1

先介绍几个基本信息:

操作系统:Win7 64Bit | 显卡:GTX970

我有一个暂存 ID3D11Texture2D ,我想对其进行编码。

我想直接通过nvEncRegisterResource使用纹理,但似乎我只能通过 D3D9 而没有 D3D11 纹理。否则我会得到一个NV_ENC_ERR_UNIMPLEMENTED

因此,我创建了一个输入缓冲区并手动填充它。

纹理的格式为DXGI_FORMAT_R8G8B8A8_UNORMDXGI_FORMAT_NV12格式可能仅从 Windows 8 开始。

输入缓冲区格式为NV_ENC_BUFFER_FORMAT_ARGB。这也应该是每个颜色通道 8 位。由于 alpha 值是互换的,我预计会出现错误的图片,但仍应对其进行编码。

到目前为止我的过程:

  • 创建 ID3D11Texture2D 渲染纹理
  • 创建 ID3D11Texture2D 暂存纹理
  • 使用 nvEncCreateInputBuffer 创建 NVENC 输入缓冲区
  • 使用 nvEncCreateBitstreamBuffer 创建 NVENC 输出缓冲区

:: 更新例程 ::

  • 渲染成渲染纹理
  • CopyResource() 到暂存纹理
  • 从暂存纹理填充 NVENC 输入缓冲区
  • 编码帧
  • 从 NVENC 输出缓冲区获取数据

如您所见,我每次更新只编码一帧。

到目前为止,一切正常,但如果我查看框架,图片大小是错误的。

在此处输入图像描述

那应该是 1280 x 720。如果我还尝试传递超过一帧,那么我会得到一个损坏的文件。

我知道在 H.264 中我的第一帧是 I 帧,接下来应该是 P 帧。

现在我的代码:

:: 创建输入缓冲区

NVENCSTATUS RenderManager::CreateInputBuffer()
{
    NVENCSTATUS nvStatus = NV_ENC_SUCCESS;

    NV_ENC_CREATE_INPUT_BUFFER createInputBufferParams = {};
    createInputBufferParams.version = NV_ENC_CREATE_INPUT_BUFFER_VER;
    createInputBufferParams.width = m_width;
    createInputBufferParams.height = m_height;
    createInputBufferParams.memoryHeap = NV_ENC_MEMORY_HEAP_SYSMEM_CACHED;
    createInputBufferParams.bufferFmt = NV_ENC_BUFFER_FORMAT_ARGB;

    nvStatus = m_pEncodeAPI->nvEncCreateInputBuffer(m_pEncoder, &createInputBufferParams);
    if (nvStatus != NV_ENC_SUCCESS) return nvStatus;

    m_pEncoderInputBuffer = createInputBufferParams.inputBuffer;

    return nvStatus;
}

:: 创建输出缓冲区

NVENCSTATUS RenderManager::CreateOutputBuffer()
{
    NVENCSTATUS nvStatus = NV_ENC_SUCCESS;

    NV_ENC_CREATE_BITSTREAM_BUFFER createBitstreamBufferParams = {};
    createBitstreamBufferParams.version = NV_ENC_CREATE_BITSTREAM_BUFFER_VER;
    createBitstreamBufferParams.size = 2 * 1024 * 1024;
    createBitstreamBufferParams.memoryHeap = NV_ENC_MEMORY_HEAP_SYSMEM_CACHED;

    nvStatus = m_pEncodeAPI->nvEncCreateBitstreamBuffer(m_pEncoder, &createBitstreamBufferParams);
    if (nvStatus != NV_ENC_SUCCESS) return nvStatus;

    m_pEncoderOutputBuffer = createBitstreamBufferParams.bitstreamBuffer;

    return nvStatus;
}

:: 填充输入缓冲区

NVENCSTATUS RenderManager::WriteInputBuffer(ID3D11Texture2D* pTexture)
{
    NVENCSTATUS nvStatus = NV_ENC_SUCCESS;
    HRESULT result = S_OK;

