Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 1 | /* |
| 2 | |
| 3 | Copyright (c) 2009, 2010, 2011 STMicroelectronics |
| 4 | Written by Christophe Lyon |
| 5 | |
| 6 | Permission is hereby granted, free of charge, to any person obtaining a copy |
| 7 | of this software and associated documentation files (the "Software"), to deal |
| 8 | in the Software without restriction, including without limitation the rights |
| 9 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| 10 | copies of the Software, and to permit persons to whom the Software is |
| 11 | furnished to do so, subject to the following conditions: |
| 12 | |
| 13 | The above copyright notice and this permission notice shall be included in |
| 14 | all copies or substantial portions of the Software. |
| 15 | |
| 16 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 17 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 18 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 19 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 20 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 21 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
| 22 | THE SOFTWARE. |
| 23 | |
| 24 | */ |
| 25 | |
Christophe Lyon | 1775be0 | 2014-07-10 13:46:54 +0200 | [diff] [blame] | 26 | #if defined(__arm__) || defined(__aarch64__) |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 27 | #include <arm_neon.h> |
| 28 | #else |
Christophe Lyon | 0dab5f7 | 2011-07-19 17:14:09 +0200 | [diff] [blame] | 29 | #include "stm-arm-neon.h" |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 30 | #endif |
| 31 | |
| 32 | #include "stm-arm-neon-ref.h" |
| 33 | |
| 34 | #define INSN vqshl |
| 35 | #define TEST_MSG "VQSHL/VQSHLQ" |
| 36 | |
| 37 | #define FNNAME1(NAME) void exec_ ## NAME (void) |
| 38 | #define FNNAME(NAME) FNNAME1(NAME) |
| 39 | |
| 40 | FNNAME (INSN) |
| 41 | { |
| 42 | /* Basic test: v3=vqshl(v1,v2), then store the result. */ |
Christophe Lyon | c1cc782 | 2015-01-20 16:04:24 +0100 | [diff] [blame] | 43 | #define TEST_VQSHL2(INSN, T3, Q, T1, T2, W, N) \ |
| 44 | Set_Neon_Cumulative_Sat(0, VECT_VAR(vector_res, T1, W, N)); \ |
| 45 | VECT_VAR(vector_res, T1, W, N) = \ |
| 46 | INSN##Q##_##T2##W(VECT_VAR(vector, T1, W, N), \ |
| 47 | VECT_VAR(vector_shift, T3, W, N)); \ |
| 48 | vst1##Q##_##T2##W(VECT_VAR(result, T1, W, N), \ |
| 49 | VECT_VAR(vector_res, T1, W, N)); \ |
| 50 | dump_neon_cumulative_sat(TEST_MSG, xSTR(INSN##Q##_##T2##W), \ |
| 51 | xSTR(T1), W, N) |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 52 | |
| 53 | /* Two auxliary macros are necessary to expand INSN */ |
| 54 | #define TEST_VQSHL1(INSN, T3, Q, T1, T2, W, N) \ |
| 55 | TEST_VQSHL2(INSN, T3, Q, T1, T2, W, N) |
| 56 | |
| 57 | #define TEST_VQSHL(T3, Q, T1, T2, W, N) \ |
| 58 | TEST_VQSHL1(INSN, T3, Q, T1, T2, W, N) |
| 59 | |
| 60 | |
| 61 | /* With ARM RVCT, we need to declare variables before any executable |
| 62 | statement */ |
| 63 | DECL_VARIABLE_ALL_VARIANTS(vector); |
| 64 | DECL_VARIABLE_ALL_VARIANTS(vector_res); |
| 65 | |
| 66 | DECL_VARIABLE_SIGNED_VARIANTS(vector_shift); |
| 67 | |
| 68 | clean_results (); |
| 69 | |
| 70 | /* Fill input vector with 0, to check saturation on limits */ |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 71 | VDUP(vector, , int, s, 8, 8, 0); |
| 72 | VDUP(vector, , int, s, 16, 4, 0); |
| 73 | VDUP(vector, , int, s, 32, 2, 0); |
| 74 | VDUP(vector, , int, s, 64, 1, 0); |
| 75 | VDUP(vector, , uint, u, 8, 8, 0); |
| 76 | VDUP(vector, , uint, u, 16, 4, 0); |
