1 /* 2 * ARM translation: M-profile MVE instructions 3 * 4 * Copyright (c) 2021 Linaro, Ltd. 5 * 6 * This library is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU Lesser General Public 8 * License as published by the Free Software Foundation; either 9 * version 2.1 of the License, or (at your option) any later version. 10 * 11 * This library is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * Lesser General Public License for more details. 15 * 16 * You should have received a copy of the GNU Lesser General Public 17 * License along with this library; if not, see <http://www.gnu.org/licenses/>. 18 */ 19 20 #include "qemu/osdep.h" 21 #include "tcg/tcg-op.h" 22 #include "tcg/tcg-op-gvec.h" 23 #include "exec/exec-all.h" 24 #include "exec/gen-icount.h" 25 #include "translate.h" 26 #include "translate-a32.h" 27 28 static inline int vidup_imm(DisasContext *s, int x) 29 { 30 return 1 << x; 31 } 32 33 /* Include the generated decoder */ 34 #include "decode-mve.c.inc" 35 36 typedef void MVEGenLdStFn(TCGv_ptr, TCGv_ptr, TCGv_i32); 37 typedef void MVEGenLdStSGFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); 38 typedef void MVEGenLdStIlFn(TCGv_ptr, TCGv_i32, TCGv_i32); 39 typedef void MVEGenOneOpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr); 40 typedef void MVEGenTwoOpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_ptr); 41 typedef void MVEGenTwoOpScalarFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); 42 typedef void MVEGenTwoOpShiftFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); 43 typedef void MVEGenLongDualAccOpFn(TCGv_i64, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i64); 44 typedef void MVEGenVADDVFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32); 45 typedef void MVEGenOneOpImmFn(TCGv_ptr, TCGv_ptr, TCGv_i64); 46 typedef void MVEGenVIDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32); 47 typedef void MVEGenVIWDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32, TCGv_i32); 48 typedef void MVEGenCmpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr); 49 typedef void MVEGenScalarCmpFn(TCGv_ptr, TCGv_ptr, TCGv_i32); 50 typedef void MVEGenVABAVFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); 51 typedef void MVEGenDualAccOpFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); 52 53 /* Return the offset of a Qn register (same semantics as aa32_vfp_qreg()) */ 54 static inline long mve_qreg_offset(unsigned reg) 55 { 56 return offsetof(CPUARMState, vfp.zregs[reg].d[0]); 57 } 58 59 static TCGv_ptr mve_qreg_ptr(unsigned reg) 60 { 61 TCGv_ptr ret = tcg_temp_new_ptr(); 62 tcg_gen_addi_ptr(ret, cpu_env, mve_qreg_offset(reg)); 63 return ret; 64 } 65 66 static bool mve_check_qreg_bank(DisasContext *s, int qmask) 67 { 68 /* 69 * Check whether Qregs are in range. For v8.1M only Q0..Q7 70 * are supported, see VFPSmallRegisterBank(). 71 */ 72 return qmask < 8; 73 } 74 75 bool mve_eci_check(DisasContext *s) 76 { 77 /* 78 * This is a beatwise insn: check that ECI is valid (not a 79 * reserved value) and note that we are handling it. 80 * Return true if OK, false if we generated an exception. 81 */ 82 s->eci_handled = true; 83 switch (s->eci) { 84 case ECI_NONE: 85 case ECI_A0: 86 case ECI_A0A1: 87 case ECI_A0A1A2: 88 case ECI_A0A1A2B0: 89 return true; 90 default: 91 /* Reserved value: INVSTATE UsageFault */ 92 gen_exception_insn(s, s->pc_curr, EXCP_INVSTATE, syn_uncategorized(), 93 default_exception_el(s)); 94 return false; 95 } 96 } 97 98 void mve_update_eci(DisasContext *s) 99 { 100 /* 101 * The helper function will always update the CPUState field, 102 * so we only need to update the DisasContext field. 103 */ 104 if (s->eci) { 105 s->eci = (s->eci == ECI_A0A1A2B0) ? ECI_A0 : ECI_NONE; 106 } 107 } 108 109 void mve_update_and_store_eci(DisasContext *s) 110 { 111 /* 112 * For insns which don't call a helper function that will call 113 * mve_advance_vpt(), this version updates s->eci and also stores 114 * it out to the CPUState field. 115 */ 116 if (s->eci) { 117 mve_update_eci(s); 118 store_cpu_field(tcg_constant_i32(s->eci << 4), condexec_bits); 119 } 120 } 121 122 static bool mve_skip_first_beat(DisasContext *s) 123 { 124 /* Return true if PSR.ECI says we must skip the first beat of this insn */ 125 switch (s->eci) { 126 case ECI_NONE: 127 return false; 128 case ECI_A0: 129 case ECI_A0A1: 130 case ECI_A0A1A2: 131 case ECI_A0A1A2B0: 132 return true; 133 default: 134 g_assert_not_reached(); 135 } 136 } 137 138 static bool do_ldst(DisasContext *s, arg_VLDR_VSTR *a, MVEGenLdStFn *fn, 139 unsigned msize) 140 { 141 TCGv_i32 addr; 142 uint32_t offset; 143 TCGv_ptr qreg; 144 145 if (!dc_isar_feature(aa32_mve, s) || 146 !mve_check_qreg_bank(s, a->qd) || 147 !fn) { 148 return false; 149 } 150 151 /* CONSTRAINED UNPREDICTABLE: we choose to UNDEF */ 152 if (a->rn == 15 || (a->rn == 13 && a->w)) { 153 return false; 154 } 155 156 if (!mve_eci_check(s) || !vfp_access_check(s)) { 157 return true; 158 } 159 160 offset = a->imm << msize; 161 if (!a->a) { 162 offset = -offset; 163 } 164 addr = load_reg(s, a->rn); 165 if (a->p) { 166 tcg_gen_addi_i32(addr, addr, offset); 167 } 168 169 qreg = mve_qreg_ptr(a->qd); 170 fn(cpu_env, qreg, addr); 171 tcg_temp_free_ptr(qreg); 172 173 /* 174 * Writeback always happens after the last beat of the insn, 175 * regardless of predication 176 */ 177 if (a->w) { 178 if (!a->p) { 179 tcg_gen_addi_i32(addr, addr, offset); 180 } 181 store_reg(s, a->rn, addr); 182 } else { 183 tcg_temp_free_i32(addr); 184 } 185 mve_update_eci(s); 186 return true; 187 } 188 189 static bool trans_VLDR_VSTR(DisasContext *s, arg_VLDR_VSTR *a) 190 { 191 static MVEGenLdStFn * const ldstfns[4][2] = { 192 { gen_helper_mve_vstrb, gen_helper_mve_vldrb }, 193 { gen_helper_mve_vstrh, gen_helper_mve_vldrh }, 194 { gen_helper_mve_vstrw, gen_helper_mve_vldrw }, 195 { NULL, NULL } 196 }; 197 return do_ldst(s, a, ldstfns[a->size][a->l], a->size); 198 } 199 200 #define DO_VLDST_WIDE_NARROW(OP, SLD, ULD, ST, MSIZE) \ 201 static bool trans_##OP(DisasContext *s, arg_VLDR_VSTR *a) \ 202 { \ 203 static MVEGenLdStFn * const ldstfns[2][2] = { \ 204 { gen_helper_mve_##ST, gen_helper_mve_##SLD }, \ 205 { NULL, gen_helper_mve_##ULD }, \ 206 }; \ 207 return do_ldst(s, a, ldstfns[a->u][a->l], MSIZE); \ 208 } 209 210 DO_VLDST_WIDE_NARROW(VLDSTB_H, vldrb_sh, vldrb_uh, vstrb_h, MO_8) 211 DO_VLDST_WIDE_NARROW(VLDSTB_W, vldrb_sw, vldrb_uw, vstrb_w, MO_8) 212 DO_VLDST_WIDE_NARROW(VLDSTH_W, vldrh_sw, vldrh_uw, vstrh_w, MO_16) 213 214 static bool do_ldst_sg(DisasContext *s, arg_vldst_sg *a, MVEGenLdStSGFn fn) 215 { 216 TCGv_i32 addr; 217 TCGv_ptr qd, qm; 218 219 if (!dc_isar_feature(aa32_mve, s) || 220 !mve_check_qreg_bank(s, a->qd | a->qm) || 221 !fn || a->rn == 15) { 222 /* Rn case is UNPREDICTABLE */ 223 return false; 224 } 225 226 if (!mve_eci_check(s) || !vfp_access_check(s)) { 227 return true; 228 } 229 230 addr = load_reg(s, a->rn); 231 232 qd = mve_qreg_ptr(a->qd); 233 qm = mve_qreg_ptr(a->qm); 234 fn(cpu_env, qd, qm, addr); 235 tcg_temp_free_ptr(qd); 236 tcg_temp_free_ptr(qm); 237 tcg_temp_free_i32(addr); 238 mve_update_eci(s); 239 return true; 240 } 241 242 /* 243 * The naming scheme here is "vldrb_sg_sh == in-memory byte loads 244 * signextended to halfword elements in register". _os_ indicates that 245 * the offsets in Qm should be scaled by the element size. 246 */ 247 /* This macro is just to make the arrays more compact in these functions */ 248 #define F(N) gen_helper_mve_##N 249 250 /* VLDRB/VSTRB (ie msize 1) with OS=1 is UNPREDICTABLE; we UNDEF */ 251 static bool trans_VLDR_S_sg(DisasContext *s, arg_vldst_sg *a) 252 { 253 static MVEGenLdStSGFn * const fns[2][4][4] = { { 254 { NULL, F(vldrb_sg_sh), F(vldrb_sg_sw), NULL }, 255 { NULL, NULL, F(vldrh_sg_sw), NULL }, 256 { NULL, NULL, NULL, NULL }, 257 { NULL, NULL, NULL, NULL } 258 }, { 259 { NULL, NULL, NULL, NULL }, 260 { NULL, NULL, F(vldrh_sg_os_sw), NULL }, 261 { NULL, NULL, NULL, NULL }, 262 { NULL, NULL, NULL, NULL } 263 } 264 }; 265 if (a->qd == a->qm) { 266 return false; /* UNPREDICTABLE */ 267 } 268 return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]); 269 } 270 271 static bool trans_VLDR_U_sg(DisasContext *s, arg_vldst_sg *a) 272 { 273 static MVEGenLdStSGFn * const fns[2][4][4] = { { 274 { F(vldrb_sg_ub), F(vldrb_sg_uh), F(vldrb_sg_uw), NULL }, 275 { NULL, F(vldrh_sg_uh), F(vldrh_sg_uw), NULL }, 276 { NULL, NULL, F(vldrw_sg_uw), NULL }, 277 { NULL, NULL, NULL, F(vldrd_sg_ud) } 278 }, { 279 { NULL, NULL, NULL, NULL }, 280 { NULL, F(vldrh_sg_os_uh), F(vldrh_sg_os_uw), NULL }, 281 { NULL, NULL, F(vldrw_sg_os_uw), NULL }, 282 { NULL, NULL, NULL, F(vldrd_sg_os_ud) } 283 } 284 }; 285 if (a->qd == a->qm) { 286 return false; /* UNPREDICTABLE */ 287 } 288 return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]); 289 } 290 291 static bool trans_VSTR_sg(DisasContext *s, arg_vldst_sg *a) 292 { 293 static MVEGenLdStSGFn * const fns[2][4][4] = { { 294 { F(vstrb_sg_ub), F(vstrb_sg_uh), F(vstrb_sg_uw), NULL }, 295 { NULL, F(vstrh_sg_uh), F(vstrh_sg_uw), NULL }, 296 { NULL, NULL, F(vstrw_sg_uw), NULL }, 297 { NULL, NULL, NULL, F(vstrd_sg_ud) } 298 }, { 299 { NULL, NULL, NULL, NULL }, 300 { NULL, F(vstrh_sg_os_uh), F(vstrh_sg_os_uw), NULL }, 301 { NULL, NULL, F(vstrw_sg_os_uw), NULL }, 302 { NULL, NULL, NULL, F(vstrd_sg_os_ud) } 303 } 304 }; 305 return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]); 306 } 307 308 #undef F 309 310 static bool do_ldst_sg_imm(DisasContext *s, arg_vldst_sg_imm *a, 311 MVEGenLdStSGFn *fn, unsigned msize) 312 { 313 uint32_t offset; 314 TCGv_ptr qd, qm; 315 316 if (!dc_isar_feature(aa32_mve, s) || 317 !mve_check_qreg_bank(s, a->qd | a->qm) || 318 !fn) { 319 return false; 320 } 321 322 if (!mve_eci_check(s) || !vfp_access_check(s)) { 323 return true; 324 } 325 326 offset = a->imm << msize; 327 if (!a->a) { 328 offset = -offset; 329 } 330 331 qd = mve_qreg_ptr(a->qd); 332 qm = mve_qreg_ptr(a->qm); 333 fn(cpu_env, qd, qm, tcg_constant_i32(offset)); 334 tcg_temp_free_ptr(qd); 335 tcg_temp_free_ptr(qm); 336 mve_update_eci(s); 337 return true; 338 } 339 340 static bool trans_VLDRW_sg_imm(DisasContext *s, arg_vldst_sg_imm *a) 341 { 342 static MVEGenLdStSGFn * const fns[] = { 343 gen_helper_mve_vldrw_sg_uw, 344 gen_helper_mve_vldrw_sg_wb_uw, 345 }; 346 if (a->qd == a->qm) { 347 return false; /* UNPREDICTABLE */ 348 } 349 return do_ldst_sg_imm(s, a, fns[a->w], MO_32); 350 } 351 352 static bool trans_VLDRD_sg_imm(DisasContext *s, arg_vldst_sg_imm *a) 353 { 354 static MVEGenLdStSGFn * const fns[] = { 355 gen_helper_mve_vldrd_sg_ud, 356 gen_helper_mve_vldrd_sg_wb_ud, 357 }; 358 if (a->qd == a->qm) { 359 return false; /* UNPREDICTABLE */ 360 } 361 return do_ldst_sg_imm(s, a, fns[a->w], MO_64); 362 } 363 364 static bool trans_VSTRW_sg_imm(DisasContext *s, arg_vldst_sg_imm *a) 365 { 366 static MVEGenLdStSGFn * const fns[] = { 367 gen_helper_mve_vstrw_sg_uw, 368 gen_helper_mve_vstrw_sg_wb_uw, 369 }; 370 return do_ldst_sg_imm(s, a, fns[a->w], MO_32); 371 } 372 373 static bool trans_VSTRD_sg_imm(DisasContext *s, arg_vldst_sg_imm *a) 374 { 375 static MVEGenLdStSGFn * const fns[] = { 376 gen_helper_mve_vstrd_sg_ud, 377 gen_helper_mve_vstrd_sg_wb_ud, 378 }; 379 return do_ldst_sg_imm(s, a, fns[a->w], MO_64); 380 } 381 382 static bool do_vldst_il(DisasContext *s, arg_vldst_il *a, MVEGenLdStIlFn *fn, 383 int addrinc) 384 { 385 TCGv_i32 rn; 386 387 if (!dc_isar_feature(aa32_mve, s) || 388 !mve_check_qreg_bank(s, a->qd) || 389 !fn || (a->rn == 13 && a->w) || a->rn == 15) { 390 /* Variously UNPREDICTABLE or UNDEF or related-encoding */ 391 return false; 392 } 393 if (!mve_eci_check(s) || !vfp_access_check(s)) { 394 return true; 395 } 396 397 rn = load_reg(s, a->rn); 398 /* 399 * We pass the index of Qd, not a pointer, because the helper must 400 * access multiple Q registers starting at Qd and working up. 401 */ 402 fn(cpu_env, tcg_constant_i32(a->qd), rn); 403 404 if (a->w) { 405 tcg_gen_addi_i32(rn, rn, addrinc); 406 store_reg(s, a->rn, rn); 407 } else { 408 tcg_temp_free_i32(rn); 409 } 410 mve_update_and_store_eci(s); 411 return true; 412 } 413 414 /* This macro is just to make the arrays more compact in these functions */ 415 #define F(N) gen_helper_mve_##N 416 417 static bool trans_VLD2(DisasContext *s, arg_vldst_il *a) 418 { 419 static MVEGenLdStIlFn * const fns[4][4] = { 420 { F(vld20b), F(vld20h), F(vld20w), NULL, }, 421 { F(vld21b), F(vld21h), F(vld21w), NULL, }, 422 { NULL, NULL, NULL, NULL }, 423 { NULL, NULL, NULL, NULL }, 424 }; 425 if (a->qd > 6) { 426 return false; 427 } 428 return do_vldst_il(s, a, fns[a->pat][a->size], 32); 429 } 430 431 static bool trans_VLD4(DisasContext *s, arg_vldst_il *a) 432 { 433 static MVEGenLdStIlFn * const fns[4][4] = { 434 { F(vld40b), F(vld40h), F(vld40w), NULL, }, 435 { F(vld41b), F(vld41h), F(vld41w), NULL, }, 436 { F(vld42b), F(vld42h), F(vld42w), NULL, }, 437 { F(vld43b), F(vld43h), F(vld43w), NULL, }, 438 }; 439 if (a->qd > 4) { 440 return false; 441 } 442 return do_vldst_il(s, a, fns[a->pat][a->size], 64); 443 } 444 445 static bool trans_VST2(DisasContext *s, arg_vldst_il *a) 446 { 447 static MVEGenLdStIlFn * const fns[4][4] = { 448 { F(vst20b), F(vst20h), F(vst20w), NULL, }, 449 { F(vst21b), F(vst21h), F(vst21w), NULL, }, 450 { NULL, NULL, NULL, NULL }, 451 { NULL, NULL, NULL, NULL }, 452 }; 453 if (a->qd > 6) { 454 return false; 455 } 456 return do_vldst_il(s, a, fns[a->pat][a->size], 32); 457 } 458 459 static bool trans_VST4(DisasContext *s, arg_vldst_il *a) 460 { 461 static MVEGenLdStIlFn * const fns[4][4] = { 462 { F(vst40b), F(vst40h), F(vst40w), NULL, }, 463 { F(vst41b), F(vst41h), F(vst41w), NULL, }, 464 { F(vst42b), F(vst42h), F(vst42w), NULL, }, 465 { F(vst43b), F(vst43h), F(vst43w), NULL, }, 466 }; 467 if (a->qd > 4) { 468 return false; 469 } 470 return do_vldst_il(s, a, fns[a->pat][a->size], 64); 471 } 472 473 #undef F 474 475 static bool trans_VDUP(DisasContext *s, arg_VDUP *a) 476 { 477 TCGv_ptr qd; 478 TCGv_i32 rt; 479 480 if (!dc_isar_feature(aa32_mve, s) || 481 !mve_check_qreg_bank(s, a->qd)) { 482 return false; 483 } 484 if (a->rt == 13 || a->rt == 15) { 485 /* UNPREDICTABLE; we choose to UNDEF */ 486 return false; 487 } 488 if (!mve_eci_check(s) || !vfp_access_check(s)) { 489 return true; 490 } 491 492 qd = mve_qreg_ptr(a->qd); 493 rt = load_reg(s, a->rt); 494 tcg_gen_dup_i32(a->size, rt, rt); 495 gen_helper_mve_vdup(cpu_env, qd, rt); 496 tcg_temp_free_ptr(qd); 497 tcg_temp_free_i32(rt); 498 mve_update_eci(s); 499 return true; 500 } 501 502 static bool do_1op(DisasContext *s, arg_1op *a, MVEGenOneOpFn fn) 503 { 504 TCGv_ptr qd, qm; 505 506 if (!dc_isar_feature(aa32_mve, s) || 507 !mve_check_qreg_bank(s, a->qd | a->qm) || 508 !fn) { 509 return false; 510 } 511 512 if (!mve_eci_check(s) || !vfp_access_check(s)) { 513 return true; 514 } 515 516 qd = mve_qreg_ptr(a->qd); 517 qm = mve_qreg_ptr(a->qm); 518 fn(cpu_env, qd, qm); 519 tcg_temp_free_ptr(qd); 520 tcg_temp_free_ptr(qm); 521 mve_update_eci(s); 522 return true; 523 } 524 525 #define DO_1OP(INSN, FN) \ 526 static bool trans_##INSN(DisasContext *s, arg_1op *a) \ 527 { \ 528 static MVEGenOneOpFn * const fns[] = { \ 529 gen_helper_mve_##FN##b, \ 530 gen_helper_mve_##FN##h, \ 531 gen_helper_mve_##FN##w, \ 532 NULL, \ 533 }; \ 534 return do_1op(s, a, fns[a->size]); \ 535 } 536 537 DO_1OP(VCLZ, vclz) 538 DO_1OP(VCLS, vcls) 539 DO_1OP(VABS, vabs) 540 DO_1OP(VNEG, vneg) 541 DO_1OP(VQABS, vqabs) 542 DO_1OP(VQNEG, vqneg) 543 DO_1OP(VMAXA, vmaxa) 544 DO_1OP(VMINA, vmina) 545 546 /* Narrowing moves: only size 0 and 1 are valid */ 547 #define DO_VMOVN(INSN, FN) \ 548 static bool trans_##INSN(DisasContext *s, arg_1op *a) \ 549 { \ 550 static MVEGenOneOpFn * const fns[] = { \ 551 gen_helper_mve_##FN##b, \ 552 gen_helper_mve_##FN##h, \ 553 NULL, \ 554 NULL, \ 555 }; \ 556 return do_1op(s, a, fns[a->size]); \ 557 } 558 559 DO_VMOVN(VMOVNB, vmovnb) 560 DO_VMOVN(VMOVNT, vmovnt) 561 DO_VMOVN(VQMOVUNB, vqmovunb) 562 DO_VMOVN(VQMOVUNT, vqmovunt) 563 DO_VMOVN(VQMOVN_BS, vqmovnbs) 564 DO_VMOVN(VQMOVN_TS, vqmovnts) 565 DO_VMOVN(VQMOVN_BU, vqmovnbu) 566 DO_VMOVN(VQMOVN_TU, vqmovntu) 567 568 static bool trans_VREV16(DisasContext *s, arg_1op *a) 569 { 570 static MVEGenOneOpFn * const fns[] = { 571 gen_helper_mve_vrev16b, 572 NULL, 573 NULL, 574 NULL, 575 }; 576 return do_1op(s, a, fns[a->size]); 577 } 578 579 static bool trans_VREV32(DisasContext *s, arg_1op *a) 580 { 581 static MVEGenOneOpFn * const fns[] = { 582 gen_helper_mve_vrev32b, 583 gen_helper_mve_vrev32h, 584 NULL, 585 NULL, 586 }; 587 return do_1op(s, a, fns[a->size]); 588 } 589 590 static bool trans_VREV64(DisasContext *s, arg_1op *a) 591 { 592 static MVEGenOneOpFn * const fns[] = { 593 gen_helper_mve_vrev64b, 594 gen_helper_mve_vrev64h, 595 gen_helper_mve_vrev64w, 596 NULL, 597 }; 598 return do_1op(s, a, fns[a->size]); 599 } 600 601 static bool trans_VMVN(DisasContext *s, arg_1op *a) 602 { 603 return do_1op(s, a, gen_helper_mve_vmvn); 604 } 605 606 static bool trans_VABS_fp(DisasContext *s, arg_1op *a) 607 { 608 static MVEGenOneOpFn * const fns[] = { 609 NULL, 610 gen_helper_mve_vfabsh, 611 gen_helper_mve_vfabss, 612 NULL, 613 }; 614 if (!dc_isar_feature(aa32_mve_fp, s)) { 615 return false; 616 } 617 return do_1op(s, a, fns[a->size]); 618 } 619 620 static bool trans_VNEG_fp(DisasContext *s, arg_1op *a) 621 { 622 static MVEGenOneOpFn * const fns[] = { 623 NULL, 624 gen_helper_mve_vfnegh, 625 gen_helper_mve_vfnegs, 626 NULL, 627 }; 628 if (!dc_isar_feature(aa32_mve_fp, s)) { 629 return false; 630 } 631 return do_1op(s, a, fns[a->size]); 632 } 633 634 static bool do_2op(DisasContext *s, arg_2op *a, MVEGenTwoOpFn fn) 635 { 636 TCGv_ptr qd, qn, qm; 637 638 if (!dc_isar_feature(aa32_mve, s) || 639 !mve_check_qreg_bank(s, a->qd | a->qn | a->qm) || 640 !fn) { 641 return false; 642 } 643 if (!mve_eci_check(s) || !vfp_access_check(s)) { 644 return true; 645 } 646 647 qd = mve_qreg_ptr(a->qd); 648 qn = mve_qreg_ptr(a->qn); 649 qm = mve_qreg_ptr(a->qm); 650 fn(cpu_env, qd, qn, qm); 651 tcg_temp_free_ptr(qd); 652 tcg_temp_free_ptr(qn); 653 tcg_temp_free_ptr(qm); 654 mve_update_eci(s); 655 return true; 656 } 657 658 #define DO_LOGIC(INSN, HELPER) \ 659 static bool trans_##INSN(DisasContext *s, arg_2op *a) \ 660 { \ 661 return do_2op(s, a, HELPER); \ 662 } 663 664 DO_LOGIC(VAND, gen_helper_mve_vand) 665 DO_LOGIC(VBIC, gen_helper_mve_vbic) 666 DO_LOGIC(VORR, gen_helper_mve_vorr) 667 DO_LOGIC(VORN, gen_helper_mve_vorn) 668 DO_LOGIC(VEOR, gen_helper_mve_veor) 669 670 DO_LOGIC(VPSEL, gen_helper_mve_vpsel) 671 672 #define DO_2OP(INSN, FN) \ 673 static bool trans_##INSN(DisasContext *s, arg_2op *a) \ 674 { \ 675 static MVEGenTwoOpFn * const fns[] = { \ 676 gen_helper_mve_##FN##b, \ 677 gen_helper_mve_##FN##h, \ 678 gen_helper_mve_##FN##w, \ 679 NULL, \ 680 }; \ 681 return do_2op(s, a, fns[a->size]); \ 682 } 683 684 DO_2OP(VADD, vadd) 685 DO_2OP(VSUB, vsub) 686 DO_2OP(VMUL, vmul) 687 DO_2OP(VMULH_S, vmulhs) 688 DO_2OP(VMULH_U, vmulhu) 689 DO_2OP(VRMULH_S, vrmulhs) 690 DO_2OP(VRMULH_U, vrmulhu) 691 DO_2OP(VMAX_S, vmaxs) 692 DO_2OP(VMAX_U, vmaxu) 693 DO_2OP(VMIN_S, vmins) 694 DO_2OP(VMIN_U, vminu) 695 DO_2OP(VABD_S, vabds) 696 DO_2OP(VABD_U, vabdu) 697 DO_2OP(VHADD_S, vhadds) 698 DO_2OP(VHADD_U, vhaddu) 699 DO_2OP(VHSUB_S, vhsubs) 700 DO_2OP(VHSUB_U, vhsubu) 701 DO_2OP(VMULL_BS, vmullbs) 702 DO_2OP(VMULL_BU, vmullbu) 703 DO_2OP(VMULL_TS, vmullts) 704 DO_2OP(VMULL_TU, vmulltu) 705 DO_2OP(VQDMULH, vqdmulh) 706 DO_2OP(VQRDMULH, vqrdmulh) 707 DO_2OP(VQADD_S, vqadds) 708 DO_2OP(VQADD_U, vqaddu) 709 DO_2OP(VQSUB_S, vqsubs) 710 DO_2OP(VQSUB_U, vqsubu) 711 DO_2OP(VSHL_S, vshls) 712 DO_2OP(VSHL_U, vshlu) 713 DO_2OP(VRSHL_S, vrshls) 714 DO_2OP(VRSHL_U, vrshlu) 715 DO_2OP(VQSHL_S, vqshls) 716 DO_2OP(VQSHL_U, vqshlu) 717 DO_2OP(VQRSHL_S, vqrshls) 718 DO_2OP(VQRSHL_U, vqrshlu) 719 DO_2OP(VQDMLADH, vqdmladh) 720 DO_2OP(VQDMLADHX, vqdmladhx) 721 DO_2OP(VQRDMLADH, vqrdmladh) 722 DO_2OP(VQRDMLADHX, vqrdmladhx) 723 DO_2OP(VQDMLSDH, vqdmlsdh) 724 DO_2OP(VQDMLSDHX, vqdmlsdhx) 725 DO_2OP(VQRDMLSDH, vqrdmlsdh) 726 DO_2OP(VQRDMLSDHX, vqrdmlsdhx) 727 DO_2OP(VRHADD_S, vrhadds) 728 DO_2OP(VRHADD_U, vrhaddu) 729 /* 730 * VCADD Qd == Qm at size MO_32 is UNPREDICTABLE; we choose not to diagnose 731 * so we can reuse the DO_2OP macro. (Our implementation calculates the 732 * "expected" results in this case.) Similarly for VHCADD. 