acdream/src/AcDream.App/Rendering/Sky/SkyRenderer.Rhi.cs

using System.Collections.Immutable;
using System.Runtime.InteropServices;
using AcDream.App.Rendering.Gpu;
using AcDream.Core.Meshing;
using AcDream.Core.Terrain;
using AcDream.Content;
using DatReaderWriter;

namespace AcDream.App.Rendering.Sky;

/// <summary>
/// Campaign V slice V6k: the sky's RHI submission arm.
///
/// <para>This is V4f's content, landed as a SECOND arm rather than a replacement,
/// for the reason §5.5.6 gave: NVIDIA rendered the V4c binary 10/10 where AMD's
/// GL stack did not, so GL keeps its raw world path through to V10 as a
/// documented fork confined to the submission seam and the RHI world path ships
/// on Vulkan. Every GL statement in the sibling file is the one it always issued;
/// everything here runs only when there is no GL context.</para>
///
/// <para>Three things differ from the GL arm, each because Vulkan bakes what GL
/// sets. The per-submesh blend function is two PIPELINES — additive for
/// sun/moon/stars, straight alpha for the dome and cloud sheets — because core
/// Vulkan 1.3 does not make blend dynamic. The <c>SkyParams</c> block is a ring
/// slice taken per draw rather than one buffer rewritten per draw, because a ring
/// allocation is distinct memory that lives until the frame retires and a
/// rewritten buffer is read at EXECUTION time, not record time (plan §5.5.8's
/// second recorded gap). And the pass is BORROWED from
/// <see cref="IWorldPassScope"/>, because the frame's one backbuffer pass
/// resolves and a second pass could not load what it left.</para>
///
/// <para>The sky is the first Vulkan consumer of set 1 binding 4. Plan §5.5.8
/// recorded that <c>UniformSkyParams</c> was missing from the uniform set layout
/// and V6i-2 added it; nothing had ever bound it until now.</para>
/// </summary>
public sealed unsafe partial class SkyRenderer
{
    private readonly IGpuDevice? _device;
    private readonly ICurrentGpuFrameSource? _frames;
    private readonly IWorldPassScope? _scope;
    private IGpuPipeline? _alphaPipeline;
    private IGpuPipeline? _additivePipeline;

    /// <summary>
    /// Slot cache for the RHI arm, keyed the same way the GL arm keys its
    /// bindless handles: one entry per (surface, wrap mode) pair, because a
    /// Vulkan table entry is a combined image sampler and the dome sampled
    /// CLAMP_TO_EDGE is a different entry from a cloud sheet sampled REPEAT.
    /// </summary>
    private readonly Dictionary<(uint SurfaceId, bool Repeat), GpuTextureSlot>
        _slotBySurfaceAndWrap = new();

    /// <summary>
    /// The sky vertex layout, taken from <see cref="Vertex"/> itself.
    ///
    /// <para>The stride is <c>sizeof(Vertex)</c> — 36 bytes, not the 32 the world
    /// mesh uses. <see cref="Vertex"/> carries a fourth member,
    /// <c>TerrainLayer</c>, which <c>sky.vert</c> does not declare and the GL arm
    /// never described to a <c>glVertexAttribPointer</c>; it is still part of the
    /// record's footprint, and getting the stride wrong scatters the dome's
    /// vertices into noise while leaving the frame otherwise plausible. The GL arm
    /// says <c>sizeof(Vertex)</c> and so does this.</para>
    /// </summary>
    internal static readonly GpuVertexLayout SkyVertexLayout = new(
        StrideBytes: (uint)sizeof(Vertex),
        ImmutableArray.Create(
            new GpuVertexAttribute(0, GpuVertexFormat.Float3, 0),
            new GpuVertexAttribute(1, GpuVertexFormat.Float3, 12),
            new GpuVertexAttribute(2, GpuVertexFormat.Float2, 24)));

