I am using SharpDX to basically render browser (chromium) output buffer on directX process.
Process is relatively simple, I intercept CEF buffer (by overriding OnPaint method) and write that to a texture2D.
Code is relatively simple:
Texture creation:
public void BuildTextureWrap() {
var oldTexture = texture;
texture = new D3D11.Texture2D(DxHandler.Device, new D3D11.Texture2DDescription() {
Width = overlay.Size.Width,
Height = overlay.Size.Height,
MipLevels = 1,
ArraySize = 1,
Format = DXGI.Format.B8G8R8A8_UNorm,
SampleDescription = new DXGI.SampleDescription(1, 0),
Usage = D3D11.ResourceUsage.Default,
BindFlags = D3D11.BindFlags.ShaderResource,
CpuAccessFlags = D3D11.CpuAccessFlags.None,
OptionFlags = D3D11.ResourceOptionFlags.None,
});
var view = new D3D11.ShaderResourceView(
DxHandler.Device,
texture,
new D3D11.ShaderResourceViewDescription {
Format = texture.Description.Format,
Dimension = D3D.ShaderResourceViewDimension.Texture2D,
Texture2D = { MipLevels = texture.Description.MipLevels },
}
);
textureWrap = new D3DTextureWrap(view, texture.Description.Width, texture.Description.Height);
if (oldTexture != null) {
obsoleteTextures.Add(oldTexture);
}
}
That piece of code is executed at start and when resize is happening.
Now when CEF OnDraw I basically copy their buffer to texture:
var destinationRegion = new D3D11.ResourceRegion {
Top = Math.Min(r.dirtyRect.y, texDesc.Height),
Bottom = Math.Min(r.dirtyRect.y + r.dirtyRect.height, texDesc.Height),
Left = Math.Min(r.dirtyRect.x, texDesc.Width),
Right = Math.Min(r.dirtyRect.x + r.dirtyRect.width, texDesc.Width),
Front = 0,
Back = 1,
};
// Draw to the target
var context = targetTexture.Device.ImmediateContext;
context.UpdateSubresource(targetTexture, 0, destinationRegion, sourceRegionPtr, rowPitch, depthPitch);
There are some more code out there but basically this is only relevant piece. Whole thing works until OnDraw happens frequently.
Apparently if I force CEF to Paint frequently, whole host process dies.
This is happening at UpdateSubresource.
So my question is, is there another, safer way to do this? (Update texture frequently)
Solution to this problem was relatively simple yet not so obvious at the beginning.
I simply moved the code responsible for updating texture inside render loop and just keep internal buffer pointer cached.