Merge pull request #109 from Devsh-Graphics-Programming/pathtracing

Path Tracer without MIS or RR

Author: devshgraphicsprogramming

3rdparty/CMakeLists.txt

@@ -416,7 +416,8 @@ if (NBL_BUILD_RADEON_RAYS)
 	option(RR_NO_TESTS "RADEON_RAYS: don't do tests" ON)
 	option(RR_ENABLE_STATIC "RADEON_RAYS: compile into a static library" ON)
 	option(RR_EMBED_KERNELS "RADEON_RAYS: so we don't have to go looking for OpenCL kernel code" ON)
-	option(RR_USE_VULKAN "RADEON_RAYS: use Vulkan (only if you have RR 3.0)" OFF)
+	set(RR_USE_VULKAN OFF CACHE BOOL "RADEON_RAYS: Cannot use Vulkan for RR 2.0 the backend is broken" FORCE)
+	set(RR_ENABLE_BACKFACE_CULL OFF CACHE BOOL "RADEON_RAYS: Don't enable backface culling (we use the memory of the ray struct for something else there)." FORCE)
 	set(MSVC_USE_STATIC_CRT ON CACHE BOOL "Use /MT flag (static CRT) when compiling in MSVC")
 	add_subdirectory(radeonrays EXCLUDE_FROM_ALL)
 
@@ -470,7 +471,7 @@ set(NBL_3RDPARTY_TARGETS
 				SPIRV 
 				Iex
 				IexMath
-				SPIRV-Tools
+				SPIRV-Tools-static
 				SPIRV-Tools-opt )
 #TODO insert more targets that are conditionally built
 if (NBL_BUILD_MITSUBA_LOADER)

3rdparty/SPIRV-Headers

@@ -105,7 +105,7 @@ void main()
 	mat2 dUV = mat2(dFdx(UV),dFdy(UV));
 
 	// "The sign of this computation is negated when the value of GL_CLIP_ORIGIN (the clip volume origin, set with glClipControl) is GL_UPPER_LEFT."
-	const bool front = (!gl_FrontFacing) != (PC.camTformDeterminant*InstData.data[InstanceIndex].determinant < 0.0);
+	const bool front = bool((InstData.data[InstanceIndex].determinantSignBit^mix(~0u,0u,gl_FrontFacing!=PC.camTformDeterminant<0.0))&0x80000000u);
 	precomp = nbl_glsl_MC_precomputeData(front);
 	material = nbl_glsl_MC_material_data_t_getOriented(InstData.data[InstanceIndex].material,precomp.frontface);
 #ifdef TEX_PREFETCH_STREAM

3rdparty/SPIRV-Tools

@@ -19,22 +19,18 @@
 class Renderer : public nbl::core::IReferenceCounted, public nbl::core::InterfaceUnmovable
 {
     public:
-		#include "drawCommon.glsl"
-		#include "raytraceCommon.glsl"
+		#include "rasterizationCommon.h"
+		#include "raytraceCommon.h"
 		#ifdef __cplusplus
 			#undef uint
+			#undef vec4
 			#undef mat4
 			#undef mat4x3
 		#endif
 
-		// No 8k yet, too many rays to store
-		_NBL_STATIC_INLINE_CONSTEXPR uint32_t MaxResolution[2] = {7680/2,4320/2};
-
-
 		Renderer(nbl::video::IVideoDriver* _driver, nbl::asset::IAssetManager* _assetManager, nbl::scene::ISceneManager* _smgr, bool useDenoiser = true);
 
-		void init(	const nbl::asset::SAssetBundle& meshes, nbl::core::smart_refctd_ptr<nbl::asset::ICPUBuffer>&& sampleSequence,
-					uint32_t rayBufferSize=(sizeof(::RadeonRays::ray)+sizeof(::RadeonRays::Intersection))*2u*MaxResolution[0]*MaxResolution[1]); // 2 samples for MIS, TODO: compute default buffer size
+		void init(const nbl::asset::SAssetBundle& meshes, nbl::core::smart_refctd_ptr<nbl::asset::ICPUBuffer>&& sampleSequence);
 
 		void deinit();
 
