I'm currently trying to add post processing to an existing C++ DirectX11 application. The task is to apply a dynamic FX shader loaded from a file directly to the backbuffer, all rendering is already done before. In short: I have to get a copy of the backbuffer, apply some post processing effects and overwrite the buffer with those afterwards. "IDXGISwapChain->Present" is called at last.

This was pretty easy to achieve in DirectX9, however I'm running into multiple issues with DirectX11 now. I only found one example for post processing on D3D11 on MSN and it was written for a Windows 8 app, which I cannot build or run on my Windows 7 desktop and the code did not make a lot of sense to me.

The first one, a missing effects framework, could be solved by using "Effects 11" you can download from MSN already.

Now I'm a bit confused on how to continue. The basic idea on how it could work is as following:

1. Load the effect from a ".fx" HLSL file and compile it using "D3DX11CompileEffectFromFile"
2. Get a copy of the backbuffer using "IDXGISwapChain->GetBuffer" and save it to a "ID3D11Texture" object
3. Get the description of the texture, to have width and height of the screen
4. Create a new render target and tell the device to use it
5. Create a fullscreen quad and draw the screen texture on it (using the loaded effect)
6. Reset the render target and update the backbuffer with the drawn quad

The following code for step one and two already works:

[spoiler]

1:

ID3DX11Effect *g_pEffect;

...

HRESULT hr;
ID3DBlob *m_pCompileBuffer;
ID3DInclude *m_pInclude = new CCustomInclude();

// Load effect file
hr = D3DX11CompileEffectFromFile(_T("shader.fx"), NULL, m_pInclude, NULL, NULL, device, &g_pEffect, &m_pCompileBuffer);
	
if (FAILED(hr))
{
	if (m_pCompileBuffer && m_pCompileBuffer->GetBufferSize() > 0)
	{
	    Log(1, "Error while compiling shader:\r\n\r\n%s\r\n", (char*)m_pCompileBuffer->GetBufferPointer());
	}
	else
	{
	    Log(1, "Error while loading shader");
	}
}

2 & 3:

ID3D11Texture2D *g_pScreenTexture;

...

hr = swapchain->GetBuffer(0, __uuidof(g_pScreenTexture), (PVOID*)&g_pScreenTexture);

if (SUCCEEDED(hr))
{
	D3D11_TEXTURE2D_DESC m_pDesc;
	g_pScreenTexture->GetDesc(&m_pDesc);
}

[/spoiler]

I'm having problems with applying the effect and drawing the fullscreen quad though ...

The vertex struct and layout for the quad is declared as following:

[spoiler]

// Vertex Structure
struct Vertex
{
	public:
		Vertex() { }
		Vertex(float x, float y, float z, float rhw) : pos(x, y, z, rhw) { }
		Vertex(float x, float y, float z, float rhw, float tex1, float tex2) : pos(x, y, z, rhw), tex(tex1, tex2) { }
		Vertex(float x, float y, float z, float rhw, D3DXVECTOR2 tex) : pos(x, y, z, rhw), tex(tex) { }
		Vertex(D3DXVECTOR4 pos, float tex1, float tex2) : pos(pos), tex(tex1, tex2) { }
		Vertex(D3DXVECTOR4 pos, D3DXVECTOR2 tex) : pos(pos), tex(tex) { }
		Vertex(D3DXVECTOR4 pos) : pos(pos) { }

		static const D3D11_INPUT_ELEMENT_DESC layout[];

	private:
		D3DXVECTOR4 pos;
		D3DXVECTOR2 tex;
};

// Vertex Layout
const D3D11_INPUT_ELEMENT_DESC Vertex::layout[] =
{
	{ "POSITION", 0, DXGI_FORMAT_R32G32B32_FLOAT, 0, 0, D3D11_INPUT_PER_VERTEX_DATA, 0 },
	{ "TEXCOORD", 0, DXGI_FORMAT_R32G32_FLOAT, 0, 16, D3D11_INPUT_PER_VERTEX_DATA, 0 }
};

[/spoiler]

The fullscreen quad is set up like this:

[spoiler]

ID3D11Buffer *g_pVertexBuffer;

...

