!57 Support Aicpu Dynamic Shape

Merge pull request !57 from caifubi/shape

inc/framework/ge_runtime/task_info.h

@@ -161,12 +161,13 @@ class TbeTaskInfo : public TaskInfo {
 class AicpuTaskInfo : public TaskInfo {
  public:
   AicpuTaskInfo(const std::string &op_name, uint32_t stream_id, const string &so_name, const std::string &kernel_name,
-                const std::string &node_def, const std::vector<void *> &input_data_addrs,
+                const std::string &node_def, const std::string &ext_info, const std::vector<void *> &input_data_addrs,
                 const std::vector<void *> &output_data_addrs, bool dump_flag)
       : TaskInfo(op_name, stream_id, TaskInfoType::AICPU, dump_flag),
         so_name_(so_name),
         kernel_name_(kernel_name),
         node_def_(node_def),
+        ext_info_(ext_info),
         input_data_addrs_(input_data_addrs),
         output_data_addrs_(output_data_addrs) {}
   ~AicpuTaskInfo() override {}
@@ -176,11 +177,13 @@ class AicpuTaskInfo : public TaskInfo {
   const std::string &node_def() const { return node_def_; }
   const std::vector<void *> &input_data_addrs() const { return input_data_addrs_; }
   const std::vector<void *> &output_data_addrs() const { return output_data_addrs_; }
+  const std::string &ext_info() const { return ext_info_; }
 
  private:
   std::string so_name_;
   std::string kernel_name_;
   std::string node_def_;
+  std::string ext_info_;
   std::vector<void *> input_data_addrs_;
   std::vector<void *> output_data_addrs_;
 };

src/ge/ge_runtime/task/aicpu_task.cc

@@ -47,10 +47,36 @@ bool AicpuTask::Distribute() {
   auto io_addrs_num = static_cast<uint32_t>(io_addrs.size());
   auto io_addrs_size = static_cast<uint32_t>(io_addrs_num * sizeof(void *));
   constexpr uint32_t io_addr_offset = sizeof(aicpu::AicpuParamHead);
-  uint32_t node_def_addr_offset = io_addr_offset + io_addrs_size;
-  uint32_t args_size =
-    sizeof(aicpu::AicpuParamHead) + io_addrs_size + static_cast<uint32_t>(task_info_->node_def().size());
-  aicpu::AicpuParamHead aicpu_param_head = {args_size, io_addrs_num};
+  uint32_t node_def_len_offset = io_addr_offset + io_addrs_size;
+  uint32_t node_def_addr_offset = node_def_len_offset + sizeof(uint32_t);
+  uint32_t args_size = sizeof(aicpu::AicpuParamHead) + io_addrs_size +
+                       static_cast<uint32_t>(task_info_->node_def().size()) + sizeof(uint32_t);
+
+  aicpu::AicpuParamHead aicpu_param_head;
+  aicpu_param_head.length = args_size;
+  aicpu_param_head.ioAddrNum = io_addrs_num;
+  auto ext_info = task_info_->ext_info();
+  uint32_t ext_size = ext_info.size();
+  if (ext_info.empty()) {
+    aicpu_param_head.extInfoLength = 0;
+    aicpu_param_head.extInfoAddr = 0;
+  } else {
+    rtError_t flag = rtMalloc(&ext_info_, ext_size, RT_MEMORY_HBM);
+    if (flag != RT_ERROR_NONE) {
+      GELOGE(RT_FAILED, "Call rt api(rtMalloc) failed, ret: 0x%X.", flag);
+      return false;
+    }
+
+    flag = rtMemcpy(ext_info_, ext_size, reinterpret_cast<void *>(ext_info.data()), ext_size, RT_MEMCPY_HOST_TO_DEVICE);
+    if (flag != RT_ERROR_NONE) {
+      GELOGE(RT_FAILED, "Call rt api(rtMemCpy) failed, ret: 0x%X.", flag);
+      return false;
+    }
+
+    GELOGI("ext info size:", ext_size);
+    aicpu_param_head.extInfoLength = ext_size;
+    aicpu_param_head.extInfoAddr = reinterpret_cast<uintptr_t>(ext_info_);
+  }
 
   // Malloc device memory for args
   rtError_t rt_ret = rtMalloc(&args_, args_size, RT_MEMORY_HBM);
@@ -76,6 +102,17 @@ bool AicpuTask::Distribute() {
       return false;
     }
   }
+
+  // Memcpy node def
+  auto size = task_info_->node_def().size();
+  rt_ret =
+    rtMemcpy(reinterpret_cast<void *>(reinterpret_cast<uint8_t *>(args_) + node_def_len_offset), sizeof(uint32_t),
+             reinterpret_cast<const void *>(&size), sizeof(uint32_t), RT_MEMCPY_HOST_TO_DEVICE);
+  if (rt_ret != RT_ERROR_NONE) {
+    GELOGE(RT_FAILED, "Call rt api(rtMemcpy) failed, ret: 0x%X.", rt_ret);
+    return false;
+  }
+
   // Memcpy node def
   rt_ret = rtMemcpy(reinterpret_cast<void *>(reinterpret_cast<uint8_t *>(args_) + node_def_addr_offset),
                     task_info_->node_def().size(), reinterpret_cast<const void *>(task_info_->node_def().data()),

src/ge/ge_runtime/task/aicpu_task.h

@@ -41,6 +41,7 @@ class AicpuTask : public TaskRepeater<AicpuTaskInfo> {
   std::shared_ptr<AicpuTaskInfo> task_info_;
   void *stream_;
   void *args_;
+  void *ext_info_;
   void *input_output_addr_;
 };
 }  // namespace model_runner