urho3d/Source/Urho3D/Graphics/OcclusionBuffer.cpp

// Copyright (c) 2008-2022 the Urho3D project
// License: MIT

#include "../Precompiled.h"

#include "../Core/WorkQueue.h"
#include "../Core/Profiler.h"
#include "../Graphics/Camera.h"
#include "../Graphics/OcclusionBuffer.h"
#include "../IO/Log.h"

#include "../DebugNew.h"

namespace Urho3D
{

enum ClipMask : unsigned
{
    CLIPMASK_X_POS = 0x1,
    CLIPMASK_X_NEG = 0x2,
    CLIPMASK_Y_POS = 0x4,
    CLIPMASK_Y_NEG = 0x8,
    CLIPMASK_Z_POS = 0x10,
    CLIPMASK_Z_NEG = 0x20,
};
URHO3D_FLAGSET(ClipMask, ClipMaskFlags);

void DrawOcclusionBatchWork(const WorkItem* item, i32 threadIndex)
{
    auto* buffer = reinterpret_cast<OcclusionBuffer*>(item->aux_);
    OcclusionBatch& batch = *reinterpret_cast<OcclusionBatch*>(item->start_);
    buffer->DrawBatch(batch, threadIndex);
}

OcclusionBuffer::OcclusionBuffer(Context* context) :
    Object(context)
{
}

OcclusionBuffer::~OcclusionBuffer() = default;

bool OcclusionBuffer::SetSize(int width, int height, bool threaded)
{
    // Force the height to an even amount of pixels for better mip generation
    if (height & 1u)
        ++height;

    if (width == width_ && height == height_)
        return true;

    if (width <= 0 || height <= 0)
        return false;

    if (!IsPowerOfTwo((unsigned)width))
    {
        URHO3D_LOGERRORF("Requested occlusion buffer width %d is not a power of two", width);
        return false;
    }

    width_ = width;
    height_ = height;

    // Build work buffers for threading
    unsigned numThreadBuffers = threaded ? GetSubsystem<WorkQueue>()->GetNumThreads() + 1 : 1;
    buffers_.Resize(numThreadBuffers);
    for (unsigned i = 0; i < numThreadBuffers; ++i)
    {
        // Reserve extra memory in case 3D clipping is not exact
        OcclusionBufferData& buffer = buffers_[i];
        buffer.dataWithSafety_ = new int[width * (height + 2) + 2];
        buffer.data_ = buffer.dataWithSafety_.Get() + width + 1;
        buffer.used_ = false;
    }

    mipBuffers_.Clear();

    // Build buffers for mip levels
    for (;;)
    {
        width = (width + 1) / 2;
        height = (height + 1) / 2;

        mipBuffers_.Push(SharedArrayPtr<DepthValue>(new DepthValue[width * height]));

        if (width <= OCCLUSION_MIN_SIZE && height <= OCCLUSION_MIN_SIZE)
            break;
    }

    URHO3D_LOGDEBUG("Set occlusion buffer size " + String(width_) + "x" + String(height_) + " with " +
             String(mipBuffers_.Size()) + " mip levels and " + String(numThreadBuffers) + " thread buffers");

    CalculateViewport();
    return true;
}

void OcclusionBuffer::SetView(Camera* camera)
{
    if (!camera)
        return;

    view_ = camera->GetView();
    projection_ = camera->GetProjection();
    viewProj_ = projection_ * view_;
    nearClip_ = camera->GetNearClip();
    farClip_ = camera->GetFarClip();
    reverseCulling_ = camera->GetReverseCulling();
    CalculateViewport();
}

void OcclusionBuffer::SetMaxTriangles(unsigned triangles)
{
    maxTriangles_ = triangles;
}

void OcclusionBuffer::SetCullMode(CullMode mode)
{
    if (reverseCulling_)
    {
        if (mode == CULL_CW)
            mode = CULL_CCW;
        else if (mode == CULL_CCW)
            mode = CULL_CW;
    }
    cullMode_ = mode;
}

void OcclusionBuffer::Reset()
{
    numTriangles_ = 0;
    batches_.Clear();
}

void OcclusionBuffer::Clear()
{
    Reset();