733 */ 734 DO_2OP(VCADD90, vcadd90) 735 DO_2OP(VCADD270, vcadd270) 736 DO_2OP(VHCADD90, vhcadd90) 737 DO_2OP(VHCADD270, vhcadd270) 738 739 static bool trans_VQDMULLB(DisasContext *s, arg_2op *a) 740 { 741 static MVEGenTwoOpFn * const fns[] = { 742 NULL, 743 gen_helper_mve_vqdmullbh, 744 gen_helper_mve_vqdmullbw, 745 NULL, 746 }; 747 if (a->size == MO_32 && (a->qd == a->qm || a->qd == a->qn)) { 748 /* UNPREDICTABLE; we choose to undef */ 749 return false; 750 } 751 return do_2op(s, a, fns[a->size]); 752 } 753 754 static bool trans_VQDMULLT(DisasContext *s, arg_2op *a) 755 { 756 static MVEGenTwoOpFn * const fns[] = { 757 NULL, 758 gen_helper_mve_vqdmullth, 759 gen_helper_mve_vqdmulltw, 760 NULL, 761 }; 762 if (a->size == MO_32 && (a->qd == a->qm || a->qd == a->qn)) { 763 /* UNPREDICTABLE; we choose to undef */ 764 return false; 765 } 766 return do_2op(s, a, fns[a->size]); 767 } 768 769 static bool trans_VMULLP_B(DisasContext *s, arg_2op *a) 770 { 771 /* 772 * Note that a->size indicates the output size, ie VMULL.P8 773 * is the 8x8->16 operation and a->size is MO_16; VMULL.P16 774 * is the 16x16->32 operation and a->size is MO_32. 775 */ 776 static MVEGenTwoOpFn * const fns[] = { 777 NULL, 778 gen_helper_mve_vmullpbh, 779 gen_helper_mve_vmullpbw, 780 NULL, 781 }; 782 return do_2op(s, a, fns[a->size]); 783 } 784 785 static bool trans_VMULLP_T(DisasContext *s, arg_2op *a) 786 { 787 /* a->size is as for trans_VMULLP_B */ 788 static MVEGenTwoOpFn * const fns[] = { 789 NULL, 790 gen_helper_mve_vmullpth, 791 gen_helper_mve_vmullptw, 792 NULL, 793 }; 794 return do_2op(s, a, fns[a->size]); 795 } 796 797 /* 798 * VADC and VSBC: these perform an add-with-carry or subtract-with-carry 799 * of the 32-bit elements in each lane of the input vectors, where the 800 * carry-out of each add is the carry-in of the next. The initial carry 801 * input is either fixed (0 for VADCI, 1 for VSBCI) or is from FPSCR.C 802 * (for VADC and VSBC); the carry out at the end is written back to FPSCR.C. 803 * These insns are subject to beat-wise execution. Partial execution 804 * of an I=1 (initial carry input fixed) insn which does not 805 * execute the first beat must start with the current FPSCR.NZCV 806 * value, not the fixed constant input. 807 */ 808 static bool trans_VADC(DisasContext *s, arg_2op *a) 809 { 810 return do_2op(s, a, gen_helper_mve_vadc); 811 } 812 813 static bool trans_VADCI(DisasContext *s, arg_2op *a) 814 { 815 if (mve_skip_first_beat(s)) { 816 return trans_VADC(s, a); 817 } 818 return do_2op(s, a, gen_helper_mve_vadci); 819 } 820 821 static bool trans_VSBC(DisasContext *s, arg_2op *a) 822 { 823 return do_2op(s, a, gen_helper_mve_vsbc); 824 } 825 826 static bool trans_VSBCI(DisasContext *s, arg_2op *a) 827 { 828 if (mve_skip_first_beat(s)) { 829 return trans_VSBC(s, a); 830 } 831 return do_2op(s, a, gen_helper_mve_vsbci); 832 } 833 834 #define DO_2OP_FP(INSN, FN) \ 835 static bool trans_##INSN(DisasContext *s, arg_2op *a) \ 836 { \ 837 static MVEGenTwoOpFn * const fns[] = { \ 838 NULL, \ 839 gen_helper_mve_##FN##h, \ 840 gen_helper_mve_##FN##s, \ 841 NULL, \ 842 }; \ 843 if (!dc_isar_feature(aa32_mve_fp, s)) { \ 844 return false; \ 845 } \ 846 return do_2op(s, a, fns[a->size]); \ 847 } 848 849 DO_2OP_FP(VADD_fp, vfadd) 850 851 static bool do_2op_scalar(DisasContext *s, arg_2scalar *a, 852 MVEGenTwoOpScalarFn fn) 853 { 854 TCGv_ptr qd, qn; 855 TCGv_i32 rm; 856 857 if (!dc_isar_feature(aa32_mve, s) || 858 !mve_check_qreg_bank(s, a->qd | a->qn) || 859 !fn) { 860 return false; 861 } 862 if (a->rm == 13 || a->rm == 15) { 863 /* UNPREDICTABLE */ 864 return false; 865 } 866 if (!mve_eci_check(s) || !vfp_access_check(s)) { 867 return true; 868 } 869 870 qd = mve_qreg_ptr(a->qd); 871 qn = mve_qreg_ptr(a->qn); 872 rm = load_reg(s, a->rm); 873 fn(cpu_env, qd, qn, rm); 874 tcg_temp_free_i32(rm); 875 tcg_temp_free_ptr(qd); 876 tcg_temp_free_ptr(qn); 877 mve_update_eci(s); 878 return true; 879 } 880 881 #define DO_2OP_SCALAR(INSN, FN) \ 882 static bool trans_##INSN(DisasContext *s, arg_2scalar *a) \ 883 { \ 884 static MVEGenTwoOpScalarFn * const fns[] = { \ 885 gen_helper_mve_##FN##b, \ 886 gen_helper_mve_##FN##h, \ 887 gen_helper_mve_##FN##w, \ 888 NULL, \ 889 }; \ 890 return do_2op_scalar(s, a, fns[a->size]); \ 891 } 892 893 DO_2OP_SCALAR(VADD_scalar, vadd_scalar) 894 DO_2OP_SCALAR(VSUB_scalar, vsub_scalar) 895 DO_2OP_SCALAR(VMUL_scalar, vmul_scalar) 896 DO_2OP_SCALAR(VHADD_S_scalar, vhadds_scalar) 897 DO_2OP_SCALAR(VHADD_U_scalar, vhaddu_scalar) 898 DO_2OP_SCALAR(VHSUB_S_scalar, vhsubs_scalar) 899 DO_2OP_SCALAR(VHSUB_U_scalar, vhsubu_scalar) 900 DO_2OP_SCALAR(VQADD_S_scalar, vqadds_scalar) 901 DO_2OP_SCALAR(VQADD_U_scalar, vqaddu_scalar) 902 DO_2OP_SCALAR(VQSUB_S_scalar, vqsubs_scalar) 903 DO_2OP_SCALAR(VQSUB_U_scalar, vqsubu_scalar) 904 DO_2OP_SCALAR(VQDMULH_scalar, vqdmulh_scalar) 905 DO_2OP_SCALAR(VQRDMULH_scalar, vqrdmulh_scalar) 906 DO_2OP_SCALAR(VBRSR, vbrsr) 907 DO_2OP_SCALAR(VMLA, vmla) 908 DO_2OP_SCALAR(VMLAS, vmlas) 909 DO_2OP_SCALAR(VQDMLAH, vqdmlah) 910 DO_2OP_SCALAR(VQRDMLAH, vqrdmlah) 911 DO_2OP_SCALAR(VQDMLASH, vqdmlash) 912 DO_2OP_SCALAR(VQRDMLASH, vqrdmlash) 913 914 static bool trans_VQDMULLB_scalar(DisasContext *s, arg_2scalar *a) 915 { 916 static MVEGenTwoOpScalarFn * const fns[] = { 917 NULL, 918 gen_helper_mve_vqdmullb_scalarh, 919 gen_helper_mve_vqdmullb_scalarw, 920 NULL, 921 }; 922 if (a->qd == a->qn && a->size == MO_32) { 923 /* UNPREDICTABLE; we choose to undef */ 924 return false; 925 } 926 return do_2op_scalar(s, a, fns[a->size]); 927 } 928 929 static bool trans_VQDMULLT_scalar(DisasContext *s, arg_2scalar *a) 930 { 931 static MVEGenTwoOpScalarFn * const fns[] = { 932 NULL, 933 gen_helper_mve_vqdmullt_scalarh, 934 gen_helper_mve_vqdmullt_scalarw, 935 NULL, 936 }; 937 if (a->qd == a->qn && a->size == MO_32) { 938 /* UNPREDICTABLE; we choose to undef */ 939 return false; 940 } 941 return do_2op_scalar(s, a, fns[a->size]); 942 } 943 944 static bool do_long_dual_acc(DisasContext *s, arg_vmlaldav *a, 945 MVEGenLongDualAccOpFn *fn) 946 { 947 TCGv_ptr qn, qm; 948 TCGv_i64 rda; 949 TCGv_i32 rdalo, rdahi; 950 951 if (!dc_isar_feature(aa32_mve, s) || 952 !mve_check_qreg_bank(s, a->qn | a->qm) || 953 !fn) { 954 return false; 955 } 956 /* 957 * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related 958 * encoding; rdalo always has bit 0 clear so cannot be 13 or 15. 