    /// <summary>
    /// The RHI arm's constructor. No GL context, no <c>Shader</c>, no
    /// <c>SamplerCache</c> and no <c>BindlessSupport</c>: the two pipelines
    /// compile <c>sky</c> from the committed SPIR-V, and the wrap mode is which
    /// sampler the slot was registered with.
    /// </summary>
    internal SkyRenderer(
        IGpuDevice device,
        ICurrentGpuFrameSource frames,
        IWorldPassScope scope,
        IDatReaderWriter dats,
        TextureCache textures)
    {
        _device = device ?? throw new ArgumentNullException(nameof(device));
        _frames = frames ?? throw new ArgumentNullException(nameof(frames));
        _scope = scope ?? throw new ArgumentNullException(nameof(scope));
        _dats = dats ?? throw new ArgumentNullException(nameof(dats));
        _textures = textures ?? throw new ArgumentNullException(nameof(textures));

        _alphaPipeline = CreateSkyPipeline("sky-alpha", GpuBlendMode.StraightAlpha);
        try
        {
            _additivePipeline = CreateSkyPipeline("sky-additive", GpuBlendMode.Additive);
        }
        catch
        {
            _alphaPipeline.Dispose();
            _alphaPipeline = null;
            throw;
        }
    }

    /// <summary>
    /// One sky pipeline. Depth is off in both directions and culling is off, which
    /// is exactly the GL arm's state bracket — the sky must never occlude scene
    /// geometry and its meshes are drawn from both sides.
    /// </summary>
    private IGpuPipeline CreateSkyPipeline(string name, GpuBlendMode blend) =>
        _device!.CreatePipeline(new GpuPipelineDescription
        {
            Name = name,
            Shaders = new GpuShaderSet("sky"),
            VertexLayout = SkyVertexLayout,
            Topology = GpuPrimitiveTopology.TriangleList,
            Blend = blend,
            Depth = GpuDepthState.Disabled,
            Cull = GpuCullMode.None,
            FrontFace = GpuFrontFace.CounterClockwise,
            AlphaToCoverage = false,
            ColorWrite = true,
            SampleCount = _scope!.SampleCount,
        });

    /// <summary>
    /// Uploads one submesh into its own device-local vertex/index buffer pair.
    /// Sky meshes are built once per GfxObj and never move, so this is the same
    /// "upload once, draw many frames" shape the GL arm's static-draw VBOs have.
    /// </summary>
    private SubMeshGpu UploadSubMeshRhi(GfxObjSubMesh sm)
    {
        IGpuDevice device = _device!;
        ReadOnlySpan<byte> vertexBytes = MemoryMarshal.AsBytes<Vertex>(sm.Vertices);
        ReadOnlySpan<byte> indexBytes = MemoryMarshal.AsBytes<uint>(sm.Indices);

        IGpuBuffer vertices = device.CreateBuffer(new GpuBufferDescription(
            $"sky-vertices-0x{sm.SurfaceId:X8}",
            Math.Max(vertexBytes.Length, 32),
            GpuBufferUsage.Vertex | GpuBufferUsage.TransferDestination,
            GpuMemoryResidency.DeviceLocal));
        IGpuBuffer indices;
        try
        {
            indices = device.CreateBuffer(new GpuBufferDescription(
                $"sky-indices-0x{sm.SurfaceId:X8}",
                Math.Max(indexBytes.Length, 4),
                GpuBufferUsage.Index | GpuBufferUsage.TransferDestination,
                GpuMemoryResidency.DeviceLocal));
        }
        catch
        {
            vertices.Dispose();
            throw;
        }

        try
        {
            if (!vertexBytes.IsEmpty)
                vertices.Upload(0, vertexBytes);
            if (!indexBytes.IsEmpty)
                indices.Upload(0, indexBytes);
        }
        catch
        {
            indices.Dispose();
            vertices.Dispose();
            throw;
        }

        return new SubMeshGpu
        {
            VertexBuffer = vertices,
            IndexBuffer = indices,
            IndexCount = sm.Indices.Length,
            SurfaceId = sm.SurfaceId,
            IsAdditive = sm.Translucency == TranslucencyKind.Additive,
            SurfLuminosity = sm.Luminosity,
            SurfDiffuse = sm.Diffuse,
            NeedsUvRepeat = sm.NeedsUvRepeat,
            SurfOpacity = sm.SurfOpacity,
            DisableFog = sm.DisableFog,
        };
    }

