1/* Compute the sum of the squares of a vector of signed shorts
2
3 *  MMX-assisted version (also used on SSE)
4
5 * The SSE2 and MMX assist routines both operate on multiples of
6 * 8 words; they differ only in their alignment requirements (8 bytes
7 * for MMX, 16 bytes for SSE2)
8
9 * Copyright 2004 Phil Karn, KA9Q
10 * May be used under the terms of the GNU Lesser Public License (LGPL)
11 */
12
13long long sumsq_mmx_assist(signed short *,int);
14
15long long sumsq_mmx(signed short *in,int cnt){
16  long long sum = 0;
17
18  /* Handle stuff before the next 8-byte boundary */
19  while(((int)in & 7) != 0 && cnt != 0){
20    sum += (long)in[0] * in[0];
21    in++;
22    cnt--;
23  }
24  sum += sumsq_mmx_assist(in,cnt);
25  in += cnt & ~7;
26  cnt &= 7;
27
28  /* Handle up to 7 words at end */
29  while(cnt != 0){
30    sum += (long)in[0] * in[0];
31    in++;
32    cnt--;
33  }
34  return sum;
35}
36