    // get data from staging texture
    D3D11_MAPPED_SUBRESOURCE mappedResource;
    result = m_pContext->Map(pTexture, 0, D3D11_MAP_READ, 0, &mappedResource);
    if (FAILED(result)) return NV_ENC_ERR_GENERIC;

    m_pContext->Unmap(pTexture, 0);

    // lock input buffer
    NV_ENC_LOCK_INPUT_BUFFER lockInputBufferParams = {};
    lockInputBufferParams.version = NV_ENC_LOCK_INPUT_BUFFER_VER;
    lockInputBufferParams.inputBuffer = m_pEncoderInputBuffer;

    nvStatus = m_pEncodeAPI->nvEncLockInputBuffer(m_pEncoder, &lockInputBufferParams);
    if (nvStatus != NV_ENC_SUCCESS) return nvStatus;

    unsigned int pitch = lockInputBufferParams.pitch;

    //ToDo: Convert R8G8B8A8 to A8R8G8B8

    // write into buffer
    memcpy(lockInputBufferParams.bufferDataPtr, mappedResource.pData, m_height * mappedResource.RowPitch);

    // unlock input buffer
    nvStatus = m_pEncodeAPI->nvEncUnlockInputBuffer(m_pEncoder, m_pEncoderInputBuffer);

    return nvStatus;
}

:: 编码帧

NVENCSTATUS RenderManager::EncodeFrame()
{
    NVENCSTATUS nvStatus = NV_ENC_SUCCESS;

    int8_t* qpDeltaMapArray = NULL;
    unsigned int qpDeltaMapArraySize = 0;

    NV_ENC_PIC_PARAMS encPicParams = {};
    encPicParams.version = NV_ENC_PIC_PARAMS_VER;
    encPicParams.inputWidth = m_width;
    encPicParams.inputHeight = m_height;
    encPicParams.inputBuffer = m_pEncoderInputBuffer;
    encPicParams.outputBitstream = m_pEncoderOutputBuffer;
    encPicParams.bufferFmt = NV_ENC_BUFFER_FORMAT_ARGB;
    encPicParams.pictureStruct = NV_ENC_PIC_STRUCT_FRAME;
    encPicParams.qpDeltaMap = qpDeltaMapArray;
    encPicParams.qpDeltaMapSize = qpDeltaMapArraySize;

    nvStatus = m_pEncodeAPI->nvEncEncodePicture(m_pEncoder, &encPicParams);

    return nvStatus;
}

:: 从输出缓冲区读取

NVENCSTATUS RenderManager::ReadOutputBuffer()
{
    NVENCSTATUS nvStatus = NV_ENC_SUCCESS;

    // lock output buffer
    NV_ENC_LOCK_BITSTREAM lockBitstreamBufferParams = {};
    lockBitstreamBufferParams.version = NV_ENC_LOCK_BITSTREAM_VER;
    lockBitstreamBufferParams.doNotWait = 0;
    lockBitstreamBufferParams.outputBitstream = m_pEncoderOutputBuffer;

    nvStatus = m_pEncodeAPI->nvEncLockBitstream(m_pEncoder, &lockBitstreamBufferParams);
    if (nvStatus != NV_ENC_SUCCESS) return nvStatus;

    void* pData = lockBitstreamBufferParams.bitstreamBufferPtr;
    unsigned int size = lockBitstreamBufferParams.bitstreamSizeInBytes;

    // read from buffer
    PlatformManager::SaveToFile("TEST", static_cast<char*>(pData), size);

    // unlock output buffer
    nvStatus = m_pEncodeAPI->nvEncUnlockBitstream(m_pEncoder, m_pEncoderOutputBuffer);

    return nvStatus;
}

m_width 和 m_height 始终为 1280 和 720。

为什么图像大小错误以及我如何用超过一帧填充输入缓冲区?

非常感谢您的帮助,

亚历山大

4

0 回答 0