gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/vuzp.c

   1 #include <arm_neon.h>
   2 #include "arm-neon-ref.h"
   3 #include "compute-ref-data.h"
   4
   5 /* Expected results splitted in several chunks.  */
   6 /* Chunk 0.  */
   7 VECT_VAR_DECL(expected0,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3,
   8                                         0xf4, 0xf5, 0xf6, 0xf7 };
   9 VECT_VAR_DECL(expected0,int,16,4) [] = { 0xfff0, 0xfff1,
  10                                          0xfff2, 0xfff3 };
  11 VECT_VAR_DECL(expected0,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
  12 VECT_VAR_DECL(expected0,int,64,1) [] = { 0x3333333333333333 };
  13 VECT_VAR_DECL(expected0,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3,
  14                                          0xf4, 0xf5, 0xf6, 0xf7 };
  15 VECT_VAR_DECL(expected0,uint,16,4) [] = { 0xfff0, 0xfff1,
  16                                           0xfff2, 0xfff3 };
  17 VECT_VAR_DECL(expected0,uint,32,2) [] = { 0xfffffff0,
  18                                           0xfffffff1 };
  19 VECT_VAR_DECL(expected0,uint,64,1) [] = { 0x3333333333333333 };
  20 VECT_VAR_DECL(expected0,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3,
  21                                          0xf4, 0xf5, 0xf6, 0xf7 };
  22 VECT_VAR_DECL(expected0,poly,16,4) [] = { 0xfff0, 0xfff1,
  23                                           0xfff2, 0xfff3 };
  24 VECT_VAR_DECL(expected0,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
  25 VECT_VAR_DECL(expected0,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3,
  26                                          0xf4, 0xf5, 0xf6, 0xf7,
  27                                          0xf8, 0xf9, 0xfa, 0xfb,
  28                                          0xfc, 0xfd, 0xfe, 0xff };
  29 VECT_VAR_DECL(expected0,int,16,8) [] = { 0xfff0, 0xfff1,
  30                                          0xfff2, 0xfff3,
  31                                          0xfff4, 0xfff5,
  32                                          0xfff6, 0xfff7 };
  33 VECT_VAR_DECL(expected0,int,32,4) [] = { 0xfffffff0, 0xfffffff1,
  34                                          0xfffffff2, 0xfffffff3 };
  35 VECT_VAR_DECL(expected0,int,64,2) [] = { 0x3333333333333333,
  36                                          0x3333333333333333 };
  37 VECT_VAR_DECL(expected0,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3,
  38                                           0xf4, 0xf5, 0xf6, 0xf7,
  39                                           0xf8, 0xf9, 0xfa, 0xfb,
  40                                           0xfc, 0xfd, 0xfe, 0xff };
  41 VECT_VAR_DECL(expected0,uint,16,8) [] = { 0xfff0, 0xfff1,
  42                                           0xfff2, 0xfff3,
  43                                           0xfff4, 0xfff5,
  44                                           0xfff6, 0xfff7 };
  45 VECT_VAR_DECL(expected0,uint,32,4) [] = { 0xfffffff0, 0xfffffff1,
  46                                           0xfffffff2, 0xfffffff3 };
  47 VECT_VAR_DECL(expected0,uint,64,2) [] = { 0x3333333333333333,
  48                                           0x3333333333333333 };
  49 VECT_VAR_DECL(expected0,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3,
  50                                           0xf4, 0xf5, 0xf6, 0xf7,
