| 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. */ |
| 7 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 8 |
VECT_VAR_DECL(expected, hfloat, 16, 4) [] = { 0xd3cb, 0xd3cb, 0xd3cb, 0xd3cb }; |
| 9 |
VECT_VAR_DECL(expected, hfloat, 16, 8) [] = { 0xc726, 0xc726, 0xc726, 0xc726, |
| 10 |
0xc726, 0xc726, 0xc726, 0xc726 }; |
| 11 |
#endif |
| 12 |
VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc2796b84, 0xc2796b84 }; |
| 13 |
VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc0e4a3d8, 0xc0e4a3d8, |
| 14 |
0xc0e4a3d8, 0xc0e4a3d8 }; |
| 15 |
|
| 16 |
/* Expected results with input=NaN. */ |
| 17 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 18 |
VECT_VAR_DECL(expected_nan, hfloat, 16, 4) [] = { 0x7e00, 0x7e00, |
| 19 |
0x7e00, 0x7e00 }; |
| 20 |
VECT_VAR_DECL(expected_nan, hfloat, 16, 8) [] = { 0x7e00, 0x7e00, |
| 21 |
0x7e00, 0x7e00, |
| 22 |
0x7e00, 0x7e00, |
| 23 |
0x7e00, 0x7e00 }; |
| 24 |
#endif |
| 25 |
VECT_VAR_DECL(expected_nan,hfloat,32,2) [] = { 0x7fc00000, 0x7fc00000 }; |
| 26 |
VECT_VAR_DECL(expected_nan,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, |
| 27 |
0x7fc00000, 0x7fc00000 }; |
| 28 |
|
| 29 |
/* Expected results with FP special inputs values (infinity, 0). */ |
| 30 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 31 |
VECT_VAR_DECL(expected_fp1, hfloat, 16, 4) [] = { 0xfc00, 0xfc00, |
| 32 |
0xfc00, 0xfc00 }; |
| 33 |
VECT_VAR_DECL(expected_fp1, hfloat, 16, 8) [] = { 0x3e00, 0x3e00, |
| 34 |
0x3e00, 0x3e00, |
| 35 |
0x3e00, 0x3e00, |
| 36 |
0x3e00, 0x3e00 }; |
| 37 |
#endif |
| 38 |
VECT_VAR_DECL(expected_fp1,hfloat,32,2) [] = { 0xff800000, 0xff800000 }; |
| 39 |
VECT_VAR_DECL(expected_fp1,hfloat,32,4) [] = { 0x3fc00000, 0x3fc00000, |
| 40 |
0x3fc00000, 0x3fc00000 }; |
| 41 |
|
| 42 |
/* Expected results with only FP special inputs values (infinity, |
| 43 |
0). */ |
| 44 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 45 |
VECT_VAR_DECL(expected_fp2, hfloat, 16, 4) [] = { 0x3e00, 0x3e00, |
| 46 |
0x3e00, 0x3e00 }; |
| 47 |
VECT_VAR_DECL(expected_fp2, hfloat, 16, 8) [] = { 0x3e00, 0x3e00, |
| 48 |
0x3e00, 0x3e00, |
| 49 |
0x3e00, 0x3e00, |
| 50 |
0x3e00, 0x3e00 }; |
| 51 |
#endif |
| 52 |
VECT_VAR_DECL(expected_fp2,hfloat,32,2) [] = { 0x3fc00000, 0x3fc00000 }; |
| 53 |
VECT_VAR_DECL(expected_fp2,hfloat,32,4) [] = { 0x3fc00000, 0x3fc00000, |
| 54 |
0x3fc00000, 0x3fc00000 }; |
| 55 |
|
| 56 |
#define TEST_MSG "VRSQRTS/VRSQRTSQ" |
| 57 |
void exec_vrsqrts(void) |
| 58 |
{ |
| 59 |
int i; |
| 60 |
|
| 61 |
/* Basic test: y=vrsqrts(x), then store the result. */ |
| 62 |
#define TEST_VRSQRTS(Q, T1, T2, W, N) \ |
| 63 |
VECT_VAR(vector_res, T1, W, N) = \ |
| 64 |
vrsqrts##Q##_##T2##W(VECT_VAR(vector, T1, W, N), \ |
| 65 |
VECT_VAR(vector2, T1, W, N)); \ |
| 66 |
vst1##Q##_##T2##W(VECT_VAR(result, T1, W, N), \ |
| 67 |
VECT_VAR(vector_res, T1, W, N)) |
| 68 |
|
| 69 |
/* No need for integer variants. */ |
| 70 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 71 |
