refactor raytracer for the shading to happen in model space (less matrix transforms)

Author: devshgraphicsprogramming

examples_tests/22.RaytracedAO/Renderer.cpp

@@ -622,7 +622,7 @@ 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_maxDepth = 5u;
 constexpr uint16_t m_UNUSED_russianRouletteDepth = 5u;
 bool extractIntegratorInfo(const ext::MitsubaLoader::CElementIntegrator& integrator, uint32_t &bxdfSamples, uint32_t &maxNEESamples)
 {
@@ -1075,9 +1075,7 @@ void Renderer::deinit()
 	}
 	m_accumulation = m_tonemapOutput = nullptr;
 
-	// release the last OpenCL object and wait for OpenCL to finish
-	for (auto i=0; i<2u; i++)
-		ocl::COpenCLHandler::ocl.pclEnqueueReleaseGLObjects(commandQueue,1u,&m_rayCountBuffer[i].asRRBuffer.second,1u,nullptr,nullptr);
+	// wait for OpenCL to finish
 	ocl::COpenCLHandler::ocl.pclFlush(commandQueue);
 	ocl::COpenCLHandler::ocl.pclFinish(commandQueue);
 	for (auto i=0; i<2u; i++)
@@ -1330,11 +1328,7 @@ void Renderer::traceBounce()
 			descriptorSets[3] = m_closestHitDS.get();
 			m_driver->bindDescriptorSets(EPBP_COMPUTE,pipelineLayout,0u,4u,descriptorSets,nullptr);
 			m_driver->bindComputePipeline(m_closestHitPipeline.get());
-			// dont ask my why this fixes the dispatch indirect! (TODO: just download the raycount)
-			auto tmp = m_driver->createDeviceLocalGPUBufferOnDedMem(16u);
-			m_driver->copyBuffer(m_rayCountBuffer[readIx].buffer.get(),tmp.get(),0u,0u,16u);
-			m_driver->dispatchIndirect(tmp.get(),sizeof(uint32_t));
-			//m_driver->dispatchIndirect(m_rayCountBuffer[readIx].buffer.get(),sizeof(uint32_t));
+			m_driver->dispatchIndirect(m_rayCountBuffer[readIx].buffer.get(),sizeof(uint32_t));
 		}
 		else
 		{

examples_tests/22.RaytracedAO/closestHit.comp

@@ -53,11 +53,12 @@ void main()
 		{
 			const uint triangleID = intersection.primid;
 			const uvec3 indices = get_triangle_indices(batchInstanceGUID,triangleID);
-
+			
+			const mat4x3 batchWorldTform = InstData.data[batchInstanceGUID].tform;
 			// clear the hit success flag
 			intersections[vertex_depth_mod_2].data[gl_GlobalInvocationID.x].shapeid = -1;
 			// obtain ray incoming direction
-			normalizedV = -ray.direction;
+			normalizedV = -inverse(mat3(batchWorldTform))*ray.direction;
 
 			// positions
 			const vec3 last_vx_pos = load_positions(indices,batchInstanceGUID);
@@ -74,15 +75,16 @@ void main()
 				// 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
+				const nbl_glsl_xoroshiro64star_state_t scramble_start_state = load_aux_vertex_attrs(compactBary,indices,batchInstanceGUID,material,outPixelLocation,vertex_depth
+#ifdef TEX_PREFETCH_STREAM
+					,mat2(0.0) // TODO: Covariance Rendering
+#endif
 				);
 
 				const vec3 hitWorldPos = dPdBary*compactBary.xy+last_vx_pos;
 
 				generate_next_rays(
-					MAX_RAYS_GENERATED,material,frontfacing,vertex_depth,
+					MAX_RAYS_GENERATED,batchWorldTform,material,frontfacing,vertex_depth,
 					scramble_start_state,sampleID,outPixelLocation,hitWorldPos,geomNormal,throughput
 				);
 			}

examples_tests/22.RaytracedAO/cull.comp

@@ -48,7 +48,7 @@ void main()
     const uint batchInstanceGUID = batchInstanceData.batchInstanceGUID;
 
     const nbl_glsl_ext_Mitsuba_Loader_instance_data_t instanceData = InstData.data[batchInstanceGUID];
-    const mat4x3 worldMatrix = InstData.data[batchInstanceGUID].tform;
+    const mat4x3 worldMatrix = instanceData.tform;
     const mat4 MVP = nbl_glsl_pseudoMul4x4with4x3(pc.data.viewProjMatrix,worldMatrix);
 
     // cull

examples_tests/22.RaytracedAO/raygen.comp

@@ -36,7 +36,7 @@ void main()
 			// vis buffer read
 			const uvec4 visBuffer = texelFetch(frontFacingTriangleIDDrawID_unorm16Bary_dBarydScreenHalf2x2,ivec2(outPixelLocation),0);
 
