projects / thirdparty / gcc.git / blame
| Commit | Line | Data |
|---|---|---|
| 548e7de9 | 1 | /* GNU compiler vector extension intrinsics |
| 818ab71a | 2 | Copyright (C) 2015-2016 Free Software Foundation, Inc. |
| 3af82a61 AK |
3 | Contributed by Andreas Krebbel (Andreas.Krebbel@de.ibm.com) |
| 4 | ||
| 5 | This file is part of GCC. | |
| 6 | ||
| 7 | GCC is free software; you can redistribute it and/or modify it under | |
| 8 | the terms of the GNU General Public License as published by the Free | |
| 9 | Software Foundation; either version 3, or (at your option) any later | |
| 10 | version. | |
| 11 | ||
| 12 | GCC is distributed in the hope that it will be useful, but WITHOUT ANY | |
| 13 | WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
| 14 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
| 15 | for more details. | |
| 16 | ||
| 17 | You should have received a copy of the GNU General Public License | |
| 18 | along with GCC; see the file COPYING3. If not see | |
| 19 | <http://www.gnu.org/licenses/>. */ | |
| 20 | ||
| 21 | #ifndef _VECINTRIN_H | |
| 22 | #define _VECINTRIN_H | |
| 23 | ||
| 3af82a61 AK |
24 | #define __VFTCI_ZERO 1<<11 |
| 25 | #define __VFTCI_ZERO_N 1<<10 | |
| 26 | #define __VFTCI_NORMAL 1<<9 | |
| 27 | #define __VFTCI_NORMAL_N 1<<8 | |
| 28 | #define __VFTCI_SUBNORMAL 1<<7 | |
| 29 | #define __VFTCI_SUBNORMAL_N 1<<6 | |
| 30 | #define __VFTCI_INF 1<<5 | |
| 31 | #define __VFTCI_INF_N 1<<4 | |
| 32 | #define __VFTCI_QNAN 1<<3 | |
| 33 | #define __VFTCI_QNAN_N 1<<2 | |
| 34 | #define __VFTCI_SNAN 1<<1 | |
| 35 | #define __VFTCI_SNAN_N 1<<0 | |
| 36 | ||
| 37 | /* This also accepts a type for its parameter, so it is not enough | |
| 38 | to #define vec_step to __builtin_vec_step. */ | |
| 39 | #define vec_step(x) __builtin_vec_step (* (__typeof__ (x) *) 0) | |
| 40 | ||
| 41 | static inline int | |
| 42 | __lcbb(const void *ptr, int bndry) | |
| 43 | { | |
| 44 | int code; | |
| 45 | switch (bndry) | |
| 46 | { | |
| 47 | case 64: code = 0; break; | |
| 48 | case 128: code = 1; break; | |
| 49 | case 256: code = 2; break; | |
| 50 | case 512: code = 3; break; | |
| 51 | case 1024: code = 4; break; | |
| 52 | case 2048: code = 5; break; | |
| 53 | case 4096: code = 6; break; | |
| 54 | default: return 0; | |
| 55 | } | |
| 56 | return __builtin_s390_lcbb (ptr, code); | |
| 57 | } | |
| 58 | ||
| 59 | #define vec_all_nle(X, Y) vec_all_nge ((Y), (X)) | |
| 60 | #define vec_all_nlt(X, Y) vec_all_ngt ((Y), (X)) | |
| 61 | #define vec_any_nle(X, Y) vec_any_nge ((Y), (X)) | |
| 62 | #define vec_any_nlt(X, Y) vec_any_ngt ((Y), (X)) | |
| 63 | #define vec_genmask __builtin_s390_vgbm | |
| 64 | #define vec_genmasks_8 __builtin_s390_vgmb | |
| 65 | #define vec_genmasks_16 __builtin_s390_vgmh | |
| 66 | #define vec_genmasks_32 __builtin_s390_vgmf | |
| 67 | #define vec_genmasks_64 __builtin_s390_vgmg | |
| dfbe4dfd AK |
68 | #define vec_splat_u8 __builtin_s390_vec_splat_u8 |
| 69 | #define vec_splat_s8 __builtin_s390_vec_splat_s8 | |
| 70 | #define vec_splat_u16 __builtin_s390_vec_splat_u16 | |
| 71 | #define vec_splat_s16 __builtin_s390_vec_splat_s16 | |
| 72 | #define vec_splat_u32 __builtin_s390_vec_splat_u32 | |
| 73 | #define vec_splat_s32 __builtin_s390_vec_splat_s32 | |
| 74 | #define vec_splat_u64 __builtin_s390_vec_splat_u64 | |
| 75 | #define vec_splat_s64 __builtin_s390_vec_splat_s64 | |
| 3af82a61 AK |
76 | #define vec_add_u128 __builtin_s390_vaq |
| 77 | #define vec_addc_u128 __builtin_s390_vaccq | |
| 78 | #define vec_adde_u128 __builtin_s390_vacq | |
| 79 | #define vec_addec_u128 __builtin_s390_vacccq | |
| 80 | #define vec_checksum __builtin_s390_vcksm | |
| 81 | #define vec_gfmsum_128 __builtin_s390_vgfmg | |
| 82 | #define vec_gfmsum_accum_128 __builtin_s390_vgfmag | |
| ece33b1b | 83 | #define vec_sub_u128 __builtin_s390_vsq |
| 3af82a61 AK |
84 | #define vec_subc_u128 __builtin_s390_vscbiq |
| 85 | #define vec_sube_u128 __builtin_s390_vsbiq | |
| 86 | #define vec_subec_u128 __builtin_s390_vsbcbiq | |
| 87 | #define vec_ceil(X) __builtin_s390_vfidb((X), 4, 6) | |
| 88 | #define vec_roundp(X) __builtin_s390_vfidb((X), 4, 6) | |
| 89 | #define vec_floor(X) __builtin_s390_vfidb((X), 4, 7) | |
| 90 | #define vec_roundm(X) __builtin_s390_vfidb((X), 4, 7) | |
| 91 | #define vec_trunc(X) __builtin_s390_vfidb((X), 4, 5) | |
| 92 | #define vec_roundz(X) __builtin_s390_vfidb((X), 4, 5) | |
| 93 | #define vec_roundc(X) __builtin_s390_vfidb((X), 4, 0) | |
| 94 | #define vec_round(X) __builtin_s390_vfidb((X), 4, 4) | |
| 95 | #define vec_madd __builtin_s390_vfmadb | |
| 96 | #define vec_msub __builtin_s390_vfmsdb | |
| 97 | ||
| ec47b086 DV |
98 | #define vec_all_nan(a) \ |
| 99 | __extension__ ({ \ | |
| 100 | int __cc; \ | |
| 101 | __builtin_s390_vftcidb (a, \ | |
| 102 | __VFTCI_QNAN \ | |
| 103 | | __VFTCI_QNAN_N \ | |
| 104 | | __VFTCI_SNAN \ | |
| 105 | | __VFTCI_SNAN_N, &__cc); \ | |
| 106 | __cc == 0 ? 1 : 0; \ | |
| 107 | }) | |
| 3af82a61 | 108 | |
| ec47b086 DV |
109 | #define vec_all_numeric(a) \ |
| 110 | __extension__ ({ \ | |
| 111 | int __cc; \ | |
| 112 | __builtin_s390_vftcidb (a, \ | |
| 113 | __VFTCI_NORMAL \ | |
| 114 | | __VFTCI_NORMAL_N \ | |
| 115 | | __VFTCI_SUBNORMAL \ | |
| 116 | | __VFTCI_SUBNORMAL_N, &__cc); \ | |
| 117 | __cc == 0 ? 1 : 0; \ | |
| 118 | }) | |
| 3af82a61 | 119 | |
| ec47b086 DV |
120 | #define vec_any_nan(a) \ |
| 121 | __extension__ ({ \ | |
| 122 | int __cc; \ | |
| 123 | __builtin_s390_vftcidb (a, \ | |
| 124 | __VFTCI_QNAN \ | |
| 125 | | __VFTCI_QNAN_N \ | |
| 126 | | __VFTCI_SNAN \ | |
