jot down some optimization ideas

Author: devshgraphicsprogramming

examples_tests/22.RaytracedAO/Renderer.cpp

@@ -264,6 +264,7 @@ Renderer::InitializationData Renderer::initSceneObjects(const SAssetBundle& mesh
 				allocParams.MDIDataBuffMinAllocCnt = 1u; //so structs from different meshbuffers are adjacent in memory
 
 
+				// TODO: after position moves to RGB21, need to split up normal from UV
 				constexpr auto combinedNormalUVAttributeIx = 1;
 				constexpr auto newEnabledAttributeMask = (0x1u<<combinedNormalUVAttributeIx)|0b1;
 
@@ -277,6 +278,9 @@ Renderer::InitializationData Renderer::initSceneObjects(const SAssetBundle& mesh
 				{
 					core::vector<const ICPUMeshBuffer*> meshBuffersToProcess;
 					meshBuffersToProcess.reserve(contents.size());
+					// TODO: Optimize! Check which triangles need normals, bin into two separate meshbuffers, dont have normals for meshbuffers where all(abs(transpose(normals)*cross(pos1-pos0,pos2-pos0))~=1.f) 
+					// TODO: Optimize! Check which materials use any textures, if meshbuffer doens't use any textures, its pipeline doesn't need UV coordinates
+					// TODO: separate pipeline for stuff without UVs and separate out the barycentric derivative FBO attachment 
 					for (const auto& asset : contents)
 					{
 						auto cpumesh = static_cast<asset::ICPUMesh*>(asset.get());

examples_tests/22.RaytracedAO/fillVisBuffer.frag

@@ -1,14 +1,14 @@
 // Copyright (C) 2018-2020 - DevSH Graphics Programming Sp. z O.O.
 // This file is part of the "Nabla Engine".
 // For conditions of distribution and use, see copyright notice in nabla.h
-#version 430 core
+#version 460 core
 #extension GL_EXT_shader_16bit_storage : require
+#include <nbl/builtin/glsl/barycentric/extensions.glsl>
 
 
 #define _NBL_GLSL_EXT_MITSUBA_LOADER_INSTANCE_DATA_BINDING_ 0
 #include "virtualGeometry.glsl"
 
-#undef NBL_GL_NV_fragment_shader_barycentric
 #include <nbl/builtin/glsl/barycentric/frag.glsl>
 layout(location = 2) flat in uint BackfacingBit_BatchInstanceGUID;
 layout(location = 3) flat in uint drawCmdFirstIndex;

examples_tests/22.RaytracedAO/fillVisBuffer.vert

@@ -1,8 +1,9 @@
 // Copyright (C) 2018-2020 - DevSH Graphics Programming Sp. z O.O.
 // This file is part of the "Nabla Engine".
 // For conditions of distribution and use, see copyright notice in nabla.h
-#version 430 core
+#version 460 core
 #extension GL_EXT_shader_16bit_storage : require
+#include <nbl/builtin/glsl/barycentric/extensions.glsl>
 
 #include "rasterizationCommon.h"
 
@@ -14,7 +15,6 @@ layout(set=2, binding=0, row_major) readonly restrict buffer PerInstancePerCamer
     DrawData_t data[];
 } instanceDataPerCamera;
 
-#undef NBL_GL_NV_fragment_shader_barycentric
 #include <nbl/builtin/glsl/barycentric/vert.glsl>
 layout(location = 2) flat out uint BackfacingBit_BatchInstanceGUID;
 layout(location = 3) flat out uint drawCmdFirstIndex;

examples_tests/22.RaytracedAO/raygen.comp

@@ -41,6 +41,7 @@ void main()
 			const uint batchInstanceGUID = bitfieldExtract(visBuffer[0],0,31-triangleIDBitcount);
 			const vec2 compactBary = unpackUnorm2x16(visBuffer[1]);
 			#ifdef TEX_PREFETCH_STREAM
+			// TODO: separate pipeline and separate out the barycentric derivative FBO attachment, only write if need to, only fetch if `needs_texture_prefetch`
 			const mat2 dBarydScreen = mat2(unpackHalf2x16(visBuffer[2]),unpackHalf2x16(visBuffer[3]));
 			#endif
 

examples_tests/22.RaytracedAO/raytraceCommon.glsl

@@ -156,6 +156,12 @@ mat2 nbl_glsl_perturbNormal_dUVdSomething()
 #define _NBL_USER_PROVIDED_MATERIAL_COMPILER_GLSL_BACKEND_FUNCTIONS_
 #include <nbl/builtin/glsl/material_compiler/common.glsl>
 
