Re: sparc object hackery and Bliss techniques?

tfj@apusapus.demon.co.uk (Trevor Jenkins) writes:
>One of the optimisation that only appears to be used by the Bliss-11
>compiler is to determine whether code should be placed before or after the
>loop out which it was being moved. They, Wulf et al, described these as
>alpha- and beta- code motion; as to which one was which I don't remember I
>still awaiting the republication of the book.


Actually they are called alpha-motion and omega-motion. You can figure
out which moves out the top of a language construct and which out the
bottom based on the position of alpha and omega in the Greek alphabet.
Also, these motions were usually out of the top and bottom of if and
select expressions rather than loops. Motions called chi- and rho-motion
were usually done on loop expressions.


>A paper, by Wulf and others, was published in the Comm of the ACM around
>the same time. (Again from memory) this described the rationale of the
>language and its design. One of the goals was that all flow-graphs were
>reducible; this was achieved but not having an true GOTO statement, though
>there were LEAVE and EXITLOOP statements so that you could terminate a
>loop in a very controlled fashion.


The CACM article predates the BLISS-11 "green" book by several years. It
is in CACM 14, 12 (December 1971), written by Wulf, Russell and Haberman
and entitled "BLISS: A Language for Systems Programming." It describes
the BLISS-10 language and not BLISS-11. (Among other things, BLISS-10 was
originally missing the LEAVE statement).


>Variants of the language were defined by Digital, who sponsored Wulf et al
>to work on it(?), Bliss-10, Bliss-11, Bliss-36, Bliss-32 and Common Bliss.
>...; the
>only acknowledged piece of software written in Common Bliss was RUNOFF.


There were several other Digital products written in Common Bliss. An
obvious one is the Common BLISS front end which I am certain that Digital
has acknowledged. Other products are EDT and CMS.
--

Return to the comp.compilers page.
Search the comp.compilers archives again.