Question

背景

使用DXVA2解码4K 30FPS视频，然后将解码后的帧直接传递给opengl进行渲染。
框架是GPU解码（DXVA2），然后是GPU渲染（Opengl），无需复制回CPU。

问题：

我在一些电脑上发现glTexImage2D耗时太长（140ms）。一台PC的GPU是intel HD 5500 + Nivida GeFoce 940M。
但是在某些PC上它只需要16ms。一台PC的GPU是intel HD 5500 + AMD R7 M260。
所有PC都更新了最新的gpu驱动程序。

源码：

以下是DXVA2解码源代码。

static int dxva2_retrieve_data(AVCodecContext *s, AVFrame *frame)
{
    LPDIRECT3DSURFACE9 surface =  (LPDIRECT3DSURFACE9)frame->data[3];
    InputStream        *ist = (InputStream *)s->opaque;
    DXVA2Context       *ctx = (DXVA2Context*)ist->hwaccel_ctx;
    D3DSURFACE_DESC    surfaceDesc;
    D3DLOCKED_RECT     LockedRect;
    HRESULT            hr;
    int                ret;

    IDirect3DSurface9_GetDesc(surface, &surfaceDesc);

    hr = IDirect3DSurface9_LockRect(surface, &LockedRect, NULL, D3DLOCK_READONLY);
    if (FAILED(hr)) {
        av_log(NULL, AV_LOG_ERROR, "Unable to lock DXVA2 surface\n");
        return AVERROR_UNKNOWN;
    }

    picture->data[0] = (uint8_t*)LockedRect.pBits;
    picture->data[1] = (uint8_t*)LockedRect.pBits + LockedRect.Pitch * surfaceDesc.Height;
    frameWidth = frame->width;
    frameHeight = frame->height;
    //glFinish() <---- commentee suggestion
    DWORD start = GetTickCount();//test time
    Draw();//opengl draw        
    //glFinish() <---- commentee suggestion
    DWORD end = GetTickCount();//test time
    printf("draw time:%d\n", end - start);//test time
    start = end - start;
    picture->data[0] = nullptr;
    picture->data[1] = nullptr;

    IDirect3DSurface9_UnlockRect(surface);
    return 0;
fail:
    return ret;
}

下面是Draw函数的代码。

void Draw()
{
    glClearColor(1.0f, 1.0f, 1.0f, 1.0f);
    glClearDepth(1.0f);
    glClear(GL_COLOR_BUFFER_BIT);
    glShadeModel(GL_SMOOTH);

    glClearColor(0.0, 255, 0.0, 0.0);
    glClear(GL_COLOR_BUFFER_BIT);

    glUseProgram(g_program);
    glActiveTexture(GL_TEXTURE0);
    glBindTexture(GL_TEXTURE_2D, g_texture[0]);
    glTexImage2D(GL_TEXTURE_2D, 0, GL_RED, frameWidth, frameHeight, 0, GL_RED, GL_UNSIGNED_BYTE, picture->data[0]);
    glUniform1i(y, 0);

    glActiveTexture(GL_TEXTURE1);
    glBindTexture(GL_TEXTURE_2D, g_texture[1]);
    glTexImage2D(GL_TEXTURE_2D, 0, GL_LUMINANCE_ALPHA, frameWidth / 2, frameHeight / 2, 0, GL_LUMINANCE_ALPHA, GL_UNSIGNED_BYTE, picture->data[1]);
    glUniform1i(uv, 1);

    DrawHorizontal();

    glBindVertexArray(0);
    glUseProgram(0);
    glBindTexture(GL_TEXTURE_2D, 0);
}

问题：

谁能告诉我为什么相同的代码在不同的PC上有不同的结果？或者给我一些建议？

glTexImage2D在某些PC上需要很长时间

0 个答案: