| 1 |
#include <arm_neon.h> |
| 2 |
#include "arm-neon-ref.h" |
| 3 |
#include "compute-ref-data.h" |
| 4 |
#include <math.h> |
| 5 |
|
| 6 |
/* Additional expected results declaration, they are initialized in |
| 7 |
each test file. */ |
| 8 |
extern ARRAY(expected_uint, uint, 8, 8); |
| 9 |
extern ARRAY(expected_uint, uint, 16, 4); |
| 10 |
extern ARRAY(expected_uint, uint, 32, 2); |
| 11 |
extern ARRAY(expected_q_uint, uint, 8, 16); |
| 12 |
extern ARRAY(expected_q_uint, uint, 16, 8); |
| 13 |
extern ARRAY(expected_q_uint, uint, 32, 4); |
| 14 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 15 |
extern ARRAY(expected_float, uint, 16, 4); |
| 16 |
extern ARRAY(expected_q_float, uint, 16, 8); |
| 17 |
extern ARRAY(expected_nan, uint, 16, 4); |
| 18 |
extern ARRAY(expected_mnan, uint, 16, 4); |
| 19 |
extern ARRAY(expected_nan2, uint, 16, 4); |
| 20 |
extern ARRAY(expected_inf, uint, 16, 4); |
| 21 |
extern ARRAY(expected_minf, uint, 16, 4); |
| 22 |
extern ARRAY(expected_inf2, uint, 16, 4); |
| 23 |
extern ARRAY(expected_mzero, uint, 16, 4); |
| 24 |
#endif |
| 25 |
extern ARRAY(expected_float, uint, 32, 2); |
| 26 |
extern ARRAY(expected_q_float, uint, 32, 4); |
| 27 |
extern ARRAY(expected_uint2, uint, 32, 2); |
| 28 |
extern ARRAY(expected_uint3, uint, 32, 2); |
| 29 |
extern ARRAY(expected_uint4, uint, 32, 2); |
| 30 |
extern ARRAY(expected_nan, uint, 32, 2); |
| 31 |
extern ARRAY(expected_mnan, uint, 32, 2); |
| 32 |
extern ARRAY(expected_nan2, uint, 32, 2); |
| 33 |
extern ARRAY(expected_inf, uint, 32, 2); |
| 34 |
extern ARRAY(expected_minf, uint, 32, 2); |
| 35 |
extern ARRAY(expected_inf2, uint, 32, 2); |
| 36 |
extern ARRAY(expected_mzero, uint, 32, 2); |
| 37 |
extern ARRAY(expected_p8, uint, 8, 8); |
| 38 |
extern ARRAY(expected_q_p8, uint, 8, 16); |
| 39 |
|
| 40 |
#define FNNAME1(NAME) exec_ ## NAME |
| 41 |
#define FNNAME(NAME) FNNAME1(NAME) |
| 42 |
|
| 43 |
void FNNAME (INSN_NAME) (void) |
| 44 |
{ |
| 45 |
/* Basic test: y=vcomp(x1,x2), then store the result. */ |
| 46 |
#define TEST_VCOMP1(INSN, Q, T1, T2, T3, W, N) \ |
| 47 |
VECT_VAR(vector_res, T3, W, N) = \ |
| 48 |
INSN##Q##_##T2##W(VECT_VAR(vector, T1, W, N), \ |
| 49 |
VECT_VAR(vector2, T1, W, N)); \ |
| 50 |
vst1##Q##_u##W(VECT_VAR(result, T3, W, N), VECT_VAR(vector_res, T3, W, N)) |
| 51 |
|
| 52 |
#define TEST_VCOMP(INSN, Q, T1, T2, T3, W, N) \ |
| 53 |
TEST_VCOMP1(INSN, Q, T1, T2, T3, W, N) |
| 54 |
|
| 55 |
/* No need for 64 bits elements. */ |
| 56 |
DECL_VARIABLE(vector, int, 8, 8); |
| 57 |
DECL_VARIABLE(vector, int, 16, 4); |
| 58 |
DECL_VARIABLE(vector, int, 32, 2); |
| 59 |
DECL_VARIABLE(vector, uint, 8, 8); |
| 60 |
DECL_VARIABLE(vector, uint, 16, 4); |
| 61 |
DECL_VARIABLE(vector, uint, 32, 2); |
| 62 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 63 |
