LVGL - RV1109 LVGL UI刷新效率优化-02
说明
前面好早写过一个文章，说明如何把LVGL移到RV1109上的操作，使用DRM方式！但出现刷新效率不高的问题！
因为一直没有真正的应用在产品中，所以也就放下了！
最近开发上需要考虑低成本，低内存的方案，所以后续考虑使用LVGL上来做UI，所以这里捡回来再炒一次！
性能分析
这里还是使用的LV_DEMO中的lv_demo_benchmark来测试的！
代码见前一相关的文:LVGL - RV1109 LVGL UI开发-01
先找到了为什么刷新FPS如此低的原因，主要耗时花在了:
lvgl_drm_flush->display_commit_ex->drm_commit_ex->drmCommit
上面！
这样导致UI的线程绘图也很慢！
解决方式
最开始想使用双缓存的方式来提交，但看代码drmCommit是整屏修改提交的！不过对drm里的这个commit底层操作还是不太了解！
还是把耗时的这个操作分开操作！使用另一个线程来commit，这样只要在另一个线程中，如是UI线程有调用lvgl_drm_flush 这个后，发送一个消息通知刷新线程drm commit操即可！
代码修改很简单，在 drm_commit_ex 中使用一个mutex和cond来通知另一个线程:
#ifdef DRM_COMMIT_THREAD
printf("commit request!!!!!!!!!!!\n");
pthread_mutex_lock(&g_commit_mutex);
pthread_cond_broadcast(&g_commit_cond);
pthread_mutex_unlock(&g_commit_mutex);
#else
ret = drmCommit(&disp->buf[num], disp->width, disp->height, 0, 0, &disp->dev, disp->plane_type);
if (ret) {
fprintf(stderr, "display commit error, ret = %d\n", ret);
}
#endif
这里用了一个宏 DRM_COMMIT_THREAD 来控制!
在另一个线程中主要处理如下，即把drmCommit动作放在这里来操作:
#ifdef DRM_COMMIT_THREAD
void* dsiplay_commit_thread_process(void* data)
{
int ret = 0;
while(g_commit_thread_start_flag)
{
pthread_mutex_lock(&g_commit_mutex);
pthread_cond_wait(&g_commit_cond, &g_commit_mutex);
pthread_mutex_unlock(&g_commit_mutex);
//printf("dsiplay_commit_thread_process!!!!!!............\n");
ret = drmCommit(&g_disp.buf[g_num], g_disp.width, g_disp.height, 0, 0, &g_disp.dev, g_disp.plane_type);
if (ret) {
fprintf(stderr, "display commit error, ret = %d\n", ret);
}
usleep(1000 * 40);
}
}
int display_commit_thread_start()
{
int ret = 0;
pthread_t pid = 0;
pthread_attr_t attr;
ret = pthread_attr_init(&attr);
if(ret != 0)
{
printf("start pthread_attr_init failure!!!\n");
return -1;
}
ret = pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
if(ret != 0)
{
pthread_attr_destroy(&attr);
printf( "pthread_attr_setdetachstate PTHREAD_CREATE_DETACHED failure!!!\n");
return -1;
}
ret = pthread_create(&pid, &attr, dsiplay_commit_thread_process, NULL);
if(ret == 0)
{
printf(" pthread_create OK! \n");
}else{
printf(" pthread_create failure!! \n");
pthread_attr_destroy(&attr);
return -1;
}
pthread_attr_destroy(&attr);
return 0;
}
#endif
性能表现
下面是 lv_demo_widgets 的统计数据:
lv_demo_widgets示例数据
FPS统计方式:
FPS统计方式,这里统计drmCommit花费的时间，看一定时长里能提交多少次:
ts = _clockTime_getCurrentMscTime();
ret = drmCommit(&g_disp.buf[g_num], g_disp.width, g_disp.height, 0, 0, &g_disp.dev, g_disp.plane_type);
if (ret) {
fprintf(stderr, "display commit error, ret = %d\n", ret);
}
commit_count++;
//usleep(1000 * 40);
commit_count_cost_time += _clockTime_checkLostMscTime(ts);
if(commit_count >= 64 || commit_count_cost_time > 1000 * 30)
{
int fps = commit_count / (commit_count_cost_time / 1000.0);
printf("FPS:%d\n", fps);
commit_count = 0;
commit_count_cost_time = 0;
}
}
FPS打印统计:
FPS:81
FPS:86
FPS:77
FPS:80
FPS:79
