Re: min/max macros

Re: min/max macros

[Mike Stump]

> Date: Thu, 11 Dec 1997 12:03:45 -0800
> From: dave madden <dhm@paradigm.webvision.com>

> min/max macros are not inherently bad, as long as you understand
> that they evaluate their arguments more than once.

Spraying a crowd with bullets from an automatic rife isn't bad, it's
only bad if people get hit...


Bad.  Period.

Re: [EGCS] Re: min/max macros

[Marc Lehmann]

> > #define max(a,b)	(((a) > (b)) ? (a) : (b))
> > #define min(a,b)	(((a) < (b)) ? (a) : (b))
> 
> That is a big no-no: Absolutely no performance benefits with any decent
> compiler but nasty side effects.

so egcs isn't a decent compiler?

for example, the macro-max(2,3) will be evaluated at compile time,
while the function-max(2,3) won't...

this get's even worse on things like:

int mult(int a, int b) { return a*b; };

mult(a,3) will result in an "imul 3" on x86... definitely
slower than a lea, for example.

      -----==-                                              |
      ----==-- _                                            |
      ---==---(_)__  __ ____  __       Marc Lehmann       +--
      --==---/ / _ \/ // /\ \/ /       pcg@goof.com       |e|
      -=====/_/_//_/\_,_/ /_/\_\                          --+
    The choice of a GNU generation                        |
                                                          |

Re: [EGCS] Re: min/max macros

[Joe Buck]

> > > #define max(a,b)	(((a) > (b)) ? (a) : (b))
> > > #define min(a,b)	(((a) < (b)) ? (a) : (b))
> > 
> > That is a big no-no: Absolutely no performance benefits with any decent
> > compiler but nasty side effects.
> 
> so egcs isn't a decent compiler?
> 
> for example, the macro-max(2,3) will be evaluated at compile time,
> while the function-max(2,3) won't...

The function max(2,3) will indeed be evaluated at compile time, if it is
inlined.

> this get's even worse on things like:
> 
> int mult(int a, int b) { return a*b; };
> 
> mult(a,3) will result in an "imul 3" on x86... definitely
> slower than a lea, for example.

Same for mult: if it is inlined, it will be evaluated at compile time if
given constant arguments.

Re: min/max macros

[Marc Lehmann]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1475 bytes --]

> > so egcs isn't a decent compiler?
> > 
> > for example, the macro-max(2,3) will be evaluated at compile time,
> > while the function-max(2,3) won't...
> 
> The function max(2,3) will indeed be evaluated at compile time, if it is
> inlined.

sorry, I was just too fast here in replyÄing (and I left the [EGCS] in, soory again ;)

> > this get's even worse on things like:
> > 
> > int mult(int a, int b) { return a*b; };
> > 
> > mult(a,3) will result in an "imul 3" on x86... definitely
> > slower than a lea, for example.
> 
> Same for mult: if it is inlined, it will be evaluated at compile time if
> given constant arguments.

no, mult(a,3) results in:

        movl 4(%esp),%eax
        imull $3,%eax,%eax
        ret

which is highly suboptimal.. a mult(4,5), though, will be evaluated at compile
time. But macros *are* faster than inline functions, despite what the
documentation tells us.


  ---------------------------------------------------------------------

      for a pentium-optimizing gcc, look at http://www.gcc.ml.org/

      -----==-                                              |
      ----==-- _                                            |
      ---==---(_)__  __ ____  __       Marc Lehmann       +--
      --==---/ / _ \/ // /\ \/ /       pcg@goof.com       |e|
      -=====/_/_//_/\_,_/ /_/\_\                          --+
    The choice of a GNU generation                        |
                                                          |

min/max macros

[Bill Ahlbrandt]

I have noticed that these macros are not always available and not always in the same place.

Specifically, with egcs, they seem to be in curses.h

I "coded" my own and used them as follows:

#define max(a,b)	(((a) > (b)) ? (a) : (b))
#define min(a,b)	(((a) < (b)) ? (a) : (b))

After discovering that max in particular was not yielding the desired results, I coded these macros as functions.  All of my problems went away.

Is it generally a bad plan to use macros like this?  Are there any known problems with egcs involving macros such as these?

Thanks in advance..

Re: min/max macros

[dave madden]

At last, an egcs question that *I* can answer! :-)  (Unfortunately,
it's not really an egcs question, but a C/C++ question.  Anyway...)

