support lod in batchTask

core/configure/proto/server_configure.proto

@@ -22,11 +22,12 @@ message EngineDesc {
   required string reloadable_type = 4;
   required string model_dir = 5;
   repeated int32 gpu_ids = 6;
-  required int32 runtime_thread_num = 7;
-  required int32 batch_infer_size = 8;
-  required int32 enable_batch_align = 9;
-  optional string version_file = 10;
-  optional string version_type = 11;
+  optional int32 runtime_thread_num = 7 [ default = 0 ];
+  optional int32 batch_infer_size = 8 [ default = 0 ];
+  optional bool enable_batch_align = 9 [ default = true ];
+  optional bool allow_split_request = 10 [ default = true ];
+  optional string version_file = 11;
+  optional string version_type = 12;
 
   /*
    * Sparse Parameter Service type. Valid types are:
@@ -39,17 +40,17 @@ message EngineDesc {
     LOCAL = 1;
     REMOTE = 2;
   }
-  optional SparseParamServiceType sparse_param_service_type = 12;
-  optional string sparse_param_service_table_name = 13;
-  optional bool enable_memory_optimization = 14;
-  optional bool enable_ir_optimization = 15;
-  optional bool use_trt = 16;
-  optional bool use_lite = 17;
-  optional bool use_xpu = 18;
-  optional bool use_gpu = 19;
-  optional bool combined_model = 20;
-  optional bool encrypted_model = 21;
-  optional bool gpu_multi_stream = 22;
+  optional SparseParamServiceType sparse_param_service_type = 13;
+  optional string sparse_param_service_table_name = 14;
+  optional bool enable_memory_optimization = 15;
+  optional bool enable_ir_optimization = 16;
+  optional bool use_trt = 17;
+  optional bool use_lite = 18;
+  optional bool use_xpu = 19;
+  optional bool use_gpu = 20;
+  optional bool combined_model = 21;
+  optional bool encrypted_model = 22;
+  optional bool gpu_multi_stream = 23;
 };
 
 // model_toolkit conf

core/predictor/framework/bsf-inl.h

@@ -26,9 +26,90 @@
 #include "core/predictor/common/inner_common.h"
 #include "core/predictor/framework/memory.h"
 
+// this file is included by bsf.h
 namespace im {
 namespace bsf {
 
+template <typename InItemT, typename OutItemT>
+bool Task<InItemT, OutItemT>::task_fetch_init(BatchTasks<TaskT>& baskTask) {
+  // 双检锁，减少加锁的粒度
+  if (!fetch_init) {
+    if (taskmeta_num > 1) {
+      // 对于task被拆分为多个taskmeta,需要加锁。
+      AutoMutex lock(task_mut);
+      task_fetch_create(baskTask);
+    } else {
+      // 对于task只有1个taskmeta,不需要加锁。
+      task_fetch_create(baskTask);
+    }
+  }
+  return true;
+}
+
+template <typename InItemT, typename OutItemT>
+bool Task<InItemT, OutItemT>::task_fetch_create(BatchTasks<TaskT>& baskTask) {
+  if (!fetch_init) {
+    vector_fetch_lod_index = baskTask.vector_fetch_lod_index;
+    set_fetch_nobatch_index = baskTask.set_fetch_nobatch_index;
+    OutVectorT taskMetaOutLodTensor;
+    size_t fetchvar_num = baskTask._batch_out.size();
+    for (size_t fetchvar_index = 0; fetchvar_index < fetchvar_num;
+         ++fetchvar_index) {
+      size_t fetchvar_bytesize_index =
+          baskTask.fetchvar_bytesize(fetchvar_index);
+      size_t fetchvar_batch = 0;
+      // 1. nobatch fetchvar情况
+      if (set_fetch_nobatch_index.size() > 0 &&
+          set_fetch_nobatch_index.find(fetchvar_index) !=
+              set_fetch_nobatch_index.end()) {
+        fetchvar_batch = 1;
+      } else if (vector_fetch_lod_index.size() > 0 &&
+                 std::find(vector_fetch_lod_index.begin(),
+                           vector_fetch_lod_index.end(),
+                           fetchvar_index) != vector_fetch_lod_index.end()) {
+        // lod fetchvar情况，此时无法确定总的shape[0]
+        // 根据task中的task_num总数开辟task_num个临时空间
+        // 每个lod型的fetchvar拷贝到对应的临时空间中
+        // 最后再计算临时空间的总量，合并fetchvar和lod
+        fetchvar_batch = 0;
+
+      } else {
+        // 普通fetchvar情况，此时该Task总的fetchvar_batch =
+        // 输入的总的batch_size()
+        fetchvar_batch = batch_size();
+      }
+      paddle::PaddleTensor tensor_out;
+      tensor_out.name = baskTask._batch_out[fetchvar_index].name;
+      tensor_out.dtype =
+          paddle::PaddleDType(baskTask._batch_out[fetchvar_index].dtype);
+      tensor_out.shape = baskTask._batch_out[fetchvar_index].shape;
+      tensor_out.shape[0] = fetchvar_batch;
+      if (fetchvar_batch != 0) {
+        // 此时 lod 为空。
+        tensor_out.lod = baskTask._batch_out[fetchvar_index].lod;
+        // resize all batch memory at one time
+        size_t databuf_size = fetchvar_batch * fetchvar_bytesize_index;
+        tensor_out.data.Resize(databuf_size);
+      } else {
+        // 当taskmeta_num = 1时，由于同时只有一个taskMeta操作task
+        // 不涉及线程安全问题，所以此时可以直接由taskMeta->task->resize->copy
+
+        // 当task被分为多个taskMeta时，需要临时对象记录
+        // 收齐后再一起合并
+        if (taskmeta_num > 1) {
+          taskMetaOutLodTensor.push_back(tensor_out);
+        }
+      }
+      outVectorT_ptr->push_back(tensor_out);
+    }
+    // outLodTensorVector实际是一个双层vector
+    // shape为taskmeta_num * vector_fetch_lod_index.size();
+    outLodTensorVector.resize(taskmeta_num, taskMetaOutLodTensor);
+    fetch_init = true;
+  }
+  return true;
+}
+
 template <typename TaskT>
 void* TaskExecutor<TaskT>::thread_entry(void* args) {
   ThreadContext<TaskT>* context = static_cast<ThreadContext<TaskT>*>(args);
@@ -156,9 +237,11 @@ TaskHandler<TaskT> TaskExecutor<TaskT>::schedule(
 
   task->inVectorT_ptr = (const InVectorT*)inVectorT_ptr;
   task->outVectorT_ptr = (OutVectorT*)outVectorT_ptr;
+  if (!task->task_init()) {
+    LOG(ERROR) << "task->init() failed";
+  }
   task->rem = task->batch_size();
   task->index.store(0, butil::memory_order_relaxed);
-
   AutoMutex lock(_mut);
   _task_queue.push_back(task);
   THREAD_COND_SIGNAL(&_cond);
@@ -168,11 +251,12 @@ TaskHandler<TaskT> TaskExecutor<TaskT>::schedule(
 
 // this function is accessed by multi thread.
 // so AutoMutex at first.
-// so batch.append_task is thread safe.
+// so batchTask.append_task is thread safe.
 // you dont need to add extra lock in append_task()
+// task is already init.
 template <typename TaskT>
 bool TaskExecutor<TaskT>::move_task_to_batch(
-    BatchTasks<TaskT>& batch) {  // NOLINT
+    BatchTasks<TaskT>& batchTask) {  // NOLINT
   AutoMutex lock(_mut);
   while (_task_queue.empty()) {
     THREAD_COND_WAIT(&_cond, &_mut);
@@ -183,9 +267,45 @@ bool TaskExecutor<TaskT>::move_task_to_batch(
     return false;
   }
 
