approx1 and approx2 cubic spline interpolation methods produce suspect values in the last bin at the boundary.

Description

• Used official installers
• Occurs on all backends
• Partial workaround is just using higher resolution so that boundary regions are small
• Bug is reliably reproducible
• Expected output to match scipy's RectBivariateSpline method, which wraps fitpack
• No error output

Example interpolating on cos(x) with comparison to scipy:

Reproducible Code and/or Steps

• Any use of approx1 or approx2 with cubic spline interpolation type

System Information

Tested with AF version 3.7.2 on Mac & 3.8.1 on Ubuntu 20.04, both via the Rust bindings.

Info from Mac for convenience:

Available backends: [OPENCL, CPU]
Active backend: CPU
Number of available devices: 1
Device number: 0
Device info: ("Intel", "CPU", "AppleClang 10.0.0.10001044", "0.0")

Checklist

• Using the latest available ArrayFire release
• GPU drivers are up to date