From: Ullrich von Bassewitz (uz_at_musoftware.de)
Date: 2001-03-19 22:40:05
On Mon, Mar 19, 2001 at 02:45:05PM -0600, Mike McCarty wrote:
> Certainly it can. The gnu C compiler does this automatically as an
> optimization even for machines which have a reasonable (read 16 bit)
> stack pointer. If a routine is declared "static" then the compiler
> knows, and can check for aliased pointers.
For one, the GNU C compiler has about one hundred times the source code size
of cc65. I would expect it to do things like this.
Second, this can only be done for routines declared static, which limits the
scope considerably.
And third, if there are external functions and calls to external functions in
this module (which is true for most C modules) the scope of such optimizations
is even more limited, since recursive calls may happen indirectly through
these external functions and the compiler does not know about it.
So I doubt that these optimizations do really save a considerable amount of
pushes and pops.
> Especially it can know if
> it is told by the programmer by...
[...]
> using an ISO C Standard compliant #pragma.
Do you really think it is realistic to expect programmers to tell the compiler
for each function if it is called recursively? Most people are not even
declaring local functions and variables as static. And what happens, if the
code gets changed but not the #pragma?
> Ultimately: You support the compiler. You don't like that solution, so
> it won't exist.
That's wrong in two places.
Everyone is welcome to send me code improvements, and a lot of people have
done so. So it's not just me who is supporting the compiler. It is your
decision to use it as is, or help improving it.
And if your solution is better than the stuff that exists now, and you can
show that with hard facts (that is: a real implementation) and not just words,
I would be happy to add your code to the compiler. If you send me code that
has no unnecessary runtime overhead and does improve the optimizer so that
lots of stack operations could be removed I would be crazy not to use it. And
I may be stubborn, but I'm not crazy:-)
Regards
Uz
P.S.: This discussion has kept me from adding the stack checking code which I
was working on, so I will try to shut up until this is finished:-)
--
Ullrich von Bassewitz uz_at_musoftware.de
----------------------------------------------------------------------
To unsubscribe from the list send mail to majordomo_at_musoftware.de with
the string "unsubscribe cc65" in the body(!) of the mail.
This archive was generated by hypermail 2.1.3 : 2001-12-14 22:05:39 CET