That's not quite true, cython supports parallel for loops either by directly calling openmp or via cython.parallel.prange etc. (which wraps openmp AFAIK) you ca

That's not quite true, cython supports parallel for loops either by directly calling openmp or via cython.parallel.prange etc. (which wraps openmp AFAIK) you can also get simd speed ups via directly calling simd intrinsics or compiling with simd support and relying on the compiler (which is admittedly is not as sophisticated as numbas simd support). For GPU support I would likely go and use cupy or arrayfire.