!3549 MapOp refactor for device augmentation (part 1)

Merge pull request !3549 from anthonyaje/device_augment_mapop_refactor1

mindspore/ccsrc/minddata/dataset/CMakeLists.txt

@@ -66,6 +66,7 @@ add_dependencies(kernels core)
 add_dependencies(engine-datasetops-source core)
 add_dependencies(engine-datasetops-source-sampler core)
 add_dependencies(engine-datasetops core)
+add_dependencies(engine-datasetops-mapop core)
 add_dependencies(engine-opt core)
 add_dependencies(engine-perf core)
 add_dependencies(engine-gnn core)
@@ -89,6 +90,7 @@ set(submodules
     $<TARGET_OBJECTS:cpp-API>
     $<TARGET_OBJECTS:engine-datasetops-source>
     $<TARGET_OBJECTS:engine-datasetops-source-sampler>
+    $<TARGET_OBJECTS:engine-datasetops-mapop>
     $<TARGET_OBJECTS:engine-gnn>
     $<TARGET_OBJECTS:engine-perf>
     $<TARGET_OBJECTS:engine-datasetops>

mindspore/ccsrc/minddata/dataset/api/datasets.cc

@@ -27,7 +27,7 @@
 #include "minddata/dataset/engine/datasetops/source/voc_op.h"
 // Dataset operator headers (in alphabetical order)
 #include "minddata/dataset/engine/datasetops/batch_op.h"
-#include "minddata/dataset/engine/datasetops/map_op.h"
+#include "minddata/dataset/engine/datasetops/map_op/map_op.h"
 #include "minddata/dataset/engine/datasetops/project_op.h"
 #include "minddata/dataset/engine/datasetops/rename_op.h"
 #include "minddata/dataset/engine/datasetops/repeat_op.h"
@@ -537,9 +537,6 @@ std::vector<std::shared_ptr<DatasetOp>> MapDataset::Build() {
   // A vector containing shared pointer to the Dataset Ops that this object will create
   std::vector<std::shared_ptr<DatasetOp>> node_ops;
 
-  // Currently default is true, and this is not exposed to user.
-  bool perf_mode = true;
-
   std::vector<std::shared_ptr<TensorOp>> tensor_ops;
 
   // Build tensorOp from tensorOperation vector
@@ -550,8 +547,7 @@ std::vector<std::shared_ptr<DatasetOp>> MapDataset::Build() {
 
   // This parameter will be removed with next rebase
   std::vector<std::string> col_orders;
-  auto map_op =
-    std::make_shared<MapOp>(input_columns_, output_columns_, tensor_ops, num_workers_, connector_que_size_, perf_mode);
+  auto map_op = std::make_shared<MapOp>(input_columns_, output_columns_, tensor_ops, num_workers_, connector_que_size_);
   if (!project_columns_.empty()) {
     auto project_op = std::make_shared<ProjectOp>(project_columns_);
     node_ops.push_back(project_op);

mindspore/ccsrc/minddata/dataset/core/client.h

@@ -39,7 +39,7 @@
 #include "minddata/dataset/engine/datasetops/batch_op.h"
 #include "minddata/dataset/engine/datasetops/dataset_op.h"
 #include "minddata/dataset/engine/datasetops/device_queue_op.h"
-#include "minddata/dataset/engine/datasetops/map_op.h"
+#include "minddata/dataset/engine/datasetops/map_op/map_op.h"
 #include "minddata/dataset/engine/datasetops/project_op.h"
 #include "minddata/dataset/engine/datasetops/rename_op.h"
 #include "minddata/dataset/engine/datasetops/repeat_op.h"

mindspore/ccsrc/minddata/dataset/core/constants.h

@@ -26,6 +26,9 @@ namespace dataset {
 using uchar = unsigned char;
 using dsize_t = int64_t;
 
+// Target devices to perform map operation
+enum class MapTargetDevice { kCpu, kGpu, kDvpp };
+
 // Possible dataset types for holding the data and client type
 enum class DatasetType { kUnknown, kArrow, kTf };
 

mindspore/ccsrc/minddata/dataset/engine/CMakeLists.txt

@@ -22,8 +22,8 @@ endif()
 
 if (ENABLE_TDTQUE)
   add_dependencies(engine engine-datasetops engine-datasetops-source engine-tdt engine-opt engine-gnn engine-perf
-                   engine-cache-client engine-cache-server)
+                   engine-cache-client engine-cache-server engine-datasetops-mapop)
 else ()
   add_dependencies(engine engine-datasetops engine-datasetops-source engine-opt engine-gnn engine-perf
-                   engine-cache-client engine-cache-server)
+                   engine-cache-client engine-cache-server engine-datasetops-mapop)
 endif ()

mindspore/ccsrc/minddata/dataset/engine/datasetops/CMakeLists.txt

 add_subdirectory(source)
+add_subdirectory(map_op)
 
