Path: ...!weretis.net!feeder9.news.weretis.net!i2pn.org!i2pn2.org!.POSTED!not-for-mail
From: mitchalsup@aol.com (MitchAlsup1)
Newsgroups: comp.arch
Subject: Re: Continuations
Date: Thu, 18 Jul 2024 20:06:49 +0000
Organization: Rocksolid Light
Message-ID: <cbd50c7559dd1275010025beb88ee89e@www.novabbs.org>
References: <v6tbki$3g9rg$1@dont-email.me> <47689j5gbdg2runh3t7oq2thodmfkalno6@4ax.com> <v71vqu$gomv$9@dont-email.me> <116d9j5651mtjmq4bkjaheuf0pgpu6p0m8@4ax.com> <f8c6c5b5863ecfc1ad45bb415f0d2b49@www.novabbs.org> <7u7e9j5dthm94vb2vdsugngjf1cafhu2i4@4ax.com> <0f7b4deb1761f4c485d1dc3b21eb7cb3@www.novabbs.org> <v78soj$1tn73$1@dont-email.me> <4bbc6af7baab612635eef0de4847ba5b@www.novabbs.org> <v792kn$1v70t$1@dont-email.me> <ef12aa647464a3ebe3bd208c13a3c40c@www.novabbs.org> <v79b56$20oq8$1@dont-email.me> <v7ahnf$2an0d$1@dont-email.me> <20240718193803.00004176@yahoo.com> <v7bi3c$2gk81$1@dont-email.me> <20240718202843.00001dd0@yahoo.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Injection-Info: i2pn2.org;
	logging-data="3748307"; mail-complaints-to="usenet@i2pn2.org";
	posting-account="65wTazMNTleAJDh/pRqmKE7ADni/0wesT78+pyiDW8A";
User-Agent: Rocksolid Light
X-Spam-Checker-Version: SpamAssassin 4.0.0
X-Rslight-Site: $2y$10$x.G2pfThlFWO07zu0oxyYuLsKRZz7pisB2dY9gm0c/mplDhFm3dfy
X-Rslight-Posting-User: ac58ceb75ea22753186dae54d967fed894c3dce8
Bytes: 2355
Lines: 23

On Thu, 18 Jul 2024 17:28:43 +0000, Michael S wrote:

> On Thu, 18 Jul 2024 17:06:52 -0000 (UTC)
> Thomas Koenig <tkoenig@netcologne.de> wrote:
>
> The idea I was thinking about was to pre-calculate R/Ea rather than
> Ea/R. And then to find fast algorithm for approximation of exp(-1/x) on
> the range of interest.

exp( -x ) = 1/exp( x )

I guess, in the past, when -x was at most 1 cycle, and FDIV was 20
cycles
and exp() was 150 cycles, you did not worry about creating an elementary
function to calculate 1/exp(x). With exp(x) now down in the 16 cycle
range, it might be different as 1/exp(x) would likely also take 16
cycles.

My 66000 exp(x) instruction has sign control on the operands so one can
calculate exp(-x) as:

       EXP    Rd,-Rx           // still 16 cycles

exp(1/x) is pretty much confined to the serial latency of FDIV and EXP.