+  TaskT* previous_task = nullptr;
   while (!_task_queue.empty()) {
     TaskT* task = _task_queue.front();
-    size_t rem = batch.append_task(task);
+
+    // 由于无法确定fetchVar是否为lod，故单个task不能拆分放到多个batchTask中，否则后续组装很难完成。
+    // 所以，task不能被拆分，即用户的请求可以合并一起预测，但不能拆分两个小部分去预测。
+    // 难点：预测前，能够知道被拆成了几个taskmeta,但只有预测后，才知道有多少个fetchvar,多少个lod的fetchvar
+    // 所以，task中想要创建taskmeta_num* lod的fetchvar num* PaddleBuf（以及Lod）
+    // 只能在notify_task中，用taskmeta->task去创建,需要在task中加锁。
+    // 原子操作不可行，因为多个线程必须等待创建好上述的PaddleBuf后才能继续。
+
+    // 对于普通的fetch，也需要加锁去创建PaddleTensor，后续才能往里拷贝。
+
+    // _batch_align为false时，即使空间小，也会全放入一个完整的Task，允许临时超限。
+    // _allow_split_request == false，则每个task不会被拆分。
+    // 默认为true，允许拆分task从而使得空间利用率最大。
+    if (!batchTask.get_allow_split_request()) {
+      if (task->batch_size() > batchTask.get_rem_size() &&
+          batchTask.get_batch_align()) {
+        break;
+      }
+    }
+
+    // combine_task_valid负责判断是否能够合并
+    // 除最外层的shape外，内层shape应一致才能合并。
+    // 否则跳出循环,放入下一个batchTask中。
+    // 以此保证batch.append_task(task)中的task的内层shape相同。
+
+    // 对于Shape[0] = 1 而!=batch的情况，因为合并时，取其中一个的值
+    // 所以要求该feedvar必须相等，才能合并。
+    // 否则跳出循环,放入下一个batchTask中。
+    // 目前没有PaddleTensor和PaddleBuff没有重载==，所以只能比较内存.
+    if (previous_task != nullptr) {
+      if (!task->combine_task_valid(previous_task)) {
+        break;
+      }
+    }
+    size_t rem = batchTask.append_task(task);
+    previous_task = task;
     if (task->rem <= 0) {
       _task_queue.pop_front();
     }
@@ -201,11 +321,12 @@ bool TaskExecutor<TaskT>::move_task_to_batch(
 // TaskT is from the SingleTon TaskExecutor`s _task_queue
 // although TaskMeta is a local variable, but several TaskMeta may points to
 // the same TaskT which is get from the SingleTon TaskExecutor`s _task_queue.
-// put TaskMeta to the local variable BatchTasks<TaskT> batch.
+// put TaskMeta to the local variable BatchTasks<TaskT> batchTask.
 
-// batch.merge_tasks() and batch.notify_tasks() has no lock.
-// BatchTasks<TaskT> batch itself is a local variable, it`s thread safe.
-// If batch.merge_tasks() and batch.notify_tasks() do something to TaskMeta
+// batchTask.merge_tasks() and batchTask.notify_tasks() has no lock.
+// BatchTasks<TaskT> batchTask itself is a local variable, it`s thread safe.
+// If batchTask.merge_tasks() and batchTask.notify_tasks() do something to
+// TaskMeta
 // you need to pay attention to that.
 // Multi-Thread deal with different TaskMeta(cause it`s created as local
 // variable)
@@ -242,11 +363,24 @@ int TaskExecutor<TaskT>::work(ThreadContext<TaskT>* context) {
       return -1;
     }
 
-    BatchTasks<TaskT> batch(_batch_size, _batch_align);
-    if (move_task_to_batch(batch)) {
-      batch.merge_tasks();
-      _fn(&batch.in(), &batch.out());
-      batch.notify_tasks();
+    // move_task_to_batch() take the original task from the `_task_queue`
+    // put the original task into its own Vector<taskmeta>
+    // the capacity of its own Vector<taskmeta> is decided by `_batch_size` or
+    // `_batch_align`
+
+    // merge_tasks() move the imput-data into `_batch_in` from its own
+    // Vector<taskmeta>.
+    // because the predictor`s input is the `_batch_in`
+
+    // notify_tasks() move the output-data into every single taskmeta from
+    // `_batch_out`.
+    // because the predictor`s output is the `_batch_out`
+    BatchTasks<TaskT> batchTask(
+        _batch_size, _batch_align, _allow_split_request);
+    if (move_task_to_batch(batchTask)) {
+      batchTask.merge_tasks();
+      _fn(&batchTask.in(), &batchTask.out());
+      batchTask.notify_tasks();
     }
   }
 

core/predictor/framework/bsf.h

@@ -16,7 +16,9 @@
 
 #include <errno.h>
 #include <algorithm>
+#include <cstring>
 #include <list>
+#include <set>
 #include <vector>
 
 #ifdef BCLOUD
@@ -46,7 +48,8 @@ static const size_t DEFAULT_BATCH_SIZE = 100;
 // `rem` don`t need to be atomic, cause the operation `put` is synchronous.
 // actually, the reason is that lock have been added outside the operation
 // `put`.
-
+template <typename TaskT>
+class BatchTasks;
 // size_t `index` records how many batch have been processing completed.
 // `index` need to be atomic, cause the operation 'notify' is asynchronous.
 template <typename InItemT, typename OutItemT>
@@ -56,7 +59,7 @@ struct Task {
   typedef InItemT InType;
   typedef OutItemT OutType;
   typedef Task<InItemT, OutItemT> TaskT;
-  typedef std::vector<int> ShapeVector;
+  typedef std::vector<size_t> ShapeVector;
   typedef std::vector<ShapeVector> VectorOfShapeVector;
 
   int read_fd;
@@ -65,7 +68,17 @@ struct Task {
   const InVectorT* inVectorT_ptr;
   OutVectorT* outVectorT_ptr;
   size_t rem;
+  size_t total_feed_batch;
+  std::set<size_t> set_feed_lod_index;
+  std::set<size_t> set_feed_nobatch_index;
+  std::vector<size_t> vector_fetch_lod_index;
+  std::set<size_t> set_fetch_nobatch_index;
   butil::atomic<size_t> index;
+  size_t taskmeta_num;
+  THREAD_MUTEX_T task_mut;
+  bool fetch_init;
+  // taskmeta_num * set_feed_lod_index.size()
+  std::vector<OutVectorT> outLodTensorVector;
 
   Task() {
     read_fd = -1;
@@ -73,11 +86,24 @@ struct Task {
     owner_tid = -1;
     inVectorT_ptr = NULL;
     outVectorT_ptr = NULL;
+    set_feed_lod_index.clear();
+    set_feed_nobatch_index.clear();
+    vector_fetch_lod_index.clear();
+    set_fetch_nobatch_index.clear();
     rem = -1;
+    total_feed_batch = 0;
+    taskmeta_num = 0;
     index.store(0, butil::memory_order_relaxed);
+    THREAD_MUTEX_INIT(&task_mut, NULL);
+    fetch_init = false;
+    outLodTensorVector.clear();
+  }
+  ~Task() {
+    THREAD_MUTEX_DESTROY(&task_mut);
+    outLodTensorVector.clear();
   }
 
-  bool check_feedvar_valid(int feedvar_index) {
+  bool check_feedvar_valid(size_t feedvar_index) {
     if (feedvar_index < 0 || inVectorT_ptr->size() <= feedvar_index) {
       LOG(ERROR) << "feedvar doesnt exsit or feedvar_index error";
       return 0;
@@ -91,20 +117,47 @@ struct Task {
     return 1;
   }
 
