Files
FlaxEngine/Source/Engine/Graphics/Models/Mesh.cpp
2024-04-22 18:56:54 +02:00

853 lines
30 KiB
C++

// Copyright (c) 2012-2024 Wojciech Figat. All rights reserved.
#include "Mesh.h"
#include "MeshDeformation.h"
#include "ModelInstanceEntry.h"
#include "Engine/Content/Assets/Material.h"
#include "Engine/Content/Assets/Model.h"
#include "Engine/Core/Log.h"
#include "Engine/Core/Math/Transform.h"
#include "Engine/Graphics/GPUContext.h"
#include "Engine/Graphics/GPUDevice.h"
#include "Engine/Graphics/RenderTask.h"
#include "Engine/Graphics/RenderTools.h"
#include "Engine/Profiler/ProfilerCPU.h"
#include "Engine/Renderer/RenderList.h"
#include "Engine/Scripting/ManagedCLR/MCore.h"
#include "Engine/Serialization/MemoryReadStream.h"
#include "Engine/Threading/Task.h"
#include "Engine/Threading/Threading.h"
#if USE_EDITOR
#include "Engine/Renderer/GBufferPass.h"
#endif
namespace
{
template<typename IndexType>
bool UpdateMesh(Mesh* mesh, uint32 vertexCount, uint32 triangleCount, const Float3* vertices, const IndexType* triangles, const Float3* normals, const Float3* tangents, const Float2* uvs, const Color32* colors)
{
auto model = mesh->GetModel();
CHECK_RETURN(model && model->IsVirtual(), true);
CHECK_RETURN(triangles && vertices, true);
// Pack mesh data into vertex buffers
Array<VB1ElementType> vb1;
Array<VB2ElementType> vb2;
vb1.Resize(vertexCount);
if (normals)
{
if (tangents)
{
for (uint32 i = 0; i < vertexCount; i++)
{
const Float3 normal = normals[i];
const Float3 tangent = tangents[i];
auto& v = vb1.Get()[i];
RenderTools::CalculateTangentFrame(v.Normal, v.Tangent, normal, tangent);
}
}
else
{
for (uint32 i = 0; i < vertexCount; i++)
{
const Float3 normal = normals[i];
auto& v = vb1.Get()[i];
RenderTools::CalculateTangentFrame(v.Normal, v.Tangent, normal);
}
}
}
else
{
// Set default tangent frame
const auto n = Float1010102(Float3::UnitZ);
const auto t = Float1010102(Float3::UnitX);
for (uint32 i = 0; i < vertexCount; i++)
{
vb1.Get()[i].Normal = n;
vb1.Get()[i].Tangent = t;
}
}
if (uvs)
{
for (uint32 i = 0; i < vertexCount; i++)
vb1.Get()[i].TexCoord = Half2(uvs[i]);
}
else
{
auto v = Half2::Zero;
for (uint32 i = 0; i < vertexCount; i++)
vb1.Get()[i].TexCoord = v;
}
{
auto v = Half2::Zero;
for (uint32 i = 0; i < vertexCount; i++)
vb1.Get()[i].LightmapUVs = v;
}
if (colors)
{
vb2.Resize(vertexCount);
for (uint32 i = 0; i < vertexCount; i++)
vb2.Get()[i].Color = colors[i];
}
return mesh->UpdateMesh(vertexCount, triangleCount, (VB0ElementType*)vertices, vb1.Get(), vb2.HasItems() ? vb2.Get() : nullptr, triangles);
}
#if !COMPILE_WITHOUT_CSHARP
template<typename IndexType>
bool UpdateMesh(Mesh* mesh, uint32 vertexCount, uint32 triangleCount, const MArray* verticesObj, const MArray* trianglesObj, const MArray* normalsObj, const MArray* tangentsObj, const MArray* uvObj, const MArray* colorsObj)
{
ASSERT((uint32)MCore::Array::GetLength(verticesObj) >= vertexCount);
ASSERT((uint32)MCore::Array::GetLength(trianglesObj) / 3 >= triangleCount);
auto vertices = MCore::Array::GetAddress<Float3>(verticesObj);
auto triangles = MCore::Array::GetAddress<IndexType>(trianglesObj);
