From 583cf40984b91ec3700f242afd105385b89bc21d Mon Sep 17 00:00:00 2001
From: ChunelFeng <chunel@foxmail.com>
Date: Mon, 27 Nov 2023 23:38:00 +0800
Subject: [PATCH] [perf] optimize all parallel dag speed. origin 21.3s, now
 12.4s

---
 .../GDynamicEngine/GDynamicEngine.cpp         | 29 +++++++++++++++++--
 .../_GEngine/GDynamicEngine/GDynamicEngine.h  |  8 ++++-
 src/GraphCtrl/GraphParam/GParamManager.inl    |  1 -
 tutorial/MyGNode/MyWriteParamNode.h           | 11 +++++--
 4 files changed, 43 insertions(+), 6 deletions(-)

diff --git a/src/GraphCtrl/GraphElement/_GEngine/GDynamicEngine/GDynamicEngine.cpp b/src/GraphCtrl/GraphElement/_GEngine/GDynamicEngine/GDynamicEngine.cpp
index 6eb27233..b8d33760 100644
--- a/src/GraphCtrl/GraphElement/_GEngine/GDynamicEngine/GDynamicEngine.cpp
+++ b/src/GraphCtrl/GraphElement/_GEngine/GDynamicEngine/GDynamicEngine.cpp
@@ -37,9 +37,13 @@ CStatus GDynamicEngine::setup(const GSortedGElementPtrSet& elements) {
 
 CStatus GDynamicEngine::run() {
     CGRAPH_FUNCTION_BEGIN
-    beforeRun();
-    asyncRunAndWait();
 
+    if (likely(total_end_size_ != total_element_arr_.size())) {
+        beforeRun();
+        asyncRunAndWait();
+    } else {
+        parallelRunAll();
+    }
     status = cur_status_;
     CGRAPH_FUNCTION_END
 }
@@ -154,4 +158,25 @@ CVoid GDynamicEngine::wait() {
     });
 }
 
+
+CVoid GDynamicEngine::parallelRunAll() {
+    /**
+     * 主要适用于dag是纯并发逻辑的情况
+     * 直接并发的执行所有的流程，从而减少调度损耗
+     * 实测效果，在32路纯并行的情况下，整体耗时从 21.5s降低到 12.5s
+     * 非纯并行逻辑，不走此函数
+     */
+    std::vector<std::future<CStatus>> futures;
+    futures.reserve(front_element_arr_.size());
+    for (auto* element : front_element_arr_) {
+        futures.emplace_back(thread_pool_->commit([element] {
+            return element->fatProcessor(CFunctionType::RUN);
+        }, calcIndex(element)));
+    }
+
+    for (auto& fut : futures) {
+        cur_status_ += fut.get();
+    }
+}
+
 CGRAPH_NAMESPACE_END
diff --git a/src/GraphCtrl/GraphElement/_GEngine/GDynamicEngine/GDynamicEngine.h b/src/GraphCtrl/GraphElement/_GEngine/GDynamicEngine/GDynamicEngine.h
index 8c60f4b5..11489faf 100644
--- a/src/GraphCtrl/GraphElement/_GEngine/GDynamicEngine/GDynamicEngine.h
+++ b/src/GraphCtrl/GraphElement/_GEngine/GDynamicEngine/GDynamicEngine.h
@@ -62,12 +62,18 @@ class GDynamicEngine : public GEngine {
     */
     CVoid wait();
 
+    /**
+     * 并发的执行所有的节点
+     * @return
+     */
+    CVoid parallelRunAll();
+
 private:
     GElementPtrArr total_element_arr_;                          // pipeline中所有的元素信息集合
     GElementPtrArr front_element_arr_;                          // 没有依赖的元素信息
     CSize total_end_size_ = 0;                                  // 图结束节点数量
     CSize finished_end_size_ = 0;                               // 执行结束节点数量
-    std::atomic<CSize> run_element_size_ { 0 };              // 执行元素的个数，用于后期校验。这里和静态不一样，需要加atomic
+    std::atomic<CSize> run_element_size_ { 0 };                 // 执行元素的个数，用于后期校验。这里和静态不一样，需要加atomic
     CStatus cur_status_;                                        // 当前全局的状态信息
 
     std::mutex lock_;
diff --git a/src/GraphCtrl/GraphParam/GParamManager.inl b/src/GraphCtrl/GraphParam/GParamManager.inl
index b26dab31..4c18ed45 100644
--- a/src/GraphCtrl/GraphParam/GParamManager.inl
+++ b/src/GraphCtrl/GraphParam/GParamManager.inl
@@ -37,7 +37,6 @@ CStatus GParamManager::create(const std::string& key, CBool backtrace) {
 template<typename T,
         c_enable_if_t<std::is_base_of<GParam, T>::value, int>>
 T* GParamManager::get(const std::string& key) {
-    CGRAPH_LOCK_GUARD lock(this->mutex_);
     auto result = params_map_.find(key);
     if (result == params_map_.end()) {
         return nullptr;
diff --git a/tutorial/MyGNode/MyWriteParamNode.h b/tutorial/MyGNode/MyWriteParamNode.h
index 131266bf..fb50088c 100644
--- a/tutorial/MyGNode/MyWriteParamNode.h
+++ b/tutorial/MyGNode/MyWriteParamNode.h
@@ -14,7 +14,7 @@
 
 class MyWriteParamNode : public CGraph::GNode {
 public:
-    CStatus init () override {
+    CStatus init() override {
         CStatus status;
         /**
          * 推荐在init()中，将可能用到的参数创建好。也支持在run的时候创建
@@ -24,7 +24,14 @@ class MyWriteParamNode : public CGraph::GNode {
         return status;
     }
 
-    CStatus run () override {
+    CStatus run() override {
+        /**
+         * 为了提高执行效率，
+         * 在【创建参数】的时候，【提供】锁保护机制
+         * 在【获取参数】的时候，【不提供】锁保护的机制
+         * 故无法通过在run()过程中，并发的通过 createGParam 和 getGParam 创建和获取参数
+         * 如果需要做此操作，请自行外部加锁
+         */
         auto* myParam = CGRAPH_GET_GPARAM_WITH_NO_EMPTY(MyParam, "param1")
         int val = 0;
         int cnt = 0;