 file(GLOB_RECURSE _CURRENT_SRC_FILES RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} "*.cc")
 set_property(SOURCE ${_CURRENT_SRC_FILES} PROPERTY COMPILE_DEFINITIONS SUBMODULE_ID=mindspore::SubModuleId::SM_MD)
@@ -9,7 +10,6 @@ set(DATASET_ENGINE_DATASETOPS_SRC_FILES
     pipeline_op.cc
     batch_op.cc
     device_queue_op.cc
-    map_op.cc
     project_op.cc
     rename_op.cc
     repeat_op.cc
@@ -37,4 +37,3 @@ if (ENABLE_PYTHON)
 endif()
 
 add_library(engine-datasetops OBJECT ${DATASET_ENGINE_DATASETOPS_SRC_FILES})
-

mindspore/ccsrc/minddata/dataset/engine/datasetops/map_op/CMakeLists.txt

+file(GLOB_RECURSE _CURRENT_SRC_FILES RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} "*.cc")
+set_property(SOURCE ${_CURRENT_SRC_FILES} PROPERTY COMPILE_DEFINITIONS SUBMODULE_ID=mindspore::SubModuleId::SM_MD)
+
+set(DATASET_ENGINE_DATASETOPS_MAPOP_SRC_FILES
+  map_op.cc
+  cpu_map_job.cc
+  gpu_map_job.cc
+  )
+
+add_library(engine-datasetops-mapop OBJECT ${DATASET_ENGINE_DATASETOPS_MAPOP_SRC_FILES})

mindspore/ccsrc/minddata/dataset/engine/datasetops/map_op/cpu_map_job.cc

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <memory>
+#include <vector>
+#include <utility>
+#include "minddata/dataset/engine/datasetops/map_op/cpu_map_job.h"
+
+namespace mindspore {
+namespace dataset {
+
+// Constructor
+CpuMapJob::CpuMapJob() = default;
+
+// Constructor
+CpuMapJob::CpuMapJob(std::vector<std::shared_ptr<TensorOp>> operations) : MapJob(operations) {}
+
+// Destructor
+CpuMapJob::~CpuMapJob() = default;
+
+// A function to execute a cpu map job
+Status CpuMapJob::Run(std::vector<TensorRow> in, std::vector<TensorRow> *out) {
+  int32_t num_rows = in.size();
+  for (int32_t row = 0; row < num_rows; row++) {
+    TensorRow input_row = in[row];
+    TensorRow result_row;
+    for (size_t i = 0; i < ops_.size(); i++) {
+      // Call compute function for cpu
+      RETURN_IF_NOT_OK(ops_[i]->Compute(input_row, &result_row));
+
+      // Assign result_row to to_process for the next TensorOp processing, except for the last TensorOp in the list.
+      if (i + 1 < ops_.size()) {
+        input_row = std::move(result_row);
+      }
+    }
+    out->push_back(std::move(result_row));
+  }
+
+  return Status::OK();
+}
+
+}  // namespace dataset
+}  // namespace mindspore

mindspore/ccsrc/minddata/dataset/engine/datasetops/map_op/cpu_map_job.h

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef DATASET_ENGINE_DATASETOPS_MAP_OP_CPU_MAP_JOB_H_
+#define DATASET_ENGINE_DATASETOPS_MAP_OP_CPU_MAP_JOB_H_
+
+#include <memory>
+#include <vector>
+#include "minddata/dataset/engine/datasetops/map_op/map_job.h"
+
+namespace mindspore {
+namespace dataset {
+class CpuMapJob : public MapJob {
+ public:
+  // Constructor
+  CpuMapJob();
+
+  // Constructor
+  explicit CpuMapJob(std::vector<std::shared_ptr<TensorOp>> operations);
+
+  // Destructor
+  ~CpuMapJob();
+
+  // A pure virtual run function to execute a cpu map job
+  Status Run(std::vector<TensorRow> in, std::vector<TensorRow> *out) override;
+};
+
+}  // namespace dataset
+}  // namespace mindspore
+
+#endif  // DATASET_ENGINE_DATASETOPS_MAP_OP_CPU_MAP_JOB_H_

mindspore/ccsrc/minddata/dataset/engine/datasetops/map_op/gpu_map_job.cc

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <memory>
+#include <vector>
+#include "minddata/dataset/engine/datasetops/map_op/gpu_map_job.h"
+
+namespace mindspore {
+namespace dataset {
+
+// Constructor
+GpuMapJob::GpuMapJob(std::vector<std::shared_ptr<TensorOp>> operations) : MapJob(operations) {}
+
+// Destructor
+GpuMapJob::~GpuMapJob() = default;
+
+// A function to execute a cpu map job
+Status GpuMapJob::Run(std::vector<TensorRow> in, std::vector<TensorRow> *out) {
+  // Do nothing for now
+  return Status::OK();
+}
+
+}  // namespace dataset
+}  // namespace mindspore