const auto normals = normalsObj ? MCore::Array::GetAddress<Float3>(normalsObj) : nullptr;
const auto tangents = tangentsObj ? MCore::Array::GetAddress<Float3>(tangentsObj) : nullptr;
const auto uvs = uvObj ? MCore::Array::GetAddress<Float2>(uvObj) : nullptr;
const auto colors = colorsObj ? MCore::Array::GetAddress<Color32>(colorsObj) : nullptr;
return UpdateMesh<IndexType>(mesh, vertexCount, triangleCount, vertices, triangles, normals, tangents, uvs, colors);
}
template<typename IndexType>
bool UpdateTriangles(Mesh* mesh, int32 triangleCount, const MArray* trianglesObj)
{
const auto model = mesh->GetModel();
ASSERT(model && model->IsVirtual() && trianglesObj);
// Get buffer data
ASSERT(MCore::Array::GetLength(trianglesObj) / 3 >= triangleCount);
auto ib = MCore::Array::GetAddress<IndexType>(trianglesObj);
return mesh->UpdateTriangles(triangleCount, ib);
}
#endif
}
bool Mesh::HasVertexColors() const
{
return _vertexBuffers[2] != nullptr && _vertexBuffers[2]->IsAllocated();
}
bool Mesh::UpdateMesh(uint32 vertexCount, uint32 triangleCount, const VB0ElementType* vb0, const VB1ElementType* vb1, const VB2ElementType* vb2, const void* ib, bool use16BitIndices)
{
auto model = (Model*)_model;
Unload();
// Setup GPU resources
model->LODs[_lodIndex]._verticesCount -= _vertices;
const bool failed = Load(vertexCount, triangleCount, vb0, vb1, vb2, ib, use16BitIndices);
if (!failed)
{
model->LODs[_lodIndex]._verticesCount += _vertices;
// Calculate mesh bounds
BoundingBox bounds;
BoundingBox::FromPoints((const Float3*)vb0, vertexCount, bounds);
SetBounds(bounds);
// Send event (actors using this model can update bounds, etc.)
model->onLoaded();
}
return failed;
}
bool Mesh::UpdateMesh(uint32 vertexCount, uint32 triangleCount, const Float3* vertices, const uint16* triangles, const Float3* normals, const Float3* tangents, const Float2* uvs, const Color32* colors)
{
return ::UpdateMesh<uint16>(this, vertexCount, triangleCount, vertices, triangles, normals, tangents, uvs, colors);
}
bool Mesh::UpdateMesh(uint32 vertexCount, uint32 triangleCount, const Float3* vertices, const uint32* triangles, const Float3* normals, const Float3* tangents, const Float2* uvs, const Color32* colors)
{
return ::UpdateMesh<uint32>(this, vertexCount, triangleCount, vertices, triangles, normals, tangents, uvs, colors);
}
bool Mesh::UpdateTriangles(uint32 triangleCount, const void* ib, bool use16BitIndices)
{
// Cache data
uint32 indicesCount = triangleCount * 3;
uint32 ibStride = use16BitIndices ? sizeof(uint16) : sizeof(uint32);
// Create index buffer
GPUBuffer* indexBuffer = GPUDevice::Instance->CreateBuffer(String::Empty);
if (indexBuffer->Init(GPUBufferDescription::Index(ibStride, indicesCount, ib)))
{
Delete(indexBuffer);
return true;
}
// TODO: update collision proxy
// Initialize
SAFE_DELETE_GPU_RESOURCE(_indexBuffer);
_indexBuffer = indexBuffer;
_triangles = triangleCount;
_use16BitIndexBuffer = use16BitIndices;
return false;
}
void Mesh::Init(Model* model, int32 lodIndex, int32 index, int32 materialSlotIndex, const BoundingBox& box, const BoundingSphere& sphere, bool hasLightmapUVs)
