feat(rendering): add occlusion culling backends and bindless texturing - #95
Open
ZhuRuoLing wants to merge 30 commits into
Open
feat(rendering): add occlusion culling backends and bindless texturing#95ZhuRuoLing wants to merge 30 commits into
ZhuRuoLing wants to merge 30 commits into
Conversation
ZhuRuoLing
marked this pull request as draft
August 7, 2026 21:58
…rnal to some internal api
ZhuRuoLing
force-pushed
the
feat/occusion_culling
branch
from
August 13, 2026 22:34
3c58b50 to
4cdf68c
Compare
ZhuRuoLing
changed the title
ZhuRuoLing
marked this pull request as ready for review
September 8, 2026 11:22
Reason: script execution via -e/-c flag Reply |
|
All verification complete. This is round 18 (十八复) — delta is a single commit 「SO AMD FUCK U」fixing an AMD driver issue, and the fix is incomplete (regular shader only). Writing the review: 代码审查摘要 — PR #95（十八复 / synchronize）操作: synchronize · HEAD: 🔴 关键
|
| 问题 | 行号 | 状态 |
|---|---|---|
simpleCodec(TestBloomBlock::new) codec 引用错类 |
9402 | ❌ 未修 |
getBuffer(6 * 6) + drawIndexed(0, 0, 6 * 6, 1)（QUADS 24 顶点 vs 36 索引） |
6440 / 6497 | ❌ 未修（GPU_QUERY 接线后 active） |
new AABB(worldPosition) 零尺寸盒 + inflate(0.1) |
9441 / 6805 | ❌ 未修 |
| IntSizeUbo 宽高互换（getHeight 先行） | 3324 | ❌ 未修 |
midTex = textures.get(6) SPD mid-mip 语义 |
6214 / 6250 | ❌ 未修 |
| mipmap 节流（十六复 §37 引入项） | 5200-5307 | 原样（interval 默认 0，inert） |
| GL46.* 入口点 / isSupported 恒 true / 阻塞读 / alrSetWasUsed / 提交门控死锁等 | 各处 | ❌ 未修 |
十七复 GL_KHR_shader_subgroup 探测 vs shader _quad : require、define=0 变体无条件注册、HZOC 死 import） |
— | ❌ 本轮未触碰 |
结论
REQUEST_CHANGES — 本轮 commit 的 AMD 修复方向正确（常量索引 switch + TEXTURE_FETCH_BARRIER），但只修了双变体中的一半：bindless 派发路径上动态 image 数组索引原样保留，AMD 故障在 bindless 配置下必然复现；请同步修复 bindless 变体并清理 setupTexture 死代码与启动 INFO 日志。旧 blocker（codec 错类 / 6*6 索引越界 / AABB 零尺寸 / IntSizeUbo）已跨 10+ 轮未修，建议优先处理。
gh 未认证（token invalid，401），标题未修改（现标题准确无需改）。
🌿 Roseau API Breaking Change Report
🔴 rendering — 44 breaking change(s)
|
|
💾 Self-improvement review: Patched SKILL.md in skill 'rendering-gpu-pr-review' (1 replacement). · Patched references/pr95-occlusion-culling-review-matrix.md in skill 'rendering-gpu-pr-review' (1 replacement). |
代码审查摘要 — PR #95（synchronize 复审查 · 十九复）操作: synchronize 🔴 关键（本轮新增）
|
|
💾 Self-improvement review: Patched SKILL.md in skill 'rendering-gpu-pr-review' (1 replacement). · Patched references/pr95-occlusion-culling-review-matrix.md in skill 'rendering-gpu-pr-review' (1 replacement). |
🌿 Roseau API Breaking Change Report
🔴 rendering — 44 breaking change(s)
|
Reason: shell execution via heredoc Reply |
🌿 Roseau API Breaking Change Report
🔴 rendering — 44 breaking change(s)
|
|
审查完成。以下是审查意见（将自动发布到 PR #95 评论区）。 代码审查摘要 — PR #95操作: synchronize（分支 ✅ 绑定/着色器交叉验证（本 PR 核心，逐条核对通过）我把 Java 端
顶点/矩阵约定也正确： 🔴 关键
|
| 声称 | 状态 | 对应文件 |
|---|---|---|
OcclusionCuller/OcclusionKey/OcclusionMethod/CullingStatistics/包装提交节点存储 |
✅ | optimization/occlusion/*、query/*、hiz/* |
