← all workloads

reflection_layout

n constant buffers, each with a structurally rich payload — vectors, matrices, an array of a nested `Light` struct, a nested `Material` struct and a scalar array. This is the only workload with a large, deeply-typed shader *parameter interface* (every other workload's interface is a single RWStructuredBuffer), so it is the one stressor for the parameter binding / layout-assignment engine and — with `-reflection-json` (see the spec's reflection_json flag) — the reflection serializer, the layout/reflection path no other workload covers. Scales by breadth = number of parameter blocks the layout engine must place and reflect. Note: layout is computed during compileInner regardless of the flag (so compileInner is the holistic signal); -reflection-json additionally runs the serializer, which is cheap today but tracked here for regression coverage. Scaling null: n scales parameters; ideal layout cost is O(n).

bucket: reflection_layout  ·  mode: target  ·  flags: -target spirv -emit-spirv-directly

Phase composition vs N (stacked sub-counters)

compileInner split into phase buckets (named leaves + (self) residuals) stacked across the sweep sizes — the top edge is compileInner, so you can see which phase drives the scaling.

reflection_layout — phase composition vs N (v2026.13, median ms) reflection_layout 14.5× over N 30→240 0.0 368 735 30 60 120 240 N reflection_layout — parseTranslationUnit reflection_layout — SemanticChecking reflection_layout — generateIR reflection_layout — frontEndExecute (self) reflection_layout — specializeModule reflection_layout — simplifyIR reflection_layout — linkIR reflection_layout — unrollLoopsInModule reflection_layout — legalizeResourceTypes reflection_layout — legalizeExistentialTypeLayout reflection_layout — performMandatoryEarlyInlining reflection_layout — performForceInlining reflection_layout — linkAndOptimizeIR (self) reflection_layout — generateOutput (self) reflection_layout — compileInner (self) phase buckets parseTranslationUnit SemanticChecking generateIR frontEndExecute (self) specializeModule simplifyIR linkIR unrollLoopsInModule legalizeResourceTypes legalizeExistentialTypeLayout performMandatoryEarlyInlining performForceInlining linkAndOptimizeIR (self) emitEntryPointsSourceFromIR generateOutput (self) compileInner (self)

Scaling analysis

floor-subtracted power-law fit (t − floor) = a·Nk; floor = the minimal workload (fixed per-compile cost), k the global exponent, top-2× the local high-end doubling ratio.

N rangefloor (ms)k (work)fit R²t(Nmin)t(Nmax)top-2×
30–240101.400.991476813.16×

Growth attribution (N=30 → N=240)

compileInner grows by 634 ms across the sweep; the mutually-exclusive phase buckets below partition that growth exactly (no nested-timer double counting). × lin is the same metric as the top-level panels, per bucket: the end point vs a linear expectation anchored to the bucket's share of the minimal floor and fitted on the low-N half — 1.0 = grew exactly linearly, >1 bends up. The super-linearity lives where × lin (and k) are red.

buckett@N=30t@N=240Δ msshare× lin∝Nk
SemanticChecking16246+23036%2.63×1.50
linkAndOptimizeIR (self)5121+11618%2.68×1.56
generateOutput (self)796+9014%2.21×1.44
legalizeExistentialTypeLayout140+396%4.23×1.84
legalizeResourceTypes139+386%4.39×1.87

Also growing (below top-5): simplifyIR (+31 ms, 5%), generateIR (+22 ms, 3%), frontEndExecute (self) (+17 ms, 3%), compileInner (self) (+14 ms, 2%), parseTranslationUnit (+13 ms, 2%).

Near-constant (≤2% of growth each): specializeModule (1→11 ms), performForceInlining (1→10 ms), linkIR (1→7 ms), performMandatoryEarlyInlining (0→2 ms), unrollLoopsInModule (0→0 ms).

Sweep numbers (median ms)

NcompileInnerfrontEndExecutegenerateOutput
30472421
60904244
12021591116
240681305359