[{"data":1,"prerenderedAt":178},["ShallowReactive",2],{"run:c2-12500-depth2-prof":3,"run-findings:c2-12500-depth2-prof":152},{"run":4,"config":17,"metrics":97,"curve":103,"assets":151},{"id":5,"title":5,"subtitle":6,"eyebrow":7,"date":8,"workspace":9,"commit":10,"tags":11,"verdict":16},"c2-12500-depth2-prof","C2 計装確認: Lego 12.5k full run (profile=true, pipelining=depth2)。gpu_gap_report で残差 4.54s を cb_gap\u002Fdrvproc\u002Fhandoff\u002Fhost_lead に分解する","Run summary · docs\u002Ffindings\u002Fp3-kernel-share-post-s3.toml §C2 — 計装基盤実装後の 12.5k 実測","2026-08-30","splat","2e16af0",[12,13,14,15],"c2","gap-report","gpu-gate","lego","partial",[18,21,24,27,30,33,36,39,42,45,48,51,54,57,60,63,66,69,72,75,78,80,83,86,88,91,94],{"key":19,"value":20},"dataset","\u002FUsers\u002Fotkrickey\u002Fdev\u002F3dgs-workspace\u002Fdatasets\u002Fnerf_synthetic\u002Flego",{"key":22,"value":23},"gt_convention","Premultiplied",{"key":25,"value":26},"iterations","12,500",{"key":28,"value":29},"seed","42",{"key":31,"value":32},"capacity","1,000,000 splats",{"key":34,"value":35},"sh_degree","3",{"key":37,"value":38},"loss","L1Ssim",{"key":40,"value":41},"lambda","0.200",{"key":43,"value":44},"ssim","window=7 sigma=1",{"key":46,"value":47},"backend.backward","Simd",{"key":49,"value":50},"backend.loss_path","Gpu",{"key":52,"value":53},"backend.emit_pairs","Original",{"key":55,"value":56},"backend.forward","F32",{"key":58,"value":59},"backend.dispatch","PerStage",{"key":61,"value":62},"backend.radix_scan","GpuHierarchical",{"key":64,"value":65},"backend.radix_scatter","SimdBallot",{"key":67,"value":68},"backend.binning","Resident",{"key":70,"value":71},"backend.tile_sort","GlobalRadix",{"key":73,"value":74},"backend.submission","Chained",{"key":76,"value":77},"backend.target_cache","true",{"key":79,"value":77},"backend.shader_fast_math",{"key":81,"value":82},"os_build","26.6.2 (25G83)",{"key":84,"value":85},"backend.ssim_clear","HostMemset",{"key":87,"value":85},"backend.raster_grad_clear",{"key":89,"value":90},"backend.backward_uniform","HostBuffer",{"key":92,"value":93},"backend.lazy_c3_scratch","false",{"key":95,"value":96},"backend.ssim_walk","PerChannel",{"psnr":98,"wallclock":99,"wallclock_regress":100,"splats":101,"final_loss":102},36.067752838134766,"2m 4s",false,406651,"5.620129e-3",{"loss":104,"splats":135},{"iters":105,"values":120},[106,107,108,109,110,111,112,113,114,115,116,117,118,119],1,1000,2000,3000,4000,5000,6000,7000,8000,9000,10000,11000,12000,12500,[121,122,123,124,125,126,127,128,129,130,131,132,133,134],0.593845546245575,0.05460316687822342,0.014249259606003761,0.008560411632061005,0.007587247993797064,0.006679520010948181,0.00622730515897274,0.006026637274771929,0.005886221304535866,0.00580056756734848,0.005698378197848797,0.005589098669588566,0.005623464