959 */ 960 if (a->rdahi == 13 || a->rdahi == 15) { 961 return false; 962 } 963 if (!mve_eci_check(s) || !vfp_access_check(s)) { 964 return true; 965 } 966 967 qn = mve_qreg_ptr(a->qn); 968 qm = mve_qreg_ptr(a->qm); 969 970 /* 971 * This insn is subject to beat-wise execution. Partial execution 972 * of an A=0 (no-accumulate) insn which does not execute the first 973 * beat must start with the current rda value, not 0. 974 */ 975 if (a->a || mve_skip_first_beat(s)) { 976 rda = tcg_temp_new_i64(); 977 rdalo = load_reg(s, a->rdalo); 978 rdahi = load_reg(s, a->rdahi); 979 tcg_gen_concat_i32_i64(rda, rdalo, rdahi); 980 tcg_temp_free_i32(rdalo); 981 tcg_temp_free_i32(rdahi); 982 } else { 983 rda = tcg_const_i64(0); 984 } 985 986 fn(rda, cpu_env, qn, qm, rda); 987 tcg_temp_free_ptr(qn); 988 tcg_temp_free_ptr(qm); 989 990 rdalo = tcg_temp_new_i32(); 991 rdahi = tcg_temp_new_i32(); 992 tcg_gen_extrl_i64_i32(rdalo, rda); 993 tcg_gen_extrh_i64_i32(rdahi, rda); 994 store_reg(s, a->rdalo, rdalo); 995 store_reg(s, a->rdahi, rdahi); 996 tcg_temp_free_i64(rda); 997 mve_update_eci(s); 998 return true; 999 } 1000 1001 static bool trans_VMLALDAV_S(DisasContext *s, arg_vmlaldav *a) 1002 { 1003 static MVEGenLongDualAccOpFn * const fns[4][2] = { 1004 { NULL, NULL }, 1005 { gen_helper_mve_vmlaldavsh, gen_helper_mve_vmlaldavxsh }, 1006 { gen_helper_mve_vmlaldavsw, gen_helper_mve_vmlaldavxsw }, 1007 { NULL, NULL }, 1008 }; 1009 return do_long_dual_acc(s, a, fns[a->size][a->x]); 1010 } 1011 1012 static bool trans_VMLALDAV_U(DisasContext *s, arg_vmlaldav *a) 1013 { 1014 static MVEGenLongDualAccOpFn * const fns[4][2] = { 1015 { NULL, NULL }, 1016 { gen_helper_mve_vmlaldavuh, NULL }, 1017 { gen_helper_mve_vmlaldavuw, NULL }, 1018 { NULL, NULL }, 1019 }; 1020 return do_long_dual_acc(s, a, fns[a->size][a->x]); 1021 } 1022 1023 static bool trans_VMLSLDAV(DisasContext *s, arg_vmlaldav *a) 1024 { 1025 static MVEGenLongDualAccOpFn * const fns[4][2] = { 1026 { NULL, NULL }, 1027 { gen_helper_mve_vmlsldavsh, gen_helper_mve_vmlsldavxsh }, 1028 { gen_helper_mve_vmlsldavsw, gen_helper_mve_vmlsldavxsw }, 1029 { NULL, NULL }, 1030 }; 1031 return do_long_dual_acc(s, a, fns[a->size][a->x]); 1032 } 1033 1034 static bool trans_VRMLALDAVH_S(DisasContext *s, arg_vmlaldav *a) 1035 { 1036 static MVEGenLongDualAccOpFn * const fns[] = { 1037 gen_helper_mve_vrmlaldavhsw, gen_helper_mve_vrmlaldavhxsw, 1038 }; 1039 return do_long_dual_acc(s, a, fns[a->x]); 1040 } 1041 1042 static bool trans_VRMLALDAVH_U(DisasContext *s, arg_vmlaldav *a) 1043 { 1044 static MVEGenLongDualAccOpFn * const fns[] = { 1045 gen_helper_mve_vrmlaldavhuw, NULL, 1046 }; 1047 return do_long_dual_acc(s, a, fns[a->x]); 1048 } 1049 1050 static bool trans_VRMLSLDAVH(DisasContext *s, arg_vmlaldav *a) 1051 { 1052 static MVEGenLongDualAccOpFn * const fns[] = { 1053 gen_helper_mve_vrmlsldavhsw, gen_helper_mve_vrmlsldavhxsw, 1054 }; 1055 return do_long_dual_acc(s, a, fns[a->x]); 1056 } 1057 1058 static bool do_dual_acc(DisasContext *s, arg_vmladav *a, MVEGenDualAccOpFn *fn) 1059 { 1060 TCGv_ptr qn, qm; 1061 TCGv_i32 rda; 1062 1063 if (!dc_isar_feature(aa32_mve, s) || 1064 !mve_check_qreg_bank(s, a->qn) || 1065 !fn) { 1066 return false; 1067 } 1068 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1069 return true; 1070 } 1071 1072 qn = mve_qreg_ptr(a->qn); 1073 qm = mve_qreg_ptr(a->qm); 1074 1075 /* 1076 * This insn is subject to beat-wise execution. Partial execution 1077 * of an A=0 (no-accumulate) insn which does not execute the first 1078 * beat must start with the current rda value, not 0. 1079 */ 1080 if (a->a || mve_skip_first_beat(s)) { 1081 rda = load_reg(s, a->rda); 1082 } else { 1083 rda = tcg_const_i32(0); 1084 } 1085 1086 fn(rda, cpu_env, qn, qm, rda); 1087 store_reg(s, a->rda, rda); 1088 tcg_temp_free_ptr(qn); 1089 tcg_temp_free_ptr(qm); 1090 1091 mve_update_eci(s); 1092 return true; 1093 } 1094 1095 #define DO_DUAL_ACC(INSN, FN) \ 1096 static bool trans_##INSN(DisasContext *s, arg_vmladav *a) \ 1097 { \ 1098 static MVEGenDualAccOpFn * const fns[4][2] = { \ 1099 { gen_helper_mve_##FN##b, gen_helper_mve_##FN##xb }, \ 1100 { gen_helper_mve_##FN##h, gen_helper_mve_##FN##xh }, \ 1101 { gen_helper_mve_##FN##w, gen_helper_mve_##FN##xw }, \ 1102 { NULL, NULL }, \ 1103 }; \ 1104 return do_dual_acc(s, a, fns[a->size][a->x]); \ 1105 } 1106 1107 DO_DUAL_ACC(VMLADAV_S, vmladavs) 1108 DO_DUAL_ACC(VMLSDAV, vmlsdav) 1109 1110 static bool trans_VMLADAV_U(DisasContext *s, arg_vmladav *a) 1111 { 1112 static MVEGenDualAccOpFn * const fns[4][2] = { 1113 { gen_helper_mve_vmladavub, NULL }, 1114 { gen_helper_mve_vmladavuh, NULL }, 1115 { gen_helper_mve_vmladavuw, NULL }, 1116 { NULL, NULL }, 1117 }; 1118 return do_dual_acc(s, a, fns[a->size][a->x]); 1119 } 1120 1121 static void gen_vpst(DisasContext *s, uint32_t mask) 1122 { 1123 /* 1124 * Set the VPR mask fields. We take advantage of MASK01 and MASK23 1125 * being adjacent fields in the register. 1126 * 1127 * Updating the masks is not predicated, but it is subject to beat-wise 1128 * execution, and the mask is updated on the odd-numbered beats. 1129 * So if PSR.ECI says we should skip beat 1, we mustn't update the 1130 * 01 mask field. 1131 */ 1132 TCGv_i32 vpr = load_cpu_field(v7m.vpr); 1133 switch (s->eci) { 1134 case ECI_NONE: 1135 case ECI_A0: 1136 /* Update both 01 and 23 fields */ 1137 tcg_gen_deposit_i32(vpr, vpr, 1138 tcg_constant_i32(mask | (mask << 4)), 1139 R_V7M_VPR_MASK01_SHIFT, 1140 R_V7M_VPR_MASK01_LENGTH + R_V7M_VPR_MASK23_LENGTH); 1141 break; 1142 case ECI_A0A1: 1143 case ECI_A0A1A2: 1144 case ECI_A0A1A2B0: 1145 /* Update only the 23 mask field */ 1146 tcg_gen_deposit_i32(vpr, vpr, 1147 tcg_constant_i32(mask), 1148 R_V7M_VPR_MASK23_SHIFT, R_V7M_VPR_MASK23_LENGTH); 1149 break; 1150 default: 1151 g_assert_not_reached(); 1152 } 1153 store_cpu_field(vpr, v7m.vpr); 1154 } 1155 1156 static bool trans_VPST(DisasContext *s, arg_VPST *a) 1157 { 1158 /* mask == 0 is a "related encoding" */ 1159 if (!dc_isar_feature(aa32_mve, s) || !a->mask) { 1160 return false; 1161 } 1162 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1163 return true; 1164 } 1165 gen_vpst(s, a->mask); 1166 mve_update_and_store_eci(s); 1167 return true; 1168 } 1169 1170 static bool trans_VPNOT(DisasContext *s, arg_VPNOT *a) 1171 { 1172 /* 1173 * Invert the predicate in VPR.P0. We have call out to 1174 * a helper because this insn itself is beatwise and can 1175 * be predicated. 