geometric normals and position derivatives need redoing

Author: devshgraphicsprogramming

examples_tests/22.RaytracedAO/Renderer.cpp

@@ -622,6 +622,8 @@ core::smart_refctd_ptr<IGPUImageView> Renderer::createScreenSizedTexture(E_FORMA
 	return m_driver->createGPUImageView(std::move(viewparams));
 }
 
+constexpr uint16_t m_maxDepth = 2u;
+constexpr uint16_t m_UNUSED_russianRouletteDepth = 5u;
 bool extractIntegratorInfo(const ext::MitsubaLoader::CElementIntegrator& integrator, uint32_t &bxdfSamples, uint32_t &maxNEESamples)
 {
 	using Enum = ext::MitsubaLoader::CElementIntegrator::Type;
@@ -673,7 +675,8 @@ void Renderer::init(const SAssetBundle& meshes,	core::smart_refctd_ptr<ICPUBuffe
 	core::smart_refctd_ptr<IGPUImageView> depthBuffer,visibilityBuffer;
 	// set up Descriptor Sets
 	{
-		// captures m_globalBackendDataDS, creates m_indirectDrawBuffers, sets up m_mdiDrawCalls ranges, creates m_additionalGlobalDS and m_cullDS, sets m_cullPushConstants and m_cullWorkgroups, creates m_perCameraRasterDS
+		// captures m_globalBackendDataDS, creates m_indirectDrawBuffers, sets up m_mdiDrawCalls ranges
+		// creates m_additionalGlobalDS and m_cullDS, sets m_cullPushConstants and m_cullWorkgroups, creates m_perCameraRasterDS
 		auto initData = initSceneObjects(meshes);
 		{
 			initSceneNonAreaLights(initData);
@@ -722,7 +725,7 @@ void Renderer::init(const SAssetBundle& meshes,	core::smart_refctd_ptr<ICPUBuffe
 			// see how much we can bump the sample count per raster pass
 			{
 				uint32_t sampleMultiplier = 0u;
-				const auto maxSSBOSize = core::min(m_driver->getMaxSSBOSize(),1024u<<20);
+				const auto maxSSBOSize = core::min(m_driver->getMaxSSBOSize(),256u<<20);
 				while (raygenBufferSize<=maxSSBOSize && intersectionBufferSize<=maxSSBOSize) // for AMD && m_maxRaysPerDispatch*WORKGROUP_SIZE<=64<<10))
 					setRayBufferSizes(++sampleMultiplier);
 				if (sampleMultiplier==1u)
@@ -774,6 +777,7 @@ void Renderer::init(const SAssetBundle& meshes,	core::smart_refctd_ptr<ICPUBuffe
 			(std::ofstream("runtime_defines.glsl")
 				<< "#define _NBL_EXT_MITSUBA_LOADER_VT_STORAGE_VIEW_COUNT " << initData.globalMeta->m_global.getVTStorageViewCount() << "\n"
 				<< initData.globalMeta->m_global.m_materialCompilerGLSL_declarations
+				<< "#define MAX_PATH_DEPTH " << m_maxDepth << "\n"
 				<< "#ifndef MAX_RAYS_GENERATED\n"
 				<< "#	define MAX_RAYS_GENERATED " << m_staticViewData.samplesPerPixelPerDispatch << "\n"
 				<< "#endif\n"
@@ -1135,8 +1139,6 @@ void Renderer::deinit()
 // one day it will just work like that
 //#include <nbl/builtin/glsl/sampling/box_muller_transform.glsl>
 
-constexpr uint16_t m_maxDepth = 2u; // TODO: = 5u
-constexpr uint16_t m_UNUSED_russianRouletteDepth = 5u;
 void Renderer::render(nbl::ITimer* timer)
 {
 	if (m_cullPushConstants.maxGlobalInstanceCount==0u)

examples_tests/22.RaytracedAO/closestHit.comp

@@ -68,23 +68,22 @@ void main()
 			const nbl_glsl_MC_oriented_material_t material = nbl_glsl_MC_material_data_t_getOriented(InstData.data[batchInstanceGUID].material,frontfacing);
 			emissive = nbl_glsl_MC_oriented_material_t_getEmissive(material);
 
