视频流GPU解码在ffempg的实现（一）-基本概念

这段时间在实现Gpu的视频流解码，遇到了很多的问题。

得到了阿里视频处理专家蔡鼎老师以及英伟达开发季光老师的指导，在这里表示感谢！

基本命令（linux下）

1.查看物理显卡

lspci  | grep -i vga
root@g1060server:/home/user# lspci  | grep -i vga
09:00.0 VGA compatible controller: ASPEED Technology, Inc. ASPEED Graphics Family (rev 30)
81:00.0 VGA compatible controller: NVIDIA Corporation Device 1c03 (rev a1)
82:00.0 VGA compatible controller: NVIDIA Corporation Device 1c03 (rev a1)

2.直接查看英伟达的物理显卡信息
有的时候因为服务器型号，GPU型号等不兼容等问题，会导致主板无法识别到插入的显卡，
我们可用下面的命令来查看主板是否识别到了显卡：

root@g1060server:/home/user# lspci | grep -i nvidia
81:00.0 VGA compatible controller: NVIDIA Corporation Device 1c03 (rev a1)
81:00.1 Audio device: NVIDIA Corporation Device 10f1 (rev a1)
82:00.0 VGA compatible controller: NVIDIA Corporation Device 1c03 (rev a1)
82:00.1 Audio device: NVIDIA Corporation Device 10f1 (rev a1)

出现上面的东西，说明主板已经识别到显卡信息

cuda版本，驱动信息

root@g1060server:/home/user# nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2013 NVIDIA Corporation
Built on Wed_Jul_17_18:36:13_PDT_2013
Cuda compilation tools, release 5.5, V5.5.0

英伟达显卡运行状态信息

root@g1060server:/home/user# nvidia-smi
modprobe: ERROR: could not insert 'nvidia_340': No such device
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

查看失败，一般没安装驱动

user@g1060server:~$ nvidia-smi
Fri Jan  5 21:50:34 2018       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 384.90                 Driver Version: 384.90                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 106...  Off  | 00000000:81:00.0  On |                  N/A |
| 32%   35C    P8    10W / 120W |   3083MiB /  6071MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   1  GeForce GTX 106...  Off  | 00000000:82:00.0 Off |                  N/A |
| 32%   37C    P8    10W / 120W |   2542MiB /  6072MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

查看成功

查看cuda驱动是否安装成功

root@g1060server:/home/user# cd /usr/local/cuda-8.0/samples/1_Utilities/deviceQuery
root@g1060server:/usr/local/cuda-8.0/samples/1_Utilities/deviceQuery# ls
deviceQuery  deviceQuery.cpp  deviceQuery.o  Makefile  NsightEclipse.xml  readme.txt
root@g1060server:/usr/local/cuda-8.0/samples/1_Utilities/deviceQuery# make
make: 没有什么可以做的为 `all'。
root@g1060server:/usr/local/cuda-8.0/samples/1_Utilities/deviceQuery# ./deviceQuery
./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 35
-> CUDA driver version is insufficient for CUDA runtime version
Result = FAIL

再次确认cuda驱动安装失败

查看cuda是否安装成功
/usr/local/cuda/extras/demo_suite/deviceQuery

root@g1060server:/home/user/mjl/test# /usr/local/cuda/extras/demo_suite/deviceQuery
/usr/local/cuda/extras/demo_suite/deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

Detected 2 CUDA Capable device(s)

Device 0: "GeForce GTX 1060 6GB"
  CUDA Driver Version / Runtime Version          9.0 / 8.0
  CUDA Capability Major/Minor version number:    6.1
  Total amount of global memory:                 6071 MBytes (6366363648 bytes)
  (10) Multiprocessors, (128) CUDA Cores/MP:     1280 CUDA Cores
  GPU Max Clock rate:                            1709 MHz (1.71 GHz)
  Memory Clock rate:                             4004 Mhz
  Memory Bus Width:                              192-bit
  L2 Cache Size:                                 1572864 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 2 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Disabled
  Device supports Unified Addressing (UVA):      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 129 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

