gcc/testsuite/gcc.target/powerpc/ssse3-palignr.c

   1 /* { dg-do run } */
   2 /* { dg-options "-O3 -mvsx -Wno-psabi" } */
   3 /* { dg-additional-options "-mdejagnu-cpu=power8" { target { ! has_arch_pwr8 } } } */
   4 /* { dg-require-effective-target p8vector_hw } */
   5
   6 #ifndef CHECK_H
   7 #define CHECK_H "ssse3-check.h"
   8 #endif
   9
  10 #ifndef TEST
  11 #define TEST ssse3_test
  12 #endif
  13
  14 #include CHECK_H
  15
  16 #include "ssse3-vals.h"
  17
  18 #include <tmmintrin.h>
  19 #include <string.h>
  20
  21 #ifndef __AVX__
  22 /* Test the 64-bit form */
  23 static void
  24 ssse3_test_palignr (__m64 *i1, __m64 *i2, unsigned int imm, __m64 *r)
  25 {
  26   switch (imm)
  27     {
  28     case 0:
  29       *r = _mm_alignr_pi8 (*i1, *i2, 0);
  30       break;
  31     case 1:
  32       *r = _mm_alignr_pi8 (*i1, *i2, 1);
  33       break;
  34     case 2:
  35       *r = _mm_alignr_pi8 (*i1, *i2, 2);
  36       break;
  37     case 3:
  38       *r = _mm_alignr_pi8 (*i1, *i2, 3);
  39       break;
  40     case 4:
  41       *r = _mm_alignr_pi8 (*i1, *i2, 4);
  42       break;
  43     case 5:
  44       *r = _mm_alignr_pi8 (*i1, *i2, 5);
  45       break;
  46     case 6:
  47       *r = _mm_alignr_pi8 (*i1, *i2, 6);
  48       break;
  49     case 7:
  50       *r = _mm_alignr_pi8 (*i1, *i2, 7);
  51       break;
  52     case 8:
  53       *r = _mm_alignr_pi8 (*i1, *i2, 8);
  54       break;
  55     case 9:
  56       *r = _mm_alignr_pi8 (*i1, *i2, 9);
  57       break;
  58     case 10:
  59       *r = _mm_alignr_pi8 (*i1, *i2, 10);
  60       break;
  61     case 11:
  62       *r = _mm_alignr_pi8 (*i1, *i2, 11);
  63       break;
  64     case 12:
  65       *r = _mm_alignr_pi8 (*i1, *i2, 12);
  66       break;
  67     case 13:
  68       *r = _mm_alignr_pi8 (*i1, *i2, 13);
  69       break;
  70     case 14:
  71       *r = _mm_alignr_pi8 (*i1, *i2, 14);
  72       break;
  73     case 15:
  74       *r = _mm_alignr_pi8 (*i1, *i2, 15);
  75       break;
  76     default:
  77       *r = _mm_alignr_pi8 (*i1, *i2, 16);
  78       break;
  79     }
  80
  81    _mm_empty();
  82 }
  83 #endif
  84
  85 /* Test the 128-bit form */
  86 static void
  87 ssse3_test_palignr128 (__m128i *i1, __m128i *i2, unsigned int imm, __m128i *r)
  88 {
  89   /* Assumes incoming pointers are 16-byte aligned */
  90
  91   switch (imm)
  92     {
  93     case 0:
  94       *r = _mm_alignr_epi8 (*i1, *i2, 0);
  95       break;
  96     case 1:
  97       *r = _mm_alignr_epi8 (*i1, *i2, 1);
  98       break;
  99     case 2:
 100       *r = _mm_alignr_epi8 (*i1, *i2, 2);
 101       break;
 102     case 3:
 103       *r = _mm_alignr_epi8 (*i1, *i2, 3);
 104       break;
 105     case 4:
 106       *r = _mm_alignr_epi8 (*i1, *i2, 4);
 107       break;
 108     case 5:
 109       *r = _mm_alignr_epi8 (*i1, *i2, 5);
 110       break;
 111     case 6:
 112       *r = _mm_alignr_epi8 (*i1, *i2, 6);
 113       break;
 114     case 7:
 115       *r = _mm_alignr_epi8 (*i1, *i2, 7);
 116       break;
 117     case 8:
 118       *r = _mm_alignr_epi8 (*i1, *i2, 8);
 119       break;
 120     case 9:
 121       *r = _mm_alignr_epi8 (*i1, *i2, 9);
 122       break;
 123     case 10:
 124       *r = _mm_alignr_epi8 (*i1, *i2, 10);
 125       break;
 126     case 11:
 127       *r = _mm_alignr_epi8 (*i1, *i2, 11);
 128       break;
 129     case 12:
 130       *r = _mm_alignr_epi8 (*i1, *i2, 12);
 131       break;
 132     case 13:
 133       *r = _mm_alignr_epi8 (*i1, *i2, 13);
 134       break;
 135     case 14:
 136       *r = _mm_alignr_epi8 (*i1, *i2, 14);
 137       break;
 138     case 15:
 139       *r = _mm_alignr_epi8 (*i1, *i2, 15);
 140       break;
 141     case 16:
 142       *r = _mm_alignr_epi8 (*i1, *i2, 16);
 143       break;
 144     case 17:
 145       *r = _mm_alignr_epi8 (*i1, *i2, 17);
