Type-generic SIMD library for optimized generic code generation
For simple operations, one can simple select block_apply(begin, end, functor), where functor is selected from SIMDTypes<float_type>::apply_##op##_##prec, where op is the operation and prec is the precision. Not all operations support all of u05, u10, and u35, so you can either examine the generated header file from SLEEF or work by trial and error.
For more complex operations, one can access the appropriate assembly instructions and work using the types and instructions contained in SIMDTypes<float_type>. For an example of this kind of use, see this Gaussian finalizer from a Fast Random Fourier Features implementation.