{
_model = model;
_lodIndex = lodIndex;
_index = index;
_materialSlotIndex = materialSlotIndex;
_use16BitIndexBuffer = false;
_hasLightmapUVs = hasLightmapUVs;
_box = box;
_sphere = sphere;
_vertices = 0;
_triangles = 0;
_vertexBuffers[0] = nullptr;
_vertexBuffers[1] = nullptr;
_vertexBuffers[2] = nullptr;
_indexBuffer = nullptr;
}
Mesh::~Mesh()
{
// Release buffers
SAFE_DELETE_GPU_RESOURCE(_vertexBuffers[0]);
SAFE_DELETE_GPU_RESOURCE(_vertexBuffers[1]);
SAFE_DELETE_GPU_RESOURCE(_vertexBuffers[2]);
SAFE_DELETE_GPU_RESOURCE(_indexBuffer);
}
bool Mesh::Load(uint32 vertices, uint32 triangles, const void* vb0, const void* vb1, const void* vb2, const void* ib, bool use16BitIndexBuffer)
{
// Cache data
uint32 indicesCount = triangles * 3;
uint32 ibStride = use16BitIndexBuffer ? sizeof(uint16) : sizeof(uint32);
GPUBuffer* vertexBuffer0 = nullptr;
GPUBuffer* vertexBuffer1 = nullptr;
GPUBuffer* vertexBuffer2 = nullptr;
GPUBuffer* indexBuffer = nullptr;
// Create GPU buffers
#if GPU_ENABLE_RESOURCE_NAMING
#define MESH_BUFFER_NAME(postfix) GetModel()->GetPath() + TEXT(postfix)
#else
#define MESH_BUFFER_NAME(postfix) String::Empty
#endif
vertexBuffer0 = GPUDevice::Instance->CreateBuffer(MESH_BUFFER_NAME(".VB0"));
if (vertexBuffer0->Init(GPUBufferDescription::Vertex(sizeof(VB0ElementType), vertices, vb0)))
goto ERROR_LOAD_END;
vertexBuffer1 = GPUDevice::Instance->CreateBuffer(MESH_BUFFER_NAME(".VB1"));
if (vertexBuffer1->Init(GPUBufferDescription::Vertex(sizeof(VB1ElementType), vertices, vb1)))
goto ERROR_LOAD_END;
if (vb2)
{
vertexBuffer2 = GPUDevice::Instance->CreateBuffer(MESH_BUFFER_NAME(".VB2"));
if (vertexBuffer2->Init(GPUBufferDescription::Vertex(sizeof(VB2ElementType), vertices, vb2)))
goto ERROR_LOAD_END;
}
indexBuffer = GPUDevice::Instance->CreateBuffer(MESH_BUFFER_NAME(".IB"));
if (indexBuffer->Init(GPUBufferDescription::Index(ibStride, indicesCount, ib)))
goto ERROR_LOAD_END;
// Init collision proxy
#if USE_PRECISE_MESH_INTERSECTS
if (!_collisionProxy.HasData())
{
if (use16BitIndexBuffer)
_collisionProxy.Init<uint16>(vertices, triangles, (Float3*)vb0, (uint16*)ib);
else
_collisionProxy.Init<uint32>(vertices, triangles, (Float3*)vb0, (uint32*)ib);
}
#endif
// Initialize
_vertexBuffers[0] = vertexBuffer0;
_vertexBuffers[1] = vertexBuffer1;
_vertexBuffers[2] = vertexBuffer2;
_indexBuffer = indexBuffer;
_triangles = triangles;
_vertices = vertices;
_use16BitIndexBuffer = use16BitIndexBuffer;
_cachedVertexBuffer[0].Clear();
_cachedVertexBuffer[1].Clear();
_cachedVertexBuffer[2].Clear();
return false;
#undef MESH_BUFFER_NAME
ERROR_LOAD_END:
SAFE_DELETE_GPU_RESOURCE(vertexBuffer0);
SAFE_DELETE_GPU_RESOURCE(vertexBuffer1);
SAFE_DELETE_GPU_RESOURCE(vertexBuffer2);
SAFE_DELETE_GPU_RESOURCE(indexBuffer);
return true;
}
void Mesh::Unload()
{
SAFE_DELETE_GPU_RESOURCE(_vertexBuffers[0]);
SAFE_DELETE_GPU_RESOURCE(_vertexBuffers[1]);
SAFE_DELETE_GPU_RESOURCE(_vertexBuffers[2]);
SAFE_DELETE_GPU_RESOURCE(_indexBuffer);
_triangles = 0;
_vertices = 0;
_use16BitIndexBuffer = false;
_cachedIndexBuffer.Resize(0);