-  // Now, it simply assume that the first dimension of data is batch.
-  // so the batch is PaddleTensor.shape[0]
+  bool combine_task_valid(Task* other_task) {
+    // TODO(HexToString): auto-padding
+    // 除最外层的shape外，内层shape应一致才能合并。
+    // 否则跳出循环,放入下一个batchTask中。
+    // 以此保证batch.append_task(task)中的task的内层shape相同。
+    if (other_task->feedvar_shape_nobatch() != feedvar_shape_nobatch()) {
+      return false;
+    }
+
+    // 对于Shape[0] = 1 而!=batch的情况，因为合并时，取其中一个的值
+    // 所以要求该feedvar必须相等，才能合并。
+    // 目前没有PaddleTensor和PaddleBuff没有重载==，所以只能比较内存.
+    for (size_t feedvar_index = 0;
+         feedvar_index < set_feed_nobatch_index.size();
+         ++feedvar_index) {
+      int result =
+          std::memcmp((*inVectorT_ptr)[feedvar_index].data.data(),
+                      (*(other_task->inVectorT_ptr))[feedvar_index].data.data(),
+                      (*inVectorT_ptr)[feedvar_index].data.length());
+      if (result != 0) return false;
+    }
+    return true;
+  }
 
-  // If batch information is added into feedvar.prototxt.
-  // we can get the information from the feedvar.prototxt instead of assume.
-  size_t feedvar_batch_size(int feedvar_index) {
+  size_t feedvar_batch_size(size_t feedvar_index) {
     if (!check_feedvar_valid(feedvar_index)) {
       return 0;
     }
-
+    // if lod, 'lod[0].size()-1' is batch.
+    // for PaddleTensor lod is vector<vector<size_t>>, so lod[0] is real lod.
+    // for example, lod = [0,3,4,6], shape = [6,340,340], batch is 3 actually.
+    // for lod, the batch < shape[0].
+    if ((*inVectorT_ptr)[feedvar_index].lod.size() > 0 &&
+        (*inVectorT_ptr)[feedvar_index].lod[0].size() > 0) {
+      return (*inVectorT_ptr)[feedvar_index].lod[0].size() - 1;
+    }
+    // if not lod, the first dimension of data `PaddleTensor.shape[0]` is batch.
     return (*inVectorT_ptr)[feedvar_index].shape[0];
   }
 
-  size_t feedvar_element_bytesize(int feedvar_index) {
+  size_t feedvar_element_bytesize(size_t feedvar_index) {
     if (!check_feedvar_valid(feedvar_index)) {
       return 0;
     }
@@ -126,7 +179,7 @@ struct Task {
 
   // Now, the implementation of this function is based on assumption
   // that shape [0] = batch_size.
-  size_t feedvar_element_num(int feedvar_index) {
+  size_t feedvar_element_num(size_t feedvar_index) {
     if (!check_feedvar_valid(feedvar_index)) {
       return 0;
     }
@@ -138,18 +191,18 @@ struct Task {
       return 1;
     }
     // start from shape[1], cause shape[0] = batch_size.
-    for (int i = 1; i < (*inVectorT_ptr)[feedvar_index].shape.size(); ++i) {
+    for (size_t i = 1; i < (*inVectorT_ptr)[feedvar_index].shape.size(); ++i) {
       element_num *= (*inVectorT_ptr)[feedvar_index].shape[i];
     }
     return element_num;
   }
 
-  size_t feedvar_bytesize(int feedvar_index) {
+  size_t feedvar_bytesize(size_t feedvar_index) {
     return feedvar_element_num(feedvar_index) *
            feedvar_element_bytesize(feedvar_index);
   }
 
-  ShapeVector feedvar_shape_nobatch(int feedvar_index) {
+  ShapeVector feedvar_shape_nobatch(size_t feedvar_index) {
     if (!check_feedvar_valid(feedvar_index)) {
       return ShapeVector();
     }
@@ -158,40 +211,165 @@ struct Task {
   }
 
   VectorOfShapeVector feedvar_shape_nobatch() {
-    VectorOfShapeVector vector_of_feedvar_shape_nobatch(inVectorT_ptr->size());
-    for (int index = 0; index < inVectorT_ptr->size(); ++index) {
-      vector_of_feedvar_shape_nobatch.push_back(feedvar_shape_nobatch(index));
+    VectorOfShapeVector vector_of_feedvar_shape_nobatch;
+    for (size_t feedvar_index = 0; feedvar_index < inVectorT_ptr->size();
+         ++feedvar_index) {
+      vector_of_feedvar_shape_nobatch.push_back(
+          feedvar_shape_nobatch(feedvar_index));
     }
     return vector_of_feedvar_shape_nobatch;
   }
 
