Add codspeed benchmarks

[AntoinePrv]

No description provided.

[codspeed]

Congrats! CodSpeed is installed 🎉

🆕 732 new benchmarks were detected.

You will start to see performance impacts in the reports once the benchmarks are run from your default branch.

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

Detected benchmarks

• fma3+avx2/abs/f64/aligned/scalar[1024]: 269.1 ns
• fma3+avx2/abs/f64/aligned/scalar[1027]: 259.9 ns
• fma3+avx2/abs/f64/aligned/scalar[2097152]: 540.5 µs
• fma3+avx2/abs/f64/aligned/scalar[64]: 15.2 ns
• fma3+avx2/abs/f64/aligned/scalar[65536]: 17.3 µs
• fma3+avx2/abs/f64/aligned/scalar[67]: 17 ns
• fma3+avx2/abs/f64/aligned/simd/map:noheader[1024]: 64.1 ns
• fma3+avx2/abs/f64/aligned/simd/map:noheader[1027]: 76.7 ns
• fma3+avx2/abs/f64/aligned/simd/map:noheader[2097152]: 372.6 µs
• fma3+avx2/abs/f64/aligned/simd/map:noheader[64]: 4.5 ns
• fma3+avx2/abs/f64/aligned/simd/map:noheader[65536]: 9.1 µs
• fma3+avx2/abs/f64/aligned/simd/map:noheader[67]: 10.7 ns
• fma3+avx2/abs/f64/aligned/simd/map:noheader+pure+unroll4[1024]: 53.2 ns
• fma3+avx2/abs/f64/aligned/simd/map:noheader+pure+unroll4[1027]: 61.3 ns
• fma3+avx2/abs/f64/aligned/simd/map:noheader+pure+unroll4[2097152]: 355.4 µs
• fma3+avx2/abs/f64/aligned/simd/map:noheader+pure+unroll4[64]: 3.5 ns
• fma3+avx2/abs/f64/aligned/simd/map:noheader+pure+unroll4[65536]: 9.3 µs
• fma3+avx2/abs/f64/aligned/simd/map:noheader+pure+unroll4[67]: 11.1 ns
• fma3+avx2/abs/f64/aligned/simd/map:pure[1024]: 87.3 ns
• fma3+avx2/abs/f64/aligned/simd/map:pure[1027]: 70.6 ns
• ...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

---

[AntoinePrv]

I don't know if codespeed is the right tool for the job at this point.

• I'm afraid the simulation mode may be too limited to model throughput, intruction-level-parallelism, dependency chains, branch predictions, and lack of support for advanced instructions (AVX512). These are all things we are trying to get right here.
• With the walltime setting, I am not sure we gain much since we are not using their runners
• We do not need the profiling since we focus on a single flat, fully inlined functions.
• The benchmarks plotting is what I would be interested but is lacking: compare multiple runs (e.g. map-scalar vs map-naive), and plot as a function of input size.