Tiny optimizations and preparation for Schussler et. al 2017

Author: devshgraphicsprogramming

examples_tests/22.RaytracedAO/closestHit.comp

@@ -69,16 +69,17 @@ void main()
 			const uvec3 indices = get_triangle_indices(batchInstanceData,triangleID);
 
 			// positions
-			vec3 geomDenormal;
-			const vec3 lastVxPos = load_positions(geomDenormal,batchInstanceData,indices);
-			const bool frontfacing = bool((batchInstanceData.determinantSignBit^floatBitsToUint(dot(geomDenormal,normalizedV)))&0x80000000u);
+			vec3 geomNormal;
+			const vec3 lastVxPos = load_positions(geomNormal,batchInstanceData,indices);
+
+			const bool frontfacing = bool((batchInstanceData.determinantSignBit^floatBitsToUint(dot(normalizedV,geomNormal)))&0x80000000u);
 
 			// get material
 			const nbl_glsl_MC_oriented_material_t material = nbl_glsl_MC_material_data_t_getOriented(batchInstanceData.material,frontfacing);
 			contrib.color = contrib.albedo = nbl_glsl_MC_oriented_material_t_getEmissive(material);
 
 			const uint pathDepth = bitfieldExtract(staticViewData.pathDepth_noRussianRouletteDepth_samplesPerPixelPerDispatch,0,8);
-			const bool _continue = vertex_depth!=pathDepth && ray.maxT==nbl_glsl_FLT_MAX; // not last vertex and not NEE path
+			const bool _continue = vertex_depth!=pathDepth && material.genchoice_count!=0u && ray.maxT==nbl_glsl_FLT_MAX; // not last vertex and has a BxDF and not NEE path
 			if (_continue)
 			{
 				// if we ever support spatially varying emissive, we'll need to hoist barycentric computation and UV fetching to the position fetching
@@ -89,7 +90,7 @@ void main()
 
 				//
 				normalizedN = load_normal_and_prefetch_textures(
-					batchInstanceData,indices,compactBary,geomDenormal,material
+					batchInstanceData,indices,compactBary,geomNormal,material
 #ifdef TEX_PREFETCH_STREAM
 					,mat2(0.0) // TODO: Covariance Rendering
 #endif
@@ -103,7 +104,7 @@ void main()
 				);
 			}
 			else
-				contrib.worldspaceNormal = normalize(geomDenormal)*nbl_glsl_MC_colorToScalar(contrib.albedo);
+				contrib.worldspaceNormal = geomNormal*nbl_glsl_MC_colorToScalar(contrib.albedo);
 		}
 		else
 			Contribution_initMiss(contrib,aovThroughputScale);

examples_tests/22.RaytracedAO/raygen.comp

@@ -79,32 +79,38 @@ void main()
 			// get material while waiting for indices
 			const nbl_glsl_MC_oriented_material_t material = nbl_glsl_MC_material_data_t_getOriented(batchInstanceData.material,frontfacing);
 			contrib.color = contrib.albedo = nbl_glsl_MC_oriented_material_t_getEmissive(material);
-
+			
 			// load vertex data
-			vec3 geomDenormal;
-			const vec3 lastVxPos = load_positions(geomDenormal,batchInstanceData,indices);
-						
-			// get initial scramble key while waiting for vertex positions
-			const nbl_glsl_xoroshiro64star_state_t scramble_start_state = texelFetch(scramblebuf,ivec2(outPixelLocation),0).rg;
+			vec3 geomNormal;
+			const vec3 lastVxPos = load_positions(geomNormal,batchInstanceData,indices);
 
-			//
-			normalizedN = load_normal_and_prefetch_textures(
-				batchInstanceData,indices,compactBary,geomDenormal,material
-				#ifdef TEX_PREFETCH_STREAM
-				,dBarydScreen
-				#endif
-			);
+			// little optimization for non-twosided materials
+			if (material.genchoice_count!=0u)
+			{			
+				// get initial scramble key while waiting for vertex positions
+				const nbl_glsl_xoroshiro64star_state_t scramble_start_state = texelFetch(scramblebuf,ivec2(outPixelLocation),0).rg;
+
+				//
+				normalizedN = load_normal_and_prefetch_textures(
+					batchInstanceData,indices,compactBary,geomNormal,material
+					#ifdef TEX_PREFETCH_STREAM
+					,dBarydScreen
+					#endif
+				);
 
-			const vec3 origin = dPdBary*compactBary+lastVxPos;
-			normalizedV = normalize(pc.cummon.camPos-origin);
+				const vec3 origin = dPdBary*compactBary+lastVxPos;
+				normalizedV = normalize(pc.cummon.camPos-origin);
 
