TA-Lib EMA benchmark: native builds and output checks
Run the original, compiler-tuned and TAKT builds on the same generated series. No account, source upload, Python packages or administrator access are needed.
Run the demo
Linux x86-64 with AVX2 and FMA, glibc, Python 3 and curl. Download the script, inspect it if you wish, then run it. It downloads only public benchmark binaries and the measurement tool; it sends no readings back.
curl -fLO https://taktcycles.com/demo/talib.py python3 talib.py
Read the result
Nine interleaved rounds compare the EMA parameter sweep over one million bars. The tool checks equal output hashes and reports cycles, instructions, wall time and uncertainty. Compare TAKT with both original and tuned builds; use the stronger baseline.
What this proves
This is one published workload, not a forecast for your program. Hash equality in this demo supplements the full bitwise checks in the package. If hardware counters are unavailable, the demo measures wall time without changing system settings. CPU features are checked before any benchmark runs.
Full measurements and equivalence checks
Bring one slow run
Tell us the language, platform, duration, frequency and outputs that must match. No code is needed for the first assessment. Rust can proceed in the dashboard; C, C++ and Python use an individual level-2 order.