add a better RWMC=OFF mode, don't try to reuse RWMC code with 2 weird cascades

Author: devsh

22.RaytracedAO/Renderer.cpp

@@ -1240,7 +1240,8 @@ void Renderer::initScreenSizedResources(
 	const float RGB19E7_MaxLuma = std::exp2(63.f);
 	if (cascadeCount<MinCascades) // rwmc OFF, store everything to cascade 0
 	{
-		cascadeCount = MinCascades;
+		// original idea was to create 2 cascades where the first starts so low that every sample gets added to it. But now we just do 1
+		cascadeCount = 0;
 		cascadeLuminanceBase = std::exp2(16.f); // just some constant to space the cascades apart
 		cascadeLuminanceStart = RGB19E7_MaxLuma;
 		std::cout << "Re-Weighting Monte Carlo = DISABLED" << std::endl;
@@ -1249,7 +1250,7 @@ void Renderer::initScreenSizedResources(
 	{
 		cascadeCount = core::min(cascadeCount,MaxCascades);
 		const float cascadeSegmentCount = cascadeCount-1;
-
+		// base is the power increment between each successive cascade, first cascade starts at Emin or 1/cascadeLuminanceBase^segmentCount scaled to max emitter radiance
 		const bool baseIsKnown = cascadeLuminanceBase>std::numeric_limits<float>::min();
 		if (core::isnan<float>(cascadeLuminanceStart))
 			cascadeLuminanceStart = baseIsKnown ? (maxEmitterRadianceLuma*std::pow(cascadeLuminanceBase,-cascadeSegmentCount)):Emin;

22.RaytracedAO/closestHit.comp

@@ -118,20 +118,25 @@ void main()
 		contrib.color *= throughput;
 		const vec3 aovThroughput = throughput*aovThroughputScale;
 		//
-		const bool pathToBeContinued = bool(rayMask);
-		if (pathToBeContinued)
-			addAccumulation(contrib.color,accumulationLocation);
-		else
+		if (isRWMCEnabled())
 		{
-			// need whole path throughput when splatting
-			contrib.color += fetchAccumulation(accumulationLocation);
-			const nbl_glsl_RWMC_SplattingParameters splat = nbl_glsl_RWMC_getCascade(staticViewData.cascadeParams,nbl_glsl_MC_colorToScalar(contrib.color)/pc.cummon.rcpFramesDispatched);
-			for (uint j=0u; j<2u; j++)
-				addAccumulationCascade(
-					contrib.color*splat.cascadeWeights[j],accumulationLocation,
-					samplesPerPixelPerDispatch,splat.lowerCascade+j
-				);
+			const bool pathToBeContinued = bool(rayMask);
+			if (pathToBeContinued)
+				addAccumulation(contrib.color,accumulationLocation);
+			else
+			{
+				// need whole path throughput when splatting
+				contrib.color += fetchAccumulation(accumulationLocation);
+				const nbl_glsl_RWMC_SplattingParameters splat = nbl_glsl_RWMC_getCascade(staticViewData.cascadeParams,nbl_glsl_MC_colorToScalar(contrib.color)/pc.cummon.rcpFramesDispatched);
+				for (uint j=0u; j<2u; j++)
+					addAccumulationCascade(
+						contrib.color*splat.cascadeWeights[j],accumulationLocation,
+						samplesPerPixelPerDispatch,splat.lowerCascade+j
+					);
+			}
 		}
+		else
+			addAccumulation(contrib.color,accumulationLocation);
 		//
 		addAlbedo(contrib.albedo*aovThroughput,accumulationLocation);
 		addWorldspaceNormal(contrib.worldspaceNormal*nbl_glsl_MC_colorToScalar(aovThroughput),accumulationLocation);

22.RaytracedAO/raygen.comp

@@ -130,38 +130,52 @@ void main()
 			{
 				const uvec3 coord = uvec3(outPixelLocation,i);
 