DECL_VARIABLE (vector, float, 16, 4); |
| 64 |
#endif |
| 65 |
DECL_VARIABLE(vector, float, 32, 2); |
| 66 |
DECL_VARIABLE(vector, int, 8, 16); |
| 67 |
DECL_VARIABLE(vector, int, 16, 8); |
| 68 |
DECL_VARIABLE(vector, int, 32, 4); |
| 69 |
DECL_VARIABLE(vector, uint, 8, 16); |
| 70 |
DECL_VARIABLE(vector, uint, 16, 8); |
| 71 |
DECL_VARIABLE(vector, uint, 32, 4); |
| 72 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 73 |
DECL_VARIABLE (vector, float, 16, 8); |
| 74 |
#endif |
| 75 |
DECL_VARIABLE(vector, float, 32, 4); |
| 76 |
|
| 77 |
DECL_VARIABLE(vector2, int, 8, 8); |
| 78 |
DECL_VARIABLE(vector2, int, 16, 4); |
| 79 |
DECL_VARIABLE(vector2, int, 32, 2); |
| 80 |
DECL_VARIABLE(vector2, uint, 8, 8); |
| 81 |
DECL_VARIABLE(vector2, uint, 16, 4); |
| 82 |
DECL_VARIABLE(vector2, uint, 32, 2); |
| 83 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 84 |
DECL_VARIABLE (vector2, float, 16, 4); |
| 85 |
#endif |
| 86 |
DECL_VARIABLE(vector2, float, 32, 2); |
| 87 |
DECL_VARIABLE(vector2, int, 8, 16); |
| 88 |
DECL_VARIABLE(vector2, int, 16, 8); |
| 89 |
DECL_VARIABLE(vector2, int, 32, 4); |
| 90 |
DECL_VARIABLE(vector2, uint, 8, 16); |
| 91 |
DECL_VARIABLE(vector2, uint, 16, 8); |
| 92 |
DECL_VARIABLE(vector2, uint, 32, 4); |
| 93 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 94 |
DECL_VARIABLE (vector2, float, 16, 8); |
| 95 |
#endif |
| 96 |
DECL_VARIABLE(vector2, float, 32, 4); |
| 97 |
|
| 98 |
DECL_VARIABLE(vector_res, uint, 8, 8); |
| 99 |
DECL_VARIABLE(vector_res, uint, 16, 4); |
| 100 |
DECL_VARIABLE(vector_res, uint, 32, 2); |
| 101 |
DECL_VARIABLE(vector_res, uint, 8, 16); |
| 102 |
DECL_VARIABLE(vector_res, uint, 16, 8); |
| 103 |
DECL_VARIABLE(vector_res, uint, 32, 4); |
| 104 |
|
| 105 |
clean_results (); |
| 106 |
|
| 107 |
/* There is no 64 bits variant, don't use the generic initializer. */ |
| 108 |
VLOAD(vector, buffer, , int, s, 8, 8); |
| 109 |
VLOAD(vector, buffer, , int, s, 16, 4); |
| 110 |
VLOAD(vector, buffer, , int, s, 32, 2); |
| 111 |
VLOAD(vector, buffer, , uint, u, 8, 8); |
| 112 |
VLOAD(vector, buffer, , uint, u, 16, 4); |
| 113 |
VLOAD(vector, buffer, , uint, u, 32, 2); |
| 114 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 115 |
VLOAD (vector, buffer, , float, f, 16, 4); |
| 116 |
#endif |
| 117 |
VLOAD(vector, buffer, , float, f, 32, 2); |
| 118 |
|
| 119 |
VLOAD(vector, buffer, q, int, s, 8, 16); |
| 120 |
VLOAD(vector, buffer, q, int, s, 16, 8); |
| 121 |
VLOAD(vector, buffer, q, int, s, 32, 4); |
| 122 |
VLOAD(vector, buffer, q, uint, u, 8, 16); |
| 123 |
VLOAD(vector, buffer, q, uint, u, 16, 8); |
| 124 |
VLOAD(vector, buffer, q, uint, u, 32, 4); |
| 125 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 126 |
VLOAD (vector, buffer, q, float, f, 16, 8); |
| 127 |
#endif |
| 128 |
VLOAD(vector, buffer, q, float, f, 32, 4); |
