Support shadow maps on a range of hardware

[This article is for Windows 8.x and Windows Phone 8.x developers writing Windows Runtime apps. If you’re developing for Windows 10, see the latest documentation]

Render higher-fidelity shadows on faster devices and faster shadows on less powerful devices. Part 4 of Walkthrough: Implement shadow volumes using depth buffers in Direct3D 11.

Comparison filter types

Only use linear filtering if the device can afford the performance penalty. Generally, Direct3D feature level 9_1 devices don't have enough power to spare for linear filtering on shadows. Use point filtering instead on these devices. When you use linear filtering, adjust the pixel shader so that it blends the shadow edges.

The ShadowMapping SDK sample creates point and linear comparison filters. The sample has a UI option to enable "shadow smoothing;" when this feature is enabled, linear filtering is used to reduce blockiness and the shader blends shadow edges for a more natural look. Try observing the difference on different types of devices and at different resolutions- or better yet, see how this looks in your game.

Create the comparison sampler for point filtering:

D3D11_SAMPLER_DESC comparisonSamplerDesc;
ZeroMemory(&comparisonSamplerDesc, sizeof(D3D11_SAMPLER_DESC));
comparisonSamplerDesc.AddressU = D3D11_TEXTURE_ADDRESS_BORDER;
comparisonSamplerDesc.AddressV = D3D11_TEXTURE_ADDRESS_BORDER;
comparisonSamplerDesc.AddressW = D3D11_TEXTURE_ADDRESS_BORDER;
comparisonSamplerDesc.BorderColor[0] = 1.0f;
comparisonSamplerDesc.BorderColor[1] = 1.0f;
comparisonSamplerDesc.BorderColor[2] = 1.0f;
comparisonSamplerDesc.BorderColor[3] = 1.0f;
comparisonSamplerDesc.MinLOD = 0.f;
comparisonSamplerDesc.MaxLOD = D3D11_FLOAT32_MAX;
comparisonSamplerDesc.MipLODBias = 0.f;
comparisonSamplerDesc.MaxAnisotropy = 0;
comparisonSamplerDesc.ComparisonFunc = D3D11_COMPARISON_LESS_EQUAL;
comparisonSamplerDesc.Filter = D3D11_FILTER_COMPARISON_MIN_MAG_MIP_POINT;

// Point filtered shadows can be faster, and may be a good choice when
// rendering on hardware with lower feature levels. This sample has a
// UI option to enable/disable filtering so you can see the difference
// in quality and speed.

DX::ThrowIfFailed(
    pD3DDevice->CreateSamplerState(
        &comparisonSamplerDesc,
        &m_comparisonSampler_point
        )
    );

Then create a sampler for linear filtering:

comparisonSamplerDesc.Filter = D3D11_FILTER_COMPARISON_MIN_MAG_MIP_LINEAR;
DX::ThrowIfFailed(
    pD3DDevice->CreateSamplerState(
        &comparisonSamplerDesc,
        &m_comparisonSampler_linear
        )
    );

Choose a sampler:

ID3D11PixelShader* pixelShader;
ID3D11SamplerState** comparisonSampler;
if (m_useLinear)
{
    pixelShader = m_shadowPixelShader_linear.Get();
    comparisonSampler = m_comparisonSampler_linear.GetAddressOf();
}
else
{
    pixelShader = m_shadowPixelShader_point.Get();
    comparisonSampler = m_comparisonSampler_point.GetAddressOf();
}

// Attach our pixel shader.
context->PSSetShader(
    pixelShader,
    nullptr,
    0
    );

context->PSSetSamplers(0, 1, comparisonSampler);
context->PSSetShaderResources(0, 1, m_shadowResourceView.GetAddressOf());

Blend shadow edges with linear filtering:

// Blends the shadow area into the lit area.
float3 light = lighting * (ambient + DplusS(N, L, NdotL, input.view));
float3 shadow = (1.0f - lighting) * ambient;
return float4(input.color * (light + shadow), 1.f);

Shadow buffer size

Larger shadow maps won't look as blocky but they take up more space in graphics memory. Experiment with different shadow map sizes in your game and observe the results in different types of devices and different display sizes. Consider an optimization like cascaded shadow maps to get better results with less graphics memory. See Common Techniques to Improve Shadow Depth Maps.

