Accelerated Rendering — Minecraft 26.1 Port

Alpha · Status and Coverage

---

English

1. About

Accelerated Rendering is a client-side rendering modification that moves vertex transformation from the CPU to the GPU using compute shaders.

Instead of multiplying every vertex by a model matrix on the CPU on every frame, the mod uploads model-space vertices once, caches the resulting mesh, and performs the transformation on the GPU. Repeated draws of the same geometry (for example, many copies of the same entity or the same dropped item) then share a single cached mesh and only differ by per-instance data such as transform, colour, light and overlay.

This repository is a port of the original Minecraft 1.21.4 mod to Minecraft 26.1.

Mod ID	acceleratedrendering
Version	1.0.15-alpha
Environment	Client only
License	MIT
Original author	Argon4W
Port author	namelessgod2008

2. Requirements

Component	Version
Minecraft	26.1
Fabric Loader	0.19.5 or newer
Java	25 or newer
Fabric API	required
Forge Config API Port	required (configuration screen)

3. Why the approach still applies in 26.1

Minecraft 26.1 rewrote its rendering architecture, so it is worth stating explicitly that vanilla 26.1 still performs vertex transformation on the CPU:

• Entities and block entities — ModelPart.Cube.compile calls matrix.transformPosition(...) for every vertex of every cube, every frame.
• Items — VertexConsumer.putBakedQuad calls matrix.transformPosition(...) for every vertex of every quad.

The vanilla vertex shaders only apply the camera rotation (gl_Position = ProjMat * ModelViewMat * vec4(Position, 1.0), where ModelViewMat contains rotation but no translation). The model transform is therefore already baked into the vertex data on the CPU.

26.1 contains no compute shaders, no instanced vertex attributes, and no indirect or multi-draw in vanilla. This mod's GPU-side transform has no vanilla equivalent, so the approach remains valid.

4. Implemented features

4.1 Entity and block-entity rendering acceleration

Model geometry produced through ModelPart is intercepted, cached as GPU meshes and transformed on the GPU. This covers living entities, other entities, and block entities whose models are rendered through the same model pipeline (for example chests).

• Mesh cache is keyed so that it survives across frames.
• Draw batching layers reproduce the vanilla order(int) grouping, so armour, trims and render layers keep their correct relative draw order.

4.2 Item rendering acceleration

Item model geometry (BakedQuad) is cached and transformed on the GPU.

• Quads belonging to one item submission are merged into a single mesh per (tintIndex, lightEmission) group, rather than one mesh per quad.
• Per-vertex normals are baked in model space; the GPU applies the pose normal matrix, which is mathematically equivalent to vanilla's pose.transformNormal(quad.direction().getUnitVec3f()).
• Per-instance colour, light and overlay are supplied at draw time, so a single cached mesh serves every instance.

4.3 Entity shadow acceleration

Entity shadows are routed through the accelerated pipeline instead of the vanilla writer. Shadow geometry, alpha and shape data are taken from the values vanilla already computed during its extraction phase.

4.4 Vanilla draw-order preservation

26.1 renders entities in two stages: a submit stage that only records data, and a render stage that writes vertices. Because the accelerated path groups geometry by batch layer, the vanilla order(int) buckets are mapped onto accelerated layers so that the original relative draw order is restored.

4.5 Shader pack compatibility (Iris)

An Iris compatibility module is included and active. It handles:

• Unwrapping of Iris-wrapped RenderType instances at the buffer entry point.
• Re-routing accelerated output into the Iris G-buffer.
• Shadow-pass awareness so that shadow rendering uses its own target.
• Adaptation to the Iris 1.11.x API surface.

4.6 Orientation culling

Per-quad orientation culling is available as a program dispatcher integrated into the core draw path, with its own configuration section.

4.7 Draw methods

Two draw methods are supported and selectable:

• BASEVERTEX (default) — uses a shared index buffer with a base vertex offset.
• INDIRECT — executes indirect draws through a mixin bridge, since the 26.1 RenderPass API has no indirect entry point.

4.8 Configuration

Configuration is provided through Forge Config API Port, with an in-game configuration screen supplied by that library and a Mod Menu entry point.

Sections include: core settings, accelerated entity rendering, accelerated item rendering, filters, Iris compatibility, culling, and mod compatibility toggles.