    // Only clear the main thread buffer. Rest are cleared on-demand when drawing the first batch
    ClearBuffer(0);
    for (OcclusionBufferData& buffer : buffers_)
        buffer.used_ = false;

    depthHierarchyDirty_ = true;
}

bool OcclusionBuffer::AddTriangles(const Matrix3x4& model, const void* vertexData, unsigned vertexSize, unsigned vertexStart,
    unsigned vertexCount)
{
    batches_.Resize(batches_.Size() + 1);
    OcclusionBatch& batch = batches_.Back();

    batch.model_ = model;
    batch.vertexData_ = vertexData;
    batch.vertexSize_ = vertexSize;
    batch.indexData_ = nullptr;
    batch.indexSize_ = 0;
    batch.drawStart_ = vertexStart;
    batch.drawCount_ = vertexCount;

    numTriangles_ += vertexCount / 3;
    return numTriangles_ <= maxTriangles_;
}

bool OcclusionBuffer::AddTriangles(const Matrix3x4& model, const void* vertexData, unsigned vertexSize, const void* indexData,
    unsigned indexSize, unsigned indexStart, unsigned indexCount)
{
    batches_.Resize(batches_.Size() + 1);
    OcclusionBatch& batch = batches_.Back();

    batch.model_ = model;
    batch.vertexData_ = vertexData;
    batch.vertexSize_ = vertexSize;
    batch.indexData_ = indexData;
    batch.indexSize_ = indexSize;
    batch.drawStart_ = indexStart;
    batch.drawCount_ = indexCount;

    numTriangles_ += indexCount / 3;
    return numTriangles_ <= maxTriangles_;
}

void OcclusionBuffer::DrawTriangles()
{
    if (buffers_.Size() == 1)
    {
        // Not threaded
        for (Vector<OcclusionBatch>::Iterator i = batches_.Begin(); i != batches_.End(); ++i)
            DrawBatch(*i, 0);

        depthHierarchyDirty_ = true;
    }
    else if (buffers_.Size() > 1)
    {
        // Threaded
        auto* queue = GetSubsystem<WorkQueue>();

        for (Vector<OcclusionBatch>::Iterator i = batches_.Begin(); i != batches_.End(); ++i)
        {
            SharedPtr<WorkItem> item = queue->GetFreeItem();
            item->priority_ = WI_MAX_PRIORITY;
            item->workFunction_ = DrawOcclusionBatchWork;
            item->aux_ = this;
            item->start_ = &(*i);
            queue->AddWorkItem(item);
        }

        queue->Complete(WI_MAX_PRIORITY);

        MergeBuffers();
        depthHierarchyDirty_ = true;
    }

    batches_.Clear();
}

void OcclusionBuffer::BuildDepthHierarchy()
{
    if (buffers_.Empty() || !depthHierarchyDirty_)
        return;

    URHO3D_PROFILE(BuildDepthHierarchy);

    // Build the first mip level from the pixel-level data
    int width = (width_ + 1) / 2;
    int height = (height_ + 1) / 2;
    if (mipBuffers_.Size())
    {
        for (int y = 0; y < height; ++y)
        {
            int* src = buffers_[0].data_ + (y * 2) * width_;
            DepthValue* dest = mipBuffers_[0].Get() + y * width;
            DepthValue* end = dest + width;

            if (y * 2 + 1 < height_)
            {
                int* src2 = src + width_;
                while (dest < end)
                {
                    int minUpper = Min(src[0], src[1]);
                    int minLower = Min(src2[0], src2[1]);
                    dest->min_ = Min(minUpper, minLower);
                    int maxUpper = Max(src[0], src[1]);
                    int maxLower = Max(src2[0], src2[1]);
                    dest->max_ = Max(maxUpper, maxLower);

                    src += 2;
                    src2 += 2;
                    ++dest;
                }
            }
            else
            {
                while (dest < end)
                {
                    dest->min_ = Min(src[0], src[1]);
                    dest->max_ = Max(src[0], src[1]);

                    src += 2;
                    ++dest;
                }
            }
        }
    }

    // Build the rest of the mip levels
    for (unsigned i = 1; i < mipBuffers_.Size(); ++i)
    {
        int prevWidth = width;
        int prevHeight = height;
        width = (width + 1) / 2;
        height = (height + 1) / 2;

        for (int y = 0; y < height; ++y)
        {
            DepthValue* src = mipBuffers_[i - 1].Get() + (y * 2) * prevWidth;
            DepthValue* dest = mipBuffers_[i].Get() + y * width;
            DepthValue* end = dest + width;