-  // At present, it is considered that the batch of all feedvar is consistent.
-  // so for each feedvar, PaddleTensor.shape[0] should be the same.
-  bool check_batch_align() {
-    int batch_size_align = feedvar_batch_size(0);
-    for (int feedvar_index = 0; feedvar_index < inVectorT_ptr->size();
+  // For each feedvar, batch should be 1 or batch_size.
+  // if feedvar-1: batch_size = 1 (always not batch).
+  // feedvar-2: batch_size = n,  batch = n.
+  // this function is not thread safe. only called when task is creating.
+  bool task_init() {
+    total_feed_batch = feedvar_batch_size(0);
+    // which means error.
+    if (total_feed_batch <= 0) return false;
+
+    for (size_t feedvar_index = 0; feedvar_index < inVectorT_ptr->size();
          ++feedvar_index) {
-      if (feedvar_batch_size(feedvar_index) != batch_size_align) {
-        return 0;
+      // TODO(HexToString): Distinguish between nobatch and batch =
+      // 1(By:HexToString)
+      // 当数据中feedvar-1: 带batch,且batch =1，shape[0] = 1
+      // feedvar-2:不带batch，由于不带batch导致shape[0] =1
+      // 此时，无法分辨是否是天然nobatch，此时set_feed_nobatch_index会漏掉
+      // 后续希望在其他地方能够区分两者。
+      if (feedvar_batch_size(feedvar_index) != total_feed_batch) {
+        // which means error.
+        if (feedvar_batch_size(feedvar_index) != 1 && total_feed_batch != 1) {
+          return false;
+        } else {
+          // which means feedvar shape[0] = 1.
+          // shape[0] does not change with batch
+          set_feed_nobatch_index.insert(feedvar_index);
+          total_feed_batch =
+              std::max(feedvar_batch_size(feedvar_index), total_feed_batch);
+        }
+      }
+      // 将lod feedvar index加入到vector中。
+      if ((*inVectorT_ptr)[feedvar_index].lod.size() > 0 &&
+          (*inVectorT_ptr)[feedvar_index].lod[0].size() > 0) {
+        set_feed_lod_index.insert(feedvar_index);
       }
     }
-    /*
-    for(int fetchvar_index = 0; fetchvar_index < outVectorT_ptr->size();
-    ++fetchvar_index) {
-      if(fetchvar_batch_size(fetchvar_index) != batch_size_align) {
-        return 0;
+    return true;
   }
+
+  size_t batch_size() { return total_feed_batch; }
+
+  // start_batch range is 0~batch_size, end_batch range is 1~batch_size
+  // start_batch should not be included, end_batch > start_batch
+  // return is (start_batch, end_batch] = [start_batch+1,end_batch]
+  // for not lod, shape0_index = [(start_batch+1)-1,end_batch-1] =
+  // [start_batch,end_batch-1] = [start_batch,end_batch)
+  // for lod, shape0_index = [lod[start_batch],lod[end_batch]-1] =
+  // [lod[start_batch],lod[end_batch])
+  // for nobatch, shape0_index = [0,1)
+  // 对于调用者，拿到shape0_index后，for(size_t myindex =shape0_index[0];
+  // myindex <shape0_index[1];myindex++)即可.
+
+  // 原始lod= [0,3,4,6] 取的batch为(start_batch = 1,end_batch =
+  // 3]，即取batch=2,3.
+  // 此时lod=[3,4,6]，处理后得到[1,3]
+  // 这样处理后，合并lod比较方便，直接加上上一个lod的结尾的值即可。
+  std::vector<std::vector<size_t>> get_feature_by_batch(size_t feedvar_index,
+                                                        size_t start_batch,
+                                                        size_t end_batch) {
+    std::vector<std::vector<size_t>> feature_vector;
+    // feature_vector是双层vector,这么设计是由于一个遍历即可处理所有的特征。
+    // feature_vector[0]是由shape0_index的范围值组成的vector,包含两个元素最小和最大值。
+    // feature_vector[1]是由lod组成的vector，包含指定batch的lod信息.
+    // feature_vector[2]是由单个元素的组成的vector，元素值为1表示是nobatch的feedvar。
+
+    // if 为 nobatch feedvar情况。
+    // else if 为带lod的feedvar情况。
+    // else为不带lod 普通feedvar情况。
+    if (set_feed_nobatch_index.size() > 0 &&
+        set_feed_nobatch_index.find(feedvar_index) !=
+            set_feed_nobatch_index.end()) {
+      feature_vector = {{0, 1}, {}, {1}};
+    } else if (set_feed_lod_index.size() > 0 &&
+               set_feed_lod_index.find(feedvar_index) !=
+                   set_feed_lod_index.end()) {
+      std::vector<size_t> feed_lod_vector(end_batch - start_batch);
+      for (size_t lod_index = start_batch + 1, vector_index = 0;
+           lod_index < end_batch + 1;
+           ++lod_index, ++vector_index) {
+        feed_lod_vector[vector_index] =
+            (*inVectorT_ptr)[feedvar_index].lod[0][lod_index] -
+            (*inVectorT_ptr)[feedvar_index].lod[0][start_batch];
+      }
+      size_t shape0_start = (*inVectorT_ptr)[feedvar_index].lod[0][start_batch];
+      size_t shape0_end = (*inVectorT_ptr)[feedvar_index].lod[0][end_batch];
+      feature_vector = {{shape0_start, shape0_end}, feed_lod_vector};
+      // feature_vector.push_back(feed_lod_vector);
+    } else {
+      feature_vector = {{start_batch, end_batch}};
+    }
+    return feature_vector;
+  }
+
+  bool combine_taskmeta() {
+    // 只有含有lod类型的fetch输出，且task被拆分为多个taskmeta的情况
+    // 才需要将数据从outLodTensorVector搬运到outVectorT_ptr
+    if (vector_fetch_lod_index.size() > 0 && taskmeta_num > 1) {
+      for (size_t index = 0; index < vector_fetch_lod_index.size(); ++index) {
+        size_t data_length = 0;
+        size_t lod_length = 0;
+        size_t feedvar_index = vector_fetch_lod_index[index];
+        // 由于PaddleTensor的resize实现，是每次都会清空，所以必须先统计总长度。
+        for (size_t taskmeta_index = 0; taskmeta_index < taskmeta_num;
+             ++taskmeta_num) {
+          data_length +=
+              outLodTensorVector[taskmeta_index][index].data.length();
+          lod_length += outLodTensorVector[taskmeta_index][index].lod[0].size();
+        }
+        // 一次性扩容PaddleTensor中的data和lod
+        paddle::PaddleTensor& fetchVarTensor = (*outVectorT_ptr)[feedvar_index];
+        fetchVarTensor.data.Resize(data_length);
+        // task中的lod补0
+        if (fetchVarTensor.lod.size() <= 0) {
+          fetchVarTensor.lod.push_back({0});
+        } else if (fetchVarTensor.lod[0].size() <= 0) {
+          fetchVarTensor.lod[0].push_back(0);
+        }
+        fetchVarTensor.lod[0].resize(lod_length + 1, 0);
+
+        //
+        size_t data_length_offset = 0;
+        size_t lod_length_offset = 0;
+        size_t once_data_length = 0;
+        size_t once_lod_length = 0;
+        size_t last_lod_value = fetchVarTensor.lod[0][lod_length_offset];
+        for (size_t taskmeta_index = 0; taskmeta_index < taskmeta_num;
+             ++taskmeta_num) {
+          void* dst_ptr = fetchVarTensor.data.data() + data_length_offset;
+          void* source_ptr =
+              outLodTensorVector[taskmeta_index][index].data.data();
+          once_data_length =
+              outLodTensorVector[taskmeta_index][index].data.length();
+          memcpy(dst_ptr, source_ptr, once_data_length);
+          once_lod_length =
+              outLodTensorVector[taskmeta_index][index].lod[0].size();
+          for (size_t once_index = 0; once_index < once_lod_length;
+               ++once_index) {
+            fetchVarTensor.lod[0][lod_length_offset + 1] =
+                last_lod_value +
+                outLodTensorVector[taskmeta_index][index].lod[0][once_index];
           }
-    */
-    return 1;
+          data_length_offset += once_data_length;
+          lod_length_offset += once_lod_length;
         }
-
-  size_t batch_size() {
-    if (check_batch_align()) {
-      return feedvar_batch_size(0);
       }
-    return 0;
     }
+    return true;
+  }
+
+  bool task_fetch_init(BatchTasks<TaskT>& baskTask);
+  bool task_fetch_create(BatchTasks<TaskT>& baskTask);
 };
 
 // `Several Task` or `part of batch in Task` can be a TaskMeta.
@@ -206,61 +384,164 @@ struct Task {
 
 // TaskMeta is necessary.
 // cause we need know the the corresponding relationship between
-// `batch_out`(which is in BatchTasks) and `outVectorT_ptr`(which is in Task).
+// `_batch_out`(which is in BatchTasks) and `outVectorT_ptr`(which is in Task).
 // especially when 1 Task be divided into several TaskMeta and be put into
 // several different BatchTasks.
+
+// begin、add、end means batch, not shape[0].
+// if not lod, batch == shape[0]. if lod, batch != shape[0]
+// for example, lod = [0,3,4,6], shape = [6,340,340]
+// there is 3 batch actually, add = 3, but shape[0] = 6.
 template <typename TaskT>
 struct TaskMeta {
-  TaskMeta(TaskT* ptr, size_t start, size_t add)
-      : task(ptr), begin(start), end(start + add) {}
+  TaskMeta(TaskT* ptr, size_t start, size_t add, size_t taskmeta_index)
+      : task(ptr),
+        begin(start),
+        end(start + add),
+        taskmeta_index(taskmeta_index) {
+    feedvar_num = ptr->inVectorT_ptr->size();
+    for (size_t feedvar_index = 0; feedvar_index < feedvar_num;
+         ++feedvar_index) {
+      std::vector<std::vector<size_t>> feature =
+          ptr->get_feature_by_batch(feedvar_index, start, start + add);
+      feed_shape0_range.push_back(feature[0]);
+      feedvar_type.push_back(feature.size());
+      if (feature.size() == 1) {
+        feed_lod_vector.push_back({});
+      } else if (feature.size() == 2) {
+        feed_lod_vector.push_back(feature[1]);
+      } else {
+        feed_lod_vector.push_back({});
+      }
+    }
+  }
 
   TaskT* task;
   size_t begin;
   size_t end;
+  size_t feedvar_num;
+  size_t taskmeta_index;
+  std::vector<std::vector<size_t>> feed_shape0_range;
+  std::vector<std::vector<size_t>> feed_lod_vector;
+  std::vector<size_t> feedvar_type;
 };
 
