Merge branch 'MatInstanceViewer' of https://github.com/Areloch/Torque3D into Preview4_0

Fixed up forward IBL
Reimplemented old forward light shader code as a baseline so it can be updated to new PBR math
This commit is contained in:
Areloch 2019-06-27 00:36:56 -05:00
commit e5bbd12d4b
19 changed files with 696 additions and 127 deletions

View file

@ -88,9 +88,121 @@ void compute4Lights( float3 wsView,
out float4 outDiffuse,
out float4 outSpecular )
{
// NOTE: The light positions and spotlight directions
// are stored in SoA order, so inLightPos[0] is the
// x coord for all 4 lights... inLightPos[1] is y... etc.
//
// This is the key to fully utilizing the vector units and
// saving a huge amount of instructions.
//
// For example this change saved more than 10 instructions
// over a simple for loop for each light.
int i;
outDiffuse = float4(0,0,0,0);
outSpecular = float4(0,0,0,0);
float4 lightVectors[3];
for ( i = 0; i < 3; i++ )
lightVectors[i] = wsPosition[i] - inLightPos[i];
float4 squareDists = 0;
for ( i = 0; i < 3; i++ )
squareDists += lightVectors[i] * lightVectors[i];
// Accumulate the dot product between the light
// vector and the normal.
//
// The normal is negated because it faces away from
// the surface and the light faces towards the
// surface... this keeps us from needing to flip
// the light vector direction which complicates
// the spot light calculations.
//
// We normalize the result a little later.
//
float4 nDotL = 0;
for ( i = 0; i < 3; i++ )
nDotL += lightVectors[i] * -wsNormal[i];
float4 rDotL = 0;
#ifndef TORQUE_BL_NOSPECULAR
// We're using the Phong specular reflection model
// here where traditionally Torque has used Blinn-Phong
// which has proven to be more accurate to real materials.
//
// We do so because its cheaper as do not need to
// calculate the half angle for all 4 lights.
//
// Advanced Lighting still uses Blinn-Phong, but the
// specular reconstruction it does looks fairly similar
// to this.
//
float3 R = reflect( wsView, -wsNormal );
for ( i = 0; i < 3; i++ )
rDotL += lightVectors[i] * R[i];
#endif
// Normalize the dots.
//
// Notice we're using the half type here to get a
// much faster sqrt via the rsq_pp instruction at
// the loss of some precision.
//
// Unless we have some extremely large point lights
// i don't believe the precision loss will matter.
//
half4 correction = (half4)rsqrt( squareDists );
nDotL = saturate( nDotL * correction );
rDotL = clamp( rDotL * correction, 0.00001, 1.0 );
// First calculate a simple point light linear
// attenuation factor.
//
// If this is a directional light the inverse
// radius should be greater than the distance
// causing the attenuation to have no affect.
//
float4 atten = saturate( 1.0 - ( squareDists * inLightInvRadiusSq ) );
#ifndef TORQUE_BL_NOSPOTLIGHT
// The spotlight attenuation factor. This is really
// fast for what it does... 6 instructions for 4 spots.
float4 spotAtten = 0;
for ( i = 0; i < 3; i++ )
spotAtten += lightVectors[i] * inLightSpotDir[i];
float4 cosAngle = ( spotAtten * correction ) - inLightSpotAngle;
atten *= saturate( cosAngle * inLightSpotFalloff );
#endif
// Finally apply the shadow masking on the attenuation.
atten *= shadowMask;
// Get the final light intensity.
float4 intensity = nDotL * atten;
// Combine the light colors for output.
outDiffuse = 0;
for ( i = 0; i < 4; i++ )
outDiffuse += intensity[i] * inLightColor[i];
// Output the specular power.
float4 specularIntensity = pow( rDotL, float4(1,1,1,1) ) * atten;
// Apply the per-light specular attenuation.
float4 specular = float4(0,0,0,1);
for ( i = 0; i < 4; i++ )
specular += float4( inLightColor[i].rgb * inLightColor[i].a * specularIntensity[i], 1 );
// Add the final specular intensity values together
// using a single dot product operation then get the
// final specular lighting color.
outSpecular = float4(1,1,1,1) * specular;
}
struct Surface