-			const uint MAX_PATH_DEPTH = 2u; // TODO: move to push constant or UBO
-			const bool _continue = vertex_depth!=MAX_PATH_DEPTH && ray.maxT==FLT_INF; // last vertex or was a NEE path
+			const bool _continue = vertex_depth!=MAX_PATH_DEPTH && ray.maxT==FLT_MAX; // last vertex or was a 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
 				const vec2 compactBary = intersection.uvwt.xy;
 
 				const mat2 dBarydScreen = mat2(0.0); // TODO: Covariance Rendering
 				const nbl_glsl_xoroshiro64star_state_t scramble_start_state = load_aux_vertex_attrs(
-					compactBary,indices,batchInstanceGUID,material,dBarydScreen,outPixelLocation,vertex_depth_mod_2
+					compactBary,indices,batchInstanceGUID,material,dBarydScreen,outPixelLocation,vertex_depth
 				);
 
 				const vec3 hitWorldPos = dPdBary*compactBary.xy+last_vx_pos;
 
 				generate_next_rays(
-					MAX_RAYS_GENERATED,material,frontfacing,vertex_depth_mod_2,
-					scramble_start_state,sampleID,outPixelLocation,hitWorldPos,normalize(geomNormal),throughput
+					MAX_RAYS_GENERATED,material,frontfacing,vertex_depth,
+					scramble_start_state,sampleID,outPixelLocation,hitWorldPos,geomNormal,throughput
 				);
 			}
 		}

examples_tests/22.RaytracedAO/main.cpp

@@ -213,11 +213,6 @@ int main()
 
 
 	auto driver = device->getVideoDriver();
-	// temporary workaround for Renderdoc v 1.14 bug
-	device->run();
-	driver->beginScene(false, false);
-	driver->endScene();
-	glFinish();
 
 
 	core::smart_refctd_ptr<Renderer> renderer = core::make_smart_refctd_ptr<Renderer>(driver,device->getAssetManager(),smgr);

examples_tests/22.RaytracedAO/raygen.comp

@@ -74,9 +74,10 @@ void main()
 
 			// generate rays
 			const uint sampleID = bitfieldExtract(pc.cummon.samplesComputed_depth,0,16);
-			const vec3 geomNormal = normalize(cross(dPdBary[0],dPdBary[1]));
+			const vec3 geomNormal = cross(dPdBary[0],dPdBary[1]);
+			const uint vertex_depth = 1u;
 			generate_next_rays(
-				staticViewData.samplesPerPixelPerDispatch,material,frontfacing,vertex_depth_mod_2,
+				staticViewData.samplesPerPixelPerDispatch,material,frontfacing,vertex_depth,
 				scramble_start_state,sampleID,outPixelLocation,hitWorldPos,geomNormal,vec3(1.0)
 			);
 		}

examples_tests/22.RaytracedAO/raytraceCommon.glsl

@@ -146,22 +146,22 @@ mat2 nbl_glsl_perturbNormal_dUVdSomething()
 #include <nbl/builtin/glsl/material_compiler/common.glsl>
 
 
-vec3 rand3d(inout nbl_glsl_xoroshiro64star_state_t scramble_state, in uint _sample)
+vec3 rand3d(inout nbl_glsl_xoroshiro64star_state_t scramble_state, in int _sample, in int depth)
 {
-	uvec3 seqVal = texelFetch(sampleSequence,int(_sample)).xyz; // TODO: add depth param!!!
+	uvec3 seqVal = texelFetch(sampleSequence,int(_sample)+(depth-1)*MAX_ACCUMULATED_SAMPLES).xyz;
 	seqVal ^= uvec3(nbl_glsl_xoroshiro64star(scramble_state),nbl_glsl_xoroshiro64star(scramble_state),nbl_glsl_xoroshiro64star(scramble_state));
     return vec3(seqVal)*uintBitsToFloat(0x2f800004u);
 }
 
