Concluding remarks

In just under 20 years, SIMD extensions have evolved rapidly to become the most efficient technique for accelerating program execution with regular calculations in a single processor or in hardware architectures with more and more cores per multi-core chip. From the initial motivation of accelerating multimedia applications, hence the name multimedia instructions used in the early days, they have become both an extension increasingly resembling the vector instructions of Cray1-type vector machines on the one hand, while at the same time introducing ad hoc instructions enabling general-purpose processors to compete in the multimedia, signal processing or cryptography fields on the other.

Compilers are becoming increasingly adept at using these instructions. For code for which compilers cannot take advantage of these instructions, the use of intrinsics - function...