1 /* PowerPC AltiVec include file. 2 Copyright (C) 2002-2018 Free Software Foundation, Inc. 3 Contributed by Aldy Hernandez (aldyh@redhat.com). 4 Rewritten by Paolo Bonzini (bonzini@gnu.org). 5 6 This file is part of GCC. 7 8 GCC is free software; you can redistribute it and/or modify it 9 under the terms of the GNU General Public License as published 10 by the Free Software Foundation; either version 3, or (at your 11 option) any later version. 12 13 GCC is distributed in the hope that it will be useful, but WITHOUT 14 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 15 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public 16 License for more details. 17 18 Under Section 7 of GPL version 3, you are granted additional 19 permissions described in the GCC Runtime Library Exception, version 20 3.1, as published by the Free Software Foundation. 21 22 You should have received a copy of the GNU General Public License and 23 a copy of the GCC Runtime Library Exception along with this program; 24 see the files COPYING3 and COPYING.RUNTIME respectively. If not, see 25 <http://www.gnu.org/licenses/>. */ 26 27 /* Implemented to conform to the specification included in the AltiVec 28 Technology Programming Interface Manual (ALTIVECPIM/D 6/1999 Rev 0). */ 29 30 #ifndef _ALTIVEC_H 31 #define _ALTIVEC_H 1 32 33 #if !defined(__VEC__) || !defined(__ALTIVEC__) 34 #error Use the "-maltivec" flag to enable PowerPC AltiVec support 35 #endif 36 37 /* If __APPLE_ALTIVEC__ is defined, the compiler supports 'vector', 38 'pixel' and 'bool' as context-sensitive AltiVec keywords (in 39 non-AltiVec contexts, they revert to their original meanings, 40 if any), so we do not need to define them as macros. */ 41 42 #if !defined(__APPLE_ALTIVEC__) 43 /* You are allowed to undef these for C++ compatibility. */ 44 #define vector __vector 45 #define pixel __pixel 46 #define bool __bool 47 #endif 48 49 /* Condition register codes for AltiVec predicates. */ 50 51 #define __CR6_EQ 0 52 #define __CR6_EQ_REV 1 53 #define __CR6_LT 2 54 #define __CR6_LT_REV 3 55 56 /* Synonyms. */ 57 #define vec_vaddcuw vec_addc 58 #define vec_vand vec_and 59 #define vec_vandc vec_andc 60 #define vec_vrfip vec_ceil 61 #define vec_vcmpbfp vec_cmpb 62 #define vec_vcmpgefp vec_cmpge 63 #define vec_vctsxs vec_cts 64 #define vec_vctuxs vec_ctu 65 #define vec_vexptefp vec_expte 66 #define vec_vrfim vec_floor 67 #define vec_lvx vec_ld 68 #define vec_lvxl vec_ldl 69 #define vec_vlogefp vec_loge 70 #define vec_vmaddfp vec_madd 71 #define vec_vmhaddshs vec_madds 72 #define vec_vmladduhm vec_mladd 73 #define vec_vmhraddshs vec_mradds 74 #define vec_vnmsubfp vec_nmsub 75 #define vec_vnor vec_nor 76 #define vec_vor vec_or 77 #define vec_vpkpx vec_packpx 78 #define vec_vperm vec_perm 79 #define vec_vrefp vec_re 80 #define vec_vrfin vec_round 81 #define vec_vrsqrtefp vec_rsqrte 82 #define vec_vsel vec_sel 83 #define vec_vsldoi vec_sld 84 #define vec_vsl vec_sll 85 #define vec_vslo vec_slo 86 #define vec_vspltisb vec_splat_s8 87 #define vec_vspltish vec_splat_s16 88 #define vec_vspltisw vec_splat_s32 89 #define vec_vsr vec_srl 90 #define vec_vsro vec_sro 91 #define vec_stvx vec_st 92 #define vec_stvxl vec_stl 93 #define vec_vsubcuw vec_subc 94 #define vec_vsum2sws vec_sum2s 95 #define vec_vsumsws vec_sums 96 #define