greatly improve world position reconstruction from depth buffer

Author: devshgraphicsprogramming

examples_tests/22.RaytracedAO/common.glsl

@@ -19,6 +19,7 @@
 	{
 		float x,y,z;
 	};
+	#define vec4 nbl::core::vectorSIMDf
 	#define mat4 nbl::core::matrix4SIMD
 	#define mat4x3 nbl::core::matrix3x4SIMD
 #endif

examples_tests/22.RaytracedAO/dirty_source/ExtraCrap.cpp

@@ -52,7 +52,7 @@ Renderer::Renderer(IVideoDriver* _driver, IAssetManager* _assetManager, scene::I
 	#endif
 		rrShapeCache(), rrInstances(), m_sceneBound(FLT_MAX,FLT_MAX,FLT_MAX,-FLT_MAX,-FLT_MAX,-FLT_MAX),
 		m_maxRaysPerDispatch(0), m_staticViewData{{0.f,0.f,0.f},0u,{0.f,0.f},{0.f,0.f},{0u,0u},0u,0u},
-		m_raytraceCommonData{core::matrix3x4SIMD(),{},0u,0.f,0u,0u,0.f},
+		m_raytraceCommonData{core::matrix4SIMD(),core::matrix3x4SIMD(),0,0,0},
 		m_indirectDrawBuffers{nullptr},m_cullPushConstants{core::matrix4SIMD(),1.f,0u,0u,0u},m_cullWorkGroups(0u),
 		m_raygenWorkGroups{0u,0u},m_resolveWorkGroups{0u,0u},
 		m_visibilityBuffer(nullptr),tmpTonemapBuffer(nullptr),m_colorBuffer(nullptr)
@@ -619,8 +619,8 @@ void Renderer::init(const SAssetBundle& meshes,
 				assert(film.cropOffsetY == 0);
 				m_staticViewData.imageDimensions = {static_cast<uint32_t>(film.cropWidth),static_cast<uint32_t>(film.cropHeight)};
 			}
-			m_staticViewData.rcpPixelSize = { 1.f/float(m_staticViewData.imageDimensions.x),1.f/float(m_staticViewData.imageDimensions.y) };
-			m_staticViewData.rcpHalfPixelSize = { 0.5f/float(m_staticViewData.imageDimensions.x),0.5f/float(m_staticViewData.imageDimensions.y) };
+			m_staticViewData.rcpPixelSize = { 2.f/float(m_staticViewData.imageDimensions.x),-2.f/float(m_staticViewData.imageDimensions.y) };
+			m_staticViewData.rcpHalfPixelSize = { 1.f/float(m_staticViewData.imageDimensions.x)-1.f,1.f-1.f/float(m_staticViewData.imageDimensions.y) };
 		}
 
 		// figure out dispatch sizes
@@ -990,7 +990,7 @@ void Renderer::deinit()
 	m_mdiDrawCalls.clear();
 	m_indirectDrawBuffers[1] = m_indirectDrawBuffers[0] = nullptr;
 
-	m_raytraceCommonData = {core::matrix3x4SIMD(),{},0u,0.f,0u,0u,0.f};
+	m_raytraceCommonData = {core::matrix4SIMD(),core::matrix3x4SIMD(),0,0,0};
 	m_staticViewData = {{0.f,0.f,0.f},0u,{0.f,0.f},{0.f,0.f},{0u,0u},0u,0u};
 	m_maxRaysPerDispatch = 0u;
 	m_sceneBound = core::aabbox3df(FLT_MAX, FLT_MAX, FLT_MAX, -FLT_MAX, -FLT_MAX, -FLT_MAX);
@@ -1062,24 +1062,10 @@ void Renderer::render(nbl::ITimer* timer)
 		m_cullPushConstants.currentCommandBufferIx ^= 0x01u;
 
 		// prepare camera data for raytracing
+		currentViewProj.getInverseTransform(m_raytraceCommonData.inverseMVP);
 		const auto cameraPosition = core::vectorSIMDf().set(camera->getAbsolutePosition());
-		{
-			auto frustum = camera->getViewFrustum();
-			core::matrix4SIMD fromCorners;
-			fromCorners[0] = cameraPosition-frustum->getFarLeftDown();
-			fromCorners[1] = cameraPosition-frustum->getFarRightDown()-fromCorners[0];
-			fromCorners[2] = cameraPosition-frustum->getFarLeftUp();
-			fromCorners[3] = cameraPosition-frustum->getFarRightUp()-fromCorners[2];
-			m_raytraceCommonData.frustumCornersToCamera = core::transpose(fromCorners).extractSub3x4();
-		}
-		m_raytraceCommonData.cameraPosition.x = cameraPosition.x;
-		m_raytraceCommonData.cameraPosition.y = cameraPosition.y;
-		m_raytraceCommonData.cameraPosition.z = cameraPosition.z;
-		{
-			auto projMat = camera->getProjectionMatrix();
-			auto* row = projMat.rows;
-			m_raytraceCommonData.depthLinearizationConstant = -double(row[3][2])/(double(row[3][2])-double(row[2][2]));
-		}
+		for (auto i=0u; i<3u; i++)
+			m_raytraceCommonData.ndcToV.rows[i] = m_raytraceCommonData.inverseMVP.rows[3]*cameraPosition[i]-m_raytraceCommonData.inverseMVP.rows[i];
 	}
 
 	// generate rays

examples_tests/22.RaytracedAO/raygen.comp

@@ -4,6 +4,7 @@
 // for per pixel inputs
 #include <nbl/builtin/glsl/utils/normal_decode.glsl>
 #include <nbl/builtin/glsl/random/xoroshiro.glsl>
+#include <nbl/builtin/glsl/utils/transform.glsl>
 
 // rng
 layout(set = 2, binding = 0) uniform usamplerBuffer sampleSequence;
@@ -29,21 +30,14 @@ vec3 nbl_glsl_MC_getNormalizedWorldSpaceN()
 	return normalizedN;
 }
 vec3 worldPosition;
-vec3 nbl_glsl_MC_getWorldSpacePosition()
-{
-	return worldPosition;
-}
 mat2x3 dPosdScreen = mat2x3(vec3(0.0),vec3(0.0)); // TODO
-mat2x3 nbl_glsl_MC_getdPos(in vec3 p)
+mat2x3 nbl_glsl_MC_getdPos()
 {
 	return dPosdScreen;
 }
 #define _NBL_USER_PROVIDED_MATERIAL_COMPILER_GLSL_BACKEND_FUNCTIONS_
 #include <nbl/builtin/glsl/material_compiler/common.glsl>
 
-//
-///
-
 
 // functions
 vec3 rand3d(in uint _sample, inout nbl_glsl_xoroshiro64star_state_t scramble_state)
@@ -53,20 +47,6 @@ vec3 rand3d(in uint _sample, inout nbl_glsl_xoroshiro64star_state_t scramble_sta
     return vec3(seqVal)*uintBitsToFloat(0x2f800004u);
 }
 
-float linearizeZBufferVal(in float nonLinearZBufferVal)
-{
-	// 1-(Ax+B)/(Cx) = y
-	// (Ax+B)/(Cx) = 1-y
-	// x = B/(C(1-y)-A)
-	// x = B/(C-A-Cy)
-	// get back original Z: `row[2][3]/(row[3][2]-row[2][2]-y*row[3][2]) = x`
-	// max Z: `B/(C-A)`
-	// positive [0,1] Z: `B/(C-A-Cy)/(B/(C-A))`
-	// positive [0,1] Z: `(C-A)/(C-A-Cy)`
-	// positive [0,1] Z: `D/(D-Cy)`
-    return 1.0/(pc.cummon.depthLinearizationConstant*nonLinearZBufferVal+1.0);
-}
-
 /*
 float maxAbs1(in float val) 
 {
@@ -157,11 +137,12 @@ void main()
 
 			// unproject
 			{
-				const mat4x3 frustumCornersToCamera = pc.cummon.frustumCornersToCamera;
-				const vec2 NDC = vec2(outputLocation)*staticViewData.rcpPixelSize+staticViewData.rcpHalfPixelSize;
+				const vec3 NDC = vec3(vec2(outputLocation)*staticViewData.rcpPixelSize+staticViewData.rcpHalfPixelSize,1.0-revdepth);
 
-				const vec3 V = mix(frustumCornersToCamera[1]*NDC.x+frustumCornersToCamera[0],frustumCornersToCamera[3]*NDC.x+frustumCornersToCamera[2],NDC.yyy); // could maybe compute this more precisely
-				worldPosition = pc.cummon.cameraPosition-V*linearizeZBufferVal(revdepth);
+				vec4 tmp = nbl_glsl_pseudoMul4x4with3x1(pc.cummon.inverseMVP,NDC);
+				worldPosition = tmp.xyz/tmp.w;
+
+				const vec3 V = nbl_glsl_pseudoMul3x4with3x1(pc.cummon.ndcToV,NDC);
 				normalizedV = normalize(V);
 			}
 
@@ -179,7 +160,6 @@ void main()
 
 				//
 				emissive = nbl_glsl_MC_oriented_material_t_getEmissive(material);
-				emissive = vec3(float(objectID),float(triangleID),0.0);
 
 				// normally we'd use MeshPackerV2's vertex attribute data for this, but now we read from temporary GBuffer
 				const vec3 normal = nbl_glsl_NormalDecode_signedSpherical(normalBuffer);
@@ -223,9 +203,9 @@ void main()
 			{
 				throughput.rgb /= float(staticViewData.samplesPerPixelPerDispatch);
 
-				const float tt = 1.0;
-				rays[realOutputID].origin = pc.cummon.cameraPosition;//nbl_glsl_MC_getWorldSpacePosition()+nbl_glsl_MC_getNormalizedWorldSpaceV()*tt;//-direction*0.5*maxT;/*+newray.direction*err?; TODO */
-				rays[realOutputID].maxT = FLT_MAX;//tt*2.0;
+				const float tt = 1.0/1024.0;
+				rays[realOutputID].origin = worldPosition+nbl_glsl_MC_getNormalizedWorldSpaceV()*tt;//-direction*0.5*maxT;/*+newray.direction*err?; TODO */
+				rays[realOutputID].maxT = tt*2.0;
 				rays[realOutputID].direction = -nbl_glsl_MC_getNormalizedWorldSpaceV();
 				rays[realOutputID].mask = -1;
 				rays[realOutputID]._active = 1;

examples_tests/22.RaytracedAO/raytraceCommon.glsl

@@ -94,13 +94,12 @@ struct StaticViewData_t
 
 struct RaytraceShaderCommonData_t
 {
-	mat4x3  frustumCornersToCamera;
-	vec3	cameraPosition;
-	uint	padding;
-	float   depthLinearizationConstant;
+	mat4	inverseMVP;
+	mat4x3  ndcToV;
 	uint    samplesComputedPerPixel;
 	uint    framesDispatched;
     float   rcpFramesDispatched;
+	float	padding0;
 };
 
 

examples_tests/22.RaytracedAO/resolve.comp

@@ -40,7 +40,7 @@ void main()
 		else
 			acc += color;
 		storeAccumulation(acc,pixelCoord);
-		acc = abs(acc);
+		acc = abs(color);
         acc.rg /= vec2(32.0,1023.0);
 		imageStore(framebuffer,pixelCoord,uvec4(nbl_glsl_encodeRGB10A2(vec4(acc,1.0)),0u,0u,0u));
 	}

include/nbl/asset/CQuantNormalCache.h

@@ -322,6 +322,7 @@ class CQuantNormalCache
 				return cacheSize;
 
 			}
+			return 0xdeadbeefBADC0FFEu;
 		}
 
 	private:

src/nbl/asset/CShaderIntrospector.cpp

@@ -699,15 +699,15 @@ static void introspectStructType(spirv_cross::Compiler& _comp, impl::SShaderMemo
             }
         }
 		else if (mtype.basetype != spirv_cross::SPIRType::Struct) // might have to ignore a few more types than structs
-			member.arrayStride = core::max(0x1u<<core::findMSB(member.size),16u);
+			member.arrayStride = core::max(0x1u<<core::findMSB(member.size),16u); // @Crisspl  NOT TRUE FOR MATRICES!
 
         if (mtype.basetype == spirv_cross::SPIRType::Struct) //recursive introspection done in DFS manner (and without recursive calls)
             _pushStack.push({member.members, mtype, member.offset});
         else
 		{
             member.mtxRowCnt = mtype.vecsize;
             member.mtxColCnt = mtype.columns;
-            if (member.mtxColCnt > 1u)
+            if (member.mtxColCnt > 1u) // @Crisspl why are you checking columns to get stride? What about row major?
                 member.mtxStride = _comp.type_struct_member_matrix_stride(_parentType, m);
         }
     }

src/nbl/ext/MitsubaLoader/CMitsubaLoader.cpp

@@ -1311,9 +1311,9 @@ SContext::tex_ass_type CMitsubaLoader::cacheTexture(SContext& ctx, uint32_t hier
 							_NBL_DEBUG_BREAK_IF(true); // TODO : replace whole texture?
 							break;
 						default:
-							return ISampler::ETC_REPEAT;
 							break;
 					}
+					return ISampler::ETC_REPEAT;
 				};
 				samplerParams.TextureWrapU = getWrapMode(tex->bitmap.wrapModeU);
 				samplerParams.TextureWrapV = getWrapMode(tex->bitmap.wrapModeV);