            if (y * 2 + 1 < prevHeight)
            {
                DepthValue* src2 = src + prevWidth;
                while (dest < end)
                {
                    int minUpper = Min(src[0].min_, src[1].min_);
                    int minLower = Min(src2[0].min_, src2[1].min_);
                    dest->min_ = Min(minUpper, minLower);
                    int maxUpper = Max(src[0].max_, src[1].max_);
                    int maxLower = Max(src2[0].max_, src2[1].max_);
                    dest->max_ = Max(maxUpper, maxLower);

                    src += 2;
                    src2 += 2;
                    ++dest;
                }
            }
            else
            {
                while (dest < end)
                {
                    dest->min_ = Min(src[0].min_, src[1].min_);
                    dest->max_ = Max(src[0].max_, src[1].max_);

                    src += 2;
                    ++dest;
                }
            }
        }
    }

    depthHierarchyDirty_ = false;
}

void OcclusionBuffer::ResetUseTimer()
{
    useTimer_.Reset();
}

bool OcclusionBuffer::IsVisible(const BoundingBox& worldSpaceBox) const
{
    if (buffers_.Empty())
        return true;

    // Transform corners to projection space
    Vector4 vertices[8];
    vertices[0] = ModelTransform(viewProj_, worldSpaceBox.min_);
    vertices[1] = ModelTransform(viewProj_, Vector3(worldSpaceBox.max_.x_, worldSpaceBox.min_.y_, worldSpaceBox.min_.z_));
    vertices[2] = ModelTransform(viewProj_, Vector3(worldSpaceBox.min_.x_, worldSpaceBox.max_.y_, worldSpaceBox.min_.z_));
    vertices[3] = ModelTransform(viewProj_, Vector3(worldSpaceBox.max_.x_, worldSpaceBox.max_.y_, worldSpaceBox.min_.z_));
    vertices[4] = ModelTransform(viewProj_, Vector3(worldSpaceBox.min_.x_, worldSpaceBox.min_.y_, worldSpaceBox.max_.z_));
    vertices[5] = ModelTransform(viewProj_, Vector3(worldSpaceBox.max_.x_, worldSpaceBox.min_.y_, worldSpaceBox.max_.z_));
    vertices[6] = ModelTransform(viewProj_, Vector3(worldSpaceBox.min_.x_, worldSpaceBox.max_.y_, worldSpaceBox.max_.z_));
    vertices[7] = ModelTransform(viewProj_, worldSpaceBox.max_);

    // Apply a far clip relative bias
    for (auto& vertice : vertices)
        vertice.z_ -= OCCLUSION_RELATIVE_BIAS;

    // Transform to screen space. If any of the corners cross the near plane, assume visible
    float minX, maxX, minY, maxY, minZ;

    if (vertices[0].z_ <= 0.0f)
        return true;

    Vector3 projected = ViewportTransform(vertices[0]);
    minX = maxX = projected.x_;
    minY = maxY = projected.y_;
    minZ = projected.z_;

    // Project the rest
    for (unsigned i = 1; i < 8; ++i)
    {
        if (vertices[i].z_ <= 0.0f)
            return true;

        projected = ViewportTransform(vertices[i]);

        if (projected.x_ < minX) minX = projected.x_;
        if (projected.x_ > maxX) maxX = projected.x_;
        if (projected.y_ < minY) minY = projected.y_;
        if (projected.y_ > maxY) maxY = projected.y_;
        if (projected.z_ < minZ) minZ = projected.z_;
    }

    // Expand the bounding box 1 pixel in each direction to be conservative and correct rasterization offset
    IntRect rect((int)(minX - 1.5f), (int)(minY - 1.5f), RoundToInt(maxX), RoundToInt(maxY));

    // If the rect is outside, let frustum culling handle
    if (rect.right_ < 0 || rect.bottom_ < 0)
        return true;
    if (rect.left_ >= width_ || rect.top_ >= height_)
        return true;

    // Clipping of rect
    if (rect.left_ < 0)
        rect.left_ = 0;
    if (rect.top_ < 0)
        rect.top_ = 0;
    if (rect.right_ >= width_)
        rect.right_ = width_ - 1;
    if (rect.bottom_ >= height_)
        rect.bottom_ = height_ - 1;

    // Convert depth to integer and apply final bias
    int z = RoundToInt(minZ) - OCCLUSION_FIXED_BIAS;

    if (!depthHierarchyDirty_)
    {
        // Start from lowest mip level and check if a conclusive result can be found
        for (int i = mipBuffers_.Size() - 1; i >= 0; --i)
        {
            int shift = i + 1;
            int width = width_ >> shift;
            int left = rect.left_ >> shift;
            int right = rect.right_ >> shift;

