🚀 go-pugleaf

Germano Caronni (caronni@nessie.cs.id.ethz.ch) wrote:
: In article <1494@eouk9.eoe.co.uk> ahaley@eoe.co.uk (Andrew Haley) writes:
: >
: >The only thing I have seen which might be faster is Khufu.  Certainly
: >IDEA (as used in PGP) isn't any faster, and if you use enough rounds
: 
: Have you tried it ? (pgp -c is not a very fair test, as there are
: some tmpfiles which let me think that pgp tries compression&encryption:-) )

I tried IDEA when it was still called PES [*] (Proposed Encryption
Standard.)  This was long before PGP became available.  My
implementation was entirely in carefully optimized 386 assembler, as
was my DES implementation, so I think that the comparision is fair.

That IDEA should not be a great deal faster than DES isn't very
surprising.  IDEA has four modular multiplications in its inner loop,
(each one of which involves multiplication, test, and shift) and a few
additions and XORs.  DES requires four mask/lookup/XOR/shift
operations in its inner loop, but has 16 rounds, compared to IDEA's 8.
The speed advantage of IDEA depends on the speed ratio of 32 bit table
lookup/ XOR to 16 bit multiply.

Interestingly enough, if you were to use a machine (such as a DSP)
with a flash multiplier things would be very different.  I would
expect IDEA to be much faster than DES on such a machine, always
assuming that the other IDEA operations could be efficiently coded.

Andrew.

[*] In fact, PES is slightly different from IDEA, but runs at almost
identical speed.