    /// <summary>
    /// The (surface, wrap) table slot on a backend with no GL texture name. The
    /// decode is <c>TextureCache</c>'s, so the pixels are the same ones the GL arm
    /// samples; what differs is that the texture is created through
    /// <see cref="IGpuDevice.CreateTexture"/> and paired with a real sampler
    /// object rather than a bindless handle.
    /// </summary>
    private uint RhiTextureTableSlot(uint surfaceId, bool repeat)
    {
        var key = (surfaceId, repeat);
        if (!_slotBySurfaceAndWrap.TryGetValue(key, out GpuTextureSlot slot))
        {
            slot = _textures.RegisterWorldSurface(surfaceId, repeat);
            _slotBySurfaceAndWrap.Add(key, slot);
        }

        return slot.Index;
    }

    /// <summary>
    /// Records one submesh into the borrowed world pass.
    ///
    /// <para>Order matters for the same reason it does in terrain's arm:
    /// <c>BindPipeline</c> re-issues the pipeline's own fixed state, and the
    /// frame-global sections are bound AFTER this renderer's own binds because
    /// those binds are what select the descriptor scope the sections must land in
    /// (plan §5.5.14 item 2).</para>
    /// </summary>
    private void DrawSubMeshRhi(SubMeshGpu sub, uint textureSlot)
    {
        if (sub.IndexCount == 0 || sub.VertexBuffer is null || sub.IndexBuffer is null)
            return;

        IWorldPassScope scope = _scope!;
        IGpuPassEncoder encoder = scope.RequireEncoder();
        IGpuFrame frame = _frames!.CurrentFrame
            ?? throw new InvalidOperationException(
                "SkyRenderer requires an open IGpuFrame (see GpuDeviceFrameLifetime).");

        encoder.BindPipeline(sub.IsAdditive ? _additivePipeline! : _alphaPipeline!);
        var pushConstants = new GpuPushConstants
        {
            // uViewProjection is unread by sky.vert — the sky carries its own
            // camera-anchored view and dome projection in SkyParams — so the
            // block's only live member here is the texture slot.
            ViewProjection = System.Numerics.Matrix4x4.Identity,
            DrawIdOffset = 0,
            LightingMode = 0,
            RenderPass = 0,
            LightDebug = 0,
            TextureIndexA = textureSlot,
            TextureIndexB = 0,
            ParamA = 0f,
            ParamB = 0f,
        };
        encoder.SetPushConstants(in pushConstants);
        encoder.BindVertexBuffer(sub.VertexBuffer, 0);
        encoder.BindIndexBuffer(sub.IndexBuffer, 0, GpuIndexType.UInt32);

        GpuRingAllocation parameters = frame.AllocateRing(
            SkyParams.SizeInBytes,
            GpuRingUsage.Uniform);
        MemoryMarshal.Write(parameters.Data, in _params);
        encoder.BindUniformBuffer(
            GpuBindingModel.UniformSkyParams,
            parameters.Buffer,
            parameters.OffsetBytes,
            SkyParams.SizeInBytes);

        WorldFrameSectionBinding.BindSceneLighting(encoder, scope.Sections, frame);
        WorldFrameSectionBinding.BindTerrainClip(encoder, scope.Sections, frame);

        encoder.DrawIndexed((uint)sub.IndexCount, 1, 0, 0, 0);
    }

    private void DisposeRhi()
    {
        List<Exception>? failures = null;
        void Attempt(Action action)
        {
            try { action(); }
            catch (Exception error) { (failures ??= []).Add(error); }
        }

        foreach (List<SubMeshGpu> subs in _gpuByGfxObj.Values)
        {
            foreach (SubMeshGpu sub in subs)
            {
                Attempt(() => sub.VertexBuffer?.Dispose());
                Attempt(() => sub.IndexBuffer?.Dispose());
            }
        }

        _gpuByGfxObj.Clear();
        // The textures themselves belong to TextureCache, which outlives this
        // renderer and releases both the image and its table slot itself; the
        // slot cache here is a lookup, not an ownership record.
        _slotBySurfaceAndWrap.Clear();

        Attempt(() => _alphaPipeline?.Dispose());
        _alphaPipeline = null;
        Attempt(() => _additivePipeline?.Dispose());
        _additivePipeline = null;

        if (failures is not null)
            throw new AggregateException("The sky renderer's RHI resources did not fully release.", failures);
    }
}