In current implementation of dcv-color-primitives image conversion may be optimized for sse2 or avx2.
However, if the image width is not multiple of the register size for that optimization, the entire image conversion is performed using x86 implementation.
A better solution is to convert the part of the image that is multiple of register size using the fastest available path and fallback with x86 path only for the remaining part.
In current implementation of dcv-color-primitives image conversion may be optimized for sse2 or avx2.
However, if the image width is not multiple of the register size for that optimization, the entire image conversion is performed using x86 implementation.
A better solution is to convert the part of the image that is multiple of register size using the fastest available path and fallback with x86 path only for the remaining part.