Newsgroups: comp.lang.forth
References: <nnd$61e0ad9a$48ed61c2@b4d945e456041481> <bc18049678801b31614fe87eba889a48@www.novabbs.com> <87cyl31kau.fsf@nightsong.com> <930672243542a2e04c6cd13d83108af9@www.novabbs.com>
From: albert@spenarnc.xs4all.nl
Subject: Re: Avoid treating the stack as an array [Re: "Back & Forth" is back!]
X-Newsreader: trn 4.0-test77 (Sep 1, 2010)
Originator: albert@cherry.(none) (albert)
Message-ID: <nnd$7f5f7952$40153562@9ebab824e23d4e9a>
Organization: KPN B.V.
Date: Tue, 17 Sep 2024 10:47:40 +0200
Path: ...!weretis.net!feeder9.news.weretis.net!usenet.blueworldhosting.com!diablo1.usenet.blueworldhosting.com!feed.abavia.com!abe007.abavia.com!abp001.abavia.com!news.kpn.nl!not-for-mail
Lines: 44
Injection-Date: Tue, 17 Sep 2024 10:47:40 +0200
Injection-Info: news.kpn.nl; mail-complaints-to="abuse@kpn.com"
Bytes: 2676

In article <930672243542a2e04c6cd13d83108af9@www.novabbs.com>,
mhx <mhx@iae.nl> wrote:
>On Mon, 16 Sep 2024 19:16:09 +0000, Paul Rubin wrote:
>
>> mhx@iae.nl (mhx) writes:
>>> This appears not to be a good idea.  The root cause is piling up too
>>> many items on the F-stack (exceeding the hardware FPU stack limits).
>>
>> I wonder if any Forth compilers use SSE instead of the x86 FPU stack.
>
>iForth would, if my tests had showed any positive effect.
>(The effect has to be substantial to outweigh the advantage of 80-bit
> floats whenever accuracy counts.)
>
>I wrote routines to process 4 floats. For unfathomable reasons, they
>are not nearly as good a pre-packaged library code. There is only
>limited potential for standard FP code to benefit from SSE. If
>parallelism can't be exploited, SSE does not seem to bring
>anything over the old FPU. But maybe my hardware was not
>good enough a few years back.
>
>With SSE I need a substantial library for special functions,
>which then become relatively slow DLL calls.
>
>The only thing wrong with the FPU is that the special stack
>overflow interrupts don't work.

In ciforth:
I added floating point support using the FPU with relatively
little work, especially because the transcendentals are easy.
I suspect that it might be not standard. E.g. F+ exhibits
more precision in 80 bits, and we are supposed to use
either IEEE 32 or 64 bits. Apparently I'm in good company
(iforth and vfxforth).
What does the language lawyers say?

>
>-marcel
-- 
Temu exploits Christians: (Disclaimer, only 10 apostles)
Last Supper Acrylic Suncatcher - 15Cm Round Stained Glass- Style Wall
Art For Home, Office And Garden Decor - Perfect For Windows, Bars,
And Gifts For Friends Family And Colleagues.