputSizeOfCase(const ComputeGraphPtr &graph, size_t input_index, const NodePtr &case_node); | |||
Status GetGearAndRealOutShapeInfo(const ComputeGraphPtr &graph, const NodePtr &node); | |||
void BuildZeroCopyTasksLookupTable(); | |||
bool is_weight_mem_has_inited_; | |||
bool is_feature_map_mem_has_inited_; | |||
@@ -1112,6 +1113,13 @@ class DavinciModel { | |||
// op name to attrs mapping | |||
std::map<std::string, std::map<std::string, std::vector<std::string>>> op_name_to_attrs_; | |||
// fields for build fast search hash table for zero copy tasks | |||
std::mutex lookup_table_build_lock_; | |||
bool lookup_table_built_{false}; | |||
std::unordered_map<size_t, std::unordered_set<ZeroCopyTask*>> label_hash2tasks_; | |||
std::unordered_map<uintptr_t, std::unordered_set<ZeroCopyTask*>> addr2specific_label_tasks_; | |||
std::unordered_map<uintptr_t, std::unordered_set<ZeroCopyTask*>> addr2default_label_tasks_; | |||
std::map<rtStream_t, rtEvent_t> stream_2_event_; | |||
}; | |||
} // namespace ge | |||
@@ -1332,6 +1332,78 @@ Status ModelManager::ExecuteModel(uint32_t model_id, rtStream_t stream, bool asy | |||
return status; | |||
} | |||
namespace { | |||
void GetGeTensorBlobs(const std::vector<GeTensor> &tensors, std::vector<DataBuffer> &blobs) { | |||
blobs.resize(tensors.size()); | |||
for (size_t i = 0; i < tensors.size(); i++) { | |||
auto &tensor = tensors[i]; | |||
auto &buf = blobs[i]; | |||
buf.data = const_cast<uint8_t *>(tensor.GetData().data()); | |||
buf.length = tensor.GetData().size(); | |||
buf.isDataSupportMemShare = false; | |||
} | |||
} | |||
void GetGeTensorDescs(const std::vector<GeTensor> &tensors, std::vector<GeTensorDesc> &descs) { | |||
descs.reserve(tensors.size()); | |||
for (auto &tensor : tensors) { | |||
descs.emplace_back(std::move(tensor.GetTensorDesc())); | |||
} | |||
} | |||
} | |||
ge::Status ModelManager::ExecuteModel(uint32_t model_id, rtStream_t stream, bool async_mode, | |||
const std::vector<GeTensor> &input_tensor, std::vector<GeTensor> &output_tensor) { | |||
InputData input_data; | |||
input_data.index = 0; | |||
input_data.model_id = model_id; | |||
OutputData output_data; | |||
output_data.index = 0; | |||
output_data.model_id = model_id; | |||
GetGeTensorBlobs(input_tensor, input_data.blobs); | |||
GetGeTensorBlobs(output_tensor, output_data.blobs); | |||
std::shared_ptr<hybrid::HybridDavinciModel> hybrid_davinci_model = GetHybridModel(model_id); | |||
if (hybrid_davinci_model != nullptr) { | |||
std::vector<GeTensorDesc> input_desc; | |||
std::vector<GeTensorDesc> output_desc; | |||
GetGeTensorDescs(input_tensor, input_desc); | |||
GetGeTensorDescs(output_tensor, output_desc); | |||
Status status = hybrid_davinci_model->Execute(input_data.blobs, input_desc, output_data.blobs, output_desc, stream); | |||
if (status == SUCCESS) { | |||
GELOGI("Execute model %u success.", model_id); | |||
} | |||
return status; | |||
} | |||
std::shared_ptr<DavinciModel> davinci_model = GetModel(model_id); | |||
GE_CHK_BOOL_RET_STATUS(davinci_model != nullptr, ACL_ERROR_GE_EXEC_MODEL_ID_INVALID, | |||
"[Get][Model] Invalid model id %u, check whether model has been loaded or not.", model_id); | |||
if (davinci_model->NeedDestroyAicpuKernel()) { | |||
GELOGI("Start to destroy specified aicpu kernel."); | |||
// Zero copy is enabled by default, no need to judge. | |||
uint64_t session_id_davinci = davinci_model->GetSessionId(); | |||
uint32_t model_id_davinci = davinci_model->GetModelId(); | |||
uint32_t sub_model_id = davinci_model->SubModelId(); | |||
Status status = DestroyAicpuKernel(session_id_davinci, model_id_davinci, sub_model_id); | |||
