We submitted new implementations of Keccak to the eBASH project. In addition to the plain C 32-bit and 64-bit implementations previously submitted, the new variants take advantage of the 64-bit MMX or 128-bit SSE2 instructions of the AMD and Intel processors.
When used on the reference processor defined by NIST, restricted to 32-bit instructions, Keccak achieves about 15 cycles/byte using SSE2 (versus 26.5 cycles/byte in plain C, on x86 katana). When unrestricted, the reference processor allows Keccak to run at about 10 cycles/byte.
The MMX variants are useful for older x86 processors.