diff --git a/docker-compose.yml b/docker-compose.yml
index 0a1846e..5b5a806 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -18,7 +18,8 @@ services:
       # --- VPU/NPU/RGA/GPU 硬件访问 ---
       - /dev/mpp_service:/dev/mpp_service
       - /dev/rga:/dev/rga
-      - /dev/dri:/dev/dri   
+      - /dev/dri:/dev/dri  
+      - /dev/dma_heap/system:/dev/dma_heap/system 
 
       - source: /sys/bus/iio/devices/iio:device0
         target: /sys/bus/iio/devices/iio:device0
diff --git a/monitor b/monitor
new file mode 100755
index 0000000..8a6a65c
Binary files /dev/null and b/monitor differ
diff --git a/rknn_perf_monitor.cpp b/rknn_perf_monitor.cpp
new file mode 100644
index 0000000..8640c24
--- /dev/null
+++ b/rknn_perf_monitor.cpp
@@ -0,0 +1,163 @@
+#include <iostream>
+#include <fstream>
+#include <sstream>
+#include <string>
+#include <vector>
+#include <thread>
+#include <chrono>
+#include <iomanip>
+#include <numeric>
+#include <algorithm>
+
+// ==========================================
+// CPU 监控类 (保持不变)
+// ==========================================
+class CpuMonitor {
+public:
+    CpuMonitor() { get_usage(); }
+
+    float get_usage() {
+        std::ifstream file("/proc/stat");
+        std::string line;
+        if (!std::getline(file, line)) return 0.0f;
+        
+        std::istringstream ss(line.substr(3));
+        unsigned long long user, nice, system, idle, iowait, irq, softirq, steal;
+        ss >> user >> nice >> system >> idle >> iowait >> irq >> softirq >> steal;
+
+        unsigned long long current_idle = idle + iowait;
+        unsigned long long current_non_idle = user + nice + system + irq + softirq + steal;
+        unsigned long long current_total = current_idle + current_non_idle;
+
+        unsigned long long total_diff = current_total - prev_total;
+        unsigned long long idle_diff = current_idle - prev_idle;
+
+        prev_idle = current_idle;
+        prev_total = current_total;
+
+        if (total_diff == 0) return 0.0f;
+        return (float)(total_diff - idle_diff) * 100.0f / total_diff;
+    }
+
+private:
+    unsigned long long prev_idle = 0;
+    unsigned long long prev_total = 0;
+};
+
+// ==========================================
+// NPU 读取函数 (修改为读取所有核心)
+// ==========================================
+std::string get_npu_usage_str() {
+    std::ifstream file("/sys/kernel/debug/rknpu/load");
+    if (!file.is_open()) return "N/A (No Perm)";
+
+    std::string line;
+    if (std::getline(file, line)) {
+        // RK3588 通常格式: "NPU load:  Core0:  0%, Core1:  0%, Core2:  0%,"
+        // 我们去掉前面的 "NPU load:" 前缀，只保留 Core 信息
+        size_t pos = line.find("NPU load:");
+        if (pos != std::string::npos) {
+            line = line.substr(pos + 9);
+        }
+        
+        // 简单的字符串清理：去掉多余空格，去掉结尾逗号
+        std::string clean_line;
+        bool last_space = false;
+        for (char c : line) {
+            if (c == ',') continue; // 去掉逗号
+            if (isspace(c)) {
+                if (!last_space) {
+                    clean_line += ' ';
+                    last_space = true;
+                }
+            } else {
+                clean_line += c;
+                last_space = false;
+            }
+        }
+        return clean_line;
+    }
+    return "Idle";
+}
+
+// ==========================================
+// RGA 读取函数 (根据你找到的路径修改)
+// ==========================================
+std::string get_rga_usage_str() {
+    std::ifstream file("/sys/kernel/debug/rkrga/load");
+    if (!file.is_open()) return "N/A";
+
+    std::string line;
+    std::stringstream ss;
+    bool found_any = false;
+    int core_idx = 0;
+
+    while (std::getline(file, line)) {
+        // 忽略第一行 "num of scheduler"
+        if (line.find("scheduler") != std::string::npos && line.find("=") != std::string::npos && line.find("%") == std::string::npos) {
+            continue;