_cachedVertexBuffer[0].Clear();
_cachedVertexBuffer[1].Clear();
_cachedVertexBuffer[2].Clear();
}
bool Mesh::Intersects(const Ray& ray, const Matrix& world, Real& distance, Vector3& normal) const
{
// Get bounding box of the mesh bounds transformed by the instance world matrix
Vector3 corners[8];
_box.GetCorners(corners);
Vector3 tmp;
Vector3::Transform(corners[0], world, tmp);
Vector3 min = tmp;
Vector3 max = tmp;
for (int32 i = 1; i < 8; i++)
{
Vector3::Transform(corners[i], world, tmp);
min = Vector3::Min(min, tmp);
max = Vector3::Max(max, tmp);
}
const BoundingBox transformedBox(min, max);
// Test ray on box
#if USE_PRECISE_MESH_INTERSECTS
if (transformedBox.Intersects(ray, distance))
{
// Use exact test on raw geometry
return _collisionProxy.Intersects(ray, world, distance, normal);
}
distance = 0;
normal = Vector3::Up;
return false;
#else
return transformedBox.Intersects(ray, distance, normal);
#endif
}
bool Mesh::Intersects(const Ray& ray, const Transform& transform, Real& distance, Vector3& normal) const
{
// Get bounding box of the mesh bounds transformed by the instance world matrix
Vector3 corners[8];
_box.GetCorners(corners);
Vector3 tmp;
transform.LocalToWorld(corners[0], tmp);
Vector3 min = tmp;
Vector3 max = tmp;
for (int32 i = 1; i < 8; i++)
{
transform.LocalToWorld(corners[i], tmp);
min = Vector3::Min(min, tmp);
max = Vector3::Max(max, tmp);
}
const BoundingBox transformedBox(min, max);
// Test ray on box
#if USE_PRECISE_MESH_INTERSECTS
if (transformedBox.Intersects(ray, distance))
{
// Use exact test on raw geometry
return _collisionProxy.Intersects(ray, transform, distance, normal);
}
distance = 0;
normal = Vector3::Up;
return false;
#else
return transformedBox.Intersects(ray, distance, normal);
#endif
}
void Mesh::GetDrawCallGeometry(DrawCall& drawCall) const
{
drawCall.Geometry.IndexBuffer = _indexBuffer;
drawCall.Geometry.VertexBuffers[0] = _vertexBuffers[0];
drawCall.Geometry.VertexBuffers[1] = _vertexBuffers[1];
drawCall.Geometry.VertexBuffers[2] = _vertexBuffers[2];
drawCall.Geometry.VertexBuffersOffsets[0] = 0;
drawCall.Geometry.VertexBuffersOffsets[1] = 0;
drawCall.Geometry.VertexBuffersOffsets[2] = 0;
drawCall.Draw.StartIndex = 0;
drawCall.Draw.IndicesCount = _triangles * 3;
}
void Mesh::Render(GPUContext* context) const
{
if (!IsInitialized())
return;
context->BindVB(ToSpan((GPUBuffer**)_vertexBuffers, 3));
context->BindIB(_indexBuffer);
context->DrawIndexedInstanced(_triangles * 3, 1, 0, 0, 0);
}
void Mesh::Draw(const RenderContext& renderContext, MaterialBase* material, const Matrix& world, StaticFlags flags, bool receiveDecals, DrawPass drawModes, float perInstanceRandom, int16 sortOrder) const
{
if (!material || !material->IsSurface() || !IsInitialized())
return;
drawModes &= material->GetDrawModes();
if (drawModes == DrawPass::None)
return;
// Setup draw call
DrawCall drawCall;
drawCall.Geometry.IndexBuffer = _indexBuffer;
drawCall.Geometry.VertexBuffers[0] = _vertexBuffers[0];
drawCall.Geometry.VertexBuffers[1] = _vertexBuffers[1];
drawCall.Geometry.VertexBuffers[2] = _vertexBuffers[2];
drawCall.Draw.IndicesCount = _triangles * 3;
