X-pop3-spooler:	POP3MAIL 2.1.0 b 3 961213 -bs-
Delivered-To:	pcg AT goof DOT com
Date:	Mon, 16 Mar 1998 17:02:59 +0200 (EET)
From:	Tuukka Toivonen <tuukkat AT ees2 DOT oulu DOT fi>
X-Sender:	tuukkat AT stekt10
Reply-To:	Tuukka Toivonen <tuukkat AT ees2 DOT oulu DOT fi>
To:	Thomas Koehler <T DOT Koehler AT pfh DOT research DOT philips DOT com>
cc:	beastium-list AT Desk DOT nl
Subject:	Re: paranoia & extra precision [was -fno-float-store in pgcc]
In-Reply-To:	<35094C21.6B3F0566@pfh.research.philips.com>
Message-ID:	<Pine.SOL.3.96.980316162141.12843A-100000@stekt10>
MIME-Version:	1.0
Sender:	Marc Lehmann <pcg AT goof DOT com>
Status:	RO
Lines:	26

On Fri, 13 Mar 1998, Thomas Koehler wrote:

>so what is the difference between the behaviour of programs compiled
>with -ffloat-store compared to setting the FPU-rounding mode to double
>via __setfpucw? (except that the -ffloat-store variante is much slower).

- Even at low precision the numbers in FPU may have much bigger/smaller
  exponent than numbers that are kept in double memory variables.
  I don't know if this violates IEEE (specs cost almost $100 :( )
- If you mix lot of both double and float, the FPU control word has to be
  set every time. Also, if the state of the CW is unknown (like
  calling a library routine) it has to be set. So using low precision
  may be actually slower than using -ffloat-store.

GCC could prepare the needed control words into memory at the beginning of
a program, so that only single fldcw instruction would be needed to 
switch to a another precision mode. Unfortunately this would interfere
if user would also change CW.

--
| Tuukka Toivonen <tuukkat AT ee DOT oulu DOT fi>       [PGP public key
| Homepage: http://www.ee.oulu.fi/~tuukkat/       available]
| Try also finger -l tuukkat AT ee DOT oulu DOT fi
| Studying information engineering at the University of Oulu
+-----------------------------------------------------------

- Raw text -