-			// unproject
+			// unproject (TODO: redo)
 			vec3 hitWorldPos;
 			{
 				const vec3 NDC = vec3(vec2(outPixelLocation)*staticViewData.rcpPixelSize+staticViewData.rcpHalfPixelSize,1.0-revdepth);
@@ -57,7 +57,11 @@ void main()
 			#ifdef TEX_PREFETCH_STREAM
 			const mat2 dBarydScreen = mat2(unpackHalf2x16(visBuffer[2]),unpackHalf2x16(visBuffer[3]));
 			#endif
-
+			
+			// TODO: redo
+			const mat4x3 batchWorldTform = InstData.data[batchInstanceGUID].tform;
+			hitWorldPos = inverse(mat3(batchWorldTform))*(hitWorldPos-batchWorldTform[3]);
+			normalizedV = inverse(mat3(batchWorldTform))*normalizedV;
 
 			const uvec3 indices = get_triangle_indices(batchInstanceGUID,triangleID);
 
@@ -68,16 +72,18 @@ void main()
 			const uint vertex_depth_mod_2 = 0x1u;
 			// load vertex data
 			load_positions(indices,batchInstanceGUID);
-			const nbl_glsl_xoroshiro64star_state_t scramble_start_state = load_aux_vertex_attrs(
-				compactBary,indices,batchInstanceGUID,material,dBarydScreen,outPixelLocation,vertex_depth_mod_2
+			const nbl_glsl_xoroshiro64star_state_t scramble_start_state = load_aux_vertex_attrs(compactBary,indices,batchInstanceGUID,material,outPixelLocation,vertex_depth_mod_2
+#ifdef TEX_PREFETCH_STREAM
+				,dBarydScreen
+#endif
 			);
 
 			// generate rays
 			const uint sampleID = bitfieldExtract(pc.cummon.samplesComputed_depth,0,16);
 			const vec3 geomNormal = cross(dPdBary[0],dPdBary[1]);
 			const uint vertex_depth = 1u;
 			generate_next_rays(
-				staticViewData.samplesPerPixelPerDispatch,material,frontfacing,vertex_depth,
+				staticViewData.samplesPerPixelPerDispatch,batchWorldTform,material,frontfacing,vertex_depth,
 				scramble_start_state,sampleID,outPixelLocation,hitWorldPos,geomNormal,vec3(1.0)
 			);
 		}

examples_tests/22.RaytracedAO/raytraceCommon.glsl

@@ -171,8 +171,12 @@ void gen_sample_ray(
 }
 
 nbl_glsl_xoroshiro64star_state_t load_aux_vertex_attrs(
-	in vec2 compactBary, in uvec3 indices, in uint batchInstanceGUID, in nbl_glsl_MC_oriented_material_t material,
-	in mat2 dBarydScreen, in uvec2 outPixelLocation, in uint vertex_depth_mod_2
+	in vec2 compactBary, in uvec3 indices, in uint batchInstanceGUID,
+	in nbl_glsl_MC_oriented_material_t material,
+	in uvec2 outPixelLocation, in uint vertex_depth_mod_2
+#ifdef TEX_PREFETCH_STREAM
+	,in mat2 dBarydScreen
+#endif
 )
 {
 	// if we ever support spatially varying emissive, we'll need to hoist barycentric computation and UV fetching to the position fetching
@@ -199,22 +203,19 @@ nbl_glsl_xoroshiro64star_state_t load_aux_vertex_attrs(
 	#endif
 	// not needed for NEE unless doing Area or Projected Solid Angle Sampling
 	const vec3 normal = normals*nbl_glsl_barycentric_expand(compactBary);
-	normalizedN.x = dot(InstData.data[batchInstanceGUID].normalMatrixRow0,normal);
-	normalizedN.y = dot(InstData.data[batchInstanceGUID].normalMatrixRow1,normal);
-	normalizedN.z = dot(InstData.data[batchInstanceGUID].normalMatrixRow2,normal);
 
 	// init scramble while waiting for getting the instance's normal matrix
 	const nbl_glsl_xoroshiro64star_state_t scramble_start_state = imageLoad(scramblebuf,ivec3(outPixelLocation,vertex_depth_mod_2)).rg;
 
 	// while waiting for the scramble state
-	normalizedN = normalize(normalizedN);
+	normalizedN = normalize(normal);
 
 	return scramble_start_state;
 }
 
 void generate_next_rays(
-	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 uint maxRaysToGen, in mat4x3 batchWorldTform, 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)
 {
 	// get material streams as well
@@ -252,6 +253,8 @@ void generate_next_rays(
 	const uint baseOutputID = atomicAdd(traceIndirect[vertex_depth_mod_2_inv].rayCount,raysToAllocate);
 	// set up dispatch indirect
 	atomicMax(traceIndirect[vertex_depth_mod_2_inv].params.num_groups_x,(baseOutputID+raysToAllocate-1u)/WORKGROUP_SIZE+1u);
+
+	const mat3 batchWorldScaleRot = mat3(batchWorldTform);
 	uint offset = 0u;
 	for (uint i=0u; i<maxRaysToGen; i++)
 	if (maxT[i]!=0.f)
@@ -261,7 +264,7 @@ void generate_next_rays(
 		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.direction = batchWorldScaleRot*direction[i];
 		newRay.time = packOutPixelLocation(outPixelLocation);
 		newRay.mask = -1;
 		newRay._active = 1;