summing a vector using SIMD

["Jack Andrews" <effbiae@xxxxxxxxx>]

hi,

the following program doesn't give me a speed-up for using SIMD.  the
program sums a vector of integers.  if i take out the line 'q=p', the
SIMD version runs faster, but is not correct.  i am on fedora core 5
running gcc:

 gcc version 4.1.0 20060304 (Red Hat 4.1.0-3)

thanks in advance,

jack

$ gcc -O3 -mmmx -msse v.c

$ time ./a.out 100000 100 0
SISD
0

real    0m0.174s
user    0m0.172s
sys     0m0.004s

$ time ./a.out 100000 100 1
SIMD
0

real    0m1.017s
user    0m1.016s
sys     0m0.000s


$ cat v.c
#include <string.h>
#include <stdio.h>
#include <stdlib.h>

#define IZ sizeof(I)
typedef long I;
typedef char C;
typedef long v4si __attribute__ ((vector_size(IZ * 4)));

simd(n, is)
I *is;
{
   v4si p, q, r;
   I *pp = (I *) & p, *qq = (I *) & q, *rr = (I *) & r, i = 0;
   bzero(qq, 4 * IZ);
   for (i = 0; i < n; i += 4) {
       memcpy(rr, is + i, 4 * IZ);
       p = q + r;
       q = p;
   }
   return qq[0]+qq[1]+qq[2]+qq[3];
}

sisd(n, is)
I *is;
{
   I i = 0, j = 0;
   for (i = 0; i < n; i++)
       j += is[i];
   return j;
}

main(c, v)
C **v;
{
   I n = atol(v[1]),
       z = atol(v[2]), m = atol(v[3]), *is = malloc(IZ * (z *= 4));
   I result;
   printf(m?"SIMD\n":"SISD\n");
   while (n--)
       result=m?simd(z, is):sisd(z, is);
   printf("%d\n",result);
}