1994-08-26 - DSPs

Header Data

From: Phil Karn <karn@unix.ka9q.ampr.org>
To: perry@imsi.com
Message Hash: 38c79e33126e8020fe07fd47b663d959ddf7a00e619689160e654a6d4cde3fa4
Message ID: <199408260422.VAA16628@unix.ka9q.ampr.org>
Reply To: <9408241340.AA03320@snark.imsi.com>
UTC Datetime: 1994-08-26 04:18:04 UTC
Raw Date: Thu, 25 Aug 94 21:18:04 PDT

Raw message

From: Phil Karn <karn@unix.ka9q.ampr.org>
Date: Thu, 25 Aug 94 21:18:04 PDT
To: perry@imsi.com
Subject: DSPs
In-Reply-To: <9408241340.AA03320@snark.imsi.com>
Message-ID: <199408260422.VAA16628@unix.ka9q.ampr.org>
MIME-Version: 1.0
Content-Type: text/plain

I'm not really an expert on DSPs, but it does seem like they should be
good for modular exponentiation given the number of multiplies required.
But I don't think I'm the first to point that out.

I think I already said that it looks like fast multiply performance is
going to be crucial for secure communications. Modular exponentiation
for public key algorithms, modulo-65537 multiplies in IDEA, and the
many DSP multiplies in CELP speech coding, just to name three.

I recently heard of a FED-STD-1016 CELP implementation that runs in
better than real time on the PowerPC. It runs in much worse than real
time on just about everything else. The PowerPC's multiply performance is
clearly the reason.