| 129 |
|
| 130 |
/* Choose init value arbitrarily, will be used for vector |
| 131 |
comparison. */ |
| 132 |
VDUP(vector2, , int, s, 8, 8, -10); |
| 133 |
VDUP(vector2, , int, s, 16, 4, -14); |
| 134 |
VDUP(vector2, , int, s, 32, 2, -16); |
| 135 |
VDUP(vector2, , uint, u, 8, 8, 0xF3); |
| 136 |
VDUP(vector2, , uint, u, 16, 4, 0xFFF2); |
| 137 |
VDUP(vector2, , uint, u, 32, 2, 0xFFFFFFF1); |
| 138 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 139 |
VDUP (vector2, , float, f, 16, 4, -15.0f); |
| 140 |
#endif |
| 141 |
VDUP(vector2, , float, f, 32, 2, -15.0f); |
| 142 |
|
| 143 |
VDUP(vector2, q, int, s, 8, 16, -4); |
| 144 |
VDUP(vector2, q, int, s, 16, 8, -10); |
| 145 |
VDUP(vector2, q, int, s, 32, 4, -14); |
| 146 |
VDUP(vector2, q, uint, u, 8, 16, 0xF4); |
| 147 |
VDUP(vector2, q, uint, u, 16, 8, 0xFFF6); |
| 148 |
VDUP(vector2, q, uint, u, 32, 4, 0xFFFFFFF2); |
| 149 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 150 |
VDUP (vector2, q, float, f, 16, 8, -14.0f); |
| 151 |
#endif |
| 152 |
VDUP(vector2, q, float, f, 32, 4, -14.0f); |
| 153 |
|
| 154 |
/* The comparison operators produce only unsigned results, which |
| 155 |
means that our tests with uint* inputs write their results in the |
| 156 |
same vectors as the int* variants. As a consequence, we have to |
| 157 |
execute and test the int* first, then the uint* ones. |
| 158 |
Same thing for float and poly8. |
| 159 |
*/ |
| 160 |
|
| 161 |
/* Apply operator named INSN_NAME. */ |
| 162 |
TEST_VCOMP(INSN_NAME, , int, s, uint, 8, 8); |
| 163 |
TEST_VCOMP(INSN_NAME, , int, s, uint, 16, 4); |
| 164 |
TEST_VCOMP(INSN_NAME, , int, s, uint, 32, 2); |
| 165 |
TEST_VCOMP(INSN_NAME, q, int, s, uint, 8, 16); |
| 166 |
TEST_VCOMP(INSN_NAME, q, int, s, uint, 16, 8); |
| 167 |
TEST_VCOMP(INSN_NAME, q, int, s, uint, 32, 4); |
| 168 |
|
| 169 |
CHECK(TEST_MSG, uint, 8, 8, PRIx8, expected, ""); |
| 170 |
CHECK(TEST_MSG, uint, 16, 4, PRIx16, expected, ""); |
| 171 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected, ""); |
| 172 |
CHECK(TEST_MSG, uint, 8, 16, PRIx8, expected, ""); |
| 173 |
CHECK(TEST_MSG, uint, 16, 8, PRIx16, expected, ""); |
| 174 |
CHECK(TEST_MSG, uint, 32, 4, PRIx32, expected, ""); |
| 175 |
|
| 176 |
/* Now the uint* variants. */ |
| 177 |
TEST_VCOMP(INSN_NAME, , uint, u, uint, 8, 8); |
| 178 |
TEST_VCOMP(INSN_NAME, , uint, u, uint, 16, 4); |
| 179 |
TEST_VCOMP(INSN_NAME, , uint, u, uint, 32, 2); |
| 180 |
TEST_VCOMP(INSN_NAME, q, uint, u, uint, 8, 16); |
| 181 |
TEST_VCOMP(INSN_NAME, q, uint, u, uint, 16, 8); |
| 182 |
TEST_VCOMP(INSN_NAME, q, uint, u, uint, 32, 4); |
| 183 |
|
| 184 |
CHECK(TEST_MSG, uint, 8, 8, PRIx8, expected_uint, ""); |
| 185 |
CHECK(TEST_MSG, uint, 16, 4, PRIx16, expected_uint, ""); |
| 186 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_uint, ""); |