  51                                           0xf8, 0xf9, 0xfa, 0xfb,
  52                                           0xfc, 0xfd, 0xfe, 0xff };
  53 VECT_VAR_DECL(expected0,poly,16,8) [] = { 0xfff0, 0xfff1,
  54                                           0xfff2, 0xfff3,
  55                                           0xfff4, 0xfff5,
  56                                           0xfff6, 0xfff7 };
  57 VECT_VAR_DECL(expected0,hfloat,32,4) [] = { 0xc1800000, 0xc1700000,
  58                                             0xc1600000, 0xc1500000 };
  59
  60 /* Chunk 1.  */
  61 VECT_VAR_DECL(expected1,int,8,8) [] = { 0x11, 0x11, 0x11, 0x11,
  62                                         0x11, 0x11, 0x11, 0x11 };
  63 VECT_VAR_DECL(expected1,int,16,4) [] = { 0x22, 0x22, 0x22, 0x22 };
  64 VECT_VAR_DECL(expected1,int,32,2) [] = { 0x33, 0x33 };
  65 VECT_VAR_DECL(expected1,int,64,1) [] = { 0x3333333333333333 };
  66 VECT_VAR_DECL(expected1,uint,8,8) [] = { 0x55, 0x55, 0x55, 0x55,
  67                                          0x55, 0x55, 0x55, 0x55 };
  68 VECT_VAR_DECL(expected1,uint,16,4) [] = { 0x66, 0x66, 0x66, 0x66 };
  69 VECT_VAR_DECL(expected1,uint,32,2) [] = { 0x77, 0x77 };
  70 VECT_VAR_DECL(expected1,uint,64,1) [] = { 0x3333333333333333 };
  71 VECT_VAR_DECL(expected1,poly,8,8) [] = { 0x55, 0x55, 0x55, 0x55,
  72                                          0x55, 0x55, 0x55, 0x55 };
  73 VECT_VAR_DECL(expected1,poly,16,4) [] = { 0x66, 0x66, 0x66, 0x66 };
  74 VECT_VAR_DECL(expected1,hfloat,32,2) [] = { 0x42066666, 0x42066666 };
  75 VECT_VAR_DECL(expected1,int,8,16) [] = { 0x11, 0x11, 0x11, 0x11,
  76                                          0x11, 0x11, 0x11, 0x11,
  77                                          0x11, 0x11, 0x11, 0x11,
  78                                          0x11, 0x11, 0x11, 0x11 };
  79 VECT_VAR_DECL(expected1,int,16,8) [] = { 0x22, 0x22, 0x22, 0x22,
  80                                          0x22, 0x22, 0x22, 0x22 };
  81 VECT_VAR_DECL(expected1,int,32,4) [] = { 0x33, 0x33, 0x33, 0x33 };
  82 VECT_VAR_DECL(expected1,int,64,2) [] = { 0x3333333333333333,
  83                                          0x3333333333333333 };
  84 VECT_VAR_DECL(expected1,uint,8,16) [] = { 0x55, 0x55, 0x55, 0x55,
  85                                           0x55, 0x55, 0x55, 0x55,
  86                                           0x55, 0x55, 0x55, 0x55,
  87                                           0x55, 0x55, 0x55, 0x55 };
  88 VECT_VAR_DECL(expected1,uint,16,8) [] = { 0x66, 0x66, 0x66, 0x66,
  89                                           0x66, 0x66, 0x66, 0x66 };
  90 VECT_VAR_DECL(expected1,uint,32,4) [] = { 0x77, 0x77, 0x77, 0x77 };
  91 VECT_VAR_DECL(expected1,uint,64,2) [] = { 0x3333333333333333,
  92                                           0x3333333333333333 };
  93 VECT_VAR_DECL(expected1,poly,8,16) [] = { 0x55, 0x55, 0x55, 0x55,
  94                                           0x55, 0x55, 0x55, 0x55,
  95                                           0x55, 0x55, 0x55, 0x55,
  96                                           0x55, 0x55, 0x55, 0x55 };
  97 VECT_VAR_DECL(expected1,poly,16,8) [] = { 0x66, 0x66, 0x66, 0x66,
  98                                           0x66, 0x66, 0x66, 0x66 };
  99 VECT_VAR_DECL(expected1,hfloat,32,4) [] = { 0x42073333, 0x42073333,