@@ -46,20 +42,32 @@ class Renderer : public nbl::core::IReferenceCounted, public nbl::core::Interfac
 
 		uint64_t getTotalSamplesComputed() const
 		{
-			const auto samplesPerDispatch = static_cast<uint64_t>(m_staticViewData.samplesPerRowPerDispatch*m_staticViewData.imageDimensions.y);
-			const auto framesDispatched = static_cast<uint64_t>(m_raytraceCommonData.framesDispatched);
+			const auto samplesPerDispatch = static_cast<uint64_t>(m_staticViewData.samplesPerPixelPerDispatch*m_staticViewData.imageDimensions.x*m_staticViewData.imageDimensions.y);
+			const auto framesDispatched = static_cast<uint64_t>(m_framesDispatched);
 			return framesDispatched*samplesPerDispatch;
 		}
+		uint64_t getTotalRaysCast() const
+		{
+			return m_totalRaysCast;
+		}
 
-
-		_NBL_STATIC_INLINE_CONSTEXPR uint32_t MaxDimensions = 6u;
+		//! Brief guideline to good path depth limits
+		// Want to see stuff with indirect lighting on the other side of a pane of glass
+		// 5 = glass frontface->glass backface->diffuse surface->diffuse surface->light
+		// Want to see through a glass box, vase, or office 
+		// 7 = glass frontface->glass backface->glass frontface->glass backface->diffuse surface->diffuse surface->light
+		// pick higher numbers for better GI and less bias
+		_NBL_STATIC_INLINE_CONSTEXPR uint32_t MaxPathDepth = 8u;
+		_NBL_STATIC_INLINE_CONSTEXPR uint32_t RandomDimsPerPathVertex = 3u;
+		// one less because the first path vertex is rasterized
+		_NBL_STATIC_INLINE_CONSTEXPR uint32_t MaxDimensions = RandomDimsPerPathVertex*(MaxPathDepth-1u);
 		static const float AntiAliasingSequence[4096][2];
     protected:
         ~Renderer();
 
 		struct InitializationData
 		{
-			InitializationData() : lights(),lightRadiances(),lightCDF(),globalMeta(nullptr) {}
+			InitializationData() : lights(),lightCDF(),globalMeta(nullptr) {}
 			InitializationData(InitializationData&& other) : InitializationData()
 			{
 				operator=(std::move(other));
@@ -69,15 +77,12 @@ class Renderer : public nbl::core::IReferenceCounted, public nbl::core::Interfac
 			inline InitializationData& operator=(InitializationData&& other)
 			{
 				lights = std::move(other.lights);
-				lightRadiances = std::move(other.lightRadiances);
 				lightCDF = std::move(other.lightCDF);
 				globalMeta = other.globalMeta;
 				return *this;
 			}
 
-
 			nbl::core::vector<SLight> lights;
-			nbl::core::vector<nbl::core::vectorSIMDf> lightRadiances;
 			union
 			{
 				nbl::core::vector<float> lightPDF;
@@ -89,7 +94,11 @@ class Renderer : public nbl::core::IReferenceCounted, public nbl::core::Interfac
 		void initSceneNonAreaLights(InitializationData& initData);
 		void finalizeScene(InitializationData& initData);
 
-		nbl::core::smart_refctd_ptr<nbl::video::IGPUImageView> createScreenSizedTexture(nbl::asset::E_FORMAT format);
+		//
+		nbl::core::smart_refctd_ptr<nbl::video::IGPUImageView> createScreenSizedTexture(nbl::asset::E_FORMAT format, uint32_t layers = 0u);
+
+		//
+		uint32_t traceBounce(uint32_t raycount);
 
 
 		// "constants"
@@ -110,34 +119,32 @@ class Renderer : public nbl::core::IReferenceCounted, public nbl::core::Interfac
 
 
 		// persistent (intialized in constructor
+		nbl::core::smart_refctd_ptr<nbl::video::IGPUBuffer> m_rayCountBuffer,m_littleDownloadBuffer;
 		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSetLayout> m_cullDSLayout;
-
-		nbl::core::smart_refctd_ptr<nbl::asset::ICPUSpecializedShader> m_visibilityBufferFillShaders[2];
-		nbl::core::smart_refctd_ptr<nbl::asset::ICPUPipelineLayout> m_visibilityBufferFillPipelineLayoutCPU;
-		nbl::core::smart_refctd_ptr<nbl::video::IGPUPipelineLayout> m_visibilityBufferFillPipelineLayoutGPU;
-		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSetLayout> m_perCameraRasterDSLayout;
-
-		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSetLayout> m_commonRaytracingDSLayout, m_raygenDSLayout, m_resolveDSLayout;
+		nbl::core::smart_refctd_ptr<const nbl::video::IGPUDescriptorSetLayout> m_perCameraRasterDSLayout;
+		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSetLayout> m_rasterInstanceDataDSLayout,m_additionalGlobalDSLayout,m_commonRaytracingDSLayout;
+		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSetLayout> m_raygenDSLayout,m_closestHitDSLayout,m_resolveDSLayout;
+		nbl::core::smart_refctd_ptr<nbl::video::IGPURenderpassIndependentPipeline> m_visibilityBufferFillPipeline;
 