 // each TaskT is already include batch in itself
 // BatchTasks need to combine several `small TaskMeta` into a new `big TaskT`.
 // The only difference between the `big TaskT` and `small TaskT` is that
-// the TaskT.inVectorT_ptr->[feedvar_index].shape[0]
-// which is actually batch_size is different.
+// the TaskT.inVectorT_ptr->[feedvar_index].shape[0] is different
+// `big TaskT`.inVectorT_ptr->[feedvar_index].shape[0] is actually batch_size .
 template <typename TaskT>
 class BatchTasks {
  public:
   typedef typename TaskT::InType InType;
   typedef typename TaskT::OutType OutType;
   typedef TaskMeta<TaskT> TaskMetaT;
+  typedef std::vector<size_t> ShapeVector;
+  typedef std::vector<ShapeVector> VectorOfShapeVector;
+  typedef std::vector<size_t> LodVector;
+  typedef std::vector<LodVector> PaddleTensorLod;
+  friend TaskT;
 
-  explicit BatchTasks(size_t batch_size, bool batch_align = true)
+  explicit BatchTasks(size_t batch_size,
+                      bool batch_align = true,
+                      bool allow_split_request = true)
       : _batch_size(batch_size),
         _rem_size(batch_size),
-        _batch_align(batch_align) {
+        _batch_align(batch_align),
+        _allow_split_request(allow_split_request) {
     _batch_in.clear();
     _batch_in_offset.clear();
+    _total_shape0_batch_in.clear();
+    _total_feed_batch = 0;
+    _batch_in_lod.clear();
+
     _batch_out.clear();
     _batch_out_offset.clear();
+    _total_fetch_batch = 0;
     _taskmeta_vector.clear();
+    set_fetch_nobatch_index.clear();
+    vector_fetch_lod_index.clear();
   }
 
   ~BatchTasks() {
     _batch_in.clear();
     _batch_in_offset.clear();
+    _total_shape0_batch_in.clear();
+    _total_feed_batch = 0;
+    _batch_in_lod.clear();
+
     _batch_out.clear();
     _batch_out_offset.clear();
+    _total_fetch_batch = 0;
     _taskmeta_vector.clear();
+    set_fetch_nobatch_index.clear();
+    vector_fetch_lod_index.clear();
   }
 
   // synchronized operation
   // because Upper level callers of this function have already locked.
+  // 能进到此函数的task都是同类task，在该函数之前已保证了这点。
   size_t append_task(TaskT* task) {
     size_t add = std::min(task->rem, _rem_size);
+    // when _batch_align == false, it means always take a whole task as TaskMeta
+    // we can temporary breakthrough the limit of BatchTask`s capacity
+    // BatchTask`s capacity is _batch_size or _rem_size
     if (!_batch_align) {
       add = task->rem;
     }
     int start_index = task->batch_size() - task->rem;
-    TaskMetaT tm(task, start_index, add);
+    TaskMetaT tm(task, start_index, add, task->taskmeta_num);
+    task->taskmeta_num += 1;
     _taskmeta_vector.push_back(tm);
-
+    if (_batch_in_offset.size() == 0) {
+      _batch_in_offset.resize(tm.feedvar_num, 0);
+    }
+    if (_total_shape0_batch_in.size() == 0) {
+      _total_shape0_batch_in.resize(tm.feedvar_num, 0);
+    }
+    if (_batch_in_lod.size() == 0) {
+      PaddleTensorLod null_lod;
+      _batch_in_lod.resize(tm.feedvar_num, null_lod);
+    }
+    _total_feed_batch += add;
+    for (size_t feedvar_index = 0; feedvar_index < tm.feedvar_num;
+         ++feedvar_index) {
+      if (tm.feedvar_type[feedvar_index] == 1) {
+        // 普通的非lod feedvar
+        // 累计计算shape0的累加值，为后面初始化PaddleTensor做准备。
+        _total_shape0_batch_in[feedvar_index] +=
+            tm.feed_shape0_range[feedvar_index][1] -
+            tm.feed_shape0_range[feedvar_index][0];
+      } else if (tm.feedvar_type[feedvar_index] == 2) {
+        // lod类型的feedvar
+        // 累计计算shape0的累加值，为后面初始化PaddleTensor做准备。
+        _total_shape0_batch_in[feedvar_index] +=
+            tm.feed_shape0_range[feedvar_index][1] -
+            tm.feed_shape0_range[feedvar_index][0];
+        // 在Lod最前面加0
+        if (_batch_in_lod[feedvar_index].size() <= 0) {
+          _batch_in_lod[feedvar_index].push_back({0});
+        } else if (_batch_in_lod[feedvar_index][0].size() <= 0) {
+          _batch_in_lod[feedvar_index][0].push_back(0);
+        }
+        // 将lod加上前一组lod的结尾最大值，组合Lod
+        size_t last_lod_value = _batch_in_lod[feedvar_index][0].back();
+        for (size_t lod_index = 0;
+             lod_index < tm.feed_lod_vector[feedvar_index].size();
+             ++lod_index) {
+          _batch_in_lod[feedvar_index][0].push_back(
+              last_lod_value + tm.feed_lod_vector[feedvar_index][lod_index]);
+        }
+      } else {
+        // tm.feedvar_type[feedvar_index] == 3
+        // nobatch类型的feedvar.
+        // 此时不累加，且值应为1
+        _total_shape0_batch_in[feedvar_index] =
+            tm.feed_shape0_range[feedvar_index][1] -
+            tm.feed_shape0_range[feedvar_index][0];
+      }
+    }
     task->rem -= add;
     _rem_size -= add;
     return _rem_size;
@@ -281,72 +562,56 @@ class BatchTasks {
   // cause maybe next time we don`t need to do the extra copy.
   // directly copy the every Task into the Predictor.
 
-  // lod is not supported.
-  // if lod is set, we should not allow to use the bsf task.
-
   // batch.merge_tasks() is thread-safe function
   // cause batch is a local variable and Task is just read, not written.
+
   void merge_tasks() {
     if (_taskmeta_vector.size() <= 0) {
       return;
     }
 
-    // Temporarily, the batch of each feedvar is consistent
-    // If not consistent, use feedvar_batch_size instead of task->batch_size().
-    int temp_batch = 0;
-    for (size_t ti = 0; ti < _taskmeta_vector.size(); ++ti) {
-      TaskMetaT& tm = _taskmeta_vector[ti];
-      temp_batch += tm.task->batch_size();
-    }
-    if (temp_batch > _batch_size) {
-      LOG(ERROR) << "_realNumber_batch_in >_batch_size, error.";
-      return;
-    }
-
-    int feedvar_num = _taskmeta_vector[0].task->inVectorT_ptr->size();
-    if (_batch_in_offset.size() == 0) {
-      _batch_in_offset.resize(feedvar_num, 0);
-      _realNumber_batch_in.resize(feedvar_num, temp_batch);
-    }
-
     for (size_t ti = 0; ti < _taskmeta_vector.size(); ++ti) {
       TaskMetaT& tm = _taskmeta_vector[ti];
 