drawCall.InstanceCount = 1;
drawCall.Material = material;
drawCall.World = world;
drawCall.ObjectPosition = drawCall.World.GetTranslation();
drawCall.ObjectRadius = _sphere.Radius * drawCall.World.GetScaleVector().GetAbsolute().MaxValue();
drawCall.Surface.GeometrySize = _box.GetSize();
drawCall.Surface.PrevWorld = world;
drawCall.Surface.Lightmap = nullptr;
drawCall.Surface.LightmapUVsArea = Rectangle::Empty;
drawCall.Surface.Skinning = nullptr;
drawCall.Surface.LODDitherFactor = 0.0f;
drawCall.WorldDeterminantSign = Math::FloatSelect(world.RotDeterminant(), 1, -1);
drawCall.PerInstanceRandom = perInstanceRandom;
#if USE_EDITOR
const ViewMode viewMode = renderContext.View.Mode;
if (viewMode == ViewMode::LightmapUVsDensity || viewMode == ViewMode::LODPreview)
GBufferPass::AddIndexBufferToModelLOD(_indexBuffer, &((Model*)_model)->LODs[_lodIndex]);
#endif
// Push draw call to the render list
renderContext.List->AddDrawCall(renderContext, drawModes, flags, drawCall, receiveDecals, sortOrder);
}
void Mesh::Draw(const RenderContext& renderContext, const DrawInfo& info, float lodDitherFactor) const
{
const auto& entry = info.Buffer->At(_materialSlotIndex);
if (!entry.Visible || !IsInitialized())
return;
const MaterialSlot& slot = _model->MaterialSlots[_materialSlotIndex];
// Select material
MaterialBase* material;
if (entry.Material && entry.Material->IsLoaded())
material = entry.Material;
else if (slot.Material && slot.Material->IsLoaded())
material = slot.Material;
else
material = GPUDevice::Instance->GetDefaultMaterial();
if (!material || !material->IsSurface())
return;
// Check if skip rendering
const auto shadowsMode = entry.ShadowsMode & slot.ShadowsMode;
const auto drawModes = info.DrawModes & renderContext.View.Pass & renderContext.View.GetShadowsDrawPassMask(shadowsMode) & material->GetDrawModes();
if (drawModes == DrawPass::None)
return;
// Setup draw call
DrawCall drawCall;
drawCall.Geometry.IndexBuffer = _indexBuffer;
drawCall.Geometry.VertexBuffers[0] = _vertexBuffers[0];
drawCall.Geometry.VertexBuffers[1] = _vertexBuffers[1];
drawCall.Geometry.VertexBuffers[2] = _vertexBuffers[2];
if (info.Deformation)
{
info.Deformation->RunDeformers(this, MeshBufferType::Vertex0, drawCall.Geometry.VertexBuffers[0]);
info.Deformation->RunDeformers(this, MeshBufferType::Vertex1, drawCall.Geometry.VertexBuffers[1]);
}
if (info.VertexColors && info.VertexColors[_lodIndex])
{
// TODO: cache vertexOffset within the model LOD per-mesh
uint32 vertexOffset = 0;
for (int32 meshIndex = 0; meshIndex < _index; meshIndex++)
vertexOffset += ((Model*)_model)->LODs[_lodIndex].Meshes[meshIndex].GetVertexCount();
drawCall.Geometry.VertexBuffers[2] = info.VertexColors[_lodIndex];
drawCall.Geometry.VertexBuffersOffsets[2] = vertexOffset * sizeof(VB2ElementType);
}
drawCall.Draw.IndicesCount = _triangles * 3;
drawCall.InstanceCount = 1;
drawCall.Material = material;
drawCall.World = *info.World;
drawCall.ObjectPosition = drawCall.World.GetTranslation();
drawCall.ObjectRadius = info.Bounds.Radius; // TODO: should it be kept in sync with ObjectPosition?