mindspore/ccsrc/minddata/dataset/engine/datasetops/map_op/gpu_map_job.h

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef DATASET_ENGINE_DATASETOPS_MAP_OP_GPU_MAP_JOB_H_
+#define DATASET_ENGINE_DATASETOPS_MAP_OP_GPU_MAP_JOB_H_
+
+#include <memory>
+#include <vector>
+#include "minddata/dataset/engine/datasetops/map_op/map_job.h"
+
+namespace mindspore {
+namespace dataset {
+class GpuMapJob : public MapJob {
+ public:
+  // Constructor
+  explicit GpuMapJob(std::vector<std::shared_ptr<TensorOp>> operations);
+
+  // Destructor
+  ~GpuMapJob();
+
+  // A pure virtual run function to execute a cpu map job
+  Status Run(std::vector<TensorRow> in, std::vector<TensorRow> *out) override;
+};
+
+}  // namespace dataset
+}  // namespace mindspore
+
+#endif  // DATASET_ENGINE_DATASETOPS_MAP_OP_GPU_MAP_JOB_H_

mindspore/ccsrc/minddata/dataset/engine/datasetops/map_op/map_job.h

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef DATASET_ENGINE_DATASETOPS_MAP_OP_MAP_JOB_H_
+#define DATASET_ENGINE_DATASETOPS_MAP_OP_MAP_JOB_H_
+
+#include <memory>
+#include <vector>
+
+#include "minddata/dataset/kernels/tensor_op.h"
+#include "minddata/dataset/core/tensor.h"
+#include "minddata/dataset/core/tensor_row.h"
+#include "minddata/dataset/util/status.h"
+
+namespace mindspore {
+namespace dataset {
+class MapJob {
+ public:
+  // Constructor
+  explicit MapJob(std::vector<std::shared_ptr<TensorOp>> operations) : ops_(operations) {}
+
+  // Constructor
+  MapJob() = default;
+
+  // Destructor
+  ~MapJob() = default;
+
+  Status AddOperation(std::shared_ptr<TensorOp> operation) {
+    ops_.push_back(operation);
+    return Status::OK();
+  }
+
+  // A pure virtual run function to execute a particular map job
+  virtual Status Run(std::vector<TensorRow> in, std::vector<TensorRow> *out) = 0;
+
+ protected:
+  std::vector<std::shared_ptr<TensorOp>> ops_;
+};
+
+}  // namespace dataset
+}  // namespace mindspore
+
+#endif  // DATASET_ENGINE_DATASETOPS_MAP_OP_MAP_JOB_H_

mindspore/ccsrc/minddata/dataset/engine/datasetops/map_op.cc → mindspore/ccsrc/minddata/dataset/engine/datasetops/map_op/map_op.cc

@@ -13,7 +13,7 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#include "minddata/dataset/engine/datasetops/map_op.h"
+#include <algorithm>
 #include <cstring>
 #include <iomanip>
 #include <iostream>
@@ -28,6 +28,9 @@
 #include "minddata/dataset/engine/db_connector.h"
 #include "minddata/dataset/engine/execution_tree.h"
 #include "minddata/dataset/engine/opt/pass.h"
+#include "minddata/dataset/engine/datasetops/map_op/map_op.h"
+#include "minddata/dataset/engine/datasetops/map_op/cpu_map_job.h"
+#include "minddata/dataset/engine/datasetops/map_op/gpu_map_job.h"
 #include "minddata/dataset/kernels/tensor_op.h"
 #include "utils/log_adapter.h"
 #include "minddata/dataset/util/task_manager.h"
@@ -35,7 +38,7 @@
 namespace mindspore {
 namespace dataset {
 // Builder constructor. Creates the builder object.
-MapOp::Builder::Builder() : build_perf_mode_(true) {
+MapOp::Builder::Builder() {
   std::shared_ptr<ConfigManager> cfg = GlobalContext::config_manager();
   build_num_workers_ = cfg->num_parallel_workers();
   build_op_connector_size_ = cfg->op_connector_size();
@@ -54,31 +57,27 @@ Status MapOp::Builder::sanityCheck() const {
 Status MapOp::Builder::Build(std::shared_ptr<MapOp> *ptr) {
   RETURN_IF_NOT_OK(sanityCheck());
   *ptr = std::make_shared<MapOp>(std::move(build_in_col_names_), std::move(build_out_col_names_),
-                                 std::move(build_tensor_funcs_), build_num_workers_, build_op_connector_size_,
-                                 build_perf_mode_);
+                                 std::move(build_tensor_funcs_), build_num_workers_, build_op_connector_size_);
   return Status::OK();
 }
 