 void gen_sample_ray(
 	out float maxT, out vec3 direction, out vec3 throughput,
-	inout nbl_glsl_xoroshiro64star_state_t scramble_state, in uint sampleID,
+	inout nbl_glsl_xoroshiro64star_state_t scramble_state, in uint sampleID, in uint depth,
 	in nbl_glsl_MC_precomputed_t precomp, in nbl_glsl_MC_instr_stream_t gcs, in nbl_glsl_MC_instr_stream_t rnps
 )
 {
 	maxT = FLT_MAX;
 
-	vec3 rand = rand3d(scramble_state,sampleID);
+	vec3 rand = rand3d(scramble_state,int(sampleID),int(depth));
 
 	float pdf;
 	nbl_glsl_LightSample s;
@@ -213,7 +213,7 @@ nbl_glsl_xoroshiro64star_state_t load_aux_vertex_attrs(
 }
 
 void generate_next_rays(
-	in uint maxRaysToGen, in nbl_glsl_MC_oriented_material_t material, in bool frontfacing, in uint vertex_depth_mod_2,
+	in uint maxRaysToGen, in nbl_glsl_MC_oriented_material_t material, in bool frontfacing, in uint vertex_depth,
 	in nbl_glsl_xoroshiro64star_state_t scramble_start_state, in uint sampleID, in uvec2 outPixelLocation,
 	in vec3 origin, vec3 geomNormal, in vec3 prevThroughput)
 {
@@ -228,7 +228,8 @@ void generate_next_rays(
 	const nbl_glsl_MC_instr_stream_t nps = nbl_glsl_MC_oriented_material_t_getNormalPrecompStream(material);
 	nbl_glsl_MC_runNormalPrecompStream(nps,precomputed);
 #endif
-
+	
+	const uint vertex_depth_mod_2 = vertex_depth&0x1u;
 	const uint vertex_depth_mod_2_inv = vertex_depth_mod_2^0x1u;
 	// prepare rays
 	uint raysToAllocate = 0u;
@@ -237,9 +238,10 @@ void generate_next_rays(
 	{
 		nbl_glsl_xoroshiro64star_state_t scramble_state = scramble_start_state;
 		// TODO: When generating NEE rays, advance the dimension, NOT the sampleID
-		gen_sample_ray(maxT[i],direction[i],nextThroughput[i],scramble_state,sampleID+i,precomputed,gcs,rnps);
-		if (i==0u)
-			imageStore(scramblebuf,ivec3(outPixelLocation,vertex_depth_mod_2_inv),uvec4(scramble_state,0u,0u));
+		gen_sample_ray(maxT[i],direction[i],nextThroughput[i],scramble_state,sampleID+i,vertex_depth,precomputed,gcs,rnps);
+// TODO: bad idea, invent something else
+//		if (i==0u)
+//			imageStore(scramblebuf,ivec3(outPixelLocation,vertex_depth_mod_2_inv),uvec4(scramble_state,0u,0u));
 		nextThroughput[i] *= prevThroughput;
 		if (any(greaterThan(nextThroughput[i],vec3(FLT_MIN))))
 			raysToAllocate++;
@@ -256,8 +258,8 @@ void generate_next_rays(
 	{
 		nbl_glsl_ext_RadeonRays_ray newRay;
 		// TODO: improve ray offsets
-		const float err = frontfacing ? (1.0/96):(-1.0/96);
-		newRay.origin = origin+geomNormal*err;
+		const float err = 1.f/96.f;
+		newRay.origin = origin+/*geomNormal/max(max(geomNormal.x,geomNormal.y),geomNormal.z)*sign(dot(geomNormal,direction[i]))*/direction[i]*err;
 		newRay.maxT = maxT[i];
 		newRay.direction = direction[i];
 		newRay.time = packOutPixelLocation(outPixelLocation);