 =>From: Bill Ahlbrandt <bahlbr@icdata.com>
 =>...
 =>I "coded" my own and used them as follows:
 =>
 =>#define max(a,b)	(((a) > (b)) ? (a) : (b))
 =>#define min(a,b)	(((a) < (b)) ? (a) : (b))
 =>
 =>Is it generally a bad plan to use macros like this?  Are there any
 =>known problems with egcs involving macros such as these? 

min/max macros are not inherently bad, as long as you understand that
they evaluate their arguments more than once.  For the simple case,
they work as you'd expect, but if you do something like:

		double x = min( sqrt(x), sqrt(y) );
		
		int		 a = 1;
		int		 b = 2;
		int		 c = max( a++, b++ );

you'll be inefficient (first example) or undefined (second example).

Min/max are better handled by inline functions; in c++ you can even
create a function template so that you can have min/max defined for
any type:

    template <class T> const T &
    min( const T &a, const T &b )
    {
        return (a < b) ? a : b;
    }

d.

Re: min/max macros

[Michael A. Benzinger]

At 11:14 AM 12/11/97 -0600, you wrote:
>#define max(a,b)	(((a) > (b)) ? (a) : (b))
>#define min(a,b)	(((a) < (b)) ? (a) : (b))
>
>After discovering that max in particular was not yielding the desired
results, I coded these macros as functions.  All of my problems went away.
>
>Is it generally a bad plan to use macros like this?  Are there any known
problems with egcs involving macros such as these?

Bill,

The functions are better than the macros for this very reason:

  int x=2;
  int y=1;

  int z = max(x++,y);

Your intentions are to have 'z' equal to '2' and 'x' equal to '3'.  However,
'x' is actually equal to '4'.  Why?  Because the macro code hides a nasty
gotcha.  The code when expanded looks like this:

  int z = (((x++) > (y)) ? (x++) : (y));

You will note that 'x++' is actuall evaluated twice.  With a function of
the Standard Library function 'max', this will not happen.  The argument
is only evaluated and incremented once, as you would expect.

Mike Benzinger
-------------------------------------------------------------
Michael A. Benzinger
Principal

SABRE Technology Solutions        Phone:  +1 817-264-6820
1 E. Kirkwood Blvd.               Fax:    +1 817-264-3504
Southlake, TX  76092              e-mail: mbenz@sabre.com
U.S.A.                                    bzinger@iName.com

Re: min/max macros

[Gerald Pfeifer]

On Thu, 11 Dec 1997, Bill Ahlbrandt wrote:
> I "coded" my own and used them as follows:
> 
> #define max(a,b)	(((a) > (b)) ? (a) : (b))
> #define min(a,b)	(((a) < (b)) ? (a) : (b))

That is a big no-no: Absolutely no performance benefits with any decent
compiler but nasty side effects.

> After discovering that max in particular was not yielding the desired
> results [...]

You didn't happen to use "max(i++,prev_max)" or similar, did you?

For that expands to 
  (((i++) > (prev_max)) ? (i++) : (prev_max))
which may increment i twice as an -- usually unintended -- side-effect.

Hope this helps,
Gerald
-- 
Gerald Pfeifer (Jerry)      Vienna University of Technology
pfeifer@dbai.tuwien.ac.at   http://www.dbai.tuwien.ac.at/~pfeifer/

Re: min/max macros

[Paul Koning]

>>>>> "Bill" == Bill Ahlbrandt <bahlbr@icdata.com> writes:

 Bill> I have noticed that these macros are not always available and
 Bill> not always in the same place.  Specifically, with egcs, they
 Bill> seem to be in curses.h

 Bill> I "coded" my own and used them as follows:

 Bill> #define max(a,b) (((a) > (b)) ? (a) : (b)) #define min(a,b)
 Bill> (((a) < (b)) ? (a) : (b))

 Bill> After discovering that max in particular was not yielding the
 Bill> desired results, I coded these macros as functions.  All of my
 Bill> problems went away.

It's always dangerous to create macros that use a macro argument more
than once.  You will get the wrong results if the macro is called with
an actual argument that has side effects (e.g., max (i++, j++) ).

Apart from that, functions have the benefit of type checking.

For these reasons, I tend to use functions when I can.  You can
declare them "static inline" to get the benefits without any
performance penalties.

	paul