interp() performance regression in 2025.1.0+

What happened?

A performance regression has been introduced between 2024.11.0 and 2025.1.0, possibly in #9881, as was the case in #10287. This issue is still present in 2025.8.0. In the example running on my machine it is a 76x performance regression. It seems to be present when interpolating 2D arrays but the degradation is more extreme with 3D arrays.

What did you expect to happen?

No performance regression.

Minimal Complete Verifiable Example

import numpy as np
import xarray as xr

import time

n_repeats = 10

# array dims
n_t = 122
n_r = 65
n_z = 65

# generate some random data to interpolate
rng = np.random.default_rng(seed=42)

data = rng.random((n_t, n_r, n_z))
data_t = np.linspace(0.01, 0.855, n_t)
data_r = np.linspace(0.04, 4., n_r)
data_z = np.linspace(-1.2, 1.2, n_z)

da = xr.DataArray(
    data=data,
    dims=["t", "r", "z"],
    coords=dict(
        t=data_t,
        r=data_r,
        z=data_z,
    ),
)

interp_z_data = rng.random((n_t))
interp_z = xr.DataArray(
    data=interp_z_data,
    dims=["t"],
    coords=dict(
        t=data_t,
    ),
)

start = time.time()
for _ in range(n_repeats):
    # regression is here:
    da.interp(z=interp_z)
end = time.time()

t_elapsed = end - start
print(f"Time elapsed: {t_elapsed}")

MVCE confirmation

• Minimal example — the example is as focused as reasonably possible to demonstrate the underlying issue in xarray.
• Complete example — the example is self-contained, including all data and the text of any traceback.
• Verifiable example — the example copy & pastes into an IPython prompt or Binder notebook, returning the result.
• New issue — a search of GitHub Issues suggests this is not a duplicate.
• Recent environment — the issue occurs with the latest version of xarray and its dependencies.

Relevant log output

pip install "xarray==2024.11.0"
python test.py
Time elapsed: 0.22251605987548828

pip install "xarray==2025.1.0"
python test.py
Time elapsed: 16.765803813934326

Anything else we need to know?

No response

Environment

Start by running the provided interp() minimal example against the stated xarray versions to reproduce the regression, then inspect the interpolation path used for 2D and 3D arrays. Done means restoring performance near the pre-2025.1.0 behavior and adding a regression test or benchmark for the demonstrated case.