1/* 2 * RISC-V translation routines for the RVV Standard Extension. 3 * 4 * Copyright (c) 2020 T-Head Semiconductor Co., Ltd. All rights reserved. 5 * 6 * This program is free software; you can redistribute it and/or modify it 7 * under the terms and conditions of the GNU General Public License, 8 * version 2 or later, as published by the Free Software Foundation. 9 * 10 * This program is distributed in the hope it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 13 * more details. 14 * 15 * You should have received a copy of the GNU General Public License along with 16 * this program. If not, see <http://www.gnu.org/licenses/>. 17 */ 18#include "tcg/tcg-op-gvec.h" 19#include "tcg/tcg-gvec-desc.h" 20#include "internals.h" 21 22static inline bool is_overlapped(const int8_t astart, int8_t asize, 23 const int8_t bstart, int8_t bsize) 24{ 25 const int8_t aend = astart + asize; 26 const int8_t bend = bstart + bsize; 27 28 return MAX(aend, bend) - MIN(astart, bstart) < asize + bsize; 29} 30 31static bool require_rvv(DisasContext *s) 32{ 33 return s->mstatus_vs != 0; 34} 35 36static bool require_rvf(DisasContext *s) 37{ 38 if (s->mstatus_fs == 0) { 39 return false; 40 } 41 42 switch (s->sew) { 43 case MO_16: 44 case MO_32: 45 return has_ext(s, RVF); 46 case MO_64: 47 return has_ext(s, RVD); 48 default: 49 return false; 50 } 51} 52 53static bool require_scale_rvf(DisasContext *s) 54{ 55 if (s->mstatus_fs == 0) { 56 return false; 57 } 58 59 switch (s->sew) { 60 case MO_8: 61 case MO_16: 62 return has_ext(s, RVF); 63 case MO_32: 64 return has_ext(s, RVD); 65 default: 66 return false; 67 } 68} 69 70/* Destination vector register group cannot overlap source mask register. */ 71static bool require_vm(int vm, int vd) 72{ 73 return (vm != 0 || vd != 0); 74} 75 76static bool require_nf(int vd, int nf, int lmul) 77{ 78 int size = nf << MAX(lmul, 0); 79 return size <= 8 && vd + size <= 32; 80} 81 82/* 83 * Vector register should aligned with the passed-in LMUL (EMUL). 84 * If LMUL < 0, i.e. fractional LMUL, any vector register is allowed. 85 */ 86static bool require_align(const int8_t val, const int8_t lmul) 87{ 88 return lmul <= 0 || extract32(val, 0, lmul) == 0; 89} 90 91/* 92 * A destination vector register group can overlap a source vector 93 * register group only if one of the following holds: 94 * 1. The destination EEW equals the source EEW. 95 * 2. The destination EEW is smaller than the source EEW and the overlap 96 * is in the lowest-numbered part of the source register group. 97 * 3. The destination EEW is greater than the source EEW, the source EMUL 98 * is at least 1, and the overlap is in the highest-numbered part of 99 * the destination register group. 100 * (Section 5.2) 101 * 102 * This function returns true if one of the following holds: 103 * * Destination vector register group does not overlap a source vector 104 * register group. 105 * * Rule 3 met. 106 * For rule 1, overlap is allowed so this function doesn't need to be called. 107 * For rule 2, (vd == vs). Caller has to check whether: (vd != vs) before 108 * calling this function. 109 */ 110static bool require_noover(const int8_t dst, const int8_t dst_lmul, 111 const int8_t src, const int8_t src_lmul) 112{ 113 int8_t dst_size = dst_lmul <= 0 ? 1 : 1 << dst_lmul; 114 int8_t src_size = src_lmul <= 0 ? 1 : 1 << src_lmul; 115 116 /* Destination EEW is greater than the source EEW, check rule 3. */ 117 if (dst_size > src_size) { 118 if (dst < src && 119 src_lmul >= 0 && 120 is_overlapped(dst, dst_size, src, src_size) && 121 !is_overlapped(dst, dst_size, src + src_size, src_size)) { 122 return true; 123 } 124 } 125 126 return !is_overlapped(dst, dst_size, src, src_size); 127} 128 129static bool do_vsetvl(DisasContext *s, int rd, int rs1, TCGv s2) 130{ 131 TCGv s1, dst; 132 133 if (!require_rvv(s) || !has_ext(s, RVV)) { 134 return false; 135 } 136 137 dst = dest_gpr(s, rd); 138 139 if (rd == 0 && rs1 == 0) { 140 s1 = tcg_temp_new(); 141 tcg_gen_mov_tl(s1, cpu_vl); 142 } else if (rs1 == 0) { 143 /* As the mask is at least one bit, RV_VLEN_MAX is >= VLMAX */ 144 s1 = tcg_constant_tl(RV_VLEN_MAX); 145 } else { 146 s1 = get_gpr(s, rs1, EXT_ZERO); 147 } 148 149 gen_helper_vsetvl(dst, cpu_env, s1, s2); 150 gen_set_gpr(s, rd, dst); 151 mark_vs_dirty(s); 152 153 tcg_gen_movi_tl(cpu_pc, s->pc_succ_insn); 154 tcg_gen_lookup_and_goto_ptr(); 155 s->base.is_jmp = DISAS_NORETURN; 156 157 if (rd == 0 && rs1 == 0) { 158 tcg_temp_free(s1); 159 } 160 161 return true; 162} 163 164static bool trans_vsetvl(DisasContext *s, arg_vsetvl *a) 165{ 166 TCGv s2 = get_gpr(s, a->rs2, EXT_ZERO); 167 return do_vsetvl(s, a->rd, a->rs1, s2); 168} 169 170static bool trans_vsetvli(DisasContext *s, arg_vsetvli *a) 171{ 172 TCGv s2 = tcg_constant_tl(a->zimm); 173 return do_vsetvl(s, a->rd, a->rs1, s2); 174} 175 176/* vector register offset from env */ 177static uint32_t vreg_ofs(DisasContext *s, int reg) 178{ 179 return offsetof(CPURISCVState, vreg) + reg * s->vlen / 8; 180} 181 182/* check functions */ 183 184/* 185 * Vector unit-stride, strided, unit-stride segment, strided segment 186 * store check function. 187 * 188 * Rules to be checked here: 189 * 1. EMUL must within the range: 1/8 <= EMUL <= 8. (Section 7.3) 190 * 2. Destination vector register number is multiples of EMUL. 191 * (Section 3.4.2, 7.3) 192 * 3. The EMUL setting must be such that EMUL * NFIELDS ≤ 8. (Section 7.8) 193 * 4. Vector register numbers accessed by the segment load or store 194 * cannot increment past 31. (Section 7.8) 195 */ 196static bool vext_check_store(DisasContext *s, int vd, int nf, uint8_t eew) 197{ 198 int8_t emul = eew - s->sew + s->lmul; 199 return (emul >= -3 && emul <= 3) && 200 require_align(vd, emul) && 201 require_nf(vd, nf, emul); 202} 203 204/* 205 * Vector unit-stride, strided, unit-stride segment, strided segment 206 * load check function. 207 * 208 * Rules to be checked here: 209 * 1. All rules applies to store instructions are applies 210 * to load instructions. 211 * 2. Destination vector register group for a masked vector 212 * instruction cannot overlap the source mask register (v0). 213 * (Section 5.3) 214 */ 215static bool vext_check_load(DisasContext *s, int vd, int nf, int vm, 216 uint8_t eew) 217{ 218 return vext_check_store(s, vd, nf, eew) && require_vm(vm, vd); 219} 220 221/* 222 * Vector indexed, indexed segment store check function. 223 * 224 * Rules to be checked here: 225 * 1. EMUL must within the range: 1/8 <= EMUL <= 8. (Section 7.3) 226 * 2. Index vector register number is multiples of EMUL. 227 * (Section 3.4.2, 7.3) 228 * 3. Destination vector register number is multiples of LMUL. 229 * (Section 3.4.2, 7.3) 230 * 4. The EMUL setting must be such that EMUL * NFIELDS ≤ 8. (Section 7.8) 231 * 5. Vector register numbers accessed by the segment load or store 232 * cannot increment past 31. (Section 7.8) 233 */ 234static bool vext_check_st_index(DisasContext *s, int vd, int vs2, int nf, 235 uint8_t eew) 236{ 237 int8_t emul = eew - s->sew + s->lmul; 238 return (emul >= -3 && emul <= 3) && 239 require_align(vs2, emul) && 240 require_align(vd, s->lmul) && 241 require_nf(vd, nf, s->lmul); 242} 243 244/* 245 * Vector indexed, indexed segment load check function. 246 * 247 * Rules to be checked here: 248 * 1. All rules applies to store instructions are applies 249 * to load instructions. 250 * 2. Destination vector register group for a masked vector 251 * instruction cannot overlap the source mask register (v0). 252 * (Section 5.3) 253 * 3. Destination vector register cannot overlap a source vector 254 * register (vs2) group. 255 * (Section 5.2) 256 * 4. Destination vector register groups cannot overlap 257 * the source vector register (vs2) group for 258 * indexed segment load instructions. (Section 7.8.3) 259 */ 260static bool vext_check_ld_index(DisasContext *s, int vd, int vs2, 261 int nf, int vm, uint8_t eew) 262{ 263 int8_t seg_vd; 264 int8_t emul = eew - s->sew + s->lmul; 265 bool ret = vext_check_st_index(s, vd, vs2, nf, eew) && 266 require_vm(vm, vd); 267 268 /* Each segment register group has to follow overlap rules. */ 269 for (int i = 0; i < nf; ++i) { 270 seg_vd = vd + (1 << MAX(s->lmul, 0)) * i; 271 272 if (eew > s->sew) { 273 if (seg_vd != vs2) { 274 ret &= require_noover(seg_vd, s->lmul, vs2, emul); 275 } 276 } else if (eew < s->sew) { 277 ret &= require_noover(seg_vd, s->lmul, vs2, emul); 278 } 279 280 /* 281 * Destination vector register groups cannot overlap 282 * the source vector register (vs2) group for 283 * indexed segment load instructions. 284 */ 285 if (nf > 1) { 286 ret &= !is_overlapped(seg_vd, 1 << MAX(s->lmul, 0), 287 vs2, 1 << MAX(emul, 0)); 288 } 289 } 290 return ret; 291} 292 293static bool vext_check_ss(DisasContext *s, int vd, int vs, int vm) 294{ 295 return require_vm(vm, vd) && 296 require_align(vd, s->lmul) && 297 require_align(vs, s->lmul); 298} 299 300/* 301 * Check function for vector instruction with format: 302 * single-width result and single-width sources (SEW = SEW op SEW) 303 * 304 * Rules to be checked here: 305 * 1. Destination vector register group for a masked vector 306 * instruction cannot overlap the source mask register (v0). 307 * (Section 5.3) 308 * 2. Destination vector register number is multiples of LMUL. 309 * (Section 3.4.2) 310 * 3. Source (vs2, vs1) vector register number are multiples of LMUL. 311 * (Section 3.4.2) 312 */ 313static bool vext_check_sss(DisasContext *s, int vd, int vs1, int vs2, int vm) 314{ 315 return vext_check_ss(s, vd, vs2, vm) && 316 require_align(vs1, s->lmul); 317} 318 319static bool vext_check_ms(DisasContext *s, int vd, int vs) 320{ 321 bool ret = require_align(vs, s->lmul); 322 if (vd != vs) { 323 ret &= require_noover(vd, 0, vs, s->lmul); 324 } 325 return ret; 326} 327 328/* 329 * Check function for maskable vector instruction with format: 330 * single-width result and single-width sources (SEW = SEW op SEW) 331 * 332 * Rules to be checked here: 333 * 1. Source (vs2, vs1) vector register number are multiples of LMUL. 334 * (Section 3.4.2) 335 * 2. Destination vector register cannot overlap a source vector 336 * register (vs2, vs1) group. 337 * (Section 5.2) 338 * 3. The destination vector register group for a masked vector 339 * instruction cannot overlap the source mask register (v0), 340 * unless the destination vector register is being written 341 * with a mask value (e.g., comparisons) or the scalar result 342 * of a reduction. (Section 5.3) 343 */ 344static bool vext_check_mss(DisasContext *s, int vd, int vs1, int vs2) 345{ 346 bool ret = vext_check_ms(s, vd, vs2) && 347 require_align(vs1, s->lmul); 348 if (vd != vs1) { 349 ret &= require_noover(vd, 0, vs1, s->lmul); 350 } 351 return ret; 352} 353 354/* 355 * Common check function for vector widening instructions 356 * of double-width result (2*SEW). 357 * 358 * Rules to be checked here: 359 * 1. The largest vector register group used by an instruction 360 * can not be greater than 8 vector registers (Section 5.2): 361 * => LMUL < 8. 362 * => SEW < 64. 363 * 2. Destination vector register number is multiples of 2 * LMUL. 364 * (Section 3.4.2) 365 * 3. Destination vector register group for a masked vector 366 * instruction cannot overlap the source mask register (v0). 367 * (Section 5.3) 368 */ 369static bool vext_wide_check_common(DisasContext *s, int vd, int vm) 370{ 371 return (s->lmul <= 2) && 372 (s->sew < MO_64) && 373 require_align(vd, s->lmul + 1) && 374 require_vm(vm, vd); 375} 376 377/* 378 * Common check function for vector narrowing instructions 379 * of single-width result (SEW) and double-width source (2*SEW). 380 * 381 * Rules to be checked here: 382 * 1. The largest vector register group used by an instruction 383 * can not be greater than 8 vector registers (Section 5.2): 384 * => LMUL < 8. 385 * => SEW < 64. 386 * 2. Source vector register number is multiples of 2 * LMUL. 387 * (Section 3.4.2) 388 * 3. Destination vector register number is multiples of LMUL. 389 * (Section 3.4.2) 390 * 4. Destination vector register group for a masked vector 391 * instruction cannot overlap the source mask register (v0). 