Conference turris::languages

    It depends on what you're doing.  The general impression around
    here is that VAX FORTRAN does the overall best job for real programs,
    but is missing stuff like inline expansion that can make a real
    difference for some programs.
    					Steve
    
    P.S.  If this has been discussed before, I'll be happy to delete
    this note.
   

re: -1,
    
>    P.S.  If this has been discussed before, I'll be happy to delete
>    this note.
   
	Deleting is really where it's at these days, eh Steve?  I know,
    why don't we all just delete all of our notes, and then start all over
    again!  Anyway, if this note is going to get deleted, then I sure as heck
    want to get in on it too!  :-)

	I was going to suggest an addition to the *Life* program which would
    have a system call to obtain the current time at the beginning and then
    again at precisely the momment that the tenth generation is updated on
    the screen.  If a set data pattern was used for consistency, it might
    prove to be a simple method of testing the speed of all the programming
    languages at run time (although alot of them have the advantage of a
    more efficient algorithm - but then who said life was fair?).

	The following pattern is simple enough, but has some interesting
    patterns which develop.  I call it the plus sign:

				X
			       XXX
			        X

							-davo

    Re .3:
       Sorry, I don't understand your comment about deleting notes.
    I only meant I'd delete MY reply (note) and move it to the other
    discussion, if there was one.  I am not a moderator of this 
    conference and have no authority to delete anyone else's notes.
    
       I'd also ask what relevance your further comments on LIFE have
    to this subject.
    					Steve

re: .4,

	You missed it, I was just refering to the recent trend of note
    deletion (both by authors and by moderators).  I was also trying to
    inject a little humor into the current state of affairs, but I keep
    forgeting that humor isn't in vogue these days.

>       I'd also ask what relevance your further comments on LIFE have
>    to this subject.

	As to whether my mention of the adding a stop-watch feature to the
    various Life programs to test the relative speed of code, I guess you're
    right: this has nothing to do with the "Fastest Code" topic of this
    discussion.  In light of this, we should just delete all of our notes
    and go home tonight knowing that we are truly relavant to our topics.

							-davo

    p.s.  Oh, I almost forgot...  --->   :-)    

    And now back to our regularly scheduled topic...
    
    Let me clarify myself a little.  What I am interested in is determining
    which Vax language compiler will generate the fastest (least CPU
    time) executable code from similar source code logic.
    
    I have found in the past that accurate results require that there
    should be as little I/O between timing points as possible.  Also
    it is important that comparisions between languges be based on the
    same algorithims, not just the same functionality.  After all even
    DCL can eventually quicksort faster than the best macro code can
    bubble sort.
    
    By the way, I'm not really interested in Macro, or language examples
    that use embedded macro, after all we all know that NOTHING can
    be faster than the best macro routines.  I want to know which languages
    can take simple straightforward source code and produce the fastest
    machine instructions.
    
    And SMILE out there!   B^)
    
    -- Barry

    If you're asking about choosing the optimal set of instructions,
    operands, etc., and aren't as concerned with techniques such
    as in-line expansion, inter-procedural optimization, etc. - I'd
    say VAX FORTRAN is the best.  But, in the Technical Languages world,
    we have four major code generators we use - VAX FORTRAN, VAX Pascal,
    BLISS and VCG (PL/I, Ada, C).  All are very good, but do different
    things well.
    
    As I think I've said in this conference before, though (now that
    I remember it), "best optimization" is the LAST thing you should
    consider when choosing a programming language among all of our
    excellent VAX compilers. All are close enough that the language
    features and other aspects make far more difference in the
    resulting application than does optimization.  Also, our compilers
    are evolving (except, unfortunately, for BLISS), and keep getting
    better each release.
    					Steve

    Re .7:

    	a)  Everything counts.  Code selection, register allocation, 
    	code motion, expansion, loop unrolling, etc...

    	b)  You say that different languages do different things well.
    	Could you expand on that some?  (This is the kind of thing that I
    	am looking for).

    	c)  I am not using any of this to choose a programming language.
    	Its just for personal interest/knowledge.  Besides, we software
    	specialists rarely have the luxury of choosing our own programming
    	language anyway.

    	-- Barry  B^)}
    

    The best ode generator I have come across to date is the one for
    CLU ... it does some incredible things to save time/space.  As far
    as I am aware, there isn't a version for VMS though.  It's worth
    pointing out that some of the optimisations can only be made because
    the language is sufficiently well-typed to allow the compiler to
    make wide-sweeping assumtions.  For example, 'string$substr' doesn't
    make a copy of the substring because it *knows* that strings cannot
    be changed.
    
    I forget who it was that said no compiler produces code as good
    as the best macro routines.  I disagree.  High level languages allow
    you to handle abstractions which are just too unwieldy in macro:
    huge programs written in a high-level languages (I exclude C and
    BLISS et al) are usually smaller and faster than their peers written
    in macro.  Don't get me wrong: a judicious (very small) sprinkling
    of assembler can be a good thing if the language is capable of using,
    for example, the 'extv' instruction.
    
    							jch