| 1 |
#include <arm_neon.h> |
| 2 |
#include "arm-neon-ref.h" |
| 3 |
#include "compute-ref-data.h" |
| 4 |
#include <math.h> |
| 5 |
|
| 6 |
/* Expected results with positive input. */ |
| 7 |
VECT_VAR_DECL(expected_positive,uint,32,2) [] = { 0xffffffff, 0xffffffff }; |
| 8 |
VECT_VAR_DECL(expected_positive,uint,32,4) [] = { 0xbf000000, 0xbf000000, |
| 9 |
0xbf000000, 0xbf000000 }; |
| 10 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 11 |
VECT_VAR_DECL(expected_positive, hfloat, 16, 4) [] = { 0x3834, 0x3834, |
| 12 |
0x3834, 0x3834 }; |
| 13 |
VECT_VAR_DECL(expected_positive, hfloat, 16, 8) [] = { 0x2018, 0x2018, |
| 14 |
0x2018, 0x2018, |
| 15 |
0x2018, 0x2018, |
| 16 |
0x2018, 0x2018 }; |
| 17 |
#endif |
| 18 |
VECT_VAR_DECL(expected_positive,hfloat,32,2) [] = { 0x3f068000, 0x3f068000 }; |
| 19 |
VECT_VAR_DECL(expected_positive,hfloat,32,4) [] = { 0x3c030000, 0x3c030000, |
| 20 |
0x3c030000, 0x3c030000 }; |
| 21 |
|
| 22 |
/* Expected results with negative input. */ |
| 23 |
VECT_VAR_DECL(expected_negative,uint,32,2) [] = { 0x80000000, 0x80000000 }; |
| 24 |
VECT_VAR_DECL(expected_negative,uint,32,4) [] = { 0xee800000, 0xee800000, |
| 25 |
0xee800000, 0xee800000 }; |
| 26 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 27 |
VECT_VAR_DECL(expected_negative, hfloat, 16, 4) [] = { 0xae64, 0xae64, |
| 28 |
0xae64, 0xae64 }; |
| 29 |
VECT_VAR_DECL(expected_negative, hfloat, 16, 8) [] = { 0xa018, 0xa018, |
| 30 |
0xa018, 0xa018, |
| 31 |
0xa018, 0xa018, |
| 32 |
0xa018, 0xa018 }; |
| 33 |
#endif |
| 34 |
VECT_VAR_DECL(expected_negative,hfloat,32,2) [] = { 0xbdcc8000, 0xbdcc8000 }; |
| 35 |
VECT_VAR_DECL(expected_negative,hfloat,32,4) [] = { 0xbc030000, 0xbc030000, |
| 36 |
0xbc030000, 0xbc030000 }; |
| 37 |
|
| 38 |
/* Expected results with FP special values (NaN, infinity). */ |
| 39 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 40 |
VECT_VAR_DECL(expected_fp1, hfloat, 16, 4) [] = { 0x7e00, 0x7e00, |
| 41 |
0x7e00, 0x7e00 }; |
| 42 |
VECT_VAR_DECL(expected_fp1, hfloat, 16, 8) [] = { 0x0, 0x0, 0x0, 0x0, |
| 43 |
0x0, 0x0, 0x0, 0x0 }; |
| 44 |
#endif |
| 45 |
VECT_VAR_DECL(expected_fp1,hfloat,32,2) [] = { 0x7fc00000, 0x7fc00000 }; |
| 46 |
VECT_VAR_DECL(expected_fp1,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 }; |
| 47 |
|
| 48 |
/* Expected results with FP special values (zero, large value). */ |
| 49 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 50 |
VECT_VAR_DECL(expected_fp2, hfloat, 16, 4) [] = { 0x7c00, 0x7c00, |
| 51 |
0x7c00, 0x7c00 }; |
| 52 |
VECT_VAR_DECL(expected_fp2, hfloat, 16, 8) [] = { 0x0, 0x0, 0x0, 0x0, |
| 53 |
0x0, 0x0, 0x0, 0x0 }; |
| 54 |
#endif |
| 55 |
VECT_VAR_DECL(expected_fp2,hfloat,32,2) [] = { 0x7f800000, 0x7f800000 }; |
| 56 |
VECT_VAR_DECL(expected_fp2,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 }; |
| 57 |
|
| 58 |
/* Expected results with FP special values (-0, -infinity). */ |
| 59 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 60 |
VECT_VAR_DECL(expected_fp3, hfloat, 16, 4) [] = { 0xfc00, 0xfc00, |
| 61 |
0xfc00, 0xfc00}; |
| 62 |