392 * (Section 5.3) 393 */ 394static bool vext_narrow_check_common(DisasContext *s, int vd, int vs2, 395 int vm) 396{ 397 return (s->lmul <= 2) && 398 (s->sew < MO_64) && 399 require_align(vs2, s->lmul + 1) && 400 require_align(vd, s->lmul) && 401 require_vm(vm, vd); 402} 403 404static bool vext_check_ds(DisasContext *s, int vd, int vs, int vm) 405{ 406 return vext_wide_check_common(s, vd, vm) && 407 require_align(vs, s->lmul) && 408 require_noover(vd, s->lmul + 1, vs, s->lmul); 409} 410 411static bool vext_check_dd(DisasContext *s, int vd, int vs, int vm) 412{ 413 return vext_wide_check_common(s, vd, vm) && 414 require_align(vs, s->lmul + 1); 415} 416 417/* 418 * Check function for vector instruction with format: 419 * double-width result and single-width sources (2*SEW = SEW op SEW) 420 * 421 * Rules to be checked here: 422 * 1. All rules in defined in widen common rules are applied. 423 * 2. Source (vs2, vs1) vector register number are multiples of LMUL. 424 * (Section 3.4.2) 425 * 3. Destination vector register cannot overlap a source vector 426 * register (vs2, vs1) group. 427 * (Section 5.2) 428 */ 429static bool vext_check_dss(DisasContext *s, int vd, int vs1, int vs2, int vm) 430{ 431 return vext_check_ds(s, vd, vs2, vm) && 432 require_align(vs1, s->lmul) && 433 require_noover(vd, s->lmul + 1, vs1, s->lmul); 434} 435 436/* 437 * Check function for vector instruction with format: 438 * double-width result and double-width source1 and single-width 439 * source2 (2*SEW = 2*SEW op SEW) 440 * 441 * Rules to be checked here: 442 * 1. All rules in defined in widen common rules are applied. 443 * 2. Source 1 (vs2) vector register number is multiples of 2 * LMUL. 444 * (Section 3.4.2) 445 * 3. Source 2 (vs1) vector register number is multiples of LMUL. 446 * (Section 3.4.2) 447 * 4. Destination vector register cannot overlap a source vector 448 * register (vs1) group. 449 * (Section 5.2) 450 */ 451static bool vext_check_dds(DisasContext *s, int vd, int vs1, int vs2, int vm) 452{ 453 return vext_check_ds(s, vd, vs1, vm) && 454 require_align(vs2, s->lmul + 1); 455} 456 457static bool vext_check_sd(DisasContext *s, int vd, int vs, int vm) 458{ 459 bool ret = vext_narrow_check_common(s, vd, vs, vm); 460 if (vd != vs) { 461 ret &= require_noover(vd, s->lmul, vs, s->lmul + 1); 462 } 463 return ret; 464} 465 466/* 467 * Check function for vector instruction with format: 468 * single-width result and double-width source 1 and single-width 469 * source 2 (SEW = 2*SEW op SEW) 470 * 471 * Rules to be checked here: 472 * 1. All rules in defined in narrow common rules are applied. 473 * 2. Destination vector register cannot overlap a source vector 474 * register (vs2) group. 475 * (Section 5.2) 476 * 3. Source 2 (vs1) vector register number is multiples of LMUL. 477 * (Section 3.4.2) 478 */ 479static bool vext_check_sds(DisasContext *s, int vd, int vs1, int vs2, int vm) 480{ 481 return vext_check_sd(s, vd, vs2, vm) && 482 require_align(vs1, s->lmul); 483} 484 485/* 486 * Check function for vector reduction instructions. 487 * 488 * Rules to be checked here: 489 * 1. Source 1 (vs2) vector register number is multiples of LMUL. 490 * (Section 3.4.2) 491 */ 492static bool vext_check_reduction(DisasContext *s, int vs2) 493{ 494 return require_align(vs2, s->lmul); 495} 496 497/* 498 * Check function for vector slide instructions. 499 * 500 * Rules to be checked here: 501 * 1. Source 1 (vs2) vector register number is multiples of LMUL. 502 * (Section 3.4.2) 503 * 2. Destination vector register number is multiples of LMUL. 504 * (Section 3.4.2) 505 * 3. Destination vector register group for a masked vector 506 * instruction cannot overlap the source mask register (v0). 507 * (Section 5.3) 508 * 4. The destination vector register group for vslideup, vslide1up, 509 * vfslide1up, cannot overlap the source vector register (vs2) group. 510 * (Section 5.2, 16.3.1, 16.3.3) 511 */ 512static bool vext_check_slide(DisasContext *s, int vd, int vs2, 513 int vm, bool is_over) 514{ 515 bool ret = require_align(vs2, s->lmul) && 516 require_align(vd, s->lmul) && 517 require_vm(vm, vd); 518 if (is_over) { 519 ret &= (vd != vs2); 520 } 521 return ret; 522} 523 524/* 525 * In cpu_get_tb_cpu_state(), set VILL if RVV was not present. 526 * So RVV is also be checked in this function. 527 */ 528static bool vext_check_isa_ill(DisasContext *s) 529{ 530 return !s->vill; 531} 532 533/* common translation macro */ 534#define GEN_VEXT_TRANS(NAME, EEW, ARGTYPE, OP, CHECK) \ 535static bool trans_##NAME(DisasContext *s, arg_##ARGTYPE * a) \ 536{ \ 537 if (CHECK(s, a, EEW)) { \ 538 return OP(s, a, EEW); \ 539 } \ 540 return false; \ 541} 542 543/* 544 *** unit stride load and store 545 */ 546typedef void gen_helper_ldst_us(TCGv_ptr, TCGv_ptr, TCGv, 547 TCGv_env, TCGv_i32); 548 549static bool ldst_us_trans(uint32_t vd, uint32_t rs1, uint32_t data, 550 gen_helper_ldst_us *fn, DisasContext *s, 551 bool is_store) 552{ 553 TCGv_ptr dest, mask; 554 TCGv base; 555 TCGv_i32 desc; 556 557 TCGLabel *over = gen_new_label(); 558 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 559 560 dest = tcg_temp_new_ptr(); 561 mask = tcg_temp_new_ptr(); 562 base = get_gpr(s, rs1, EXT_NONE); 563 564 /* 565 * As simd_desc supports at most 256 bytes, and in this implementation, 566 * the max vector group length is 2048 bytes. So split it into two parts. 567 * 568 * The first part is vlen in bytes, encoded in maxsz of simd_desc. 569 * The second part is lmul, encoded in data of simd_desc. 570 */ 571 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 572 573 tcg_gen_addi_ptr(dest, cpu_env, vreg_ofs(s, vd)); 574 tcg_gen_addi_ptr(mask, cpu_env, vreg_ofs(s, 0)); 575 576 fn(dest, mask, base, cpu_env, desc); 577 578 tcg_temp_free_ptr(dest); 579 tcg_temp_free_ptr(mask); 580 581 if (!is_store) { 582 mark_vs_dirty(s); 583 } 584 585 gen_set_label(over); 586 return true; 587} 588 589static bool ld_us_op(DisasContext *s, arg_r2nfvm *a, uint8_t eew) 590{ 591 uint32_t data = 0; 592 gen_helper_ldst_us *fn; 593 static gen_helper_ldst_us * const fns[2][4] = { 594 /* masked unit stride load */ 595 { gen_helper_vle8_v_mask, gen_helper_vle16_v_mask, 596 gen_helper_vle32_v_mask, gen_helper_vle64_v_mask }, 597 /* unmasked unit stride load */ 598 { gen_helper_vle8_v, gen_helper_vle16_v, 599 gen_helper_vle32_v, gen_helper_vle64_v } 600 }; 601 602 fn = fns[a->vm][eew]; 603 if (fn == NULL) { 604 return false; 605 } 606 607 data = FIELD_DP32(data, VDATA, VM, a->vm); 608 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 609 data = FIELD_DP32(data, VDATA, NF, a->nf); 610 return ldst_us_trans(a->rd, a->rs1, data, fn, s, false); 611} 612 613static bool ld_us_check(DisasContext *s, arg_r2nfvm* a, uint8_t eew) 614{ 615 return require_rvv(s) && 616 vext_check_isa_ill(s) && 617 vext_check_load(s, a->rd, a->nf, a->vm, eew); 618} 619 620GEN_VEXT_TRANS(vle8_v, MO_8, r2nfvm, ld_us_op, ld_us_check) 621GEN_VEXT_TRANS(vle16_v, MO_16, r2nfvm, ld_us_op, ld_us_check) 622GEN_VEXT_TRANS(vle32_v, MO_32, r2nfvm, ld_us_op, ld_us_check) 623GEN_VEXT_TRANS(vle64_v, MO_64, r2nfvm, ld_us_op, ld_us_check) 624 625static bool st_us_op(DisasContext *s, arg_r2nfvm *a, uint8_t eew) 626{ 627 uint32_t data = 0; 628 gen_helper_ldst_us *fn; 629 static gen_helper_ldst_us * const fns[2][4] = { 630 /* masked unit stride store */ 631 { gen_helper_vse8_v_mask, gen_helper_vse16_v_mask, 632 gen_helper_vse32_v_mask, gen_helper_vse64_v_mask }, 633 /* unmasked unit stride store */ 634 { gen_helper_vse8_v, gen_helper_vse16_v, 635 gen_helper_vse32_v, gen_helper_vse64_v } 636 }; 637 638 fn = fns[a->vm][eew]; 639 if (fn == NULL) { 640 return false; 641 } 642 643 data = FIELD_DP32(data, VDATA, VM, a->vm); 644 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 645 data = FIELD_DP32(data, VDATA, NF, a->nf); 646 return ldst_us_trans(a->rd, a->rs1, data, fn, s, true); 647} 648 649static bool st_us_check(DisasContext *s, arg_r2nfvm* a, uint8_t eew) 650{ 651 return require_rvv(s) && 652 vext_check_isa_ill(s) && 653 vext_check_store(s, a->rd, a->nf, eew); 654} 655 656GEN_VEXT_TRANS(vse8_v, MO_8, r2nfvm, st_us_op, st_us_check) 657GEN_VEXT_TRANS(vse16_v, MO_16, r2nfvm, st_us_op, st_us_check) 658GEN_VEXT_TRANS(vse32_v, MO_32, r2nfvm, st_us_op, st_us_check) 659GEN_VEXT_TRANS(vse64_v, MO_64, r2nfvm, st_us_op, st_us_check) 660 661/* 662 *** stride load and store 663 */ 664typedef void gen_helper_ldst_stride(TCGv_ptr, TCGv_ptr, TCGv, 665 TCGv, TCGv_env, TCGv_i32); 666 667static bool ldst_stride_trans(uint32_t vd, uint32_t rs1, uint32_t rs2, 668 uint32_t data, gen_helper_ldst_stride *fn, 669 DisasContext *s, bool is_store) 670{ 671 TCGv_ptr dest, mask; 672 TCGv base, stride; 673 TCGv_i32 desc; 674 675 TCGLabel *over = gen_new_label(); 676 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 677 678 dest = tcg_temp_new_ptr(); 679 mask = tcg_temp_new_ptr(); 680 base = get_gpr(s, rs1, EXT_NONE); 681 stride = get_gpr(s, rs2, EXT_NONE); 682 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 683 684 tcg_gen_addi_ptr(dest, cpu_env, vreg_ofs(s, vd)); 685 tcg_gen_addi_ptr(mask, cpu_env, vreg_ofs(s, 0)); 686 687 fn(dest, mask, base, stride, cpu_env, desc); 688 689 tcg_temp_free_ptr(dest); 690 tcg_temp_free_ptr(mask); 691 692 if (!is_store) { 693 mark_vs_dirty(s); 694 } 695 696 gen_set_label(over); 697 return true; 698} 699 700static bool ld_stride_op(DisasContext *s, arg_rnfvm *a, uint8_t eew) 701{ 702 uint32_t data = 0; 703 gen_helper_ldst_stride *fn; 704 static gen_helper_ldst_stride * const fns[4] = { 705 gen_helper_vlse8_v, gen_helper_vlse16_v, 706 gen_helper_vlse32_v, gen_helper_vlse64_v 707 }; 708 709 fn = fns[eew]; 710 if (fn == NULL) { 711 return false; 712 } 713 714 data = FIELD_DP32(data, VDATA, VM, a->vm); 715 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 716 data = FIELD_DP32(data, VDATA, NF, a->nf); 717 return ldst_stride_trans(a->rd, a->rs1, a->rs2, data, fn, s, false); 718} 719 720static bool ld_stride_check(DisasContext *s, arg_rnfvm* a, uint8_t eew) 721{ 722 return require_rvv(s) && 723 vext_check_isa_ill(s) && 724 vext_check_load(s, a->rd, a->nf, a->vm, eew); 725} 726 727GEN_VEXT_TRANS(vlse8_v, MO_8, rnfvm, ld_stride_op, ld_stride_check) 728GEN_VEXT_TRANS(vlse16_v, MO_16, rnfvm, ld_stride_op, ld_stride_check) 729GEN_VEXT_TRANS(vlse32_v, MO_32, rnfvm, ld_stride_op, ld_stride_check) 730GEN_VEXT_TRANS(vlse64_v, MO_64, rnfvm, ld_stride_op, ld_stride_check) 731 732static bool st_stride_op(DisasContext *s, arg_rnfvm *a, uint8_t eew) 733{ 734 uint32_t data = 0; 735 gen_helper_ldst_stride *fn; 736 static gen_helper_ldst_stride * const fns[4] = { 737 /* masked stride store */ 738 gen_helper_vsse8_v, gen_helper_vsse16_v, 739 gen_helper_vsse32_v, gen_helper_vsse64_v 740 }; 741 742 data = FIELD_DP32(data, VDATA, VM, a->vm); 743 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 744 data = FIELD_DP32(data, VDATA, NF, a->nf); 745 fn = fns[eew]; 746 if (fn == NULL) { 747 return false; 748 } 749 750 return ldst_stride_trans(a->rd, a->rs1, a->rs2, data, fn, s, true); 751} 752 753static bool st_stride_check(DisasContext *s, arg_rnfvm* a, uint8_t eew) 754{ 755 return require_rvv(s) && 756 vext_check_isa_ill(s) && 757 vext_check_store(s, a->rd, a->nf, eew); 758} 759 760GEN_VEXT_TRANS(vsse8_v, MO_8, rnfvm, st_stride_op, st_stride_check) 761GEN_VEXT_TRANS(vsse16_v, MO_16, rnfvm, st_stride_op, st_stride_check) 762GEN_VEXT_TRANS(vsse32_v, MO_32, rnfvm, st_stride_op, st_stride_check) 763GEN_VEXT_TRANS(vsse64_v, MO_64, rnfvm, st_stride_op, st_stride_check) 764 765/* 766 *** index load and store 767 */ 768typedef void gen_helper_ldst_index(TCGv_ptr, TCGv_ptr, TCGv, 769 TCGv_ptr, TCGv_env, TCGv_i32); 770 771static bool ldst_index_trans(uint32_t vd, uint32_t rs1, uint32_t vs2, 772 uint32_t data, gen_helper_ldst_index *fn, 773 DisasContext *s, bool is_store) 774{ 775 TCGv_ptr dest, mask, index; 776 TCGv base; 777 TCGv_i32 desc; 778 779 TCGLabel *over = gen_new_label(); 780 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 781 782 dest = tcg_temp_new_ptr(); 783 mask = tcg_temp_new_ptr(); 784 index = tcg_temp_new_ptr(); 785 base = get_gpr(s, rs1, EXT_NONE); 786 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 787 788 tcg_gen_addi_ptr(dest, cpu_env, vreg_ofs(s, vd)); 789 tcg_gen_addi_ptr(index, cpu_env, vreg_ofs(s, vs2)); 790 tcg_gen_addi_ptr(mask, cpu_env, vreg_ofs(s, 0)); 791 792 fn(dest, mask, base, index, cpu_env, desc); 793 794 tcg_temp_free_ptr(dest); 795 tcg_temp_free_ptr(mask); 796 tcg_temp_free_ptr(index); 797 798 if (!is_store) { 799 mark_vs_dirty(s); 800 } 801 802 gen_set_label(over); 803 return true; 804} 805 806static bool ld_index_op(DisasContext *s, arg_rnfvm *a, uint8_t seq) 807{ 808 uint32_t data = 0; 809 gen_helper_ldst_index *fn; 810 static gen_helper_ldst_index * const fns[7][4] = { 811 { gen_helper_vlxb_v_b, gen_helper_vlxb_v_h, 812 gen_helper_vlxb_v_w, gen_helper_vlxb_v_d }, 813 { NULL, gen_helper_vlxh_v_h, 814 gen_helper_vlxh_v_w, gen_helper_vlxh_v_d }, 815 { NULL, NULL, 816 gen_helper_vlxw_v_w, gen_helper_vlxw_v_d }, 817 { gen_helper_vlxe_v_b, gen_helper_vlxe_v_h, 818 gen_helper_vlxe_v_w, gen_helper_vlxe_v_d }, 819 { gen_helper_vlxbu_v_b, gen_helper_vlxbu_v_h, 820 gen_helper_vlxbu_v_w, gen_helper_vlxbu_v_d }, 821 { NULL, gen_helper_vlxhu_v_h, 822 gen_helper_vlxhu_v_w, gen_helper_vlxhu_v_d }, 823 { NULL, NULL, 824 gen_helper_vlxwu_v_w, gen_helper_vlxwu_v_d }, 825 }; 826 827 fn = fns[seq][s->sew]; 828 if (fn == NULL) { 829 return false; 830 } 831 832 data = FIELD_DP32(data, VDATA, VM, a->vm); 833 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 834 data = FIELD_DP32(data, VDATA, NF, a->nf); 835 return ldst_index_trans(a->rd, a->rs1, a->rs2, data, fn, s, false); 836} 837 838/* 839 * For vector indexed segment loads, the destination vector register 840 * groups cannot overlap the source vector register group (specified by 841 * `vs2`), else an illegal instruction exception is raised. 