Fixing workspace mismatch

ge/hybrid/node_executor/aicore/aicore_node_executor.cc

@@ -66,7 +66,7 @@ Status AiCoreNodeExecutor::LoadTask(const HybridModel &model, const NodePtr &nod
   }
 
   AiCoreTaskBuilder builder(node->GetOpDesc(), *task_defs);
-  std::unique_ptr<NodeTask> node_task;
+  std::unique_ptr<AiCoreNodeTask> node_task;
   GE_CHK_STATUS_RET(builder.BuildTask(node_task, true, is_single_op),
                     "[%s] Failed to build op tasks.", node->GetName().c_str());
   task = std::move(node_task);
@@ -99,7 +99,7 @@ Status AiCoreNodeExecutor::GenNodeKey(const NodePtr &node, std::string &node_key
   return SUCCESS;
 }
 
-bool AiCoreNodeTaskRegistry::AddTask(const std::string &node_key, const std::shared_ptr<NodeTask> task) {
+bool AiCoreNodeTaskRegistry::AddTask(const std::string &node_key, const std::shared_ptr<AiCoreNodeTask> &task) {
   GE_CHECK_NOTNULL(task);
   std::lock_guard<std::mutex> lock(mutex_);
   auto iter = reg_node_tasks_.find(node_key);
@@ -111,7 +111,7 @@ bool AiCoreNodeTaskRegistry::AddTask(const std::string &node_key, const std::sha
   return ret.second;
 }
 
-std::shared_ptr<NodeTask> AiCoreNodeTaskRegistry::GetTask(const std::string &node_key) {
+std::shared_ptr<AiCoreNodeTask> AiCoreNodeTaskRegistry::GetTask(const std::string &node_key) {
   std::lock_guard<std::mutex> lock(mutex_);
   auto iter = reg_node_tasks_.find(node_key);
   return (iter != reg_node_tasks_.end()) ? iter->second : nullptr;
@@ -140,9 +140,12 @@ Status AiCoreNodeExecutor::CompileTask(const HybridModel &model,
 
   auto node_key = std::to_string(model.GetModelId()) + "/" + shape_key;
   GELOGD("NodeKey for %s = %s", node->GetName().c_str(), node_key.c_str());
-  task = registry.GetTask(node_key);
+  auto aicore_task = registry.GetTask(node_key);
   if (task != nullptr) {
+    // The workspaces needed by a operator may differ with different shapes
+    op_desc->SetWorkspaceBytes(aicore_task->GetWorkspaceSizes());
     GELOGI("AiCoreNodeExecutor(%s) CompileTask Skip.", node->GetName().c_str());
+    task = std::move(aicore_task);
     return SUCCESS;
   }
 
@@ -153,16 +156,18 @@ Status AiCoreNodeExecutor::CompileTask(const HybridModel &model,
   GELOGD("successfully generated task_defs: %s", node->GetName().c_str());
 
   AiCoreTaskBuilder builder(node->GetOpDesc(), task_defs);
-  std::unique_ptr<NodeTask> node_task;
+  std::unique_ptr<AiCoreNodeTask> node_task;
   GE_CHK_STATUS_RET(builder.BuildTask(node_task, false), "[%s] Failed to build op tasks.", node->GetName().c_str());
-  task = std::move(node_task);
+  node_task->SetWorkspaceSizes(op_desc->GetWorkspaceBytes());
+  aicore_task = std::move(node_task);
   GELOGD("successfully created node task: %s", node->GetName().c_str());
 
-  if (!registry.AddTask(node_key, task)) {
+  if (!registry.AddTask(node_key, aicore_task)) {
     GELOGE(INTERNAL_ERROR, "Add NodeTask failed, op name = %s.", node->GetName().c_str());
     return INTERNAL_ERROR;
   }
 
+  task = std::move(aicore_task);
   GELOGI("AiCoreNodeExecutor(%s) CompileTask End.", node->GetName().c_str());
   return SUCCESS;
 }
@@ -247,6 +252,14 @@ bool AiCoreNodeTask::IsSupportDynamicShape() {
   return true;
 }
 