vec_vrfiz vec_trunc 97 #define vec_vxor vec_xor 98 99 /* Functions that are resolved by the backend to one of the 100 typed builtins. */ 101 #define vec_vaddfp __builtin_vec_vaddfp 102 #define vec_addc __builtin_vec_addc 103 #define vec_adde __builtin_vec_adde 104 #define vec_addec __builtin_vec_addec 105 #define vec_vaddsws __builtin_vec_vaddsws 106 #define vec_vaddshs __builtin_vec_vaddshs 107 #define vec_vaddsbs __builtin_vec_vaddsbs 108 #define vec_vavgsw __builtin_vec_vavgsw 109 #define vec_vavguw __builtin_vec_vavguw 110 #define vec_vavgsh __builtin_vec_vavgsh 111 #define vec_vavguh __builtin_vec_vavguh 112 #define vec_vavgsb __builtin_vec_vavgsb 113 #define vec_vavgub __builtin_vec_vavgub 114 #define vec_ceil __builtin_vec_ceil 115 #define vec_cmpb __builtin_vec_cmpb 116 #define vec_vcmpeqfp __builtin_vec_vcmpeqfp 117 #define vec_cmpge __builtin_vec_cmpge 118 #define vec_vcmpgtfp __builtin_vec_vcmpgtfp 119 #define vec_vcmpgtsw __builtin_vec_vcmpgtsw 120 #define vec_vcmpgtuw __builtin_vec_vcmpgtuw 121 #define vec_vcmpgtsh __builtin_vec_vcmpgtsh 122 #define vec_vcmpgtuh __builtin_vec_vcmpgtuh 123 #define vec_vcmpgtsb __builtin_vec_vcmpgtsb 124 #define vec_vcmpgtub __builtin_vec_vcmpgtub 125 #define vec_vcfsx __builtin_vec_vcfsx 126 #define vec_vcfux __builtin_vec_vcfux 127 #define vec_cts __builtin_vec_cts 128 #define vec_ctu __builtin_vec_ctu 129 #define vec_cpsgn __builtin_vec_copysign 130 #define vec_double __builtin_vec_double 131 #define vec_expte __builtin_vec_expte 132 #define vec_floor __builtin_vec_floor 133 #define vec_loge __builtin_vec_loge 134 #define vec_madd __builtin_vec_madd 135 #define vec_madds __builtin_vec_madds 136 #define vec_mtvscr __builtin_vec_mtvscr 137 #define vec_vmaxfp __builtin_vec_vmaxfp 138 #define vec_vmaxsw __builtin_vec_vmaxsw 139 #define vec_vmaxsh __builtin_vec_vmaxsh 140 #define vec_vmaxsb __builtin_vec_vmaxsb 141 #define vec_vminfp __builtin_vec_vminfp 142 #define vec_vminsw __builtin_vec_vminsw 143 #define vec_vminsh __builtin_vec_vminsh 144 #define vec_vminsb __builtin_vec_vminsb 145 #define vec_mradds __builtin_vec_mradds 146 #define vec_vmsumshm __builtin_vec_vmsumshm 147 #define vec_vmsumuhm __builtin_vec_vmsumuhm 148 #define vec_vmsummbm __builtin_vec_vmsummbm 149 #define vec_vmsumubm __builtin_vec_vmsumubm 150 #define vec_vmsumshs __builtin_vec_vmsumshs 151 #define vec_vmsumuhs __builtin_vec_vmsumuhs 152 #define vec_vmulesb __builtin_vec_vmulesb 153 #define vec_vmulesh __builtin_vec_vmulesh 154 #define vec_vmuleuh __builtin_vec_vmuleuh 155 #define vec_vmuleub __builtin_vec_vmuleub 156 #define vec_vmulosh __builtin_vec_vmulosh 157 #define vec_vmulouh __builtin_vec_vmulouh 158 #define vec_vmulosb __builtin_vec_vmulosb 159 #define vec_vmuloub __builtin_vec_vmuloub 160 #define vec_nmsub __builtin_vec_nmsub 161 #define vec_packpx __builtin_vec_packpx 162 #define vec_vpkswss __builtin_vec_vpkswss 163 #define vec_vpkuwus __builtin_vec_vpkuwus 164 #define vec_vpkshss __builtin_vec_vpkshss 165 #define vec_vpkuhus __builtin_vec_vpkuhus 166 #define vec_vpkswus __builtin_vec_vpkswus 167 #define vec_vpkshus __builtin_vec_vpkshus 168 #define vec_re __builtin_vec_re 169 #define vec_round __builtin_vec_round 170 #define vec_recipdiv __builtin_vec_recipdiv 171 #define vec_rlmi __builtin_vec_rlmi 172 #define vec_vrlnm __builtin_vec_rlnm 173 #define vec_rlnm(a,b,c) (__builtin_vec_rlnm((a),((b)<<8)|(c))) 174 #define vec_rsqrt __builtin_vec_rsqrt 175 #define vec_rsqrte __builtin_vec_rsqrte 176 #define vec_vsubfp __builtin_vec_vsubfp 177 #define vec_subc __builtin_vec_subc 178 #define vec_vsubsws __builtin_vec_vsubsws 179 #define vec_vsubshs __builtin_vec_vsubshs 180 #define vec_vsubsbs __builtin_vec_vsubsbs 181 #define vec_sum4s __builtin_vec_sum4s 182 #define vec_vsum4shs __builtin_vec_vsum4shs 183 #define vec_vsum4sbs __builtin_vec_vsum4sbs 184 #define vec_vsum4ubs __builtin_vec_vsum4ubs 185 #define vec_sum2s __builtin_vec_sum2s 186 #define vec_sums __builtin_vec_sums 187 #define vec_trunc __builtin_vec_trunc 188 #define vec_vupkhpx __builtin_vec_vupkhpx 189 #define vec_vupkhsh __builtin_vec_vupkhsh 190 #define vec_vupkhsb __builtin_vec_vupkhsb 191 #define vec_vupklpx __builtin_vec_vupklpx 192 #define vec_vupklsh __builtin_vec_vupklsh 193 #define vec_vupklsb __builtin_vec_vupklsb 194 #define vec_abs __builtin_vec_abs 195 #define vec_nabs __builtin_vec_nabs 196 #define vec_abss __builtin_vec_abss 197 #define vec_add __builtin_vec_add 198 #define vec_adds __builtin_vec_adds 199 #define vec_and __builtin_vec_and 200 #define vec_andc __builtin_vec_andc 201 #define vec_avg __builtin_vec_avg 202 #define vec_cmpeq __builtin_vec_cmpeq 203 #define vec_cmpne __builtin_vec_cmpne 204 #define vec_cmpgt __builtin_vec_cmpgt 205 #define vec_ctf __builtin_vec_ctf 206 #define vec_dst __builtin_vec_dst 207 #define vec_dstst __builtin_vec_dstst 208 #define vec_dststt __builtin_vec_dststt 209 #define vec_dstt __builtin_vec_dstt 210 #define vec_ld __builtin_vec_ld 211 #define vec_lde __builtin_vec_lde 212 #define vec_ldl __builtin_vec_ldl 213 #define vec_lvebx __builtin_vec_lvebx 214 #define vec_lvehx __builtin_vec_lvehx 215 #define vec_lvewx __builtin_vec_lvewx 216 #define vec_neg __builtin_vec_neg 217 #define vec_pmsum_be __builtin_vec_vpmsum 218 #define vec_shasigma_be __builtin_crypto_vshasigma 219 /* Cell only intrinsics. */ 220 #ifdef __PPU__ 221 #define vec_lvlx __builtin_vec_lvlx 222 #define vec_lvlxl __builtin_vec_lvlxl 223 #define vec_lvrx __builtin_vec_lvrx 224 #define vec_lvrxl __builtin_vec_lvrxl 225 #endif 226 #define vec_lvsl __builtin_vec_lvsl 227 #define vec_lvsr __builtin_vec_lvsr 228 #define vec_max __builtin_vec_max 229 #define vec_mergee __builtin_vec_vmrgew 230 #define vec_mergeh __builtin_vec_mergeh 231 #define vec_mergel __builtin_vec_mergel 232 #define vec_mergeo __builtin_vec_vmrgow 233 #define vec_min __builtin_vec_min 234 #define vec_mladd __builtin_vec_mladd 235 #define vec_msum __builtin_vec_msum 236 #define vec_msums __builtin_vec_msums 237 #define vec_mul __builtin_vec_mul 238 #define vec_mule __builtin_vec_mule 239 #define vec_mulo __builtin_vec_mulo 240 #define vec_nor __builtin_vec_nor 241 #define vec_or __builtin_vec_or 242 #define vec_pack __builtin_vec_pack 243 #define vec_packs __builtin_vec_packs 244 #define vec_packsu __builtin_vec_packsu 245 #define vec_perm __builtin_vec_perm 246 #define vec_rl __builtin_vec_rl 247 #define vec_sel __builtin_vec_sel 248 #define vec_sl __builtin_vec_sl 249 #define vec_sld __builtin_vec_sld 250 #define vec_sldw __builtin_vsx_xxsldwi 251 #define vec_sll __builtin_vec_sll 252 #define vec_slo __builtin_vec_slo 253 #define