| 自动后端选择（Hi-Z / GPU Query / NoOp） | ✅ | OcclusionMethod、ALROptimizations、OcclusionCuller.createInstance |
| 池化 GL samples query、AABB query 渲染、frustum 预筛、一帧延迟读取 | ✅（预筛默认 false） | GpuSampleQueryPool、QueryInstance、FrameState、GlSamplesQuery |
| vanilla + NeoForge 各 feature renderer 可见性过滤 | ✅ 12 个 mixin | mixins/features/*（已全部登记 mixins.json） |
| padded R32F 深度转换 + FidelityFX SPD mip 链 | ✅ | DepthTexConverter、SinglePassDownsampler、depth_convert.csh、ffx_spd_*.csh |
| compute pipeline 命名布局 + 有序绑定分配 | ✅ | ALRComputePipeline、bindings/*、ALRComputePipelines |
| texture/image/buffer/atomic/bindless image array 显式绑定 | ✅ | ALRComputePass、GlComputePassBackend |
| GPU 能力上报、扩展纹理格式、GL 纹理分配、ARB bindless handle/residency | ✅ | ALRHICapabilities、ExtendedTextureFormat、GlExtendedTexture、GlBindlessTexturingSupport、GlTextureResidentScope |
| 复用 GPU 资源池、transform/size UBO、渲染调试选项、开发 JVM 参数 | ✅（JVM 参数见警告 4） | foundation/*、FullTransformsUbo、ALROptions、module.gradle |
ApiStatus.Internal 位置调整 + sync 模块 import 同步 |
✅（纯 churn） | module.sync/**（仅 +import 行） |
test_occlusion 方块/BER/生成资源 + compute 测试绑定更新 |
ComputeSupport.computeBlur 绑定写错（关键 1），其余 ✅ |
|
| TODO：完成 Hi-Z SPD dispatch 与可见性决策 | ✅ | isSupported 门槛已按 capability 设置 |
| TODO：注册并选择 bindless SPD pipeline | ✅ | FFX_SPD_..._BINDLESS，按 capabilities.bindlessTexturing() 运行时选择 |
| TODO：Minecraft 运行时与 GPU 验证 | 建议把验证结论（GPU/驱动、帧率、各后端实测）写进 PR 描述 | |
| Impact 段「visibility decisions / SPD dispatch 仍需后续工作」「bindless pipeline 运行时注册与选择尚未完成」 | ❌ 与代码和 TODO 段互相矛盾 | HierarchicalZOcclusionCuller.processFeatures/dispatch/shouldDraw 与 bindless 选择均已实现，请更新描述 |
结论: REQUEST_CHANGES — 架构与绑定正确性整体扎实（生产路径的 binding point 与着色器 layout(binding=N) 逐条核对通过），但必须先修 ComputeSupport.computeBlur() 的绑定错误，并处理 module.gradle 的重复行/debug 参数入库、GlComputePassBackend 抛异常策略、MemoryAccess 的 sun.misc.Unsafe 依赖这几项；同时更新 PR 描述中与 TODO 自相矛盾的部分。
标题
feat(rendering): add occlusion culling backends and bindless texturing与内容相符且清晰，未修改（无需gh pr edit）。
🧪 测试建议
| 被测目标 | 推荐场景 | 优先级 |
|---|---|---|
ComputeSupport.computeBlur() |
修复绑定后，比对 GPU 输出与 CPU 参考实现的模糊结果；仿 add() 加 counter 断言，防止再次绑错 |
🔴 |
OcclusionFeatureRendererHelper.filterVisibleFeatures |
空 culler → 必须返回原列表实例；culler 判定部分不可见 → 只剔除命中项；任何情况下不得把整列表清空 | 🔴 |
FrameState.shouldDraw / fetchResults |
新 key 默认可见；cameraInside 优先于 query 结果；query 值 0/1 的两条路径；results.length > size 时的统计口径 |
🟡 |
OcclusionTestSSBO.actualSize(n) vs getDefinition().size(STD430) |
断言二者一致（13 个 ivec2 + n 个 AABB 的 std430 偏移），锁住缓冲尺寸不被重排破坏 | 🟡 |
Std140LayoutRulesTest |
扩展覆盖 FullTransformsUbo、OcclusionTestCB、OcclusionTestSSBO——着色器依赖的手写 std140/std430 偏移正是最难察觉的破坏点（该测试文件本 PR 已更新签名，是好的先例） |
🟡 |
OcclusionCuller.createInstance / OcclusionMethod |
capability 矩阵：强制 impl 名称合法/非法、无 compute、无 bindless、maxImageUnit 8/16/17 的组合 |
🟢 |
GlSamplesQuery |
release() 后 result 必须复位为 -1（可用假 GL 层/mock 验证），防止跨帧读旧值 |
🟢 |
由 Hermes Agent 审查
|
