🚀 go-pugleaf

In article <9363:Sep2521:41:1290@kramden.acf.nyu.edu> brnstnd@kramden.acf.nyu.edu (Dan Bernstein) writes:
>  quickpick
>  fast(define builtin_bitcount 0)
>    count = bitcounts[j & 65535] + bitcounts[(j >> 16) % 65535];
>  fast(define builtin_bitcount 1)
>    unsigned long i;
>    for(i = j;i;i >>= 1) count += (i & 1);
>  endquickpick

Ugh.  Are you then going to standardize *everything*?

(N.B.  The machines mentioned that had the bit-counting were, I suspect, the
CDC Cyber and Cray machines, which have a Pop Count instruction.)

The Elxsi has an instruction for passing a message (up to 836 bytes, I
think) to a "port", which is usually another process (yet need not be).  Are
you going to straddle your "language" with the semantics of this
instruction, simply because someone may find it useful?

Some of the highly-specialized graphics processors have instructions to
rotate three-dimensional polygons in memory; yet most of them have vastly to
slightly different semantics; are you going to straddle the "language" with
the semantics of *all* of them?

Now, all that said:  C has the #ifdef route, which is close to what you
want.  (The compiler won't automatically know that it can do the popcount,
for example, but it's easy enough to add conditional expressions to check
for which machine you're on.)  Combine this with inline functions and
assembly a la gcc, and it's hard to get any better *and* portable.

-- 
-----------------+
Sean Eric Fagan  | "Never knock on Death's door:  ring the bell and 
seanf@sco.COM    |   run away!  Death really hates that!"
uunet!sco!seanf  |     -- Dr. Mike Stratford (Matt Frewer, "Doctor, Doctor")
(408) 458-1422   | Any opinions expressed are my own, not my employers'.