vec_splat __builtin_vec_splat 254 #define vec_sr __builtin_vec_sr 255 #define vec_sra __builtin_vec_sra 256 #define vec_srl __builtin_vec_srl 257 #define vec_sro __builtin_vec_sro 258 #define vec_st __builtin_vec_st 259 #define vec_ste __builtin_vec_ste 260 #define vec_stl __builtin_vec_stl 261 #define vec_stvebx __builtin_vec_stvebx 262 #define vec_stvehx __builtin_vec_stvehx 263 #define vec_stvewx __builtin_vec_stvewx 264 /* Cell only intrinsics. */ 265 #ifdef __PPU__ 266 #define vec_stvlx __builtin_vec_stvlx 267 #define vec_stvlxl __builtin_vec_stvlxl 268 #define vec_stvrx __builtin_vec_stvrx 269 #define vec_stvrxl __builtin_vec_stvrxl 270 #endif 271 #define vec_sub __builtin_vec_sub 272 #define vec_subs __builtin_vec_subs 273 #define vec_sum __builtin_vec_sum 274 #define vec_unpackh __builtin_vec_unpackh 275 #define vec_unpackl __builtin_vec_unpackl 276 #define vec_vaddubm __builtin_vec_vaddubm 277 #define vec_vaddubs __builtin_vec_vaddubs 278 #define vec_vadduhm __builtin_vec_vadduhm 279 #define vec_vadduhs __builtin_vec_vadduhs 280 #define vec_vadduwm __builtin_vec_vadduwm 281 #define vec_vadduws __builtin_vec_vadduws 282 #define vec_vcmpequb __builtin_vec_vcmpequb 283 #define vec_vcmpequh __builtin_vec_vcmpequh 284 #define vec_vcmpequw __builtin_vec_vcmpequw 285 #define vec_vmaxub __builtin_vec_vmaxub 286 #define vec_vmaxuh __builtin_vec_vmaxuh 287 #define vec_vmaxuw __builtin_vec_vmaxuw 288 #define vec_vminub __builtin_vec_vminub 289 #define vec_vminuh __builtin_vec_vminuh 290 #define vec_vminuw __builtin_vec_vminuw 291 #define vec_vmrghb __builtin_vec_vmrghb 292 #define vec_vmrghh __builtin_vec_vmrghh 293 #define vec_vmrghw __builtin_vec_vmrghw 294 #define vec_vmrglb __builtin_vec_vmrglb 295 #define vec_vmrglh __builtin_vec_vmrglh 296 #define vec_vmrglw __builtin_vec_vmrglw 297 #define vec_vpkuhum __builtin_vec_vpkuhum 298 #define vec_vpkuwum __builtin_vec_vpkuwum 299 #define vec_vrlb __builtin_vec_vrlb 300 #define vec_vrlh __builtin_vec_vrlh 301 #define vec_vrlw __builtin_vec_vrlw 302 #define vec_vslb __builtin_vec_vslb 303 #define vec_vslh __builtin_vec_vslh 304 #define vec_vslw __builtin_vec_vslw 305 #define vec_vspltb __builtin_vec_vspltb 306 #define vec_vsplth __builtin_vec_vsplth 307 #define vec_vspltw __builtin_vec_vspltw 308 #define vec_vsrab __builtin_vec_vsrab 309 #define vec_vsrah __builtin_vec_vsrah 310 #define vec_vsraw __builtin_vec_vsraw 311 #define vec_vsrb __builtin_vec_vsrb 312 #define vec_vsrh __builtin_vec_vsrh 313 #define vec_vsrw __builtin_vec_vsrw 314 #define vec_vsububs __builtin_vec_vsububs 315 #define vec_vsububm __builtin_vec_vsububm 316 #define vec_vsubuhm __builtin_vec_vsubuhm 317 #define vec_vsubuhs __builtin_vec_vsubuhs 318 #define vec_vsubuwm __builtin_vec_vsubuwm 319 #define vec_vsubuws __builtin_vec_vsubuws 320 #define vec_xor __builtin_vec_xor 321 322 #define vec_extract __builtin_vec_extract 323 #define vec_insert __builtin_vec_insert 324 #define vec_splats __builtin_vec_splats 325 #define vec_promote __builtin_vec_promote 326 327 #ifdef __VSX__ 328 /* VSX additions */ 329 #define vec_div __builtin_vec_div 330 #define vec_mul __builtin_vec_mul 331 #define vec_msub __builtin_vec_msub 332 #define vec_nmadd __builtin_vec_nmadd 333 #define vec_nearbyint __builtin_vec_nearbyint 334 #define vec_rint __builtin_vec_rint 335 #define vec_sqrt __builtin_vec_sqrt 336 #define vec_vsx_ld __builtin_vec_vsx_ld 337 #define vec_vsx_st __builtin_vec_vsx_st 338 #define vec_xl __builtin_vec_vsx_ld 339 #define vec_xst __builtin_vec_vsx_st 340 341 /* Note, xxsldi and xxpermdi were added as __builtin_vsx_<xxx> functions 342 instead of __builtin_vec_<xxx> */ 343 #define vec_xxsldwi __builtin_vsx_xxsldwi 344 #define vec_xxpermdi __builtin_vsx_xxpermdi 345 #endif 346 347 #ifdef _ARCH_PWR8 348 /* Vector additions added in ISA 2.07. */ 349 #define vec_eqv __builtin_vec_eqv 350 #define vec_nand __builtin_vec_nand 351 #define vec_orc __builtin_vec_orc 352 #define vec_vaddcuq __builtin_vec_vaddcuq 353 #define vec_vaddudm __builtin_vec_vaddudm 354 #define vec_vadduqm __builtin_vec_vadduqm 355 #define vec_vbpermq __builtin_vec_vbpermq 356 #define vec_bperm __builtin_vec_vbperm_api 357 #define vec_vclz __builtin_vec_vclz 358 #define vec_cntlz __builtin_vec_vclz 359 #define vec_vclzb __builtin_vec_vclzb 360 #define vec_vclzd __builtin_vec_vclzd 361 #define vec_vclzh __builtin_vec_vclzh 362 #define vec_vclzw __builtin_vec_vclzw 363 #define vec_vaddecuq __builtin_vec_vaddecuq 364 #define vec_vaddeuqm __builtin_vec_vaddeuqm 365 #define vec_vsubecuq __builtin_vec_vsubecuq 366 #define vec_vsubeuqm __builtin_vec_vsubeuqm 367 #define vec_vgbbd __builtin_vec_vgbbd 368 #define vec_gb __builtin_vec_vgbbd 369 #define vec_vmaxsd __builtin_vec_vmaxsd 370 #define vec_vmaxud __builtin_vec_vmaxud 371 #define vec_vminsd __builtin_vec_vminsd 372 #define vec_vminud __builtin_vec_vminud 373 #define vec_vmrgew __builtin_vec_vmrgew 374 #define vec_vmrgow __builtin_vec_vmrgow 375 #define vec_vpksdss __builtin_vec_vpksdss 376 #define vec_vpksdus __builtin_vec_vpksdus 377 #define vec_vpkudum __builtin_vec_vpkudum 378 #define vec_vpkudus __builtin_vec_vpkudus 379 #define vec_vpopcnt __builtin_vec_vpopcnt 380 #define vec_vpopcntb __builtin_vec_vpopcntb 381 #define vec_vpopcntd __builtin_vec_vpopcntd 382 #define vec_vpopcnth __builtin_vec_vpopcnth 383 #define vec_vpopcntw __builtin_vec_vpopcntw 384 #define vec_popcnt __builtin_vec_vpopcntu 385 #define vec_popcntb __builtin_vec_vpopcntub 386 #define vec_popcnth __builtin_vec_vpopcntuh 387 #define vec_popcntw __builtin_vec_vpopcntuw 388 #define vec_popcntd __builtin_vec_vpopcntud 389 #define vec_vrld __builtin_vec_vrld 390 #define vec_vsld __builtin_vec_vsld 391 #define vec_vsrad __builtin_vec_vsrad 392 #define vec_vsrd __builtin_vec_vsrd 393 #define vec_vsubcuq __builtin_vec_vsubcuq 394 #define vec_vsubudm __builtin_vec_vsubudm 395 #define vec_vsubuqm __builtin_vec_vsubuqm 396 #define vec_vupkhsw __builtin_vec_vupkhsw 397 #define vec_vupklsw __builtin_vec_vupklsw 398 #endif 399 400 #ifdef __POWER9_VECTOR__ 401 /* Vector additions added in ISA 3.0. */ 402 #define vec_vctz __builtin_vec_vctz 403 #define vec_cnttz __builtin_vec_vctz 404 #define vec_vctzb __builtin_vec_vctzb 405 #define vec_vctzd __builtin_vec_vctzd 406 #define vec_vctzh __builtin_vec_vctzh 407 #define vec_vctzw __builtin_vec_vctzw 408 #define vec_vextract4b __builtin_vec_vextract4b 409 #define vec_vinsert4b __builtin_vec_vinsert4b 410 #define vec_vprtyb __builtin_vec_vprtyb 411 #define vec_vprtybd __builtin_vec_vprtybd 412 #define vec_vprtybw __builtin_vec_vprtybw 413 414 #ifdef _ARCH_PPC64 415 #define vec_vprtybq __builtin_vec_vprtybq 416 #endif 417 418 #define vec_absd __builtin_vec_vadu 419 #define vec_absdb __builtin_vec_vadub 420 #define vec_absdh __builtin_vec_vaduh 421 #define vec_absdw __builtin_vec_vaduw 422 423 #define vec_slv __builtin_vec_vslv 424 #define vec_srv __builtin_vec_vsrv 425 426 #define vec_extract_exp __builtin_vec_extract_exp 427 #define vec_extract_sig __builtin_vec_extract_sig 428 #define vec_insert_exp __builtin_vec_insert_exp 429 #define vec_test_data_class __builtin_vec_test_data_class 430 431 #define scalar_extract_exp __builtin_vec_scalar_extract_exp 432 #define scalar_extract_sig __builtin_vec_scalar_extract_sig 433 #define scalar_insert_exp __builtin_vec_scalar_insert_exp 434 #define scalar_test_data_class __builtin_vec_scalar_test_data_class 435 #define scalar_test_neg __builtin_vec_scalar_test_neg 436 437 #define scalar_cmp_exp_gt __builtin_vec_scalar_cmp_exp_gt 438 #define scalar_cmp_exp_lt __builtin_vec_scalar_cmp_exp_lt 439 #define scalar_cmp_exp_eq __builtin_vec_scalar_cmp_exp_eq 440 #define scalar_cmp_exp_unordered __builtin_vec_scalar_cmp_exp_unordered 441 442 #ifdef _ARCH_PPC64 443 #define vec_xl_len __builtin_vec_lxvl 444 #define vec_xst_len __builtin_vec_stxvl 445 #endif 446 447 #define vec_cmpnez __builtin_vec_vcmpnez 448 449 #define vec_cntlz_lsbb __builtin_vec_vclzlsbb 450 #define vec_cnttz_lsbb __builtin_vec_vctzlsbb 451 452 #define vec_xlx __builtin_vec_vextulx 453 #define vec_xrx __builtin_vec_vexturx 454 455 #define vec_revb __builtin_vec_revb 456 #endif 457 458 /* Predicates. 459 For C++, we use templates in order to allow non-parenthesized arguments. 460 For C, instead, we use macros since non-parenthesized arguments were 461 not allowed even in older GCC implementation of AltiVec. 462 463 In the future, we may add more magic to the back-end, so that no 464 one- or two-argument macros are used. */ 465 466 #ifdef __cplusplus__ 467 #define __altivec_unary_pred(NAME, CALL) \ 468 template <class T> int NAME (T a1) { return CALL; } 469 470 #define __altivec_scalar_pred(NAME, CALL) \ 471 template <class T, class U> int NAME (T a1, U a2) { return CALL; } 472 473 /* Given the vec_step of a type, return the corresponding bool type. */ 474 template <int STEP> class __altivec_bool_ret { }; 475 template <> class __altivec_bool_ret <4> { 476 typedef __vector __bool int __ret; 477 }; 478 template <> class __altivec_bool_ret <8> { 479 typedef __vector __bool short __ret; 480 }; 481 template <> class __altivec_bool_ret <16> { 482 typedef __vector __bool char __ret; 483 }; 484 485 /* Be very liberal in the pairs we accept. Mistakes such as passing 486 a `vector char' and `vector short' will be caught by the middle-end, 487 while any attempt to detect them here would produce hard to understand 488 error messages involving the implementation details of AltiVec. */ 489 #define __altivec_binary_pred(NAME, CALL) \ 490 template <class T, class U> \ 491 typename __altivec_bool_ret <vec_step (T)>::__ret \ 492 NAME (T a1, U a2) \ 493 { \ 494 return CALL; \ 495 } 496 497 __altivec_binary_pred(vec_cmplt, 498 __builtin_vec_cmpgt (a2, a1)) 499 __altivec_binary_pred(vec_cmple, 500 __builtin_vec_cmpge (a2, a1)) 501 502 __altivec_scalar_pred(vec_all_in, 503 __builtin_altivec_vcmpbfp_p (__CR6_EQ, a1, a2)) 504 __altivec_scalar_pred(vec_any_out, 