FPS:83
FPS:76
FPS:77
FPS:93
FPS:82
FPS:83
FPS:72
FPS:82
FPS:80
FPS:80
CPU占用
top - 15:57:46 up 6:57, 2 users, load average: 1.40, 0.84, 0.85
Tasks: 106 total, 3 running, 46 sleeping, 1 stopped, 0 zombie
%Cpu0 : 3.0/21.8 25[||||||||||||||||||||||||| ]
%Cpu1 : 25.7/23.8 50[|||||||||||||||||||||||||||||||||||||||||||||||||| ]
GiB Mem : 7.9/0.876 [ ]
GiB Swap: 0.0/0.000 [ ]
PID USER PR NI VIRT RES %CPU %MEM TIME+ S COMMAND
1 root 20 0 2.0m 0.2m 0.0 0.0 0:00.51 S init
102 root 20 0 2.0m 0.2m 0.0 0.0 0:00.05 S `- syslogd
105 root 20 0 2.0m 0.2m 0.0 0.0 0:00.04 S `- klogd
lv_demo_benchmark统计数据:
FPS打印统计
Result of FPS:47
FPS:80
Result of FPS:40
FPS:78
Result of FPS:46
FPS:74
Result of FPS:40
FPS:74
Result of FPS:42
FPS:83
Result of FPS:28
FPS:79
Result of FPS:50
Result of FPS:51
FPS:77
Result of FPS:48
FPS:77
Result of FPS:47
FPS:73
Result of FPS:42
FPS:75
Result of FPS:41
FPS:79
Result of FPS:49
FPS:75
Result of FPS:51
FPS:77
Result of FPS:50
FPS:78
Result of FPS:50
Result of FPS:50
FPS:80
Result of FPS:48
FPS:76
CPU占用
top - 16:03:47 up 7:03, 2 users, load average: 2.10, 1.93, 1.38
Tasks: 107 total, 2 running, 45 sleeping, 1 stopped, 0 zombie
%Cpu0 : 1.0/11.8 13[||||||||||||| ]
%Cpu1 : 68.6/16.7 85[|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||| ]
GiB Mem : 7.8/0.876 [ ]
GiB Swap: 0.0/0.000 [ ]
PID USER PR NI VIRT RES %CPU %MEM TIME+ S COMMAND
1 root 20 0 2.0m 0.2m 0.0 0.0 0:00.51 S init
102 root 20 0 2.0m 0.2m 0.0 0.0 0:00.05 S `- syslogd
105 root 20 0 2.0m 0.2m 0.0 0.0 0:00.04 S `- klogd
120 root 20 0 2.7m 1.9m 0.0 0.2 0:00.22 S `- udevd
573 root 20 0 1.9m 0.1m 0.0 0.0 0:00.00 S `- dropbear
1113 root 20 0 2.3m 1.6m 0.0 0.2 0:12.49 S `- dropbear
1114 root 20 0 2.0m 1.5m 0.0 0.2 0:00.01 S `- sh
1189 root 20 0 2.8m 1.6m 0.0 0.2 0:28.82 T `- top
1220 root 20 0 2.8m 1.6m 1.9 0.2 0:07.22 R `- top
628 root 20 0 1.4m 0.1m 0.0 0.0 0:00.00 S `- input-event-dae
851 root 20 0 2.0m 0.3m 0.0 0.0 0:00.02 S `- login
852 root 20 0 2.0m 1.6m 0.0 0.2 0:00.62 S `- sh
893 root 20 0 1.4m 0.2m 0.0 0.0 0:00.32 S `- wdt_deamon
1225 root 20 0 39.4m 14.7m 84.5 1.6 0:25.70 R `- pupa_face
2 root 20 0 0.0m 0.0m 0.0 0.0 0:00.03 S kthreadd
3 root 0 -20 0.0m 0.0m 0.0 0.0 0:00.00 I `- rcu_gp
4 root 0 -20 0.0m 0.0m 0.0 0.0 0:00.00 I `- rcu_par_gp
8 root 0 -20 0.0m 0.0m 0.0 0.0 0:00.00 I `- mm_percpu_wq
9 root 20 0 0.0m 0.0m 0.0 0.0 0:01.78 S `- ksoftirqd/0
10 root 20 0 0.0m 0.0m 0.0 0.0 1:17.11 I `- rcu_preempt
11 root 20 0 0.0m 0.0m 0.0 0.0 0:00.00 I `- rcu_sched
12 root 20 0 0.0m 0.0m 0.0 0.0 0:00.00 I `- rcu_bh
13 root rt 0 0.0m 0.0m 0.0 0.0 0:00.11 S `- migration/0
14 root 20 0 0.0m 0.0m 0.0 0.0 0:00.00 S `- cpuhp/0
15 root 20 0 0.0m 0.0m 0.0 0.0 0:00.00 S `- cpuhp/1
16 root rt 0 0.0m 0.0m 0.0 0.0 0:00.02 S `- migration/1
分开线程处理后，刷新率立即就上来了，UI线程的FPS在跑 lv_demo_benchmark时，平均commit可以跑到70FPS.
