Point light shadows work wrong, how can I debug it?

[SC5Shout]

I've got a problem with my point light shadows.

I've seen several articles, videos, questions:

- https://learnopengl.com/Advanced-Lighting/Shadows/Point-Shadows
- https://developer.nvidia.com/gpugem...ows/chapter-12-omnidirectional-shadow-mapping
- https://ogldev.org/www/tutorial43/tutorial43.html
-

- and others...

and I wrote this (pseudo) code:

```cpp
struct PointShadowMatrix
{
glm::mat4 viewProjection[6];
};
const glm::mat4 shadowProj = glm:erspective(Math::HALF_PI, 1.0f, 25.0f, 1.0f);

const glm::vec3 lightPos = light.getPosition();

PointShadowMatrix shadowData;
shadowData.viewProjection[0] = shadowProj * glm::lookAt(lightPos, lightPos + glm::vec3(1.0, 0.0, 0.0), glm::vec3(0.0, 1.0, 0.0));
shadowData.viewProjection[1] = shadowProj * glm::lookAt(lightPos, lightPos + glm::vec3(-1.0, 0.0, 0.0), glm::vec3(0.0, 1.0, 0.0));
shadowData.viewProjection[2] = shadowProj * glm::lookAt(lightPos, lightPos + glm::vec3(0.0, 1.0, 0.0), glm::vec3(0.0, 0.0, -1.0));
shadowData.viewProjection[3] = shadowProj * glm::lookAt(lightPos, lightPos + glm::vec3(0.0, -1.0, 0.0), glm::vec3(0.0, 0.0, 1.0));
shadowData.viewProjection[4] = shadowProj * glm::lookAt(lightPos, lightPos + glm::vec3(0.0, 0.0, 1.0), glm::vec3(0.0, 1.0, 0.0));
shadowData.viewProjection[5] = shadowProj * glm::lookAt(lightPos, lightPos + glm::vec3(0.0, 0.0, -1.0), glm::vec3(0.0, 1.0, 0.0));

shadowConstantBuffer.AddData(shadowData);

auto textureCube = TextureCube(
.Width = 1024,
.Height = 1024,
.MipLevels = 1,
.Format = DXGI_FORMAT_R24G8_TYPELESS,
.BindFlags = D3D11_BIND_DEPTH_STENCIL | D3D11_BIND_SHADER_RESOURCE
);

auto cubeDepthRenderTarget = CubeRenderTarget(
.resource = textureCube,
.FirstArraySlice = 0,
.MipSlice = 0
);

auto shader = Shader(.path = "PointShadowShader.hlsl");

shader.Bind();
cubeDepthRenderTarget.Bind();
shadowConstantBuffer.Bind();

SetViewport(.w = 1024, .h = 1024);
ClearDepth(cubeDepthRenderTarget, 0.0f);

DrawDepth();
```

PointShadowShader:

```hlsl
struct VertexInput
{
float3 position : POSITION;

nointerpolation float4 modelToWorld1 : I_MODEL_TO_WORLD_ONE;
nointerpolation float4 modelToWorld2 : I_MODEL_TO_WORLD_TWO;
nointerpolation float4 modelToWorld3 : I_MODEL_TO_WORLD_THREE;
nointerpolation float4 modelToWorld4 : I_MODEL_TO_WORLD_FOUR;
};

float4 VSMain(VertexInput input) : SV_POSITION
{
float4x4 model = float4x4(input.modelToWorld1, input.modelToWorld2, input.modelToWorld3, input.modelToWorld4);
return mul(float4(input.position, 1.0f), model);
}

cbuffer PointShadowData : register(b4)
{
row_major float4x4 pointShadowViewProjection[6];
};

struct GS_OUTPUT
{
float4 position : SV_POSITION;
float3 worldPosition : WORLD_POSITION;
uint RTIndex : SV_RenderTargetArrayIndex;
};

[maxvertexcount(18)]
void GSMain(triangle float4 position[3] : SV_POSITION, inout TriangleStream<GS_OUTPUT> OutStream)
{
for (uint face = 0; face < 6; ++face)
{
GS_OUTPUT output;

output.RTIndex = face;

for (uint v = 0; v < 3; ++v)
{
output.worldPosition = position[v].xyz;
output.position = mul(position[v], pointShadowViewProjection[face]);
OutStream.Append(output);
}
OutStream.RestartStrip();
}
}

cbuffer LightInfo : register(b1)
{
float3 lightPos;
};

struct PixelInput
{
float4 position : SV_POSITION;
float3 worldPosition : WORLD_POSITION;
};

float PSMain(PixelInput input) : SV_DEPTH
{
//I tried input.position instead of input.worldPosition, it didn't fixed the problem
//I tried NVidia's approach to calculate squared distance instead of length
//I tried using just a length, without dividing by far plane
//Nothing fixed the problem
return length(input.worldPosition - lightPos) / 25.0f;
}
```

The depth pass looks good to me (light space perspective):

but the shading version does not:


Another screenshot, now from behind the wall:

the way I use the depth map to calculate shadows:

```hlsl
float PointLightShadows(float3 worldPosition, float3 lightPosition)
{
float3 fragToLight = worldPosition - lightPosition;
float distance = length(fragToLight);

return t_pointShadowMap.SampleCmpLevelZero(g_shadowBorder, float4(fragToLight, lightIndex), distance / 25.0f);}
```

As I wrote before, I've seen all these articles and I tried every approach authors show there, but nothing worked.

The depth buffer is cleared to `0`, depth function is `GREATER`

what is wrong with my code? How can I fix it? What things should I consider checking while debugging this problem?

 

[SC5Shout]

I've cleared the depth buffer to 1.0f, set depthFunc to LESS, and sampler ComparisonFunction to LESS and it works now. My question is why? What was wrong with the reversed depth version?

 

[JoeJ]

Are you blind?
Everyone can see the bug.
But here, i give you a hint: