have to read position in raygen anyway, so stop reconstructing it

Author: devshgraphicsprogramming

examples_tests/22.RaytracedAO/Renderer.cpp

@@ -53,7 +53,7 @@ Renderer::Renderer(IVideoDriver* _driver, IAssetManager* _assetManager, scene::I
 		m_prevView(), m_sceneBound(FLT_MAX,FLT_MAX,FLT_MAX,-FLT_MAX,-FLT_MAX,-FLT_MAX),
 		m_maxRaysPerDispatch(0), m_framesDispatched(0u),
 		m_staticViewData{{0.f,0.f,0.f},0u,{0.f,0.f},{0.f,0.f},{0u,0u},0u,0u},
-		m_raytraceCommonData{core::matrix4SIMD(),core::matrix3x4SIMD(),0,0,0},
+		m_raytraceCommonData{core::matrix3x4SIMD(),vec3(),0,0,0},
 		m_indirectDrawBuffers{nullptr},m_cullPushConstants{core::matrix4SIMD(),1.f,0u,0u,0u},m_cullWorkGroups(0u),
 		m_raygenWorkGroups{0u,0u},m_visibilityBuffer(nullptr),m_colorBuffer(nullptr)
 {
@@ -1102,7 +1102,7 @@ void Renderer::deinit()
 	m_indirectDrawBuffers[1] = m_indirectDrawBuffers[0] = nullptr;
 	m_indexBuffer = nullptr;
 
-	m_raytraceCommonData = {core::matrix4SIMD(),core::matrix3x4SIMD(),0,0,0,0u};
+	m_raytraceCommonData = {core::matrix3x4SIMD(),vec3(),0,0,0,0u};
 	m_staticViewData = {{0.f,0.f,0.f},0u,{0.f,0.f},{0.f,0.f},{0u,0u},0u,0u};
 	m_framesDispatched = 0u;
 	m_maxRaysPerDispatch = 0u;
@@ -1229,10 +1229,12 @@ void Renderer::render(nbl::ITimer* timer)
 		m_cullPushConstants.currentCommandBufferIx ^= 0x01u;
 
 		// prepare camera data for raytracing
-		modifiedViewProj.getInverseTransform(m_raytraceCommonData.inverseMVP);
+		core::matrix4SIMD inverseMVP;
+		modifiedViewProj.getInverseTransform(inverseMVP);
 		const auto cameraPosition = core::vectorSIMDf().set(camera->getAbsolutePosition());
 		for (auto i=0u; i<3u; i++)
-			m_raytraceCommonData.ndcToV.rows[i] = m_raytraceCommonData.inverseMVP.rows[3]*cameraPosition[i]-m_raytraceCommonData.inverseMVP.rows[i];
+			m_raytraceCommonData.ndcToV.rows[i] = inverseMVP.rows[3]*cameraPosition[i]-inverseMVP.rows[i];
+		//m_raytraceCommonData.camPos = ;
 	}
 	// path trace
 	m_raytraceCommonData.depth = 0u;

examples_tests/22.RaytracedAO/raygen.comp

@@ -37,12 +37,8 @@ void main()
 			const uvec4 visBuffer = texelFetch(frontFacingTriangleIDDrawID_unorm16Bary_dBarydScreenHalf2x2,ivec2(outPixelLocation),0);
 
 			// unproject (TODO: redo)
-			vec3 hitWorldPos;
 			{
 				const vec3 NDC = vec3(vec2(outPixelLocation)*staticViewData.rcpPixelSize+staticViewData.rcpHalfPixelSize,1.0-revdepth);
-				
-				const vec4 tmp = nbl_glsl_pseudoMul4x4with3x1(pc.cummon.inverseMVP,NDC);
-				hitWorldPos = tmp.xyz/tmp.w;
 
 				const vec3 V = nbl_glsl_pseudoMul3x4with3x1(pc.cummon.ndcToV,NDC);
 				normalizedV = normalize(V);
@@ -68,12 +64,14 @@ void main()
 
 			const uint vertex_depth_mod_2 = 0x1u;
 			// load vertex data
-			load_positions(indices,batchInstanceData);
+			const vec3 last_vx_pos = load_positions(indices,batchInstanceData);
 			const nbl_glsl_xoroshiro64star_state_t scramble_start_state = load_aux_vertex_attrs(compactBary,indices,batchInstanceData,material,outPixelLocation,vertex_depth_mod_2
 #ifdef TEX_PREFETCH_STREAM
 				,dBarydScreen
 #endif
 			);
+			
+			const vec3 hitWorldPos = dPdBary*compactBary.xy+last_vx_pos;
 
 			// generate rays
 			const uint sampleID = bitfieldExtract(pc.cummon.samplesComputed_depth,0,16);

examples_tests/22.RaytracedAO/raytraceCommon.glsl

@@ -261,16 +261,17 @@ for (uint i=1u; i!=vertex_depth; i++)
 	// set up dispatch indirect
 	atomicMax(traceIndirect[vertex_depth_mod_2_inv].params.num_groups_x,(baseOutputID+raysToAllocate-1u)/WORKGROUP_SIZE+1u);
 
+	// TODO: improve ray offset (maybe using smooth normal wouldn't be a sin)
+	const vec3 absGeomNormal = abs(geomNormal);
+	geomNormal /= max(max(absGeomNormal.x,absGeomNormal.y),absGeomNormal.z);
 	uint offset = 0u;
 	for (uint i=0u; i<maxRaysToGen; i++)
 	if (maxT[i]!=0.f)
 	{
 		nbl_glsl_ext_RadeonRays_ray newRay;
-		// TODO: improve ray offsets
-		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.origin = origin+uintBitsToFloat(floatBitsToUint(direction[i])^floatBitsToUint(dot(geomNormal,direction[i]))&0x80000000u)/96.f;
 		newRay.maxT = maxT[i];
-		newRay.direction = direction[i]; // normalize after ? (doesn't non-uniform scale screw up BxDF eval and generation?)
+		newRay.direction = direction[i];
 		newRay.time = packOutPixelLocation(outPixelLocation);
 		newRay.mask = -1;
 		newRay._active = 1;

examples_tests/22.RaytracedAO/raytraceCommon.h

@@ -91,8 +91,8 @@ struct StaticViewData_t
 
 struct RaytraceShaderCommonData_t
 {
-	mat4	inverseMVP;
 	mat4x3  ndcToV;
+	vec3	camPos;
 	float   rcpFramesDispatched;
 #ifdef __cplusplus
 	uint16_t samplesComputed;
@@ -102,6 +102,7 @@ struct RaytraceShaderCommonData_t
 #endif
 	uint padding0;
 	uint padding1;
+	uint padding2;
 };
 
 #endif