On Thu, 30 Apr 2026 13:30:22 GMT, Andrew Dinn <[email protected]> wrote:

> If I understand correctly this patch employs a different algorithm when 
> interpreting the kyber MultJava method vs when executing it compiled 
> (admitted that they are functionally equivalent). Does this have any 
> implications for reliability or security that might need to be considered?

No, both methods run in constant time and both are reliable. As the inputs to 
this method are of very limited size (the entries in the input arrays, as well 
as the constants used are in the [-3329, 3329] range) it is possible to run 
(and I did that) a test that proves that the partial results stay in the range 
where they should. Correctness is tested the same way as before (for the API 
level functionality).

-------------

PR Comment: https://git.openjdk.org/jdk/pull/30991#issuecomment-4352983656

Reply via email to