Hey! inline function can do it for float and double or anything with template, and don''t use C-casts! use reinterpret_cast

The C++ integrist

(just kidding)

Doesn''t that require an extra copy?
&x->p

baskuenen

Obfuscate: tr.v. -cated, -cating, -cates. 1. a. To render obscure. b. To darken.
2. To confuse: his emotions obfuscated his judgment.
[LLat. obfuscare, to darken : ob(intensive) + Lat. fuscare,
to darken < fuscus, dark.] -obfuscation n. obfuscatory adj


I don''t know if you''ve checked the site. You should check it... Their laws are mainly:

1. To write the most Obscure/Obfuscated C program under the rules below.
   
2. To show the importance of programming style, in an ironic way.
   
3. To stress C compilers with unusual code.
   
4. To illustrate some of the subtleties of the C language.
   
5. To provide a safe forum for poor C code. :-)
   

So you may see that it is very fun... Some examples are even artisticly interesting

Thanks for the parenthesis. I don''t know if the compiler will accept them though, because (8+4) have no addresses... And by this same fact you cannot send constants or expressions... But here is a way (not tested though) in C++:

    // Still it is NOT LISP...  Please.#define fabs(x) (*((float*)(*((long*)(&((*(new float))=(x)))) &= 0x7FFFFFFF)))             

I love this fabs defines
Maybe a bit too many parenthesis... I just don''t want to mess with operator priority.
Now this should work, but I don''t know if the compiler will yield or collect the garbage for this...
WinNT will though.

They try to prevent death, I try to promote life.

Magmai Kai Holmlor: You mean typecasting to a pointer and back to a value again? I think a good compiler will optimize this.

Poltras: Ah - Obfuscate is their name. Now I get it. But what is their URL? Could it be I missed it?

You can''t use reinterpret cast:
''reinterpret_cast'' : cannot convert from ''float'' to ''int''

...
the idea was to make the fabs fast... not bloated!

quote:
float x= -10.5; int *p = (int *)&x *p &= 0x7FFFFFFF;

this *does* dereference a pointer, *p, p contains the address of x, not the value of x
All those defines use this method so they dereference a pointer as well...

quote:
#define fabs(x) (*((float*)(*((long*)(&((*(new float))=(x)))) &= 0x7FFFFFFF)))

Good God, your leaking!

    //C++''ine waytemplate <class T>T fabs(T)	{	return((x<0)?-x:x);	}    

quote: Original post by Magmai Kai Holmlor

Poltras said:

#define fabs(x) (*((float*)(*((long*)(&((*(new float))=(x)))) &= 0x7FFFFFFF)))


Good God, your leaking!

NO!!! I'm making fun... that's different.

quote:

            //C++'ine waytemplate <class T>T fabs(T)	{	return((x<0)?-x:x);	}            

And what for ANSI C??? Worse, you were talking about speed, and my solution is more speed-efficient (Anyway, who will do 30M fabs in a second?).
Your solution is not "look-what-I-did-it's-cool" enough

Just kidding.

I don't want to participate into flame wars, I just want to start them.

Edited by - Poltras on November 2, 2000 11:57:46 AM

quote: Original post by Poltras

            // This changes to absolute the value of the parameter// and return the value.#define fabs(x) (*((float*)(*((long*)&x) &= 0x7FFFFFFF)))        [/source] <hr height=1 noshade></BLOCKQUOTE></font> Your second version leaks a float every time you call it; you never match the new w/ a delete.Secondly, the above will always access potentially invalid memory addresses, as you cast a long to a float* (Bad!)Here''s a better way to do it:[source]struct fabs_func{public:    fabs_func( float f ): t(f)    {        *reinterpret_cast<long*>(&t)&=0x7fffffff;    }    operator float() const    {        return t;    }    float t;};    

Note that this will not work for doubles. Although it doesn''t take much to hack this for doubles.

Getting back to the original question, though, this is a pointless optimization. fabs takes 3 clocks on a 486 and 1 on a P5 and above. FP multiplication is much faster than integer multiplication. All these optimizations are pretty damn stupid.

MSN

quote: Original post by msn12b

Original post by Poltras

                // This changes to absolute the value of the parameter// and return the value.#define fabs(x) (*((float*)(*((long*)&x) &= 0x7FFFFFFF)))        [/source] <hr height=1 noshade></BLOCKQUOTE></font>     

Your second version leaks a float every time you call it; you never match the new w/ a delete.

I know the second version will memleak 4 bytes everytimes we call it, and I warned it...

quote:
Secondly, the above will always access potentially invalid memory addresses, as you cast a long to a float* (Bad!)

Note that this will not work for doubles. Although it doesn''t take much to hack this for doubles.

mmmmh?? No... wait... you''re right! Damn!

Is there a solution? Oh yeah:

// Just for fun again (I know this is crap, but...)
#define fabs(x) (*((float*)&((*((long*)&(x)) &= 0x7FFFFFFF))))
<br><br>Tadaaa!  It works!  <img src="wink.gif" width=15 height=15 align=middle><br><br><br><BLOCKQUOTE><font size=1>quote:<hr height=1 noshade><br>Getting back to the original question, though, this is a pointless optimization. fabs takes 3 clocks on a 486 and 1 on a P5 and above. FP multiplication is much faster than integer multiplication. All these optimizations are pretty damn stupid.<br> <hr height=1 noshade></BLOCKQUOTE></font> <br>Like I was saying, I''m only doing it for fun...  and fabs takes a bit more time...  it is not a single ASM instruction since you need to store and restore the value into registry.<br>    </pre> 

quote: Original post by Poltras
Like I was saying, I''m only doing it for fun... and fabs takes a bit more time... it is not a single ASM instruction since you need to store and restore the value into registry.

Hello? Are you joking or are you actually this stupid? You''re doing a FP compare immediately following the fabs()... even if you''re doing a simple fabs on a float, it is still faster to load and save from the FPU; the CPU can pair integer instructions w/ the FP instructions.

MSN

Raise your hands if you''re tired of msn''s crappy attitude.
{raises hand}