Well, its finally been done – Sgàil has been updated to fix the error described in this post which allowed second pre-image and collision attacks. An error in the reference implementation has been remedied, and a number of typographical corrections have been made to the specification. The known answer tests and intermediate value results have also been regenerated.
Although version 0.4 is considered to be immune to the previous design error and was submitted to NIST, it has not yet been approved/rejected as an official change. Updates will be posted as they happen.
Changelogs have been included in the Supporting_Documentation directory.
The performance of the reference implementation has dropped from 62 cycles/byte to 71 cycles/byte in asymptopic behaviour due to the corrections made to the key schedule. It is expected that an optimised version can improve this to at least 40 cycles/byte; I’m looking at coding this in the near future.
UPDATED 11th December 2009: I really should have updated this about six months ago; the corrections were not accepted by NIST and Sgàil did not progress to Round 2.