Re: Q: unboxed values and polymorphism

[fjh@mundook.cs.mu.OZ.AU (Fergus Henderson)]

dewar@cs.nyu.edu (Robert Dewar) writes:

>Hannes said
>
>"Let's assume your fixnums have the low bits `00' as tags. To add
>two fixnums, you would extract them from the word, perform the
>usual addition and put the result in a variant record? I'd just
>add the words."
>
>So what, you would get the same code!

That's presuming the compiler is smart enough, which is a very big IF.
Have you tried this with GNAT?

I tried a simple example in C, and found that GNU C on an Alpha
generated worse code using bitfields than using the approach
suggested by Hannes above; in particular, it does not avoid the
unnecessary shifts on the operands and result of the addition.
I doubt if GNAT would do any better.

(The test case I used is available on request, should anyone wish
to duplicate the results of my experimentation.)

--
Fergus Henderson <fjh@cs.mu.oz.au>   |  "I have always known that the pursuit
WWW: <http://www.cs.mu.oz.au/~fjh>   |  of excellence is a lethal habit"
PGP: finger fjh@128.250.37.3         |     -- the last words of T. S. Garp.