// Create the vertex buffer
Vertex quad[] =
{
	Vertex(D3DXVECTOR4(-1.0f, 1.0f, 0.5f, 1.0f), D3DXVECTOR2(0.0f, 0.0f)),
	Vertex(D3DXVECTOR4(1.0f, 1.0f, 0.5f, 1.0f), D3DXVECTOR2(1.0f, 0.0f)),
	Vertex(D3DXVECTOR4(-1.0f, -1.0f, 0.5f, 1.0f), D3DXVECTOR2(0.0f, 1.0f)),
	Vertex(D3DXVECTOR4(1.0f, -1.0f, 0.5f, 1.0f), D3DXVECTOR2(1.0f, 1.0f))
};

D3D11_BUFFER_DESC m_pVertexDesc = { sizeof(Vertex) * ARRAYSIZE(quad), D3D11_USAGE_DEFAULT, D3D11_BIND_VERTEX_BUFFER, 0, 0 };
D3D11_SUBRESOURCE_DATA m_pVertexData = { quad, 0, 0 }; 
device->CreateBuffer(&m_pVertexDesc, &m_pVertexData, &g_pVertexBuffer);

[/spoiler]

Drawing currently looks similar to the next code:

[spoiler]

// Update vertex declaration
UINT m_iStrides = sizeof(Vertex);
UINT m_iOffsets = 0;
context->IASetVertexBuffers(0, 1, &g_pVertexBuffer, &m_iStrides, &m_iOffsets);
context->IASetPrimitiveTopology(D3D10_PRIMITIVE_TOPOLOGY_TRIANGLESTRIP);

// Begin drawing
D3DX11_TECHNIQUE_DESC m_pTechDesc;
g_pEffect->GetTechniqueByIndex(0)->GetDesc(&m_pTechDesc);

for (UINT iPass = 0; iPass < m_pTechDesc.Passes; iPass++)
{
	g_pEffect->GetTechniqueByIndex(0)->GetPassByIndex(iPass)->Apply(NULL, context);
	context->Draw(4, 0);
}

[/spoiler]

I'm clearing the render target to the color red before drawing the quad, but the screen is red then only, so sadly it doesn't seem to draw the quad at all...

I don't know if there is any way to check that.

Also, the a post processing shader normally requires me to pass the screen texture to it, so it can apply the effects on every pixel of it. The Effects Framework provides the "ID3DX11Effect->GetVariableByName" and allows one to set the variable to a specific object / data, but it has no definition for a texture. The nearest thing I found is "Variable->AsShaderResource()->SetResource(&MyResource)", but is that really the most efficient way, to create a shader resource for the screen texture to pass it to the pixel shader?

I'm sorry for the tons of code in this post, but I found it the easiest way to show what I got already. In DirectX9 the steps described earlier worked without any problems, but it wasn't required to create a vertex buffer here anyway, so the whole thing was shorter and easier to achieve.

I hope somebody has done something similar / post processing before in D3D11/D3D10 and can help me out here, I would really appreciate it.

Thank you and cheers,

Crosire

We should actually make that a big red blinking sticky in this subforum

The nearest thing I found is "Variable->AsShaderResource()->SetResource(&MyResource)", but is that really the most efficient way, to create a shader resource for the screen texture to pass it to the pixel shader ?

Thank you for your answers already! I now set the position layout to a float4, using "DXGI_FORMAT_R32G32B32A32_FLOAT", makes a bit more sense.

I'm used to PS 2.0, so the new HLSL format is a bit strange to me and I'm sure there is a mistake in there.