VECT_VAR_DECL(expected_fp3, hfloat, 16, 8) [] = { 0x8000, 0x8000, |
| 63 |
0x8000, 0x8000, |
| 64 |
0x8000, 0x8000, |
| 65 |
0x8000, 0x8000 }; |
| 66 |
#endif |
| 67 |
VECT_VAR_DECL(expected_fp3,hfloat,32,2) [] = { 0xff800000, 0xff800000 }; |
| 68 |
VECT_VAR_DECL(expected_fp3,hfloat,32,4) [] = { 0x80000000, 0x80000000, |
| 69 |
0x80000000, 0x80000000 }; |
| 70 |
|
| 71 |
/* Expected results with FP special large negative value. */ |
| 72 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 73 |
VECT_VAR_DECL(expected_fp4, hfloat, 16, 4) [] = { 0x8000, 0x8000, |
| 74 |
0x8000, 0x8000 }; |
| 75 |
#endif |
| 76 |
VECT_VAR_DECL(expected_fp4,hfloat,32,2) [] = { 0x80000000, 0x80000000 }; |
| 77 |
|
| 78 |
#define TEST_MSG "VRECPE/VRECPEQ" |
| 79 |
void exec_vrecpe(void) |
| 80 |
{ |
| 81 |
int i; |
| 82 |
|
| 83 |
/* Basic test: y=vrecpe(x), then store the result. */ |
| 84 |
#define TEST_VRECPE(Q, T1, T2, W, N) \ |
| 85 |
VECT_VAR(vector_res, T1, W, N) = \ |
| 86 |
vrecpe##Q##_##T2##W(VECT_VAR(vector, T1, W, N)); \ |
| 87 |
vst1##Q##_##T2##W(VECT_VAR(result, T1, W, N), \ |
| 88 |
VECT_VAR(vector_res, T1, W, N)) |
| 89 |
|
| 90 |
/* No need for 64 bits variants. */ |
| 91 |
DECL_VARIABLE(vector, uint, 32, 2); |
| 92 |
DECL_VARIABLE(vector, uint, 32, 4); |
| 93 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 94 |
DECL_VARIABLE(vector, float, 16, 4); |
| 95 |
DECL_VARIABLE(vector, float, 16, 8); |
| 96 |
#endif |
| 97 |
DECL_VARIABLE(vector, float, 32, 2); |
| 98 |
DECL_VARIABLE(vector, float, 32, 4); |
| 99 |
|
| 100 |
DECL_VARIABLE(vector_res, uint, 32, 2); |
| 101 |
DECL_VARIABLE(vector_res, uint, 32, 4); |
| 102 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 103 |
DECL_VARIABLE(vector_res, float, 16, 4); |
| 104 |
DECL_VARIABLE(vector_res, float, 16, 8); |
| 105 |
#endif |
| 106 |
DECL_VARIABLE(vector_res, float, 32, 2); |
| 107 |
DECL_VARIABLE(vector_res, float, 32, 4); |
| 108 |
|
| 109 |
clean_results (); |
| 110 |
|
| 111 |
/* Choose init value arbitrarily, positive. */ |
| 112 |
VDUP(vector, , uint, u, 32, 2, 0x12345678); |
| 113 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 114 |
VDUP(vector, , float, f, 16, 4, 1.9f); |
| 115 |
#endif |
| 116 |
VDUP(vector, , float, f, 32, 2, 1.9f); |
| 117 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 118 |
VDUP(vector, q, float, f, 16, 8, 125.0f); |
| 119 |
#endif |
| 120 |
VDUP(vector, q, uint, u, 32, 4, 0xABCDEF10); |
| 121 |
VDUP(vector, q, float, f, 32, 4, 125.0f); |
| 122 |
|
| 123 |
/* Apply the operator. */ |
| 124 |
TEST_VRECPE(, uint, u, 32, 2); |
| 125 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 126 |
TEST_VRECPE(, float, f, 16, 4); |
| 127 |
#endif |
| 128 |
TEST_VRECPE(, float, f, 32, 2); |
| 129 |
TEST_VRECPE(q, uint, u, 32, 4); |
| 130 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 131 |
TEST_VRECPE(q, float, f, 16, 8); |
| 132 |
#endif |
| 133 |
TEST_VRECPE(q, float, f, 32, 4); |
| 134 |
|
| 135 |
#define CMT " (positive input)" |
| 136 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_positive, CMT); |
| 137 |
CHECK(TEST_MSG, uint, 32, 4, PRIx32, expected_positive, CMT); |
