Change the max primitive count for meshlets

AnKi/Renderer/GBuffer.cpp

@@ -71,16 +71,13 @@ Error GBuffer::initInternal()
 	}
 
 	// FB descr
-	AttachmentLoadOperation loadop = AttachmentLoadOperation::kDontCare;
-#if ANKI_EXTRA_CHECKS
-	loadop = AttachmentLoadOperation::kClear;
-#endif
+	const AttachmentLoadOperation loadop = AttachmentLoadOperation::kClear;
 
 	m_fbDescr.m_colorAttachmentCount = kGBufferColorRenderTargetCount;
 	for(U i = 0; i < kGBufferColorRenderTargetCount; ++i)
 	{
 		m_fbDescr.m_colorAttachments[i].m_loadOperation = loadop;
-		m_fbDescr.m_colorAttachments[i].m_clearValue.m_colorf = {1.0f, 0.0f, 1.0f, 0.0f};
+		m_fbDescr.m_colorAttachments[i].m_clearValue.m_colorf = {0.0f, 0.0f, 0.0f, 0.0f};
 	}
 
 	m_fbDescr.m_colorAttachments[3].m_loadOperation = AttachmentLoadOperation::kClear;

AnKi/ShaderCompiler/Dxc.cpp

@@ -89,6 +89,7 @@ Error compileHlslToSpirv(CString src, ShaderType shaderType, Bool compileWith16b
 	dxcArgs.emplaceBack("-Wfatal-errors");
 	dxcArgs.emplaceBack("-Wundef");
 	dxcArgs.emplaceBack("-Wno-unused-const-variable");
+	dxcArgs.emplaceBack("-Wno-unused-parameter");
 	dxcArgs.emplaceBack("-HV");
 	dxcArgs.emplaceBack("2021");
 	dxcArgs.emplaceBack("-E");

AnKi/Shaders/Functions.hlsl

@@ -738,3 +738,9 @@ void projectSphereView(Vec3 sphereCenter, F32 sphereRadius, F32 P00, F32 P11, ou
 	aabbMin = Vec2(minx * P00, miny * P11);
 	aabbMax = Vec2(maxx * P00, maxy * P11);
 }
+
+template<typename T>
+T barycentricInterpolation(T a, T b, T c, Vec3 barycentrics)
+{
+	return a * barycentrics.x + b * barycentrics.y + c * barycentrics.z;
+}

AnKi/Shaders/GBufferGeneric.ankiprog

@@ -408,7 +408,7 @@ main(U32 svGroupId : SV_GROUPID, U32 svGroupIndex : SV_GROUPINDEX, in payload Ta
 		{
 			MeshPerVertOut output;
 
-			UnpackedMeshVertex vert = loadVertex(meshlet, idx, ANKI_BONES, meshlet.m_positionScale, meshlet.m_positionTranslation);
+			UnpackedMeshVertex vert = loadVertex(meshlet, idx, ANKI_BONES);
 
 			const Mat3x4 worldTransform = g_gpuScene.Load<Mat3x4>(payload.m_worldTransformsOffset);
 			const Mat3x4 prevWorldTransform = g_gpuScene.Load<Mat3x4>(payload.m_worldTransformsOffset + sizeof(Mat3x4));
@@ -453,7 +453,7 @@ main(U32 svGroupId : SV_GROUPID, U32 svGroupIndex : SV_GROUPINDEX, in payload Ta
 #	endif
 
 	// Write the primitive data
-	const U32 primLoopCount = kMaxPrimitivesPerMeshlet / ANKI_MESH_SHADER_THREADGROUP_SIZE;
+	const U32 primLoopCount = (kMaxPrimitivesPerMeshlet + ANKI_MESH_SHADER_THREADGROUP_SIZE - 1u) / ANKI_MESH_SHADER_THREADGROUP_SIZE;
 	[unroll] for(U32 l = 0; l < primLoopCount; ++l)
 	{
 		const U32 idx = l * ANKI_MESH_SHADER_THREADGROUP_SIZE + svGroupIndex;

AnKi/Shaders/Include/Common.h

@@ -786,13 +786,12 @@ constexpr F32 kShadowsPolygonOffsetUnits = 2.75f;
 
 constexpr U32 kMaxMipsSinglePassDownsamplerCanProduce = 12u;
 
-constexpr U32 kMaxPrimitivesPerMeshlet = 128;
+constexpr U32 kMaxPrimitivesPerMeshlet = 124; ///< nVidia prefers 126 but meshoptimizer choks with that value.
 constexpr U32 kMaxVerticesPerMeshlet = 128;
 #define ANKI_TASK_SHADER_THREADGROUP_SIZE 64u
 constexpr U32 kMeshletGroupSize = ANKI_TASK_SHADER_THREADGROUP_SIZE;
 
 #define ANKI_MESH_SHADER_THREADGROUP_SIZE 32u
-static_assert(kMaxPrimitivesPerMeshlet % ANKI_MESH_SHADER_THREADGROUP_SIZE == 0);
 static_assert(kMaxVerticesPerMeshlet % ANKI_MESH_SHADER_THREADGROUP_SIZE == 0);
 
 /// Assume that a render state bucket can't go beyond 100M triangles. This helps ground some memory allocations.

AnKi/Shaders/MaterialShadersCommon.hlsl

@@ -66,11 +66,11 @@ UnpackedMeshVertex loadVertex(GpuSceneMeshLod mlod, U32 svVertexId, Bool bones)
 	return v;
 }
 
-UnpackedMeshVertex loadVertex(MeshletGeometryDescriptor meshlet, U32 vertexIndex, Bool bones, F32 positionScale, Vec3 positionTranslation)
+UnpackedMeshVertex loadVertex(MeshletGeometryDescriptor meshlet, U32 vertexIndex, Bool bones)
 {
 	UnpackedMeshVertex v;
 	v.m_position = g_unifiedGeom_R16G16B16A16_Unorm[meshlet.m_vertexOffsets[(U32)VertexStreamId::kPosition] + vertexIndex];
-	v.m_position = v.m_position * positionScale + positionTranslation;
+	v.m_position = v.m_position * meshlet.m_positionScale + meshlet.m_positionTranslation;
 
 	v.m_normal = g_unifiedGeom_R8G8B8A8_Snorm[meshlet.m_vertexOffsets[(U32)VertexStreamId::kNormal] + vertexIndex].xyz;
 	v.m_uv = g_unifiedGeom_R32G32_Sfloat[meshlet.m_vertexOffsets[(U32)VertexStreamId::kUv] + vertexIndex];