5. Not implemented in this build

The following parts of the upstream mod are excluded from compilation in this alpha. Source files are retained in the repository but are not compiled or registered.

5.1 Item rendering

• GUI item rendering — features/items/gui/** and features/items/mixins/gui/** (GUI batching, accelerated blit / fill / gradient, string rendering).
• Whole-model baking path — ItemRendererMixin, ModelBlockRendererMixin, features/items/mixins/models/**, features/items/mixins/accessors/**. These depend on classes removed in 26.1 (ItemRenderer, BakedModel, SimpleBakedModel, MultiPartBakedModel, WeightedBakedModel) and have no direct equivalent in the 26.1 architecture.
• Block/item colour helpers — BlockLayerColors, ColorHelper, EmptyBlockColor, ItemLayerColors, IAcceleratedBakedModel. 26.1 replaced the block colour system with block tint sources; this acceleration path uses the tint layers carried by item submissions instead.

5.2 Dedicated feature modules

Excluded entirely, pending porting:

• GeckoLib (features/geckolib)
• FTB (features/ftb)
• Modern UI (features/modernui)
• Text rendering (features/text)
• Touhou Little Maid (features/tlm)
• Create (features/create)
• Entity Model Features (features/emf)
• Simple Bedrock Model (features/simplebedrockmodel)

The matching configuration entries exist but have no effect until these modules are ported.

5.3 Filters

features/filter/** is excluded except for the FilterType enum (which the configuration model depends on). Filter configuration entries are present but the filtering logic is not compiled into this build.

5.4 Mod compatibility modules

Excluded, pending porting:

• ImmediatelyFast
• TweakerMore
• Tweakeroo
• Xaero's Minimap / World Map
• Roughly Enough Items
• Trinkets
• Sophisticated Backpacks
• Forge Config API Port (the compatibility module; the library itself is still required)

5.5 Miscellaneous

• InventoryScreenMixin
• ParticleEngineMixin
• NeoForge helper classes (ConfigScreen, IQuadTransformer, UnitTextureAtlasSprite)

5.6 Texture pixel download

TextureUtils.downloadTexture returns null in this build, because reading texture pixels in 26.1 requires a GpuDevice command encoder. The culled mesh collector degrades safely when no texture is available.

6. Known limitations

• Alpha quality. This is a work in progress port; feature coverage is incomplete and behaviour may change between builds.
• GUI acceleration is absent. Item and text rendering inside containers and menus is not accelerated.
• Indirect draw requires the mixin bridge. Because 26.1's RenderPass has no indirect entry point, the INDIRECT draw method relies on a mixin into the command encoder. The BASEVERTEX method is the default.
• Diagnostics are currently enabled. The build prints per-second diagnostic counters prefixed with [AR-FRAME] and occasional [AR-SLOW] lines to the game log. These are temporary and will be removed.
• Performance characteristics are not documented here. This alpha makes no performance claims.

7. Compatibility notes

• Sodium and Iris are detected at runtime; compatibility mixin configurations use a plugin with a presence check so that they are only applied when the target mod is installed.
• The shadow acceleration mixin uses an elevated mixin priority so that it runs ahead of Sodium's own shadow handling, allowing the accelerated pipeline to take over when its guards are satisfied.
• Mixin configurations that target absent mods are not registered, so a missing optional dependency does not cause a crash.

8. Summary table

Area	Status
Entity / block-entity model rendering	Implemented
Item model rendering	Implemented
Entity shadows	Implemented
Vanilla draw-order preservation	Implemented
Iris / shader pack support	Implemented
Orientation culling	Implemented
Indirect draw	Implemented (mixin bridge)
Configuration screen	Implemented
Item GUI rendering	Not implemented
Whole-model baking path	Not implemented
Text rendering	Not implemented
Filters	Not implemented
GeckoLib / FTB / ModernUI / TLM / Create / EMF / SBM	Not implemented
Mod compatibility modules (see 5.4)	Not implemented
Texture pixel download	Not implemented