The shader is just a simple monochrome testing effect:

[spoiler]

Texture2D colorTex : register(t0);
SamplerState _sampler { Filter = MIN_MAG_MIP_POINT; AddressU = Clamp; AddressV = Clamp; };

float4 PostProcess_PS(float3 normal : NORMAL, float2 coord : TEXCOORD0) : SV_TARGET
{
    float4 result = colorTex.Sample(_sampler, coord);
    result.rgb = dot(float3(0.18, 0.41, 0.41), result.rgb);
    result.a = 1.0;
    return result;
}

technique10 PostProcess
{
    pass p0
    {
        SetPixelShader(CompileShader(ps_4_0, PostProcess_PS()));
    }
}

[/spoiler]

Thanks so far, you helped me a lot. The quad is now drawing fine and the shader / effect gets loaded and executed too.

FX:

[spoiler]

Texture2D colorTex;
SamplerState colorSampler { AddressU = Clamp; AddressV = Clamp; };

struct VS_INPUT
{
    float4 pos : POSITION;
    float2 txcoord : TEXCOORD0;
};
struct VS_OUTPUT
{
    float4 pos : SV_POSITION;
    float2 txcoord : TEXCOORD0;
};

VS_OUTPUT PostProcess_VS(VS_INPUT IN)
{
    VS_OUTPUT OUT;
	
    OUT.pos = IN.pos;
    OUT.txcoord = IN.txcoord;
	
    return OUT;
}

float4 PostProcess_PS(VS_OUTPUT IN) : SV_TARGET
{
    float4 color;

    //color = colorTex.Sample(colorSampler, IN.txcoord);
    //color.rgb = dot(float3(0.18, 0.41, 0.41), color.rgb);
    color.r = 1.0;
    color.g = 0.0;
    color.b = 0.0;
    color.a = 1.0;

    return color;
}

technique11 PostProcess
{
    pass p0
    {
        SetPixelShader(CompileShader(ps_4_0, PostProcess_PS()));
        SetVertexShader(CompileShader(vs_4_0, PostProcess_VS()));
    }
}

[/spoiler]

The full screen gets covered in red as expected now. The only thing left is to pass the screen texture to the pixel shader and let it do its job.

However, as soon as I remove the comments from the two lines above and comment out the other color code, I just get a black screen. I tried some different shader code and it's always just black, lowering the alpha will show the original image a bit through the now transparent quad, as long as I disable render target clearing.

I'm sending the texture to the shader with this code:

ID3D11ShaderResourceView *m_pResource;
D3D11_SHADER_RESOURCE_VIEW_DESC m_pResourceDesc = { DXGI_FORMAT_R32_FLOAT, D3D10_SRV_DIMENSION_TEXTURE2D };
m_pResourceDesc.Texture2D.MipLevels = 1;
m_pResourceDesc.Texture2D.MostDetailedMip = 0;
device->CreateShaderResourceView(g_pScreenTexture11, &m_pResourceDesc, &m_pResource);

g_pEffect11->GetVariableByName("colorTex")->AsShaderResource()->SetResource(m_pResource);

"g_pScreenTexture11" contains the full screen, checked that via "D3DX11SaveTextureToFile". "g_pEffect11" also links to the effect, the other shader code works perfectly fine. The shader layout is the same as the vertex input layout declared in the C++ code too (as seen earlier), so I'm a bit lost here.

Thanks again for all help so far, it's greatly appreciated. I'm really close to the target now ... Just this little thing.

Cheers,

Crosire

• Check if the texcoords by outputting float4(IN.txcoord, 0, 1). This should give a black-red gradient from left-to-right and a black-green gradient from top to bottom (and yellow bottom right).
• Your sampler is a bit bare for my taste. Really set all the needed values (not sure what the effects use per default). Or set the sampler explicitly to NULL (you should find the corresponding variable type), this will give you the default sampler. If that gives you something useful, work from those values. For a postprocess effect you e.g. want point sampling.

Texcoords are set correctly, I get the expected green/red gradient image with no blue and alpha at 100%.

I tried it by using "D3DX11CreateShaderResourceViewFromFile" and sent an existing image to the shader by that and I get that image with the effects applied drawn on the screen, so the shader is fully working. It's just not reading the data from the texture object (which has it, tested) and using that one when sending it to the shader.