-      for (int index = 0; index < feedvar_num; ++index) {
+      for (size_t feedvar_index = 0; feedvar_index < tm.feedvar_num;
+           ++feedvar_index) {
         const paddle::PaddleTensor& feedVarTensor =
-            (*tm.task->inVectorT_ptr)[index];
-        size_t feedvar_bytesize = tm.task->feedvar_bytesize(index);
+            (*tm.task->inVectorT_ptr)[feedvar_index];
+        size_t feedvar_bytesize = tm.task->feedvar_bytesize(feedvar_index);
 
         if (ti == 0) {
-          if (feedVarTensor.lod.size() > 0 && feedVarTensor.lod[0].size() > 0) {
-            LOG(ERROR) << "lod Tensor is not supported now.";
-            return;
-          }
+          // Create the entire tensor at once
           // for now, we assume that every task feedvar_bytesize is the same.
           // which means we dont support auto embedding.
           // but for different feedvar, it is different.
           paddle::PaddleTensor paddleTensor;
           paddleTensor.dtype = feedVarTensor.dtype;
           paddleTensor.name = feedVarTensor.name;
-          paddleTensor.lod = feedVarTensor.lod;
+          paddleTensor.lod = _batch_in_lod[feedvar_index];
           paddleTensor.shape = feedVarTensor.shape;
-          paddleTensor.shape[0] = _realNumber_batch_in[index];
+          paddleTensor.shape[0] = _total_shape0_batch_in[feedvar_index];
           paddleTensor.data.Resize(feedvar_bytesize *
-                                   _realNumber_batch_in[index]);
+                                   _total_shape0_batch_in[feedvar_index]);
           _batch_in.push_back(paddleTensor);
         }
 
-        void* dst_ptr = _batch_in[index].data.data() + _batch_in_offset[index];
+        void* dst_ptr = _batch_in[feedvar_index].data.data() +
+                        _batch_in_offset[feedvar_index];
         void* source_ptr =
-            feedVarTensor.data.data() + feedvar_bytesize * tm.begin;
-        size_t length = feedvar_bytesize * (tm.end - tm.begin);
+            feedVarTensor.data.data() +
+            feedvar_bytesize * tm.feed_shape0_range[feedvar_index][0];
+        size_t length =
+            feedvar_bytesize * (tm.feed_shape0_range[feedvar_index][1] -
+                                tm.feed_shape0_range[feedvar_index][0]);
         memcpy(dst_ptr, source_ptr, length);
-        _batch_in_offset[index] += length;
+        // nobatch类型的feedvar，不叠加.
+        if (tm.feedvar_type[feedvar_index] != 3)
+          _batch_in_offset[feedvar_index] += length;
       }
     }
   }
 
-  bool check_fetchvar_valid(int fetchvar_index) {
+  bool check_fetchvar_valid(size_t fetchvar_index) {
     if (fetchvar_index < 0 || _batch_out.size() <= fetchvar_index) {
       LOG(ERROR) << "fetchvar doesnt exsit or fetchvar_index error";
       return 0;
@@ -360,19 +625,11 @@ class BatchTasks {
     return 1;
   }
 
-  size_t fetchvar_batch_size(int fetchvar_index) {
+  size_t fetchvar_element_bytesize(size_t fetchvar_index) {
     if (!check_fetchvar_valid(fetchvar_index)) {
       return 0;
     }
-
-    return _batch_out[fetchvar_index].shape[0];
-  }
-
-  size_t fetchvar_element_bytesize(int fetchvar_index) {
-    if (!check_fetchvar_valid(fetchvar_index)) {
-      return 0;
-    }
-    int dtype = _batch_out[fetchvar_index].dtype;
+    size_t dtype = _batch_out[fetchvar_index].dtype;
     if (dtype == paddle::PaddleDType::INT64) {
       return sizeof(int64_t);
     }
@@ -390,7 +647,7 @@ class BatchTasks {
 
   // Now, the implementation of this function is based on assumption
   // that shape [0] = batch_size.
-  size_t fetchvar_element_num(int fetchvar_index) {
+  size_t fetchvar_element_num(size_t fetchvar_index) {
     if (!check_fetchvar_valid(fetchvar_index)) {
       return 0;
     }
@@ -400,35 +657,66 @@ class BatchTasks {
       return 1;
     }
     // start from shape[1], cause shape[0] = batch_size.
-    for (int i = 1; i < _batch_out[fetchvar_index].shape.size(); ++i) {
+    for (size_t i = 1; i < _batch_out[fetchvar_index].shape.size(); ++i) {
       element_num *= _batch_out[fetchvar_index].shape[i];
     }
     return element_num;
   }
 
-  size_t fetchvar_bytesize(int fetchvar_index) {
+  size_t fetchvar_bytesize(size_t fetchvar_index) {
     return fetchvar_element_num(fetchvar_index) *
            fetchvar_element_bytesize(fetchvar_index);
   }
 
-  bool check_fetchvar_batch_align() {
-    int batch_size_align = fetchvar_batch_size(0);
-
-    for (int fetchvar_index = 0; fetchvar_index < _batch_out.size();
-         ++fetchvar_index) {
-      if (fetchvar_batch_size(fetchvar_index) != batch_size_align) {
+  size_t fetchvar_batch_size(size_t fetchvar_index) {
+    if (!check_fetchvar_valid(fetchvar_index)) {
       return 0;
     }
+    // if lod, 'lod[0].size()-1' is batch.
+    // for PaddleTensor lod is vector<vector<size_t>>, so lod[0] is real lod.
+    // for example, lod = [0,3,4,6], shape = [6,340,340], batch is 3 actually.
+    // for lod, the batch < shape[0].
+    if (_batch_out[fetchvar_index].lod.size() > 0 &&
+        _batch_out[fetchvar_index].lod[0].size() > 0) {
+      return _batch_out[fetchvar_index].lod[0].size() - 1;
     }
-
-    return 1;
+    // if not lod, the first dimension of data `PaddleTensor.shape[0]` is batch.
+    return _batch_out[fetchvar_index].shape[0];
   }
 
-  size_t fetchvar_batch_size() {
-    if (check_fetchvar_batch_align()) {
-      return fetchvar_batch_size(0);
+  size_t fetchvar_batch_size() { return _total_fetch_batch; }
+
+  bool deal_batch_out() {
+    _total_fetch_batch = fetchvar_batch_size(0);
+    if (_total_fetch_batch <= 0) return false;
+    for (size_t fetchvar_index = 0; fetchvar_index < _batch_out.size();
+         ++fetchvar_index) {
+      // TODO(HexToString): Distinguish between nobatch and batch =
+      // 1(By:HexToString)
+      // 当数据中fetchvar-1: 带batch,且batch =1，shape[0] = 1
+      // fetchvar-2:不带batch，由于不带batch导致shape[0] =1
+      // 此时，无法分辨是否是天然nobatch，此时set_fetch_nobatch_index会漏掉
+      // 后续希望在其他地方能够区分两者。
+      if (fetchvar_batch_size(fetchvar_index) != _total_fetch_batch) {
+        // which means error.
+        if (fetchvar_batch_size(fetchvar_index) != 1 &&
+            _total_fetch_batch != 1) {
+          return false;
+        } else {
+          // which means fetchvar shape[0] = 1.
+          // shape[0] does not change with batch
+          set_fetch_nobatch_index.insert(fetchvar_index);
+          _total_fetch_batch =
+              std::max(fetchvar_batch_size(fetchvar_index), _total_fetch_batch);
+        }
+      }
+      // 将lod fetchvar index加入到vector中。
+      if (_batch_out[fetchvar_index].lod.size() > 0 &&
+          _batch_out[fetchvar_index].lod[0].size() > 0) {
+        vector_fetch_lod_index.push_back(fetchvar_index);
       }
-    return 0;
+    }
+    return true;
   }
 
   void notify_tasks() {
@@ -436,12 +724,16 @@ class BatchTasks {
       LOG(ERROR) << "_taskmeta_vector.size() <=0, error.";
       return;
     }
-    if (_realNumber_batch_in[0] != fetchvar_batch_size()) {
+    // 根据_batch_out，求出输出的整体batch
+    // 并将lod类型和nobatch类型的fetchvar的index记录到set中，方便后续查看。
+    deal_batch_out();
+    // 若输出的batch不是1，且不与输入batch对应，则错误
+    if (_total_feed_batch != _total_fetch_batch && _total_fetch_batch != 1) {
       LOG(ERROR) << "_batch_out`s batch != _batch_in`s batch, error.";
       return;
     }
 