            DepthValue* buffer = mipBuffers_[i].Get();
            DepthValue* row = buffer + (rect.top_ >> shift) * width;
            DepthValue* endRow = buffer + (rect.bottom_ >> shift) * width;
            bool allOccluded = true;

            while (row <= endRow)
            {
                DepthValue* src = row + left;
                DepthValue* end = row + right;
                while (src <= end)
                {
                    if (z <= src->min_)
                        return true;
                    if (z <= src->max_)
                        allOccluded = false;
                    ++src;
                }
                row += width;
            }

            if (allOccluded)
                return false;
        }
    }

    // If no conclusive result, finally check the pixel-level data
    int* row = buffers_[0].data_ + rect.top_ * width_;
    int* endRow = buffers_[0].data_ + rect.bottom_ * width_;
    while (row <= endRow)
    {
        int* src = row + rect.left_;
        int* end = row + rect.right_;
        while (src <= end)
        {
            if (z <= *src)
                return true;
            ++src;
        }
        row += width_;
    }

    return false;
}

unsigned OcclusionBuffer::GetUseTimer()
{
    return useTimer_.GetMSec(false);
}


void OcclusionBuffer::DrawBatch(const OcclusionBatch& batch, i32 threadIndex)
{
    assert(threadIndex >= 0);

    // If buffer not yet used, clear it
    if (threadIndex > 0 && !buffers_[threadIndex].used_)
    {
        ClearBuffer(threadIndex);
        buffers_[threadIndex].used_ = true;
    }

    Matrix4 modelViewProj = viewProj_ * batch.model_;

    // Theoretical max. amount of vertices if each of the 6 clipping planes doubles the triangle count
    Vector4 vertices[64 * 3];

    if (!batch.indexData_)
    {
        const unsigned char* srcData = ((const unsigned char*)batch.vertexData_) + batch.drawStart_ * batch.vertexSize_;

        unsigned index = 0;
        while (index + 2 < batch.drawCount_)
        {
            const Vector3& v0 = *((const Vector3*)(&srcData[index * batch.vertexSize_]));
            const Vector3& v1 = *((const Vector3*)(&srcData[(index + 1) * batch.vertexSize_]));
            const Vector3& v2 = *((const Vector3*)(&srcData[(index + 2) * batch.vertexSize_]));

            vertices[0] = ModelTransform(modelViewProj, v0);
            vertices[1] = ModelTransform(modelViewProj, v1);
            vertices[2] = ModelTransform(modelViewProj, v2);
            DrawTriangle(vertices, threadIndex);

            index += 3;
        }
    }
    else
    {
        const auto* srcData = (const unsigned char*)batch.vertexData_;

        // 16-bit indices
        if (batch.indexSize_ == sizeof(unsigned short))
        {
            const unsigned short* indices = ((const unsigned short*)batch.indexData_) + batch.drawStart_;
            const unsigned short* indicesEnd = indices + batch.drawCount_;

            while (indices < indicesEnd)
            {
                const Vector3& v0 = *((const Vector3*)(&srcData[indices[0] * batch.vertexSize_]));
                const Vector3& v1 = *((const Vector3*)(&srcData[indices[1] * batch.vertexSize_]));
                const Vector3& v2 = *((const Vector3*)(&srcData[indices[2] * batch.vertexSize_]));

                vertices[0] = ModelTransform(modelViewProj, v0);
                vertices[1] = ModelTransform(modelViewProj, v1);
                vertices[2] = ModelTransform(modelViewProj, v2);
                DrawTriangle(vertices, threadIndex);

                indices += 3;
            }
        }
        else
        {
            const unsigned* indices = ((const unsigned*)batch.indexData_) + batch.drawStart_;
            const unsigned* indicesEnd = indices + batch.drawCount_;

            while (indices < indicesEnd)
            {
                const Vector3& v0 = *((const Vector3*)(&srcData[indices[0] * batch.vertexSize_]));
                const Vector3& v1 = *((const Vector3*)(&srcData[indices[1] * batch.vertexSize_]));
                const Vector3& v2 = *((const Vector3*)(&srcData[indices[2] * batch.vertexSize_]));

                vertices[0] = ModelTransform(modelViewProj, v0);
                vertices[1] = ModelTransform(modelViewProj, v1);
                vertices[2] = ModelTransform(modelViewProj, v2);
                DrawTriangle(vertices, threadIndex);

