You write:
>> The ppc version of bn_mul_comba4 produces an incorrect result because
>> one of the products added into r[5] is wrong.
...
>Isn't it amazing for how long can a bug go unnoticed? This one was
>present in original submission from 2004.

Presumably nobody has used the assmbler code for 32-bit ppc.

>How did you find that
>bn_mul_comba4 is broken?

I fed various random inputs into both the C and assembler versions and
compared them. I think bntest also fails because of this bug, though it was
difficult to run it as I'm cross-compiling and the tests seem to want to
both run the test and perform the analysis on the same machine, so I had
to run the test by hand.

I believe that bn_div_words is also broken (from a similar comparison)
though I haven't bothered investigating it in detail as it doesn't seem
to make much difference to the speed, so the C version is adequate.
-- 
Charles Bryant - [email protected]
______________________________________________________________________
OpenSSL Project                                 http://www.openssl.org
Development Mailing List                       [email protected]
Automated List Manager                           [email protected]

Reply via email to