 100                                             0x42073333, 0x42073333 };
 101
 102 #ifndef INSN_NAME
 103 #define INSN_NAME vuzp
 104 #define TEST_MSG "VUZP/VUZPQ"
 105 #endif
 106
 107 #define FNNAME1(NAME) exec_ ## NAME
 108 #define FNNAME(NAME) FNNAME1(NAME)
 109
 110 void FNNAME (INSN_NAME) (void)
 111 {
 112   /* In this case, output variables are arrays of vectors.  */
 113 #define DECL_VUZP(T1, W, N)                                             \
 114   VECT_ARRAY_TYPE(T1, W, N, 2) VECT_ARRAY_VAR(result_vec, T1, W, N, 2); \
 115   VECT_VAR_DECL(result_bis, T1, W, N)[2 * N]
 116
 117   /* We need to use a temporary result buffer (result_bis), because
 118      the one used for other tests is not large enough. A subset of the
 119      result data is moved from result_bis to result, and it is this
 120      subset which is used to check the actual behaviour. The next
 121      macro enables to move another chunk of data from result_bis to
 122      result.  */
 123 #define TEST_VUZP(INSN, Q, T1, T2, W, N)                                \
 124   VECT_ARRAY_VAR(result_vec, T1, W, N, 2) =                             \
 125     INSN##Q##_##T2##W(VECT_VAR(vector1, T1, W, N),                      \
 126                       VECT_VAR(vector2, T1, W, N));                     \
 127   vst2##Q##_##T2##W(VECT_VAR(result_bis, T1, W, N),                     \
 128                     VECT_ARRAY_VAR(result_vec, T1, W, N, 2));           \
 129   memcpy(VECT_VAR(result, T1, W, N), VECT_VAR(result_bis, T1, W, N),    \
 130          sizeof(VECT_VAR(result, T1, W, N)));
 131
 132   /* Overwrite "result" with the contents of "result_bis"[X].  */
 133 #define TEST_EXTRA_CHUNK(T1, W, N, X)                                   \
 134   memcpy(VECT_VAR(result, T1, W, N), &(VECT_VAR(result_bis, T1, W, N)[X*N]), \
 135          sizeof(VECT_VAR(result, T1, W, N)));
 136
 137   DECL_VARIABLE_ALL_VARIANTS(vector1);
 138   DECL_VARIABLE_ALL_VARIANTS(vector2);
 139
 140   /* We don't need 64 bits variants.  */
 141 #define DECL_ALL_VUZP()                         \
 142   DECL_VUZP(int, 8, 8);                         \
 143   DECL_VUZP(int, 16, 4);                        \
 144   DECL_VUZP(int, 32, 2);                        \
 145   DECL_VUZP(uint, 8, 8);                        \
 146   DECL_VUZP(uint, 16, 4);                       \
 147   DECL_VUZP(uint, 32, 2);                       \
 148   DECL_VUZP(poly, 8, 8);                        \
 149   DECL_VUZP(poly, 16, 4);                       \
 150   DECL_VUZP(float, 32, 2);                      \
 151   DECL_VUZP(int, 8, 16);                        \
 152   DECL_VUZP(int, 16, 8);                        \
 153   DECL_VUZP(int, 32, 4);                        \
 154   DECL_VUZP(uint, 8, 16);                       \
 155   DECL_VUZP(uint, 16, 8);                       \
 156   DECL_VUZP(uint, 32, 4);                       \
 157   DECL_VUZP(poly, 8, 16);                       \
 158   DECL_VUZP(poly, 16, 8);                       \
 159   DECL_VUZP(float, 32, 4)
 160
 161   DECL_ALL_VUZP();
 162
 163   /* Initialize input "vector" from "buffer".  */
 164   TEST_MACRO_ALL_VARIANTS_2_5(VLOAD, vector1, buffer);
 165   VLOAD(vector1, buffer, , float, f, 32, 2);
 166   VLOAD(vector1, buffer, q, float, f, 32, 4);
 167