drawCall.Surface.GeometrySize = _box.GetSize();
drawCall.Surface.PrevWorld = info.DrawState->PrevWorld;
drawCall.Surface.Lightmap = (info.Flags & StaticFlags::Lightmap) != StaticFlags::None ? info.Lightmap : nullptr;
drawCall.Surface.LightmapUVsArea = info.LightmapUVs ? *info.LightmapUVs : Rectangle::Empty;
drawCall.Surface.Skinning = nullptr;
drawCall.Surface.LODDitherFactor = lodDitherFactor;
drawCall.WorldDeterminantSign = Math::FloatSelect(drawCall.World.RotDeterminant(), 1, -1);
drawCall.PerInstanceRandom = info.PerInstanceRandom;
#if USE_EDITOR
const ViewMode viewMode = renderContext.View.Mode;
if (viewMode == ViewMode::LightmapUVsDensity || viewMode == ViewMode::LODPreview)
GBufferPass::AddIndexBufferToModelLOD(_indexBuffer, &((Model*)_model)->LODs[_lodIndex]);
#endif
// Push draw call to the render list
renderContext.List->AddDrawCall(renderContext, drawModes, info.Flags, drawCall, entry.ReceiveDecals, info.SortOrder);
}
void Mesh::Draw(const RenderContextBatch& renderContextBatch, const DrawInfo& info, float lodDitherFactor) const
{
const auto& entry = info.Buffer->At(_materialSlotIndex);
if (!entry.Visible || !IsInitialized())
return;
const MaterialSlot& slot = _model->MaterialSlots[_materialSlotIndex];
// Select material
MaterialBase* material;
if (entry.Material && entry.Material->IsLoaded())
material = entry.Material;
else if (slot.Material && slot.Material->IsLoaded())
material = slot.Material;
else
material = GPUDevice::Instance->GetDefaultMaterial();
if (!material || !material->IsSurface())
return;
// Setup draw call
DrawCall drawCall;
drawCall.Geometry.IndexBuffer = _indexBuffer;
drawCall.Geometry.VertexBuffers[0] = _vertexBuffers[0];
drawCall.Geometry.VertexBuffers[1] = _vertexBuffers[1];
drawCall.Geometry.VertexBuffers[2] = _vertexBuffers[2];
if (info.Deformation)
{
info.Deformation->RunDeformers(this, MeshBufferType::Vertex0, drawCall.Geometry.VertexBuffers[0]);
info.Deformation->RunDeformers(this, MeshBufferType::Vertex1, drawCall.Geometry.VertexBuffers[1]);
}
if (info.VertexColors && info.VertexColors[_lodIndex])
{
// TODO: cache vertexOffset within the model LOD per-mesh
uint32 vertexOffset = 0;
for (int32 meshIndex = 0; meshIndex < _index; meshIndex++)
vertexOffset += ((Model*)_model)->LODs[_lodIndex].Meshes[meshIndex].GetVertexCount();
drawCall.Geometry.VertexBuffers[2] = info.VertexColors[_lodIndex];
drawCall.Geometry.VertexBuffersOffsets[2] = vertexOffset * sizeof(VB2ElementType);
}
drawCall.Draw.IndicesCount = _triangles * 3;
drawCall.InstanceCount = 1;
drawCall.Material = material;
drawCall.World = *info.World;
drawCall.ObjectPosition = drawCall.World.GetTranslation();
drawCall.ObjectRadius = info.Bounds.Radius; // TODO: should it be kept in sync with ObjectPosition?
drawCall.Surface.GeometrySize = _box.GetSize();
drawCall.Surface.PrevWorld = info.DrawState->PrevWorld;
drawCall.Surface.Lightmap = (info.Flags & StaticFlags::Lightmap) != StaticFlags::None ? info.Lightmap : nullptr;
drawCall.Surface.LightmapUVsArea = info.LightmapUVs ? *info.LightmapUVs : Rectangle::Empty;
drawCall.Surface.Skinning = nullptr;
drawCall.Surface.LODDitherFactor = lodDitherFactor;
drawCall.WorldDeterminantSign = Math::FloatSelect(drawCall.World.RotDeterminant(), 1, -1);
drawCall.PerInstanceRandom = info.PerInstanceRandom;
#if USE_EDITOR
const ViewMode viewMode = renderContextBatch.GetMainContext().View.Mode;
if (viewMode == ViewMode::LightmapUVsDensity || viewMode == ViewMode::LODPreview)
GBufferPass::AddIndexBufferToModelLOD(_indexBuffer, &((Model*)_model)->LODs[_lodIndex]);
#endif
// Push draw call to the render lists
const auto shadowsMode = entry.ShadowsMode & slot.ShadowsMode;
const auto drawModes = info.DrawModes & material->GetDrawModes();