| 187 |
CHECK(TEST_MSG, uint, 8, 16, PRIx8, expected_q_uint, ""); |
| 188 |
CHECK(TEST_MSG, uint, 16, 8, PRIx16, expected_q_uint, ""); |
| 189 |
CHECK(TEST_MSG, uint, 32, 4, PRIx32, expected_q_uint, ""); |
| 190 |
|
| 191 |
/* The float variants. */ |
| 192 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 193 |
TEST_VCOMP (INSN_NAME, , float, f, uint, 16, 4); |
| 194 |
CHECK (TEST_MSG, uint, 16, 4, PRIx16, expected_float, ""); |
| 195 |
#endif |
| 196 |
TEST_VCOMP(INSN_NAME, , float, f, uint, 32, 2); |
| 197 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_float, ""); |
| 198 |
|
| 199 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 200 |
TEST_VCOMP (INSN_NAME, q, float, f, uint, 16, 8); |
| 201 |
CHECK (TEST_MSG, uint, 16, 8, PRIx16, expected_q_float, ""); |
| 202 |
#endif |
| 203 |
TEST_VCOMP(INSN_NAME, q, float, f, uint, 32, 4); |
| 204 |
CHECK(TEST_MSG, uint, 32, 4, PRIx32, expected_q_float, ""); |
| 205 |
|
| 206 |
/* Some "special" input values to test some corner cases. */ |
| 207 |
/* Extra tests to have 100% coverage on all the variants. */ |
| 208 |
VDUP(vector2, , uint, u, 32, 2, 0xFFFFFFF0); |
| 209 |
TEST_VCOMP(INSN_NAME, , uint, u, uint, 32, 2); |
| 210 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_uint2, "uint 0xfffffff0"); |
| 211 |
|
| 212 |
VDUP(vector2, , int, s, 32, 2, -15); |
| 213 |
TEST_VCOMP(INSN_NAME, , int, s, uint, 32, 2); |
| 214 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_uint3, "int -15"); |
| 215 |
|
| 216 |
VDUP(vector2, , float, f, 32, 2, -16.0f); |
| 217 |
TEST_VCOMP(INSN_NAME, , float, f, uint, 32, 2); |
| 218 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_uint4, "float -16.0f"); |
| 219 |
|
| 220 |
|
| 221 |
/* Extra FP tests with special values (NaN, ....). */ |
| 222 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 223 |
VDUP (vector, , float, f, 16, 4, 1.0); |
| 224 |
VDUP (vector2, , float, f, 16, 4, NAN); |
| 225 |
TEST_VCOMP (INSN_NAME, , float, f, uint, 16, 4); |
| 226 |
CHECK (TEST_MSG, uint, 16, 4, PRIx16, expected_nan, "FP special (NaN)"); |
| 227 |
|
| 228 |
VDUP (vector, , float, f, 16, 4, 1.0); |
| 229 |
VDUP (vector2, , float, f, 16, 4, -NAN); |
| 230 |
TEST_VCOMP (INSN_NAME, , float, f, uint, 16, 4); |
| 231 |
CHECK (TEST_MSG, uint, 16, 4, PRIx16, expected_mnan, " FP special (-NaN)"); |
| 232 |
|
| 233 |
VDUP (vector, , float, f, 16, 4, NAN); |
| 234 |
VDUP (vector2, , float, f, 16, 4, 1.0); |
| 235 |
TEST_VCOMP (INSN_NAME, , float, f, uint, 16, 4); |
| 236 |
CHECK (TEST_MSG, uint, 16, 4, PRIx16, expected_nan2, " FP special (NaN)"); |
| 237 |
|
| 238 |
VDUP (vector, , float, f, 16, 4, 1.0); |
| 239 |
VDUP (vector2, , float, f, 16, 4, HUGE_VALF); |
| 240 |
TEST_VCOMP (INSN_NAME, , float, f, uint, 16, 4); |
| 241 |
CHECK (TEST_MSG, uint, 16, 4, PRIx16, expected_inf, " FP special (inf)"); |