842 */ 843static bool ld_index_check(DisasContext *s, arg_rnfvm* a, uint8_t eew) 844{ 845 return require_rvv(s) && 846 vext_check_isa_ill(s) && 847 vext_check_ld_index(s, a->rd, a->rs2, a->nf, a->vm, eew); 848} 849 850GEN_VEXT_TRANS(vlxb_v, 0, rnfvm, ld_index_op, ld_index_check) 851GEN_VEXT_TRANS(vlxh_v, 1, rnfvm, ld_index_op, ld_index_check) 852GEN_VEXT_TRANS(vlxw_v, 2, rnfvm, ld_index_op, ld_index_check) 853GEN_VEXT_TRANS(vlxe_v, 3, rnfvm, ld_index_op, ld_index_check) 854GEN_VEXT_TRANS(vlxbu_v, 4, rnfvm, ld_index_op, ld_index_check) 855GEN_VEXT_TRANS(vlxhu_v, 5, rnfvm, ld_index_op, ld_index_check) 856GEN_VEXT_TRANS(vlxwu_v, 6, rnfvm, ld_index_op, ld_index_check) 857 858static bool st_index_op(DisasContext *s, arg_rnfvm *a, uint8_t seq) 859{ 860 uint32_t data = 0; 861 gen_helper_ldst_index *fn; 862 static gen_helper_ldst_index * const fns[4][4] = { 863 { gen_helper_vsxb_v_b, gen_helper_vsxb_v_h, 864 gen_helper_vsxb_v_w, gen_helper_vsxb_v_d }, 865 { NULL, gen_helper_vsxh_v_h, 866 gen_helper_vsxh_v_w, gen_helper_vsxh_v_d }, 867 { NULL, NULL, 868 gen_helper_vsxw_v_w, gen_helper_vsxw_v_d }, 869 { gen_helper_vsxe_v_b, gen_helper_vsxe_v_h, 870 gen_helper_vsxe_v_w, gen_helper_vsxe_v_d } 871 }; 872 873 fn = fns[seq][s->sew]; 874 if (fn == NULL) { 875 return false; 876 } 877 878 data = FIELD_DP32(data, VDATA, VM, a->vm); 879 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 880 data = FIELD_DP32(data, VDATA, NF, a->nf); 881 return ldst_index_trans(a->rd, a->rs1, a->rs2, data, fn, s, true); 882} 883 884static bool st_index_check(DisasContext *s, arg_rnfvm* a, uint8_t eew) 885{ 886 return require_rvv(s) && 887 vext_check_isa_ill(s) && 888 vext_check_st_index(s, a->rd, a->rs2, a->nf, eew); 889} 890 891GEN_VEXT_TRANS(vsxb_v, 0, rnfvm, st_index_op, st_index_check) 892GEN_VEXT_TRANS(vsxh_v, 1, rnfvm, st_index_op, st_index_check) 893GEN_VEXT_TRANS(vsxw_v, 2, rnfvm, st_index_op, st_index_check) 894GEN_VEXT_TRANS(vsxe_v, 3, rnfvm, st_index_op, st_index_check) 895 896/* 897 *** unit stride fault-only-first load 898 */ 899static bool ldff_trans(uint32_t vd, uint32_t rs1, uint32_t data, 900 gen_helper_ldst_us *fn, DisasContext *s) 901{ 902 TCGv_ptr dest, mask; 903 TCGv base; 904 TCGv_i32 desc; 905 906 TCGLabel *over = gen_new_label(); 907 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 908 909 dest = tcg_temp_new_ptr(); 910 mask = tcg_temp_new_ptr(); 911 base = get_gpr(s, rs1, EXT_NONE); 912 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 913 914 tcg_gen_addi_ptr(dest, cpu_env, vreg_ofs(s, vd)); 915 tcg_gen_addi_ptr(mask, cpu_env, vreg_ofs(s, 0)); 916 917 fn(dest, mask, base, cpu_env, desc); 918 919 tcg_temp_free_ptr(dest); 920 tcg_temp_free_ptr(mask); 921 mark_vs_dirty(s); 922 gen_set_label(over); 923 return true; 924} 925 926static bool ldff_op(DisasContext *s, arg_r2nfvm *a, uint8_t seq) 927{ 928 uint32_t data = 0; 929 gen_helper_ldst_us *fn; 930 static gen_helper_ldst_us * const fns[7][4] = { 931 { gen_helper_vlbff_v_b, gen_helper_vlbff_v_h, 932 gen_helper_vlbff_v_w, gen_helper_vlbff_v_d }, 933 { NULL, gen_helper_vlhff_v_h, 934 gen_helper_vlhff_v_w, gen_helper_vlhff_v_d }, 935 { NULL, NULL, 936 gen_helper_vlwff_v_w, gen_helper_vlwff_v_d }, 937 { gen_helper_vleff_v_b, gen_helper_vleff_v_h, 938 gen_helper_vleff_v_w, gen_helper_vleff_v_d }, 939 { gen_helper_vlbuff_v_b, gen_helper_vlbuff_v_h, 940 gen_helper_vlbuff_v_w, gen_helper_vlbuff_v_d }, 941 { NULL, gen_helper_vlhuff_v_h, 942 gen_helper_vlhuff_v_w, gen_helper_vlhuff_v_d }, 943 { NULL, NULL, 944 gen_helper_vlwuff_v_w, gen_helper_vlwuff_v_d } 945 }; 946 947 fn = fns[seq][s->sew]; 948 if (fn == NULL) { 949 return false; 950 } 951 952 data = FIELD_DP32(data, VDATA, VM, a->vm); 953 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 954 data = FIELD_DP32(data, VDATA, NF, a->nf); 955 return ldff_trans(a->rd, a->rs1, data, fn, s); 956} 957 958GEN_VEXT_TRANS(vlbff_v, 0, r2nfvm, ldff_op, ld_us_check) 959GEN_VEXT_TRANS(vlhff_v, 1, r2nfvm, ldff_op, ld_us_check) 960GEN_VEXT_TRANS(vlwff_v, 2, r2nfvm, ldff_op, ld_us_check) 961GEN_VEXT_TRANS(vleff_v, 3, r2nfvm, ldff_op, ld_us_check) 962GEN_VEXT_TRANS(vlbuff_v, 4, r2nfvm, ldff_op, ld_us_check) 963GEN_VEXT_TRANS(vlhuff_v, 5, r2nfvm, ldff_op, ld_us_check) 964GEN_VEXT_TRANS(vlwuff_v, 6, r2nfvm, ldff_op, ld_us_check) 965 966/* 967 *** Vector Integer Arithmetic Instructions 968 */ 969#define MAXSZ(s) (s->vlen >> (3 - s->lmul)) 970 971static bool opivv_check(DisasContext *s, arg_rmrr *a) 972{ 973 return require_rvv(s) && 974 vext_check_isa_ill(s) && 975 vext_check_sss(s, a->rd, a->rs1, a->rs2, a->vm); 976} 977 978typedef void GVecGen3Fn(unsigned, uint32_t, uint32_t, 979 uint32_t, uint32_t, uint32_t); 980 981static inline bool 982do_opivv_gvec(DisasContext *s, arg_rmrr *a, GVecGen3Fn *gvec_fn, 983 gen_helper_gvec_4_ptr *fn) 984{ 985 TCGLabel *over = gen_new_label(); 986 if (!opivv_check(s, a)) { 987 return false; 988 } 989 990 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 991 992 if (a->vm && s->vl_eq_vlmax) { 993 gvec_fn(s->sew, vreg_ofs(s, a->rd), 994 vreg_ofs(s, a->rs2), vreg_ofs(s, a->rs1), 995 MAXSZ(s), MAXSZ(s)); 996 } else { 997 uint32_t data = 0; 998 999 data = FIELD_DP32(data, VDATA, VM, a->vm); 1000 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 1001 tcg_gen_gvec_4_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), 1002 vreg_ofs(s, a->rs1), vreg_ofs(s, a->rs2), 1003 cpu_env, s->vlen / 8, s->vlen / 8, data, fn); 1004 } 1005 mark_vs_dirty(s); 1006 gen_set_label(over); 1007 return true; 1008} 1009 1010/* OPIVV with GVEC IR */ 1011#define GEN_OPIVV_GVEC_TRANS(NAME, SUF) \ 1012static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1013{ \ 1014 static gen_helper_gvec_4_ptr * const fns[4] = { \ 1015 gen_helper_##NAME##_b, gen_helper_##NAME##_h, \ 1016 gen_helper_##NAME##_w, gen_helper_##NAME##_d, \ 1017 }; \ 1018 return do_opivv_gvec(s, a, tcg_gen_gvec_##SUF, fns[s->sew]); \ 1019} 1020 1021GEN_OPIVV_GVEC_TRANS(vadd_vv, add) 1022GEN_OPIVV_GVEC_TRANS(vsub_vv, sub) 1023 1024typedef void gen_helper_opivx(TCGv_ptr, TCGv_ptr, TCGv, TCGv_ptr, 1025 TCGv_env, TCGv_i32); 1026 1027static bool opivx_trans(uint32_t vd, uint32_t rs1, uint32_t vs2, uint32_t vm, 1028 gen_helper_opivx *fn, DisasContext *s) 1029{ 1030 TCGv_ptr dest, src2, mask; 1031 TCGv src1; 1032 TCGv_i32 desc; 1033 uint32_t data = 0; 1034 1035 TCGLabel *over = gen_new_label(); 1036 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 1037 1038 dest = tcg_temp_new_ptr(); 1039 mask = tcg_temp_new_ptr(); 1040 src2 = tcg_temp_new_ptr(); 1041 src1 = get_gpr(s, rs1, EXT_NONE); 1042 1043 data = FIELD_DP32(data, VDATA, VM, vm); 1044 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 1045 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 1046 1047 tcg_gen_addi_ptr(dest, cpu_env, vreg_ofs(s, vd)); 1048 tcg_gen_addi_ptr(src2, cpu_env, vreg_ofs(s, vs2)); 1049 tcg_gen_addi_ptr(mask, cpu_env, vreg_ofs(s, 0)); 1050 1051 fn(dest, mask, src1, src2, cpu_env, desc); 1052 1053 tcg_temp_free_ptr(dest); 1054 tcg_temp_free_ptr(mask); 1055 tcg_temp_free_ptr(src2); 1056 mark_vs_dirty(s); 1057 gen_set_label(over); 1058 return true; 1059} 1060 1061static bool opivx_check(DisasContext *s, arg_rmrr *a) 1062{ 1063 return require_rvv(s) && 1064 vext_check_isa_ill(s) && 1065 vext_check_ss(s, a->rd, a->rs2, a->vm); 1066} 1067 1068typedef void GVecGen2sFn(unsigned, uint32_t, uint32_t, TCGv_i64, 1069 uint32_t, uint32_t); 1070 1071static inline bool 1072do_opivx_gvec(DisasContext *s, arg_rmrr *a, GVecGen2sFn *gvec_fn, 1073 gen_helper_opivx *fn) 1074{ 1075 if (!opivx_check(s, a)) { 1076 return false; 1077 } 1078 1079 if (a->vm && s->vl_eq_vlmax) { 1080 TCGv_i64 src1 = tcg_temp_new_i64(); 1081 1082 tcg_gen_ext_tl_i64(src1, get_gpr(s, a->rs1, EXT_SIGN)); 1083 gvec_fn(s->sew, vreg_ofs(s, a->rd), vreg_ofs(s, a->rs2), 1084 src1, MAXSZ(s), MAXSZ(s)); 1085 1086 tcg_temp_free_i64(src1); 1087 mark_vs_dirty(s); 1088 return true; 1089 } 1090 return opivx_trans(a->rd, a->rs1, a->rs2, a->vm, fn, s); 1091} 1092 1093/* OPIVX with GVEC IR */ 1094#define GEN_OPIVX_GVEC_TRANS(NAME, SUF) \ 1095static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1096{ \ 1097 static gen_helper_opivx * const fns[4] = { \ 1098 gen_helper_##NAME##_b, gen_helper_##NAME##_h, \ 1099 gen_helper_##NAME##_w, gen_helper_##NAME##_d, \ 1100 }; \ 1101 return do_opivx_gvec(s, a, tcg_gen_gvec_##SUF, fns[s->sew]); \ 1102} 1103 1104GEN_OPIVX_GVEC_TRANS(vadd_vx, adds) 1105GEN_OPIVX_GVEC_TRANS(vsub_vx, subs) 1106 1107static void gen_vec_rsub8_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b) 1108{ 1109 tcg_gen_vec_sub8_i64(d, b, a); 1110} 1111 1112static void gen_vec_rsub16_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b) 1113{ 1114 tcg_gen_vec_sub16_i64(d, b, a); 1115} 1116 1117static void gen_rsub_i32(TCGv_i32 ret, TCGv_i32 arg1, TCGv_i32 arg2) 1118{ 1119 tcg_gen_sub_i32(ret, arg2, arg1); 1120} 1121 1122static void gen_rsub_i64(TCGv_i64 ret, TCGv_i64 arg1, TCGv_i64 arg2) 1123{ 1124 tcg_gen_sub_i64(ret, arg2, arg1); 1125} 1126 1127static void gen_rsub_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b) 1128{ 1129 tcg_gen_sub_vec(vece, r, b, a); 1130} 1131 1132static void tcg_gen_gvec_rsubs(unsigned vece, uint32_t dofs, uint32_t aofs, 1133 TCGv_i64 c, uint32_t oprsz, uint32_t maxsz) 1134{ 1135 static const TCGOpcode vecop_list[] = { INDEX_op_sub_vec, 0 }; 1136 static const GVecGen2s rsub_op[4] = { 1137 { .fni8 = gen_vec_rsub8_i64, 1138 .fniv = gen_rsub_vec, 1139 .fno = gen_helper_vec_rsubs8, 1140 .opt_opc = vecop_list, 1141 .vece = MO_8 }, 1142 { .fni8 = gen_vec_rsub16_i64, 1143 .fniv = gen_rsub_vec, 1144 .fno = gen_helper_vec_rsubs16, 1145 .opt_opc = vecop_list, 1146 .vece = MO_16 }, 1147 { .fni4 = gen_rsub_i32, 1148 .fniv = gen_rsub_vec, 1149 .fno = gen_helper_vec_rsubs32, 1150 .opt_opc = vecop_list, 1151 .vece = MO_32 }, 1152 { .fni8 = gen_rsub_i64, 1153 .fniv = gen_rsub_vec, 1154 .fno = gen_helper_vec_rsubs64, 1155 .opt_opc = vecop_list, 1156 .prefer_i64 = TCG_TARGET_REG_BITS == 64, 1157 .vece = MO_64 }, 1158 }; 1159 1160 tcg_debug_assert(vece <= MO_64); 1161 tcg_gen_gvec_2s(dofs, aofs, oprsz, maxsz, c, &rsub_op[vece]); 1162} 1163 1164GEN_OPIVX_GVEC_TRANS(vrsub_vx, rsubs) 1165 1166typedef enum { 1167 IMM_ZX, /* Zero-extended */ 1168 IMM_SX, /* Sign-extended */ 1169 IMM_TRUNC_SEW, /* Truncate to log(SEW) bits */ 1170 IMM_TRUNC_2SEW, /* Truncate to log(2*SEW) bits */ 1171} imm_mode_t; 1172 1173static int64_t extract_imm(DisasContext *s, uint32_t imm, imm_mode_t