ceph/src/isa-l/erasure_code/ppc64le/gf_vect_dot_prod_vsx.c

   1 #include "ec_base_vsx.h"
   2
   3 void gf_vect_dot_prod_vsx(int len, int vlen, unsigned char *gftbls,
   4                           unsigned char **src, unsigned char *dest)
   5 {
   6         unsigned char *s, *t0;
   7         vector unsigned char vX1, vY1;
   8         vector unsigned char vX2, vY2;
   9         vector unsigned char vX3, vY3;
  10         vector unsigned char vX4, vY4;
  11         vector unsigned char vX5, vY5;
  12         vector unsigned char vX6, vY6;
  13         vector unsigned char vX7, vY7;
  14         vector unsigned char vX8, vY8;
  15         vector unsigned char vhi0, vlo0;
  16         int i, j, head;
  17
  18         if (vlen < 128) {
  19                 gf_vect_mul_vsx(len, &gftbls[0 * 32 * vlen], src[0], (unsigned char *)dest);
  20
  21                 for (j = 1; j < vlen; j++) {
  22                         gf_vect_mad_vsx(len, vlen, j, gftbls, src[j], dest);
  23                 }
  24                 return;
  25         }
  26
  27         t0 = (unsigned char *)dest;
  28
  29         head = len % 128;
  30         if (head != 0) {
  31                 gf_vect_dot_prod_base(head, vlen, &gftbls[0 * 32 * vlen], src, t0);
  32         }
  33
  34         for (i = head; i < len - 127; i += 128) {
  35                 vY1 = vY1 ^ vY1;
  36                 vY2 = vY2 ^ vY2;
  37                 vY3 = vY3 ^ vY3;
  38                 vY4 = vY4 ^ vY4;
  39
  40                 vY5 = vY5 ^ vY5;
  41                 vY6 = vY6 ^ vY6;
  42                 vY7 = vY7 ^ vY7;
  43                 vY8 = vY8 ^ vY8;
  44
  45                 unsigned char *g0 = &gftbls[0 * 32 * vlen];
  46
  47                 for (j = 0; j < vlen; j++) {
  48                         s = (unsigned char *)src[j];
  49                         vX1 = vec_xl(0, s + i);
  50                         vX2 = vec_xl(16, s + i);
  51                         vX3 = vec_xl(32, s + i);
  52                         vX4 = vec_xl(48, s + i);
  53
  54                         vlo0 = EC_vec_xl(0, g0);
  55                         vhi0 = EC_vec_xl(16, g0);
  56
  57                         vX5 = vec_xl(64, s + i);
  58                         vX6 = vec_xl(80, s + i);
  59                         vX7 = vec_xl(96, s + i);
  60                         vX8 = vec_xl(112, s + i);
  61
  62                         vY1 = vY1 ^ EC_vec_permxor(vhi0, vlo0, vX1);
  63                         vY2 = vY2 ^ EC_vec_permxor(vhi0, vlo0, vX2);
  64                         vY3 = vY3 ^ EC_vec_permxor(vhi0, vlo0, vX3);
  65                         vY4 = vY4 ^ EC_vec_permxor(vhi0, vlo0, vX4);
  66
  67                         vY5 = vY5 ^ EC_vec_permxor(vhi0, vlo0, vX5);
  68                         vY6 = vY6 ^ EC_vec_permxor(vhi0, vlo0, vX6);
  69                         vY7 = vY7 ^ EC_vec_permxor(vhi0, vlo0, vX7);
  70                         vY8 = vY8 ^ EC_vec_permxor(vhi0, vlo0, vX8);
  71
  72                         g0 += 32;
  73                 }
  74                 vec_xst(vY1, 0, t0 + i);
  75                 vec_xst(vY2, 16, t0 + i);
  76                 vec_xst(vY3, 32, t0 + i);
  77                 vec_xst(vY4, 48, t0 + i);
  78
  79                 vec_xst(vY5, 64, t0 + i);
  80                 vec_xst(vY6, 80, t0 + i);
  81                 vec_xst(vY7, 96, t0 + i);
  82                 vec_xst(vY8, 112, t0 + i);
  83         }
  84         return;
  85 }