Adjust ray origin offsets and continuation thresholds, also get rid of firstIndexSSBO (cause I shoved it into instance data)

Author: devshgraphicsprogramming

examples_tests/22.RaytracedAO/Renderer.cpp

@@ -336,7 +336,6 @@ Renderer::InitializationData Renderer::initSceneObjects(const SAssetBundle& mesh
 					cullData.reserve(batchInstanceBoundTotal);
 
 					newInstanceDataBuffer = core::make_smart_refctd_ptr<ICPUBuffer>(sizeof(ext::MitsubaLoader::instance_data_t)*batchInstanceBoundTotal);
-					retval.mdiFirstIndices.resize(batchInstanceBoundTotal);
 				}
 				// actually commit the physical memory, compute batches and set up instance data
 				{
@@ -346,7 +345,6 @@ Renderer::InitializationData Renderer::initSceneObjects(const SAssetBundle& mesh
 					auto* vertexPtr = reinterpret_cast<const float*>(cpump->getPackerDataStore().vertexBuffer->getPointer());
 					auto* mdiPtr = reinterpret_cast<DrawElementsIndirectCommand_t*>(cpump->getPackerDataStore().MDIDataBuffer->getPointer());
 					auto* newInstanceData = reinterpret_cast<ext::MitsubaLoader::instance_data_t*>(newInstanceDataBuffer->getPointer());
-					auto batchFirstIndexIt = retval.mdiFirstIndices.begin();
 
 					constexpr uint32_t kIndicesPerTriangle = 3u;
 					core::vector<CPUMeshPacker::CombinedDataOffsetTable> cdot(mdiBoundMax);
@@ -455,7 +453,6 @@ Renderer::InitializationData Renderer::initSceneObjects(const SAssetBundle& mesh
 									}
 
 									newInstanceData++;
-									*(batchFirstIndexIt++) = firstIndex;
 								}
 								for (auto j=thisShapeInstancesBeginIx; j!=rrInstances.size(); j++)
 									rr->AttachShape(rrInstances[j]);
@@ -484,7 +481,6 @@ Renderer::InitializationData Renderer::initSceneObjects(const SAssetBundle& mesh
 				);
 				instanceDataDescPtr->buffer = {0u,cullData.size()*sizeof(ext::MitsubaLoader::instance_data_t)};
 				instanceDataDescPtr->desc = std::move(newInstanceDataBuffer); // TODO: trim the buffer
-				retval.mdiFirstIndices.resize(cullData.size());
 				{
 					auto gpump = core::make_smart_refctd_ptr<GPUMeshPacker>(m_driver,cpump.get());
 					const auto& dataStore = gpump->getPackerDataStore();
@@ -874,7 +870,10 @@ void Renderer::init(const SAssetBundle& meshes,	core::smart_refctd_ptr<ICPUBuffe
 			auto createEmptyInteropBufferAndSetUpInfo = [&](IGPUDescriptorSet::SDescriptorInfo* info, InteropBuffer& interopBuffer, size_t size) -> void
 			{
 				if (static_cast<COpenGLDriver*>(m_driver)->runningInRenderdoc()) // makes Renderdoc capture the modifications done by OpenCL
+				{
 					interopBuffer.buffer = m_driver->createUpStreamingGPUBufferOnDedMem(size);
+					//interopBuffer.buffer->getBoundMemory()->mapMemoryRange(IDriverMemoryAllocation::EMCAF_WRITE,{0u,size})
+				}
 				else
 					interopBuffer.buffer = m_driver->createDeviceLocalGPUBufferOnDedMem(size);
 				interopBuffer.asRRBuffer = m_rrManager->linkBuffer(interopBuffer.buffer.get(), CL_MEM_READ_ONLY);
@@ -912,7 +911,6 @@ void Renderer::init(const SAssetBundle& meshes,	core::smart_refctd_ptr<ICPUBuffe
 			core::smart_refctd_ptr<IGPUBuffer> _staticViewDataBuffer;
 			{
 				_staticViewDataBuffer = createFilledBufferAndSetUpInfoFromStruct(infos+0,m_staticViewData);
-				createFilledBufferAndSetUpInfoFromVector(infos+1,initData.mdiFirstIndices);
 				{
 					constexpr auto ScrambleStateChannels = 2u;
 					auto tmpBuff = m_driver->createCPUSideGPUVisibleGPUBufferOnDedMem(sizeof(uint32_t)*ScrambleStateChannels*renderPixelCount);
@@ -1369,6 +1367,7 @@ uint32_t Renderer::traceBounce(uint32_t raycount)
 		m_driver->copyBuffer(m_rayCountBuffer.get(),m_littleDownloadBuffer.get(),sizeof(uint32_t)*m_raytraceCommonData.rayCountWriteIx,0u,sizeof(uint32_t));
 		static_assert(core::isPoT(RAYCOUNT_N_BUFFERING),"Raycount Buffer needs to be PoT sized!");
 		glFinish(); // sync CPU to GL
+		//auto start = std::chrono::steady_clock::now();
 		const uint32_t nextTraceRaycount = *reinterpret_cast<uint32_t*>(m_littleDownloadBuffer->getBoundMemory()->getMappedPointer());
 		if (nextTraceRaycount==0u)
 			return 0u;
@@ -1396,6 +1395,9 @@ uint32_t Renderer::traceBounce(uint32_t raycount)
 		ocl::COpenCLHandler::ocl.pclEnqueueReleaseGLObjects(commandQueue, objCount, clObjects, 1u, &raycastDone, &released);
 		ocl::COpenCLHandler::ocl.pclFlush(commandQueue);
 		ocl::COpenCLHandler::ocl.pclWaitForEvents(1u,&released);
+		//const auto elapsed = std::chrono::duration_cast<std::chrono::microseconds>(std::chrono::steady_clock::now()-start).count();
+		//if (elapsed>100000ull)
+			//printf("Path Vertex: %d took %d us\n",m_raytraceCommonData.depth,elapsed);
 		return nextTraceRaycount;
 	}
 	else