                indices += 3;
            }
        }
    }
}

inline Vector4 OcclusionBuffer::ModelTransform(const Matrix4& transform, const Vector3& vertex) const
{
    return Vector4(
        transform.m00_ * vertex.x_ + transform.m01_ * vertex.y_ + transform.m02_ * vertex.z_ + transform.m03_,
        transform.m10_ * vertex.x_ + transform.m11_ * vertex.y_ + transform.m12_ * vertex.z_ + transform.m13_,
        transform.m20_ * vertex.x_ + transform.m21_ * vertex.y_ + transform.m22_ * vertex.z_ + transform.m23_,
        transform.m30_ * vertex.x_ + transform.m31_ * vertex.y_ + transform.m32_ * vertex.z_ + transform.m33_
    );
}

inline Vector3 OcclusionBuffer::ViewportTransform(const Vector4& vertex) const
{
    float invW = 1.0f / vertex.w_;
    return Vector3(
        invW * vertex.x_ * scaleX_ + offsetX_,
        invW * vertex.y_ * scaleY_ + offsetY_,
        invW * vertex.z_ * OCCLUSION_Z_SCALE
    );
}

inline Vector4 OcclusionBuffer::ClipEdge(const Vector4& v0, const Vector4& v1, float d0, float d1) const
{
    float t = d0 / (d0 - d1);
    return v0 + t * (v1 - v0);
}

inline float OcclusionBuffer::SignedArea(const Vector3& v0, const Vector3& v1, const Vector3& v2) const
{
    float aX = v0.x_ - v1.x_;
    float aY = v0.y_ - v1.y_;
    float bX = v2.x_ - v1.x_;
    float bY = v2.y_ - v1.y_;
    return aX * bY - aY * bX;
}

void OcclusionBuffer::CalculateViewport()
{
    // Add half pixel offset due to 3D frustum culling
    scaleX_ = 0.5f * width_;
    scaleY_ = -0.5f * height_;
    offsetX_ = 0.5f * width_ + 0.5f;
    offsetY_ = 0.5f * height_ + 0.5f;
    projOffsetScaleX_ = projection_.m00_ * scaleX_;
    projOffsetScaleY_ = projection_.m11_ * scaleY_;
}

void OcclusionBuffer::DrawTriangle(Vector4* vertices, i32 threadIndex)
{
    assert(threadIndex >= 0);

    ClipMaskFlags clipMask{};
    ClipMaskFlags andClipMask{};
    bool drawOk = false;
    Vector3 projected[3];

    // Build the clip plane mask for the triangle
    for (unsigned i = 0; i < 3; ++i)
    {
        ClipMaskFlags vertexClipMask{};

        if (vertices[i].x_ > vertices[i].w_)
            vertexClipMask |= CLIPMASK_X_POS;
        if (vertices[i].x_ < -vertices[i].w_)
            vertexClipMask |= CLIPMASK_X_NEG;
        if (vertices[i].y_ > vertices[i].w_)
            vertexClipMask |= CLIPMASK_Y_POS;
        if (vertices[i].y_ < -vertices[i].w_)
            vertexClipMask |= CLIPMASK_Y_NEG;
        if (vertices[i].z_ > vertices[i].w_)
            vertexClipMask |= CLIPMASK_Z_POS;
        if (vertices[i].z_ < 0.0f)
            vertexClipMask |= CLIPMASK_Z_NEG;

        clipMask |= vertexClipMask;

        if (!i)
            andClipMask = vertexClipMask;
        else
            andClipMask &= vertexClipMask;
    }

    // If triangle is fully behind any clip plane, can reject quickly
    if (andClipMask)
        return;

    // Check if triangle is fully inside
    if (!clipMask)
    {
        projected[0] = ViewportTransform(vertices[0]);
        projected[1] = ViewportTransform(vertices[1]);
        projected[2] = ViewportTransform(vertices[2]);

        bool clockwise = SignedArea(projected[0], projected[1], projected[2]) < 0.0f;
        if (cullMode_ == CULL_NONE || (cullMode_ == CULL_CCW && clockwise) || (cullMode_ == CULL_CW && !clockwise))
        {
            DrawTriangle2D(projected, clockwise, threadIndex);
            drawOk = true;
        }
    }
    else
    {
        bool triangles[64];

