← all workloads

inlining

n [ForceInline] functions in bounded-depth groups. The inliner collapses each group into its group-top; the entry sums all group-tops, giving the SSA simplifier large bodies to chew on. Scales by breadth (bounded nesting). Scaling null: n scales call sites with bounded inlining depth; ideal cost is O(n).

bucket: inlining  ·  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.

inlining — phase composition vs N (v2026.12, median ms) inlining 35.8× over N 100→800 0.0 1162 2325 100 200 400 800 N inlining — parseTranslationUnit inlining — SemanticChecking inlining — generateIR inlining — frontEndExecute (self) inlining — specializeModule inlining — simplifyIR inlining — linkIR inlining — unrollLoopsInModule inlining — legalizeResourceTypes inlining — legalizeExistentialTypeLayout inlining — performMandatoryEarlyInlining inlining — performForceInlining inlining — linkAndOptimizeIR (self) inlining — generateOutput (self) inlining — 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×
100–800141.840.9916021534.63×

Growth attribution (N=100 → N=800)

compileInner grows by 2093 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=100t@N=800Δ msshare× lin∝Nk
simplifyIR241944+191992%6.12×2.11
SemanticChecking1568+533%0.94×0.94
specializeModule551+462%1.57×1.24

Near-constant (≤2% of growth each): generateIR (6→32 ms), linkAndOptimizeIR (self) (3→29 ms), performForceInlining (1→5 ms), legalizeResourceTypes (0→4 ms), legalizeExistentialTypeLayout (0→4 ms), linkIR (1→4 ms), parseTranslationUnit (1→4 ms), generateOutput (self) (3→4 ms), compileInner (self) (1→2 ms), performMandatoryEarlyInlining (0→1 ms), frontEndExecute (self) (0→1 ms), unrollLoopsInModule (0→0 ms).

Sweep numbers (median ms)

NcompileInnersimplifyIRlinkAndOptimizeIR
100602435
20014288106
400465369408
800215319442043