Add Linux arm64 builds.

[jenskeiner]

This adds Linux builds on arm64 architecture to broaden the platforms we test on.

[codspeed-hq]

CodSpeed Performance Report

Merging #169 will create unknown performance changes

_{Comparing feature/linux-arm (f73b22b) with develop (b4dae23)}

Summary

🆕 704 new
⏩ 88 skipped¹

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_1d[128/400]	N/A	3.3 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_1d[256/800]	N/A	13.1 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_1d[32/100]	N/A	203.6 µs	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_1d[512/1600]	N/A	52.4 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_1d[64/200]	N/A	811.2 µs	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_1d_omp[128/400]	N/A	14.3 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_1d_omp[256/800]	N/A	27.8 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_1d_omp[32/100]	N/A	10.7 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_1d_omp[512/1600]	N/A	72.3 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_1d_omp[64/200]	N/A	11.2 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_2d[16/16/500]	N/A	9.4 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_2d[32/32/1000]	N/A	75.5 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_2d[64/64/2000]	N/A	610.1 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_2d_omp[16/16/500]	N/A	27.4 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_2d_omp[32/32/1000]	N/A	89.9 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_2d_omp[64/64/2000]	N/A	621.6 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_3d[16/16/16/1000]	N/A	303.4 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_3d[4/4/4/250]	N/A	1.2 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_3d[8/8/8/500]	N/A	18.8 ms	N/A
🆕	ubuntu-24.04-arm_clang_kaiserbessel_double_openmp/nfft_adjoint_direct_3d_omp[16/16/16/1000]	N/A	325.2 ms	N/A
...	...	...	...	...

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

Footnotes

1. 88 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[fargito]

Hello, CodSpeed team member here,

This error message is misleading

Too many benchmarks in a single upload

The performance report could not be generated because there were too many benchmarks in a single upload to CodSpeed. We recommend sharding your benchmarks into smaller uploads, see the documentation for more information.

Actually, the issue in this case is the number of jobs in your workflow run. We put fixing this issue on our roadmap. In the meantime, maybe you can try and reduce the number of jobs in the matrix, or run CodSpeed only on a subset to make it work.

[jenskeiner]

Hello, CodSpeed team member here,

This error message is misleading

Too many benchmarks in a single upload

The performance report could not be generated because there were too many benchmarks in a single upload to CodSpeed. We recommend sharding your benchmarks into smaller uploads, see the documentation for more information.

Actually, the issue in this case is the number of jobs in your workflow run. We put fixing this issue on our roadmap. In the meantime, maybe you can try and reduce the number of jobs in the matrix, or run CodSpeed only on a subset to make it work.

Thanks for the heads-up @fargito. We have quite a few different configurations to test, but the overall number of jobs can probably be reduced. We're working on it. Otherwise, this has been working quite well so far, so good job.

I was also testing running benchmarks on macOS recently. After some patches to instrument-hooks I got codspeed-cpp to compile. I could also link against the library, but running the benchmarks inside the provided GitHub Action failed with a runtime error that I could not yet get to the bottom of.