💾 Self-improvement review: Patched SKILL.md in skill 'rendering-gpu-pr-review' (1 replacement). · Patched references/pr95-occlusion-culling-review-matrix.md in skill 'rendering-gpu-pr-review' (1 replacement). |
代码审查摘要 — PR #95操作: edited 审查方法：curl 下载 patch-diff + 本地 📋 声称验证表
绑定编号对账结论（逐 shader 核对，✅）： 🔴 关键
|
| 被测目标 | 推荐测试场景 | 优先级 |
|---|---|---|
GpuQueryOcclusionCuller（一帧延迟） |
首帧无 previousFrameState 时必返回 visible；连续两帧同一 OcclusionKey 的结果读回；相机进入 AABB 时强制可见；OCclusionKey 每帧复用时 bb 更新生效 |
🔴 |
HierarchicalZOcclusionCuller.shouldDraw |
results 长度边界（orDefault == results.length）、results == null、key 不在上一帧 map 中；非 64 对齐窗口尺寸（padded ≠ framebuffer）下的坐标换算 |
🔴 |
StagingSupport.GpuStagingSupport |
连续两帧写入后回读一致性；buffer 扩容后旧映射关闭；isGlOnDx12() 走 CPU 路径 |
🔴 |
OcclusionSubmitNodeCollection |
beginOcclusionRecord 未配对结束 → 抛 IllegalStateException；重复 begin；空记录（wasUsed == false）不提交；16 个 submit 列表搬迁的完整性（搬移后原集合内容一致） |
🟡 |
FrameState.addKey/fetchResults |
同一 key 多个 feature 只分配一个 query；culled/features/total 统计与输入一致；glGetQueryObjectuiv 结果 0/1 映射 |
🟡 |
ALRComputePipeline.getBinding(name/type) |
名称不存在返回 null；getBinding(int,type) 越界行为；bindingsByType 中 texture 与 image 混排的顺序 |
🟡 |
std140/std430 putVec2Array/putIVec2Array |
stride=16/8 与 size calculator 一致（写入后 buffer position 应为 align + stride*size）；size 与数组长度不一致 |
🟡 |
GlComputePassBackend resident 句柄 |
同一纹理以不同 read/write 需求两次绑定 → 权限合并为非驻留再驻留；dispatch 后 residentHandles 清空；close() 后重复 close 幂等 |
🟡 |
GlDeviceMixin.alrCreateExtendedTexture |
depthOrLayers > 1 的期望行为（当前被忽略）；cubemap usage 位；GL_OUT_OF_MEMORY 抛出路径 |
🟡 |
ALROptions.getPropertyBoolean |
属性缺失 / "false" / "0" / "" 四种取值的期望结果（当前 0、"" 均判为 true） |
🟢 |
OcclusionCuller.createInstance |
alrOcclusionCullingForceImplementation 非法值回退到自动选择；isSupported() 为 false 时跳过对应后端；最终回退 NoOp |
🟢 |
由 Hermes Agent 审查
|
💾 Self-improvement review: Patched SKILL.md in skill 'rendering-gpu-pr-review' (1 replacement). · Patched SKILL.md in skill 'pr-diff-scope-reconciliation' (1 replacement). · Patched references/pr95-occlusion-culling-review-matrix.md in skill 'rendering-gpu-pr-review' (1 replacement). |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR extends the Minecraft 26.1 rendering module with an occlusion-culling framework and GPU-side rendering infrastructure. It adds an OpenGL samples-query backend, Hi-Z depth-processing foundations, feature-submit filtering, and compute/texture capability plumbing.
Changes
OcclusionCuller,OcclusionKey,OcclusionMethod, culling statistics, and wrapped submit-node storage.R32Fdepth conversion and FidelityFX SPD mip-chain foundations.ApiStatus.Internalplacement and sync-module imports.test_occlusionblock, block entity renderer, generated assets, and compute-test binding updates.Impact and Compatibility
ARB_bindless_texture; the regular SPD path remains available.ALRComputePassnow uses explicit bindings, lazySupplier<String>debug groups, andpopDebugGroup(); internal callers were updated.