| 242 |
|
| 243 |
VDUP (vector, , float, f, 16, 4, 1.0); |
| 244 |
VDUP (vector2, , float, f, 16, 4, -HUGE_VALF); |
| 245 |
TEST_VCOMP (INSN_NAME, , float, f, uint, 16, 4); |
| 246 |
CHECK (TEST_MSG, uint, 16, 4, PRIx16, expected_minf, " FP special (-inf)"); |
| 247 |
|
| 248 |
VDUP (vector, , float, f, 16, 4, HUGE_VALF); |
| 249 |
VDUP (vector2, , float, f, 16, 4, 1.0); |
| 250 |
TEST_VCOMP (INSN_NAME, , float, f, uint, 16, 4); |
| 251 |
CHECK (TEST_MSG, uint, 16, 4, PRIx16, expected_inf2, " FP special (inf)"); |
| 252 |
|
| 253 |
VDUP (vector, , float, f, 16, 4, -0.0); |
| 254 |
VDUP (vector2, , float, f, 16, 4, 0.0); |
| 255 |
TEST_VCOMP (INSN_NAME, , float, f, uint, 16, 4); |
| 256 |
CHECK (TEST_MSG, uint, 16, 4, PRIx16, expected_mzero, " FP special (-0.0)"); |
| 257 |
#endif |
| 258 |
|
| 259 |
VDUP(vector, , float, f, 32, 2, 1.0); |
| 260 |
VDUP(vector2, , float, f, 32, 2, NAN); |
| 261 |
TEST_VCOMP(INSN_NAME, , float, f, uint, 32, 2); |
| 262 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_nan, "FP special (NaN)"); |
| 263 |
|
| 264 |
VDUP(vector, , float, f, 32, 2, 1.0); |
| 265 |
VDUP(vector2, , float, f, 32, 2, -NAN); |
| 266 |
TEST_VCOMP(INSN_NAME, , float, f, uint, 32, 2); |
| 267 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_mnan, " FP special (-NaN)"); |
| 268 |
|
| 269 |
VDUP(vector, , float, f, 32, 2, NAN); |
| 270 |
VDUP(vector2, , float, f, 32, 2, 1.0); |
| 271 |
TEST_VCOMP(INSN_NAME, , float, f, uint, 32, 2); |
| 272 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_nan2, " FP special (NaN)"); |
| 273 |
|
| 274 |
VDUP(vector, , float, f, 32, 2, 1.0); |
| 275 |
VDUP(vector2, , float, f, 32, 2, HUGE_VALF); |
| 276 |
TEST_VCOMP(INSN_NAME, , float, f, uint, 32, 2); |
| 277 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_inf, " FP special (inf)"); |
| 278 |
|
| 279 |
VDUP(vector, , float, f, 32, 2, 1.0); |
| 280 |
VDUP(vector2, , float, f, 32, 2, -HUGE_VALF); |
| 281 |
TEST_VCOMP(INSN_NAME, , float, f, uint, 32, 2); |
| 282 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_minf, " FP special (-inf)"); |
| 283 |
|
| 284 |
VDUP(vector, , float, f, 32, 2, HUGE_VALF); |
| 285 |
VDUP(vector2, , float, f, 32, 2, 1.0); |
| 286 |
TEST_VCOMP(INSN_NAME, , float, f, uint, 32, 2); |
| 287 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_inf2, " FP special (inf)"); |
| 288 |
|
| 289 |
VDUP(vector, , float, f, 32, 2, -0.0); |
| 290 |
VDUP(vector2, , float, f, 32, 2, 0.0); |
| 291 |
TEST_VCOMP(INSN_NAME, , float, f, uint, 32, 2); |
| 292 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_mzero, " FP special (-0.0)"); |
| 293 |
|
| 294 |
#ifdef EXTRA_TESTS |
| 295 |
EXTRA_TESTS(); |
| 296 |
#endif |
| 297 |
} |
| 298 |
|
| 299 |
int main (void) |
| 300 |
{ |
| 301 |
FNNAME (INSN_NAME) (); |
| 302 |
|
| 303 |
return 0; |
| 304 |
} |