И не где-нибудь, а на Cell BE.

Coconut is a developing system for high-assurance, high-performance software. It was used to develop a library of special functions for the Cell BE processor, which is distributed in the Cell BE SDK 3.0 as MASS. Average performance is 4X better than the alternative hand-tuned C library, SimdMath.<br />
<br />
Coconut has been successful where patterns of efficient hardware-specific computation can be captured as higher-order functions and encoded in a Domain Specific Language embedded in Haskell. Patterns include efficient control structures not expressible in C, e.g., the MultiLoop, and efficient uses of SIMD instructions which require significant compile-time computation for pattern specialization. Some patterns interact with a novel instruction scheduler called Explicitly Staged Software Pipelining, based on a min-cut approach, which outperforms SWING modulo scheduling in our tests.

PS
Я понимаю, что обогнать в четыре раза одну библиотеку не то же самое, что обогнать четыре библиотеки.

Но тогда бы заголовок был бы не столь увлекательным.

К тому же этот заголовок помещается в поле ввода.