1 /* 2 * Copyright(c) 2019-2022 Qualcomm Innovation Center, Inc. All Rights Reserved. 3 * 4 * This program is free software; you can redistribute it and/or modify 5 * it under the terms of the GNU General Public License as published by 6 * the Free Software Foundation; either version 2 of the License, or 7 * (at your option) any later version. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, see <http://www.gnu.org/licenses/>. 16 */ 17 18 #include "qemu/osdep.h" 19 #include "cpu.h" 20 #include "internal.h" 21 #include "tcg/tcg-op.h" 22 #include "tcg/tcg-op-gvec.h" 23 #include "insn.h" 24 #include "opcodes.h" 25 #include "translate.h" 26 #define QEMU_GENERATE /* Used internally by macros.h */ 27 #include "macros.h" 28 #include "mmvec/macros.h" 29 #undef QEMU_GENERATE 30 #include "gen_tcg.h" 31 #include "gen_tcg_hvx.h" 32 33 static inline void gen_log_predicated_reg_write(int rnum, TCGv val, int slot) 34 { 35 TCGv zero = tcg_constant_tl(0); 36 TCGv slot_mask = tcg_temp_new(); 37 38 tcg_gen_andi_tl(slot_mask, hex_slot_cancelled, 1 << slot); 39 tcg_gen_movcond_tl(TCG_COND_EQ, hex_new_value[rnum], slot_mask, zero, 40 val, hex_new_value[rnum]); 41 if (HEX_DEBUG) { 42 /* 43 * Do this so HELPER(debug_commit_end) will know 44 * 45 * Note that slot_mask indicates the value is not written 46 * (i.e., slot was cancelled), so we create a true/false value before 47 * or'ing with hex_reg_written[rnum]. 48 */ 49 tcg_gen_setcond_tl(TCG_COND_EQ, slot_mask, slot_mask, zero); 50 tcg_gen_or_tl(hex_reg_written[rnum], hex_reg_written[rnum], slot_mask); 51 } 52 53 tcg_temp_free(slot_mask); 54 } 55 56 static inline void gen_log_reg_write(int rnum, TCGv val) 57 { 58 tcg_gen_mov_tl(hex_new_value[rnum], val); 59 if (HEX_DEBUG) { 60 /* Do this so HELPER(debug_commit_end) will know */ 61 tcg_gen_movi_tl(hex_reg_written[rnum], 1); 62 } 63 } 64 65 static void gen_log_predicated_reg_write_pair(int rnum, TCGv_i64 val, int slot) 66 { 67 TCGv val32 = tcg_temp_new(); 68 TCGv zero = tcg_constant_tl(0); 69 TCGv slot_mask = tcg_temp_new(); 70 71 tcg_gen_andi_tl(slot_mask, hex_slot_cancelled, 1 << slot); 72 /* Low word */ 73 tcg_gen_extrl_i64_i32(val32, val); 74 tcg_gen_movcond_tl(TCG_COND_EQ, hex_new_value[rnum], 75 slot_mask, zero, 76 val32, hex_new_value[rnum]); 77 /* High word */ 78 tcg_gen_extrh_i64_i32(val32, val); 79 tcg_gen_movcond_tl(TCG_COND_EQ, hex_new_value[rnum + 1], 80 slot_mask, zero, 81 val32, hex_new_value[rnum + 1]); 82 if (HEX_DEBUG) { 83 /* 84 * Do this so HELPER(debug_commit_end) will know 85 * 86 * Note that slot_mask indicates the value is not written 87 * (i.e., slot was cancelled), so we create a true/false value before 88 * or'ing with hex_reg_written[rnum]. 