 // Constructor of MapOp
 MapOp::MapOp(const std::vector<std::string> &in_col_names, const std::vector<std::string> &out_col_names,
-             std::vector<std::shared_ptr<TensorOp>> tensor_funcs, int32_t num_workers, int32_t op_connector_size,
-             bool perf_mode)
+             std::vector<std::shared_ptr<TensorOp>> tensor_funcs, int32_t num_workers, int32_t op_connector_size)
     : ParallelOp(num_workers, op_connector_size),
       tfuncs_(std::move(tensor_funcs)),
       in_columns_(in_col_names),
-      out_columns_(out_col_names),
-      perf_mode_(perf_mode) {
+      out_columns_(out_col_names) {
   // If caller didn't specify the out_col_names, assume they are same as the in_columns.
   if (out_columns_.empty() || out_columns_[0].empty()) {
     out_columns_ = in_columns_;
   }
-  MS_LOG(DEBUG) << "Performance Mode in map operator is " << perf_mode_ << ".";
 }
 
 // The number of threads consuming data from previous op's output Connector.
 int32_t MapOp::num_consumers() const {
   // When Performance Mode is on, there is only one thread consuming from the previous Connector.
-  return perf_mode_ == true ? 1 : num_workers_;
+  return 1;
 }
 
 // A print method typically used for debugging
@@ -106,36 +105,98 @@ void MapOp::Print(std::ostream &out, bool show_all) const {
   }
 }
 
+// A helper function that fetch worker map job from local queues and extract the data and map job list
+Status MapOp::FetchNextWork(uint32_t worker_id, std::unique_ptr<DataBuffer> *db,
+                            std::vector<std::shared_ptr<MapJob>> *job_list) {
+  std::unique_ptr<MapWorkerJob> worker_job;
+  // Fetch the next worker job and data buffer
+  RETURN_IF_NOT_OK(local_queues_[worker_id]->PopFront(&worker_job));
+  // Extract the databuffer and job list from the map worker job.
+  *db = std::move(worker_job->databuffer);
+  *job_list = std::move(worker_job->jobs);
+
+  return Status::OK();
+}
+
+Status MapOp::GenerateWorkerJob(const std::unique_ptr<MapWorkerJob> *worker_job) {
+  std::shared_ptr<MapJob> map_job = nullptr;
+  MapTargetDevice prev_target;
+  for (size_t i = 0; i < tfuncs_.size(); i++) {
+    // Currently we only have CPU as the device target
+    // In the future, we will have heuristic or control from user to select target device
+    // MapTargetDevice target_device;
+    // RETURN_IF_NOT_OK(SelectTarget(tfuncs_[i], &target_device));
+    MapTargetDevice target_device = MapTargetDevice::kCpu;
+
+    switch (target_device) {
+      case MapTargetDevice::kCpu:
+        // If there is no existing map_job, we will create one.
+        // map_job could be nullptr when we are at the first tensor op or when the target device of the prev op
+        // is different with that of the current op.
+        if (map_job == nullptr) {
+          map_job = std::make_shared<CpuMapJob>();
+        }
+        map_job->AddOperation(tfuncs_[i]);
+        break;
+
+      case MapTargetDevice::kGpu:
+        break;
+
+      case MapTargetDevice::kDvpp:
+        break;
+
+      default:
+        break;
+    }
+
+    // Push map_job into worker_job if one of the two conditions is true:
+    // 1) It is the last tensor operation in tfuncs_
+    // 2) The the target device of the current tensor operation is different with previous one
+    if ((i + 1 == tfuncs_.size()) || ((i != 0) && (prev_target != target_device))) {
+      (*worker_job)->jobs.push_back(std::move(map_job));
+    }
+
+    prev_target = target_device;
+  }
+
+  return Status::OK();
+}
+
 // This class functor will provide the master loop that drives the logic for performing the work
 Status MapOp::operator()() {
-  if (perf_mode_) {
-    // Create and register the local queues.
-    local_queues_.Init(num_workers_, oc_queue_size_);
-    Status rc = local_queues_.Register(tree_->AllTasks());
-    if (rc.IsError()) {
-      TaskManager::FindMe()->Post();
-      return rc;
-    }
+  // Create and register the local queues.
+  local_queues_.Init(num_workers_, oc_queue_size_);
+  Status rc = local_queues_.Register(tree_->AllTasks());
+  if (rc.IsError()) {
+    TaskManager::FindMe()->Post();
+    return rc;
   }
 
   // The operator class just starts off threads by calling the tree_ function
-  Status rc = tree_->LaunchWorkers(num_workers_, std::bind(&MapOp::WorkerEntry, this, std::placeholders::_1));
+  rc = tree_->LaunchWorkers(num_workers_, std::bind(&MapOp::WorkerEntry, this, std::placeholders::_1));
   // Synchronize with TaskManager
   TaskManager::FindMe()->Post();
   RETURN_IF_NOT_OK(rc);
 
