On ARM with CV_SIMD_16F, getMulFunc_, getDivFunc_, and getAddWeightedFunc_ used float16 for 8-bit integer SIMD kernels, leading to precision loss and overflow (> 65504). Switch to float32 to match x86 and ensure precision.