Turing speed

From APIDesign

Revision as of 08:00, 9 March 2018

Measuring speed of a language is a biased task. Every language comes with certain built-ins suitable for tasks the language is optimized for. Authors of the language usually spend enormous amount of time to optimize these parts and thus comparing languages on their (different) built-ins makes little sense.

The Real Speed

On the other hand, each programming language shall be Turing complete (e.g. support general programming with **if**, **while**, **for** statements). Thus I propose to compare speed of languages on so called Turing speed: measure how fast the language is when performing generic computation (preferably without any intrinsic built-ins).

I am running such experiment with my polyglot sieve of Eratosthenes (https://github.com/jtulach/sieve/) - I am trying to implement simple, long running algorithm in many different languages and compare their speed. See my sieve project (https://github.com/jtulach/sieve/) for the actual results. In addition to that I am helping with OracleLabs’ Graal/Truffle project (https://graalvm.github.io) that aims to implement the fastest virtual machine on the Planet. The project offers various languages with boosted performance: **Ruby**, **R**, **Python**, **JavaScript**, etc. with the vision to make them all really fast from the point of Turing speed.

As far as I can tell we are on the right track, as our **Ruby** implementation is ten times faster than any other implementation. Other languages are on par with it. Default R is probably slower than Python when it comes to Turing speed, but with OracleLabs’ GraalVM (http://www.oracle.com/technetwork/oracle-labs/program-languages/overview/index.html) it is no longer true - both the languages become as fast as native can be!