sgi cc and gcc

[David Bourguignon <David dot Bourguignon at imag dot fr>]

Hi all,

I've got strange performance results on two different platforms: one with sgi cc
and a mips processor and one with gnu gcc and an intel processor.
I was looking for the difference between two alternative constructs (see code
below). My benchmark consisted in calling *_add() many times.
Here is the time elapsed for each construct:

Vec2d
sgi cc:  15.018956
gcc:     46.187340

Vec2ds
sgi cc:  47.569683
gcc:     22.387057

I knew that sgi cc was unable to perform ISO C 1999 inline optimization, so the
difference between the two compilers is understandable for Vec2ds. What is more
surprising is the difference for Vec2d: it seems that gcc don't know how to
optimize functions taking arrays in parameters.

Do you have a better explanation ?
(If you are interested I can send you the code for the entire benchmark).

Thank you in advance,

	David.


/******************************************************************************/

#ifdef __sgi
#define __inline__ static __inline
#else
#define __inline__ static inline
#endif

typedef double Vec2d[4];
typedef const double const_Vec2d[4];
typedef struct _Vec2ds Vec2ds;

struct _Vec2ds {
  double x, y;
};

__inline__ void vec2d_add(Vec2d v, const_Vec2d a, const_Vec2d b) {
  v[0] = a[0] + b[0];
  v[1] = a[1] + b[1];
}
__inline__ Vec2ds *
vec2ds_add(Vec2ds *v, const Vec2ds *a, const Vec2ds *b) {
  v->x = a->x + b->x;
  v->y = a->y + b->y;
  return (v);
}

/******************************************************************************/

-- 
iMAGIS project, GRAVIR lab
INRIA Rhône-Alpes, 655 avenue de l'Europe
38330 Montbonnot Saint Martin, France
http://www-imagis.imag.fr/~David.Bourguignon/index.gb.html