| 138 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 139 |
CHECK_FP(TEST_MSG, float, 16, 4, PRIx16, expected_positive, CMT); |
| 140 |
CHECK_FP(TEST_MSG, float, 16, 8, PRIx16, expected_positive, CMT); |
| 141 |
#endif |
| 142 |
CHECK_FP(TEST_MSG, float, 32, 2, PRIx32, expected_positive, CMT); |
| 143 |
CHECK_FP(TEST_MSG, float, 32, 4, PRIx32, expected_positive, CMT); |
| 144 |
|
| 145 |
/* Choose init value arbitrarily,negative. */ |
| 146 |
VDUP(vector, , uint, u, 32, 2, 0xFFFFFFFF); |
| 147 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 148 |
VDUP(vector, , float, f, 16, 4, -10.0f); |
| 149 |
#endif |
| 150 |
VDUP(vector, , float, f, 32, 2, -10.0f); |
| 151 |
VDUP(vector, q, uint, u, 32, 4, 0x89081234); |
| 152 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 153 |
VDUP(vector, q, float, f, 16, 8, -125.0f); |
| 154 |
#endif |
| 155 |
VDUP(vector, q, float, f, 32, 4, -125.0f); |
| 156 |
|
| 157 |
/* Apply the operator. */ |
| 158 |
TEST_VRECPE(, uint, u, 32, 2); |
| 159 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 160 |
TEST_VRECPE(, float, f, 16, 4); |
| 161 |
#endif |
| 162 |
TEST_VRECPE(, float, f, 32, 2); |
| 163 |
TEST_VRECPE(q, uint, u, 32, 4); |
| 164 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 165 |
TEST_VRECPE(q, float, f, 16, 8); |
| 166 |
#endif |
| 167 |
TEST_VRECPE(q, float, f, 32, 4); |
| 168 |
|
| 169 |
#undef CMT |
| 170 |
#define CMT " (negative input)" |
| 171 |
CHECK(TEST_MSG, uint, 32, 2, PRIx32, expected_negative, CMT); |
| 172 |
CHECK(TEST_MSG, uint, 32, 4, PRIx32, expected_negative, CMT); |
| 173 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 174 |
CHECK_FP(TEST_MSG, float, 16, 4, PRIx16, expected_negative, CMT); |
| 175 |
CHECK_FP(TEST_MSG, float, 16, 8, PRIx16, expected_negative, CMT); |
| 176 |
#endif |
| 177 |
CHECK_FP(TEST_MSG, float, 32, 2, PRIx32, expected_negative, CMT); |
| 178 |
CHECK_FP(TEST_MSG, float, 32, 4, PRIx32, expected_negative, CMT); |
| 179 |
|
| 180 |
/* Test FP variants with special input values (NaN, infinity). */ |
| 181 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 182 |
VDUP(vector, , float, f, 16, 4, NAN); |
| 183 |
VDUP(vector, q, float, f, 16, 8, HUGE_VALF); |
| 184 |
#endif |
| 185 |
VDUP(vector, , float, f, 32, 2, NAN); |
| 186 |
VDUP(vector, q, float, f, 32, 4, HUGE_VALF); |
| 187 |
|
| 188 |
/* Apply the operator. */ |
| 189 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 190 |
TEST_VRECPE(, float, f, 16, 4); |
| 191 |
TEST_VRECPE(q, float, f, 16, 8); |
| 192 |
#endif |
| 193 |
TEST_VRECPE(, float, f, 32, 2); |
| 194 |
TEST_VRECPE(q, float, f, 32, 4); |
| 195 |
|
| 196 |
#undef CMT |
| 197 |
#define CMT " FP special (NaN, infinity)" |
| 198 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 199 |
CHECK_FP(TEST_MSG, float, 16, 4, PRIx16, expected_fp1, CMT); |
| 200 |
CHECK_FP(TEST_MSG, float, 16, 8, PRIx16, expected_fp1, CMT); |
| 201 |
#endif |
| 202 |
CHECK_FP(TEST_MSG, float, 32, 2, PRIx32, expected_fp1, CMT); |
| 203 |
CHECK_FP(TEST_MSG, float, 32, 4, PRIx32, expected_fp1, CMT); |
| 204 |
|
| 205 |