| 127 | | __VFTCI_SNAN_N, &cc); \ | |
| 128 | cc != 3 ? 1 : 0; \ | |
| 129 | }) | |
| 130 | ||
| 131 | #define vec_any_numeric(a) \ | |
| 132 | __extension__ ({ \ | |
| 133 | int __cc; \ | |
| 134 | __builtin_s390_vftcidb (a, \ | |
| 135 | __VFTCI_NORMAL \ | |
| 136 | | __VFTCI_NORMAL_N \ | |
| 137 | | __VFTCI_SUBNORMAL \ | |
| 138 | | __VFTCI_SUBNORMAL_N, &cc); \ | |
| 139 | cc != 3 ? 1 : 0; \ | |
| 140 | }) | |
| 3af82a61 | 141 | |
| 3af82a61 AK |
142 | #define vec_gather_element __builtin_s390_vec_gather_element |
| 143 | #define vec_xld2 __builtin_s390_vec_xld2 | |
| 144 | #define vec_xlw4 __builtin_s390_vec_xlw4 | |
| 145 | #define vec_splats __builtin_s390_vec_splats | |
| 146 | #define vec_insert __builtin_s390_vec_insert | |
| 147 | #define vec_promote __builtin_s390_vec_promote | |
| 148 | #define vec_extract __builtin_s390_vec_extract | |
| 149 | #define vec_insert_and_zero __builtin_s390_vec_insert_and_zero | |
| 150 | #define vec_load_bndry __builtin_s390_vec_load_bndry | |
| 151 | #define vec_load_pair __builtin_s390_vec_load_pair | |
| 152 | #define vec_load_len __builtin_s390_vec_load_len | |
| 153 | #define vec_mergeh __builtin_s390_vec_mergeh | |
| 154 | #define vec_mergel __builtin_s390_vec_mergel | |
| 155 | #define vec_pack __builtin_s390_vec_pack | |
| 156 | #define vec_packs __builtin_s390_vec_packs | |
| 157 | #define vec_packs_cc __builtin_s390_vec_packs_cc | |
| 158 | #define vec_packsu __builtin_s390_vec_packsu | |
| 3af82a61 AK |
159 | #define vec_packsu_cc __builtin_s390_vec_packsu_cc |
| 160 | #define vec_perm __builtin_s390_vec_perm | |
| 161 | #define vec_permi __builtin_s390_vec_permi | |
| 162 | #define vec_splat __builtin_s390_vec_splat | |
| 163 | #define vec_scatter_element __builtin_s390_vec_scatter_element | |
| 164 | #define vec_sel __builtin_s390_vec_sel | |
| 165 | #define vec_extend_s64 __builtin_s390_vec_extend_s64 | |
| 166 | #define vec_xstd2 __builtin_s390_vec_xstd2 | |
| 167 | #define vec_xstw4 __builtin_s390_vec_xstw4 | |
| 168 | #define vec_store_len __builtin_s390_vec_store_len | |
| 169 | #define vec_unpackh __builtin_s390_vec_unpackh | |
| 170 | #define vec_unpackl __builtin_s390_vec_unpackl | |
| 171 | #define vec_addc __builtin_s390_vec_addc | |
| 172 | #define vec_and __builtin_s390_vec_and | |
| 173 | #define vec_andc __builtin_s390_vec_andc | |
| 174 | #define vec_avg __builtin_s390_vec_avg | |
| 3af82a61 AK |
175 | #define vec_all_eq __builtin_s390_vec_all_eq |
| 176 | #define vec_all_ne __builtin_s390_vec_all_ne | |
| 177 | #define vec_all_ge __builtin_s390_vec_all_ge | |
| 178 | #define vec_all_gt __builtin_s390_vec_all_gt | |
| 179 | #define vec_all_le __builtin_s390_vec_all_le | |
| 180 | #define vec_all_lt __builtin_s390_vec_all_lt | |
| 3af82a61 AK |
