structs in c causing big size diffrences.

[frea]

Hi!
I can't understand one thing. I've been writing a small program, and at one point i needed to use a struct more. Merely by adding it, and doing a single instruction using it the size jumped by 1.5kb uncompressed. I disassembled 2 versions, one with this single instruction one without. the asm code differed very slightly. ( ~12 lines in ~2k lines were different )

So, if the code doesn't differ, why am i getting such a huge change?

Code: [Select]

typedef struct Primitive_
{



unsigned int type;



float pos[ 3 ], col[ 3 ];



float rotx, roty;



union



{





float sphere;





float cube[ 3 ];



};
} Primitive;


Code: [Select]

Primitive prim[ PRIMITIVENUM ];
(...)
sub( vector1, prim[ id ].pos, vector2 ); //this line adds 1.5kb, sub is also called previously on different vectors
Stripping the contents of Primitive doesn't reduce size.

[taj]

How big is PRIMITIVENUM?
Say around 50?

Thats your 1.5k there.

--
DBF debugging service.

[frea]

#define PRIMITIVENUM 2

changing to 10000 makes no size difference. ( all arrays are in .bss section )

edit: mistake, it causes diffrence.
But that what is surpricing that the difference is between PRIMITIVEMNUM beeing 2 and 3. What the hell? ( even if no code uses the array! )

edit2: it is even more werid. For purpouses of finding why i have such problem i have commented out the rest of the function. I tried compiling with -Os, and the size dropped greatly, now there is no jump on primnum 2 to 3. That what is interesting that by uncommenting the rest of function i get ~1.5kb again but, with -O2 i don't get 2x 1.5 kb but only 1x1.5kb. The uncommented part gives sixty-something byte diffrence in file size if used with -O2.

So the file size doesn't differ to much with -O2/-Os, but it
has a werid growth when (un)commenting .


edit3: don't ask me . I've switched computers and now no werid behaviour. At home everything's ok .

[Jim]

Maybe depending on size the compiler is bringing in a library routine like memcpy or mul to do some internal sums instead of doing them inline?  But it sounds like your problem has gone away for now
Jim

[frea]

It seems that the compiler at labs is somewhat broken :p. I comment ~30 function calls and get 20bytes less uncompressed, although it ereases ~100 lines of asm.