NetNews Usenet Archive 1992 #26/NN_1992_26.iso/spool/comp/arch/10510

Internet Message Format | 1992-11-08 | 1.6 KB

Path: sparky!uunet!pmafire!news.dell.com!swrinde!zaphod.mps.ohio-state.edu!cs.utexas.edu!sun-barr!news2me.EBay.Sun.COM!exodus.Eng.Sun.COM!flayout.Eng.Sun.COM!tremblay
From: [e-mail] (Marc Tremblay)
Newsgroups: comp.arch
Subject: Re: Optimizing Old Code for i586
Date: 7 Nov 1992 18:11:14 GMT
Organization: Sun Microsystems, Mt. View, Ca.
Lines: 23
Message-ID: <[e-mail]>
References: <[e-mail]>
NNTP-Posting-Host: flayout

In article <[e-mail]> [e-mail] (Elan Feingold) writes:
>Write a program that takes a .exe file and performs the following analysis
>on it:
>    a) Break up the code into windows, bounded by branches.
>    b) Rearrange the code within the windows following the i586
>       idea of being able to execute two instructions in a cycle if
>       there are no dependancies.

This basic block scheduling is very limited since branches occur approximately
every 5 instructions (integer programs running on SPARC). It will buy you
some (around 10%) but much better performance can be achieved if the code
is recompiled with a compiler that can schedule instructions across basic
blocks, especially for integer programs. Key instructions to move "up"
in the code are loads, since they may have a long latency due to a cache
miss. Moving loads pass conditional branches introduces problems such
as possible additional page faults, segmentation error, etc. That's why
it is desirable to have some kind of support for speculative loads,
which brings the question:

Does the P5 have support for speculative loads?

- Marc Tremblay.
Sun Microsystems.