| 77 | VDUP(vector, , uint, u, 32, 2, 0); |
| 78 | VDUP(vector, , uint, u, 64, 1, 0); |
| 79 | VDUP(vector, q, int, s, 8, 16, 0); |
| 80 | VDUP(vector, q, int, s, 16, 8, 0); |
| 81 | VDUP(vector, q, int, s, 32, 4, 0); |
| 82 | VDUP(vector, q, int, s, 64, 2, 0); |
| 83 | VDUP(vector, q, uint, u, 8, 16, 0); |
| 84 | VDUP(vector, q, uint, u, 16, 8, 0); |
| 85 | VDUP(vector, q, uint, u, 32, 4, 0); |
| 86 | VDUP(vector, q, uint, u, 64, 2, 0); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 87 | |
| 88 | /* Choose init value arbitrarily, will be used as shift amount */ |
| 89 | /* Use values equal or one-less-than the type width to check |
| 90 | behaviour on limits */ |
| 91 | /* Shift all lanes by 7 ... */ |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 92 | VDUP(vector_shift, , int, s, 8, 8, 7); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 93 | /* except: lane 0 (by 6), lane 1 (by 8) and lane 2 (by 9) */ |
| 94 | TEST_VSET_LANE(vector_shift, , int, s, 8, 8, 0, 6); |
| 95 | TEST_VSET_LANE(vector_shift, , int, s, 8, 8, 1, 8); |
| 96 | TEST_VSET_LANE(vector_shift, , int, s, 8, 8, 2, 9); |
| 97 | |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 98 | VDUP(vector_shift, , int, s, 16, 4, 15); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 99 | TEST_VSET_LANE(vector_shift, , int, s, 16, 4, 0, 14); |
| 100 | TEST_VSET_LANE(vector_shift, , int, s, 16, 4, 1, 16); |
| 101 | TEST_VSET_LANE(vector_shift, , int, s, 16, 4, 2, 17); |
| 102 | |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 103 | VDUP(vector_shift, , int, s, 32, 2, 31); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 104 | TEST_VSET_LANE(vector_shift, , int, s, 32, 2, 1, 30); |
| 105 | |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 106 | VDUP(vector_shift, , int, s, 64, 1, 63); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 107 | |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 108 | VDUP(vector_shift, q, int, s, 8, 16, 8); |
| 109 | VDUP(vector_shift, q, int, s, 16, 8, 16); |
| 110 | VDUP(vector_shift, q, int, s, 32, 4, 32); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 111 | TEST_VSET_LANE(vector_shift, q, int, s, 32, 4, 1, 33); |
| 112 | |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 113 | VDUP(vector_shift, q, int, s, 64, 2, 64); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 114 | TEST_VSET_LANE(vector_shift, q, int, s, 64, 2, 1, 62); |
| 115 | |
Christophe Lyon | 4a6e5cc | 2014-06-03 22:47:52 +0200 | [diff] [blame] | 116 | fprintf(ref_file, "\n%s cumulative saturation output:\n", |
| 117 | TEST_MSG " (with input = 0)"); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 118 | TEST_MACRO_ALL_VARIANTS_1_5(TEST_VQSHL, int); |
| 119 | dump_results_hex2 (TEST_MSG, " (with input = 0)"); |
| 120 | |
| 121 | |
| 122 | /* Use negative shift amounts */ |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 123 | VDUP(vector_shift, , int, s, 8, 8, -1); |
| 124 | VDUP(vector_shift, , int, s, 16, 4, -2); |
| 125 | VDUP(vector_shift, , int, s, 32, 2, -3); |
| 126 | VDUP(vector_shift, , int, s, 64, 1, -4); |
| 127 | VDUP(vector_shift, q, int, s, 8, 16, -7); |
| 128 | VDUP(vector_shift, q, int, s, 16, 8, -11); |
| 129 | VDUP(vector_shift, q, int, s, 32, 4, -13); |
| 130 | VDUP(vector_shift, q, int, s, 64, 2, -20); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 131 | |
Christophe Lyon | 4a6e5cc | 2014-06-03 22:47:52 +0200 | [diff] [blame] | 132 | fprintf(ref_file, "\n%s cumulative saturation output:\n", |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 133 | TEST_MSG " (input 0 and negative shift amount)"); |