505 __builtin_altivec_vcmpbfp_p (__CR6_EQ_REV, a1, a2)) 506 507 __altivec_unary_pred(vec_all_nan, 508 __builtin_altivec_vcmpeq_p (__CR6_EQ, a1, a1)) 509 __altivec_unary_pred(vec_any_nan, 510 __builtin_altivec_vcmpeq_p (__CR6_LT_REV, a1, a1)) 511 512 __altivec_unary_pred(vec_all_numeric, 513 __builtin_altivec_vcmpeq_p (__CR6_LT, a1, a1)) 514 __altivec_unary_pred(vec_any_numeric, 515 __builtin_altivec_vcmpeq_p (__CR6_EQ_REV, a1, a1)) 516 517 __altivec_scalar_pred(vec_all_eq, 518 __builtin_vec_vcmpeq_p (__CR6_LT, a1, a2)) 519 520 #ifndef __POWER9_VECTOR__ 521 __altivec_scalar_pred(vec_all_ne, 522 __builtin_vec_vcmpeq_p (__CR6_EQ, a1, a2)) 523 __altivec_scalar_pred(vec_any_eq, 524 __builtin_vec_vcmpeq_p (__CR6_EQ_REV, a1, a2)) 525 #else 526 __altivec_scalar_pred(vec_all_nez, 527 __builtin_vec_vcmpnez_p (__CR6_LT, a1, a2)) 528 __altivec_scalar_pred(vec_any_eqz, 529 __builtin_vec_vcmpnez_p (__CR6_LT_REV, a1, a2)) 530 __altivec_scalar_pred(vec_all_ne, 531 __builtin_vec_vcmpne_p (a1, a2)) 532 __altivec_scalar_pred(vec_any_eq, 533 __builtin_vec_vcmpae_p (a1, a2)) 534 #endif 535 536 __altivec_scalar_pred(vec_any_ne, 537 __builtin_vec_vcmpeq_p (__CR6_LT_REV, a1, a2)) 538 539 __altivec_scalar_pred(vec_all_gt, 540 __builtin_vec_vcmpgt_p (__CR6_LT, a1, a2)) 541 __altivec_scalar_pred(vec_all_lt, 542 __builtin_vec_vcmpgt_p (__CR6_LT, a2, a1)) 543 __altivec_scalar_pred(vec_any_gt, 544 __builtin_vec_vcmpgt_p (__CR6_EQ_REV, a1, a2)) 545 __altivec_scalar_pred(vec_any_lt, 546 __builtin_vec_vcmpgt_p (__CR6_EQ_REV, a2, a1)) 547 548 __altivec_scalar_pred(vec_all_ngt, 549 __builtin_altivec_vcmpgt_p (__CR6_EQ, a1, a2)) 550 __altivec_scalar_pred(vec_all_nlt, 551 __builtin_altivec_vcmpgt_p (__CR6_EQ, a2, a1)) 552 __altivec_scalar_pred(vec_any_ngt, 553 __builtin_altivec_vcmpgt_p (__CR6_LT_REV, a1, a2)) 554 __altivec_scalar_pred(vec_any_nlt, 555 __builtin_altivec_vcmpgt_p (__CR6_LT_REV, a2, a1)) 556 557 /* __builtin_vec_vcmpge_p is vcmpgefp for floating-point vector types, 558 while for integer types it is converted to __builtin_vec_vcmpgt_p, 559 with inverted args and condition code. */ 560 __altivec_scalar_pred(vec_all_le, 561 __builtin_vec_vcmpge_p (__CR6_LT, a2, a1)) 562 __altivec_scalar_pred(vec_all_ge, 563 __builtin_vec_vcmpge_p (__CR6_LT, a1, a2)) 564 __altivec_scalar_pred(vec_any_le, 565 __builtin_vec_vcmpge_p (__CR6_EQ_REV, a2, a1)) 566 __altivec_scalar_pred(vec_any_ge, 567 __builtin_vec_vcmpge_p (__CR6_EQ_REV, a1, a2)) 568 569 __altivec_scalar_pred(vec_all_nge, 570 __builtin_altivec_vcmpge_p (__CR6_EQ, a1, a2)) 571 __altivec_scalar_pred(vec_all_nle, 572 __builtin_altivec_vcmpge_p (__CR6_EQ, a2, a1)) 573 __altivec_scalar_pred(vec_any_nge, 574 __builtin_altivec_vcmpge_p (__CR6_LT_REV, a1, a2)) 575 __altivec_scalar_pred(vec_any_nle, 576 __builtin_altivec_vcmpge_p (__CR6_LT_REV, a2, a1)) 577 578 #undef __altivec_scalar_pred 579 #undef __altivec_unary_pred 580 #undef __altivec_binary_pred 581 #else 582 #define vec_cmplt(a1, a2) __builtin_vec_cmpgt ((a2), (a1)) 583 #define vec_cmple(a1, a2) __builtin_vec_cmpge ((a2), (a1)) 584 585 #define vec_all_in(a1, a2) __builtin_altivec_vcmpbfp_p (__CR6_EQ, (a1), (a2)) 586 #define vec_any_out(a1, a2) __builtin_altivec_vcmpbfp_p (__CR6_EQ_REV, (a1), (a2)) 587 588 #define vec_all_nan(a1) __builtin_vec_vcmpeq_p (__CR6_EQ, (a1), (a1)) 589 #define