Path: utzoo!attcan!uunet!husc6!ogccse!littlei!intelisc!omepd!randys
From: randys@omews2.intel.com (Randy Steck)
Newsgroups: comp.arch
Subject: Re: RISC v. CISC --more misconceptions
Message-ID: <3914@omepd>
Date: 11 Nov 88 01:50:05 GMT
References: <156@gloom.UUCP> <18931@apple.Apple.COM> <40@sopwith.UUCP> <998 <10774@cup.portal.com>
Sender: news@omepd
Reply-To: randys@omews2.UUCP (Randy Steck)
Organization: Intel Corp., Hillsboro
Lines: 23

In article <10774@cup.portal.com> PLS@cup.portal.com (Paul L Schauble) writes:
>> Find the distance to the next 1 in a bit stream FAST
>
>Scan for the first non-zero word. Load word and do a floating normalize. Look
>at the exponent. 
>
>  ++PLS

I'm afraid that it isn't quite this simple.  Leaving aside the complexities of
what FP format you use to load it with, you first of all have to determine
what exponent you are going to use.  Zero exponents are a special case.  You
have to preset the exponent, shift and mask the bit stream, do the
normalization, shift and mask to get the exponent value out, and unbias it
from the original input.  Hmmmmmm.  Sounds messy to me.

However, this is the right idea.  On the 80960, we used the floating point
normalizer, without all the checks on the values, to implement the scan_bit
operation (finds the first set bit in a word).  The thing that saved us here
was independent access to the exponent values and fast transfer and operation
on bit values.

Randy Steck
Intel Corp.