Files
2nd/10_Wiki/Topics/Backend/Render State.md
T
Antigravity Agent f8b21af4be Wiki cleanup: error-doc removal, dedup merge, link normalization
10_Wiki/Topics 대규모 정리:
- 오류 캡처/미완성 stub 문서 227개 제거
- 교차폴더 중복 43클러스터 병합 (63파일 → redirect)
- 링크명 정규화: 깨진 링크 수정·redirect 직결·개념 매핑 ~2,400건
- 카테고리 MOC 6개 신규 생성
- Graph 섹션 미해결 related-keyword 링크 10,058건 제거

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-20 23:52:15 +09:00

4.7 KiB

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack
id title category status canonical_id aliases duplicate_of source_trust_level confidence_score verification_status tags raw_sources last_reinforced github_commit tech_stack
wiki-2026-0508-render-state Render State 10_Wiki/Topics verified self
GPU Render State
Pipeline State
Graphics State
none A 0.9 applied
graphics
gpu
rendering
pipeline
2026-05-10 pending
language framework
hlsl/glsl vulkan/d3d12

Render State

매 한 줄

"매 GPU 명령마다 어떤 컨텍스트에서 그릴지 결정하는 파라미터 묶음". 매 blend mode, depth test, rasterizer setup, shader binding 의 collection. 매 modern Vulkan/D3D12 에서는 PSO (Pipeline State Object) 의 immutable bake — state change cost 의 minimization.

매 핵심

매 구성 요소

  • Rasterizer state: cull mode, fill mode, depth bias.
  • Depth-stencil state: depth test, stencil ops, write masks.
  • Blend state: src/dst factors, blend op, write mask.
  • Input layout: vertex attribute format.
  • Shader stages: VS / PS / GS / HS / DS / CS bindings.
  • Descriptor sets / root signature: resource bindings.

매 진화

  • OpenGL/D3D9-11: per-state setter — glEnable(GL_DEPTH_TEST) style. Driver 의 lazy validation cost 큼.
  • D3D12 / Vulkan / Metal: PSO bake at create time. Runtime change = bind 1 PSO.
  • WebGPU 2025: Vulkan style PSO + 동일 abstractions.

매 응용

  1. Forward / deferred 의 pass 별 PSO 분리.
  2. Shadow pass 의 depth-only PSO.
  3. UI overlay 의 alpha-blend PSO.

💻 패턴

Vulkan PSO 생성

VkGraphicsPipelineCreateInfo info{};
info.stageCount = 2;
info.pStages = stages;             // VS + PS
info.pVertexInputState = &vi;
info.pInputAssemblyState = &ia;
info.pRasterizationState = &rs;    // cull, fill
info.pDepthStencilState = &ds;     // test, write
info.pColorBlendState = &cb;       // alpha blend
info.layout = pipelineLayout;
info.renderPass = rp;
vkCreateGraphicsPipelines(dev, cache, 1, &info, nullptr, &pso);

State sort 의 draw call batching

std::sort(drawables.begin(), drawables.end(), [](auto& a, auto& b){
  if (a.psoHash != b.psoHash) return a.psoHash < b.psoHash;
  if (a.materialHash != b.materialHash) return a.materialHash < b.materialHash;
  return a.depth < b.depth;
});
for (auto& d : drawables) {
  if (d.psoHash != lastPso) cmd.bindPipeline(d.pso);
  if (d.materialHash != lastMat) cmd.bindDescriptorSets(...);
  cmd.draw(...);
}

Dynamic state subset

VkPipelineDynamicStateCreateInfo dyn{};
VkDynamicState ds[] = {VK_DYNAMIC_STATE_VIEWPORT, VK_DYNAMIC_STATE_SCISSOR};
dyn.dynamicStateCount = 2;
dyn.pDynamicStates = ds;
// PSO 는 viewport/scissor 없이 bake → runtime 에서 cmd.setViewport()

D3D12 root signature + PSO

CD3DX12_PIPELINE_STATE_STREAM_DESC desc;
desc.pRootSignature = rootSig;
desc.VS = {vsBlob, vsSize};
desc.PS = {psBlob, psSize};
desc.RasterizerState = CD3DX12_RASTERIZER_DESC(D3D12_DEFAULT);
desc.BlendState = CD3DX12_BLEND_DESC(D3D12_DEFAULT);
desc.DepthStencilState = CD3DX12_DEPTH_STENCIL_DESC(D3D12_DEFAULT);
device->CreateGraphicsPipelineState(&desc, IID_PPV_ARGS(&pso));

Shader hot-reload + PSO cache

auto hash = HashShaders(vs, ps) ^ HashState(rs, ds, cb);
if (auto it = cache.find(hash); it != cache.end()) return it->second;
auto pso = CreatePso(...);
cache[hash] = pso;
return pso;

Bindless descriptor + PSO 단순화

// Descriptor heap 1개 + bindless indexing → PSO 변경 거의 없음
ConstantBuffer<uint> meshIdx;
Texture2D<float4> textures[] : register(t0, space0);
float4 color = textures[NonUniformResourceIndex(materialId)].Sample(...);

매 결정 기준

상황 Approach
AAA 엔진, 수만 draw call PSO + sort by state hash
Indie, simple 2D Dynamic state OK, PSO 1-2개
Shader 자주 바뀜 (dev) PSO cache + hot-reload
Mobile (tile-based) Render pass 단위 minimization

기본값: PSO bake at level load + state-sorted draw queue.

🔗 Graph

🤖 LLM 활용

언제: state-sort 알고리즘 작성, PSO cache 설계, Vulkan boilerplate 생성. 언제 X: GPU vendor 별 micro-optimization (driver 의 black-box).

안티패턴

  • Per-draw PSO recreation: stutter 발생. Cache + warmup 필수.
  • State leak: legacy GL 코드 의 glEnable 후 disable 누락 → 다음 frame corruption.
  • Over-granular PSO: 1만 PSO → memory + driver overhead.

🧪 검증 / 중복

  • Verified (Vulkan spec, D3D12 docs, GPU Gems).
  • 신뢰도 A.

🕓 Changelog

날짜 변경
2026-05-08 Phase 1
2026-05-10 Manual cleanup — Render state full coverage