compiler optimize issue

[Cmdr0Sunburn]

im using MSVC++6
when i do a register int something;
it never makes it a register, even when no functions are being called so it doesnt need to puch/pop to save it.( it would be clearly faster, why doesnt the compiler see that?) i also code assembly , is there any way i could force it to make it a register?

[Zaei]

The register modifier is only a hint for the compiler, but I believe that MSVC6 ignores it totally.

Z.

[Cmdr0Sunburn]

what about VC7?, oh man why would they even bother to put that statment in this c++ book is it doesnt use the register.

[parksie]

The register keyword has always just been a hint. Most of the time, if you get your optimisation options right, the compiler can do a better job than the developer, especially if you have a properly targeted compiler (such as Intel's).

If you're desperate for the speed you can always whack in some assembler

[transcendental]

Originally posted by Cmdr0Sunburn
what about VC7?, oh man why would they even bother to put that statment in this c++ book is it doesnt use the register.

This is because x86 platform has too few registers. It cannot afford to let you have them. And VC can optimise the program better if the programmer let it have all the registers (which are already too few).

Maybe on other platforms perhaps, the C++ compiler does allow 'register' keyword?

[jim mcnamara]

Things you can do to allow optimization to happen the most efficiently:

Use as many local variables as possible. Avoid globals
Keep all variable lengths a power of two:
ie., avoid

char b[3];

because it DOES NOT save memory. The compiler will try to word- align everthing when you do this.

The best possible form of a loop for optimization is:

int a=1;
while (a){
    // do something
    // check for exit by a=(b==c) or something like that
}