| 134 | TEST_MACRO_ALL_VARIANTS_1_5(TEST_VQSHL, int); |
| 135 | dump_results_hex2 (TEST_MSG, " (input 0 and negative shift amount)"); |
| 136 | |
| 137 | /* Test again, with predefined input values */ |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 138 | TEST_MACRO_ALL_VARIANTS_2_5(VLOAD, vector, buffer); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 139 | |
| 140 | /* Choose init value arbitrarily, will be used as shift amount */ |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 141 | VDUP(vector_shift, , int, s, 8, 8, 1); |
| 142 | VDUP(vector_shift, , int, s, 16, 4, 3); |
| 143 | VDUP(vector_shift, , int, s, 32, 2, 8); |
| 144 | VDUP(vector_shift, , int, s, 64, 1, -3); |
| 145 | VDUP(vector_shift, q, int, s, 8, 16, 10); |
| 146 | VDUP(vector_shift, q, int, s, 16, 8, 12); |
| 147 | VDUP(vector_shift, q, int, s, 32, 4, 32); |
| 148 | VDUP(vector_shift, q, int, s, 64, 2, 63); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 149 | |
Christophe Lyon | 4a6e5cc | 2014-06-03 22:47:52 +0200 | [diff] [blame] | 150 | fprintf(ref_file, "\n%s cumulative saturation output:\n", TEST_MSG); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 151 | TEST_MACRO_ALL_VARIANTS_1_5(TEST_VQSHL, int); |
| 152 | dump_results_hex (TEST_MSG); |
| 153 | |
| 154 | /* Use negative shift amounts */ |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 155 | VDUP(vector_shift, , int, s, 8, 8, -1); |
| 156 | VDUP(vector_shift, , int, s, 16, 4, -2); |
| 157 | VDUP(vector_shift, , int, s, 32, 2, -3); |
| 158 | VDUP(vector_shift, , int, s, 64, 1, -4); |
| 159 | VDUP(vector_shift, q, int, s, 8, 16, -7); |
| 160 | VDUP(vector_shift, q, int, s, 16, 8, -11); |
| 161 | VDUP(vector_shift, q, int, s, 32, 4, -13); |
| 162 | VDUP(vector_shift, q, int, s, 64, 2, -20); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 163 | |
Christophe Lyon | 4a6e5cc | 2014-06-03 22:47:52 +0200 | [diff] [blame] | 164 | fprintf(ref_file, "\n%s cumulative saturation output:\n", |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 165 | TEST_MSG " (negative shift amount)"); |
| 166 | TEST_MACRO_ALL_VARIANTS_1_5(TEST_VQSHL, int); |
| 167 | dump_results_hex2 (TEST_MSG, " (negative shift amount)"); |
| 168 | |
| 169 | /* Use large shift amounts */ |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 170 | VDUP(vector_shift, , int, s, 8, 8, 8); |
| 171 | VDUP(vector_shift, , int, s, 16, 4, 16); |
| 172 | VDUP(vector_shift, , int, s, 32, 2, 32); |
| 173 | VDUP(vector_shift, , int, s, 64, 1, 64); |
| 174 | VDUP(vector_shift, q, int, s, 8, 16, 8); |
| 175 | VDUP(vector_shift, q, int, s, 16, 8, 16); |
| 176 | VDUP(vector_shift, q, int, s, 32, 4, 32); |
| 177 | VDUP(vector_shift, q, int, s, 64, 2, 64); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 178 | |
Christophe Lyon | 4a6e5cc | 2014-06-03 22:47:52 +0200 | [diff] [blame] | 179 | fprintf(ref_file, "\n%s cumulative saturation output:\n", |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 180 | TEST_MSG " (large shift amount, negative input)"); |
| 181 | TEST_MACRO_ALL_VARIANTS_1_5(TEST_VQSHL, int); |
| 182 | dump_results_hex2 (TEST_MSG, " (large shift amount, negative input)"); |
| 183 | |
| 184 | /* Fill input vector with max value, to check saturation on limits */ |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 185 | VDUP(vector, , int, s, 8, 8, 0x7F); |
| 186 | VDUP(vector, , int, s, 16, 4, 0x7FFF); |
| 187 | VDUP(vector, , int, s, 32, 2, 0x7FFFFFFF); |