1176 */ 1177 if (!dc_isar_feature(aa32_mve, s)) { 1178 return false; 1179 } 1180 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1181 return true; 1182 } 1183 1184 gen_helper_mve_vpnot(cpu_env); 1185 mve_update_eci(s); 1186 return true; 1187 } 1188 1189 static bool trans_VADDV(DisasContext *s, arg_VADDV *a) 1190 { 1191 /* VADDV: vector add across vector */ 1192 static MVEGenVADDVFn * const fns[4][2] = { 1193 { gen_helper_mve_vaddvsb, gen_helper_mve_vaddvub }, 1194 { gen_helper_mve_vaddvsh, gen_helper_mve_vaddvuh }, 1195 { gen_helper_mve_vaddvsw, gen_helper_mve_vaddvuw }, 1196 { NULL, NULL } 1197 }; 1198 TCGv_ptr qm; 1199 TCGv_i32 rda; 1200 1201 if (!dc_isar_feature(aa32_mve, s) || 1202 a->size == 3) { 1203 return false; 1204 } 1205 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1206 return true; 1207 } 1208 1209 /* 1210 * This insn is subject to beat-wise execution. Partial execution 1211 * of an A=0 (no-accumulate) insn which does not execute the first 1212 * beat must start with the current value of Rda, not zero. 1213 */ 1214 if (a->a || mve_skip_first_beat(s)) { 1215 /* Accumulate input from Rda */ 1216 rda = load_reg(s, a->rda); 1217 } else { 1218 /* Accumulate starting at zero */ 1219 rda = tcg_const_i32(0); 1220 } 1221 1222 qm = mve_qreg_ptr(a->qm); 1223 fns[a->size][a->u](rda, cpu_env, qm, rda); 1224 store_reg(s, a->rda, rda); 1225 tcg_temp_free_ptr(qm); 1226 1227 mve_update_eci(s); 1228 return true; 1229 } 1230 1231 static bool trans_VADDLV(DisasContext *s, arg_VADDLV *a) 1232 { 1233 /* 1234 * Vector Add Long Across Vector: accumulate the 32-bit 1235 * elements of the vector into a 64-bit result stored in 1236 * a pair of general-purpose registers. 1237 * No need to check Qm's bank: it is only 3 bits in decode. 1238 */ 1239 TCGv_ptr qm; 1240 TCGv_i64 rda; 1241 TCGv_i32 rdalo, rdahi; 1242 1243 if (!dc_isar_feature(aa32_mve, s)) { 1244 return false; 1245 } 1246 /* 1247 * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related 1248 * encoding; rdalo always has bit 0 clear so cannot be 13 or 15. 1249 */ 1250 if (a->rdahi == 13 || a->rdahi == 15) { 1251 return false; 1252 } 1253 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1254 return true; 1255 } 1256 1257 /* 1258 * This insn is subject to beat-wise execution. Partial execution 1259 * of an A=0 (no-accumulate) insn which does not execute the first 1260 * beat must start with the current value of RdaHi:RdaLo, not zero. 1261 */ 1262 if (a->a || mve_skip_first_beat(s)) { 1263 /* Accumulate input from RdaHi:RdaLo */ 1264 rda = tcg_temp_new_i64(); 1265 rdalo = load_reg(s, a->rdalo); 1266 rdahi = load_reg(s, a->rdahi); 1267 tcg_gen_concat_i32_i64(rda, rdalo, rdahi); 1268 tcg_temp_free_i32(rdalo); 1269 tcg_temp_free_i32(rdahi); 1270 } else { 1271 /* Accumulate starting at zero */ 1272 rda = tcg_const_i64(0); 1273 } 1274 1275 qm = mve_qreg_ptr(a->qm); 1276 if (a->u) { 1277 gen_helper_mve_vaddlv_u(rda, cpu_env, qm, rda); 1278 } else { 1279 gen_helper_mve_vaddlv_s(rda, cpu_env, qm, rda); 1280 } 1281 tcg_temp_free_ptr(qm); 1282 1283 rdalo = tcg_temp_new_i32(); 1284 rdahi = tcg_temp_new_i32(); 1285 tcg_gen_extrl_i64_i32(rdalo, rda); 1286 tcg_gen_extrh_i64_i32(rdahi, rda); 1287 store_reg(s, a->rdalo, rdalo); 1288 store_reg(s, a->rdahi, rdahi); 1289 tcg_temp_free_i64(rda); 1290 mve_update_eci(s); 1291 return true; 1292 } 1293 1294 static bool do_1imm(DisasContext *s, arg_1imm *a, MVEGenOneOpImmFn *fn) 1295 { 1296 TCGv_ptr qd; 1297 uint64_t imm; 1298 1299 if (!dc_isar_feature(aa32_mve, s) || 1300 !mve_check_qreg_bank(s, a->qd) || 1301 !fn) { 1302 return false; 1303 } 1304 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1305 return true; 1306 } 1307 1308 imm = asimd_imm_const(a->imm, a->cmode, a->op); 1309 1310 qd = mve_qreg_ptr(a->qd); 1311 fn(cpu_env, qd, tcg_constant_i64(imm)); 1312 tcg_temp_free_ptr(qd); 1313 mve_update_eci(s); 1314 return true; 1315 } 1316 1317 static bool trans_Vimm_1r(DisasContext *s, arg_1imm *a) 1318 { 1319 /* Handle decode of cmode/op here between VORR/VBIC/VMOV */ 1320 MVEGenOneOpImmFn *fn; 1321 1322 if ((a->cmode & 1) && a->cmode < 12) { 1323 if (a->op) { 1324 /* 1325 * For op=1, the immediate will be inverted by asimd_imm_const(), 1326 * so the VBIC becomes a logical AND operation. 1327 */ 1328 fn = gen_helper_mve_vandi; 1329 } else { 1330 fn = gen_helper_mve_vorri; 1331 } 1332 } else { 1333 /* There is one unallocated cmode/op combination in this space */ 1334 if (a->cmode == 15 && a->op == 1) { 1335 return false; 1336 } 1337 /* asimd_imm_const() sorts out VMVNI vs VMOVI for us */ 1338 fn = gen_helper_mve_vmovi; 1339 } 1340 return do_1imm(s, a, fn); 1341 } 1342 1343 static bool do_2shift(DisasContext *s, arg_2shift *a, MVEGenTwoOpShiftFn fn, 1344 bool negateshift) 1345 { 1346 TCGv_ptr qd, qm; 1347 int shift = a->shift; 1348 1349 if (!dc_isar_feature(aa32_mve, s) || 1350 !mve_check_qreg_bank(s, a->qd | a->qm) || 1351 !fn) { 1352 return false; 1353 } 1354 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1355 return true; 1356 } 1357 1358 /* 1359 * When we handle a right shift insn using a left-shift helper 1360 * which permits a negative shift count to indicate a right-shift, 1361 * we must negate the shift count. 1362 */ 1363 if (negateshift) { 1364 shift = -shift; 1365 } 1366 1367 qd = mve_qreg_ptr(a->qd); 1368 qm = mve_qreg_ptr(a->qm); 1369 fn(cpu_env, qd, qm, tcg_constant_i32(shift)); 1370 tcg_temp_free_ptr(qd); 1371 tcg_temp_free_ptr(qm); 1372 mve_update_eci(s); 1373 return true; 1374 } 1375 1376 #define DO_2SHIFT(INSN, FN, NEGATESHIFT) \ 1377 static bool trans_##INSN(DisasContext *s, arg_2shift *a) \ 1378 { \ 1379 static MVEGenTwoOpShiftFn * const fns[] = { \ 1380 gen_helper_mve_##FN##b, \ 1381 gen_helper_mve_##FN##h, \ 1382 gen_helper_mve_##FN##w, \ 1383 NULL, \ 1384 }; \ 1385 return do_2shift(s, a, fns[a->size], NEGATESHIFT); \ 1386 } 1387 1388 DO_2SHIFT(VSHLI, vshli_u, false) 1389 DO_2SHIFT(VQSHLI_S, vqshli_s, false) 1390 DO_2SHIFT(VQSHLI_U, vqshli_u, false) 1391 DO_2SHIFT(VQSHLUI, vqshlui_s, false) 1392 /* These right shifts use a left-shift helper with negated shift count */ 1393 DO_2SHIFT(VSHRI_S, vshli_s, true) 1394 DO_2SHIFT(VSHRI_U, vshli_u, true) 1395 DO_2SHIFT(VRSHRI_S, vrshli_s, true) 1396 DO_2SHIFT(VRSHRI_U, vrshli_u, true) 1397 1398 DO_2SHIFT(VSRI, vsri, false) 1399 DO_2SHIFT(VSLI, vsli, false) 1400 1401 static bool do_2shift_scalar(DisasContext *s, arg_shl_scalar *a, 1402 MVEGenTwoOpShiftFn *fn) 1403 { 1404 TCGv_ptr qda; 1405 TCGv_i32 rm; 1406 1407 if (!dc_isar_feature(aa32_mve, s) || 1408 !mve_check_qreg_bank(s, a->qda) || 1409 a->rm == 13 || a->rm == 15 || !fn) { 1410 /* Rm cases are UNPREDICTABLE */ 1411 return false; 1412 } 1413 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1414 return true; 1415 } 1416 1417 qda = mve_qreg_ptr(a->qda); 1418 rm = load_reg(s, a->rm); 1419 fn(cpu_env, qda, qda, rm); 1420 tcg_temp_free_ptr(qda); 1421 tcg_temp_free_i32(rm); 1422 mve_update_eci(s); 1423 return true; 1424 } 1425 1426 #define DO_2SHIFT_SCALAR(INSN, FN) \ 1427 static bool trans_##INSN(DisasContext *s, arg_shl_scalar *a) \ 1428 { \ 1429 static MVEGenTwoOpShiftFn * const fns[] = { \ 1430 gen_helper_mve_##FN##b, \ 1431 gen_helper_mve_##FN##h, \ 