-			// generate rays
-			const uint vertex_depth = 1u;
-			generate_next_rays(
-				samplesPerPixelPerDispatch,material,frontfacing,vertex_depth,
-				scramble_start_state,pc.cummon.samplesComputed,outPixelLocation,origin,
-				vec3(pc.cummon.rcpFramesDispatched),1.f,contrib.albedo,contrib.worldspaceNormal
-			);
+				// generate rays
+				const uint vertex_depth = 1u;
+				generate_next_rays(
+					samplesPerPixelPerDispatch,material,frontfacing,vertex_depth,
+					scramble_start_state,pc.cummon.samplesComputed,outPixelLocation,origin,
+					vec3(pc.cummon.rcpFramesDispatched),1.f,contrib.albedo,contrib.worldspaceNormal
+				);
+			}
+			else
+				contrib.worldspaceNormal = geomNormal*nbl_glsl_MC_colorToScalar(contrib.albedo);
 		}
 		else
 			Contribution_initMiss(contrib,1.f);

examples_tests/22.RaytracedAO/raytraceCommon.glsl

@@ -154,7 +154,7 @@ bool has_world_transform(in nbl_glsl_ext_Mitsuba_Loader_instance_data_t batchIns
 
 #include <nbl/builtin/glsl/barycentric/utils.glsl>
 mat2x3 dPdBary;
-vec3 load_positions(out vec3 geomDenormal, in nbl_glsl_ext_Mitsuba_Loader_instance_data_t batchInstanceData, in uvec3 indices)
+vec3 load_positions(out vec3 geomNormal, in nbl_glsl_ext_Mitsuba_Loader_instance_data_t batchInstanceData, in uvec3 indices)
 {
 	mat3 positions = mat3(
 		nbl_glsl_fetchVtxPos(indices[0],batchInstanceData),
@@ -167,7 +167,7 @@ vec3 load_positions(out vec3 geomDenormal, in nbl_glsl_ext_Mitsuba_Loader_instan
 	//
 	for (int i=0; i<2; i++)
 		dPdBary[i] = positions[i]-positions[2];
-	geomDenormal = cross(dPdBary[0],dPdBary[1]);
+	geomNormal = normalize(cross(dPdBary[0],dPdBary[1]));
 	//
 	if (tform)
 		positions[2] += batchInstanceData.tform[3];
@@ -196,7 +196,7 @@ bool needs_texture_prefetch(in nbl_glsl_ext_Mitsuba_Loader_instance_data_t batch
 
 vec3 load_normal_and_prefetch_textures(
 	in nbl_glsl_ext_Mitsuba_Loader_instance_data_t batchInstanceData,
-	in uvec3 indices, in vec2 compactBary, in vec3 geomDenormal,
+	in uvec3 indices, in vec2 compactBary, in vec3 geomNormal,
 	in nbl_glsl_MC_oriented_material_t material
 #ifdef TEX_PREFETCH_STREAM
 	,in mat2 dBarydScreen
@@ -224,7 +224,6 @@ vec3 load_normal_and_prefetch_textures(
 	// the rest is always only needed for continuing rays
 
 
-	vec3 normal = geomDenormal;
 	// while waiting for the scramble state
 	// TODO: optimize, add loads more flags to control this
 	const bool needsSmoothNormals = true;
@@ -237,18 +236,20 @@ vec3 load_normal_and_prefetch_textures(
 		);
 
 		// not needed for NEE unless doing Area or Projected Solid Angle Sampling
-		const vec3 smoothNormal = normals*nbl_glsl_barycentric_expand(compactBary);
-		// TODO: first check wouldn't be needed if we had `needsSmoothNormals` implemented
-		if (!isnan(smoothNormal.x) && has_world_transform(batchInstanceData))
+		vec3 smoothNormal = normals*nbl_glsl_barycentric_expand(compactBary);
+		if (has_world_transform(batchInstanceData))
 		{
-			normal = vec3(
+			smoothNormal = vec3(
 				dot(batchInstanceData.normalMatrixRow0,smoothNormal),
 				dot(batchInstanceData.normalMatrixRow1,smoothNormal),
 				dot(batchInstanceData.normalMatrixRow2,smoothNormal)
 			);
 		}
+		// TODO: this check wouldn't be needed if we had `needsSmoothNormals` implemented
+		if (!isnan(smoothNormal.x))
+			return normalize(smoothNormal);
 	}
-	return normalize(normal);
+	return geomNormal;
 }
 
 #include <nbl/builtin/glsl/sampling/quantized_sequence.glsl>

include/nbl/builtin/glsl/material_compiler/common_invariant_declarations.glsl

@@ -37,6 +37,7 @@ struct nbl_glsl_MC_instr_stream_t
 // (in case of precomp.NdotV<0.0, currInteraction will be set with -precomp.N)
 struct nbl_glsl_MC_precomputed_t
 {
+	// TODO: shadingN and geomN
 	vec3 N;
 	vec3 V;
 	bool frontface;
@@ -82,6 +83,7 @@ void nbl_glsl_MC_finalizeMicrofacet(inout nbl_glsl_MC_microfacet_t mf)
 struct nbl_glsl_MC_oriented_material_t
 {
 	uvec2 emissive;
+	// TODO: derive/define upper bounds for instruction counts and bitpack them!
 	uint prefetch_offset;
 	uint prefetch_count;
 	uint instr_offset;