-  if (perf_mode_) {
-    int64_t que_id = 0;
-    std::unique_ptr<DataBuffer> buff;
-    bool is_eof = false;
-    // Draining output connector of the previous op and distribute it to local queues.
-    // Stop when all worker threads are finished (received EOF).
-    while (!is_eof) {
-      RETURN_IF_NOT_OK(child_[0]->GetNextBuffer(&buff, 0));
-      is_eof = buff->eof();
-      RETURN_IF_NOT_OK(local_queues_[que_id]->Add(std::move(buff)));
-      que_id = (que_id + 1) % num_workers_;
-    }
+  int64_t que_id = 0;
+  std::unique_ptr<DataBuffer> buff;
+  bool is_eof = false;
+  // Drain output connector of the previous op, generate jobs for worker threads, and distribute them via local queues
+  // Stop when all worker threads are finished (received EOF)
+  while (!is_eof) {
+    RETURN_IF_NOT_OK(child_[0]->GetNextBuffer(&buff, 0));
+    is_eof = buff->eof();
+
+    // Create an empty map worker job to be populated by a databuffer and map jobs
+    std::unique_ptr<MapWorkerJob> worker_job = std::make_unique<MapWorkerJob>();
+    worker_job->databuffer = std::move(buff);
+
+    // Populate map worker job for a worker to execute
+    RETURN_IF_NOT_OK(GenerateWorkerJob(&worker_job));
+
+    // Push map worker job to the corresponding worker's queue
+    RETURN_IF_NOT_OK(local_queues_[que_id]->Add(std::move(worker_job)));
+    que_id = (que_id + 1) % num_workers_;
   }
 
   return Status::OK();
@@ -148,12 +209,11 @@ Status MapOp::operator()() {
 Status MapOp::WorkerEntry(int32_t worker_id) {
   // Handshake with TaskManager that thread creation is successful.
   TaskManager::FindMe()->Post();
-  std::unique_ptr<DataBuffer> in_buffer;
 
-  // Getting a databuffer to work on.
-  // Perform the first fetch here outside of the loop.  This allows us to execute one-time only
-  // initializations that happen after the first fetch.
-  RETURN_IF_NOT_OK(FetchNextBuffer(&in_buffer, worker_id));
+  std::unique_ptr<DataBuffer> in_buffer;
+  std::vector<std::shared_ptr<MapJob>> job_list;
+  // Fetch next data buffer and map job list
+  RETURN_IF_NOT_OK(FetchNextWork(worker_id, &in_buffer, &job_list));
 
   // Sanity check the databuffer.
   // Special case: if there's more threads than buffers, some threads simply get the final control
@@ -175,7 +235,8 @@ Status MapOp::WorkerEntry(int32_t worker_id) {
     if (in_buffer->eoe()) {
       // Calling base class EoeReceived to forward eoe buffer.
       RETURN_IF_NOT_OK(EoeReceived(worker_id));
-      RETURN_IF_NOT_OK(FetchNextBuffer(&in_buffer, worker_id));
+      // Fetch next data buffer and map job list
+      RETURN_IF_NOT_OK(FetchNextWork(worker_id, &in_buffer, &job_list));
       continue;
     } else if (in_buffer->eof()) {
       // Calling base class EofReceived to forward eof buffer.
@@ -185,76 +246,77 @@ Status MapOp::WorkerEntry(int32_t worker_id) {
 
     std::unique_ptr<TensorQTable> new_tensor_table(std::make_unique<TensorQTable>());
     // Perform the compute function of TensorOp(s) and store the result in new_tensor_table.
-    RETURN_IF_NOT_OK(WorkerCompute(in_buffer.get(), new_tensor_table.get()));
-
+    RETURN_IF_NOT_OK(WorkerCompute(in_buffer.get(), new_tensor_table.get(), job_list));
     // Replace the TensorTable in DataBuffer with the new one.
     in_buffer->set_tensor_table(std::move(new_tensor_table));
-
     // Push the buffer onto the connector for next operator to consume.
     RETURN_IF_NOT_OK(out_connector_->Add(static_cast<int>(worker_id), std::move(in_buffer)));
-
-    // Fetch the next buffer and loop back to the top.
-    RETURN_IF_NOT_OK(FetchNextBuffer(&in_buffer, worker_id));
+    // Fetch next data buffer and map job list
+    RETURN_IF_NOT_OK(FetchNextWork(worker_id, &in_buffer, &job_list));
   }
-
   return Status::OK();
 }
 
-Status MapOp::WorkerCompute(DataBuffer *in_buffer, TensorQTable *new_tensor_table) {
-  // Getting number of rows and cols in this buffer.
+Status MapOp::WorkerCompute(DataBuffer *in_buffer, TensorQTable *new_tensor_table,
+                            const std::vector<std::shared_ptr<MapJob>> &job_list) {
   int32_t num_rows = in_buffer->NumRows();
   int32_t num_cols = in_buffer->NumCols();
 
