Implementation of arange algorithm

[vvish]

Adresses #1 (xtensor-stack/xsimd#121)

[JohanMabille]

Thanks for the fix! Let us know it's ready for review.

[vvish]

Thanks for the fix! Let us know it's ready for review.

Thank you. Please have a look.

I also have a question: are there any CI driven benchmarks in the stack that could also be integrated in this repository? I expect these algorithms to have different relative performance on different platforms and was wondering if something already exists in the stack that can help to make informed decision about applicability and drive further optimizations.

[JohanMabille]

Thanks again for this. We only have "hand-crafted" benchmarks in xsimd.

We use Google Benchmark in xtensor, but not sure there's a lot we can reuse here. I'm open to suggestions.

[vvish]

Hi @JohanMabille .
Unfortunately only recently found again some time to work on benchmark integration.
I prepared some benchmarks in the fork https://github.com/vvish/xsimd-algorithm/blob/main/benchmark/bench_algorithms.cpp . They measure timings for xsimd and analogous std algorithm implementations. Should provide better visibility into performance on various target platforms.
Corresponding trivial CI job uses google/benchmark-supplied tool to compare results. This information can be reported in job summary as displayed here https://github.com/vvish/xsimd-algorithm/actions/runs/34756839542

Experiments with comparison against previous runs was unsuccessful as GH Actions seem to have huge timing fluctuations so it was not reliable.

What do you think?