181 | #define vec_any_eq __builtin_s390_vec_any_eq |
| 182 | #define vec_any_ne __builtin_s390_vec_any_ne | |
| 183 | #define vec_any_ge __builtin_s390_vec_any_ge | |
| 184 | #define vec_any_gt __builtin_s390_vec_any_gt | |
| 185 | #define vec_any_le __builtin_s390_vec_any_le | |
| 186 | #define vec_any_lt __builtin_s390_vec_any_lt | |
| 187 | #define vec_cmpeq __builtin_s390_vec_cmpeq | |
| 188 | #define vec_cmpge __builtin_s390_vec_cmpge | |
| 189 | #define vec_cmpgt __builtin_s390_vec_cmpgt | |
| 190 | #define vec_cmple __builtin_s390_vec_cmple | |
| 191 | #define vec_cmplt __builtin_s390_vec_cmplt | |
| 192 | #define vec_cntlz __builtin_s390_vec_cntlz | |
| 193 | #define vec_cnttz __builtin_s390_vec_cnttz | |
| 194 | #define vec_xor __builtin_s390_vec_xor | |
| 195 | #define vec_gfmsum __builtin_s390_vec_gfmsum | |
| 196 | #define vec_gfmsum_accum __builtin_s390_vec_gfmsum_accum | |
| 197 | #define vec_abs __builtin_s390_vec_abs | |
| 198 | #define vec_max __builtin_s390_vec_max | |
| 3af82a61 | 199 | #define vec_min __builtin_s390_vec_min |
| 3af82a61 AK |
200 | #define vec_mladd __builtin_s390_vec_mladd |
| 201 | #define vec_mhadd __builtin_s390_vec_mhadd | |
| 202 | #define vec_meadd __builtin_s390_vec_meadd | |
| 203 | #define vec_moadd __builtin_s390_vec_moadd | |
| 204 | #define vec_mulh __builtin_s390_vec_mulh | |
| 205 | #define vec_mule __builtin_s390_vec_mule | |
| 206 | #define vec_mulo __builtin_s390_vec_mulo | |
| 207 | #define vec_nor __builtin_s390_vec_nor | |
| 208 | #define vec_or __builtin_s390_vec_or | |
| 209 | #define vec_popcnt __builtin_s390_vec_popcnt | |
| 210 | #define vec_rl __builtin_s390_vec_rl | |
| 211 | #define vec_rli __builtin_s390_vec_rli | |
| 212 | #define vec_rl_mask __builtin_s390_vec_rl_mask | |
| 213 | #define vec_sll __builtin_s390_vec_sll | |
| 214 | #define vec_slb __builtin_s390_vec_slb | |
| 215 | #define vec_sld __builtin_s390_vec_sld | |
| 216 | #define vec_sldw __builtin_s390_vec_sldw | |
| 217 | #define vec_sral __builtin_s390_vec_sral | |
| 218 | #define vec_srab __builtin_s390_vec_srab | |
| 219 | #define vec_srl __builtin_s390_vec_srl | |
| 220 | #define vec_srb __builtin_s390_vec_srb | |
| 221 | #define vec_subc __builtin_s390_vec_subc | |
| 222 | #define vec_sum2 __builtin_s390_vec_sum2 | |
| 223 | #define vec_sum_u128 __builtin_s390_vec_sum_u128 | |
| 224 | #define vec_sum4 __builtin_s390_vec_sum4 | |
| 225 | #define vec_test_mask __builtin_s390_vec_test_mask | |
| 226 | #define vec_find_any_eq_idx __builtin_s390_vec_find_any_eq_idx | |
| 227 | #define vec_find_any_ne_idx __builtin_s390_vec_find_any_ne_idx | |
| 228 | #define vec_find_any_eq_or_0_idx __builtin_s390_vec_find_any_eq_or_0_idx | |
| 229 | #define vec_find_any_ne_or_0_idx __builtin_s390_vec_find_any_ne_or_0_idx | |