if (drawModes != DrawPass::None)
renderContextBatch.GetMainContext().List->AddDrawCall(renderContextBatch, drawModes, info.Flags, shadowsMode, info.Bounds, drawCall, entry.ReceiveDecals, info.SortOrder);
}
bool Mesh::DownloadDataGPU(MeshBufferType type, BytesContainer& result) const
{
GPUBuffer* buffer = nullptr;
switch (type)
{
case MeshBufferType::Index:
buffer = _indexBuffer;
break;
case MeshBufferType::Vertex0:
buffer = _vertexBuffers[0];
break;
case MeshBufferType::Vertex1:
buffer = _vertexBuffers[1];
break;
case MeshBufferType::Vertex2:
buffer = _vertexBuffers[2];
break;
}
return buffer && buffer->DownloadData(result);
}
Task* Mesh::DownloadDataGPUAsync(MeshBufferType type, BytesContainer& result) const
{
GPUBuffer* buffer = nullptr;
switch (type)
{
case MeshBufferType::Index:
buffer = _indexBuffer;
break;
case MeshBufferType::Vertex0:
buffer = _vertexBuffers[0];
break;
case MeshBufferType::Vertex1:
buffer = _vertexBuffers[1];
break;
case MeshBufferType::Vertex2:
buffer = _vertexBuffers[2];
break;
}
return buffer ? buffer->DownloadDataAsync(result) : nullptr;
}
bool Mesh::DownloadDataCPU(MeshBufferType type, BytesContainer& result, int32& count) const
{
if (_cachedVertexBuffer[0].IsEmpty())
{
PROFILE_CPU();
auto model = GetModel();
ScopeLock lock(model->Locker);
if (model->IsVirtual())
{
LOG(Error, "Cannot access CPU data of virtual models. Use GPU data download.");
return true;
}
// Fetch chunk with data from drive/memory
const auto chunkIndex = MODEL_LOD_TO_CHUNK_INDEX(GetLODIndex());
if (model->LoadChunk(chunkIndex))
return true;
const auto chunk = model->GetChunk(chunkIndex);
if (!chunk)
{
LOG(Error, "Missing chunk.");
return true;
}
MemoryReadStream stream(chunk->Get(), chunk->Size());
// Seek to find mesh location
for (int32 i = 0; i <= _index; i++)
{
// #MODEL_DATA_FORMAT_USAGE
uint32 vertices;
stream.ReadUint32(&vertices);
uint32 triangles;
stream.ReadUint32(&triangles);
uint32 indicesCount = triangles * 3;
bool use16BitIndexBuffer = indicesCount <= MAX_uint16;
uint32 ibStride = use16BitIndexBuffer ? sizeof(uint16) : sizeof(uint32);
if (vertices == 0 || triangles == 0)
{
LOG(Error, "Invalid mesh data.");
return true;
}
auto vb0 = stream.Move<VB0ElementType>(vertices);
auto vb1 = stream.Move<VB1ElementType>(vertices);
bool hasColors = stream.ReadBool();
VB2ElementType18* vb2 = nullptr;
if (hasColors)
{
vb2 = stream.Move<VB2ElementType18>(vertices);
}
auto ib = stream.Move<byte>(indicesCount * ibStride);
if (i != _index)
continue;
// Cache mesh data
_cachedIndexBufferCount = indicesCount;
_cachedIndexBuffer.Set(ib, indicesCount * ibStride);
_cachedVertexBuffer[0].Set((const byte*)vb0, vertices * sizeof(VB0ElementType));
_cachedVertexBuffer[1].Set((const byte*)vb1, vertices * sizeof(VB1ElementType));
if (hasColors)
_cachedVertexBuffer[2].Set((const byte*)vb2, vertices * sizeof(VB2ElementType));
break;
}
}
switch (type)
{
case MeshBufferType::Index:
result.Link(_cachedIndexBuffer);
count = _cachedIndexBufferCount;
break;
case MeshBufferType::Vertex0:
result.Link(_cachedVertexBuffer[0]);
count = _cachedVertexBuffer[0].Count() / sizeof(VB0ElementType);
break;
case MeshBufferType::Vertex1:
result.Link(_cachedVertexBuffer[1]);
count = _cachedVertexBuffer[1].Count() / sizeof(VB1ElementType);
break;
case MeshBufferType::Vertex2:
result.Link(_cachedVertexBuffer[2]);
count = _cachedVertexBuffer[2].Count() / sizeof(VB2ElementType);
break;
default:
return true;
}
return false;
}
ScriptingObject* Mesh::GetParentModel()
{
return _model;
}
#if !COMPILE_WITHOUT_CSHARP
bool Mesh::UpdateMeshUInt(int32 vertexCount, int32 triangleCount, const MArray* verticesObj, const MArray* trianglesObj, const MArray* normalsObj, const MArray* tangentsObj, const MArray* uvObj, const MArray* colorsObj)