-    int fetchvar_num = _batch_out.size();
+    size_t fetchvar_num = _batch_out.size();
     if (_batch_out_offset.size() == 0) {
       _batch_out_offset.resize(fetchvar_num, 0);
     }
@@ -451,44 +743,130 @@ class BatchTasks {
       size_t begin = _taskmeta_vector[ti].begin;
       size_t end = _taskmeta_vector[ti].end;
       size_t add = end - begin;
+      size_t taskmeta_index = _taskmeta_vector[ti].taskmeta_index;
+      // 对task中的outVectorT_ptr进行初始化
+      // 如果是lod输出+多个taskmeta，此时对outLodTensorVector也需要初始化
+      if (!task->task_fetch_init(*this)) {
+        LOG(ERROR) << " task_fetch_init error.";
+        return;
+      }
+      size_t fetch_lod_index = 0;
 
-      for (int index = 0; index < fetchvar_num; ++index) {
-        // the task->outVectorT_ptr is null before core->run().
-        // first time we should copy from _batch_out
-        // so we need init.
-        size_t fetchvar_bytesize_index = fetchvar_bytesize(index);
-        if (task->outVectorT_ptr->size() <= index) {
-          paddle::PaddleTensor tensor_out;
-          tensor_out.name = _batch_out[index].name;
-          tensor_out.dtype = paddle::PaddleDType(_batch_out[index].dtype);
-          tensor_out.shape = _batch_out[index].shape;
-          tensor_out.shape[0] = task->batch_size();
-          tensor_out.lod = _batch_out[index].lod;
-          // resize all batch memory at one time
-          size_t databuf_size = task->batch_size() * fetchvar_bytesize_index;
-          tensor_out.data.Resize(databuf_size);
-          task->outVectorT_ptr->push_back(tensor_out);
-        }
-
-        paddle::PaddleTensor& fetchVarTensor = (*task->outVectorT_ptr)[index];
+      for (size_t fetchvar_index = 0; fetchvar_index < fetchvar_num;
+           ++fetchvar_index) {
+        size_t fetchvar_bytesize_index = fetchvar_bytesize(fetchvar_index);
+
+        if (set_fetch_nobatch_index.size() > 0 &&
+            set_fetch_nobatch_index.find(fetchvar_index) !=
+                set_fetch_nobatch_index.end()) {
+          // nobatch fetchvar情况
+          // 无论输入是多少batch，该index的fetchvar始终就shape[0] = 1
+          paddle::PaddleTensor& fetchVarTensor =
+              (*task->outVectorT_ptr)[fetchvar_index];
+          void* dst_ptr = fetchVarTensor.data.data();
+          size_t length = fetchvar_bytesize_index * 1;
+          void* source_ptr = _batch_out[fetchvar_index].data.data();
+          memcpy(dst_ptr, source_ptr, length);
+        } else if (vector_fetch_lod_index.size() > 0 &&
+                   std::find(vector_fetch_lod_index.begin(),
+                             vector_fetch_lod_index.end(),
+                             fetchvar_index) != vector_fetch_lod_index.end()) {
+          // lod fetchvar情况，此时无法确定总的shape[0]
+          // 根据task中的task_num总数开辟task_num个临时空间
+          // 每个lod型的fetchvar拷贝到对应的临时空间中
+          // 最后再计算临时空间的总量，合并fetchvar和lod
+          size_t last_batch = _batch_out_offset[fetchvar_index];
+          size_t shape0_index_start =
+              _batch_out[fetchvar_index].lod[0][last_batch];
+          size_t shape0_index_end =
+              _batch_out[fetchvar_index].lod[0][last_batch + add];
+          size_t shape0_length = shape0_index_end - shape0_index_start;
+          // task被拆分为多个taskmeta时，不能直接拷入task->outVectorT_ptr
+          // 此时,先拷入task->outLodTensorVector[taskmeta_index]
+          // 当task所有的taskmeta都完成时，再按照顺序进行拷贝回task->outVectorT_ptr。
+          if (task->taskmeta_num > 1) {
+            paddle::PaddleTensor& fetchVarTensor =
+                task->outLodTensorVector[taskmeta_index][fetch_lod_index];
+            size_t length = fetchvar_bytesize_index * shape0_length;
+            fetchVarTensor.data.Resize(length);
+            void* dst_ptr = fetchVarTensor.data.data();
+            void* source_ptr = _batch_out[fetchvar_index].data.data() +
+                               shape0_index_start * fetchvar_bytesize_index;
+            memcpy(dst_ptr, source_ptr, length);
+            // 由于是拆分的各个lod，不要补0，在最后合并给Task中的outVectorT_ptr时再补。
+            if (fetchVarTensor.lod.size() <= 0) {
+              fetchVarTensor.lod.push_back({});
+            }
+            fetchVarTensor.lod[0].resize(add, 0);
+            size_t last_lod_value =
+                _batch_out[fetchvar_index].lod[0][last_batch];
+            for (size_t lod_index = last_batch + 1, my_index = 0;
+                 lod_index < last_batch + add + 1;
+                 ++lod_index, ++my_index) {
+              fetchVarTensor.lod[0][my_index] =
+                  (_batch_out[fetchvar_index].lod[0][lod_index] -
+                   last_lod_value);
+            }
+          } else {
+            // task未被拆分为多个taskmeta，故只有某个线程中的taskmeta会操作task不存在多线程竞争
+            // 此时resize后，直接写入task->outVectorT_ptr中即可。
+            paddle::PaddleTensor& fetchVarTensor =
+                (*task->outVectorT_ptr)[fetchvar_index];
+            size_t length = fetchvar_bytesize_index * shape0_length;
+            fetchVarTensor.data.Resize(length);
+            void* dst_ptr = fetchVarTensor.data.data();
+            void* source_ptr = _batch_out[fetchvar_index].data.data() +
+                               shape0_index_start * fetchvar_bytesize_index;
+            memcpy(dst_ptr, source_ptr, length);
 
+            // task中的lod补0
+            if (fetchVarTensor.lod.size() <= 0) {
+              fetchVarTensor.lod.push_back({0});
+            } else if (fetchVarTensor.lod[0].size() <= 0) {
+              fetchVarTensor.lod[0].push_back(0);
+            }
+            // 将合并的lod信息对应的batch，拆分到task中。
+            // 注意，此时需要去掉前面lod导致的前置积累。
+            // 例如: 合lod = [0,2,5;7,10]，是由两组batch=2的task合并后预测的。
+            // 此时拆分，第一组时，都减去0,得到[2,5]+(由于前面已经补了0了) =
+            // [0,2,5]
+            // 第二组，都需要减5,得到[2,5]，这样处理才对。
+            fetchVarTensor.lod[0].resize(add + 1, 0);
+            size_t last_lod_value =
+                _batch_out[fetchvar_index].lod[0][last_batch];
+            for (size_t lod_index = last_batch + 1, my_index = 1;
+                 lod_index < last_batch + add + 1;
+                 ++lod_index, ++my_index) {
+              fetchVarTensor.lod[0][my_index] =
+                  (_batch_out[fetchvar_index].lod[0][lod_index] -
+                   last_lod_value);
+            }
+          }
+          fetch_lod_index++;
+        } else {
+          // 普通fetchvar情况，此时该Task总的fetchvar_batch =
+          // 输入的总的batch_size()
+          // 输出的batch应与输入的batch对应相等。
+          paddle::PaddleTensor& fetchVarTensor =
+              (*task->outVectorT_ptr)[fetchvar_index];
           void* dst_ptr =
               fetchVarTensor.data.data() + fetchvar_bytesize_index * begin;
           size_t length = fetchvar_bytesize_index * add;
-        if (_batch_out_offset[index] + length >
-            fetchvar_batch_size() * fetchvar_bytesize(index)) {
-          LOG(ERROR) << "_batch_out is less than taskmeta, error.";
-          return;
-        }
           void* source_ptr =
-            _batch_out[index].data.data() + _batch_out_offset[index];
+              _batch_out[fetchvar_index].data.data() +
+              _batch_out_offset[fetchvar_index] * fetchvar_bytesize_index;
 
           memcpy(dst_ptr, source_ptr, length);
-        _batch_out_offset[index] += length;
+        }
+        _batch_out_offset[fetchvar_index] += add;
       }
 
