gcc/testsuite/gcc.dg/vmx/newton-1.c

   1 /* { dg-do compile } */
   2 #include <altivec.h>
   3
   4 #define SPLAT76 ((vector unsigned char)\
   5                  {0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3})
   6 #define SPLAT54 ((vector unsigned char)\
   7                  {4,5,6,7,4,5,6,7,4,5,6,7,4,5,6,7})
   8 #define SPLAT32 ((vector unsigned char)\
   9                  {8,9,10,11,8,9,10,11,8,9,10,11,8,9,10,11})
  10 #define SPLAT10 ((vector unsigned char)\
  11                  {12,13,14,15,12,13,14,15,12,13,14,15,12,13,14,15})
  12 #define INTERLEAVE ((vector unsigned char)\
  13                     {0,1,16,17,4,5,20,21,8,9,24,25,12,13,28,29})
  14
  15 long real_32_manytaps (long ntaps, vector signed short *c, long ndat,
  16                        vector signed short *x, vector signed short *y)
  17 {
  18   long i, j, op, ndatavec, ncoefvec;
  19   vector signed short x0, x1;
  20   vector signed short coef;
  21   vector signed short cr10, cr32, cr54, cr76;
  22   vector signed int y_even, y_odd;
  23   vector signed short *x1p;
  24
  25   op = 0;
  26   ndatavec = ndat >> 3;
  27   ncoefvec = ntaps >> 3;
  28
  29   for (i = 0; i < ndatavec; i += 1) {
  30     x0 = x[i];
  31
  32     y_even = ((vector signed int){0x8000,0x8000,0x8000,0x8000});
  33     y_odd = ((vector signed int){0x8000,0x8000,0x8000,0x8000});
  34
  35     j = 0;
  36     x1p = x + 1 + i;
  37
  38     do {
  39
  40       coef = c[j];
  41       x1 = x1p[j];
  42
  43       cr10 = vec_perm(coef, coef, SPLAT10);
  44       y_odd = vec_msums(cr10, x1, y_odd);
  45       y_even = vec_msums(cr10, vec_sld(x0, x1, 14), y_even);
  46
  47       cr32 = vec_perm(coef, coef, SPLAT32);
  48       y_odd = vec_msums(cr32, vec_sld(x0, x1, 12), y_odd);
  49       y_even = vec_msums(cr32, vec_sld(x0, x1, 10), y_even);
  50
  51       cr54 = vec_perm(coef, coef, SPLAT54);
  52       y_odd = vec_msums(cr54, vec_sld(x0, x1, 8), y_odd);
  53       y_even = vec_msums(cr54, vec_sld(x0, x1, 6), y_even);
  54
  55       cr76 = vec_perm(coef, coef, SPLAT76);
  56       y_odd = vec_msums(cr76, vec_sld(x0, x1, 4), y_odd);
  57       y_even = vec_msums(cr76, vec_sld(x0, x1, 2), y_even);
  58
  59       x0 = x1;
  60
  61     } while (++j < ncoefvec);
  62     y[op++] = (vector signed short) vec_perm(y_even, y_odd, INTERLEAVE);
  63
  64   }
  65
  66   return op*8;
  67 }