---

中文

1. 简介

Accelerated Rendering 是一个客户端渲染优化模组，它使用计算着色器把顶点变换从 CPU 搬到 GPU。

原版每一帧都要在 CPU 上把每个顶点乘以模型矩阵；本模组改为：把模型空间顶点上传一次、 缓存成网格，变换交给 GPU 完成。同一几何被重复绘制时（例如大量同种实体、或大量同一种 掉落物），它们共享同一份缓存网格，彼此之间只差「逐实例数据」——变换矩阵、颜色、 光照、overlay。

本仓库是原 Minecraft 1.21.4 版本到 Minecraft 26.1 的移植。

项	值
模组 ID	acceleratedrendering
版本	1.0.15-alpha
运行环境	仅客户端
许可证	MIT
原作者	Argon4W
移植	namelessgod2008

2. 环境要求

组件	版本
Minecraft	26.1
Fabric Loader	0.19.5 或更高
Java	25 或更高
Fabric API	必需
Forge Config API Port	必需（提供配置界面）

3. 为什么这个思路在 26.1 依然成立

Minecraft 26.1 重写了整个渲染架构，因此有必要明确说明：26.1 原版仍然在 CPU 侧做 顶点变换。

• 实体与方块实体 —— ModelPart.Cube.compile 每帧对每个方块的每个顶点调用 matrix.transformPosition(...)。
• 物品 —— VertexConsumer.putBakedQuad 对每个 quad 的每个顶点调用 matrix.transformPosition(...)。

原版顶点着色器只负责相机旋转（gl_Position = ProjMat * ModelViewMat * vec4(Position, 1.0)， 其中 ModelViewMat 只含旋转、不含平移），模型变换早已在 CPU 侧烘进顶点数据。

26.1 原版既没有计算着色器，也没有实例化顶点属性，也没有 indirect / multi-draw。 本模组的 GPU 侧变换在 26.1 原版中没有对应实现，因此该思路依然有效。

4. 已实现的功能

4.1 实体与方块实体渲染加速

经由 ModelPart 产出的模型几何会被拦截、缓存为 GPU 网格并在 GPU 上变换。 覆盖生物实体、其它实体，以及通过同一套模型管线渲染的方块实体（例如箱子）。

• 网格缓存按可跨帧复用的方式建立索引。
• 绘制批次层（layer）还原了原版 order(int) 的分组，因此盔甲、纹饰、各渲染层的 相对绘制顺序保持正确。

4.2 物品渲染加速

物品模型几何（BakedQuad）会被缓存并在 GPU 上变换。

• 同一次物品提交内的 quad 会按 (tintIndex, lightEmission) 分组，合并成每个分组 一个网格，而不是每个 quad 一个网格。
• 逐顶点法线以模型空间烘焙；GPU 侧再应用姿态法线矩阵，与原本的 pose.transformNormal(quad.direction().getUnitVec3f()) 数学等价。
• 逐实例的颜色、光照、overlay 在绘制时提供，因此一份缓存网格可服务所有实例。

4.3 实体阴影加速

实体阴影改由加速管线写入，不再走原版写入器。阴影的几何、透明度与形状数据直接取自 原版在提取阶段已经算好的结果。

4.4 原版绘制顺序还原

26.1 的实体渲染是两段式：提交段只记录数据，渲染段才写顶点。由于加速路径按批次层分组 几何，本模组把原版的 order(int) 桶映射为加速层，从而还原原本的相对绘制顺序。

4.5 光影包兼容（Iris）

内置并启用 Iris 兼容模块，负责：

• 在缓冲入口处解包 Iris 包装过的 RenderType 实例。
• 把加速输出重定向进 Iris 的 G-buffer。
• 感知阴影阶段，使阴影渲染使用自己的渲染目标。
• 适配 Iris 1.11.x 的 API 形态。

4.6 朝向剔除

逐 quad 的朝向剔除以「程序调度器」的形式内建于核心绘制路径，并有独立的配置分区。

4.7 绘制方式

支持两种绘制方式，可配置：

• BASEVERTEX（默认）—— 使用共享索引缓冲 + 基准顶点偏移。
• INDIRECT —— 由于 26.1 的 RenderPass API 没有 indirect 入口，该方式通过 mixin 桥接执行间接绘制。