| 188 | VDUP(vector, , int, s, 64, 1, 0x7FFFFFFFFFFFFFFFLL); |
| 189 | VDUP(vector, , uint, u, 8, 8, 0xFF); |
| 190 | VDUP(vector, , uint, u, 16, 4, 0xFFFF); |
| 191 | VDUP(vector, , uint, u, 32, 2, 0xFFFFFFFF); |
| 192 | VDUP(vector, , uint, u, 64, 1, 0xFFFFFFFFFFFFFFFFULL); |
| 193 | VDUP(vector, q, int, s, 8, 16, 0x7F); |
| 194 | VDUP(vector, q, int, s, 16, 8, 0x7FFF); |
| 195 | VDUP(vector, q, int, s, 32, 4, 0x7FFFFFFF); |
| 196 | VDUP(vector, q, int, s, 64, 2, 0x7FFFFFFFFFFFFFFFLL); |
| 197 | VDUP(vector, q, uint, u, 8, 16, 0xFF); |
| 198 | VDUP(vector, q, uint, u, 16, 8, 0xFFFF); |
| 199 | VDUP(vector, q, uint, u, 32, 4, 0xFFFFFFFF); |
| 200 | VDUP(vector, q, uint, u, 64, 2, 0xFFFFFFFFFFFFFFFFULL); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 201 | |
| 202 | /* Shift by -1 */ |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 203 | VDUP(vector_shift, , int, s, 8, 8, -1); |
| 204 | VDUP(vector_shift, , int, s, 16, 4, -1); |
| 205 | VDUP(vector_shift, , int, s, 32, 2, -1); |
| 206 | VDUP(vector_shift, , int, s, 64, 1, -1); |
| 207 | VDUP(vector_shift, q, int, s, 8, 16, -1); |
| 208 | VDUP(vector_shift, q, int, s, 16, 8, -1); |
| 209 | VDUP(vector_shift, q, int, s, 32, 4, -1); |
| 210 | VDUP(vector_shift, q, int, s, 64, 2, -1); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 211 | |
Christophe Lyon | 4a6e5cc | 2014-06-03 22:47:52 +0200 | [diff] [blame] | 212 | fprintf(ref_file, "\n%s cumulative saturation output:\n", |
| 213 | TEST_MSG " (check cumulative saturation)"); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 214 | TEST_MACRO_ALL_VARIANTS_1_5(TEST_VQSHL, int); |
Christophe Lyon | 4a6e5cc | 2014-06-03 22:47:52 +0200 | [diff] [blame] | 215 | dump_results_hex2 (TEST_MSG, " (check cumulative saturation)"); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 216 | |
| 217 | /* Use large shift amounts */ |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 218 | VDUP(vector_shift, , int, s, 8, 8, 8); |
| 219 | VDUP(vector_shift, , int, s, 16, 4, 16); |
| 220 | VDUP(vector_shift, , int, s, 32, 2, 32); |
| 221 | VDUP(vector_shift, , int, s, 64, 1, 64); |
| 222 | VDUP(vector_shift, q, int, s, 8, 16, 8); |
| 223 | VDUP(vector_shift, q, int, s, 16, 8, 16); |
| 224 | VDUP(vector_shift, q, int, s, 32, 4, 32); |
| 225 | VDUP(vector_shift, q, int, s, 64, 2, 64); |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 226 | |
Christophe Lyon | 4a6e5cc | 2014-06-03 22:47:52 +0200 | [diff] [blame] | 227 | fprintf(ref_file, "\n%s cumulative saturation output:\n", |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 228 | TEST_MSG " (large shift amount, positive input)"); |
| 229 | TEST_MACRO_ALL_VARIANTS_1_5(TEST_VQSHL, int); |
| 230 | dump_results_hex2 (TEST_MSG, " (large shift amount, positive input)"); |
| 231 | |
| 232 | /* Check 64 bits saturation */ |
Christophe Lyon | f205367 | 2014-12-16 10:26:00 +0100 | [diff] [blame] | 233 | VDUP(vector, , int, s, 64, 1, -10); |
| 234 | VDUP(vector_shift, , int, s, 64, 1, 64); |
| 235 | VDUP(vector, q, int, s, 64, 2, 10); |
| 236 | VDUP(vector_shift, q, int, s, 64, 2, 64); |
Christophe Lyon | 4a6e5cc | 2014-06-03 22:47:52 +0200 | [diff] [blame] | 237 | fprintf(ref_file, "\n%s cumulative saturation output:\n", |
Christophe Lyon | 073831a | 2011-01-24 17:37:40 +0100 | [diff] [blame] | 238 | TEST_MSG " (check saturation on 64 bits)"); |
| 239 | TEST_MACRO_ALL_VARIANTS_1_5(TEST_VQSHL, int); |
| 240 | dump_results_hex2 (TEST_MSG, " (check saturation on 64 bits)"); |
| 241 | } |