GPU/CPU内存访问同步操作开发指南 (C/C++)

场景介绍

NativeFence是管理fenceFd同步状态的模块。开发者可通过其接口实现以下功能：设置阻塞时间、永久阻塞、关闭fenceFd以及检查其有效性。

接口说明

接口名	描述
OH_NativeFence_IsValid (int fenceFd)	检查fenceFd是否有效。
OH_NativeFence_Wait (int fenceFd, uint32_t timeout)	阻塞传入的fenceFd，超时参数指定了允许等待的最长时间。
OH_NativeFence_WaitForever (int fenceFd)	永久阻塞传入的fenceFd。
OH_NativeFence_Close (int fenceFd)	关闭fenceFd。

详细的接口说明请参考NativeFence。

开发步骤

以下步骤描述了如何使用NativeFence提供的Native API接口。

添加动态链接库

CMakeLists.txt中添加以下库文件。

libnative_fence.so

头文件

#include <native_fence/native_fence.h>
#include <cstring>
#include <iostream>
#include <linux/sync_file.h>
#include <signal.h>
#include <sys/signalfd.h>
#include <unistd.h>

1. 使用signalfd()接口创建fenceFd。

   sigset_t mask;
   sigemptyset(&mask);
   sigaddset(&mask, SIGINT); // Monitor SIGINT signal (Ctrl C)
   sigaddset(&mask, SIGURG); // Generated when urgent data or out of band data arrives at the socket
   sigprocmask(SIG_BLOCK, &mask, NULL);
   int sfd = signalfd(-1, &mask, 0);

2. 判断传入的fenceFd是否合法。

   bool isValid = OH_NativeFence_IsValid(INVALID_FD);
   if (!isValid) {
       DRAWING_LOGW("fenceFd is invalid");
   }

3. 调用OH_NativeFence_Wait阻塞接口，等待fence完成后进行下一步操作。

   constexpr uint32_t TIMEOUT_MS = 5000;
   // ...
   bool result = OH_NativeFence_Wait(INVALID_FD, TIMEOUT_MS);

4. 调用OH_NativeFence_WaitForever阻塞接口，等待fence完成后进行下一步操作。

   bool result2 = false;
   auto startTime = std::chrono::steady_clock::now();
   result2 = OH_NativeFence_WaitForever(sfd);
   auto endTime = std::chrono::steady_clock::now();
   auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(endTime - startTime).count();
   if (result2) {
       DRAWING_LOGI("SyncFenceWaitForever has an event occurring result2 %{public}d, cost_time: %{public}d",
           result2, duration);
   } else {
       DRAWING_LOGI("SyncFenceWaitForever timeout with no event occurrence"
           "result2 %{public}d, cost_time: %{public}d", result2, duration);
   }

5. GPU或CPU发送同步信号(signal)，通知fenceFd解除阻塞。

6. 关闭fenceFd。

   OH_NativeFence_Close(sfd);