+const vector<int64_t> &AiCoreNodeTask::GetWorkspaceSizes() const {
+  return workspace_sizes_;
+}
+
+void AiCoreNodeTask::SetWorkspaceSizes(const vector<int64_t> &workspace_sizes) {
+  workspace_sizes_ = workspace_sizes;
+}
+
 TaskCompilerFactory &TaskCompilerFactory::GetInstance() {
   static TaskCompilerFactory instance;
   return instance;

ge/hybrid/node_executor/aicore/aicore_node_executor.h

@@ -24,7 +24,6 @@
 
 namespace ge {
 namespace hybrid {
-
 class TaskCompiler {
  public:
   TaskCompiler() = default;
@@ -42,11 +41,11 @@ class AiCoreNodeTaskRegistry {
     return instance;
   }
 
-  std::shared_ptr<NodeTask> GetTask(const std::string &node_key);
+  std::shared_ptr<AiCoreNodeTask> GetTask(const std::string &node_key);
-  bool AddTask(const std::string &node_key, const std::shared_ptr<NodeTask> task);
+  bool AddTask(const std::string &node_key, const std::shared_ptr<AiCoreNodeTask> &task);
  private:
   AiCoreNodeTaskRegistry() = default;
-  std::map<std::string, std::shared_ptr<NodeTask>> reg_node_tasks_;
+  std::map<std::string, std::shared_ptr<AiCoreNodeTask>> reg_node_tasks_;
   std::mutex mutex_;
 };
 
@@ -59,8 +58,12 @@ class AiCoreNodeTask : public NodeTask {
 
   Status UpdateArgs(TaskContext &context) override;
   Status ExecuteAsync(TaskContext &context, std::function<void()> done_callback) override;
+
+  const vector<int64_t> &GetWorkspaceSizes() const;
+  void SetWorkspaceSizes(const vector<int64_t> &workspace_sizes);
  private:
   std::vector<std::unique_ptr<AiCoreOpTask>> tasks_;
+  std::vector<int64_t> workspace_sizes_;
 };
 
 class AiCoreNodeExecutor : public NodeExecutor {

ge/hybrid/node_executor/aicore/aicore_task_builder.cc

@@ -37,7 +37,7 @@ AiCoreTaskBuilder::AiCoreTaskBuilder(const OpDescPtr &op_desc, const std::vector
     : op_desc_(op_desc), task_defs_(task_defs) {
 }
 
-Status AiCoreTaskBuilder::BuildTask(std::unique_ptr<NodeTask> &node_task,
+Status AiCoreTaskBuilder::BuildTask(std::unique_ptr<AiCoreNodeTask> &node_task,
                                     bool ignore_failure_on_atomic,
                                     bool is_single_op) {
   GE_CHECK_NOTNULL(op_desc_);

ge/hybrid/node_executor/aicore/aicore_task_builder.h

@@ -27,6 +27,7 @@
 
 namespace ge {
 namespace hybrid {
+class AiCoreNodeTask;
 class AiCoreKernelRegistry {
  public:
   ~AiCoreKernelRegistry() = default;
@@ -47,7 +48,9 @@ class AiCoreTaskBuilder {
   AiCoreTaskBuilder(const OpDescPtr &op_desc, const std::vector<domi::TaskDef> &task_defs);
   ~AiCoreTaskBuilder() = default;
 
-  Status BuildTask(std::unique_ptr<NodeTask> &node_task, bool ignore_failure_on_atomic, bool is_single_op = false);
+  Status BuildTask(std::unique_ptr<AiCoreNodeTask> &node_task,
+                   bool ignore_failure_on_atomic,
+                   bool is_single_op = false);
 
  private:
   bool ExpectAtomicAddrCleanTask();

ge/hybrid/node_executor/aicore/aicore_task_compiler.cc

@@ -61,11 +61,11 @@ Status AiCoreTaskCompiler::CompileOp(const NodePtr &node, std::vector<domi::Task
   GE_CHECK_NOTNULL(node);
   GELOGI("AiCoreTaskCompiler(%s) CompileOp Start.", node->GetName().c_str());
 
-
+  auto op_desc = node->GetOpDesc();
+  op_desc->SetWorkspaceBytes({});
   GE_CHK_STATUS_RET_NOLOG(DoCompileOp(node));
   GELOGD("successfully compiled op: %s", node->GetName().c_str());
 
-  auto op_desc = node->GetOpDesc();
   std::vector<int64_t> input_offsets(op_desc->GetInputsSize(), kMemBase);
   std::vector<int64_t> output_offsets(op_desc->GetOutputsSize(), kMemBase);
   op_desc->SetInputOffset(input_offsets);