CPU占用率上,不过明显的上升了，使用单线程方式时，CPU占用率在只在5%上下，使用多线程方式时，CPU平均在15%左右，连续操作甚至50%以上！
所以还是还是需要控制一下FPS，这里在dsiplay_commit_thread_process加一个usleep操作.
在跑lv_demo_widgets这个示例时，第二个tab页的 Analytics 界面，能看到仪表盘的每一秒的跳动了，之前的代码是直接跳秒的。
加usleep控制一下刷新
FPS打印统计
FPS:24
Result of FPS:40
Result of FPS:37
Result of FPS:38
FPS:24
Result of FPS:33
Result of FPS:25
Result of FPS:25
FPS:24
Result of FPS:25
Result of FPS:23
Result of FPS:35
FPS:24
Result of FPS:26
Result of FPS:29
Result of FPS:25
Result of FPS:29
FPS:24
Result of FPS:24
Result of FPS:21
Result of FPS:16
FPS:24
CPU占用
top - 16:17:17 up 7:16, 2 users, load average: 1.08, 0.66, 0.86
Tasks: 106 total, 1 running, 46 sleeping, 1 stopped, 0 zombie
%Cpu(s): 21.3/19.3 41[|||||||||||||||||||||||||||||||||||||||| ]
GiB Mem : 7.8/0.876 [ ]
GiB Swap: 0.0/0.000 [ ]
PID USER PR NI VIRT RES %CPU %MEM TIME+ S COMMAND
1 root 20 0 2.0m 0.2m 0.0 0.0 0:00.51 S init
102 root 20 0 2.0m 0.2m 0.0 0.0 0:00.05 S `- syslogd
105 root 20 0 2.0m 0.2m 0.0 0.0 0:00.04 S `- klogd
120 root 20 0 2.7m 1.9m 0.0 0.2 0:00.22 S `- udevd
573 root 20 0 1.9m 0.1m 0.0 0.0 0:00.00 S `- dropbear
1113 root 20 0 2.3m 1.6m 1.0 0.2 0:13.36 S `- dropbear
1114 root 20 0 2.0m 1.5m 0.0 0.2 0:00.01 S `- sh
1189 root 20 0 2.8m 1.6m 0.0 0.2 0:28.82 T `- top
1220 root 20 0 2.8m 1.6m 1.9 0.2 0:24.25 R `- top
628 root 20 0 1.4m 0.1m 0.0 0.0 0:00.00 S `- input-event-dae
851 root 20 0 2.0m 0.3m 0.0 0.0 0:00.02 S `- login
852 root 20 0 2.0m 1.6m 0.0 0.2 0:00.63 S `- sh
893 root 20 0 1.4m 0.2m 0.0 0.0 0:00.33 S `- wdt_deamon
1234 root 20 0 39.4m 14.5m 62.1 1.6 1:05.77 S `- pupa_face
2 root 20 0 0.0m 0.0m 0.0 0.0 0:00.04 S kthreadd
3 root 0 -20 0.0m 0.0m 0.0 0.0 0:00.00 I `- rcu_gp
4 root 0 -20 0.0m 0.0m 0.0 0.0 0:00.00 I `- rcu_par_gp
8 root 0 -20 0.0m 0.0m 0.0 0.0 0:00.00 I `- mm_percpu_wq
9 root 20 0 0.0m 0.0m 0.0 0.0 0:01.84 S `- ksoftirqd/0
10 root 20 0 0.0m 0.0m 1.0 0.0 1:19.85 I `- rcu_preempt
11 root 20 0 0.0m 0.0m 0.0 0.0 0:00.00 I `- rcu_sched
在使用单线程时，lv_demo_widgets在操作，如果是在刷新界面时，很明显有时是无法反应你的触摸操作的，改用多线程后，响应基本是很实时.