Automatic generation of high speed elliptic curve cryptography code

Abstract

Apparently, trust is a rare commodity when power, money or life itself are at stake. History is full of examples. Julius Caesar did not trust his generals, so that: ``If he had anything confidential to say, he wrote it in cipher, that is, by so changing the order of the letters of the alphabet, that not a word could be made out. If anyone wishes to decipher these, and get at their meaning, he must substitute the fourth letter of the alphabet, namely D, for A, and so with the others.'' And so the history of cryptography began moving its first steps. Nowadays, encryption has decayed from being an emperor's prerogative and became a daily life operation. Cryptography is pervasive, ubiquitous and, the best of all, completely transparent to the unaware user. Each time we buy something on the Internet we use it. Each time we search something on Google we use it. Everything without (almost) realizing that it silently protects our privacy and our secrets. Encryption is a very interesting instrument in the toolbox of because it has very few side effects, at least on the user side. A particularly important one is the intrinsic slow down that its use imposes in the communications. High speed cryptography is very important for the Internet, where busy servers proliferate. Being faster is a double advantage: more throughput and less server overhead. In this context, however, the public key algorithms starts with a big handicap. They have very bad performances if compared to their symmetric counterparts. Due to this reason their use is often reduced to the essential operations, most notably key exchanges and digital signatures. The high speed public key cryptography challenge is a very practical topic with serious repercussions in our technocentric world. Using weak algorithms with a reduced key length to increase the performances of a system can lead to catastrophic results. In 1985, Miller and Koblitz independently proposed to use the group of rational points of an elliptic curve over a finite field to create an asymmetric algorithm. Elliptic Curve Cryptography (ECC) is based on a problem known as the ECDLP (Elliptic Curve Discrete Logarithm Problem) and offers several advantages with respect to other more traditional encryption systems such as RSA and DSA. The main benefit is that it requires smaller keys to provide the same security level since breaking the ECDLP is much harder. In addition, a good ECC implementation can be very efficient both in time and memory consumption, thus being a good candidate for performing high speed public key cryptography. Moreover, some elliptic curve based techniques are known to be extremely resilient to quantum computing attacks, such as the SIDH (Supersingular Isogeny Diffie-Hellman). Traditional elliptic curve cryptography implementations are optimized by hand taking into account the mathematical properties of the underlying algebraic structures, the target machine architecture and the compiler facilities. This process is time consuming, requires a high degree of expertise and, ultimately, error prone. This dissertation' ultimate goal is to automatize the whole optimization process of cryptographic code, with a special focus on ECC. The framework presented in this thesis is able to produce high speed cryptographic code by automatically choosing the best algorithms and applying a number of code-improving techniques inspired by the compiler theory. Its central component is a flexible and powerful compiler able to translate an algorithm written in a high level language and produce a highly optimized C code for a particular algebraic structure and hardware platform. The system is generic enough to accommodate a wide array of number theory related algorithms, however this document focuses only on optimizing primitives based on elliptic curves defined over binary fields