examples_tests/22.RaytracedAO/Renderer.h

@@ -67,7 +67,7 @@ class Renderer : public nbl::core::IReferenceCounted, public nbl::core::Interfac
 
 		struct InitializationData
 		{
-			InitializationData() : mdiFirstIndices(), lights(),lightCDF(),globalMeta(nullptr) {}
+			InitializationData() : lights(),lightCDF(),globalMeta(nullptr) {}
 			InitializationData(InitializationData&& other) : InitializationData()
 			{
 				operator=(std::move(other));
@@ -76,14 +76,12 @@ class Renderer : public nbl::core::IReferenceCounted, public nbl::core::Interfac
 
 			inline InitializationData& operator=(InitializationData&& other)
 			{
-				mdiFirstIndices = std::move(other.mdiFirstIndices);
 				lights = std::move(other.lights);
 				lightCDF = std::move(other.lightCDF);
 				globalMeta = other.globalMeta;
 				return *this;
 			}
 
-			nbl::core::vector<uint32_t> mdiFirstIndices;
 			nbl::core::vector<SLight> lights;
 			union
 			{

examples_tests/22.RaytracedAO/closestHit.comp

@@ -65,7 +65,7 @@ void main()
 			// clear the hit success flag
 			intersections[gl_GlobalInvocationID.x].shapeid = -1;
 
-			const uvec3 indices = get_triangle_indices(batchInstanceGUID,triangleID);
+			const uvec3 indices = get_triangle_indices(batchInstanceData,triangleID);
 
 			// positions
 			mat2x3 dPdBary_error; vec3 lastVxPos_error;

examples_tests/22.RaytracedAO/raygen.comp

@@ -46,7 +46,7 @@ void main()
 
 			//
 			const nbl_glsl_ext_Mitsuba_Loader_instance_data_t batchInstanceData = InstData.data[batchInstanceGUID];
-			const uvec3 indices = get_triangle_indices(batchInstanceGUID,triangleID);
+			const uvec3 indices = get_triangle_indices(batchInstanceData,triangleID);
 
 			// get material while waiting for indices
 			const nbl_glsl_MC_oriented_material_t material = nbl_glsl_MC_material_data_t_getOriented(batchInstanceData.material,frontfacing);

examples_tests/22.RaytracedAO/raytraceCommon.glsl

@@ -23,10 +23,6 @@ layout(set = 2, binding = 0, row_major) uniform StaticViewData
 {
 	StaticViewData_t staticViewData;
 };
-layout(set = 2, binding = 1) readonly restrict buffer ExtraBatchData
-{
-	uint firstIndex[];
-} extraBatchData;
 // rng
 layout(set = 2, binding = 2, rg32ui) uniform uimage2DArray scramblebuf;
 layout(set = 2, binding = 3) uniform usamplerBuffer sampleSequence;
@@ -51,9 +47,9 @@ void clear_raycount()
 }
 
 //
-uvec3 get_triangle_indices(in uint batchInstanceGUID, in uint triangleID)
+uvec3 get_triangle_indices(in nbl_glsl_ext_Mitsuba_Loader_instance_data_t batchInstanceData, in uint triangleID)
 {
-	const uint baseTriangleVertex = triangleID*3u+extraBatchData.firstIndex[batchInstanceGUID];
+	const uint baseTriangleVertex = triangleID*3u+batchInstanceData.padding0;
 	return uvec3(
 		nbl_glsl_VG_fetchTriangleVertexIndex(baseTriangleVertex,0u),
 		nbl_glsl_VG_fetchTriangleVertexIndex(baseTriangleVertex,1u),
@@ -294,15 +290,15 @@ for (uint i=1u; i!=vertex_depth; i++)
 //		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(nbl_glsl_FLT_MIN))))
+		if (max(max(nextThroughput[i].x,nextThroughput[i].y),nextThroughput[i].z)>exp2(-19.f)) // TODO: reverse tonemap to adjust the threshold
 			raysToAllocate++;
 		else
 			maxT[i] = 0.f;
 	}
 	// TODO: investigate workgroup reductions here
 	const uint baseOutputID = atomicAdd(rayCount[pc.cummon.rayCountWriteIx],raysToAllocate);
 
-	float ray_offset = 0.00001f; // I pulled the constant out of my @$$
+	float ray_offset = dot(abs(normalizedN),abs(origin))*nbl_glsl_numeric_limits_float_epsilon(9u)*1.03f+nbl_glsl_numeric_limits_float_epsilon(20u); // I pulled the constants out of my @$$
 	// TODO: in the future run backward error analysis of
 	// dot(mat3(WorldToObj)*(origin+offset*geomNormal/length(geomNormal))+(WorldToObj-vx_pos[1]),geomNormal)
 	// where