+      // index是局部变量，fetch_add是原子操作，成功则返回原值。
+      // 只有最后一个taskmeta都完成后，该线程的index+add才能>task->batch_size()
+      // 故只有一个线程能进入if{}内.不会造成多线程竞争的问题。
       size_t index = task->index.fetch_add(add);
       if ((index + add) >= task->batch_size()) {
+        task->combine_taskmeta();
         char c = 0;
         while (write(task->write_fd, &c, 1) != 1 && errno == EINTR) {
         }
@@ -503,17 +881,32 @@ class BatchTasks {
 
   size_t task_size() { return _taskmeta_vector.size(); }
 
+  const size_t get_rem_size() { return _rem_size; }
+
+  bool get_batch_align() { return _batch_align; }
+
+  bool get_allow_split_request() { return _allow_split_request; }
+
  private:
   std::vector<TaskMetaT> _taskmeta_vector;
   typename TaskT::InVectorT _batch_in;
   std::vector<size_t> _batch_in_offset;
-  std::vector<size_t> _realNumber_batch_in;
+  std::vector<size_t> _total_shape0_batch_in;
+  size_t _total_feed_batch;
+  std::vector<PaddleTensorLod> _batch_in_lod;
+
   typename TaskT::OutVectorT _batch_out;
   std::vector<size_t> _batch_out_offset;
-  std::vector<size_t> _realNumber_batch_out;
+  // std::vector<size_t> _total_shape0_batch_out;
+  size_t _total_fetch_batch;
+  // std::vector<PaddleTensorLod>  _batch_out_lod;
+  std::set<size_t> set_fetch_nobatch_index;
+  std::vector<size_t> vector_fetch_lod_index;
+
   size_t _rem_size;
   size_t _batch_size;
   bool _batch_align;
+  bool _allow_split_request;
 };
 
 // BSF task handle
@@ -589,6 +982,8 @@ class TaskExecutor {
   typedef typename TaskT::OutVectorT OutVectorT;
   typedef std::vector<TaskT> TaskArrayT;
   typedef baidu::paddle_serving::predictor::MempoolWrapper MempoolWrapper;
+  typedef std::vector<size_t> ShapeVector;
+  typedef std::vector<ShapeVector> VectorOfShapeVector;
 
   TaskExecutor()
       : _stop(false),
@@ -617,7 +1012,11 @@ class TaskExecutor {
 
   void set_batch_size(size_t batch_size) { _batch_size = batch_size; }
 
-  void set_batch_align(size_t batch_align) { _batch_align = batch_align; }
+  void set_batch_align(bool batch_align) { _batch_align = batch_align; }
+
+  void set_allow_split_request(bool allow_split_request) {
+    _allow_split_request = allow_split_request;
+  }
 
   void set_thread_init_fn(boost::function<int(void*)> init_fn,
                           void** contexts = NULL) {
@@ -642,7 +1041,7 @@ class TaskExecutor {
 
   TaskHandler<TaskT> schedule(const void*, void*);
 
-  bool move_task_to_batch(BatchTasks<TaskT>& batch);  // NOLINT
+  bool move_task_to_batch(BatchTasks<TaskT>& batchTask);  // NOLINT
 
  private:
   TaskExecutor(TaskExecutor<TaskT> const& other) = delete;
@@ -670,6 +1069,7 @@ class TaskExecutor {
 
   size_t _batch_size;
   bool _batch_align;
+  bool _allow_split_request;
 
   boost::function<void(const void*, void*)> _fn;
 };
@@ -687,12 +1087,12 @@ class TaskExecutorVector {
 
   void resize(int size) { _vector_executor.resize(size); }
 
-  TaskExecutor<TaskT>& operator[](int index) {
-    if (_vector_executor.size() <= index || index <= -1) {
-      LOG(ERROR) << "_vector_executor.size() <= index or <= -1";
-      throw "_vector_executor.size() <= index or <= -1";
+  TaskExecutor<TaskT>& operator[](int task_index) {
+    if (_vector_executor.size() <= task_index || task_index <= -1) {
+      LOG(ERROR) << "_vector_executor.size() <= task_index or <= -1";
+      throw "_vector_executor.size() <= task_index or <= -1";
     }
-    return _vector_executor[index];
+    return _vector_executor[task_index];
   }
 
  private:
@@ -717,8 +1117,8 @@ class TaskManager {
   typedef typename TaskT::InVectorT InVectorT;
   typedef typename TaskT::OutVectorT OutVectorT;
 
-  explicit TaskManager(uint32_t index)  // NOLINT
-      : _model_index(index) {}
+  explicit TaskManager(uint32_t model_index)  // NOLINT
+      : _model_index(model_index) {}
 
   ~TaskManager() { wait(); }
 

core/predictor/framework/infer.cpp

@@ -26,6 +26,7 @@ int ReloadableInferEngine::proc_initialize_impl(
   _infer_thread_num = conf.runtime_thread_num();
   _infer_batch_size = conf.batch_infer_size();
   _infer_batch_align = conf.enable_batch_align();
+  _allow_split_request = conf.allow_split_request();
 
   _conf = conf;
 
@@ -56,9 +57,6 @@ int ReloadableInferEngine::proc_initialize(const configure::EngineDesc& conf,
   }
 
   // init bsf framework
-  im::bsf::TaskExecutorVector<TaskT>::instance()[_model_index]
-      .set_thread_init_fn(
-          boost::bind(&InferEngine::thrd_initialize_impl, this));
   im::bsf::TaskExecutorVector<TaskT>::instance()[_model_index]
       .set_thread_init_fn(
           boost::bind(&InferEngine::thrd_initialize_impl, this));
@@ -71,6 +69,8 @@ int ReloadableInferEngine::proc_initialize(const configure::EngineDesc& conf,
       _infer_batch_size);
   im::bsf::TaskExecutorVector<TaskT>::instance()[_model_index].set_batch_align(
       _infer_batch_align);
+  im::bsf::TaskExecutorVector<TaskT>::instance()[_model_index]
+      .set_allow_split_request(_allow_split_request);
   if (im::bsf::TaskExecutorVector<TaskT>::instance()[_model_index].start(
           _infer_thread_num) != 0) {
     LOG(ERROR) << "Failed start bsf executor, threads:" << _infer_thread_num;
@@ -79,7 +79,8 @@ int ReloadableInferEngine::proc_initialize(const configure::EngineDesc& conf,
 
   LOG(WARNING) << "Enable batch schedule framework, thread_num:"
                << _infer_thread_num << ", batch_size:" << _infer_batch_size
-               << ", enable_batch_align:" << _infer_batch_align;
+               << ", enable_batch_align:" << _infer_batch_align
+               << ", allow_split_request:" << _allow_split_request;
   return 0;
 }
 

core/predictor/framework/infer.h

@@ -165,6 +165,8 @@ class ReloadableInferEngine : public InferEngine {
   // Need to align batch_size in inferring
   bool _infer_batch_align;
 
+  // allow to split request in inferring
+  bool _allow_split_request;
   // model version
   uint64_t _version;
 };