D3D 9 Rendering

[Gaius Julius Caesar]

Hello! First of all id like to express my joy towards this library. libtheoraplayer was the best and free solution to my opengl project - it's fast, it works and very easy to use. However, i started to port my project to d3d and encountered slight difficulty with video playback. Frankly speaking, im total newbie in dx environment, so the question is: how do you pass texture data via frame loop?

frame=clip->getNextFrame();
if (frame)
{...

It would very nice to have full direct x example included into examples section.

Thank you for your attention!

[Krešimir Špes]

Hi Caesar,

check this thread for more info: https://sourceforge.net/p/libtheoraplayer/forum/general/thread/9d59e9a8/

Also, for a general texture writing tutorial, search google for "DirectX procedural textures".

[David]

here is some basic code from my engine that might help (dx9).

load/play clip and create dynamic texture...

//-----------------------------------------------------
//Theora
pTexture2DVideo[(DWORD)Texture2DVideo].pClip = gTheora_mgr->createVideoClip((char*)pTexture2DVideo[(DWORD)Texture2DVideo].FilePath.c_str(),
TH_RGB,
0,
bUsePowerOf2);
pTexture2DVideo[(DWORD)Texture2DVideo].pClip->play();

pTexture2DVideo[(DWORD)Texture2DVideo].w = pTexture2DVideo[(DWORD)Texture2DVideo].pClip->getWidth();
pTexture2DVideo[(DWORD)Texture2DVideo].h = pTexture2DVideo[(DWORD)Texture2DVideo].pClip->getHeight();

if(ContinuosPlay)
{
pTexture2DVideo[(DWORD)Texture2DVideo].pClip->setAutoRestart(true);
}
else
{
pTexture2DVideo[(DWORD)Texture2DVideo].pClip->setAutoRestart(false);
}

//create direct x texture
HRESULT hr;
hr = gD3D_Device->CreateTexture(pTexture2DVideo[(DWORD)Texture2DVideo].w,
pTexture2DVideo[(DWORD)Texture2DVideo].h,
1,
D3DUSAGE_DYNAMIC,
D3DFMT_A8R8G8B8,
D3DPOOL_DEFAULT,
&pTexture2DVideo[(DWORD)Texture2DVideo].pTexture,
NULL);

draw video to texture...

bool FrameIsActive = false;
TheoraVideoFrame *frame = pTexture2DVideo[Index].pClip->getNextFrame();
if (frame)
{
FrameIsActive = true;

BYTE  *pTexBuffer; // Bitmap buffer, texture buffer
LONG  lTexPitch;                // Pitch of bitmap, texture

BYTE * pTextureBytes = NULL;
UINT row, col;
// Lock the Texture surface
D3DLOCKED_RECT d3dlr;
pTexture2DVideo[Index].pTexture_Surf->LockRect( &d3dlr, 0, D3DLOCK_DISCARD);

// Get the texture buffer & pitch
pTexBuffer = static_cast<byte *>(d3dlr.pBits);
lTexPitch = d3dlr.Pitch;
pTexBuffer = pTexBuffer;

uint8_t* mb = (uint8_t*)frame->getBuffer();

for( row = 0; row < (UINT)pTexture2DVideo[Index].pClip->getHeight(); row++)
{
    pTextureBytes = pTexBuffer;

    for( col = 0; col < (UINT)pTexture2DVideo[Index].pClip->getWidth(); col ++ )
    {
        //copying from RGB to RGBA(== D3DFMT_A8R8G8B8 of texture)
        pTextureBytes[2] = mb[0];//r
        pTextureBytes[1] = mb[1];//g
        pTextureBytes[0] = mb[2];//b
        //make black alpha if needed...
        if(!pTexture2DVideo[Index].bAlphaEnable)
        {
            pTextureBytes[3] = 0xff;//a
        }
        else
        {
            pTextureBytes[3] = mb[3];//a
        }

        //Texture uses 4 bytes (=32bit Format D3DFMT_A8R8G8B8)
        pTextureBytes +=4;
        //Bitmap(Video) uses 3 bytes (=24bit MEDIASUBTYPE_RGB24 )
        mb += 3;
    }
    pTexBuffer += lTexPitch;
}
//unlock the texture surface
pTexture2DVideo[Index].pTexture_Surf->UnlockRect();

// be sure to pop the frame from the frame queue when you're done
pTexture2DVideo[Index].pClip->popFrame();
}

[Krešimir Špes]

Hi David,

from the looks of it, your code seems to be ok.

I'd just change the way you're copying data to speed things up. You're doing a double for loop and copying each pixel individually which is very slow. What you should be doing is just do a memcpy() from the theoravideoframe to the texture buffer. If you chose the texture format correctly when creating a video clip, the library will already prepare the frames in the format you need. This is better because pixel conversion is best done in the theora worker threads in a hardware accelerated way then you doing it in the main thread unoptimised.

[Konstanty Kalicki]

You won't be able to memcpy whole buffer, because DX textures have padded rows. However even using memcpy to copy whole rows will speed things significantly. Here is a snippet from my code hastly eddited to fit your naming convention:

for(s32 row =0;row<last_frame->getHeight();row++){
memcpy( &bits[row*lTexPitch], &mb[rowlast_frame->getHeight()4 ], last_frame->getHeight()*4 );
}

Edit: this code will work with RGBA texture and video, you will have to ensure both buffer and dx tex are in same format

[Krešimir Špes]

oh yeah, forgot about that part of directx,buy yeah, a series of line memcpy's should be good.