472591877,0.005620129406452179,{"iters":136,"values":137},[106,107,108,109,110,111,112,113,114,115,116,117,118,119],[138,139,140,141,142,143,144,145,146,147,148,149,150,101],5207,8231,181432,295154,346716,372651,387182,395655,400216,402434,403506,404662,405929,{"has_renders":100,"has_splat":100},[153],{"id":154,"title":155,"date":8,"status":156,"polarity":157,"category":158,"axes":159,"tags":161,"task_code":170,"related_runs":171,"delta_psnr":172,"delta_wallclock":173,"rank":174,"verdict":175,"impact_summary":176,"detail_path":177},"p3-c2-bubble-attribution","C2 実測 — 残差 bubble は host-late 支配 (7.2〜8.1s)、refine tick drain は A7 で半減済み (3.0〜3.8s)。handoff は定常帯で headroom ほぼ無し","stable","mixed","measurement",[106,160],3,[12,13,162,163,164,165,166,167,168,15,169],"cb-gap","host-lead","drvproc","handoff","residual-bubble","refine-drain","post-m4-a7","measurement-only","C2: 12.5k の残差 bubble 帰属計測と分析",[5],"n=1 self-eval: 36.068 dB (100 views)。C2 は測定目的の run で PSNR gate 対象ではないが、p3-final-stack の n=3 mean (36.219 dB) からの乖離は n=1 ノイズ帯として妥当な範囲","train elapsed 124.75s (n=1)。p3-final-stack n=3 mean 124.393s (σ0.756%) の guard 内 — profile=true + gap report 計装のオーバーヘッドは計測可能な歪みを生んでいない、と読める","high","residual-is-host-late-dominated \u002F handoff-headroom-thin \u002F refine-drain-already-halved-by-a7","`p3-kernel-share-post-s3.toml` §4\u002F§8 の宿題 (「残差は既存計装では分解不能、実装後も 12.5k 実測がまだ無い」) を、`2026-08-30-c2-12500-depth2-prof.toml` (M4+A7 default 込み、S3 gate arm B と [backend] bit-exact) の 1 run で埋めた。\u003Cstrong>reliable=true \u002F dropped=0 \u002F kernel_invalid=0 \u002F sum\u002Funion=1.000000x\u003C\u002Fstrong> — 信頼性ゲート健全、全列使用可。wall 124.75s \u002F busy 113.627s (91.1%) \u002F sync-tax 11.126〜11.129s は `p3-final-stack` の n=3 mean (124.393s, σ0.756%) の guard 内 (n=1 の点推定として妥当)。\u003Cstrong>Σcb_gap=11.085s が sync-tax 全量の 99.6% を説明\u003C\u002Fstrong>し、`p50×n` (band 別合計 7.244s) と `mean×n` (11.081s) の差 3.837s を refine tick 由来の外れ値と読むと、\u003Cstrong>残差 (通常境界) 7.2〜8.1s (5.8〜6.5%) \u002F refine tick drain 3.0〜3.8s (2.4〜3.1%)\u003C\u002Fstrong> にブラケットできる。refine 側は `ts_refine`(2.869s)+`ts_opacity_decay`(0.185s)=3.054s の独立 host 実測が下限とほぼ一致し、クロスチェックが成立した。\u003Cstrong>旧 3 分解 (refine 6.26s \u002F encode(B) 3.41s \u002F 残差 4.54s、旧 sync-tax 全量 14.206s) との整合\u003C\u002Fstrong>: 旧「残差」+「encode(B)」= 4.54+3.41=7.95s ≈ 新「通常境界」7.24s (wall が 151.70→124.75s に縮んだ分を割り引けば近い) — \u003Cstrong>旧 encode(B) バケツは独立した加算成分ではなく、非重複だった encode コストが host-late として `cb_gap` に同じ形で現れていた\u003C\u002Fstrong>と判明し、旧 3 分解を「refine drain \u002F それ以外 (通常境界)」の 2 分解へ訂正する形になった。\u003Cstrong>帰属の結論: 通常境界の bubble は host-late が支配的、driver 処理 (`drvproc`) と GPU キュー待ち (`handoff`) は定常帯で中央値がほぼゼロ (headroom 薄い)\u003C\u002Fstrong> — task が例示した「handoff → CB 統合」は主因ではないと実測が明示的に否定した。","\u002Ffindings\u002Fp3-c2-bubble-attribution\u002F",1790227323336]