 168   /* Choose arbitrary initialization values.  */
 169   VDUP(vector2, , int, s, 8, 8, 0x11);
 170   VDUP(vector2, , int, s, 16, 4, 0x22);
 171   VDUP(vector2, , int, s, 32, 2, 0x33);
 172   VDUP(vector2, , uint, u, 8, 8, 0x55);
 173   VDUP(vector2, , uint, u, 16, 4, 0x66);
 174   VDUP(vector2, , uint, u, 32, 2, 0x77);
 175   VDUP(vector2, , poly, p, 8, 8, 0x55);
 176   VDUP(vector2, , poly, p, 16, 4, 0x66);
 177   VDUP(vector2, , float, f, 32, 2, 33.6f);
 178
 179   VDUP(vector2, q, int, s, 8, 16, 0x11);
 180   VDUP(vector2, q, int, s, 16, 8, 0x22);
 181   VDUP(vector2, q, int, s, 32, 4, 0x33);
 182   VDUP(vector2, q, uint, u, 8, 16, 0x55);
 183   VDUP(vector2, q, uint, u, 16, 8, 0x66);
 184   VDUP(vector2, q, uint, u, 32, 4, 0x77);
 185   VDUP(vector2, q, poly, p, 8, 16, 0x55);
 186   VDUP(vector2, q, poly, p, 16, 8, 0x66);
 187   VDUP(vector2, q, float, f, 32, 4, 33.8f);
 188
 189 #define TEST_ALL_VUZP(INSN)                     \
 190   TEST_VUZP(INSN, , int, s, 8, 8);              \
 191   TEST_VUZP(INSN, , int, s, 16, 4);             \
 192   TEST_VUZP(INSN, , int, s, 32, 2);             \
 193   TEST_VUZP(INSN, , uint, u, 8, 8);             \
 194   TEST_VUZP(INSN, , uint, u, 16, 4);            \
 195   TEST_VUZP(INSN, , uint, u, 32, 2);            \
 196   TEST_VUZP(INSN, , poly, p, 8, 8);             \
 197   TEST_VUZP(INSN, , poly, p, 16, 4);            \
 198   TEST_VUZP(INSN, , float, f, 32, 2);           \
 199   TEST_VUZP(INSN, q, int, s, 8, 16);            \
 200   TEST_VUZP(INSN, q, int, s, 16, 8);            \
 201   TEST_VUZP(INSN, q, int, s, 32, 4);            \
 202   TEST_VUZP(INSN, q, uint, u, 8, 16);           \
 203   TEST_VUZP(INSN, q, uint, u, 16, 8);           \
 204   TEST_VUZP(INSN, q, uint, u, 32, 4);           \
 205   TEST_VUZP(INSN, q, poly, p, 8, 16);           \
 206   TEST_VUZP(INSN, q, poly, p, 16, 8);           \
 207   TEST_VUZP(INSN, q, float, f, 32, 4)
 208
 209 #define TEST_ALL_EXTRA_CHUNKS()                 \
 210   TEST_EXTRA_CHUNK(int, 8, 8, 1);               \
 211   TEST_EXTRA_CHUNK(int, 16, 4, 1);              \
 212   TEST_EXTRA_CHUNK(int, 32, 2, 1);              \
 213   TEST_EXTRA_CHUNK(uint, 8, 8, 1);              \
 214   TEST_EXTRA_CHUNK(uint, 16, 4, 1);             \
 215   TEST_EXTRA_CHUNK(uint, 32, 2, 1);             \
 216   TEST_EXTRA_CHUNK(poly, 8, 8, 1);              \
 217   TEST_EXTRA_CHUNK(poly, 16, 4, 1);             \
 218   TEST_EXTRA_CHUNK(float, 32, 2, 1);            \
 219   TEST_EXTRA_CHUNK(int, 8, 16, 1);              \
 220   TEST_EXTRA_CHUNK(int, 16, 8, 1);              \
 221   TEST_EXTRA_CHUNK(int, 32, 4, 1);              \
 222   TEST_EXTRA_CHUNK(uint, 8, 16, 1);             \
 223   TEST_EXTRA_CHUNK(uint, 16, 8, 1);             \
 224   TEST_EXTRA_CHUNK(uint, 32, 4, 1);             \
 225   TEST_EXTRA_CHUNK(poly, 8, 16, 1);             \
 226   TEST_EXTRA_CHUNK(poly, 16, 8, 1);             \
 227   TEST_EXTRA_CHUNK(float, 32, 4, 1)
 228
 229   clean_results ();
 230
 231   /* Execute the tests.  */
 232   TEST_ALL_VUZP(INSN_NAME);
 233
 234   CHECK_RESULTS_NAMED (TEST_MSG, expected0, "(chunk 0)");
 235
 236   TEST_ALL_EXTRA_CHUNKS();
 237   CHECK_RESULTS_NAMED (TEST_MSG, expected1, "(chunk 1)");
 238 }
 239
 240 int main (void)
 241 {
 242   FNNAME (INSN_NAME) ();
 243
 244   return 0;
 245 }