/* Test FP variants with special input values (zero, large value). */ |
| 206 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 207 |
VDUP(vector, , float, f, 16, 4, 0.0f); |
| 208 |
VDUP(vector, q, float, f, 16, 8, 8.97229e37f /*9.0e37f*/); |
| 209 |
#endif |
| 210 |
VDUP(vector, , float, f, 32, 2, 0.0f); |
| 211 |
VDUP(vector, q, float, f, 32, 4, 8.97229e37f /*9.0e37f*/); |
| 212 |
|
| 213 |
/* Apply the operator. */ |
| 214 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 215 |
TEST_VRECPE(, float, f, 16, 4); |
| 216 |
TEST_VRECPE(q, float, f, 16, 8); |
| 217 |
#endif |
| 218 |
TEST_VRECPE(, float, f, 32, 2); |
| 219 |
TEST_VRECPE(q, float, f, 32, 4); |
| 220 |
|
| 221 |
#undef CMT |
| 222 |
#define CMT " FP special (zero, large value)" |
| 223 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 224 |
CHECK_FP(TEST_MSG, float, 16, 4, PRIx16, expected_fp2, CMT); |
| 225 |
CHECK_FP(TEST_MSG, float, 16, 8, PRIx16, expected_fp2, CMT); |
| 226 |
#endif |
| 227 |
CHECK_FP(TEST_MSG, float, 32, 2, PRIx32, expected_fp2, CMT); |
| 228 |
CHECK_FP(TEST_MSG, float, 32, 4, PRIx32, expected_fp2, CMT); |
| 229 |
|
| 230 |
/* Test FP variants with special input values (-0, -infinity). */ |
| 231 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 232 |
VDUP(vector, , float, f, 16, 4, -0.0f); |
| 233 |
VDUP(vector, q, float, f, 16, 8, -HUGE_VALF); |
| 234 |
#endif |
| 235 |
VDUP(vector, , float, f, 32, 2, -0.0f); |
| 236 |
VDUP(vector, q, float, f, 32, 4, -HUGE_VALF); |
| 237 |
|
| 238 |
/* Apply the operator. */ |
| 239 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 240 |
TEST_VRECPE(, float, f, 16, 4); |
| 241 |
TEST_VRECPE(q, float, f, 16, 8); |
| 242 |
#endif |
| 243 |
TEST_VRECPE(, float, f, 32, 2); |
| 244 |
TEST_VRECPE(q, float, f, 32, 4); |
| 245 |
|
| 246 |
#undef CMT |
| 247 |
#define CMT " FP special (-0, -infinity)" |
| 248 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 249 |
CHECK_FP(TEST_MSG, float, 16, 4, PRIx16, expected_fp3, CMT); |
| 250 |
CHECK_FP(TEST_MSG, float, 16, 8, PRIx16, expected_fp3, CMT); |
| 251 |
#endif |
| 252 |
CHECK_FP(TEST_MSG, float, 32, 2, PRIx32, expected_fp3, CMT); |
| 253 |
CHECK_FP(TEST_MSG, float, 32, 4, PRIx32, expected_fp3, CMT); |
| 254 |
|
| 255 |
/* Test FP variants with special input values (large negative value). */ |
| 256 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 257 |
VDUP(vector, , float, f, 16, 4, -9.0e37f); |
| 258 |
#endif |
| 259 |
VDUP(vector, , float, f, 32, 2, -9.0e37f); |
| 260 |
|
| 261 |
/* Apply the operator. */ |
| 262 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 263 |
TEST_VRECPE(, float, f, 16, 4); |
| 264 |
#endif |
| 265 |
TEST_VRECPE(, float, f, 32, 2); |
| 266 |
|
| 267 |
#undef CMT |
| 268 |
#define CMT " FP special (large negative value)" |
| 269 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 270 |
CHECK_FP(TEST_MSG, float, 16, 4, PRIx16, expected_fp4, CMT); |
| 271 |
#endif |
| 272 |
CHECK_FP(TEST_MSG, float, 32, 2, PRIx32, expected_fp4, CMT); |
| 273 |
} |
| 274 |
|
| 275 |
int main (void) |
| 276 |
{ |
| 277 |
exec_vrecpe (); |
| 278 |
return 0; |
| 279 |
} |