Adding options to switch kernel at runtime

[DiamonDinoia]

No description provided.

[DiamonDinoia]

The test now catches the problem spotted by @lu1and10 : https://github.com/flatironinstitute/finufft/actions/runs/20095171541/job/57652014814?pr=763

11/23 Test #11: run_accuracy_test_float ..........***Failed    0.39 sec
tol=1.000000e-01 req=2.000000e-01 rel_err=1.046724e+00 -> FAILED (decade=1e-1)

[lu1and10]

very nice, should merge this once master fixed.

[DiamonDinoia]

@lu1and10 the failure in the test might have been another issue that it took me a while to spot.

I think this is ready for review now.

[lu1and10]

by another issue, do you mean ns for upsampfac 2.0 in float can't be 2?
if (std::is_same_v<T, float> && upsampfac == 2.00) ns = std::max(ns, 3);

[DiamonDinoia]

I moved kernel functions in kernel.h kernel.cpp make a static_library for finufft and cufinufft. A bit of cleanup.

Reason why I moved the kernel functions there is that I can link the test against this library and clamp eps to the max_eps allowed per sigma. This function takes kernel_type as input so we can return something different for different kernels.

by another issue, do you mean ns for upsampfac 2.0 in float can't be 2? if (std::is_same_v<T, float> && upsampfac == 2.00) ns = std::max(ns, 3);

yes, if you keep ns = 2 in that case it gives 0(1) errors for tol exactly 1e-1

[DiamonDinoia]

I noticed that there are again formatting changes. I will revert them tomorrow. (I'll have time hopefully)

[lu1and10]

I noticed that there are again formatting changes. I will revert them tomorrow. (I'll have time hopefully)

No worries, it's not urgent and I can try to help with it if you don't have time.

[lu1and10]

@DiamonDinoia I guess the format is not a problem for this PR, if you run clang-format -i -style=file src/*.cpp in master, there are some .cpp files that haven't been formatted for a while since the cpp was missing in the .pre-commit-config.yaml.
Maybe we could just make a single commit only to format the src/*.cpp files in master and put it in .git-blame-ignore-revs, what do you think.

[DiamonDinoia]

That might result in more work in fixing conflicts. Reverting them is not much work, just tedious

…

On Wednesday, December 10, 2025, Libin Lu ***@***.***> wrote: *lu1and10* left a comment (flatironinstitute/finufft#763) <https://urldefense.com/v3/__https://github.com/flatironinstitute/finufft/pull/763*issuecomment-3638567654__;Iw!!DSb-azq1wVFtOg!T8IusygBvyknExvROjXI6wkux72FNwAxjLuOCBvm7Hk144UehNsmuJrx32seNYQY07jM3Kz73QRs1uFaA3MmvOeSi1e4v7ZQ$> @DiamonDinoia <https://urldefense.com/v3/__https://github.com/DiamonDinoia__;!!DSb-azq1wVFtOg!T8IusygBvyknExvROjXI6wkux72FNwAxjLuOCBvm7Hk144UehNsmuJrx32seNYQY07jM3Kz73QRs1uFaA3MmvOeSi5G44096$> I guess the format is not a problem for this PR, if you run clang-format -i -style=file src/*.cpp in master, there are some .cpp files that haven't been formatted for a while since the cpp was missing in the .pre-commit-config.yaml. Maybe we could just make a single commit only to format the src/*.cpp files in master and put it in .git-blame-ignore-revs, what do you think. — Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/flatironinstitute/finufft/pull/763*issuecomment-3638567654__;Iw!!DSb-azq1wVFtOg!T8IusygBvyknExvROjXI6wkux72FNwAxjLuOCBvm7Hk144UehNsmuJrx32seNYQY07jM3Kz73QRs1uFaA3MmvOeSi1e4v7ZQ$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/ACGKNQJ35LFZZV6EA462GZD4BBVU7AVCNFSM6AAAAACOQPZ7BOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMMZYGU3DONRVGQ__;!!DSb-azq1wVFtOg!T8IusygBvyknExvROjXI6wkux72FNwAxjLuOCBvm7Hk144UehNsmuJrx32seNYQY07jM3Kz73QRs1uFaA3MmvOeSi1YLXM6K$> . You are receiving this because you were mentioned.Message ID: ***@***.***>