Re: Can optimization-specific codes exist ?

escargo@mirage.skypoint.com (David S Cargo)
4 Oct 1999 12:20:37 -0400

          From comp.compilers

Related articles
Can optimization-specific codes exist ? compusm@chollian.net (Chango Cho) (1999-09-28)
Re: Can optimization-specific codes exist ? gneuner@dyn.com (1999-10-03)
Re: Can optimization-specific codes exist ? adrian@dcs.rhbnc.ac.uk (1999-10-04)
Re: Can optimization-specific codes exist ? escargo@mirage.skypoint.com (1999-10-04)
Re: Can optimization-specific codes exist ? tgl@netcom.com (Tom Lane) (1999-10-06)
Re: Can optimization-specific codes exist ? adrian@dcs.rhbnc.ac.uk (1999-10-06)
| List of all articles for this month |

From: escargo@mirage.skypoint.com (David S Cargo)
Newsgroups: comp.compilers
Date: 4 Oct 1999 12:20:37 -0400
Organization: none
References: 99-09-118 99-10-011
Keywords: optimize, architecture

>>Can optimization-specific(only dedicated to optimization) codes exist?
>>this question is generally related with the problems of
>>'codification', I think. Environments that surround 'code' begin to
>>change dramatically. I want to think about 'code' itself - as a
>>turn-point.
>
>Digital's PDPs used to allow custom opcodes to be defined which could
>then be included in programs.


Back in my earlier days with Honeywell, we did an experiment of
putting part of a compiler into PDP microcode. It was a compiler for
the ATLAS language (Abbreviated Test Language for Avionics Systems, an
IEEE standard as I recall). The compiler we used was home grown and
very table driven. We were able to put a lot of the lookup code into
microcode, and then put assembly language calls to the new
instructions into the Fortran that made up the compiler; we were even
able to add calls to the compiler into a text editor (also written in
Fortran) so that people could write programs and then check the syntax
quickly. (The syntax check was for statements only, not whole
programs.)


The speedup from use of microcode was substantial (30 or 40 percent)
but not radical. Effectively the inner loop of the table lookup was
all in microcode, and the instruction fetch penalties were eliminated.
A system with a big instruction cache and instruction decoding might
have likewise shown measurable improvement.


David S. Cargo


Post a followup to this message

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