Benchmarks

Repeatable performance proof, not marketing numbers.

The public harness separates check, bytecode, native build, native run, and optional external language comparisons.

startup

Minimal process launch and program startup.

sum_loop

Integer loop lowering and arithmetic throughput.

string_scan

Byte scanning, nested loops, and accumulation.

file_io

Binary write, size check, readback, accumulation, and cleanup.

Comparison Fixtures

Comparison sources are visible in the repository. Missing optional toolchains are reported as skipped, not as benchmark results.

Windows

powershell -ExecutionPolicy Bypass -File .\benchmarks\run.ps1 -IncludeExternal

Linux / WSL

bash benchmarks/run.sh ./TezzNative-language/bin/tezzc-linux-x64 --iterations 3 --include-external

CSV rows use tezznative.benchmark-result.v1. Metadata files use tezznative.benchmark-run.v1.

Elapsed time, peak memory where available, binary size, command, exit code, timeout status, and output hash.

Bytecode, native build, native run, and external language runs are separate rows.

Category	Gate
JSON parsing	Waiting for a stable JSON fixture.
HTTP throughput	Waiting for a local load driver and stable server fixture.
Matrix math	Waiting for a real supported numeric backend gate.
GPU/NPU/LLM	Experimental only; fallback results are not acceleration results.