if (status != SUCCESS) { | |||
GELOGW("Destroy specified aicpu kernel failed, session id is %lu, model id is %u.", session_id_davinci, | |||
model_id_davinci); | |||
} | |||
} | |||
Status status = davinci_model->NnExecute(stream, async_mode, input_data, output_data); | |||
if (status == SUCCESS) { | |||
GELOGD("Execute model %u success.", model_id); | |||
} | |||
return status; | |||
} | |||
Status ModelManager::CreateAicpuSession(uint64_t session_id) { | |||
std::lock_guard<std::recursive_mutex> lock(map_mutex_); | |||
auto it = sess_ids_.find(session_id); | |||
@@ -157,6 +157,9 @@ class FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY ModelManager { | |||
const std::vector<GeTensorDesc> &input_desc, OutputData &output_data, | |||
std::vector<GeTensorDesc> &output_desc); | |||
ge::Status ExecuteModel(uint32_t model_id, rtStream_t stream, bool async_mode, const std::vector<GeTensor> &inputs, | |||
std::vector<GeTensor> &outputs); | |||
ge::Status SyncExecuteModel(uint32_t model_id, const std::vector<GeTensor> &inputs, std::vector<GeTensor> &outputs); | |||
/// | |||
@@ -54,6 +54,10 @@ Status ZeroCopyTask::SetTaskArgsOffset(uintptr_t addr, size_t offset) { | |||
return SUCCESS; | |||
} | |||
const std::map<uintptr_t, std::set<size_t >>& ZeroCopyTask::GetTaskArgsOffset() const { | |||
return task_addr_offset_; | |||
} | |||
/** | |||
* @ingroup ge | |||
* @brief Save orignal data of task args. | |||
@@ -80,17 +84,16 @@ void ZeroCopyTask::SetOriginalArgs(const void *info, size_t size) { | |||
Status ZeroCopyTask::UpdateTaskParam(uintptr_t addr, void *buffer_addr) { | |||
auto iter = task_addr_offset_.find(addr); | |||
if (iter != task_addr_offset_.end()) { | |||
auto &cur_pair = *iter; | |||
auto dst_addr = reinterpret_cast<uintptr_t>(static_cast<uint8_t *>(buffer_addr)); | |||
uint8_t *args_info = args_info_.data(); | |||
for (auto offset : cur_pair.second) { | |||
auto dst_addr = static_cast<uint8_t *>(buffer_addr); | |||
GELOGI("[ZCPY] %s update task, args_addr: %p, size: %zu, offset: %zu, virtual_addr: 0x%lx, user_data_addr: %p", | |||
name_.c_str(), args_addr_, args_size_, offset, addr, buffer_addr); | |||
*reinterpret_cast<uintptr_t *>(args_info + offset)= reinterpret_cast<uintptr_t>(dst_addr); | |||
is_updated_ = true; | |||
for (auto offset : iter->second) { | |||
auto ¤t_addr = *reinterpret_cast<uintptr_t *>(args_info + offset); | |||
if (current_addr != dst_addr) { | |||
current_addr = dst_addr; | |||
is_updated_ = true; | |||
} | |||
} | |||
} | |||
return SUCCESS; | |||
} | |||
@@ -46,6 +46,8 @@ class ZeroCopyTask { | |||
*/ | |||
ge::Status SetTaskArgsOffset(uintptr_t addr, size_t offset); | |||
const std::map<uintptr_t, std::set<size_t >>& GetTaskArgsOffset() const; | |||
/** | |||
* @ingroup ge | |||
* @brief Is need zero copy. | |||
@@ -341,12 +341,14 @@ Status InnerSession::RunGraphWithStreamAsync(uint32_t graph_id, rtStream_t strea | |||
} | |||
UpdateThreadContext(graph_id); | |||
vector<GeTensor> ge_inputs; | |||
ge_inputs.reserve(inputs.size()); | |||
for (auto &item : inputs) { | |||
ge_inputs.emplace_back(TensorAdapter::AsGeTensor(item)); | |||
ge_inputs.emplace_back(TensorAdapter::AsGeTensorShared(item)); | |||
} | |||
vector<GeTensor> ge_outputs; | |||
ge_outputs.reserve(outputs.size()); | |||
for (auto &item : outputs) { | |||
ge_outputs.emplace_back(TensorAdapter::AsGeTensor(item)); | |||
ge_outputs.emplace_back(TensorAdapter::AsGeTensorShared(item)); | |||
} | |||
Status ret = graph_manager_.RunGraphWithStreamAsync(graph_id, stream, session_id_, ge_inputs, ge_outputs); | |||
domi::GetContext().out_nodes_map.clear(); | |||