 146       break;
 147     case 18:
 148       *r = _mm_alignr_epi8 (*i1, *i2, 18);
 149       break;
 150     case 19:
 151       *r = _mm_alignr_epi8 (*i1, *i2, 19);
 152       break;
 153     case 20:
 154       *r = _mm_alignr_epi8 (*i1, *i2, 20);
 155       break;
 156     case 21:
 157       *r = _mm_alignr_epi8 (*i1, *i2, 21);
 158       break;
 159     case 22:
 160       *r = _mm_alignr_epi8 (*i1, *i2, 22);
 161       break;
 162     case 23:
 163       *r = _mm_alignr_epi8 (*i1, *i2, 23);
 164       break;
 165     case 24:
 166       *r = _mm_alignr_epi8 (*i1, *i2, 24);
 167       break;
 168     case 25:
 169       *r = _mm_alignr_epi8 (*i1, *i2, 25);
 170       break;
 171     case 26:
 172       *r = _mm_alignr_epi8 (*i1, *i2, 26);
 173       break;
 174     case 27:
 175       *r = _mm_alignr_epi8 (*i1, *i2, 27);
 176       break;
 177     case 28:
 178       *r = _mm_alignr_epi8 (*i1, *i2, 28);
 179       break;
 180     case 29:
 181       *r = _mm_alignr_epi8 (*i1, *i2, 29);
 182       break;
 183     case 30:
 184       *r = _mm_alignr_epi8 (*i1, *i2, 30);
 185       break;
 186     case 31:
 187       *r = _mm_alignr_epi8 (*i1, *i2, 31);
 188       break;
 189     default:
 190       *r = _mm_alignr_epi8 (*i1, *i2, 32);
 191       break;
 192     }
 193 }
 194
 195 /* Routine to manually compute the results */
 196 static void
 197 compute_correct_result_128 (signed char *i1, signed char *i2, unsigned int imm,
 198                             signed char *r)
 199 {
 200   signed char buf [32];
 201   int i;
 202
 203   memcpy (&buf[0], i2, 16);
 204   memcpy (&buf[16], i1, 16);
 205
 206   for (i = 0; i < 16; i++)
 207     if (imm >= 32 || imm + i >= 32)
 208       r[i] = 0;
 209     else
 210       r[i] = buf[imm + i];
 211 }
 212
 213 #ifndef __AVX__
 214 static void
 215 compute_correct_result_64 (signed char *i1, signed char *i2, unsigned int imm,
 216                            signed char *r)
 217 {
 218   signed char buf [16];
 219   int i;
 220
 221   /* Handle the first half */
 222   memcpy (&buf[0], &i2[0], 8);
 223   memcpy (&buf[8], &i1[0], 8);
 224
 225   for (i = 0; i < 8; i++)
 226     if (imm >= 16 || imm + i >= 16)
 227       r[i] = 0;
 228     else
 229       r[i] = buf[imm + i];
 230
 231   /* Handle the second half */
 232   memcpy (&buf[0], &i2[8], 8);
 233   memcpy (&buf[8], &i1[8], 8);
 234
 235   for (i = 0; i < 8; i++)
 236     if (imm >= 16 || imm + i >= 16)
 237       r[i + 8] = 0;
 238     else
 239       r[i + 8] = buf[imm + i];
 240 }
 241 #endif
 242
 243 static void
 244 TEST (void)
 245 {
 246   int i;
 247   union data r __attribute__ ((aligned(16)));
 248   union data ck;
 249   unsigned int imm;
 250   int fail = 0;
 251
 252   for (i = 0; i < ARRAY_SIZE (vals) - 1; i++)
 253     for (imm = 0; imm < 100; imm++)
 254       {
 255 #ifndef __AVX__
 256         /* Manually compute the result */
 257         compute_correct_result_64 (&vals[i + 0].b[0],
 258                                    &vals[i + 1].b[0], imm, &ck.b[0]);
 259
 260         /* Run the 64-bit tests */
 261         ssse3_test_palignr (&vals[i + 0].ll[0],
 262                             &vals[i + 1].ll[0], imm, &r.ll[0]);
 263         ssse3_test_palignr (&vals[i + 0].ll[1],
 264                             &vals[i + 1].ll[1], imm, &r.ll[1]);
 265         fail += chk_128 (ck.m[0], r.m[0]);
 266 #endif
 267
 268         /* Recompute the results for 128-bits */
 269         compute_correct_result_128 (&vals[i + 0].b[0],
 270                                     &vals[i + 1].b[0], imm, &ck.b[0]);
 271
 272         /* Run the 128-bit tests */
 273         ssse3_test_palignr128 (&vals[i + 0].m[0],
 274                                &vals[i + 1].m[0], imm, &r.m[0]);
 275         fail += chk_128 (ck.m[0], r.m[0]);
 276       }
 277
 278   if (fail != 0)
 279     abort ();
 280 }