← all workloads

module_link

n importable modules plus a main that imports and uses all of them. The harness precompiles each module to .slang-module, then compiles main against them, stressing module read + linkIR. Scaling null: n scales module count, each O(1); ideal load+link cost is O(n).

bucket: module_link  ·  mode: link  ·  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.

module_link — phase composition vs N (v2026.13, median ms) module_link 8.9× over N 50→400 0.0 272 544 50 100 200 400 N module_link — parseTranslationUnit module_link — SemanticChecking module_link — generateIR module_link — frontEndExecute (self) module_link — specializeModule module_link — simplifyIR module_link — linkIR module_link — unrollLoopsInModule module_link — legalizeResourceTypes module_link — legalizeExistentialTypeLayout module_link — performMandatoryEarlyInlining module_link — performForceInlining module_link — linkAndOptimizeIR (self) module_link — generateOutput (self) module_link — 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×
50–400101.140.999575042.26×

Growth attribution (N=50 → N=400)

compileInner grows by 447 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=50t@N=400Δ msshare× lin∝Nk
SemanticChecking36289+25357%1.12×1.08
linkIR549+4410%1.34×1.13
generateOutput (self)543+388%1.69×1.29
linkAndOptimizeIR (self)335+327%1.99×1.33
simplifyIR231+296%1.69×1.27

Also growing (below top-5): legalizeResourceTypes (+15 ms, 3%), legalizeExistentialTypeLayout (+15 ms, 3%), specializeModule (+13 ms, 3%).

Near-constant (≤2% of growth each): generateIR (2→11 ms), parseTranslationUnit (0→1 ms), performForceInlining (0→0 ms), performMandatoryEarlyInlining (0→0 ms), unrollLoopsInModule (0→0 ms), compileInner (self) (0→0 ms), frontEndExecute (self) (0→0 ms).

Sweep numbers (median ms)

NcompileInnerlinkIR
50575
1001079
20022320
40050449