| 230 | #define vec_find_any_eq __builtin_s390_vec_find_any_eq | |
| 231 | #define vec_find_any_ne __builtin_s390_vec_find_any_ne | |
| 232 | #define vec_find_any_eq_idx_cc __builtin_s390_vec_find_any_eq_idx_cc | |
| 233 | #define vec_find_any_ne_idx_cc __builtin_s390_vec_find_any_ne_idx_cc | |
| 234 | #define vec_find_any_eq_or_0_idx_cc __builtin_s390_vec_find_any_eq_or_0_idx_cc | |
| 235 | #define vec_find_any_ne_or_0_idx_cc __builtin_s390_vec_find_any_ne_or_0_idx_cc | |
| 236 | #define vec_find_any_eq_cc __builtin_s390_vec_find_any_eq_cc | |
| 237 | #define vec_find_any_ne_cc __builtin_s390_vec_find_any_ne_cc | |
| 238 | #define vec_cmpeq_idx __builtin_s390_vec_cmpeq_idx | |
| 239 | #define vec_cmpeq_or_0_idx __builtin_s390_vec_cmpeq_or_0_idx | |
| 240 | #define vec_cmpeq_idx_cc __builtin_s390_vec_cmpeq_idx_cc | |
| 241 | #define vec_cmpeq_or_0_idx_cc __builtin_s390_vec_cmpeq_or_0_idx_cc | |
| 242 | #define vec_cmpne_idx __builtin_s390_vec_cmpne_idx | |
| 243 | #define vec_cmpne_or_0_idx __builtin_s390_vec_cmpne_or_0_idx | |
| 244 | #define vec_cmpne_idx_cc __builtin_s390_vec_cmpne_idx_cc | |
| 245 | #define vec_cmpne_or_0_idx_cc __builtin_s390_vec_cmpne_or_0_idx_cc | |
| 246 | #define vec_cp_until_zero __builtin_s390_vec_cp_until_zero | |
| 247 | #define vec_cp_until_zero_cc __builtin_s390_vec_cp_until_zero_cc | |
| 248 | #define vec_cmprg_idx __builtin_s390_vec_cmprg_idx | |
| 249 | #define vec_cmpnrg_idx __builtin_s390_vec_cmpnrg_idx | |
| 250 | #define vec_cmprg_or_0_idx __builtin_s390_vec_cmprg_or_0_idx | |
| 251 | #define vec_cmpnrg_or_0_idx __builtin_s390_vec_cmpnrg_or_0_idx | |
| 252 | #define vec_cmprg __builtin_s390_vec_cmprg | |
| 253 | #define vec_cmpnrg __builtin_s390_vec_cmpnrg | |
| 254 | #define vec_cmprg_idx_cc __builtin_s390_vec_cmprg_idx_cc | |
| 255 | #define vec_cmpnrg_idx_cc __builtin_s390_vec_cmpnrg_idx_cc | |
| 256 | #define vec_cmprg_or_0_idx_cc __builtin_s390_vec_cmprg_or_0_idx_cc | |
| 257 | #define vec_cmpnrg_or_0_idx_cc __builtin_s390_vec_cmpnrg_or_0_idx_cc | |
| 258 | #define vec_cmprg_cc __builtin_s390_vec_cmprg_cc | |
| 259 | #define vec_cmpnrg_cc __builtin_s390_vec_cmpnrg_cc | |
| 260 | #define vec_all_nge __builtin_s390_vec_all_nge | |
| 261 | #define vec_all_ngt __builtin_s390_vec_all_ngt | |
| 262 | #define vec_any_nge __builtin_s390_vec_any_nge | |
| 263 | #define vec_any_ngt __builtin_s390_vec_any_ngt | |
| 264 | #define vec_ctd __builtin_s390_vec_ctd | |
| 265 | #define vec_ctd_s64 __builtin_s390_vec_ctd_s64 | |
| 266 | #define vec_ctd_u64 __builtin_s390_vec_ctd_u64 | |
| 267 | #define vec_ctsl __builtin_s390_vec_ctsl | |
| 268 | #define vec_ctul __builtin_s390_vec_ctul | |
| 269 | #define vec_ld2f __builtin_s390_vec_ld2f | |
| 270 | #define vec_st2f __builtin_s390_vec_st2f | |
| 3af82a61 | 271 | #endif /* _VECINTRIN_H */ |