imm_mode) 1174{ 1175 switch (imm_mode) { 1176 case IMM_ZX: 1177 return extract64(imm, 0, 5); 1178 case IMM_SX: 1179 return sextract64(imm, 0, 5); 1180 case IMM_TRUNC_SEW: 1181 return extract64(imm, 0, s->sew + 3); 1182 case IMM_TRUNC_2SEW: 1183 return extract64(imm, 0, s->sew + 4); 1184 default: 1185 g_assert_not_reached(); 1186 } 1187} 1188 1189static bool opivi_trans(uint32_t vd, uint32_t imm, uint32_t vs2, uint32_t vm, 1190 gen_helper_opivx *fn, DisasContext *s, 1191 imm_mode_t imm_mode) 1192{ 1193 TCGv_ptr dest, src2, mask; 1194 TCGv src1; 1195 TCGv_i32 desc; 1196 uint32_t data = 0; 1197 1198 TCGLabel *over = gen_new_label(); 1199 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 1200 1201 dest = tcg_temp_new_ptr(); 1202 mask = tcg_temp_new_ptr(); 1203 src2 = tcg_temp_new_ptr(); 1204 src1 = tcg_constant_tl(extract_imm(s, imm, imm_mode)); 1205 1206 data = FIELD_DP32(data, VDATA, VM, vm); 1207 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 1208 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 1209 1210 tcg_gen_addi_ptr(dest, cpu_env, vreg_ofs(s, vd)); 1211 tcg_gen_addi_ptr(src2, cpu_env, vreg_ofs(s, vs2)); 1212 tcg_gen_addi_ptr(mask, cpu_env, vreg_ofs(s, 0)); 1213 1214 fn(dest, mask, src1, src2, cpu_env, desc); 1215 1216 tcg_temp_free_ptr(dest); 1217 tcg_temp_free_ptr(mask); 1218 tcg_temp_free_ptr(src2); 1219 mark_vs_dirty(s); 1220 gen_set_label(over); 1221 return true; 1222} 1223 1224typedef void GVecGen2iFn(unsigned, uint32_t, uint32_t, int64_t, 1225 uint32_t, uint32_t); 1226 1227static inline bool 1228do_opivi_gvec(DisasContext *s, arg_rmrr *a, GVecGen2iFn *gvec_fn, 1229 gen_helper_opivx *fn, imm_mode_t imm_mode) 1230{ 1231 if (!opivx_check(s, a)) { 1232 return false; 1233 } 1234 1235 if (a->vm && s->vl_eq_vlmax) { 1236 gvec_fn(s->sew, vreg_ofs(s, a->rd), vreg_ofs(s, a->rs2), 1237 extract_imm(s, a->rs1, imm_mode), MAXSZ(s), MAXSZ(s)); 1238 mark_vs_dirty(s); 1239 return true; 1240 } 1241 return opivi_trans(a->rd, a->rs1, a->rs2, a->vm, fn, s, imm_mode); 1242} 1243 1244/* OPIVI with GVEC IR */ 1245#define GEN_OPIVI_GVEC_TRANS(NAME, IMM_MODE, OPIVX, SUF) \ 1246static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1247{ \ 1248 static gen_helper_opivx * const fns[4] = { \ 1249 gen_helper_##OPIVX##_b, gen_helper_##OPIVX##_h, \ 1250 gen_helper_##OPIVX##_w, gen_helper_##OPIVX##_d, \ 1251 }; \ 1252 return do_opivi_gvec(s, a, tcg_gen_gvec_##SUF, \ 1253 fns[s->sew], IMM_MODE); \ 1254} 1255 1256GEN_OPIVI_GVEC_TRANS(vadd_vi, IMM_SX, vadd_vx, addi) 1257 1258static void tcg_gen_gvec_rsubi(unsigned vece, uint32_t dofs, uint32_t aofs, 1259 int64_t c, uint32_t oprsz, uint32_t maxsz) 1260{ 1261 TCGv_i64 tmp = tcg_constant_i64(c); 1262 tcg_gen_gvec_rsubs(vece, dofs, aofs, tmp, oprsz, maxsz); 1263} 1264 1265GEN_OPIVI_GVEC_TRANS(vrsub_vi, IMM_SX, vrsub_vx, rsubi) 1266 1267/* Vector Widening Integer Add/Subtract */ 1268 1269/* OPIVV with WIDEN */ 1270static bool opivv_widen_check(DisasContext *s, arg_rmrr *a) 1271{ 1272 return require_rvv(s) && 1273 vext_check_isa_ill(s) && 1274 vext_check_dss(s, a->rd, a->rs1, a->rs2, a->vm); 1275} 1276 1277static bool do_opivv_widen(DisasContext *s, arg_rmrr *a, 1278 gen_helper_gvec_4_ptr *fn, 1279 bool (*checkfn)(DisasContext *, arg_rmrr *)) 1280{ 1281 if (checkfn(s, a)) { 1282 uint32_t data = 0; 1283 TCGLabel *over = gen_new_label(); 1284 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 1285 1286 data = FIELD_DP32(data, VDATA, VM, a->vm); 1287 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 1288 tcg_gen_gvec_4_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), 1289 vreg_ofs(s, a->rs1), 1290 vreg_ofs(s, a->rs2), 1291 cpu_env, s->vlen / 8, s->vlen / 8, 1292 data, fn); 1293 mark_vs_dirty(s); 1294 gen_set_label(over); 1295 return true; 1296 } 1297 return false; 1298} 1299 1300#define GEN_OPIVV_WIDEN_TRANS(NAME, CHECK) \ 1301static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1302{ \ 1303 static gen_helper_gvec_4_ptr * const fns[3] = { \ 1304 gen_helper_##NAME##_b, \ 1305 gen_helper_##NAME##_h, \ 1306 gen_helper_##NAME##_w \ 1307 }; \ 1308 return do_opivv_widen(s, a, fns[s->sew], CHECK); \ 1309} 1310 1311GEN_OPIVV_WIDEN_TRANS(vwaddu_vv, opivv_widen_check) 1312GEN_OPIVV_WIDEN_TRANS(vwadd_vv, opivv_widen_check) 1313GEN_OPIVV_WIDEN_TRANS(vwsubu_vv, opivv_widen_check) 1314GEN_OPIVV_WIDEN_TRANS(vwsub_vv, opivv_widen_check) 1315 1316/* OPIVX with WIDEN */ 1317static bool opivx_widen_check(DisasContext *s, arg_rmrr *a) 1318{ 1319 return require_rvv(s) && 1320 vext_check_isa_ill(s) && 1321 vext_check_ds(s, a->rd, a->rs2, a->vm); 1322} 1323 1324static bool do_opivx_widen(DisasContext *s, arg_rmrr *a, 1325 gen_helper_opivx *fn) 1326{ 1327 if (opivx_widen_check(s, a)) { 1328 return opivx_trans(a->rd, a->rs1, a->rs2, a->vm, fn, s); 1329 } 1330 return false; 1331} 1332 1333#define GEN_OPIVX_WIDEN_TRANS(NAME) \ 1334static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1335{ \ 1336 static gen_helper_opivx * const fns[3] = { \ 1337 gen_helper_##NAME##_b, \ 1338 gen_helper_##NAME##_h, \ 1339 gen_helper_##NAME##_w \ 1340 }; \ 1341 return do_opivx_widen(s, a, fns[s->sew]); \ 1342} 1343 1344GEN_OPIVX_WIDEN_TRANS(vwaddu_vx) 1345GEN_OPIVX_WIDEN_TRANS(vwadd_vx) 1346GEN_OPIVX_WIDEN_TRANS(vwsubu_vx) 1347GEN_OPIVX_WIDEN_TRANS(vwsub_vx) 1348 1349/* WIDEN OPIVV with WIDEN */ 1350static bool opiwv_widen_check(DisasContext *s, arg_rmrr *a) 1351{ 1352 return require_rvv(s) && 1353 vext_check_isa_ill(s) && 1354 vext_check_dds(s, a->rd, a->rs1, a->rs2, a->vm); 1355} 1356 1357static bool do_opiwv_widen(DisasContext *s, arg_rmrr *a, 1358 gen_helper_gvec_4_ptr *fn) 1359{ 1360 if (opiwv_widen_check(s, a)) { 1361 uint32_t data = 0; 1362 TCGLabel *over = gen_new_label(); 1363 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 1364 1365 data = FIELD_DP32(data, VDATA, VM, a->vm); 1366 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 1367 tcg_gen_gvec_4_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), 1368 vreg_ofs(s, a->rs1), 1369 vreg_ofs(s, a->rs2), 1370 cpu_env, s->vlen / 8, s->vlen / 8, data, fn); 1371 mark_vs_dirty(s); 1372 gen_set_label(over); 1373 return true; 1374 } 1375 return false; 1376} 1377 1378#define GEN_OPIWV_WIDEN_TRANS(NAME) \ 1379static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1380{ \ 1381 static gen_helper_gvec_4_ptr * const fns[3] = { \ 1382 gen_helper_##NAME##_b, \ 1383 gen_helper_##NAME##_h, \ 1384 gen_helper_##NAME##_w \ 1385 }; \ 1386 return do_opiwv_widen(s, a, fns[s->sew]); \ 1387} 1388 1389GEN_OPIWV_WIDEN_TRANS(vwaddu_wv) 1390GEN_OPIWV_WIDEN_TRANS(vwadd_wv) 1391GEN_OPIWV_WIDEN_TRANS(vwsubu_wv) 1392GEN_OPIWV_WIDEN_TRANS(vwsub_wv) 1393 1394/* WIDEN OPIVX with WIDEN */ 1395static bool opiwx_widen_check(DisasContext *s, arg_rmrr *a) 1396{ 1397 return require_rvv(s) && 1398 vext_check_isa_ill(s) && 1399 vext_check_dd(s, a->rd, a->rs2, a->vm); 1400} 1401 1402static bool do_opiwx_widen(DisasContext *s, arg_rmrr *a, 1403 gen_helper_opivx *fn) 1404{ 1405 if (opiwx_widen_check(s, a)) { 1406 return opivx_trans(a->rd, a->rs1, a->rs2, a->vm, fn, s); 1407 } 1408 return false; 1409} 1410 1411#define GEN_OPIWX_WIDEN_TRANS(NAME) \ 1412static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1413{ \ 1414 static gen_helper_opivx * const fns[3] = { \ 1415 gen_helper_##NAME##_b, \ 1416 gen_helper_##NAME##_h, \ 1417 gen_helper_##NAME##_w \ 1418 }; \ 1419 return do_opiwx_widen(s, a, fns[s->sew]); \ 1420} 1421 1422GEN_OPIWX_WIDEN_TRANS(vwaddu_wx) 1423GEN_OPIWX_WIDEN_TRANS(vwadd_wx) 1424GEN_OPIWX_WIDEN_TRANS(vwsubu_wx) 1425GEN_OPIWX_WIDEN_TRANS(vwsub_wx) 1426 1427/* Vector Integer Add-with-Carry / Subtract-with-Borrow Instructions */ 1428/* OPIVV without GVEC IR */ 1429#define GEN_OPIVV_TRANS(NAME, CHECK) \ 1430static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1431{ \ 1432 if (CHECK(s, a)) { \ 1433 uint32_t data = 0; \ 1434 static gen_helper_gvec_4_ptr * const fns[4] = { \ 1435 gen_helper_##NAME##_b, gen_helper_##NAME##_h, \ 1436 gen_helper_##NAME##_w, gen_helper_##NAME##_d, \ 1437 }; \ 1438 TCGLabel *over = gen_new_label(); \ 1439 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); \ 1440 \ 1441 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 1442 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 1443 tcg_gen_gvec_4_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), \ 1444 vreg_ofs(s, a->rs1), \ 1445 vreg_ofs(s, a->rs2), cpu_env, \ 1446 s->vlen / 8, s->vlen / 8, data, \ 1447 fns[s->sew]); \ 1448 mark_vs_dirty(s); \ 1449 gen_set_label(over); \ 1450 return true; \ 1451 } \ 1452 return false; \ 1453} 1454 1455/* 1456 * For vadc and vsbc, an illegal instruction exception is raised if the 1457 * destination vector register is v0 and LMUL > 1. (Section 12.3) 1458 */ 1459static bool opivv_vadc_check(DisasContext *s, arg_rmrr *a) 1460{ 1461 return require_rvv(s) && 1462 vext_check_isa_ill(s) && 1463 (a->rd != 0) && 1464 vext_check_sss(s, a->rd, a->rs1, a->rs2, a->vm); 1465} 1466 1467GEN_OPIVV_TRANS(vadc_vvm, opivv_vadc_check) 1468GEN_OPIVV_TRANS(vsbc_vvm, opivv_vadc_check) 1469 1470/* 1471 * For vmadc and vmsbc, an illegal instruction exception is raised if the 1472 * destination vector register overlaps a source vector register group. 1473 */ 1474static bool opivv_vmadc_check(DisasContext *s, arg_rmrr *a) 1475{ 1476 return require_rvv(s) && 1477 vext_check_isa_ill(s) && 1478 vext_check_mss(s, a->rd, a->rs1, a->rs2); 1479} 1480 1481GEN_OPIVV_TRANS(vmadc_vvm, opivv_vmadc_check) 1482GEN_OPIVV_TRANS(vmsbc_vvm, opivv_vmadc_check) 1483 1484static bool opivx_vadc_check(DisasContext *s, arg_rmrr *a) 1485{ 1486 return require_rvv(s) && 1487 vext_check_isa_ill(s) && 1488 (a->rd != 0) && 1489 vext_check_ss(s, a->rd, a->rs2, a->vm); 1490} 1491 1492/* OPIVX without GVEC IR */ 1493#define GEN_OPIVX_TRANS(NAME, CHECK) \ 1494static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1495{ \ 1496 if (CHECK(s, a)) { \ 1497 static gen_helper_opivx * const fns[4] = { \ 1498 gen_helper_##NAME##_b, gen_helper_##NAME##_h, \ 1499 gen_helper_##NAME##_w, gen_helper_##NAME##_d, \ 1500 }; \ 1501 \ 1502 return opivx_trans(a->rd, a->rs1, a->rs2, a->vm, fns[s->sew], s);\ 1503 } \ 1504 return false; \ 1505} 1506 1507GEN_OPIVX_TRANS(vadc_vxm, opivx_vadc_check) 1508GEN_OPIVX_TRANS(vsbc_vxm, opivx_vadc_check) 1509 1510static bool opivx_vmadc_check(DisasContext *s, arg_rmrr *a) 1511{ 1512 return require_rvv(s) && 1513 vext_check_isa_ill(s) && 1514 vext_check_ms(s, a->rd, a->rs2); 1515} 1516 1517GEN_OPIVX_TRANS(vmadc_vxm, opivx_vmadc_check) 1518GEN_OPIVX_TRANS(vmsbc_vxm, opivx_vmadc_check) 1519 1520/* OPIVI without GVEC IR */ 1521#define GEN_OPIVI_TRANS(NAME, IMM_MODE, OPIVX, CHECK) \ 1522static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1523{ \ 1524 if (CHECK(s, a)) { \ 1525 static gen_helper_opivx * const fns[4] = { \ 1526 gen_helper_##OPIVX##_b, gen_helper_##OPIVX##_h, \ 1527 gen_helper_##OPIVX##_w, gen_helper_##OPIVX##_d, \ 1528 }; \ 1529 return opivi_trans(a->rd, a->rs1, a->rs2, a->vm, \ 1530 fns[s->sew], s, IMM_MODE); \ 1531 } \ 1532 return false; \ 1533} 1534 1535GEN_OPIVI_TRANS(vadc_vim, IMM_SX, vadc_vxm, opivx_vadc_check) 1536GEN_OPIVI_TRANS(vmadc_vim, IMM_SX, vmadc_vxm, opivx_vmadc_check) 1537 1538/* Vector Bitwise Logical Instructions */ 1539GEN_OPIVV_GVEC_TRANS(vand_vv, and) 1540GEN_OPIVV_GVEC_TRANS(vor_vv, or) 1541GEN_OPIVV_GVEC_TRANS(vxor_vv, xor) 1542GEN_OPIVX_GVEC_TRANS(vand_vx, ands) 1543GEN_OPIVX_GVEC_TRANS(vor_vx, ors) 1544GEN_OPIVX_GVEC_TRANS(vxor_vx, xors) 1545GEN_OPIVI_GVEC_TRANS(vand_vi, IMM_SX, vand_vx, andi) 1546GEN_OPIVI_GVEC_TRANS(vor_vi, IMM_SX, vor_vx, ori) 1547GEN_OPIVI_GVEC_TRANS(vxor_vi, IMM_SX, vxor_vx, xori) 1548 1549/* Vector Single-Width Bit Shift Instructions */ 1550GEN_OPIVV_GVEC_TRANS(vsll_vv, shlv) 1551GEN_OPIVV_GVEC_TRANS(vsrl_vv, shrv) 1552GEN_OPIVV_GVEC_TRANS(vsra_vv, sarv) 1553 1554typedef void GVecGen2sFn32(unsigned, uint32_t, uint32_t, TCGv_i32, 1555 uint32_t, uint32_t); 1556 1557static inline bool 1558do_opivx_gvec_shift(DisasContext *s, arg_rmrr *a, GVecGen2sFn32 *gvec_fn, 1559 gen_helper_opivx *fn) 1560{ 1561 if (!opivx_check(s, a)) { 1562 return false; 1563 } 1564 1565 if (a->vm && s->vl_eq_vlmax) { 1566 TCGv_i32 src1 = tcg_temp_new_i32(); 