I've heard from multiple sources that the best way to crank out performance while rendering 2D text to the screen is to roll out your own font class. The font class should hold a cache of text being drawn to the screen, and only create new resources when the specified text does not already exist within the cache. This makes perfect sense to me.

As for the implementation, I figured I would create the font & text using GDI (or possibly GDI+), then slap it onto a texture in D3D, freeing GDI resources as I go along. I've also read that this is basically how ID3DXFont does the job. My question, however, is how would you get the image from a GDI device context onto a ID3D10ShaderResourceView texture? And would it be better to use GDI+ instead of GDI?

Thank you for your thoughts.

So just as a general test of ideas, I tried this:


bool TextClass::Initialize(ID3D10Device* device)
{
int bmpWidth = 64;
int bmpHeight = 16;
LPCWSTR strText = TEXT("Test");
RECT rcText = { 0, 0, bmpWidth, bmpHeight };
COLORREF hbrColor = RGB(255, 0, 0);
HDC hDC = CreateCompatibleDC(NULL);
DWORD* pSrcData = 0;
BITMAPINFO bmi = { sizeof( BITMAPINFOHEADER ), bmpWidth, bmpHeight, 1, 32, BI_RGB, 0, 0, 0, 0, 0 };
HBITMAP hTempBmp = CreateDIBSection( hDC, &bmi, DIB_RGB_COLORS, (void**)&pSrcData, NULL, 0 );
HFONT NewFont = CreateFont(0, 0, 0, 0, FW_DONTCARE, FALSE, FALSE, FALSE, DEFAULT_CHARSET, OUT_OUTLINE_PRECIS, CLIP_DEFAULT_PRECIS, CLEARTYPE_QUALITY, VARIABLE_PITCH, TEXT("Impact"));
HBRUSH NewBrush = CreateSolidBrush(hbrColor);
SelectObject(hDC, NewFont);
SelectObject(hDC, NewBrush);
DrawText(hDC, strText, 4, &rcText, DT_LEFT | DT_WORDBREAK);
GdiFlush();
DeleteObject(NewBrush);
DeleteObject(NewFont);
ReleaseDC(NULL, hDC);
D3DX10_IMAGE_LOAD_INFO info;
info.Width = (rcText.right - rcText.left);
info.Height = (rcText.bottom - rcText.top);
info.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
info.Usage = D3D10_USAGE_DEFAULT;
info.MipLevels = 1;
info.Depth = 1;
info.CpuAccessFlags = 0;
info.BindFlags = D3D10_BIND_SHADER_RESOURCE;
info.MiscFlags = 0;
info.Filter = 0;
info.FirstMipLevel = 1;
info.MipFilter = 0;
info.pSrcInfo = 0;
if(FAILED( D3DX10CreateTextureFromMemory( device, pSrcData, (4 * info.Width) * info.Height, &info, NULL, &m_pTexture, NULL) ))
{
return false;
}
D3D10_SHADER_RESOURCE_VIEW_DESC srvDesc;
SecureZeroMemory(&srvDesc, sizeof(D3D10_SHADER_RESOURCE_VIEW_DESC));
m_pTexture2D = (ID3D10Texture2D*)m_pTexture;
srvDesc.Format = info.Format;
srvDesc.ViewDimension = D3D10_SRV_DIMENSION_TEXTURE2D;
srvDesc.Texture2D.MostDetailedMip = info.MipLevels;
if(FAILED( device->CreateShaderResourceView( m_pTexture, &srvDesc, &m_pSRView ) ))
{
return false;
}
return true;
}

... But it always fails on D3DX10CreateTextureFromMemory. I don't know if this is a compatibility issue or if I'm just going about this improperly, but AFAIK it looks like should work.

I'm using D3D10, so if it were possible to use those suggestions then I'd imagine it being a rather painful rewriting process.

For that, just create a Texture2D and pass the image data through the pInitialData parameter of CreateTexture2D.

Thanks! I was able to get the fonts generated, cached, and even added into a resource view. When I save the bitmap data out to a file, it comes out precisely as expected.
But when I try to render the resource view, I see nothing being rendered. Is there a simple way to output the texture data to a file from a resource?
Just trying to make sure the data is actually copying over properly before continuing on. If it saves as expected, then it's probably just my matrix/shader setup.

Thanks! It appears to have copied over properly, except the image is rendered upside-down so I assume I'll need to reverse the pixel order in the GDI image before copying, unless there's a more efficient way to handle this.

EDIT: In the BITMAPINFOHEADER, when you set biHeight to be negative, it reverses the image along the y-axis.
All is good, thanks again!