        // Initial triangle
        triangles[0] = true;
        unsigned numTriangles = 1;

        if (clipMask & CLIPMASK_X_POS)
            ClipVertices(Vector4(-1.0f, 0.0f, 0.0f, 1.0f), vertices, triangles, numTriangles);
        if (clipMask & CLIPMASK_X_NEG)
            ClipVertices(Vector4(1.0f, 0.0f, 0.0f, 1.0f), vertices, triangles, numTriangles);
        if (clipMask & CLIPMASK_Y_POS)
            ClipVertices(Vector4(0.0f, -1.0f, 0.0f, 1.0f), vertices, triangles, numTriangles);
        if (clipMask & CLIPMASK_Y_NEG)
            ClipVertices(Vector4(0.0f, 1.0f, 0.0f, 1.0f), vertices, triangles, numTriangles);
        if (clipMask & CLIPMASK_Z_POS)
            ClipVertices(Vector4(0.0f, 0.0f, -1.0f, 1.0f), vertices, triangles, numTriangles);
        if (clipMask & CLIPMASK_Z_NEG)
            ClipVertices(Vector4(0.0f, 0.0f, 1.0f, 0.0f), vertices, triangles, numTriangles);

        // Draw each accepted triangle
        for (unsigned i = 0; i < numTriangles; ++i)
        {
            if (triangles[i])
            {
                unsigned index = i * 3;
                projected[0] = ViewportTransform(vertices[index]);
                projected[1] = ViewportTransform(vertices[index + 1]);
                projected[2] = ViewportTransform(vertices[index + 2]);

                bool clockwise = SignedArea(projected[0], projected[1], projected[2]) < 0.0f;
                if (cullMode_ == CULL_NONE || (cullMode_ == CULL_CCW && clockwise) || (cullMode_ == CULL_CW && !clockwise))
                {
                    DrawTriangle2D(projected, clockwise, threadIndex);
                    drawOk = true;
                }
            }
        }
    }

    if (drawOk)
        ++numTriangles_;
}

void OcclusionBuffer::ClipVertices(const Vector4& plane, Vector4* vertices, bool* triangles, unsigned& numTriangles)
{
    unsigned num = numTriangles;

    for (unsigned i = 0; i < num; ++i)
    {
        if (triangles[i])
        {
            unsigned index = i * 3;
            float d0 = plane.DotProduct(vertices[index]);
            float d1 = plane.DotProduct(vertices[index + 1]);
            float d2 = plane.DotProduct(vertices[index + 2]);

            // If all vertices behind the plane, reject triangle
            if (d0 < 0.0f && d1 < 0.0f && d2 < 0.0f)
            {
                triangles[i] = false;
                continue;
            }
            // If 2 vertices behind the plane, create a new triangle in-place
            else if (d0 < 0.0f && d1 < 0.0f)
            {
                vertices[index] = ClipEdge(vertices[index], vertices[index + 2], d0, d2);
                vertices[index + 1] = ClipEdge(vertices[index + 1], vertices[index + 2], d1, d2);
            }
            else if (d0 < 0.0f && d2 < 0.0f)
            {
                vertices[index] = ClipEdge(vertices[index], vertices[index + 1], d0, d1);
                vertices[index + 2] = ClipEdge(vertices[index + 2], vertices[index + 1], d2, d1);
            }
            else if (d1 < 0.0f && d2 < 0.0f)
            {
                vertices[index + 1] = ClipEdge(vertices[index + 1], vertices[index], d1, d0);
                vertices[index + 2] = ClipEdge(vertices[index + 2], vertices[index], d2, d0);
            }
            // 1 vertex behind the plane: create one new triangle, and modify one in-place
            else if (d0 < 0.0f)
            {
                unsigned newIdx = numTriangles * 3;
                triangles[numTriangles] = true;
                ++numTriangles;

                vertices[newIdx] = ClipEdge(vertices[index], vertices[index + 2], d0, d2);
                vertices[newIdx + 1] = vertices[index] = ClipEdge(vertices[index], vertices[index + 1], d0, d1);
                vertices[newIdx + 2] = vertices[index + 2];
            }
            else if (d1 < 0.0f)
            {
                unsigned newIdx = numTriangles * 3;
                triangles[numTriangles] = true;
                ++numTriangles;

                vertices[newIdx + 1] = ClipEdge(vertices[index + 1], vertices[index], d1, d0);
                vertices[newIdx + 2] = vertices[index + 1] = ClipEdge(vertices[index + 1], vertices[index + 2], d1, d2);
                vertices[newIdx] = vertices[index];
            }
            else if (d2 < 0.0f)
            {
                unsigned newIdx = numTriangles * 3;
                triangles[numTriangles] = true;
                ++numTriangles;

                vertices[newIdx + 2] = ClipEdge(vertices[index + 2], vertices[index + 1], d2, d1);
                vertices[newIdx] = vertices[index + 2] = ClipEdge(vertices[index + 2], vertices[index], d2, d0);
                vertices[newIdx + 1] = vertices[index + 1];
            }
        }
    }
}