1432 gen_helper_mve_##FN##w, \ 1433 NULL, \ 1434 }; \ 1435 return do_2shift_scalar(s, a, fns[a->size]); \ 1436 } 1437 1438 DO_2SHIFT_SCALAR(VSHL_S_scalar, vshli_s) 1439 DO_2SHIFT_SCALAR(VSHL_U_scalar, vshli_u) 1440 DO_2SHIFT_SCALAR(VRSHL_S_scalar, vrshli_s) 1441 DO_2SHIFT_SCALAR(VRSHL_U_scalar, vrshli_u) 1442 DO_2SHIFT_SCALAR(VQSHL_S_scalar, vqshli_s) 1443 DO_2SHIFT_SCALAR(VQSHL_U_scalar, vqshli_u) 1444 DO_2SHIFT_SCALAR(VQRSHL_S_scalar, vqrshli_s) 1445 DO_2SHIFT_SCALAR(VQRSHL_U_scalar, vqrshli_u) 1446 1447 #define DO_VSHLL(INSN, FN) \ 1448 static bool trans_##INSN(DisasContext *s, arg_2shift *a) \ 1449 { \ 1450 static MVEGenTwoOpShiftFn * const fns[] = { \ 1451 gen_helper_mve_##FN##b, \ 1452 gen_helper_mve_##FN##h, \ 1453 }; \ 1454 return do_2shift(s, a, fns[a->size], false); \ 1455 } 1456 1457 DO_VSHLL(VSHLL_BS, vshllbs) 1458 DO_VSHLL(VSHLL_BU, vshllbu) 1459 DO_VSHLL(VSHLL_TS, vshllts) 1460 DO_VSHLL(VSHLL_TU, vshlltu) 1461 1462 #define DO_2SHIFT_N(INSN, FN) \ 1463 static bool trans_##INSN(DisasContext *s, arg_2shift *a) \ 1464 { \ 1465 static MVEGenTwoOpShiftFn * const fns[] = { \ 1466 gen_helper_mve_##FN##b, \ 1467 gen_helper_mve_##FN##h, \ 1468 }; \ 1469 return do_2shift(s, a, fns[a->size], false); \ 1470 } 1471 1472 DO_2SHIFT_N(VSHRNB, vshrnb) 1473 DO_2SHIFT_N(VSHRNT, vshrnt) 1474 DO_2SHIFT_N(VRSHRNB, vrshrnb) 1475 DO_2SHIFT_N(VRSHRNT, vrshrnt) 1476 DO_2SHIFT_N(VQSHRNB_S, vqshrnb_s) 1477 DO_2SHIFT_N(VQSHRNT_S, vqshrnt_s) 1478 DO_2SHIFT_N(VQSHRNB_U, vqshrnb_u) 1479 DO_2SHIFT_N(VQSHRNT_U, vqshrnt_u) 1480 DO_2SHIFT_N(VQSHRUNB, vqshrunb) 1481 DO_2SHIFT_N(VQSHRUNT, vqshrunt) 1482 DO_2SHIFT_N(VQRSHRNB_S, vqrshrnb_s) 1483 DO_2SHIFT_N(VQRSHRNT_S, vqrshrnt_s) 1484 DO_2SHIFT_N(VQRSHRNB_U, vqrshrnb_u) 1485 DO_2SHIFT_N(VQRSHRNT_U, vqrshrnt_u) 1486 DO_2SHIFT_N(VQRSHRUNB, vqrshrunb) 1487 DO_2SHIFT_N(VQRSHRUNT, vqrshrunt) 1488 1489 static bool trans_VSHLC(DisasContext *s, arg_VSHLC *a) 1490 { 1491 /* 1492 * Whole Vector Left Shift with Carry. The carry is taken 1493 * from a general purpose register and written back there. 1494 * An imm of 0 means "shift by 32". 1495 */ 1496 TCGv_ptr qd; 1497 TCGv_i32 rdm; 1498 1499 if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) { 1500 return false; 1501 } 1502 if (a->rdm == 13 || a->rdm == 15) { 1503 /* CONSTRAINED UNPREDICTABLE: we UNDEF */ 1504 return false; 1505 } 1506 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1507 return true; 1508 } 1509 1510 qd = mve_qreg_ptr(a->qd); 1511 rdm = load_reg(s, a->rdm); 1512 gen_helper_mve_vshlc(rdm, cpu_env, qd, rdm, tcg_constant_i32(a->imm)); 1513 store_reg(s, a->rdm, rdm); 1514 tcg_temp_free_ptr(qd); 1515 mve_update_eci(s); 1516 return true; 1517 } 1518 1519 static bool do_vidup(DisasContext *s, arg_vidup *a, MVEGenVIDUPFn *fn) 1520 { 1521 TCGv_ptr qd; 1522 TCGv_i32 rn; 1523 1524 /* 1525 * Vector increment/decrement with wrap and duplicate (VIDUP, VDDUP). 1526 * This fills the vector with elements of successively increasing 1527 * or decreasing values, starting from Rn. 1528 */ 1529 if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) { 1530 return false; 1531 } 1532 if (a->size == MO_64) { 1533 /* size 0b11 is another encoding */ 1534 return false; 1535 } 1536 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1537 return true; 1538 } 1539 1540 qd = mve_qreg_ptr(a->qd); 1541 rn = load_reg(s, a->rn); 1542 fn(rn, cpu_env, qd, rn, tcg_constant_i32(a->imm)); 1543 store_reg(s, a->rn, rn); 1544 tcg_temp_free_ptr(qd); 1545 mve_update_eci(s); 1546 return true; 1547 } 1548 1549 static bool do_viwdup(DisasContext *s, arg_viwdup *a, MVEGenVIWDUPFn *fn) 1550 { 1551 TCGv_ptr qd; 1552 TCGv_i32 rn, rm; 1553 1554 /* 1555 * Vector increment/decrement with wrap and duplicate (VIWDUp, VDWDUP) 1556 * This fills the vector with elements of successively increasing 1557 * or decreasing values, starting from Rn. Rm specifies a point where 1558 * the count wraps back around to 0. The updated offset is written back 1559 * to Rn. 1560 */ 1561 if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) { 1562 return false; 1563 } 1564 if (!fn || a->rm == 13 || a->rm == 15) { 1565 /* 1566 * size 0b11 is another encoding; Rm == 13 is UNPREDICTABLE; 1567 * Rm == 13 is VIWDUP, VDWDUP. 1568 */ 1569 return false; 1570 } 1571 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1572 return true; 1573 } 1574 1575 qd = mve_qreg_ptr(a->qd); 1576 rn = load_reg(s, a->rn); 1577 rm = load_reg(s, a->rm); 1578 fn(rn, cpu_env, qd, rn, rm, tcg_constant_i32(a->imm)); 1579 store_reg(s, a->rn, rn); 1580 tcg_temp_free_ptr(qd); 1581 tcg_temp_free_i32(rm); 1582 mve_update_eci(s); 1583 return true; 1584 } 1585 1586 static bool trans_VIDUP(DisasContext *s, arg_vidup *a) 1587 { 1588 static MVEGenVIDUPFn * const fns[] = { 1589 gen_helper_mve_vidupb, 1590 gen_helper_mve_viduph, 1591 gen_helper_mve_vidupw, 1592 NULL, 1593 }; 1594 return do_vidup(s, a, fns[a->size]); 1595 } 1596 1597 static bool trans_VDDUP(DisasContext *s, arg_vidup *a) 1598 { 1599 static MVEGenVIDUPFn * const fns[] = { 1600 gen_helper_mve_vidupb, 1601 gen_helper_mve_viduph, 1602 gen_helper_mve_vidupw, 1603 NULL, 1604 }; 1605 /* VDDUP is just like VIDUP but with a negative immediate */ 1606 a->imm = -a->imm; 1607 return do_vidup(s, a, fns[a->size]); 1608 } 1609 1610 static bool trans_VIWDUP(DisasContext *s, arg_viwdup *a) 1611 { 1612 static MVEGenVIWDUPFn * const fns[] = { 1613 gen_helper_mve_viwdupb, 1614 gen_helper_mve_viwduph, 1615 gen_helper_mve_viwdupw, 1616 NULL, 1617 }; 1618 return do_viwdup(s, a, fns[a->size]); 1619 } 1620 1621 static bool trans_VDWDUP(DisasContext *s, arg_viwdup *a) 1622 { 1623 static MVEGenVIWDUPFn * const fns[] = { 1624 gen_helper_mve_vdwdupb, 1625 gen_helper_mve_vdwduph, 1626 gen_helper_mve_vdwdupw, 1627 NULL, 1628 }; 1629 return do_viwdup(s, a, fns[a->size]); 1630 } 1631 1632 static bool do_vcmp(DisasContext *s, arg_vcmp *a, MVEGenCmpFn *fn) 1633 { 1634 TCGv_ptr qn, qm; 1635 1636 if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) || 1637 !fn) { 1638 return false; 1639 } 1640 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1641 return true; 1642 } 1643 1644 qn = mve_qreg_ptr(a->qn); 1645 qm = mve_qreg_ptr(a->qm); 1646 fn(cpu_env, qn, qm); 1647 tcg_temp_free_ptr(qn); 1648 tcg_temp_free_ptr(qm); 1649 if (a->mask) { 1650 /* VPT */ 1651 gen_vpst(s, a->mask); 1652 } 1653 mve_update_eci(s); 1654 return true; 1655 } 1656 1657 static bool do_vcmp_scalar(DisasContext *s, arg_vcmp_scalar *a, 1658 MVEGenScalarCmpFn *fn) 1659 { 1660 TCGv_ptr qn; 1661 TCGv_i32 rm; 1662 1663 if (!dc_isar_feature(aa32_mve, s) || !fn || a->rm == 13) { 1664 return false; 1665 } 1666 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1667 return true; 1668 } 1669 1670 qn = mve_qreg_ptr(a->qn); 