vec_any_nan(a1) __builtin_vec_vcmpeq_p (__CR6_LT_REV, (a1), (a1)) 590 591 #define vec_all_numeric(a1) __builtin_vec_vcmpeq_p (__CR6_LT, (a1), (a1)) 592 #define vec_any_numeric(a1) __builtin_vec_vcmpeq_p (__CR6_EQ_REV, (a1), (a1)) 593 594 #define vec_all_eq(a1, a2) __builtin_vec_vcmpeq_p (__CR6_LT, (a1), (a2)) 595 596 #ifdef __POWER9_VECTOR__ 597 #define vec_all_nez(a1, a2) __builtin_vec_vcmpnez_p (__CR6_LT, (a1), (a2)) 598 #define vec_any_eqz(a1, a2) __builtin_vec_vcmpnez_p (__CR6_LT_REV, (a1), (a2)) 599 #define vec_all_ne(a1, a2) __builtin_vec_vcmpne_p ((a1), (a2)) 600 #define vec_any_eq(a1, a2) __builtin_vec_vcmpae_p ((a1), (a2)) 601 #else 602 #define vec_all_ne(a1, a2) __builtin_vec_vcmpeq_p (__CR6_EQ, (a1), (a2)) 603 #define vec_any_eq(a1, a2) __builtin_vec_vcmpeq_p (__CR6_EQ_REV, (a1), (a2)) 604 #endif 605 606 #define vec_any_ne(a1, a2) __builtin_vec_vcmpeq_p (__CR6_LT_REV, (a1), (a2)) 607 608 #define vec_all_gt(a1, a2) __builtin_vec_vcmpgt_p (__CR6_LT, (a1), (a2)) 609 #define vec_all_lt(a1, a2) __builtin_vec_vcmpgt_p (__CR6_LT, (a2), (a1)) 610 #define vec_any_gt(a1, a2) __builtin_vec_vcmpgt_p (__CR6_EQ_REV, (a1), (a2)) 611 #define vec_any_lt(a1, a2) __builtin_vec_vcmpgt_p (__CR6_EQ_REV, (a2), (a1)) 612 613 #define vec_all_ngt(a1, a2) __builtin_vec_vcmpgt_p (__CR6_EQ, (a1), (a2)) 614 #define vec_all_nlt(a1, a2) __builtin_vec_vcmpgt_p (__CR6_EQ, (a2), (a1)) 615 #define vec_any_ngt(a1, a2) __builtin_vec_vcmpgt_p (__CR6_LT_REV, (a1), (a2)) 616 #define vec_any_nlt(a1, a2) __builtin_vec_vcmpgt_p (__CR6_LT_REV, (a2), (a1)) 617 618 /* __builtin_vec_vcmpge_p is vcmpgefp for floating-point vector types, 619 while for integer types it is converted to __builtin_vec_vcmpgt_p, 620 with inverted args and condition code. */ 621 #define vec_all_le(a1, a2) __builtin_vec_vcmpge_p (__CR6_LT, (a2), (a1)) 622 #define vec_all_ge(a1, a2) __builtin_vec_vcmpge_p (__CR6_LT, (a1), (a2)) 623 #define vec_any_le(a1, a2) __builtin_vec_vcmpge_p (__CR6_EQ_REV, (a2), (a1)) 624 #define vec_any_ge(a1, a2) __builtin_vec_vcmpge_p (__CR6_EQ_REV, (a1), (a2)) 625 626 #define vec_all_nge(a1, a2) __builtin_vec_vcmpge_p (__CR6_EQ, (a1), (a2)) 627 #define vec_all_nle(a1, a2) __builtin_vec_vcmpge_p (__CR6_EQ, (a2), (a1)) 628 #define vec_any_nge(a1, a2) __builtin_vec_vcmpge_p (__CR6_LT_REV, (a1), (a2)) 629 #define vec_any_nle(a1, a2) __builtin_vec_vcmpge_p (__CR6_LT_REV, (a2), (a1)) 630 #endif 631 632 /* These do not accept vectors, so they do not have a __builtin_vec_* 633 counterpart. */ 634 #define vec_dss(x) __builtin_altivec_dss((x)) 635 #define vec_dssall() __builtin_altivec_dssall () 636 #define vec_mfvscr() ((__vector unsigned short) __builtin_altivec_mfvscr ()) 637 #define vec_splat_s8(x) __builtin_altivec_vspltisb ((x)) 638 #define vec_splat_s16(x) __builtin_altivec_vspltish ((x)) 639 #define vec_splat_s32(x) __builtin_altivec_vspltisw ((x)) 640 #define vec_splat_u8(x) ((__vector unsigned char) vec_splat_s8 ((x))) 641 #define vec_splat_u16(x) ((__vector unsigned short) vec_splat_s16 ((x))) 642 #define vec_splat_u32(x) ((__vector unsigned int) vec_splat_s32 ((x))) 643 644 /* This also accepts a type for its parameter, so it is not enough 645 to #define vec_step to __builtin_vec_step. */ 646 #define vec_step(x) __builtin_vec_step (* (__typeof__ (x) *) 0) 647 648 #endif /* _ALTIVEC_H */ 649