// Code based on Chris Hecker's Perspective Texture Mapping series in the Game Developer magazine
// Also available online at http://chrishecker.com/Miscellaneous_Technical_Articles

/// %Gradients of a software rasterized triangle.
struct Gradients
{
    /// Construct from vertices.
    explicit Gradients(const Vector3* vertices)
    {
        float invdX = 1.0f / (((vertices[1].x_ - vertices[2].x_) *
                               (vertices[0].y_ - vertices[2].y_)) -
                              ((vertices[0].x_ - vertices[2].x_) *
                               (vertices[1].y_ - vertices[2].y_)));

        float invdY = -invdX;

        dInvZdX_ = invdX * (((vertices[1].z_ - vertices[2].z_) * (vertices[0].y_ - vertices[2].y_)) -
                            ((vertices[0].z_ - vertices[2].z_) * (vertices[1].y_ - vertices[2].y_)));

        dInvZdY_ = invdY * (((vertices[1].z_ - vertices[2].z_) * (vertices[0].x_ - vertices[2].x_)) -
                            ((vertices[0].z_ - vertices[2].z_) * (vertices[1].x_ - vertices[2].x_)));

        dInvZdXInt_ = (int)dInvZdX_;
    }

    /// Integer horizontal gradient.
    int dInvZdXInt_;
    /// Horizontal gradient.
    float dInvZdX_;
    /// Vertical gradient.
    float dInvZdY_;
};

/// %Edge of a software rasterized triangle.
struct Edge
{
    /// Construct from gradients and top & bottom vertices.
    Edge(const Gradients& gradients, const Vector3& top, const Vector3& bottom, int topY)
    {
        float height = (bottom.y_ - top.y_);
        float slope = (height != 0.0f) ? (bottom.x_ - top.x_) / height : 0.0f;
        float yPreStep = (float)(topY + 1) - top.y_;
        float xPreStep = slope * yPreStep;

        x_ = RoundToInt((xPreStep + top.x_) * OCCLUSION_X_SCALE);
        xStep_ = RoundToInt(slope * OCCLUSION_X_SCALE);
        invZ_ = RoundToInt(top.z_ + xPreStep * gradients.dInvZdX_ + yPreStep * gradients.dInvZdY_);
        invZStep_ = RoundToInt(slope * gradients.dInvZdX_ + gradients.dInvZdY_);
    }

    /// X coordinate.
    int x_;
    /// X coordinate step.
    int xStep_;
    /// Inverse Z.
    int invZ_;
    /// Inverse Z step.
    int invZStep_;
};

void OcclusionBuffer::DrawTriangle2D(const Vector3* vertices, bool clockwise, i32 threadIndex)
{
    assert(threadIndex >= 0);

    int top, middle, bottom;
    bool middleIsRight;

    // Sort vertices in Y-direction
    if (vertices[0].y_ < vertices[1].y_)
    {
        if (vertices[2].y_ < vertices[0].y_)
        {
            top = 2;
            middle = 0;
            bottom = 1;
            middleIsRight = true;
        }
        else
        {
            top = 0;
            if (vertices[1].y_ < vertices[2].y_)
            {
                middle = 1;
                bottom = 2;
                middleIsRight = true;
            }
            else
            {
                middle = 2;
                bottom = 1;
                middleIsRight = false;
            }
        }
    }
    else
    {
        if (vertices[2].y_ < vertices[1].y_)
        {
            top = 2;
            middle = 1;
            bottom = 0;
            middleIsRight = false;
        }
        else
        {
            top = 1;
            if (vertices[0].y_ < vertices[2].y_)
            {
                middle = 0;
                bottom = 2;
                middleIsRight = false;
            }
            else
            {
                middle = 2;
                bottom = 0;
                middleIsRight = true;
            }
        }
    }

    auto topY = (int)vertices[top].y_;
    auto middleY = (int)vertices[middle].y_;
    auto bottomY = (int)vertices[bottom].y_;

    // Check for degenerate triangle
    if (topY == bottomY)
        return;

    // Reverse middleIsRight test if triangle is counterclockwise
    if (!clockwise)
        middleIsRight = !middleIsRight;

    const bool topDegenerate = topY == middleY;
    const bool bottomDegenerate = middleY == bottomY;