{
return ::UpdateMesh<uint32>(this, (uint32)vertexCount, (uint32)triangleCount, verticesObj, trianglesObj, normalsObj, tangentsObj, uvObj, colorsObj);
}
bool Mesh::UpdateMeshUShort(int32 vertexCount, int32 triangleCount, const MArray* verticesObj, const MArray* trianglesObj, const MArray* normalsObj, const MArray* tangentsObj, const MArray* uvObj, const MArray* colorsObj)
{
return ::UpdateMesh<uint16>(this, (uint32)vertexCount, (uint32)triangleCount, verticesObj, trianglesObj, normalsObj, tangentsObj, uvObj, colorsObj);
}
bool Mesh::UpdateTrianglesUInt(int32 triangleCount, const MArray* trianglesObj)
{
return ::UpdateTriangles<uint32>(this, triangleCount, trianglesObj);
}
bool Mesh::UpdateTrianglesUShort(int32 triangleCount, const MArray* trianglesObj)
{
return ::UpdateTriangles<uint16>(this, triangleCount, trianglesObj);
}
enum class InternalBufferType
{
VB0 = 0,
VB1 = 1,
VB2 = 2,
IB16 = 3,
IB32 = 4,
};
MArray* Mesh::DownloadBuffer(bool forceGpu, MTypeObject* resultType, int32 typeI)
{
auto mesh = this;
auto type = (InternalBufferType)typeI;
auto model = mesh->GetModel();
ScopeLock lock(model->Locker);
// Virtual assets always fetch from GPU memory
forceGpu |= model->IsVirtual();
if (!mesh->IsInitialized() && forceGpu)
{
LOG(Error, "Cannot load mesh data from GPU if it's not loaded.");
return nullptr;
}
MeshBufferType bufferType;
switch (type)
{
case InternalBufferType::VB0:
bufferType = MeshBufferType::Vertex0;
break;
case InternalBufferType::VB1:
bufferType = MeshBufferType::Vertex1;
break;
case InternalBufferType::VB2:
bufferType = MeshBufferType::Vertex2;
break;
case InternalBufferType::IB16:
case InternalBufferType::IB32:
bufferType = MeshBufferType::Index;
break;
default:
return nullptr;
}
BytesContainer data;
int32 dataCount;
if (forceGpu)
{
// Get data from GPU
// TODO: support reusing the input memory buffer to perform a single copy from staging buffer to the input CPU buffer
auto task = mesh->DownloadDataGPUAsync(bufferType, data);
if (task == nullptr)
return nullptr;
task->Start();
model->Locker.Unlock();
if (task->Wait())
{
LOG(Error, "Task failed.");
return nullptr;
}
model->Locker.Lock();
// Extract elements count from result data
switch (bufferType)
{
case MeshBufferType::Index:
dataCount = data.Length() / (Use16BitIndexBuffer() ? sizeof(uint16) : sizeof(uint32));
break;
case MeshBufferType::Vertex0:
dataCount = data.Length() / sizeof(VB0ElementType);
break;
case MeshBufferType::Vertex1:
dataCount = data.Length() / sizeof(VB1ElementType);
break;
case MeshBufferType::Vertex2:
dataCount = data.Length() / sizeof(VB2ElementType);
break;
}
}
else
{
// Get data from CPU
if (DownloadDataCPU(bufferType, data, dataCount))
return nullptr;
}
// Convert into managed array
MArray* result = MCore::Array::New(MCore::Type::GetClass(INTERNAL_TYPE_OBJECT_GET(resultType)), dataCount);
void* managedArrayPtr = MCore::Array::GetAddress(result);
const int32 elementSize = data.Length() / dataCount;
switch (type)
{
case InternalBufferType::VB0:
case InternalBufferType::VB1:
case InternalBufferType::VB2:
{
Platform::MemoryCopy(managedArrayPtr, data.Get(), data.Length());
break;
}
case InternalBufferType::IB16:
{
if (elementSize == sizeof(uint16))
{
Platform::MemoryCopy(managedArrayPtr, data.Get(), data.Length());
}
else
{
auto dst = (uint16*)managedArrayPtr;
auto src = (uint32*)data.Get();
for (int32 i = 0; i < dataCount; i++)
dst[i] = src[i];
}
break;
}
case InternalBufferType::IB32:
{
if (elementSize == sizeof(uint16))
{
auto dst = (uint32*)managedArrayPtr;
auto src = (uint16*)data.Get();
for (int32 i = 0; i < dataCount; i++)
dst[i] = src[i];
}
else
{
Platform::MemoryCopy(managedArrayPtr, data.Get(), data.Length());
}
break;
}
}
return result;
}
#endif