Device 1: "GeForce GTX 1060 6GB"
  CUDA Driver Version / Runtime Version          9.0 / 8.0
  CUDA Capability Major/Minor version number:    6.1
  Total amount of global memory:                 6073 MBytes (6367739904 bytes)
  (10) Multiprocessors, (128) CUDA Cores/MP:     1280 CUDA Cores
  GPU Max Clock rate:                            1709 MHz (1.71 GHz)
  Memory Clock rate:                             4004 Mhz
  Memory Bus Width:                              192-bit
  L2 Cache Size:                                 1572864 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 2 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Disabled
  Device supports Unified Addressing (UVA):      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 130 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
> Peer access from GeForce GTX 1060 6GB (GPU0) -> GeForce GTX 1060 6GB (GPU1) : Yes
> Peer access from GeForce GTX 1060 6GB (GPU1) -> GeForce GTX 1060 6GB (GPU0) : Yes

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 9.0, CUDA Runtime Version = 8.0, NumDevs = 2, Device0 = GeForce GTX 1060 6GB, Device1 = GeForce GTX 1060 6GB
Result = PASS

查看成功

主要流程

要想实现ffempg的GPU化，必须要要对ffempg的解码流程有基本的认识才能改造（因为GPU也是这个流程，不过中间一部分用GPU运算）

我在http://www.cnblogs.com/baldermurphy/p/7828337.html 中曾经帖出过CPU解码的流程

主要流程如下

    avformat_network_init();
    av_register_all();//1.注册各种编码解码模块，如果3.3及以上版本，里面包含GPU解码模块
  
    std::string tempfile = “xxxx”;//视频流地址

    avformat_find_stream_info(format_context_, nullptr）//2.拉取一小段数据流分析，便于得到数据的基本格式
    if (AVMEDIA_TYPE_VIDEO == enc->codec_type && video_stream_index_ < 0)//3.筛选出视频流
    codec_ = avcodec_find_decoder(enc->codec_id);//4.找到对应的解码器
    codec_context_ = avcodec_alloc_context3(codec_);//5.创建解码器对应的结构体
    
    av_read_frame(format_context_, &packet_); //6.读取数据包
    
    avcodec_send_packet(codec_context_, &packet_) //7.发出解码
    avcodec_receive_frame(codec_context_, yuv_frame_) //8.接收解码 
    
    sws_scale(y2r_sws_context_, yuv_frame_->data, yuv_frame_->linesize, 0, codec_context_->height, rgb_data_, rgb_line_size_) //9.数据格式转换

 GPU解码需要改变4,7,8,9这几个步骤，也就是

找到gpu解码器，

拉取数据给GPU解码器，

得到解码后的数据，

数据格式使用gpu转换（如果需要的话，如nv12转bgra），

最终的格式由具体的需求确定，比如很多opengl的互操作，转成指定的格式（bgra），共用一段内存，数据立刻刷新，连拷贝都不用；

如果是转化成图片，又是另一种需求(bgr)；

适用场景的匹配

不得不提的一点是，GPU运算是一个很好的功能，可是也要看需求和场景，如果不考虑这个，可能得不偿失

比如一个极端的例子，opencv里面也有实现图片的解码，可是在追求效率的时候不会使用它的，

因为一张图片数据上传到GPU（非并行，很耗时），解码（非常快），GPU显存拷贝到内存（非并行，很耗时）

在上传和拷贝出来的就花了几百毫秒，而图片数据的操作很频繁，这会导致cpu占用率的得不到很好的缓解，甚至有的时候，不降反升，解码虽然快，可是用户的体验是慢，而且CPU，GPU都占用了

主要的几个网站

英伟达推荐的ffempg的gpu解码sdk

https://developer.nvidia.com/nvidia-video-codec-sdk

检查显存泄露的工具

http://docs.nvidia.com/cuda/cuda-memcheck/index.html#device-side-allocation-checking