+  std::vector<TensorRow> job_input_table;
+  std::vector<TensorRow> original_table;
+
+  // Prepare the data that we need from in_buffer
   for (int32_t r = 0; r < num_rows; r++) {
     // to_process   : A vector of Tensors only holding cols in input_columns.
-    // result_row;  : A vector of Tensors to hold the result after Compute().
-    // cur_row      : A vector of Tensors holding all the columns from DataBuffer.
-    TensorRow to_process, result_row, cur_row;
+    // cur_row      : A vector of Tensors holding all the cols from DataBuffer.
+    TensorRow to_process, cur_row;
     RETURN_IF_NOT_OK(in_buffer->PopRow(&cur_row));
+    // From the current row, select the Tensor that need to be passed to TensorOp
+    (void)std::transform(to_process_indices_.begin(), to_process_indices_.end(), std::back_inserter(to_process),
+                         [&cur_row](const auto &it) { return std::move(cur_row[it]); });
+    job_input_table.push_back(std::move(to_process));
+    original_table.push_back(std::move(cur_row));
+  }
 
-    // Populate the Tensor from the current row to be processed by TensorOp
-    for (const auto &idx : to_process_indices_) {
-      to_process.push_back(std::move(cur_row[idx]));
-    }
-
-    // Looping over multiple TensorOps supplied in to MapOp.
-    // The assumption is that the result of one TensorOp matches the required input to the next TensorOp.
-    for (size_t i = 0; i < tfuncs_.size(); i++) {
-      // TensorOp can operate on single col or multiple cols. MapOp always call compute for multiple cols.
-      // TensorOp base class will call the single column Compute() depending on the ops.
-      // Note: The columns of the result_row is not preallocated, the compute function of each tensor op are
-      // required to resize/push back the result_row
-      RETURN_IF_NOT_OK(tfuncs_[i]->Compute(to_process, &result_row));
-
-      // Assign result_row to to_process for the next TensorOp processing, except for the last TensorOp in the list.
-      if (i + 1 < tfuncs_.size()) {
-        to_process = std::move(result_row);
-      }
+  // Variable to keep the result after executing the job.
+  std::vector<TensorRow> result_table;
+  // Executing the list of jobs
+  for (size_t i = 0; i < job_list.size(); i++) {
+    // Executre MapJob.
+    RETURN_IF_NOT_OK(job_list[i]->Run(job_input_table, &result_table));
+    // Assign the pocessed data as an input for the next job processing, except for the last TensorOp in the list.
+    if (i + 1 < job_list.size()) {
+      job_input_table = std::move(result_table);
     }
+  }
 
-    if (out_columns_.size() != result_row.size()) {
-      return Status(StatusCode::kUnexpectedError, __LINE__, __FILE__,
-                    "Result of a tensorOp doesn't match output column names");
-    }
+  // Sanity check a row in result_table
+  if (!result_table.empty() && out_columns_.size() != result_table[0].size()) {
+    return Status(StatusCode::kUnexpectedError, __LINE__, __FILE__,
+                  "Result of a tensorOp doesn't match output column names");
+  }
 
+  // Merging the data processed by job (result_table) with the data that are not used.
+  for (int32_t r = 0; r < num_rows; r++) {
+    TensorRow out_row;
     if (in_columns_.size() == out_columns_.size()) {
-      for (size_t i = 0; i < result_row.size(); i++) {
-        cur_row[to_process_indices_[i]] = std::move(result_row[i]);
+      // Place the processed tensor back into the original index of the input tensor
+      for (size_t i = 0; i < result_table[r].size(); i++) {
+        original_table[r][to_process_indices_[i]] = std::move(result_table[r][i]);
       }
-      new_tensor_table->push_back(std::move(cur_row));
+      out_row = std::move(original_table[r]);
     } else {
-      // Add the columns we did not touch to the result_row.
+      // Append the data in the original table that we did not use to the end of each row in result_table.
       for (int32_t i = 0; i < num_cols; i++) {
         if (keep_input_columns_[i]) {
-          result_row.push_back(std::move(cur_row[i]));
+          result_table[r].push_back(std::move(original_table[r][i]));
         }
       }
-
-      // Add this final result_row to our new TensorTable.
-      new_tensor_table->push_back(std::move(result_row));
+      out_row = std::move(result_table[r]);
     }
+    // Add this final out_row to our new TensorTable.
+    new_tensor_table->push_back(std::move(out_row));
   }
-
   return Status::OK();
 }
 
@@ -288,13 +350,11 @@ Status MapOp::ValidateInColumns(const std::unordered_map<std::string, int32_t> &
 Status MapOp::InitPrivateVariable(std::unordered_map<std::string, int32_t> *col_name_id_map) {
   // If input_columns is empty(), The col at index-0 will be picked.
   if (in_columns_.empty()) {
-    for (const auto &pair : *col_name_id_map) {
-      if (pair.second == 0) {
-        MS_LOG(INFO) << "Input columns empty for map op, will apply to the first column in the current table.";
-        in_columns_.push_back(pair.first);
-        break;
-      }
-    }
+    auto itr =
+      std::find_if(col_name_id_map->begin(), col_name_id_map->end(), [](const auto &it) { return it.second == 0; });
+    CHECK_FAIL_RETURN_UNEXPECTED(itr != col_name_id_map->end(), "Column name id map doesn't have id 0");
+    MS_LOG(INFO) << "Input columns empty for map op, will apply to the first column in the current table.";
+    in_columns_.push_back(itr->first);
 
