1994-12-29 - Re: Pentium FDIV

Header Data

From: Jim Gillogly <jim@acm.org>
To: cypherpunks@toad.com
Message Hash: 75d9e1fccfd00d6cba4054fc12297410d45942388d62822ea22e93bb762855a5
Message ID: <199412291807.KAA19437@mycroft.rand.org>
Reply To: <199412291710.AA05400@poboy.b17c.ingr.com>
UTC Datetime: 1994-12-29 18:07:40 UTC
Raw Date: Thu, 29 Dec 94 10:07:40 PST

Raw message

From: Jim Gillogly <jim@acm.org>
Date: Thu, 29 Dec 94 10:07:40 PST
To: cypherpunks@toad.com
Subject: Re: Pentium FDIV
In-Reply-To: <199412291710.AA05400@poboy.b17c.ingr.com>
Message-ID: <199412291807.KAA19437@mycroft.rand.org>
MIME-Version: 1.0
Content-Type: text/plain



> paul@poboy.b17c.ingr.com (Paul Robichaux) writes:
> I don't know how you missed it; it was certainly here. Let me summarize:

> - crypto typically uses integers
> - RSA crypto uses large integers
> - the Pentium FDIV bug only affects floats

Yes, but Prof. Nicely found the problem because he was using FDIV when
doing something or other with large potential primes.  For most of our
crypto stuff integer is fine, but for some operations on 33- to 50-bit
integers it may be faster to use the floating point ops, since IEEE fp is
supposed to be exact for integers up to some number of bits in the 50's.

	Jim Gillogly
	Sterday, 7 Afteryule S.R. 1995, 18:07





Thread