1567 1568 tcg_gen_trunc_tl_i32(src1, get_gpr(s, a->rs1, EXT_NONE)); 1569 tcg_gen_extract_i32(src1, src1, 0, s->sew + 3); 1570 gvec_fn(s->sew, vreg_ofs(s, a->rd), vreg_ofs(s, a->rs2), 1571 src1, MAXSZ(s), MAXSZ(s)); 1572 1573 tcg_temp_free_i32(src1); 1574 mark_vs_dirty(s); 1575 return true; 1576 } 1577 return opivx_trans(a->rd, a->rs1, a->rs2, a->vm, fn, s); 1578} 1579 1580#define GEN_OPIVX_GVEC_SHIFT_TRANS(NAME, SUF) \ 1581static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1582{ \ 1583 static gen_helper_opivx * const fns[4] = { \ 1584 gen_helper_##NAME##_b, gen_helper_##NAME##_h, \ 1585 gen_helper_##NAME##_w, gen_helper_##NAME##_d, \ 1586 }; \ 1587 \ 1588 return do_opivx_gvec_shift(s, a, tcg_gen_gvec_##SUF, fns[s->sew]); \ 1589} 1590 1591GEN_OPIVX_GVEC_SHIFT_TRANS(vsll_vx, shls) 1592GEN_OPIVX_GVEC_SHIFT_TRANS(vsrl_vx, shrs) 1593GEN_OPIVX_GVEC_SHIFT_TRANS(vsra_vx, sars) 1594 1595GEN_OPIVI_GVEC_TRANS(vsll_vi, IMM_ZX, vsll_vx, shli) 1596GEN_OPIVI_GVEC_TRANS(vsrl_vi, IMM_ZX, vsrl_vx, shri) 1597GEN_OPIVI_GVEC_TRANS(vsra_vi, IMM_ZX, vsra_vx, sari) 1598 1599/* Vector Narrowing Integer Right Shift Instructions */ 1600static bool opivv_narrow_check(DisasContext *s, arg_rmrr *a) 1601{ 1602 return require_rvv(s) && 1603 vext_check_isa_ill(s) && 1604 vext_check_sds(s, a->rd, a->rs1, a->rs2, a->vm); 1605} 1606 1607/* OPIVV with NARROW */ 1608#define GEN_OPIVV_NARROW_TRANS(NAME) \ 1609static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1610{ \ 1611 if (opivv_narrow_check(s, a)) { \ 1612 uint32_t data = 0; \ 1613 static gen_helper_gvec_4_ptr * const fns[3] = { \ 1614 gen_helper_##NAME##_b, \ 1615 gen_helper_##NAME##_h, \ 1616 gen_helper_##NAME##_w, \ 1617 }; \ 1618 TCGLabel *over = gen_new_label(); \ 1619 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); \ 1620 \ 1621 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 1622 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 1623 tcg_gen_gvec_4_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), \ 1624 vreg_ofs(s, a->rs1), \ 1625 vreg_ofs(s, a->rs2), cpu_env, \ 1626 s->vlen / 8, s->vlen / 8, data, \ 1627 fns[s->sew]); \ 1628 mark_vs_dirty(s); \ 1629 gen_set_label(over); \ 1630 return true; \ 1631 } \ 1632 return false; \ 1633} 1634GEN_OPIVV_NARROW_TRANS(vnsra_vv) 1635GEN_OPIVV_NARROW_TRANS(vnsrl_vv) 1636 1637static bool opivx_narrow_check(DisasContext *s, arg_rmrr *a) 1638{ 1639 return require_rvv(s) && 1640 vext_check_isa_ill(s) && 1641 vext_check_sd(s, a->rd, a->rs2, a->vm); 1642} 1643 1644/* OPIVX with NARROW */ 1645#define GEN_OPIVX_NARROW_TRANS(NAME) \ 1646static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1647{ \ 1648 if (opivx_narrow_check(s, a)) { \ 1649 static gen_helper_opivx * const fns[3] = { \ 1650 gen_helper_##NAME##_b, \ 1651 gen_helper_##NAME##_h, \ 1652 gen_helper_##NAME##_w, \ 1653 }; \ 1654 return opivx_trans(a->rd, a->rs1, a->rs2, a->vm, fns[s->sew], s);\ 1655 } \ 1656 return false; \ 1657} 1658 1659GEN_OPIVX_NARROW_TRANS(vnsra_vx) 1660GEN_OPIVX_NARROW_TRANS(vnsrl_vx) 1661 1662/* OPIVI with NARROW */ 1663#define GEN_OPIVI_NARROW_TRANS(NAME, IMM_MODE, OPIVX) \ 1664static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1665{ \ 1666 if (opivx_narrow_check(s, a)) { \ 1667 static gen_helper_opivx * const fns[3] = { \ 1668 gen_helper_##OPIVX##_b, \ 1669 gen_helper_##OPIVX##_h, \ 1670 gen_helper_##OPIVX##_w, \ 1671 }; \ 1672 return opivi_trans(a->rd, a->rs1, a->rs2, a->vm, \ 1673 fns[s->sew], s, IMM_MODE); \ 1674 } \ 1675 return false; \ 1676} 1677 1678GEN_OPIVI_NARROW_TRANS(vnsra_vi, IMM_ZX, vnsra_vx) 1679GEN_OPIVI_NARROW_TRANS(vnsrl_vi, IMM_ZX, vnsrl_vx) 1680 1681/* Vector Integer Comparison Instructions */ 1682/* 1683 * For all comparison instructions, an illegal instruction exception is raised 1684 * if the destination vector register overlaps a source vector register group 1685 * and LMUL > 1. 1686 */ 1687static bool opivv_cmp_check(DisasContext *s, arg_rmrr *a) 1688{ 1689 return require_rvv(s) && 1690 vext_check_isa_ill(s) && 1691 vext_check_mss(s, a->rd, a->rs1, a->rs2); 1692} 1693 1694GEN_OPIVV_TRANS(vmseq_vv, opivv_cmp_check) 1695GEN_OPIVV_TRANS(vmsne_vv, opivv_cmp_check) 1696GEN_OPIVV_TRANS(vmsltu_vv, opivv_cmp_check) 1697GEN_OPIVV_TRANS(vmslt_vv, opivv_cmp_check) 1698GEN_OPIVV_TRANS(vmsleu_vv, opivv_cmp_check) 1699GEN_OPIVV_TRANS(vmsle_vv, opivv_cmp_check) 1700 1701static bool opivx_cmp_check(DisasContext *s, arg_rmrr *a) 1702{ 1703 return require_rvv(s) && 1704 vext_check_isa_ill(s) && 1705 vext_check_ms(s, a->rd, a->rs2); 1706} 1707 1708GEN_OPIVX_TRANS(vmseq_vx, opivx_cmp_check) 1709GEN_OPIVX_TRANS(vmsne_vx, opivx_cmp_check) 1710GEN_OPIVX_TRANS(vmsltu_vx, opivx_cmp_check) 1711GEN_OPIVX_TRANS(vmslt_vx, opivx_cmp_check) 1712GEN_OPIVX_TRANS(vmsleu_vx, opivx_cmp_check) 1713GEN_OPIVX_TRANS(vmsle_vx, opivx_cmp_check) 1714GEN_OPIVX_TRANS(vmsgtu_vx, opivx_cmp_check) 1715GEN_OPIVX_TRANS(vmsgt_vx, opivx_cmp_check) 1716 1717GEN_OPIVI_TRANS(vmseq_vi, IMM_SX, vmseq_vx, opivx_cmp_check) 1718GEN_OPIVI_TRANS(vmsne_vi, IMM_SX, vmsne_vx, opivx_cmp_check) 1719GEN_OPIVI_TRANS(vmsleu_vi, IMM_ZX, vmsleu_vx, opivx_cmp_check) 1720GEN_OPIVI_TRANS(vmsle_vi, IMM_SX, vmsle_vx, opivx_cmp_check) 1721GEN_OPIVI_TRANS(vmsgtu_vi, IMM_ZX, vmsgtu_vx, opivx_cmp_check) 1722GEN_OPIVI_TRANS(vmsgt_vi, IMM_SX, vmsgt_vx, opivx_cmp_check) 1723 1724/* Vector Integer Min/Max Instructions */ 1725GEN_OPIVV_GVEC_TRANS(vminu_vv, umin) 1726GEN_OPIVV_GVEC_TRANS(vmin_vv, smin) 1727GEN_OPIVV_GVEC_TRANS(vmaxu_vv, umax) 1728GEN_OPIVV_GVEC_TRANS(vmax_vv, smax) 1729GEN_OPIVX_TRANS(vminu_vx, opivx_check) 1730GEN_OPIVX_TRANS(vmin_vx, opivx_check) 1731GEN_OPIVX_TRANS(vmaxu_vx, opivx_check) 1732GEN_OPIVX_TRANS(vmax_vx, opivx_check) 1733 1734/* Vector Single-Width Integer Multiply Instructions */ 1735GEN_OPIVV_GVEC_TRANS(vmul_vv, mul) 1736GEN_OPIVV_TRANS(vmulh_vv, opivv_check) 1737GEN_OPIVV_TRANS(vmulhu_vv, opivv_check) 1738GEN_OPIVV_TRANS(vmulhsu_vv, opivv_check) 1739GEN_OPIVX_GVEC_TRANS(vmul_vx, muls) 1740GEN_OPIVX_TRANS(vmulh_vx, opivx_check) 1741GEN_OPIVX_TRANS(vmulhu_vx, opivx_check) 1742GEN_OPIVX_TRANS(vmulhsu_vx, opivx_check) 1743 1744/* Vector Integer Divide Instructions */ 1745GEN_OPIVV_TRANS(vdivu_vv, opivv_check) 1746GEN_OPIVV_TRANS(vdiv_vv, opivv_check) 1747GEN_OPIVV_TRANS(vremu_vv, opivv_check) 1748GEN_OPIVV_TRANS(vrem_vv, opivv_check) 1749GEN_OPIVX_TRANS(vdivu_vx, opivx_check) 1750GEN_OPIVX_TRANS(vdiv_vx, opivx_check) 1751GEN_OPIVX_TRANS(vremu_vx, opivx_check) 1752GEN_OPIVX_TRANS(vrem_vx, opivx_check) 1753 1754/* Vector Widening Integer Multiply Instructions */ 1755GEN_OPIVV_WIDEN_TRANS(vwmul_vv, opivv_widen_check) 1756GEN_OPIVV_WIDEN_TRANS(vwmulu_vv, opivv_widen_check) 1757GEN_OPIVV_WIDEN_TRANS(vwmulsu_vv, opivv_widen_check) 1758GEN_OPIVX_WIDEN_TRANS(vwmul_vx) 1759GEN_OPIVX_WIDEN_TRANS(vwmulu_vx) 1760GEN_OPIVX_WIDEN_TRANS(vwmulsu_vx) 1761 1762/* Vector Single-Width Integer Multiply-Add Instructions */ 1763GEN_OPIVV_TRANS(vmacc_vv, opivv_check) 1764GEN_OPIVV_TRANS(vnmsac_vv, opivv_check) 1765GEN_OPIVV_TRANS(vmadd_vv, opivv_check) 1766GEN_OPIVV_TRANS(vnmsub_vv, opivv_check) 1767GEN_OPIVX_TRANS(vmacc_vx, opivx_check) 1768GEN_OPIVX_TRANS(vnmsac_vx, opivx_check) 1769GEN_OPIVX_TRANS(vmadd_vx, opivx_check) 1770GEN_OPIVX_TRANS(vnmsub_vx, opivx_check) 1771 1772/* Vector Widening Integer Multiply-Add Instructions */ 1773GEN_OPIVV_WIDEN_TRANS(vwmaccu_vv, opivv_widen_check) 1774GEN_OPIVV_WIDEN_TRANS(vwmacc_vv, opivv_widen_check) 1775GEN_OPIVV_WIDEN_TRANS(vwmaccsu_vv, opivv_widen_check) 1776GEN_OPIVX_WIDEN_TRANS(vwmaccu_vx) 1777GEN_OPIVX_WIDEN_TRANS(vwmacc_vx) 1778GEN_OPIVX_WIDEN_TRANS(vwmaccsu_vx) 1779GEN_OPIVX_WIDEN_TRANS(vwmaccus_vx) 1780 1781/* Vector Integer Merge and Move Instructions */ 1782static bool trans_vmv_v_v(DisasContext *s, arg_vmv_v_v *a) 1783{ 1784 if (require_rvv(s) && 1785 vext_check_isa_ill(s) && 1786 /* vmv.v.v has rs2 = 0 and vm = 1 */ 1787 vext_check_sss(s, a->rd, a->rs1, 0, 1)) { 1788 if (s->vl_eq_vlmax) { 1789 tcg_gen_gvec_mov(s->sew, vreg_ofs(s, a->rd), 1790 vreg_ofs(s, a->rs1), 1791 MAXSZ(s), MAXSZ(s)); 1792 } else { 1793 uint32_t data = FIELD_DP32(0, VDATA, LMUL, s->lmul); 1794 static gen_helper_gvec_2_ptr * const fns[4] = { 1795 gen_helper_vmv_v_v_b, gen_helper_vmv_v_v_h, 1796 gen_helper_vmv_v_v_w, gen_helper_vmv_v_v_d, 1797 }; 1798 TCGLabel *over = gen_new_label(); 1799 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 1800 1801 tcg_gen_gvec_2_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, a->rs1), 1802 cpu_env, s->vlen / 8, s->vlen / 8, data, 1803 fns[s->sew]); 1804 gen_set_label(over); 1805 } 1806 mark_vs_dirty(s); 1807 return true; 1808 } 1809 return false; 1810} 1811 1812typedef void gen_helper_vmv_vx(TCGv_ptr, TCGv_i64, TCGv_env, TCGv_i32); 1813static bool trans_vmv_v_x(DisasContext *s, arg_vmv_v_x *a) 1814{ 1815 if (require_rvv(s) && 1816 vext_check_isa_ill(s) && 1817 /* vmv.v.x has rs2 = 0 and vm = 1 */ 1818 vext_check_ss(s, a->rd, 0, 1)) { 1819 TCGv s1; 1820 TCGLabel *over = gen_new_label(); 1821 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 1822 1823 s1 = get_gpr(s, a->rs1, EXT_SIGN); 1824 1825 if (s->vl_eq_vlmax) { 1826 tcg_gen_gvec_dup_tl(s->sew, vreg_ofs(s, a->rd), 1827 MAXSZ(s), MAXSZ(s), s1); 1828 } else { 1829 TCGv_i32 desc; 1830 TCGv_i64 s1_i64 = tcg_temp_new_i64(); 1831 TCGv_ptr dest = tcg_temp_new_ptr(); 1832 uint32_t data = FIELD_DP32(0, VDATA, LMUL, s->lmul); 1833 static gen_helper_vmv_vx * const fns[4] = { 1834 gen_helper_vmv_v_x_b, gen_helper_vmv_v_x_h, 1835 gen_helper_vmv_v_x_w, gen_helper_vmv_v_x_d, 1836 }; 1837 1838 tcg_gen_ext_tl_i64(s1_i64, s1); 1839 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 1840 tcg_gen_addi_ptr(dest, cpu_env, vreg_ofs(s, a->rd)); 1841 fns[s->sew](dest, s1_i64, cpu_env, desc); 1842 1843 tcg_temp_free_ptr(dest); 1844 tcg_temp_free_i64(s1_i64); 1845 } 1846 1847 mark_vs_dirty(s); 1848 gen_set_label(over); 1849 return true; 1850 } 1851 return false; 1852} 1853 1854static bool trans_vmv_v_i(DisasContext *s, arg_vmv_v_i *a) 1855{ 1856 if (require_rvv(s) && 1857 vext_check_isa_ill(s) && 1858 /* vmv.v.i has rs2 = 0 and vm = 1 */ 1859 vext_check_ss(s, a->rd, 0, 1)) { 1860 int64_t simm = sextract64(a->rs1, 0, 5); 1861 if (s->vl_eq_vlmax) { 1862 tcg_gen_gvec_dup_imm(s->sew, vreg_ofs(s, a->rd), 1863 MAXSZ(s), MAXSZ(s), simm); 1864 mark_vs_dirty(s); 1865 } else { 1866 TCGv_i32 desc; 1867 TCGv_i64 s1; 1868 TCGv_ptr dest; 1869 uint32_t data = FIELD_DP32(0, VDATA, LMUL, s->lmul); 1870 static gen_helper_vmv_vx * const fns[4] = { 1871 gen_helper_vmv_v_x_b, gen_helper_vmv_v_x_h, 1872 gen_helper_vmv_v_x_w, gen_helper_vmv_v_x_d, 1873 }; 1874 TCGLabel *over = gen_new_label(); 1875 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 1876 1877 s1 = tcg_constant_i64(simm); 1878 dest = tcg_temp_new_ptr(); 1879 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 1880 tcg_gen_addi_ptr(dest, cpu_env, vreg_ofs(s, a->rd)); 1881 fns[s->sew](dest, s1, cpu_env, desc); 1882 1883 tcg_temp_free_ptr(dest); 1884 mark_vs_dirty(s); 1885 gen_set_label(over); 1886 } 1887 return true; 1888 } 1889 return false; 1890} 1891 1892GEN_OPIVV_TRANS(vmerge_vvm, opivv_vadc_check) 1893GEN_OPIVX_TRANS(vmerge_vxm, opivx_vadc_check) 1894GEN_OPIVI_TRANS(vmerge_vim, IMM_SX, vmerge_vxm, opivx_vadc_check) 1895 1896/* 1897 *** Vector Fixed-Point Arithmetic Instructions 1898 */ 1899 1900/* Vector Single-Width Saturating Add and Subtract */ 1901GEN_OPIVV_TRANS(vsaddu_vv, opivv_check) 1902GEN_OPIVV_TRANS(vsadd_vv, opivv_check) 1903GEN_OPIVV_TRANS(vssubu_vv, opivv_check) 1904GEN_OPIVV_TRANS(vssub_vv, opivv_check) 1905GEN_OPIVX_TRANS(vsaddu_vx, opivx_check) 1906GEN_OPIVX_TRANS(vsadd_vx, opivx_check) 1907GEN_OPIVX_TRANS(vssubu_vx, opivx_check) 1908GEN_OPIVX_TRANS(vssub_vx, opivx_check) 1909GEN_OPIVI_TRANS(vsaddu_vi, IMM_ZX, vsaddu_vx, opivx_check) 1910GEN_OPIVI_TRANS(vsadd_vi, IMM_SX, vsadd_vx, opivx_check) 1911 1912/* Vector Single-Width Averaging Add and Subtract */ 1913GEN_OPIVV_TRANS(vaadd_vv, opivv_check) 1914GEN_OPIVV_TRANS(vasub_vv, opivv_check) 1915GEN_OPIVX_TRANS(vaadd_vx, opivx_check) 1916GEN_OPIVX_TRANS(vasub_vx, opivx_check) 1917GEN_OPIVI_TRANS(vaadd_vi, 0, vaadd_vx, opivx_check) 1918 1919/* Vector Single-Width Fractional Multiply with Rounding and Saturation */ 1920GEN_OPIVV_TRANS(vsmul_vv, opivv_check) 1921GEN_OPIVX_TRANS(vsmul_vx, opivx_check) 1922 1923/* Vector Widening Saturating Scaled Multiply-Add */ 1924GEN_OPIVV_WIDEN_TRANS(vwsmaccu_vv, opivv_widen_check) 1925GEN_OPIVV_WIDEN_TRANS(vwsmacc_vv, opivv_widen_check) 1926GEN_OPIVV_WIDEN_TRANS(vwsmaccsu_vv, opivv_widen_check) 1927GEN_OPIVX_WIDEN_TRANS(vwsmaccu_vx) 1928GEN_OPIVX_WIDEN_TRANS(vwsmacc_vx) 1929GEN_OPIVX_WIDEN_TRANS(vwsmaccsu_vx) 1930GEN_OPIVX_WIDEN_TRANS(vwsmaccus_vx) 1931 1932/* Vector Single-Width Scaling Shift Instructions */ 1933GEN_OPIVV_TRANS(vssrl_vv, opivv_check) 1934GEN_OPIVV_TRANS(vssra_vv, opivv_check) 1935GEN_OPIVX_TRANS(vssrl_vx, opivx_check) 1936GEN_OPIVX_TRANS(vssra_vx, opivx_check) 1937GEN_OPIVI_TRANS(vssrl_vi, IMM_ZX, vssrl_vx, opivx_check) 1938GEN_OPIVI_TRANS(vssra_vi, IMM_SX, vssra_vx, opivx_check) 1939 1940/* Vector Narrowing Fixed-Point Clip Instructions */ 1941GEN_OPIVV_NARROW_TRANS(vnclipu_vv) 1942GEN_OPIVV_NARROW_TRANS(vnclip_vv) 1943GEN_OPIVX_NARROW_TRANS(vnclipu_vx) 1944GEN_OPIVX_NARROW_TRANS(vnclip_vx) 1945GEN_OPIVI_NARROW_TRANS(vnclipu_vi, IMM_ZX, vnclipu_vx) 1946GEN_OPIVI_NARROW_TRANS(vnclip_vi, IMM_ZX, vnclip_vx) 1947 1948/* 1949 *** Vector Float Point Arithmetic Instructions 1950 */ 1951 1952/* 1953 * As RVF-only cpus always have values NaN-boxed to 64-bits, 1954 * RVF and RVD can be treated equally. 