1671 if (a->rm == 15) { 1672 /* Encoding Rm=0b1111 means "constant zero" */ 1673 rm = tcg_constant_i32(0); 1674 } else { 1675 rm = load_reg(s, a->rm); 1676 } 1677 fn(cpu_env, qn, rm); 1678 tcg_temp_free_ptr(qn); 1679 tcg_temp_free_i32(rm); 1680 if (a->mask) { 1681 /* VPT */ 1682 gen_vpst(s, a->mask); 1683 } 1684 mve_update_eci(s); 1685 return true; 1686 } 1687 1688 #define DO_VCMP(INSN, FN) \ 1689 static bool trans_##INSN(DisasContext *s, arg_vcmp *a) \ 1690 { \ 1691 static MVEGenCmpFn * const fns[] = { \ 1692 gen_helper_mve_##FN##b, \ 1693 gen_helper_mve_##FN##h, \ 1694 gen_helper_mve_##FN##w, \ 1695 NULL, \ 1696 }; \ 1697 return do_vcmp(s, a, fns[a->size]); \ 1698 } \ 1699 static bool trans_##INSN##_scalar(DisasContext *s, \ 1700 arg_vcmp_scalar *a) \ 1701 { \ 1702 static MVEGenScalarCmpFn * const fns[] = { \ 1703 gen_helper_mve_##FN##_scalarb, \ 1704 gen_helper_mve_##FN##_scalarh, \ 1705 gen_helper_mve_##FN##_scalarw, \ 1706 NULL, \ 1707 }; \ 1708 return do_vcmp_scalar(s, a, fns[a->size]); \ 1709 } 1710 1711 DO_VCMP(VCMPEQ, vcmpeq) 1712 DO_VCMP(VCMPNE, vcmpne) 1713 DO_VCMP(VCMPCS, vcmpcs) 1714 DO_VCMP(VCMPHI, vcmphi) 1715 DO_VCMP(VCMPGE, vcmpge) 1716 DO_VCMP(VCMPLT, vcmplt) 1717 DO_VCMP(VCMPGT, vcmpgt) 1718 DO_VCMP(VCMPLE, vcmple) 1719 1720 static bool do_vmaxv(DisasContext *s, arg_vmaxv *a, MVEGenVADDVFn fn) 1721 { 1722 /* 1723 * MIN/MAX operations across a vector: compute the min or 1724 * max of the initial value in a general purpose register 1725 * and all the elements in the vector, and store it back 1726 * into the general purpose register. 1727 */ 1728 TCGv_ptr qm; 1729 TCGv_i32 rda; 1730 1731 if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) || 1732 !fn || a->rda == 13 || a->rda == 15) { 1733 /* Rda cases are UNPREDICTABLE */ 1734 return false; 1735 } 1736 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1737 return true; 1738 } 1739 1740 qm = mve_qreg_ptr(a->qm); 1741 rda = load_reg(s, a->rda); 1742 fn(rda, cpu_env, qm, rda); 1743 store_reg(s, a->rda, rda); 1744 tcg_temp_free_ptr(qm); 1745 mve_update_eci(s); 1746 return true; 1747 } 1748 1749 #define DO_VMAXV(INSN, FN) \ 1750 static bool trans_##INSN(DisasContext *s, arg_vmaxv *a) \ 1751 { \ 1752 static MVEGenVADDVFn * const fns[] = { \ 1753 gen_helper_mve_##FN##b, \ 1754 gen_helper_mve_##FN##h, \ 1755 gen_helper_mve_##FN##w, \ 1756 NULL, \ 1757 }; \ 1758 return do_vmaxv(s, a, fns[a->size]); \ 1759 } 1760 1761 DO_VMAXV(VMAXV_S, vmaxvs) 1762 DO_VMAXV(VMAXV_U, vmaxvu) 1763 DO_VMAXV(VMAXAV, vmaxav) 1764 DO_VMAXV(VMINV_S, vminvs) 1765 DO_VMAXV(VMINV_U, vminvu) 1766 DO_VMAXV(VMINAV, vminav) 1767 1768 static bool do_vabav(DisasContext *s, arg_vabav *a, MVEGenVABAVFn *fn) 1769 { 1770 /* Absolute difference accumulated across vector */ 1771 TCGv_ptr qn, qm; 1772 TCGv_i32 rda; 1773 1774 if (!dc_isar_feature(aa32_mve, s) || 1775 !mve_check_qreg_bank(s, a->qm | a->qn) || 1776 !fn || a->rda == 13 || a->rda == 15) { 1777 /* Rda cases are UNPREDICTABLE */ 1778 return false; 1779 } 1780 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1781 return true; 1782 } 1783 1784 qm = mve_qreg_ptr(a->qm); 1785 qn = mve_qreg_ptr(a->qn); 1786 rda = load_reg(s, a->rda); 1787 fn(rda, cpu_env, qn, qm, rda); 1788 store_reg(s, a->rda, rda); 1789 tcg_temp_free_ptr(qm); 1790 tcg_temp_free_ptr(qn); 1791 mve_update_eci(s); 1792 return true; 1793 } 1794 1795 #define DO_VABAV(INSN, FN) \ 1796 static bool trans_##INSN(DisasContext *s, arg_vabav *a) \ 1797 { \ 1798 static MVEGenVABAVFn * const fns[] = { \ 1799 gen_helper_mve_##FN##b, \ 1800 gen_helper_mve_##FN##h, \ 1801 gen_helper_mve_##FN##w, \ 1802 NULL, \ 1803 }; \ 1804 return do_vabav(s, a, fns[a->size]); \ 1805 } 1806 1807 DO_VABAV(VABAV_S, vabavs) 1808 DO_VABAV(VABAV_U, vabavu) 1809 1810 static bool trans_VMOV_to_2gp(DisasContext *s, arg_VMOV_to_2gp *a) 1811 { 1812 /* 1813 * VMOV two 32-bit vector lanes to two general-purpose registers. 1814 * This insn is not predicated but it is subject to beat-wise 1815 * execution if it is not in an IT block. For us this means 1816 * only that if PSR.ECI says we should not be executing the beat 1817 * corresponding to the lane of the vector register being accessed 1818 * then we should skip perfoming the move, and that we need to do 1819 * the usual check for bad ECI state and advance of ECI state. 1820 * (If PSR.ECI is non-zero then we cannot be in an IT block.) 1821 */ 1822 TCGv_i32 tmp; 1823 int vd; 1824 1825 if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) || 1826 a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15 || 1827 a->rt == a->rt2) { 1828 /* Rt/Rt2 cases are UNPREDICTABLE */ 1829 return false; 1830 } 1831 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1832 return true; 1833 } 1834 1835 /* Convert Qreg index to Dreg for read_neon_element32() etc */ 1836 vd = a->qd * 2; 1837 1838 if (!mve_skip_vmov(s, vd, a->idx, MO_32)) { 1839 tmp = tcg_temp_new_i32(); 1840 read_neon_element32(tmp, vd, a->idx, MO_32); 1841 store_reg(s, a->rt, tmp); 1842 } 1843 if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) { 1844 tmp = tcg_temp_new_i32(); 1845 read_neon_element32(tmp, vd + 1, a->idx, MO_32); 1846 store_reg(s, a->rt2, tmp); 1847 } 1848 1849 mve_update_and_store_eci(s); 1850 return true; 1851 } 1852 1853 static bool trans_VMOV_from_2gp(DisasContext *s, arg_VMOV_to_2gp *a) 1854 { 1855 /* 1856 * VMOV two general-purpose registers to two 32-bit vector lanes. 1857 * This insn is not predicated but it is subject to beat-wise 1858 * execution if it is not in an IT block. For us this means 1859 * only that if PSR.ECI says we should not be executing the beat 1860 * corresponding to the lane of the vector register being accessed 1861 * then we should skip perfoming the move, and that we need to do 1862 * the usual check for bad ECI state and advance of ECI state. 1863 * (If PSR.ECI is non-zero then we cannot be in an IT block.) 1864 */ 1865 TCGv_i32 tmp; 1866 int vd; 1867 1868 if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) || 1869 a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15) { 1870 /* Rt/Rt2 cases are UNPREDICTABLE */ 1871 return false; 1872 } 1873 if (!mve_eci_check(s) || !vfp_access_check(s)) { 1874 return true; 1875 } 1876 1877 /* Convert Qreg idx to Dreg for read_neon_element32() etc */ 1878 vd = a->qd * 2; 1879 1880 if (!mve_skip_vmov(s, vd, a->idx, MO_32)) { 1881 tmp = load_reg(s, a->rt); 1882 write_neon_element32(tmp, vd, a->idx, MO_32); 1883 tcg_temp_free_i32(tmp); 1884 } 1885 if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) { 1886 tmp = load_reg(s, a->rt2); 1887 write_neon_element32(tmp, vd + 1, a->idx, MO_32); 1888 tcg_temp_free_i32(tmp); 1889 } 1890 1891 mve_update_and_store_eci(s); 1892 return true; 1893 } 1894