DECL_VARIABLE(vector, float, 16, 4); |
| 72 |
DECL_VARIABLE(vector, float, 16, 8); |
| 73 |
#endif |
| 74 |
DECL_VARIABLE(vector, float, 32, 2); |
| 75 |
DECL_VARIABLE(vector, float, 32, 4); |
| 76 |
|
| 77 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 78 |
DECL_VARIABLE(vector2, float, 16, 4); |
| 79 |
DECL_VARIABLE(vector2, float, 16, 8); |
| 80 |
#endif |
| 81 |
DECL_VARIABLE(vector2, float, 32, 2); |
| 82 |
DECL_VARIABLE(vector2, float, 32, 4); |
| 83 |
|
| 84 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 85 |
DECL_VARIABLE(vector_res, float, 16, 4); |
| 86 |
DECL_VARIABLE(vector_res, float, 16, 8); |
| 87 |
#endif |
| 88 |
DECL_VARIABLE(vector_res, float, 32, 2); |
| 89 |
DECL_VARIABLE(vector_res, float, 32, 4); |
| 90 |
|
| 91 |
clean_results (); |
| 92 |
|
| 93 |
/* Choose init value arbitrarily. */ |
| 94 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 95 |
VDUP(vector, , float, f, 16, 4, 12.9f); |
| 96 |
VDUP(vector, q, float, f, 16, 8, 9.1f); |
| 97 |
#endif |
| 98 |
VDUP(vector, , float, f, 32, 2, 12.9f); |
| 99 |
VDUP(vector, q, float, f, 32, 4, 9.1f); |
| 100 |
|
| 101 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 102 |
VDUP(vector2, , float, f, 16, 4, 9.9f); |
| 103 |
VDUP(vector2, q, float, f, 16, 8, 1.9f); |
| 104 |
#endif |
| 105 |
VDUP(vector2, , float, f, 32, 2, 9.9f); |
| 106 |
VDUP(vector2, q, float, f, 32, 4, 1.9f); |
| 107 |
|
| 108 |
/* Apply the operator. */ |
| 109 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 110 |
TEST_VRSQRTS(, float, f, 16, 4); |
| 111 |
TEST_VRSQRTS(q, float, f, 16, 8); |
| 112 |
#endif |
| 113 |
TEST_VRSQRTS(, float, f, 32, 2); |
| 114 |
TEST_VRSQRTS(q, float, f, 32, 4); |
| 115 |
|
| 116 |
#define CMT "" |
| 117 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 118 |
CHECK_FP(TEST_MSG, float, 16, 4, PRIx16, expected, CMT); |
| 119 |
CHECK_FP(TEST_MSG, float, 16, 8, PRIx16, expected, CMT); |
| 120 |
#endif |
| 121 |
CHECK_FP(TEST_MSG, float, 32, 2, PRIx32, expected, CMT); |
| 122 |
CHECK_FP(TEST_MSG, float, 32, 4, PRIx32, expected, CMT); |
| 123 |
|
| 124 |
|
| 125 |
/* Test FP variants with special input values (NaN). */ |
| 126 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 127 |
VDUP(vector, , float, f, 16, 4, NAN); |
| 128 |
VDUP(vector2, q, float, f, 16, 8, NAN); |
| 129 |
#endif |
| 130 |
VDUP(vector, , float, f, 32, 2, NAN); |
| 131 |
VDUP(vector2, q, float, f, 32, 4, NAN); |
| 132 |
|
| 133 |
/* Apply the operator. */ |
| 134 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 135 |
TEST_VRSQRTS(, float, f, 16, 4); |
| 136 |
TEST_VRSQRTS(q, float, f, 16, 8); |
| 137 |
#endif |
| 138 |
TEST_VRSQRTS(, float, f, 32, 2); |
| 139 |
TEST_VRSQRTS(q, float, f, 32, 4); |
| 140 |
|
| 141 |
#undef CMT |
| 142 |
#define CMT " FP special (NAN) and normal values" |
| 143 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 144 |
CHECK_FP(TEST_MSG, float, 16, 4, PRIx16, expected_nan, CMT); |
| 145 |
CHECK_FP(TEST_MSG, float, 16, 8, PRIx16, expected_nan, CMT); |