1955 * We don't have to deal with the cases of: SEW > FLEN. 1956 * 1957 * If SEW < FLEN, check whether input fp register is a valid 1958 * NaN-boxed value, in which case the least-significant SEW bits 1959 * of the f regsiter are used, else the canonical NaN value is used. 1960 */ 1961static void do_nanbox(DisasContext *s, TCGv_i64 out, TCGv_i64 in) 1962{ 1963 switch (s->sew) { 1964 case 1: 1965 gen_check_nanbox_h(out, in); 1966 break; 1967 case 2: 1968 gen_check_nanbox_s(out, in); 1969 break; 1970 case 3: 1971 tcg_gen_mov_i64(out, in); 1972 break; 1973 default: 1974 g_assert_not_reached(); 1975 } 1976} 1977 1978/* Vector Single-Width Floating-Point Add/Subtract Instructions */ 1979 1980/* 1981 * If the current SEW does not correspond to a supported IEEE floating-point 1982 * type, an illegal instruction exception is raised. 1983 */ 1984static bool opfvv_check(DisasContext *s, arg_rmrr *a) 1985{ 1986 return require_rvv(s) && 1987 require_rvf(s) && 1988 vext_check_isa_ill(s) && 1989 vext_check_sss(s, a->rd, a->rs1, a->rs2, a->vm); 1990} 1991 1992/* OPFVV without GVEC IR */ 1993#define GEN_OPFVV_TRANS(NAME, CHECK) \ 1994static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 1995{ \ 1996 if (CHECK(s, a)) { \ 1997 uint32_t data = 0; \ 1998 static gen_helper_gvec_4_ptr * const fns[3] = { \ 1999 gen_helper_##NAME##_h, \ 2000 gen_helper_##NAME##_w, \ 2001 gen_helper_##NAME##_d, \ 2002 }; \ 2003 TCGLabel *over = gen_new_label(); \ 2004 gen_set_rm(s, 7); \ 2005 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); \ 2006 \ 2007 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 2008 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 2009 tcg_gen_gvec_4_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), \ 2010 vreg_ofs(s, a->rs1), \ 2011 vreg_ofs(s, a->rs2), cpu_env, \ 2012 s->vlen / 8, s->vlen / 8, data, \ 2013 fns[s->sew - 1]); \ 2014 mark_vs_dirty(s); \ 2015 gen_set_label(over); \ 2016 return true; \ 2017 } \ 2018 return false; \ 2019} 2020GEN_OPFVV_TRANS(vfadd_vv, opfvv_check) 2021GEN_OPFVV_TRANS(vfsub_vv, opfvv_check) 2022 2023typedef void gen_helper_opfvf(TCGv_ptr, TCGv_ptr, TCGv_i64, TCGv_ptr, 2024 TCGv_env, TCGv_i32); 2025 2026static bool opfvf_trans(uint32_t vd, uint32_t rs1, uint32_t vs2, 2027 uint32_t data, gen_helper_opfvf *fn, DisasContext *s) 2028{ 2029 TCGv_ptr dest, src2, mask; 2030 TCGv_i32 desc; 2031 TCGv_i64 t1; 2032 2033 TCGLabel *over = gen_new_label(); 2034 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 2035 2036 dest = tcg_temp_new_ptr(); 2037 mask = tcg_temp_new_ptr(); 2038 src2 = tcg_temp_new_ptr(); 2039 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 2040 2041 tcg_gen_addi_ptr(dest, cpu_env, vreg_ofs(s, vd)); 2042 tcg_gen_addi_ptr(src2, cpu_env, vreg_ofs(s, vs2)); 2043 tcg_gen_addi_ptr(mask, cpu_env, vreg_ofs(s, 0)); 2044 2045 /* NaN-box f[rs1] */ 2046 t1 = tcg_temp_new_i64(); 2047 do_nanbox(s, t1, cpu_fpr[rs1]); 2048 2049 fn(dest, mask, t1, src2, cpu_env, desc); 2050 2051 tcg_temp_free_ptr(dest); 2052 tcg_temp_free_ptr(mask); 2053 tcg_temp_free_ptr(src2); 2054 tcg_temp_free_i64(t1); 2055 mark_vs_dirty(s); 2056 gen_set_label(over); 2057 return true; 2058} 2059 2060/* 2061 * If the current SEW does not correspond to a supported IEEE floating-point 2062 * type, an illegal instruction exception is raised 2063 */ 2064static bool opfvf_check(DisasContext *s, arg_rmrr *a) 2065{ 2066 return require_rvv(s) && 2067 require_rvf(s) && 2068 vext_check_isa_ill(s) && 2069 vext_check_ss(s, a->rd, a->rs2, a->vm); 2070} 2071 2072/* OPFVF without GVEC IR */ 2073#define GEN_OPFVF_TRANS(NAME, CHECK) \ 2074static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 2075{ \ 2076 if (CHECK(s, a)) { \ 2077 uint32_t data = 0; \ 2078 static gen_helper_opfvf *const fns[3] = { \ 2079 gen_helper_##NAME##_h, \ 2080 gen_helper_##NAME##_w, \ 2081 gen_helper_##NAME##_d, \ 2082 }; \ 2083 gen_set_rm(s, 7); \ 2084 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 2085 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 2086 return opfvf_trans(a->rd, a->rs1, a->rs2, data, \ 2087 fns[s->sew - 1], s); \ 2088 } \ 2089 return false; \ 2090} 2091 2092GEN_OPFVF_TRANS(vfadd_vf, opfvf_check) 2093GEN_OPFVF_TRANS(vfsub_vf, opfvf_check) 2094GEN_OPFVF_TRANS(vfrsub_vf, opfvf_check) 2095 2096/* Vector Widening Floating-Point Add/Subtract Instructions */ 2097static bool opfvv_widen_check(DisasContext *s, arg_rmrr *a) 2098{ 2099 return require_rvv(s) && 2100 require_rvf(s) && 2101 vext_check_isa_ill(s) && 2102 vext_check_dss(s, a->rd, a->rs1, a->rs2, a->vm); 2103} 2104 2105/* OPFVV with WIDEN */ 2106#define GEN_OPFVV_WIDEN_TRANS(NAME, CHECK) \ 2107static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 2108{ \ 2109 if (CHECK(s, a)) { \ 2110 uint32_t data = 0; \ 2111 static gen_helper_gvec_4_ptr * const fns[2] = { \ 2112 gen_helper_##NAME##_h, gen_helper_##NAME##_w, \ 2113 }; \ 2114 TCGLabel *over = gen_new_label(); \ 2115 gen_set_rm(s, 7); \ 2116 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); \ 2117 \ 2118 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 2119 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 2120 tcg_gen_gvec_4_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), \ 2121 vreg_ofs(s, a->rs1), \ 2122 vreg_ofs(s, a->rs2), cpu_env, \ 2123 s->vlen / 8, s->vlen / 8, data, \ 2124 fns[s->sew - 1]); \ 2125 mark_vs_dirty(s); \ 2126 gen_set_label(over); \ 2127 return true; \ 2128 } \ 2129 return false; \ 2130} 2131 2132GEN_OPFVV_WIDEN_TRANS(vfwadd_vv, opfvv_widen_check) 2133GEN_OPFVV_WIDEN_TRANS(vfwsub_vv, opfvv_widen_check) 2134 2135static bool opfvf_widen_check(DisasContext *s, arg_rmrr *a) 2136{ 2137 return require_rvv(s) && 2138 require_rvf(s) && 2139 vext_check_isa_ill(s) && 2140 vext_check_ds(s, a->rd, a->rs2, a->vm); 2141} 2142 2143/* OPFVF with WIDEN */ 2144#define GEN_OPFVF_WIDEN_TRANS(NAME) \ 2145static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 2146{ \ 2147 if (opfvf_widen_check(s, a)) { \ 2148 uint32_t data = 0; \ 2149 static gen_helper_opfvf *const fns[2] = { \ 2150 gen_helper_##NAME##_h, gen_helper_##NAME##_w, \ 2151 }; \ 2152 gen_set_rm(s, 7); \ 2153 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 2154 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 2155 return opfvf_trans(a->rd, a->rs1, a->rs2, data, \ 2156 fns[s->sew - 1], s); \ 2157 } \ 2158 return false; \ 2159} 2160 2161GEN_OPFVF_WIDEN_TRANS(vfwadd_vf) 2162GEN_OPFVF_WIDEN_TRANS(vfwsub_vf) 2163 2164static bool opfwv_widen_check(DisasContext *s, arg_rmrr *a) 2165{ 2166 return require_rvv(s) && 2167 require_rvf(s) && 2168 vext_check_isa_ill(s) && 2169 vext_check_dds(s, a->rd, a->rs1, a->rs2, a->vm); 2170} 2171 2172/* WIDEN OPFVV with WIDEN */ 2173#define GEN_OPFWV_WIDEN_TRANS(NAME) \ 2174static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 2175{ \ 2176 if (opfwv_widen_check(s, a)) { \ 2177 uint32_t data = 0; \ 2178 static gen_helper_gvec_4_ptr * const fns[2] = { \ 2179 gen_helper_##NAME##_h, gen_helper_##NAME##_w, \ 2180 }; \ 2181 TCGLabel *over = gen_new_label(); \ 2182 gen_set_rm(s, 7); \ 2183 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); \ 2184 \ 2185 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 2186 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 2187 tcg_gen_gvec_4_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), \ 2188 vreg_ofs(s, a->rs1), \ 2189 vreg_ofs(s, a->rs2), cpu_env, \ 2190 s->vlen / 8, s->vlen / 8, data, \ 2191 fns[s->sew - 1]); \ 2192 mark_vs_dirty(s); \ 2193 gen_set_label(over); \ 2194 return true; \ 2195 } \ 2196 return false; \ 2197} 2198 2199GEN_OPFWV_WIDEN_TRANS(vfwadd_wv) 2200GEN_OPFWV_WIDEN_TRANS(vfwsub_wv) 2201 2202static bool opfwf_widen_check(DisasContext *s, arg_rmrr *a) 2203{ 2204 return require_rvv(s) && 2205 require_rvf(s) && 2206 vext_check_isa_ill(s) && 2207 vext_check_dd(s, a->rd, a->rs2, a->vm); 2208} 2209 2210/* WIDEN OPFVF with WIDEN */ 2211#define GEN_OPFWF_WIDEN_TRANS(NAME) \ 2212static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ 2213{ \ 2214 if (opfwf_widen_check(s, a)) { \ 2215 uint32_t data = 0; \ 2216 static gen_helper_opfvf *const fns[2] = { \ 2217 gen_helper_##NAME##_h, gen_helper_##NAME##_w, \ 2218 }; \ 2219 gen_set_rm(s, 7); \ 2220 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 2221 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 2222 return opfvf_trans(a->rd, a->rs1, a->rs2, data, \ 2223 fns[s->sew - 1], s); \ 2224 } \ 2225 return false; \ 2226} 2227 2228GEN_OPFWF_WIDEN_TRANS(vfwadd_wf) 2229GEN_OPFWF_WIDEN_TRANS(vfwsub_wf) 2230 2231/* Vector Single-Width Floating-Point Multiply/Divide Instructions */ 2232GEN_OPFVV_TRANS(vfmul_vv, opfvv_check) 2233GEN_OPFVV_TRANS(vfdiv_vv, opfvv_check) 2234GEN_OPFVF_TRANS(vfmul_vf, opfvf_check) 2235GEN_OPFVF_TRANS(vfdiv_vf, opfvf_check) 2236GEN_OPFVF_TRANS(vfrdiv_vf, opfvf_check) 2237 2238/* Vector Widening Floating-Point Multiply */ 2239GEN_OPFVV_WIDEN_TRANS(vfwmul_vv, opfvv_widen_check) 2240GEN_OPFVF_WIDEN_TRANS(vfwmul_vf) 2241 2242/* Vector Single-Width Floating-Point Fused Multiply-Add Instructions */ 2243GEN_OPFVV_TRANS(vfmacc_vv, opfvv_check) 2244GEN_OPFVV_TRANS(vfnmacc_vv, opfvv_check) 2245GEN_OPFVV_TRANS(vfmsac_vv, opfvv_check) 2246GEN_OPFVV_TRANS(vfnmsac_vv, opfvv_check) 2247GEN_OPFVV_TRANS(vfmadd_vv, opfvv_check) 2248GEN_OPFVV_TRANS(vfnmadd_vv, opfvv_check) 2249GEN_OPFVV_TRANS(vfmsub_vv, opfvv_check) 2250GEN_OPFVV_TRANS(vfnmsub_vv, opfvv_check) 2251GEN_OPFVF_TRANS(vfmacc_vf, opfvf_check) 2252GEN_OPFVF_TRANS(vfnmacc_vf, opfvf_check) 2253GEN_OPFVF_TRANS(vfmsac_vf, opfvf_check) 2254GEN_OPFVF_TRANS(vfnmsac_vf, opfvf_check) 2255GEN_OPFVF_TRANS(vfmadd_vf, opfvf_check) 2256GEN_OPFVF_TRANS(vfnmadd_vf, opfvf_check) 2257GEN_OPFVF_TRANS(vfmsub_vf, opfvf_check) 2258GEN_OPFVF_TRANS(vfnmsub_vf, opfvf_check) 2259 2260/* Vector Widening Floating-Point Fused Multiply-Add Instructions */ 2261GEN_OPFVV_WIDEN_TRANS(vfwmacc_vv, opfvv_widen_check) 2262GEN_OPFVV_WIDEN_TRANS(vfwnmacc_vv, opfvv_widen_check) 2263GEN_OPFVV_WIDEN_TRANS(vfwmsac_vv, opfvv_widen_check) 2264GEN_OPFVV_WIDEN_TRANS(vfwnmsac_vv, opfvv_widen_check) 2265GEN_OPFVF_WIDEN_TRANS(vfwmacc_vf) 2266GEN_OPFVF_WIDEN_TRANS(vfwnmacc_vf) 2267GEN_OPFVF_WIDEN_TRANS(vfwmsac_vf) 2268GEN_OPFVF_WIDEN_TRANS(vfwnmsac_vf) 2269 2270/* Vector Floating-Point Square-Root Instruction */ 2271 2272/* 2273 * If the current SEW does not correspond to a supported IEEE floating-point 2274 * type, an illegal instruction exception is raised 2275 */ 2276static bool opfv_check(DisasContext *s, arg_rmr *a) 2277{ 2278 return require_rvv(s) && 2279 require_rvf(s) && 2280 vext_check_isa_ill(s) && 2281 /* OPFV instructions ignore vs1 check */ 2282 vext_check_ss(s, a->rd, a->rs2, a->vm); 2283} 2284 2285#define GEN_OPFV_TRANS(NAME, CHECK) \ 2286static bool trans_##NAME(DisasContext *s, arg_rmr *a) \ 2287{ \ 2288 if (CHECK(s, a)) { \ 2289 uint32_t data = 0; \ 2290 static gen_helper_gvec_3_ptr * const fns[3] = { \ 2291 gen_helper_##NAME##_h, \ 2292 gen_helper_##NAME##_w, \ 2293 gen_helper_##NAME##_d, \ 2294 }; \ 2295 TCGLabel *over = gen_new_label(); \ 2296 gen_set_rm(s, 7); \ 2297 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); \ 2298 \ 2299 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 2300 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 2301 tcg_gen_gvec_3_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), \ 2302 vreg_ofs(s, a->rs2), cpu_env, \ 2303 s->vlen / 8, s->vlen / 8, data, \ 2304 fns[s->sew - 1]); \ 2305 mark_vs_dirty(s); \ 2306 gen_set_label(over); \ 2307 return true; \ 2308 } \ 2309 return false; \ 2310} 2311 2312GEN_OPFV_TRANS(vfsqrt_v, opfv_check) 2313 2314/* Vector Floating-Point MIN/MAX Instructions */ 2315GEN_OPFVV_TRANS(vfmin_vv, opfvv_check) 2316GEN_OPFVV_TRANS(vfmax_vv, opfvv_check) 2317GEN_OPFVF_TRANS(vfmin_vf, opfvf_check) 2318GEN_OPFVF_TRANS(vfmax_vf, opfvf_check) 2319 2320/* Vector Floating-Point Sign-Injection Instructions */ 2321GEN_OPFVV_TRANS(vfsgnj_vv, opfvv_check) 2322GEN_OPFVV_TRANS(vfsgnjn_vv, opfvv_check) 2323GEN_OPFVV_TRANS(vfsgnjx_vv, opfvv_check) 2324GEN_OPFVF_TRANS(vfsgnj_vf, opfvf_check) 2325GEN_OPFVF_TRANS(vfsgnjn_vf, opfvf_check) 2326GEN_OPFVF_TRANS(vfsgnjx_vf, opfvf_check) 2327 2328/* Vector Floating-Point Compare Instructions */ 2329static bool opfvv_cmp_check(DisasContext *s, arg_rmrr *a) 2330{ 2331 return require_rvv(s) && 2332 require_rvf(s) && 2333 vext_check_isa_ill(s) && 2334 vext_check_mss(s, a->rd, a->rs1, a->rs2); 2335} 2336 2337GEN_OPFVV_TRANS(vmfeq_vv, opfvv_cmp_check) 2338GEN_OPFVV_TRANS(vmfne_vv, opfvv_cmp_check) 2339GEN_OPFVV_TRANS(vmflt_vv, opfvv_cmp_check) 2340GEN_OPFVV_TRANS(vmfle_vv, opfvv_cmp_check) 2341GEN_OPFVV_TRANS(vmford_vv, opfvv_cmp_check) 2342 2343static bool opfvf_cmp_check(DisasContext *s, arg_rmrr *a) 2344{ 2345 return require_rvv(s) && 2346 require_rvf(s) && 2347 vext_check_isa_ill(s) && 2348 vext_check_ms(s, a->rd, a->rs2); 2349} 2350 2351GEN_OPFVF_TRANS(vmfeq_vf, opfvf_cmp_check) 2352GEN_OPFVF_TRANS(vmfne_vf, opfvf_cmp_check) 2353GEN_OPFVF_TRANS(vmflt_vf, opfvf_cmp_check) 2354GEN_OPFVF_TRANS(vmfle_vf, opfvf_cmp_check) 2355GEN_OPFVF_TRANS(vmfgt_vf, opfvf_cmp_check) 2356GEN_OPFVF_TRANS(vmfge_vf, opfvf_cmp_check) 2357GEN_OPFVF_TRANS(vmford_vf, opfvf_cmp_check) 2358 2359/* Vector Floating-Point Classify Instruction */ 2360GEN_OPFV_TRANS(vfclass_v, opfv_check) 2361 2362/* Vector Floating-Point Merge Instruction */ 2363GEN_OPFVF_TRANS(vfmerge_vfm, opfvf_check) 2364 2365static bool trans_vfmv_v_f(DisasContext *s, arg_vfmv_v_f *a) 2366{ 2367 if (require_rvv(s) && 2368 require_rvf(s) && 2369 vext_check_isa_ill(s) && 2370 require_align(a->rd, s->lmul)) { 2371 if (s->vl_eq_vlmax) { 2372 tcg_gen_gvec_dup_i64(s->sew, vreg_ofs(s, a->rd), 2373 MAXSZ(s), MAXSZ(s), cpu_fpr[a->rs1]); 2374 mark_vs_dirty(s); 2375 } else { 2376 TCGv_ptr dest; 2377 TCGv_i32 desc; 2378 uint32_t data = FIELD_DP32(0, VDATA, LMUL, s->lmul); 2379 static gen_helper_vmv_vx * const fns[3] = { 2380 gen_helper_vmv_v_x_h, 2381 gen_helper_vmv_v_x_w, 2382 gen_helper_vmv_v_x_d, 2383 }; 2384 TCGLabel *over = gen_new_label(); 2385 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 2386 2387 dest = tcg_temp_new_ptr(); 2388 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 2389 tcg_gen_addi_ptr(dest, cpu_env, vreg_ofs(s, a->rd)); 2390 fns[s->sew - 1](dest, cpu_fpr[a->rs1], cpu_env, desc); 2391 2392 tcg_temp_free_ptr(dest); 2393 mark_vs_dirty(s); 2394 gen_set_label(over); 2395 } 2396 return true; 2397 } 2398 return false; 2399} 2400 2401/* Single-Width Floating-Point/Integer Type-Convert Instructions */ 2402GEN_OPFV_TRANS(vfcvt_xu_f_v, opfv_check) 2403GEN_OPFV_TRANS(vfcvt_x_f_v, opfv_check) 2404GEN_OPFV_TRANS(vfcvt_f_xu_v, opfv_check) 2405GEN_OPFV_TRANS(vfcvt_f_x_v, opfv_check) 2406 2407/* Widening Floating-Point/Integer Type-Convert Instructions */ 2408 2409/* 2410 * If the current SEW does not correspond to a supported IEEE floating-point 2411 * type, an illegal instruction exception is raised 2412 */ 2413static bool opfv_widen_check(DisasContext *s, arg_rmr *a) 2414{ 2415 return require_rvv(s) && 2416 require_scale_rvf(s) && 2417 (s->sew != MO_8) && 2418 vext_check_isa_ill(s) && 2419 vext_check_ds(s, a->rd, a->rs2, a->vm); 2420} 2421 2422#define GEN_OPFV_WIDEN_TRANS(NAME) \ 2423static bool trans_##NAME(DisasContext *s, arg_rmr *a) \ 2424{ \ 2425 if (opfv_widen_check(s, a)) { \ 2426 uint32_t data = 0; \ 2427 static gen_helper_gvec_3_ptr * const fns[2] = { \ 2428 gen_helper_##NAME##_h, \ 2429 gen_helper_##NAME##_w, \ 2430 }; \ 2431 TCGLabel *over = gen_new_label(); \ 2432 gen_set_rm(s, 7); \ 2433 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); \ 2434 \ 2435 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 2436 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 2437 tcg_gen_gvec_3_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), \ 2438 vreg_ofs(s, a->rs2), cpu_env, \ 2439 s->vlen / 8, s->vlen / 8, data, \ 2440 fns[s->sew - 1]); \ 2441 mark_vs_dirty(s); \ 2442 gen_set_label(over); \ 2443 return true; \ 2444 } \ 2445 return false; \ 2446} 2447 2448GEN_OPFV_WIDEN_TRANS(vfwcvt_xu_f_v) 2449GEN_OPFV_WIDEN_TRANS(vfwcvt_x_f_v) 2450GEN_OPFV_WIDEN_TRANS(vfwcvt_f_xu_v) 2451GEN_OPFV_WIDEN_TRANS(vfwcvt_f_x_v) 2452GEN_OPFV_WIDEN_TRANS(vfwcvt_f_f_v) 2453 2454/* Narrowing Floating-Point/Integer Type-Convert Instructions */ 2455 2456/* 2457 * If the current SEW does not correspond to a supported IEEE floating-point 2458 * type, an illegal instruction exception is raised 2459 */ 2460static bool opfv_narrow_check(DisasContext *s, arg_rmr *a) 2461{ 2462 return require_rvv(s) && 2463 require_rvf(s) && 2464 (s->sew != MO_64) && 2465 vext_check_isa_ill(s) && 2466 /* OPFV narrowing instructions ignore vs1 check */ 2467 vext_check_sd(s, a->rd, a->rs2, a->vm); 2468} 2469 2470#define GEN_OPFV_NARROW_TRANS(NAME) \ 2471static bool trans_##NAME(DisasContext *s, arg_rmr *a) \ 2472{ \ 2473 if (opfv_narrow_check(s, a)) { \ 2474 uint32_t data = 0; \ 2475 static gen_helper_gvec_3_ptr * const fns[2] = { \ 2476 gen_helper_##NAME##_h, \ 2477 gen_helper_##NAME##_w, \ 2478 }; \ 2479 TCGLabel *over = gen_new_label(); \ 2480 gen_set_rm(s, 7); \ 2481 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); \ 2482 \ 2483 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 2484 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 2485 tcg_gen_gvec_3_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), \ 2486 vreg_ofs(s, a->rs2), cpu_env, \ 2487 s->vlen / 8, s->vlen / 8, data, \ 2488 fns[s->sew - 1]); \ 2489 mark_vs_dirty(s); \ 2490 gen_set_label(over); \ 2491 return true; \ 2492 } \ 2493 return false; \ 2494} 2495 2496GEN_OPFV_NARROW_TRANS(vfncvt_xu_f_v) 2497GEN_OPFV_NARROW_TRANS(vfncvt_x_f_v) 2498GEN_OPFV_NARROW_TRANS(vfncvt_f_xu_v) 2499GEN_OPFV_NARROW_TRANS(vfncvt_f_x_v) 2500GEN_OPFV_NARROW_TRANS(vfncvt_f_f_v) 2501 2502/* 2503 *** Vector Reduction Operations 2504 */ 2505/* Vector Single-Width Integer Reduction Instructions */ 2506static bool reduction_check(DisasContext *s, arg_rmrr *a) 2507{ 2508 return require_rvv(s) && 2509 vext_check_isa_ill(s) && 2510 vext_check_reduction(s, a->rs2); 2511} 2512 2513GEN_OPIVV_TRANS(vredsum_vs, reduction_check) 2514GEN_OPIVV_TRANS(vredmaxu_vs, reduction_check) 2515GEN_OPIVV_TRANS(vredmax_vs, reduction_check) 2516GEN_OPIVV_TRANS(vredminu_vs, reduction_check) 2517GEN_OPIVV_TRANS(vredmin_vs, reduction_check) 2518GEN_OPIVV_TRANS(vredand_vs, reduction_check) 2519GEN_OPIVV_TRANS(vredor_vs, reduction_check) 2520GEN_OPIVV_TRANS(vredxor_vs, reduction_check) 2521 2522/* Vector Widening Integer Reduction Instructions */ 2523static bool reduction_widen_check(DisasContext *s, arg_rmrr *a) 2524{ 2525 return reduction_check(s, a) && (s->sew < MO_64); 2526} 2527 2528GEN_OPIVV_WIDEN_TRANS(vwredsum_vs, reduction_widen_check) 2529GEN_OPIVV_WIDEN_TRANS(vwredsumu_vs, reduction_widen_check) 2530 2531/* Vector Single-Width Floating-Point Reduction Instructions */ 2532GEN_OPFVV_TRANS(vfredsum_vs, reduction_check) 2533GEN_OPFVV_TRANS(vfredmax_vs, reduction_check) 2534GEN_OPFVV_TRANS(vfredmin_vs, reduction_check) 2535 2536/* Vector Widening Floating-Point Reduction Instructions */ 2537GEN_OPFVV_WIDEN_TRANS(vfwredsum_vs, reduction_check) 2538 2539/* 2540 *** Vector Mask Operations 2541 */ 2542 2543/* Vector Mask-Register Logical Instructions */ 2544#define GEN_MM_TRANS(NAME) \ 2545static bool trans_##NAME(DisasContext *s, arg_r *a) \ 2546{ \ 2547 if (vext_check_isa_ill(s)) { \ 2548 uint32_t data = 0; \ 2549 gen_helper_gvec_4_ptr *fn = gen_helper_##NAME; \ 2550 TCGLabel *over = gen_new_label(); \ 2551 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); \ 2552 \ 2553 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 2554 tcg_gen_gvec_4_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), \ 2555 vreg_ofs(s, a->rs1), \ 2556 vreg_ofs(s, a->rs2), cpu_env, \ 2557 s->vlen / 8, s->vlen / 8, data, fn); \ 2558 mark_vs_dirty(s); \ 2559 gen_set_label(over); \ 2560 return true; \ 2561 } \ 2562 return false; \ 2563} 2564 2565GEN_MM_TRANS(vmand_mm) 2566GEN_MM_TRANS(vmnand_mm) 2567GEN_MM_TRANS(vmandnot_mm) 2568GEN_MM_TRANS(vmxor_mm) 2569GEN_MM_TRANS(vmor_mm) 2570GEN_MM_TRANS(vmnor_mm) 2571GEN_MM_TRANS(vmornot_mm) 2572GEN_MM_TRANS(vmxnor_mm) 2573 2574/* Vector mask population count vmpopc */ 2575static bool trans_vmpopc_m(DisasContext *s, arg_rmr *a) 2576{ 2577 if (require_rvv(s) && 2578 vext_check_isa_ill(s)) { 2579 TCGv_ptr src2, mask; 2580 TCGv dst; 2581 TCGv_i32 desc; 2582 uint32_t data = 0; 2583 data = FIELD_DP32(data, VDATA, VM, a->vm); 2584 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 2585 2586 mask = tcg_temp_new_ptr(); 2587 src2 = tcg_temp_new_ptr(); 2588 dst = dest_gpr(s, a->rd); 2589 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 2590 2591 tcg_gen_addi_ptr(src2, cpu_env, vreg_ofs(s, a->rs2)); 2592 tcg_gen_addi_ptr(mask, cpu_env, vreg_ofs(s, 0)); 2593 2594 gen_helper_vmpopc_m(dst, mask, src2, cpu_env, desc); 2595 gen_set_gpr(s, a->rd, dst); 2596 2597 tcg_temp_free_ptr(mask); 2598 tcg_temp_free_ptr(src2); 2599 return true; 2600 } 2601 return false; 2602} 2603 2604/* vmfirst find-first-set mask bit */ 2605static bool trans_vmfirst_m(DisasContext *s, arg_rmr *a) 2606{ 2607 if (require_rvv(s) && 2608 vext_check_isa_ill(s)) { 2609 TCGv_ptr src2, mask; 2610 TCGv dst; 2611 TCGv_i32 desc; 2612 uint32_t data = 0; 2613 data = FIELD_DP32(data, VDATA, VM, a->vm); 2614 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 2615 2616 mask = tcg_temp_new_ptr(); 2617 src2 = tcg_temp_new_ptr(); 2618 dst = dest_gpr(s, a->rd); 2619 desc = tcg_constant_i32(simd_desc(s->vlen / 8, s->vlen / 8, data)); 2620 2621 tcg_gen_addi_ptr(src2, cpu_env, vreg_ofs(s, a->rs2)); 2622 tcg_gen_addi_ptr(mask, cpu_env, vreg_ofs(s, 0)); 2623 2624 gen_helper_vmfirst_m(dst, mask, src2, cpu_env, desc); 2625 gen_set_gpr(s, a->rd, dst); 2626 2627 tcg_temp_free_ptr(mask); 2628 tcg_temp_free_ptr(src2); 2629 return true; 2630 } 2631 return false; 2632} 2633 2634/* vmsbf.m set-before-first mask bit */ 2635/* vmsif.m set-includ-first mask bit */ 2636/* vmsof.m set-only-first mask bit */ 2637#define GEN_M_TRANS(NAME) \ 2638static bool trans_##NAME(DisasContext *s, arg_rmr *a) \ 2639{ \ 2640 if (vext_check_isa_ill(s)) { \ 2641 uint32_t data = 0; \ 2642 gen_helper_gvec_3_ptr *fn = gen_helper_##NAME; \ 2643 TCGLabel *over = gen_new_label(); \ 2644 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); \ 2645 \ 2646 data = FIELD_DP32(data, VDATA, VM, a->vm); \ 2647 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ 2648 tcg_gen_gvec_3_ptr(vreg_ofs(s, a->rd), \ 2649 vreg_ofs(s, 0), vreg_ofs(s, a->rs2), \ 2650 cpu_env, s->vlen / 8, s->vlen / 8, \ 2651 data, fn); \ 2652 mark_vs_dirty(s); \ 2653 gen_set_label(over); \ 2654 return true; \ 2655 } \ 2656 return false; \ 2657} 2658 2659GEN_M_TRANS(vmsbf_m) 2660GEN_M_TRANS(vmsif_m) 2661GEN_M_TRANS(vmsof_m) 2662 2663/* Vector Iota Instruction */ 2664static bool trans_viota_m(DisasContext *s, arg_viota_m *a) 2665{ 2666 if (require_rvv(s) && 2667 vext_check_isa_ill(s) && 