 
 		// scene specific data
-		nbl::ext::RadeonRays::MockSceneManager m_mock_smgr;
-		nbl::ext::RadeonRays::Manager::MeshBufferRRShapeCache rrShapeCache;
-		nbl::ext::RadeonRays::Manager::NblInstanceRRInstanceCache rrInstances;
+		nbl::core::vector<::RadeonRays::Shape*> rrShapes;
+		nbl::core::vector<::RadeonRays::Shape*> rrInstances;
 
 		nbl::core::matrix3x4SIMD m_prevView;
+		nbl::core::matrix4x3 m_prevCamTform;
 		nbl::core::aabbox3df m_sceneBound;
-		uint32_t m_maxRaysPerDispatch;
+		uint32_t m_framesDispatched;
+		vec2 m_rcpPixelSize;
+		uint64_t m_totalRaysCast;
 		StaticViewData_t m_staticViewData;
 		RaytraceShaderCommonData_t m_raytraceCommonData;
 
+		nbl::core::smart_refctd_ptr<nbl::video::IGPUBuffer> m_indexBuffer;
 		nbl::core::smart_refctd_ptr<nbl::video::IGPUBuffer> m_indirectDrawBuffers[2];
 		struct MDICall
 		{
-			nbl::asset::SBufferBinding<const nbl::video::IGPUBuffer> vertexBindings[nbl::video::IGPUMeshBuffer::MAX_ATTR_BUF_BINDING_COUNT];
-			nbl::core::smart_refctd_ptr<const nbl::video::IGPUBuffer> indexBuffer;
-			nbl::core::smart_refctd_ptr<const nbl::video::IGPURenderpassIndependentPipeline> pipeline;
-			uint32_t mdiOffset, mdiCount;
+			uint32_t mdiOffset,mdiCount;
 		};
 		nbl::core::vector<MDICall> m_mdiDrawCalls;
 		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSet> m_cullDS;
@@ -146,23 +153,23 @@ class Renderer : public nbl::core::IReferenceCounted, public nbl::core::Interfac
 
 		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSet> m_perCameraRasterDS;
 
-		nbl::core::smart_refctd_ptr<nbl::video::IGPUPipelineLayout> m_cullPipelineLayout, m_raygenPipelineLayout, m_resolvePipelineLayout;
-		nbl::core::smart_refctd_ptr<nbl::video::IGPUComputePipeline> m_cullPipeline, m_raygenPipeline, m_resolvePipeline;
-		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSet> m_globalBackendDataDS,m_commonRaytracingDS,m_raygenDS;
+		nbl::core::smart_refctd_ptr<nbl::video::IGPUComputePipeline> m_cullPipeline,m_raygenPipeline,m_closestHitPipeline,m_resolvePipeline;
+		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSet> m_globalBackendDataDS,m_additionalGlobalDS;
+		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSet> m_commonRaytracingDS[2];
+		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSet> m_rasterInstanceDataDS,m_raygenDS,m_resolveDS;
+		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSet> m_closestHitDS[2];
 		uint32_t m_raygenWorkGroups[2];
 
 		struct InteropBuffer
 		{
 			nbl::core::smart_refctd_ptr<nbl::video::IGPUBuffer> buffer;
 			std::pair<::RadeonRays::Buffer*, cl_mem> asRRBuffer = { nullptr,0u };
 		};
-		InteropBuffer m_rayCountBuffer,m_rayBuffer,m_intersectionBuffer;
-
-		nbl::core::smart_refctd_ptr<nbl::video::IGPUDescriptorSet> m_resolveDS;
-		uint32_t m_resolveWorkGroups[2];
+		InteropBuffer m_rayBuffer[2];
+		InteropBuffer m_intersectionBuffer[2];
 
 		nbl::core::smart_refctd_ptr<nbl::video::IGPUImageView> m_accumulation,m_tonemapOutput;
-		nbl::video::IFrameBuffer* m_visibilityBuffer,* m_colorBuffer,* tmpTonemapBuffer;
+		nbl::video::IFrameBuffer* m_visibilityBuffer,* m_colorBuffer;
 
 	#ifdef _NBL_BUILD_OPTIX_
 		nbl::core::smart_refctd_ptr<nbl::ext::OptiX::IDenoiser> m_denoiser;