Shadow buffer depth

Greater precision in the shadow buffer will give more accurate depth test results, which helps avoid issues like z-buffer fighting. But like larger shadow maps, greater precision takes up more memory. Experiment with different depth precision types in your game - DXGI_FORMAT_R24G8_TYPELESS versus DXGI_FORMAT_R16_TYPELESS - and observe the speed and quality on different feature levels.

Optimizing precompiled shaders

Windows Store apps can use dynamic shader compilation, but it's faster to use dynamic shader linking. You can also use compiler directives and #ifdef blocks to create different versions of shaders. For example, the ShadowMapping SDK sample outputs two different versions of the pixel shader - one for point filtered shadows with hard edges, one for linear filtered shadows with blended edges. This is done by opening the Visual Studio project file in a text editor and adding multiple <FxcCompiler> entries for the HLSL (each with the appropriate preprocessor definitions). Note that this necessitates different filenames; in this case, Visual Studio appends _point and _linear to the different versions of the shader.

The project file entry for the linear filtered version of the shader defines LINEAR:

<FxCompile Include="Content\ShadowPixelShader.hlsl">
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">Pixel</ShaderType>
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|ARM'">Pixel</ShaderType>
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Pixel</ShaderType>
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Pixel</ShaderType>
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">Pixel</ShaderType>
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|x64'">Pixel</ShaderType>
  <DisableOptimizations Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">false</DisableOptimizations>
  <EnableDebuggingInformation Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">true</EnableDebuggingInformation>
  <DisableOptimizations Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">false</DisableOptimizations>
  <DisableOptimizations Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">false</DisableOptimizations>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">$(OutDir)%(Filename)_linear.cso</ObjectFileOutput>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|ARM'">$(OutDir)%(Filename)_linear.cso</ObjectFileOutput>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(OutDir)%(Filename)_linear.cso</ObjectFileOutput>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(OutDir)%(Filename)_linear.cso</ObjectFileOutput>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">$(OutDir)%(Filename)_linear.cso</ObjectFileOutput>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|x64'">$(OutDir)%(Filename)_linear.cso</ObjectFileOutput>
  <PreprocessorDefinitions Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">LINEAR</PreprocessorDefinitions>
  <PreprocessorDefinitions Condition="'$(Configuration)|$(Platform)'=='Release|ARM'">LINEAR</PreprocessorDefinitions>
  <PreprocessorDefinitions Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">LINEAR</PreprocessorDefinitions>
  <PreprocessorDefinitions Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">LINEAR</PreprocessorDefinitions>
  <PreprocessorDefinitions Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">LINEAR</PreprocessorDefinitions>
  <PreprocessorDefinitions Condition="'$(Configuration)|$(Platform)'=='Release|x64'">LINEAR</PreprocessorDefinitions>
</FxCompile>

The project file entry for the linear filtered version of the shader does not include preprocessor definitions:

<FxCompile Include="Content\ShadowPixelShader.hlsl">
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">Pixel</ShaderType>
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|ARM'">Pixel</ShaderType>
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Pixel</ShaderType>
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Pixel</ShaderType>
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">Pixel</ShaderType>
  <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|x64'">Pixel</ShaderType>
  <DisableOptimizations Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">false</DisableOptimizations>
  <EnableDebuggingInformation Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">true</EnableDebuggingInformation>
  <DisableOptimizations Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">false</DisableOptimizations>
  <DisableOptimizations Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">false</DisableOptimizations>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">$(OutDir)%(Filename)_point.cso</ObjectFileOutput>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|ARM'">$(OutDir)%(Filename)_point.cso</ObjectFileOutput>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(OutDir)%(Filename)_point.cso</ObjectFileOutput>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(OutDir)%(Filename)_point.cso</ObjectFileOutput>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">$(OutDir)%(Filename)_point.cso</ObjectFileOutput>
  <ObjectFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|x64'">$(OutDir)%(Filename)_point.cso</ObjectFileOutput>
</FxCompile>