    Gradients gradients(vertices);
    Edge topToBottom(gradients, vertices[top], vertices[bottom], topY);

    int* bufferData = buffers_[threadIndex].data_;

    if (middleIsRight)
    {
        // Top half
        if (!topDegenerate)
        {
            Edge topToMiddle(gradients, vertices[top], vertices[middle], topY);
            int* row = bufferData + topY * width_;
            int* endRow = bufferData + middleY * width_;
            while (row < endRow)
            {
                int invZ = topToBottom.invZ_;
                int* dest = row + (topToBottom.x_ >> 16u);
                int* end = row + (topToMiddle.x_ >> 16u);
                while (dest < end)
                {
                    if (invZ < *dest)
                        *dest = invZ;
                    invZ += gradients.dInvZdXInt_;
                    ++dest;
                }

                topToBottom.x_ += topToBottom.xStep_;
                topToBottom.invZ_ += topToBottom.invZStep_;
                topToMiddle.x_ += topToMiddle.xStep_;
                row += width_;
            }
        }

        // Bottom half
        if (!bottomDegenerate)
        {
            Edge middleToBottom(gradients, vertices[middle], vertices[bottom], middleY);
            int* row = bufferData + middleY * width_;
            int* endRow = bufferData + bottomY * width_;
            while (row < endRow)
            {
                int invZ = topToBottom.invZ_;
                int* dest = row + (topToBottom.x_ >> 16u);
                int* end = row + (middleToBottom.x_ >> 16u);
                while (dest < end)
                {
                    if (invZ < *dest)
                        *dest = invZ;
                    invZ += gradients.dInvZdXInt_;
                    ++dest;
                }

                topToBottom.x_ += topToBottom.xStep_;
                topToBottom.invZ_ += topToBottom.invZStep_;
                middleToBottom.x_ += middleToBottom.xStep_;
                row += width_;
            }
        }
    }
    else
    {
        // Top half
        if (!topDegenerate)
        {
            Edge topToMiddle(gradients, vertices[top], vertices[middle], topY);
            int* row = bufferData + topY * width_;
            int* endRow = bufferData + middleY * width_;
            while (row < endRow)
            {
                int invZ = topToMiddle.invZ_;
                int* dest = row + (topToMiddle.x_ >> 16u);
                int* end = row + (topToBottom.x_ >> 16u);
                while (dest < end)
                {
                    if (invZ < *dest)
                        *dest = invZ;
                    invZ += gradients.dInvZdXInt_;
                    ++dest;
                }

                topToMiddle.x_ += topToMiddle.xStep_;
                topToMiddle.invZ_ += topToMiddle.invZStep_;
                topToBottom.x_ += topToBottom.xStep_;
                row += width_;
            }
        }

        // Bottom half
        if (!bottomDegenerate)
        {
            Edge middleToBottom(gradients, vertices[middle], vertices[bottom], middleY);
            int* row = bufferData + middleY * width_;
            int* endRow = bufferData + bottomY * width_;
            while (row < endRow)
            {
                int invZ = middleToBottom.invZ_;
                int* dest = row + (middleToBottom.x_ >> 16u);
                int* end = row + (topToBottom.x_ >> 16u);
                while (dest < end)
                {
                    if (invZ < *dest)
                        *dest = invZ;
                    invZ += gradients.dInvZdXInt_;
                    ++dest;
                }

                middleToBottom.x_ += middleToBottom.xStep_;
                middleToBottom.invZ_ += middleToBottom.invZStep_;
                topToBottom.x_ += topToBottom.xStep_;
                row += width_;
            }
        }
    }
}

void OcclusionBuffer::MergeBuffers()
{
    URHO3D_PROFILE(MergeBuffers);

    for (unsigned i = 1; i < buffers_.Size(); ++i)
    {
        if (!buffers_[i].used_)
            continue;

        int* src = buffers_[i].data_;
        int* dest = buffers_[0].data_;
        int count = width_ * height_;

        while (count--)
        {
            // If thread buffer's depth value is closer, overwrite the original
            if (*src < *dest)
                *dest = *src;
            ++src;
            ++dest;
        }
    }
}

void OcclusionBuffer::ClearBuffer(i32 threadIndex)
{
    assert(threadIndex >= 0);

    if (threadIndex >= buffers_.Size())
        return;

    int* dest = buffers_[threadIndex].data_;
    int count = width_ * height_;
    auto fillValue = (int)OCCLUSION_Z_SCALE;

    while (count--)
        *dest++ = fillValue;
}

}