     // If caller didn't specify the out_col_names, assume they are same as the input_columns.
     // This was done in the constructor, but if input columns was empty to start we have to redo it here.

mindspore/ccsrc/minddata/dataset/engine/datasetops/map_op.h → mindspore/ccsrc/minddata/dataset/engine/datasetops/map_op/map_op.h

@@ -24,6 +24,7 @@
 #include "minddata/dataset/engine/datasetops/parallel_op.h"
 #include "minddata/dataset/kernels/tensor_op.h"
 #include "minddata/dataset/util/queue.h"
+#include "minddata/dataset/engine/datasetops/map_op/map_job.h"
 
 namespace mindspore {
 namespace dataset {
@@ -107,13 +108,6 @@ class MapOp : public ParallelOp {
       return *this;
     }
 
-    // Setter method.
-    // @return Builder setter method returns reference to the builder.
-    Builder &SetPerformanceMode(bool perf_mode) {
-      build_perf_mode_ = perf_mode;
-      return *this;
-    }
-
     // The builder "build" method creates the final object.
     // @param ptr The shared_ptr to the new MapOp object
     // @return Status
@@ -125,7 +119,6 @@ class MapOp : public ParallelOp {
     std::vector<std::shared_ptr<TensorOp>> build_tensor_funcs_;
     int32_t build_num_workers_;
     int32_t build_op_connector_size_;
-    bool build_perf_mode_;  // Default true.
 
     // Check if the required parameters are set by the builder.
     // @return Status The error code return
@@ -140,8 +133,7 @@ class MapOp : public ParallelOp {
   // @param num_workers The number of worker threads.
   // @param op_connector_size The size of each queue in the connector.
   MapOp(const std::vector<std::string> &in_col_names, const std::vector<std::string> &out_col_names,
-        std::vector<std::shared_ptr<TensorOp>> tensor_funcs, int32_t num_workers, int32_t op_connector_size,
-        bool perf_mode);
+        std::vector<std::shared_ptr<TensorOp>> tensor_funcs, int32_t num_workers, int32_t op_connector_size);
 
   // Destructor
   ~MapOp() = default;
@@ -164,6 +156,8 @@ class MapOp : public ParallelOp {
   // Class functor operator () override.
   // All dataset ops operate by launching a thread (see ExecutionTree). This class functor will
   // provide the master loop that drives the logic for performing the work
+  // This main thread creates local queues, pulls databuffers from the previous
+  // op's Connector and distributes them to the local queues. Workers pull from the local queues.
   // @return Status The error code return
   Status operator()() override;
 
@@ -189,12 +183,24 @@ class MapOp : public ParallelOp {
   const auto &TFuncs() const { return tfuncs_; }
 
  private:
-  // Local queues where worker threads can pop from.
-  // Popping directly from the Connector can block if the previous designated threads haven't pop.
-  // Setting the size of these queues to 0 is essentially the same as pulling directly from Connector.
-  QueueList<std::unique_ptr<DataBuffer>> local_queues_;
+  // A unit of job for map worker thread.
+  // MapWorkerJob holds a list of MapJob where each MapJob can be a CpuMapJob, GpuMapJob or DvppMapJob.
+  struct MapWorkerJob {
+    std::vector<std::shared_ptr<MapJob>> jobs;
+    std::unique_ptr<DataBuffer> databuffer;
+  };
+
+  // A helper function to create jobs for workers.
+  Status GenerateWorkerJob(const std::unique_ptr<MapWorkerJob> *worker_job);
 
-  // Static variables to be ready by worker threads, no modification and readonly
+  // A helper function that fetch worker map job from local queues and extract the data and map job list
+  Status FetchNextWork(uint32_t worker_id, std::unique_ptr<DataBuffer> *db,
+                       std::vector<std::shared_ptr<MapJob>> *job_list);
+
+  // Local queues where worker threads get a job from
+  QueueList<std::unique_ptr<MapWorkerJob>> local_queues_;
+
+  //  Tensorops to be read and applied by worker threads
   std::vector<std::shared_ptr<TensorOp>> tfuncs_;
 
   // Variable to store the column name that the tensorOps are consuming
@@ -209,13 +215,6 @@ class MapOp : public ParallelOp {
   // Indices of the columns to process.
   std::vector<size_t> to_process_indices_;
 
-  // Performance mode is when the main thread creates local queues, pulls databuffers from the previous
-  // op's Connector and distributes them to the local queues. Workers pull from the local queues.
-  // If this flag is false, each worker pulls directly from the Connector. This use less resources
-  // (thread and memory), but when the computation cost is heavy (e.g. DecodeOp) and fluctuating, it can
-  // cause additional blocking because pop calls to Connector from the threads are synchronized to enforce the order.
-  bool perf_mode_;
-
   // Private function for worker/thread to loop continuously. It comprises the main
   // logic of MapOp: getting the data from previous Op, validating user specified column names,
   // applying a list of TensorOps to each of the data, process the results and then
@@ -224,25 +223,12 @@ class MapOp : public ParallelOp {
   // @return Status The error code return
   Status WorkerEntry(int32_t worker_id) override;  //  In: workerId assigned by tree_
 