@ -159,7 +271,7 @@ inline Surface createForwardSurface(float4 baseColor, float3 normal, float4 pbrP
surface.V = normalize(wsEyePos - surface.P);
surface.baseColor = baseColor;
const float minRoughness=1e-4;
surface.roughness = clamp(1.0 - pbrProperties.b, minRoughness, 1); //t3d uses smoothness, so we convert to roughness.
surface.roughness = clamp(1.0 - pbrProperties.b, minRoughness, 1.0); //t3d uses smoothness, so we convert to roughness.
surface.roughness_brdf = surface.roughness * surface.roughness;
surface.metalness = pbrProperties.a;
surface.ao = pbrProperties.g;
@ -302,13 +414,13 @@ float3 boxProject(float3 wsPosition, float3 wsReflectVec, float4x4 worldToObj, f
}
float4 computeForwardProbes(Surface surface,
float cubeMips, float numProbes, float4x4 worldToObjArray[MAX_FORWARD_PROBES], float4 probeConfigData[MAX_FORWARD_PROBES],
float cubeMips, int numProbes, float4x4 worldToObjArray[MAX_FORWARD_PROBES], float4 probeConfigData[MAX_FORWARD_PROBES],
float4 inProbePosArray[MAX_FORWARD_PROBES], float4 bbMinArray[MAX_FORWARD_PROBES], float4 bbMaxArray[MAX_FORWARD_PROBES], float4 inRefPosArray[MAX_FORWARD_PROBES],
float hasSkylight, TORQUE_SAMPLER2D(BRDFTexture),
TORQUE_SAMPLERCUBE(skylightIrradMap), TORQUE_SAMPLERCUBE(skylightSpecularMap),
TORQUE_SAMPLERCUBEARRAY(irradianceCubemapAR), TORQUE_SAMPLERCUBEARRAY(specularCubemapAR))
float skylightCubemapIdx, TORQUE_SAMPLER2D(BRDFTexture),
TORQUE_SAMPLERCUBEARRAY(irradianceCubemapAR), TORQUE_SAMPLERCUBEARRAY(specularCubemapAR))
{
int i = 0;
int i = 0;
float alpha = 1;
float blendFactor[MAX_FORWARD_PROBES];
float blendSum = 0;
float blendFacSum = 0;
@ -332,6 +444,8 @@ float4 computeForwardProbes(Surface surface,
if (contribution[i] > 0.0)
probehits++;
}
else
continue;
contribution[i] = max(contribution[i], 0);
@ -360,8 +474,43 @@ float4 computeForwardProbes(Surface surface,
{
blendFactor[i] *= invBlendSumWeighted;
contribution[i] *= blendFactor[i];
alpha -= contribution[i];
}
}
else
alpha -= blendSum;
#if DEBUGVIZ_ATTENUATION == 1
float contribAlpha = 1;
for (i = 0; i < numProbes; ++i)
{
contribAlpha -= contribution[i];
}
return float4(1 - contribAlpha, 1 - contribAlpha, 1 - contribAlpha, 1);
#endif
#if DEBUGVIZ_CONTRIB == 1
float3 probeContribColors[4];
probeContribColors[0] = float3(1,0,0);
probeContribColors[1] = float3(0,1,0);
probeContribColors[2] = float3(0,0,1);
probeContribColors[3] = float3(1,1,0);
float3 finalContribColor = float3(0, 0, 0);
float contribAlpha = 1;
for (i = 0; i < numProbes; ++i)
{
finalContribColor += contribution[i] *probeContribColors[i].rgb;
contribAlpha -= contribution[i];
}
//Skylight coloration for anything not covered by probes above
if(skylightCubemapIdx != -1)
finalContribColor += float3(0.3, 0.3, 0.3) * contribAlpha;
return float4(finalContribColor, 1);
#endif
float3 irradiance = float3(0, 0, 0);
float3 specular = float3(0, 0, 0);
@ -369,7 +518,6 @@ float4 computeForwardProbes(Surface surface,
// Radiance (Specular)
float lod = surface.roughness*cubeMips;
float alpha = 1;
for (i = 0; i < numProbes; ++i)
{
float contrib = contribution[i];
@ -384,10 +532,10 @@ float4 computeForwardProbes(Surface surface,
}
}
if (hasSkylight && alpha > 0.001)
if(skylightCubemapIdx != -1 && alpha >= 0.001)
{
irradiance += TORQUE_TEXCUBELOD(skylightIrradMap, float4(surface.R, 0)).xyz;
specular = TORQUE_TEXCUBELOD(skylightSpecularMap, float4(surface.R, lod)).xyz;
irradiance += TORQUE_TEXCUBEARRAYLOD(irradianceCubemapAR, surface.R, skylightCubemapIdx, 0).xyz * alpha;
specular += TORQUE_TEXCUBEARRAYLOD(specularCubemapAR, surface.R, skylightCubemapIdx, lod).xyz * alpha;
}
float3 F = FresnelSchlickRoughness(surface.NdotV, surface.f0, surface.roughness);
@ -403,7 +551,7 @@ float4 computeForwardProbes(Surface surface,
//final diffuse color
float3 diffuse = kD * irradiance * surface.baseColor.rgb;
float4 finalColor = float4(diffuse + specular, 1);
//finalColor.rgb += abs(surface.N);
float4 finalColor = float4(diffuse + specular * surface.ao, 1.0);
return finalColor;
}