[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]


Re your conjecture:
    Date: 30 January 1981 16:32-EST
    From: Glenn S. Burke <GSB at MIT-ML>
    Subject: optimizers
    .  .  .
    For one, it is a complete loss that to utilize such a feature
    one must use multiple values, ESPECIALLY in Maclisp.  I might feel
    differently if the support were small and efficient, but .  .  .
I don't know if there is a polite way to say this, but you seem to 
have lost all perspective on this facility:
    (VALUES x y)
Compiles into code that is exactly 2 (simple) instructions executions
longer than merely setq'ing X and Y 
    (MULTIPLE-VALUE (X Y) (<m-v-function> ...))
Compiles into code that is exactly 3 (simple) instructions executions
longer than merely calling the function and setq'ing X and Y, except 
when there are too few return values, in which case an ** autoloadable ** 
function would be called.  (Even the out-of-core file needn't be loaded if 
one supplies his own "patch-over" function here).  (Checking for failure 
to return "multiple-values" could be done, at a cost of about another 
2 instructions, but this check is debatable.)

If the top cell of an arg to "Filter/optimizer" is to be copied, 
then surely that copying itself will take more time than those few
instructions.  If one has 50 (!) "Filter/optimizer" functions in his
system, *** and if it could be proven that the two-value return is indeed 
two words longer than any other alternative **,  then he will lose all of 
100 words out of his address space.  Really big deal.

Using VALUES and MULTIPLE-VALUE doesn't require one to use VALUES-LIST
or MULTIPLE-VALUE-LIST;  but even if one did use them, I don't think
a MacLISP implementation of them could be made significantly smaller
and more efficient.  Also, the macro support lives in the MLMAC file,
which isn't generally loaded in a user's system;  with all that space
reclaimed in the compiler recently, how could one carp about a couple
hundred words of new macros?  especially for such a winning idea
as multiple values.