Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!crdgw1!uunet!comp.vuw.ac.nz!waikato.ac.nz!ldo
From: ldo@waikato.ac.nz (Lawrence D'Oliveiro, Waikato University)
Newsgroups: comp.arch
Subject: Bitfield instructions--a good idea?
Message-ID: <1991Apr15.193425.3436@waikato.ac.nz>
Date: 15 Apr 91 07:34:25 GMT
Organization: University of Waikato, Hamilton, New Zealand
Lines: 29

What do RISC designers think of bitfield instructions? (A la
the VAX and 680x0 [x >= 2] processors: extract so many bits at
a certain offset, insert so many bits at a certain offset, etc.)

I was doing some work on some GIF image decompression code on a
68020 processor (a Macintosh LC)--a lot of bit manipulation involved.
I looked up the instruction timings for the bitfield instructions in
my 68020 handbook, and was mildly dismayed to find that they took about
twice as long to execute as a shift or mask instruction!

But I tried using them anyway. Motorola's bit-numbering scheme
in these instructions was the opposite of the natural order for
solving my problem, so I first had to reverse the order of each
bufferful of bytes after reading them in, before I could start
extracting the compressed data symbols.

The result? I sped up the code by more than 20%. I know the
compiler I was using generated pretty lousy code, but it allowed
the insertion of in-line machine code, which was how I was
fine-tuning the time-consuming parts.

Do other people find these instructions useful?

Lawrence D'Oliveiro                       fone: +64-71-562-889
Computer Services Dept                     fax: +64-71-384-066
University of Waikato            electric mail: ldo@waikato.ac.nz
Hamilton, New Zealand    37^ 47' 26" S, 175^ 19' 7" E, GMT+12:00
To someone with a hammer and a screwdriver, every problem looks
like a nail with threads.