2668 require_noover(a->rd, s->lmul, a->rs2, 0) && 2669 require_vm(a->vm, a->rd) && 2670 require_align(a->rd, s->lmul)) { 2671 uint32_t data = 0; 2672 TCGLabel *over = gen_new_label(); 2673 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 2674 2675 data = FIELD_DP32(data, VDATA, VM, a->vm); 2676 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 2677 static gen_helper_gvec_3_ptr * const fns[4] = { 2678 gen_helper_viota_m_b, gen_helper_viota_m_h, 2679 gen_helper_viota_m_w, gen_helper_viota_m_d, 2680 }; 2681 tcg_gen_gvec_3_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), 2682 vreg_ofs(s, a->rs2), cpu_env, 2683 s->vlen / 8, s->vlen / 8, data, fns[s->sew]); 2684 mark_vs_dirty(s); 2685 gen_set_label(over); 2686 return true; 2687 } 2688 return false; 2689} 2690 2691/* Vector Element Index Instruction */ 2692static bool trans_vid_v(DisasContext *s, arg_vid_v *a) 2693{ 2694 if (require_rvv(s) && 2695 vext_check_isa_ill(s) && 2696 require_align(a->rd, s->lmul) && 2697 require_vm(a->vm, a->rd)) { 2698 uint32_t data = 0; 2699 TCGLabel *over = gen_new_label(); 2700 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 2701 2702 data = FIELD_DP32(data, VDATA, VM, a->vm); 2703 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 2704 static gen_helper_gvec_2_ptr * const fns[4] = { 2705 gen_helper_vid_v_b, gen_helper_vid_v_h, 2706 gen_helper_vid_v_w, gen_helper_vid_v_d, 2707 }; 2708 tcg_gen_gvec_2_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), 2709 cpu_env, s->vlen / 8, s->vlen / 8, 2710 data, fns[s->sew]); 2711 mark_vs_dirty(s); 2712 gen_set_label(over); 2713 return true; 2714 } 2715 return false; 2716} 2717 2718/* 2719 *** Vector Permutation Instructions 2720 */ 2721 2722/* Integer Extract Instruction */ 2723 2724static void load_element(TCGv_i64 dest, TCGv_ptr base, 2725 int ofs, int sew) 2726{ 2727 switch (sew) { 2728 case MO_8: 2729 tcg_gen_ld8u_i64(dest, base, ofs); 2730 break; 2731 case MO_16: 2732 tcg_gen_ld16u_i64(dest, base, ofs); 2733 break; 2734 case MO_32: 2735 tcg_gen_ld32u_i64(dest, base, ofs); 2736 break; 2737 case MO_64: 2738 tcg_gen_ld_i64(dest, base, ofs); 2739 break; 2740 default: 2741 g_assert_not_reached(); 2742 break; 2743 } 2744} 2745 2746/* offset of the idx element with base regsiter r */ 2747static uint32_t endian_ofs(DisasContext *s, int r, int idx) 2748{ 2749#ifdef HOST_WORDS_BIGENDIAN 2750 return vreg_ofs(s, r) + ((idx ^ (7 >> s->sew)) << s->sew); 2751#else 2752 return vreg_ofs(s, r) + (idx << s->sew); 2753#endif 2754} 2755 2756/* adjust the index according to the endian */ 2757static void endian_adjust(TCGv_i32 ofs, int sew) 2758{ 2759#ifdef HOST_WORDS_BIGENDIAN 2760 tcg_gen_xori_i32(ofs, ofs, 7 >> sew); 2761#endif 2762} 2763 2764/* Load idx >= VLMAX ? 0 : vreg[idx] */ 2765static void vec_element_loadx(DisasContext *s, TCGv_i64 dest, 2766 int vreg, TCGv idx, int vlmax) 2767{ 2768 TCGv_i32 ofs = tcg_temp_new_i32(); 2769 TCGv_ptr base = tcg_temp_new_ptr(); 2770 TCGv_i64 t_idx = tcg_temp_new_i64(); 2771 TCGv_i64 t_vlmax, t_zero; 2772 2773 /* 2774 * Mask the index to the length so that we do 2775 * not produce an out-of-range load. 2776 */ 2777 tcg_gen_trunc_tl_i32(ofs, idx); 2778 tcg_gen_andi_i32(ofs, ofs, vlmax - 1); 2779 2780 /* Convert the index to an offset. */ 2781 endian_adjust(ofs, s->sew); 2782 tcg_gen_shli_i32(ofs, ofs, s->sew); 2783 2784 /* Convert the index to a pointer. */ 2785 tcg_gen_ext_i32_ptr(base, ofs); 2786 tcg_gen_add_ptr(base, base, cpu_env); 2787 2788 /* Perform the load. */ 2789 load_element(dest, base, 2790 vreg_ofs(s, vreg), s->sew); 2791 tcg_temp_free_ptr(base); 2792 tcg_temp_free_i32(ofs); 2793 2794 /* Flush out-of-range indexing to zero. */ 2795 t_vlmax = tcg_constant_i64(vlmax); 2796 t_zero = tcg_constant_i64(0); 2797 tcg_gen_extu_tl_i64(t_idx, idx); 2798 2799 tcg_gen_movcond_i64(TCG_COND_LTU, dest, t_idx, 2800 t_vlmax, dest, t_zero); 2801 2802 tcg_temp_free_i64(t_idx); 2803} 2804 2805static void vec_element_loadi(DisasContext *s, TCGv_i64 dest, 2806 int vreg, int idx) 2807{ 2808 load_element(dest, cpu_env, endian_ofs(s, vreg, idx), s->sew); 2809} 2810 2811static bool trans_vext_x_v(DisasContext *s, arg_r *a) 2812{ 2813 TCGv_i64 tmp = tcg_temp_new_i64(); 2814 TCGv dest = dest_gpr(s, a->rd); 2815 2816 if (a->rs1 == 0) { 2817 /* Special case vmv.x.s rd, vs2. */ 2818 vec_element_loadi(s, tmp, a->rs2, 0); 2819 } else { 2820 /* This instruction ignores LMUL and vector register groups */ 2821 int vlmax = s->vlen >> (3 + s->sew); 2822 vec_element_loadx(s, tmp, a->rs2, cpu_gpr[a->rs1], vlmax); 2823 } 2824 2825 tcg_gen_trunc_i64_tl(dest, tmp); 2826 gen_set_gpr(s, a->rd, dest); 2827 2828 tcg_temp_free_i64(tmp); 2829 return true; 2830} 2831 2832/* Integer Scalar Move Instruction */ 2833 2834static void store_element(TCGv_i64 val, TCGv_ptr base, 2835 int ofs, int sew) 2836{ 2837 switch (sew) { 2838 case MO_8: 2839 tcg_gen_st8_i64(val, base, ofs); 2840 break; 2841 case MO_16: 2842 tcg_gen_st16_i64(val, base, ofs); 2843 break; 2844 case MO_32: 2845 tcg_gen_st32_i64(val, base, ofs); 2846 break; 2847 case MO_64: 2848 tcg_gen_st_i64(val, base, ofs); 2849 break; 2850 default: 2851 g_assert_not_reached(); 2852 break; 2853 } 2854} 2855 2856/* 2857 * Store vreg[idx] = val. 2858 * The index must be in range of VLMAX. 2859 */ 2860static void vec_element_storei(DisasContext *s, int vreg, 2861 int idx, TCGv_i64 val) 2862{ 2863 store_element(val, cpu_env, endian_ofs(s, vreg, idx), s->sew); 2864} 2865 2866/* vmv.s.x vd, rs1 # vd[0] = rs1 */ 2867static bool trans_vmv_s_x(DisasContext *s, arg_vmv_s_x *a) 2868{ 2869 if (vext_check_isa_ill(s)) { 2870 /* This instruction ignores LMUL and vector register groups */ 2871 int maxsz = s->vlen >> 3; 2872 TCGv_i64 t1; 2873 TCGLabel *over = gen_new_label(); 2874 2875 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 2876 tcg_gen_gvec_dup_imm(SEW64, vreg_ofs(s, a->rd), maxsz, maxsz, 0); 2877 if (a->rs1 == 0) { 2878 goto done; 2879 } 2880 2881 t1 = tcg_temp_new_i64(); 2882 tcg_gen_extu_tl_i64(t1, cpu_gpr[a->rs1]); 2883 vec_element_storei(s, a->rd, 0, t1); 2884 tcg_temp_free_i64(t1); 2885 mark_vs_dirty(s); 2886 done: 2887 gen_set_label(over); 2888 return true; 2889 } 2890 return false; 2891} 2892 2893/* Floating-Point Scalar Move Instructions */ 2894static bool trans_vfmv_f_s(DisasContext *s, arg_vfmv_f_s *a) 2895{ 2896 if (!s->vill && has_ext(s, RVF) && 2897 (s->mstatus_fs != 0) && (s->sew != 0)) { 2898 unsigned int len = 8 << s->sew; 2899 2900 vec_element_loadi(s, cpu_fpr[a->rd], a->rs2, 0); 2901 if (len < 64) { 2902 tcg_gen_ori_i64(cpu_fpr[a->rd], cpu_fpr[a->rd], 2903 MAKE_64BIT_MASK(len, 64 - len)); 2904 } 2905 2906 mark_fs_dirty(s); 2907 return true; 2908 } 2909 return false; 2910} 2911 2912/* vfmv.s.f vd, rs1 # vd[0] = rs1 (vs2=0) */ 2913static bool trans_vfmv_s_f(DisasContext *s, arg_vfmv_s_f *a) 2914{ 2915 if (!s->vill && has_ext(s, RVF) && (s->sew != 0)) { 2916 TCGv_i64 t1; 2917 /* The instructions ignore LMUL and vector register group. */ 2918 uint32_t vlmax = s->vlen >> 3; 2919 2920 /* if vl == 0, skip vector register write back */ 2921 TCGLabel *over = gen_new_label(); 2922 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 2923 2924 /* zeroed all elements */ 2925 tcg_gen_gvec_dup_imm(SEW64, vreg_ofs(s, a->rd), vlmax, vlmax, 0); 2926 2927 /* NaN-box f[rs1] as necessary for SEW */ 2928 t1 = tcg_temp_new_i64(); 2929 if (s->sew == MO_64 && !has_ext(s, RVD)) { 2930 tcg_gen_ori_i64(t1, cpu_fpr[a->rs1], MAKE_64BIT_MASK(32, 32)); 2931 } else { 2932 tcg_gen_mov_i64(t1, cpu_fpr[a->rs1]); 2933 } 2934 vec_element_storei(s, a->rd, 0, t1); 2935 tcg_temp_free_i64(t1); 2936 mark_vs_dirty(s); 2937 gen_set_label(over); 2938 return true; 2939 } 2940 return false; 2941} 2942 2943/* Vector Slide Instructions */ 2944static bool slideup_check(DisasContext *s, arg_rmrr *a) 2945{ 2946 return require_rvv(s) && 2947 vext_check_isa_ill(s) && 2948 vext_check_slide(s, a->rd, a->rs2, a->vm, true); 2949} 2950 2951GEN_OPIVX_TRANS(vslideup_vx, slideup_check) 2952GEN_OPIVX_TRANS(vslide1up_vx, slideup_check) 2953GEN_OPIVI_TRANS(vslideup_vi, IMM_ZX, vslideup_vx, slideup_check) 2954 2955static bool slidedown_check(DisasContext *s, arg_rmrr *a) 2956{ 2957 return require_rvv(s) && 2958 vext_check_isa_ill(s) && 2959 vext_check_slide(s, a->rd, a->rs2, a->vm, false); 2960} 2961 2962GEN_OPIVX_TRANS(vslidedown_vx, slidedown_check) 2963GEN_OPIVX_TRANS(vslide1down_vx, slidedown_check) 2964GEN_OPIVI_TRANS(vslidedown_vi, IMM_ZX, vslidedown_vx, slidedown_check) 2965 2966/* Vector Register Gather Instruction */ 2967static bool vrgather_vv_check(DisasContext *s, arg_rmrr *a) 2968{ 2969 return require_rvv(s) && 2970 vext_check_isa_ill(s) && 2971 require_align(a->rd, s->lmul) && 2972 require_align(a->rs1, s->lmul) && 2973 require_align(a->rs2, s->lmul) && 2974 (a->rd != a->rs2 && a->rd != a->rs1) && 2975 require_vm(a->vm, a->rd); 2976} 2977 2978GEN_OPIVV_TRANS(vrgather_vv, vrgather_vv_check) 2979 2980static bool vrgather_vx_check(DisasContext *s, arg_rmrr *a) 2981{ 2982 return require_rvv(s) && 2983 vext_check_isa_ill(s) && 2984 require_align(a->rd, s->lmul) && 2985 require_align(a->rs2, s->lmul) && 2986 (a->rd != a->rs2) && 2987 require_vm(a->vm, a->rd); 2988} 2989 2990/* vrgather.vx vd, vs2, rs1, vm # vd[i] = (x[rs1] >= VLMAX) ? 0 : vs2[rs1] */ 2991static bool trans_vrgather_vx(DisasContext *s, arg_rmrr *a) 2992{ 2993 if (!vrgather_vx_check(s, a)) { 2994 return false; 2995 } 2996 2997 if (a->vm && s->vl_eq_vlmax) { 2998 int vlmax = s->vlen; 2999 TCGv_i64 dest = tcg_temp_new_i64(); 3000 3001 if (a->rs1 == 0) { 3002 vec_element_loadi(s, dest, a->rs2, 0); 3003 } else { 3004 vec_element_loadx(s, dest, a->rs2, cpu_gpr[a->rs1], vlmax); 3005 } 3006 3007 tcg_gen_gvec_dup_i64(s->sew, vreg_ofs(s, a->rd), 3008 MAXSZ(s), MAXSZ(s), dest); 3009 tcg_temp_free_i64(dest); 3010 mark_vs_dirty(s); 3011 } else { 3012 static gen_helper_opivx * const fns[4] = { 3013 gen_helper_vrgather_vx_b, gen_helper_vrgather_vx_h, 3014 gen_helper_vrgather_vx_w, gen_helper_vrgather_vx_d 3015 }; 3016 return opivx_trans(a->rd, a->rs1, a->rs2, a->vm, fns[s->sew], s); 3017 } 3018 return true; 3019} 3020 3021/* vrgather.vi vd, vs2, imm, vm # vd[i] = (imm >= VLMAX) ? 0 : vs2[imm] */ 3022static bool trans_vrgather_vi(DisasContext *s, arg_rmrr *a) 3023{ 3024 if (!vrgather_vx_check(s, a)) { 3025 return false; 3026 } 3027 3028 if (a->vm && s->vl_eq_vlmax) { 3029 if (a->rs1 >= s->vlen) { 3030 tcg_gen_gvec_dup_imm(SEW64, vreg_ofs(s, a->rd), 3031 MAXSZ(s), MAXSZ(s), 0); 3032 } else { 3033 tcg_gen_gvec_dup_mem(s->sew, vreg_ofs(s, a->rd), 3034 endian_ofs(s, a->rs2, a->rs1), 3035 MAXSZ(s), MAXSZ(s)); 3036 } 3037 mark_vs_dirty(s); 3038 } else { 3039 static gen_helper_opivx * const fns[4] = { 3040 gen_helper_vrgather_vx_b, gen_helper_vrgather_vx_h, 3041 gen_helper_vrgather_vx_w, gen_helper_vrgather_vx_d 3042 }; 3043 return opivi_trans(a->rd, a->rs1, a->rs2, a->vm, fns[s->sew], 3044 s, IMM_ZX); 3045 } 3046 return true; 3047} 3048 3049/* 3050 * Vector Compress Instruction 3051 * 3052 * The destination vector register group cannot overlap the 3053 * source vector register group or the source mask register. 3054 */ 3055static bool vcompress_vm_check(DisasContext *s, arg_r *a) 3056{ 3057 return require_rvv(s) && 3058 vext_check_isa_ill(s) && 3059 require_align(a->rd, s->lmul) && 3060 require_align(a->rs2, s->lmul) && 3061 (a->rd != a->rs2) && 3062 !is_overlapped(a->rd, 1 << MAX(s->lmul, 0), a->rs1, 1); 3063} 3064 3065static bool trans_vcompress_vm(DisasContext *s, arg_r *a) 3066{ 3067 if (vcompress_vm_check(s, a)) { 3068 uint32_t data = 0; 3069 static gen_helper_gvec_4_ptr * const fns[4] = { 3070 gen_helper_vcompress_vm_b, gen_helper_vcompress_vm_h, 3071 gen_helper_vcompress_vm_w, gen_helper_vcompress_vm_d, 3072 }; 3073 TCGLabel *over = gen_new_label(); 3074 tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_vl, 0, over); 3075 3076 data = FIELD_DP32(data, VDATA, LMUL, s->lmul); 3077 tcg_gen_gvec_4_ptr(vreg_ofs(s, a->rd), vreg_ofs(s, 0), 3078 vreg_ofs(s, a->rs1), vreg_ofs(s, a->rs2), 3079 cpu_env, s->vlen / 8, s->vlen / 8, data, 3080 fns[s->sew]); 3081 mark_vs_dirty(s); 3082 gen_set_label(over); 3083 return true; 3084 } 3085 return false; 3086} 3087