-  // Private helper function for getting the next buffer
-  // When PerformanceMode is enabled, workers pop from the local queue.
-  // Otherwise, workers pop from the first child output Connector.
-  // @param p_buffer - the buffer to return
-  // @return Status return code
-  Status FetchNextBuffer(std::unique_ptr<DataBuffer> *p_buffer, int32_t worker_id) {
-    if (perf_mode_) {
-      RETURN_IF_NOT_OK(local_queues_[worker_id]->PopFront(p_buffer));
-    } else {
-      RETURN_IF_NOT_OK(child_[0]->GetNextBuffer(p_buffer, worker_id));
-    }
-    return Status::OK();
-  }
-
   // Private function for worker thread to perform TensorOp's compute function and get the result.
   // @param in_buffer A raw pointer to the DataBuffer. A raw pointer is fine because this function doesn't manage memory
   //     and is not shared with other threads.
   // @param[out] new_tensor_table A new Tensor Table to be populated in this function.
-  Status WorkerCompute(DataBuffer *in_buffer, TensorQTable *new_tensor_table);
+  Status WorkerCompute(DataBuffer *in_buffer, TensorQTable *new_tensor_table,
+                       const std::vector<std::shared_ptr<MapJob>> &job_list);
 
   // Private function that create the final column name to index mapping and
   // get indices of the columns this mapop does not use.

mindspore/ccsrc/minddata/dataset/engine/opt/optional/tensor_op_fusion_pass.cc

@@ -17,7 +17,7 @@
 #include <memory>
 #include "minddata/dataset/engine/opt/optional/tensor_op_fusion_pass.h"
 #include "minddata/dataset/kernels/image/decode_op.h"
-#include "minddata/dataset/engine/datasetops/map_op.h"
+#include "minddata/dataset/engine/datasetops/map_op/map_op.h"
 #include "minddata/dataset/kernels/image/random_crop_decode_resize_op.h"
 
 namespace mindspore {

mindspore/ccsrc/minddata/dataset/engine/opt/pass.cc

@@ -24,7 +24,7 @@
 #include "minddata/dataset/engine/datasetops/dataset_op.h"
 #include "minddata/dataset/engine/datasetops/device_queue_op.h"
 #include "minddata/dataset/engine/datasetops/epoch_ctrl_op.h"
-#include "minddata/dataset/engine/datasetops/map_op.h"
+#include "minddata/dataset/engine/datasetops/map_op/map_op.h"
 #include "minddata/dataset/engine/datasetops/project_op.h"
 #include "minddata/dataset/engine/datasetops/rename_op.h"
 #include "minddata/dataset/engine/datasetops/repeat_op.h"

tests/ut/cpp/dataset/map_op_test.cc

@@ -645,16 +645,14 @@ TEST_F(MindDataTestMapOp, ImageFolder_Decode_Repeat_Resize) {
   map_decode_builder.SetInColNames({"image"})
     .SetOutColNames({})
     .SetTensorFuncs(func_list)
-    .SetNumWorkers(14)
-    .SetPerformanceMode(false);
+    .SetNumWorkers(14);
   rc = map_decode_builder.Build(&map_decode_map);
   EXPECT_TRUE(rc.IsOk());
 
   map_resize_builder.SetInColNames({"image"})
     .SetOutColNames({})
     .SetTensorFuncs(func_list2)
-    .SetNumWorkers(15)
-    .SetPerformanceMode(false);
+    .SetNumWorkers(15);
   rc = map_resize_builder.Build(&map_resize_op);
   EXPECT_TRUE(rc.IsOk());
 
@@ -739,5 +737,3 @@ TEST_F(MindDataTestMapOp, ImageFolder_Decode_Repeat_Resize_NoInputColumns) {
   }
   EXPECT_TRUE(i == 88);
 }
-
-

tests/ut/cpp/dataset/rename_op_test.cc

@@ -19,7 +19,6 @@
 #include <string>
 #include "minddata/dataset/core/client.h"
 #include "minddata/dataset/core/constants.h"
-#include "minddata/dataset/engine/datasetops/map_op.h"
 #include "minddata/dataset/engine/datasetops/rename_op.h"
 #include "common/common.h"
 #include "common/utils.h"

tests/ut/cpp/dataset/zip_op_test.cc

@@ -23,7 +23,6 @@
 #include <thread>
 #include "minddata/dataset/core/client.h"
 #include "minddata/dataset/core/constants.h"
-#include "minddata/dataset/engine/datasetops/map_op.h"
 #include "minddata/dataset/engine/datasetops/zip_op.h"
 #include "minddata/dataset/core/tensor.h"
 #include "minddata/dataset/core/config_manager.h"