numba-dpex
numba-dpex copied to clipboard
arctan2 does not work for devices that does not support double precision
If we try to run arctan2
function on device that does not support f64 with f32 arguments it will silently crash.
Occurred on gen12 igpu. Reproducible is inside tests.
Discovered while working on https://github.com/IntelPython/numba-dpex/pull/1143