4.8 配置

配置由 Forge Config API Port 提供，配置界面由该库自带，并通过 Mod Menu 提供入口。

配置分区包括：核心设置、实体加速、物品加速、过滤器、Iris 兼容、剔除、模组兼容开关。

5. 本版本尚未实现的部分

以下上游功能在本 alpha 版本中被排除出编译。源码文件仍保留在仓库中，但不参与编译、 也未注册。

5.1 物品渲染相关

• GUI 物品渲染 —— features/items/gui/** 与 features/items/mixins/gui/** （GUI 批处理、加速 blit / fill / gradient、字符串渲染）。
• 整模型烘焙路径 —— ItemRendererMixin、ModelBlockRendererMixin、 features/items/mixins/models/**、features/items/mixins/accessors/**。 这些依赖 26.1 已移除的类（ItemRenderer、BakedModel、SimpleBakedModel、 MultiPartBakedModel、WeightedBakedModel），在 26.1 架构下没有直接对应物。
• 方块/物品着色辅助类 —— BlockLayerColors、ColorHelper、EmptyBlockColor、 ItemLayerColors、IAcceleratedBakedModel。26.1 已用 block tint source 取代旧的 方块着色体系；本加速路径改用物品提交自带的 tint layer。

5.2 独立功能模块

以下模块整体排除，待移植：

• GeckoLib（features/geckolib）
• FTB（features/ftb）
• Modern UI（features/modernui）
• 文本渲染（features/text）
• 车万女仆 TLM（features/tlm）
• Create（features/create）
• Entity Model Features（features/emf）
• Simple Bedrock Model（features/simplebedrockmodel）

对应的配置项已经存在，但在这些模块移植完成前不会生效。

5.3 过滤器

features/filter/** 除 FilterType 枚举（配置模型依赖它）外整体排除。 过滤器配置项存在，但过滤逻辑未编入本版本。

5.4 模组兼容模块

以下兼容模块排除，待移植：

• ImmediatelyFast
• TweakerMore
• Tweakeroo
• Xaero 小地图 / 世界地图
• REI（物品管理器）
• Trinkets
• Sophisticated Backpacks
• Forge Config API Port（指其兼容模块；库本身仍然是必需的）

5.5 其它

• InventoryScreenMixin
• ParticleEngineMixin
• NeoForge 辅助类（ConfigScreen、IQuadTransformer、UnitTextureAtlasSprite）

5.6 纹理像素下载

本版本中 TextureUtils.downloadTexture 返回 null —— 26.1 读取纹理像素需要 GpuDevice 的 command encoder。在拿不到纹理时，剔除网格收集器会安全降级。

6. 已知限制

• Alpha 质量。 这是进行中的移植版本；功能覆盖不完整，行为可能随版本变化。
• 没有 GUI 加速。 容器与菜单内的物品、文本渲染未加速。
• INDIRECT 绘制依赖 mixin 桥接。 因为 26.1 的 RenderPass 没有 indirect 入口， INDIRECT 方式依赖对命令编码器的 mixin。默认方式是 BASEVERTEX。
• 诊断代码当前处于开启状态。 本版本会向游戏日志每秒打印以 [AR-FRAME] 为前缀的 诊断计数，并偶尔打印 [AR-SLOW]。这些是临时代码，后续会移除。
• 本文档不涉及性能。 本 alpha 版本不对性能作任何声明。

7. 兼容性说明

• 运行时检测 Sodium 与 Iris；兼容用的 mixin 配置带有存在性判定插件，只有目标模组 已安装时才会应用。
• 阴影加速 mixin 使用了更高的 mixin 优先级，以便先于 Sodium 自身的阴影处理执行， 在守卫条件满足时由加速管线接管。
• 指向不存在模组的 mixin 配置不会注册，因此缺少可选依赖不会导致崩溃。

8. 状态总览

领域	状态
实体 / 方块实体模型渲染	已实现
物品模型渲染	已实现
实体阴影	已实现
原版绘制顺序还原	已实现
Iris / 光影包支持	已实现
朝向剔除	已实现
Indirect 绘制	已实现（mixin 桥接）
配置界面	已实现
物品 GUI 渲染	未实现
整模型烘焙路径	未实现
文本渲染	未实现
过滤器	未实现
GeckoLib / FTB / ModernUI / TLM / Create / EMF / SBM	未实现
模组兼容模块（见 5.4）	未实现
纹理像素下载	未实现