+
+bool needs_texture_prefetch(in nbl_glsl_ext_Mitsuba_Loader_instance_data_t batchInstanceData)
+{
+	return true;
+}
+
 nbl_glsl_xoroshiro64star_state_t load_aux_vertex_attrs(
 	in nbl_glsl_ext_Mitsuba_Loader_instance_data_t batchInstanceData,
 	in uvec3 indices, in vec2 compactBary, in vec3 geomDenormal,
@@ -168,18 +174,21 @@ nbl_glsl_xoroshiro64star_state_t load_aux_vertex_attrs(
 {
 	// if we ever support spatially varying emissive, we'll need to hoist barycentric computation and UV fetching to the position fetching
 	#ifdef TEX_PREFETCH_STREAM
-	const mat3x2 uvs = mat3x2(
-		nbl_glsl_fetchVtxUV(indices[0],batchInstanceData),
-		nbl_glsl_fetchVtxUV(indices[1],batchInstanceData),
-		nbl_glsl_fetchVtxUV(indices[2],batchInstanceData)
-	);
+	if (needs_texture_prefetch(batchInstanceData))
+	{
+		const mat3x2 uvs = mat3x2(
+			nbl_glsl_fetchVtxUV(indices[0],batchInstanceData),
+			nbl_glsl_fetchVtxUV(indices[1],batchInstanceData),
+			nbl_glsl_fetchVtxUV(indices[2],batchInstanceData)
+		);
 
-	const nbl_glsl_MC_instr_stream_t tps = nbl_glsl_MC_oriented_material_t_getTexPrefetchStream(material);
+		const nbl_glsl_MC_instr_stream_t tps = nbl_glsl_MC_oriented_material_t_getTexPrefetchStream(material);
 
-	dUVdBary = mat2(uvs[0]-uvs[2],uvs[1]-uvs[2]);
-	const vec2 UV = dUVdBary*compactBary+uvs[2];
-	const mat2 dUVdScreen = nbl_glsl_applyChainRule2D(dUVdBary,dBarydScreen);
-	nbl_glsl_MC_runTexPrefetchStream(tps,UV,dUVdScreen);
+		dUVdBary = mat2(uvs[0]-uvs[2],uvs[1]-uvs[2]);
+		const vec2 UV = dUVdBary*compactBary+uvs[2];
+		const mat2 dUVdScreen = nbl_glsl_applyChainRule2D(dUVdBary,dBarydScreen);
+		nbl_glsl_MC_runTexPrefetchStream(tps,UV,dUVdScreen);
+	}
 	#endif
 	// the rest is always only needed for continuing
 

examples_tests/41.VisibilityBuffer/fillVBuffer.frag

@@ -1,4 +1,4 @@
-#version 430 core
+#version 460 core
 #include <nbl/builtin/glsl/barycentric/extensions.glsl>
 
 #include "common.glsl"

include/nbl/builtin/glsl/barycentric/extensions.glsl

@@ -6,7 +6,7 @@
 #define _NBL_BUILTIN_GLSL_BARYCENTRIC_EXTENSIONS_INCLUDED_
 
 #ifdef NBL_GL_NV_fragment_shader_barycentric
-#extension GL_NV_fragment_shader_barycentric : require
+#extension GL_NV_fragment_shader_barycentric : enable
 #elif defined(NBL_GL_AMD_shader_explicit_vertex_parameter)
 #extension GL_AMD_shader_explicit_vertex_parameter : enable
 #endif