In code:

This works:

ID3D11ShaderResourceView *m_pResource;
D3DX11CreateShaderResourceViewFromFile(device, L"tex.bmp", NULL, NULL, &m_pResource, NULL);
g_pEffect11->GetVariableByName("colorTex")->AsShaderResource()->SetResource(m_pResource);

The code from my previous post not:

ID3D11ShaderResourceView *m_pResource;
D3D11_SHADER_RESOURCE_VIEW_DESC m_pResourceDesc = { DXGI_FORMAT_R32_FLOAT, D3D10_SRV_DIMENSION_TEXTURE2D };
m_pResourceDesc.Texture2D.MipLevels = 1;
m_pResourceDesc.Texture2D.MostDetailedMip = 0;

device->CreateShaderResourceView(g_pScreenTexture11, &m_pResourceDesc, &m_pResource);

g_pEffect11->GetVariableByName("colorTex")->AsShaderResource()->SetResource(m_pResource);

After looking at it again, I found some mistakes already:

• D3D11_SRV_DIMENSION_TEXTURE2D instead of D3D10_SRV_DIMENSION_TEXTURE2D (which outputs the same value in the end (4), but it just makes more sense)
• DXGI_FORMAT_R32G32B32A32_FLOAT instead of DXGI_FORMAT_R32_FLOAT (The image isn't made of red color only obviously, I want the full RGBA color code here)

Still it doesn't work with the corrected code. Does anybody see another error in here?

And I just have to thank unbird again, you pushed me in the right directions :)

I think it might be because the screen texture gets created by the swapchain interface in "IDXGISwapChain->GetBuffer" as I pass a null texture interface object to it. It then probably gets initialized without "D3D11_BIND_SHADER_RESOURCE", which makes it unusable for the shader resource view object.

I tried a different approach now:

Code which gets executed on startup:

// Create screen resource
D3D11_TEXTURE2D_DESC m_pTextureDesc = { m_pDesc.Width, m_pDesc.Height, 1, 1, DXGI_FORMAT_R32G32B32A32_FLOAT, 1, 0, D3D11_USAGE_DEFAULT, D3D11_BIND_SHADER_RESOURCE, 0, 0 };
device->CreateTexture2D(&m_pTextureDesc, NULL, &g_pScreenTexture11);

// Create shader resource view
D3D11_SHADER_RESOURCE_VIEW_DESC m_pResourceDesc = { m_pTextureDesc.Format, D3D11_SRV_DIMENSION_TEXTURE2D };
device->CreateShaderResourceView(g_pScreenTexture11, &m_pResourceDesc, &g_pScreenView11);

Code which gets executed every frame:

ID3D11Resource *m_pBackBuffer;

hr = swapchain->GetBuffer(0, __uuidof(m_pBackBuffer), (LPVOID*)&m_pBackBuffer);

if (SUCCEEDED(hr) && g_pScreenTexture11)
{
	// Update screen texture
	context->CopyResource(g_pScreenTexture11, m_pBackBuffer);

	// Update effect parameters
	g_pEffect11->GetVariableByName("colorTex")->AsShaderResource()->SetResource(g_pScreenView11);

	...
}

I thought the shader view interface only stores a pointer to the screen texture, so it will retrieve the updated data, when I change the texture object it was initalized with.

The code now fails at "context->CopyResource(g_pScreenTexture11, m_pBackBuffer);" though. The texture is just empty.

I tried this one too:

ID3D11Resource *tex;
g_pScreenSurface11->GetResource(&tex);
D3DX11SaveTextureToFile(context, g_pScreenTexture11, D3DX11_IFF_BMP, L"tex.bmp");

But it crashes the application at the second line already, I have no idea why.

I'm going to try and put together a quick testing program as I cannot bind PIX to any other software with my project applied. PIX tries to hook the DirectX Exported Functions which was already done by my hook, so it just crashes.