# SLeeLa Pi 20,000-Digit Native Math Benchmark **Date:** 2026-11-09 **Status:** Historical PASS (original branch run; current-master rerun is now automated) **Branch:** `pi-20000-sleelactl-native-math` **Test commit:** `2feed4fb627adc22196766f55dd7706fa9812773` **Workflow run:** [Pi 20,000-digit native math benchmark](https://github.com/mearvk/SLeeLa/actions/runs/42690168886) ## Result > Historical result from the `pi-20000-sleelactl-native-math` branch. This report preserves the original successful run and does not claim that current `master` has reproduced it. The test exercised VM big-integer opcodes; it did not demonstrate `@native` VM-to-`sleelactl` worker dispatch. SLeeLa generated **20,000 fractional digits of π** with the Chudnovsky implementation in `PiChudnovsky.sleela`. The complete output contains 20,002 ASCII characters (`3.` plus 20,000 fractional digits). An independent `mpmath` reference, evaluated at 20,080 decimal places, was compared character-for-character with the SLeeLa output. Every digit matched. - Verification: **PASS — all 20,000 fractional digits** - Output SHA-256: `977f8d4251b4f633100b4d2dbc214b874c45353b0e3debe41046753777fd6efc` - Prefix: `3.14159265358979323846264338327950288419716939937510` - Final 52 characters: `2429681062037765788371669091094180744878140490755178` ## Execution and measurements | Measurement | Result | |---|---:| | Runner OS | Ubuntu 26.04.1 LTS | | Runtime build | PASS (`make -C impl clean all`) | | Trusted manifest verification | PASS (12,499 items) | | `sleelactl run math` native subject smoke test | PASS | | Pi generation | PASS | | Independent digit comparison | PASS | | Wall-clock time for Pi generation | 10.72 s | | User CPU time | 9.87 s | | System CPU time | 0.84 s | | CPU utilization | 99% | | Peak resident memory | 49,724 KiB (about 48.6 MiB) | | Process exit status | 0 | The Pi generation command was: ```sh timeout 30m impl/build/sleela run pi-generator/PiChudnovsky20k.sleela ``` The registered native math subject was separately exercised through: ```sh sh bin/sleelactl run math ``` That distinction matters: `sleelactl run math` runs the native math subject demonstration; the arbitrary-precision Pi program is executed by the freshly built SLeeLa runtime. Both stages passed in the same CI job. ## Implementation under test - Program: `pi-generator/PiChudnovsky.sleela` - Algorithm: Chudnovsky series with binary splitting - Precision target: 20,000 digits after the decimal point - Arithmetic: SLeeLa VM arbitrary-precision integer operations (`bigmul`, `bigadd`, `bigsub`, `bigdiv`, `bigisqrt`, and `bigdivmod`) ## Reproduction Run the workflow on branch `pi-20000-sleelactl-native-math`, or dispatch the workflow manually. The job regenerates the SHA-256 source manifest, builds the runtime, executes the native math subject smoke test, runs the Pi program, and compares its output against the independent `mpmath` reference. The workflow preserves the output, timing report, and verification summary as the `sleela-pi-20000-benchmark` artifact. ## Interpretation and limitations This is a successful correctness and single-run performance result on the specified hosted Linux runner. The 10.72-second timing and memory figure are measurements of this run, not a cross-platform guarantee or a statistically characterized benchmark. No speedup ratio is claimed: comparisons with other algorithms or earlier scalar math tests would not be like-for-like without matched workloads, hardware, and measurement conditions. The workflow and its artifacts are the authoritative record for this result.