+        }
+
+        // 寻找包含 "%" 的行，通常负载都在这些行里
+        if (line.find("%") != std::string::npos) {
+            // 尝试提取这一行里的最后一个数字
+            // 假设格式类似: "scheduler[0]: full_load = 0.0%, load = 12.5%"
+            // 或者简单点: "load = 12%"
+            
+            // 简单粗暴法：找到 "%" 前面的数字
+            size_t percent_pos = line.rfind('%'); // 找最后一个 %
+            if (percent_pos != std::string::npos && percent_pos > 0) {
+                // 向前回溯找到数字起始位置
+                size_t num_end = percent_pos;
+                size_t num_start = num_end;
+                while (num_start > 0 && (isdigit(line[num_start-1]) || line[num_start-1] == '.')) {
+                    num_start--;
+                }
+                
+                if (num_start < num_end) {
+                    std::string num_str = line.substr(num_start, num_end - num_start);
+                    try {
+                        float load = std::stof(num_str);
+                        ss << "C" << core_idx++ << ":" << std::fixed << std::setprecision(0) << load << "% ";
+                        found_any = true;
+                    } catch (...) {}
+                }
+            }
+        }
+    }
+
+    if (!found_any) return "Idle (fmt err)";
+    return ss.str();
+}
+// ==========================================
+// 主函数
+// ==========================================
+int main() {
+    CpuMonitor cpu_mon;
+    
+    std::cout << "=== RK3588 System Monitor ===" << std::endl;
+    // 打印更宽的表头
+    std::cout << "----------------------------------------------------------------------------------" << std::endl;
+    std::cout << "|  CPU   |              NPU (3 Cores)             |           RGA (Cores)        |" << std::endl;
+    std::cout << "----------------------------------------------------------------------------------" << std::endl;
+
+    while (true) {
+        float cpu = cpu_mon.get_usage();
+        std::string npu = get_npu_usage_str();
+        std::string rga = get_rga_usage_str();
+
+        // 格式化输出
+        // CPU: 占用 6 格
+        // NPU: 占用 38 格 (预留给 Core0: xx% Core1: xx% Core2: xx%)
+        // RGA: 占用 28 格
+        std::cout << "\r| " 
+                  << std::fixed << std::setprecision(1) << std::setw(5) << cpu << "% | "
+                  << std::left << std::setw(38) << npu << " | "
+                  << std::setw(28) << rga << " |" 
+                  << std::flush;
+
+        std::this_thread::sleep_for(std::chrono::milliseconds(1000));
+    }
+
+    return 0;
+}
\ No newline at end of file
diff --git a/src/rknn_test/DmaBuffer.h b/src/rknn_test/DmaBuffer.h
new file mode 100644
index 0000000..ed682b7
--- /dev/null
+++ b/src/rknn_test/DmaBuffer.h
@@ -0,0 +1,53 @@
+#include <fcntl.h>
+#include <linux/dma-heap.h>
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <unistd.h>
+
+#include <iostream>
+#include <string>
+
+class DmaBuffer {
+public:
+	int fd;
+	void* vaddr;  // 虚拟地址（仅供调试或CPU必须介入时使用，如画字）
+	size_t size;
+
+	DmaBuffer(size_t size, const std::string& heap_name = "/dev/dma_heap/system") {
+		this->size = size;
+		this->vaddr = MAP_FAILED;
+		this->fd = -1;
+
+		int heap_fd = open(heap_name.c_str(), O_RDONLY | O_CLOEXEC);
+		if (heap_fd < 0) {
+			perror("Failed to open dma heap");
+			return;
+		}
+
+		struct dma_heap_allocation_data data = {0};
+		data.len = size;
+		data.fd_flags = O_RDWR | O_CLOEXEC;
+
+		if (ioctl(heap_fd, DMA_HEAP_IOCTL_ALLOC, &data) < 0) {
+			perror("dma heap alloc failed");
+			close(heap_fd);
+			return;
+		}
+
+		this->fd = data.fd;
+		close(heap_fd);
+
+		// 如果需要CPU访问（例如写OSD文字），则映射，否则可以不映射
+		this->vaddr = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, this->fd, 0);
+		if (this->vaddr == MAP_FAILED) {
+			perror("mmap failed");
+		}
+	}
+
+	~DmaBuffer() {
+		if (vaddr != MAP_FAILED)
+			munmap(vaddr, size);
+		if (fd >= 0)
+			close(fd);
+	}
+};
\ No newline at end of file