| 146 |
#endif |
| 147 |
CHECK_FP(TEST_MSG, float, 32, 2, PRIx32, expected_nan, CMT); |
| 148 |
CHECK_FP(TEST_MSG, float, 32, 4, PRIx32, expected_nan, CMT); |
| 149 |
|
| 150 |
|
| 151 |
/* Test FP variants with special input values (infinity, 0). */ |
| 152 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 153 |
VDUP(vector, , float, f, 16, 4, HUGE_VALF); |
| 154 |
VDUP(vector, q, float, f, 16, 8, 0.0f); |
| 155 |
/* Restore a normal value in vector2. */ |
| 156 |
VDUP(vector2, q, float, f, 16, 8, 3.2f); |
| 157 |
#endif |
| 158 |
VDUP(vector, , float, f, 32, 2, HUGE_VALF); |
| 159 |
VDUP(vector, q, float, f, 32, 4, 0.0f); |
| 160 |
/* Restore a normal value in vector2. */ |
| 161 |
VDUP(vector2, q, float, f, 32, 4, 3.2f); |
| 162 |
|
| 163 |
/* Apply the operator. */ |
| 164 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 165 |
TEST_VRSQRTS(, float, f, 16, 4); |
| 166 |
TEST_VRSQRTS(q, float, f, 16, 8); |
| 167 |
#endif |
| 168 |
TEST_VRSQRTS(, float, f, 32, 2); |
| 169 |
TEST_VRSQRTS(q, float, f, 32, 4); |
| 170 |
|
| 171 |
#undef CMT |
| 172 |
#define CMT " FP special (infinity, 0) and normal values" |
| 173 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 174 |
CHECK_FP(TEST_MSG, float, 16, 4, PRIx16, expected_fp1, CMT); |
| 175 |
CHECK_FP(TEST_MSG, float, 16, 8, PRIx16, expected_fp1, CMT); |
| 176 |
#endif |
| 177 |
CHECK_FP(TEST_MSG, float, 32, 2, PRIx32, expected_fp1, CMT); |
| 178 |
CHECK_FP(TEST_MSG, float, 32, 4, PRIx32, expected_fp1, CMT); |
| 179 |
|
| 180 |
|
| 181 |
/* Test FP variants with only special input values (infinity, 0). */ |
| 182 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 183 |
VDUP(vector, , float, f, 16, 4, HUGE_VALF); |
| 184 |
VDUP(vector, q, float, f, 16, 8, 0.0f); |
| 185 |
VDUP(vector2, , float, f, 16, 4, -0.0f); |
| 186 |
VDUP(vector2, q, float, f, 16, 8, HUGE_VALF); |
| 187 |
#endif |
| 188 |
VDUP(vector, , float, f, 32, 2, HUGE_VALF); |
| 189 |
VDUP(vector, q, float, f, 32, 4, 0.0f); |
| 190 |
VDUP(vector2, , float, f, 32, 2, -0.0f); |
| 191 |
VDUP(vector2, q, float, f, 32, 4, HUGE_VALF); |
| 192 |
|
| 193 |
/* Apply the operator. */ |
| 194 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 195 |
TEST_VRSQRTS(, float, f, 16, 4); |
| 196 |
TEST_VRSQRTS(q, float, f, 16, 8); |
| 197 |
#endif |
| 198 |
TEST_VRSQRTS(, float, f, 32, 2); |
| 199 |
TEST_VRSQRTS(q, float, f, 32, 4); |
| 200 |
|
| 201 |
#undef CMT |
| 202 |
#define CMT " only FP special (infinity, 0)" |
| 203 |
#if defined (__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) |
| 204 |
CHECK_FP(TEST_MSG, float, 16, 4, PRIx16, expected_fp2, CMT); |
| 205 |
CHECK_FP(TEST_MSG, float, 16, 8, PRIx16, expected_fp2, CMT); |
| 206 |
#endif |
| 207 |
CHECK_FP(TEST_MSG, float, 32, 2, PRIx32, expected_fp2, CMT); |
| 208 |
CHECK_FP(TEST_MSG, float, 32, 4, PRIx32, expected_fp2, CMT); |
| 209 |
} |
| 210 |
|
| 211 |
int main (void) |
| 212 |
{ |
| 213 |
exec_vrsqrts (); |
| 214 |
return 0; |
| 215 |
} |