-				nbl_glsl_RWMC_SplattingParameters splat = nbl_glsl_RWMC_getCascade(staticViewData.cascadeParams,luma);
-				const bool pathToBeContinued = bool((rayMask>>i)&0x1u);
-				if (pathToBeContinued)
+				if (isRWMCEnabled())
 				{
-					storeAccumulation(contrib.color*pc.cummon.rcpFramesDispatched,coord);
-					splat.cascadeWeights = vec2(0.f,0.f);
-				}
-				const uint higherCascade = splat.lowerCascade+1u;
-				const uint cascadeCount = staticViewData.cascadeParams.penultimateCascadeIx+2u;
-				for (uint cascadeIx=0u; cascadeIx<cascadeCount; cascadeIx++)
-				{
-					float weight = 0.f;
-					if (cascadeIx==splat.lowerCascade)
-						weight = splat.cascadeWeights[0];
-					else if (cascadeIx==higherCascade)
-						weight = splat.cascadeWeights[1];
-					nextSampleAccumulationCascade(
-						contrib.color*weight,coord,samplesPerPixelPerDispatch,
-						cascadeIx,pc.cummon.rcpFramesDispatched
-					);
+					nbl_glsl_RWMC_SplattingParameters splat = nbl_glsl_RWMC_getCascade(staticViewData.cascadeParams,luma);
+
+					// ray out of ray-payload, using first layer of accumulation texture to store emissive along path so far
+					const bool pathToBeContinued = bool((rayMask>>i)&0x1u);
+					if (pathToBeContinued)
+					{
+						storeAccumulation(contrib.color*pc.cummon.rcpFramesDispatched,coord);
+						splat.cascadeWeights = vec2(0.f,0.f);
+					}
+
+					const uint higherCascade = splat.lowerCascade+1u;
+					const uint cascadeCount = staticViewData.cascadeParams.penultimateCascadeIx+2u;
+					for (uint cascadeIx=0u; cascadeIx<cascadeCount; cascadeIx++)
+					{
+						float weight = 0.f;
+						if (cascadeIx==splat.lowerCascade)
+							weight = splat.cascadeWeights[0];
+						else if (cascadeIx==higherCascade)
+							weight = splat.cascadeWeights[1];
+						nextSampleAccumulationCascade(
+							contrib.color*weight,coord,samplesPerPixelPerDispatch,
+							cascadeIx,pc.cummon.rcpFramesDispatched
+						);
+					}
 				}
 
 				const bool hideEnvmap = bool(staticViewData.sampleSequenceStride_hideEnvmap>>31);
 				// clear accumulations totally if beginning a new frame
 				if (firstFrame)
 				{
+					storeAccumulation(contrib.color,coord);
 					storeAlbedo(contrib.albedo,coord);
 					storeWorldspaceNormal(contrib.worldspaceNormal,coord);
 					storeMask(hideEnvmap&&(!hit) ? 1.f:0.f,coord);
 				}
 				else
 				{
+					if (!isRWMCEnabled())
+					{
+						const vec3 prev = fetchAccumulation(coord);
+						const vec3 delta = (contrib.color-prev)*pc.cummon.rcpFramesDispatched;
+						if (any(greaterThan(delta,vec3(exp2(-19.f)))))
+							storeAccumulation(prev+delta,coord);
+					}
 					addAlbedo(contrib.albedo,coord,pc.cummon.rcpFramesDispatched);
 					addWorldspaceNormal(contrib.worldspaceNormal,coord,pc.cummon.rcpFramesDispatched);
 					if (hideEnvmap)

22.RaytracedAO/raytraceCommon.glsl

@@ -74,6 +74,11 @@ uvec3 get_triangle_indices(in nbl_glsl_ext_Mitsuba_Loader_instance_data_t batchI
 #include <nbl/builtin/glsl/format/decode.glsl>
 #include <nbl/builtin/glsl/format/encode.glsl>
 
+bool isRWMCEnabled()
+{
+	return staticViewData.cascadeParams.penultimateCascadeIx!=uint(-2);
+}
+
 vec3 fetchAccumulation(in uvec3 coord)
 {
 	const uvec2 data = imageLoad(accumulation,ivec3(coord)).rg;

22.RaytracedAO/resolve.comp

@@ -61,6 +61,10 @@ vec3 nbl_glsl_RWMC_sampleCascadeTexel(ivec2 coord, in ivec2 offset, in uint casc
 	return value/float(samplesPerPixelPerDispatch);
 }
 
+bool isRWMCEnabled()
+{
+	return staticViewData.cascadeParams.penultimateCascadeIx!=uint(-2);
+}
 
 void main()
 {
@@ -69,7 +73,12 @@ void main()
 	{
 		samplesPerPixelPerDispatch = bitfieldExtract(staticViewData.maxPathDepth_noRussianRouletteDepth_samplesPerPixelPerDispatch,16,16);
 
-		vec3 acc = nbl_glsl_RWMC_reweight(pc.rwmcReweightingParams,pixelCoord);
+		vec3 acc;
+		if (isRWMCEnabled())
+			acc = nbl_glsl_RWMC_reweight(pc.rwmcReweightingParams,pixelCoord);
+		else // its a pretty ok function, reusing it
+			acc = nbl_glsl_RWMC_sampleCascadeTexel(pixelCoord,ivec2(0,0),-1);
+
 		vec3 alb = texelFetch(albedoSamples,ivec3(pixelCoord,0),0).rgb;
 		vec3 nml = nbl_glsl_decodeRGB10A2_SNORM(texelFetch(normalSamples,ivec3(pixelCoord,0),0).r).xyz;
 		float msk = texelFetch(maskSamples,ivec3(pixelCoord,0),0).r;