Terrain shader: Fix incorrect tangent normals sampled using seamless UV method

[wheatleymf]

Issue

Currently, terrain shader samples tangent normal maps using seamless UVs 'as is', which means that normals are not transformed to be inline with the new UV rotation angle. This results in some issues with NdotL shading. Here's a comparison between current, and corrected normals:

Fix

The solution for this is to multiply tangent normal with 2x2 rotation matrix that we can access from Terrain_SampleSeamless2D in TerrainCommon.hlsl. This is probably very stupid, but I had to add an output out float2x2 uvAngle so I can access this matrix from other shaders:

float2 Terrain_SampleSeamlessUV( float2 uv, out float2x2 uvAngle )

To avoid breaking any shaders that rely on a previous implementation of this function, I've added an overload that has an old set of arguments:

float2 Terrain_SampleSeamlessUV( float2 uv ) 
{
    float2x2 dummy;
    return Terrain_SampleSeamlessUV( uv, dummy ); 
}

In terrain.shader, inside the Terrain_Splat4 function, tangent normal is transformed like this:

<...>
float3 normal = ComputeNormalFromRGTexture( nho.rg );
normal.xz *= g_TerrainMaterials[ i ].normalstrength;
            
// Rotate normal map to match seamless UV angle and avoid shading errors 
normal.xy = mul( uvRotation, normal.xy );

normal = normalize( normal );

albedos[i] = SrgbGammaToLinear( bcr.rgb );
<...>

I am not entirely sure if that's a correct way to approach this, and I absolutely don't know if it was a good idea to change seamless UV function like I did, so please let me know if my solution is stupid.

[sampavlovic]

sexy hunk

[lolleko]

Could update your branch, please 🙏

[wheatleymf]

Could update your branch, please 🙏

Should be done now

[sboxbot]

This PR has been merged upstream.