89 */ 90 tcg_gen_setcond_tl(TCG_COND_EQ, slot_mask, slot_mask, zero); 91 tcg_gen_or_tl(hex_reg_written[rnum], hex_reg_written[rnum], slot_mask); 92 tcg_gen_or_tl(hex_reg_written[rnum + 1], hex_reg_written[rnum + 1], 93 slot_mask); 94 } 95 96 tcg_temp_free(val32); 97 tcg_temp_free(slot_mask); 98 } 99 100 static void gen_log_reg_write_pair(int rnum, TCGv_i64 val) 101 { 102 /* Low word */ 103 tcg_gen_extrl_i64_i32(hex_new_value[rnum], val); 104 if (HEX_DEBUG) { 105 /* Do this so HELPER(debug_commit_end) will know */ 106 tcg_gen_movi_tl(hex_reg_written[rnum], 1); 107 } 108 109 /* High word */ 110 tcg_gen_extrh_i64_i32(hex_new_value[rnum + 1], val); 111 if (HEX_DEBUG) { 112 /* Do this so HELPER(debug_commit_end) will know */ 113 tcg_gen_movi_tl(hex_reg_written[rnum + 1], 1); 114 } 115 } 116 117 static inline void gen_log_pred_write(DisasContext *ctx, int pnum, TCGv val) 118 { 119 TCGv base_val = tcg_temp_new(); 120 121 tcg_gen_andi_tl(base_val, val, 0xff); 122 123 /* 124 * Section 6.1.3 of the Hexagon V67 Programmer's Reference Manual 125 * 126 * Multiple writes to the same preg are and'ed together 127 * If this is the first predicate write in the packet, do a 128 * straight assignment. Otherwise, do an and. 129 */ 130 if (!test_bit(pnum, ctx->pregs_written)) { 131 tcg_gen_mov_tl(hex_new_pred_value[pnum], base_val); 132 } else { 133 tcg_gen_and_tl(hex_new_pred_value[pnum], 134 hex_new_pred_value[pnum], base_val); 135 } 136 tcg_gen_ori_tl(hex_pred_written, hex_pred_written, 1 << pnum); 137 138 tcg_temp_free(base_val); 139 } 140 141 static inline void gen_read_p3_0(TCGv control_reg) 142 { 143 tcg_gen_movi_tl(control_reg, 0); 144 for (int i = 0; i < NUM_PREGS; i++) { 145 tcg_gen_deposit_tl(control_reg, control_reg, hex_pred[i], i * 8, 8); 146 } 147 } 148 149 /* 150 * Certain control registers require special handling on read 151 * HEX_REG_P3_0 aliased to the predicate registers 152 * -> concat the 4 predicate registers together 153 * HEX_REG_PC actual value stored in DisasContext 154 * -> assign from ctx->base.pc_next 155 * HEX_REG_QEMU_*_CNT changes in current TB in DisasContext 156 * -> add current TB changes to existing reg value 157 */ 158 static inline void gen_read_ctrl_reg(DisasContext *ctx, const int reg_num, 159 TCGv dest) 160 { 161 if (reg_num == HEX_REG_P3_0) { 162 gen_read_p3_0(dest); 163 } else if (reg_num == HEX_REG_PC) { 164 tcg_gen_movi_tl(dest, ctx->base.pc_next); 165 } else if (reg_num == HEX_REG_QEMU_PKT_CNT) { 166 tcg_gen_addi_tl(dest, hex_gpr[HEX_REG_QEMU_PKT_CNT], 167 ctx->num_packets); 168 } else if (reg_num == HEX_REG_QEMU_INSN_CNT) { 169 tcg_gen_addi_tl(dest, hex_gpr[HEX_REG_QEMU_INSN_CNT], 170 ctx->num_insns); 171 } else if (reg_num == HEX_REG_QEMU_HVX_CNT) { 172 tcg_gen_addi_tl(dest, hex_gpr[HEX_REG_QEMU_HVX_CNT], 173 ctx->num_hvx_insns); 174 } else { 175 tcg_gen_mov_tl(dest, hex_gpr[reg_num]); 176 } 177 } 178 179 static inline void gen_read_ctrl_reg_pair(DisasContext *ctx, const int reg_num, 180 TCGv_i64 dest) 181 { 182 if (reg_num == HEX_REG_P3_0) { 183 TCGv p3_0 = tcg_temp_new(); 184 gen_read_p3_0(p3_0); 185 tcg_gen_concat_i32_i64(dest, p3_0, hex_gpr[reg_num + 1]); 186 tcg_temp_free(p3_0); 187 } else if (reg_num == HEX_REG_PC - 1) { 188 TCGv pc = tcg_constant_tl(ctx->base.pc_next); 189 tcg_gen_concat_i32_i64(dest, hex_gpr[reg_num], pc); 190 } else if (reg_num == HEX_REG_QEMU_PKT_CNT) { 191 TCGv pkt_cnt = tcg_temp_new(); 192 TCGv insn_cnt = tcg_temp_new(); 193 tcg_gen_addi_tl(pkt_cnt, hex_gpr[HEX_REG_QEMU_PKT_CNT], 194 ctx->num_packets); 195 tcg_gen_addi_tl(insn_cnt, hex_gpr[HEX_REG_QEMU_INSN_CNT], 196 ctx->num_insns); 197 tcg_gen_concat_i32_i64(dest, pkt_cnt, insn_cnt); 198 tcg_temp_free(pkt_cnt); 199 tcg_temp_free(insn_cnt); 200 } else if (reg_num == HEX_REG_QEMU_HVX_CNT) { 201 TCGv hvx_cnt = tcg_temp_new(); 202 tcg_gen_addi_tl(hvx_cnt, hex_gpr[HEX_REG_QEMU_HVX_CNT], 203 ctx->num_hvx_insns); 204 tcg_gen_concat_i32_i64(dest, hvx_cnt, hex_gpr[reg_num + 1]); 205 tcg_temp_free(hvx_cnt); 206 } else { 207 tcg_gen_concat_i32_i64(dest, 208 hex_gpr[reg_num], 209 hex_gpr[reg_num + 1]); 210 } 211 } 212 213 static void gen_write_p3_0(DisasContext *ctx, TCGv control_reg) 214 { 215 TCGv hex_p8 = tcg_temp_new(); 216 for (int i = 0; i < NUM_PREGS; i++) { 217 tcg_gen_extract_tl(hex_p8, control_reg, i * 8, 8); 218 gen_log_pred_write(ctx, i, hex_p8); 219 ctx_log_pred_write(ctx, i); 220 } 221 tcg_temp_free(hex_p8); 222 } 223 224 /* 225 * Certain control registers require special handling on write 226 * HEX_REG_P3_0 aliased to the predicate registers 227 * -> break the value across 4 predicate registers 228 * HEX_REG_QEMU_*_CNT changes in current TB in DisasContext 229 * -> clear the changes 230 */ 231 static inline void gen_write_ctrl_reg(DisasContext *ctx, int reg_num, 232 TCGv val) 233 { 234 if (reg_num == HEX_REG_P3_0) { 235 gen_write_p3_0(ctx, val); 236 } else { 237 gen_log_reg_write(reg_num, val); 238 ctx_log_reg_write(ctx, reg_num); 239 if (reg_num == HEX_REG_QEMU_PKT_CNT) { 240 ctx->num_packets = 0; 241 } 242 if (reg_num == HEX_REG_QEMU_INSN_CNT) { 243 ctx->num_insns = 0; 244 } 245 if (reg_num == HEX_REG_QEMU_HVX_CNT) { 246 ctx->num_hvx_insns = 0; 247 } 248 } 249 } 250 251 static inline void gen_write_ctrl_reg_pair(DisasContext *ctx, int reg_num, 252 TCGv_i64 val) 253 { 254 if (reg_num == HEX_REG_P3_0) { 255 TCGv val32 = tcg_temp_new(); 256 tcg_gen_extrl_i64_i32(val32, val); 257 gen_write_p3_0(ctx, val32); 258 tcg_gen_extrh_i64_i32(val32, val); 259 gen_log_reg_write(reg_num + 1, val32); 260 tcg_temp_free(val32); 261 ctx_log_reg_write(ctx, reg_num + 1); 262 } else { 263 gen_log_reg_write_pair(reg_num, val); 264 ctx_log_reg_write_pair(ctx, reg_num); 265 if (reg_num == HEX_REG_QEMU_PKT_CNT) { 266 ctx->num_packets = 0; 267 ctx->num_insns = 0; 268 } 269 if (reg_num == HEX_REG_QEMU_HVX_CNT) { 270 ctx->num_hvx_insns = 0; 271 } 272 } 273 } 274 275 static TCGv gen_get_byte(TCGv result, int N, TCGv src, bool sign) 276 { 277 if (sign) { 278 tcg_gen_sextract_tl(result, src, N * 8, 8); 279 } else { 280 tcg_gen_extract_tl(result, src, N * 8, 8); 281 } 282 return result; 283 } 284 285 static TCGv gen_get_byte_i64(TCGv result, int N, TCGv_i64 src, bool sign) 286 { 287 TCGv_i64 res64 = tcg_temp_new_i64(); 288 if (sign) { 289 tcg_gen_sextract_i64(res64, src, N * 8, 8); 290 } else { 291 tcg_gen_extract_i64(res64, src, N * 8, 8); 292 } 293 tcg_gen_extrl_i64_i32(result, res64); 294 tcg_temp_free_i64(res64); 295 296 return result; 297 } 298 299 static inline TCGv gen_get_half(TCGv result, int N, TCGv src, bool sign) 300 { 301 if (sign) { 302 tcg_gen_sextract_tl(result, src, N * 16, 16); 303 } else { 304 tcg_gen_extract_tl(result, src, N * 16, 16); 305 } 306 return result; 307 } 308 309 static inline void gen_set_half(int N, TCGv result, TCGv src) 310 { 311 tcg_gen_deposit_tl(result, result, src, N * 16, 16); 312 } 313 314 static inline void gen_set_half_i64(int N, TCGv_i64 result, TCGv src) 315 { 316 TCGv_i64 src64 = tcg_temp_new_i64(); 317 tcg_gen_extu_i32_i64(src64, src); 318 tcg_gen_deposit_i64(result, result, src64, N * 16, 16); 319 tcg_temp_free_i64(src64); 320 } 321 322 static void gen_set_byte_i64(int N, TCGv_i64 result, TCGv src) 323 { 324 TCGv_i64 src64 = tcg_temp_new_i64(); 325 tcg_gen_extu_i32_i64(src64, src); 326 tcg_gen_deposit_i64(result, result, src64, N * 8, 8); 327 tcg_temp_free_i64(src64); 328 } 329 330 static inline void gen_load_locked4u(TCGv dest, TCGv vaddr, int mem_index) 331 { 332 tcg_gen_qemu_ld32u(dest, vaddr, mem_index); 333 tcg_gen_mov_tl(hex_llsc_addr, vaddr); 334 tcg_gen_mov_tl(hex_llsc_val, dest); 335 } 336 337 static inline void gen_load_locked8u(TCGv_i64 dest, TCGv vaddr, int mem_index) 338 { 339 tcg_gen_qemu_ld64(dest, vaddr, mem_index); 340 tcg_gen_mov_tl(hex_llsc_addr, vaddr); 341 tcg_gen_mov_i64(hex_llsc_val_i64, dest); 342 } 343 344 static inline void gen_store_conditional4(DisasContext *ctx, 345 TCGv pred, TCGv vaddr, TCGv src) 346 { 347 TCGLabel *fail = gen_new_label(); 348 TCGLabel *done = gen_new_label(); 349 TCGv one, zero, tmp; 350 351 tcg_gen_brcond_tl(TCG_COND_NE, vaddr, hex_llsc_addr, fail); 352 353 one = tcg_constant_tl(0xff); 354 zero = tcg_constant_tl(0); 355 tmp = tcg_temp_new(); 356 tcg_gen_atomic_cmpxchg_tl(tmp, hex_llsc_addr, hex_llsc_val, src, 357 ctx->mem_idx, MO_32); 358 tcg_gen_movcond_tl(TCG_COND_EQ, pred, tmp, hex_llsc_val, 359 one, zero); 360 tcg_temp_free(tmp); 361 tcg_gen_br(done); 362 363 gen_set_label(fail); 364 tcg_gen_movi_tl(pred, 0); 365 366 gen_set_label(done); 367 tcg_gen_movi_tl(hex_llsc_addr, ~0); 368 } 369 370 static inline void gen_store_conditional8(DisasContext *ctx, 371 TCGv pred, TCGv vaddr, TCGv_i64 src) 372 { 373 TCGLabel *fail = gen_new_label(); 374 TCGLabel *done = gen_new_label(); 375 TCGv_i64 one, zero, tmp; 376 377 tcg_gen_brcond_tl(TCG_COND_NE, vaddr, hex_llsc_addr, fail); 378 379 one = tcg_constant_i64(0xff); 380 zero = tcg_constant_i64(0); 381 tmp = tcg_temp_new_i64(); 382 tcg_gen_atomic_cmpxchg_i64(tmp, hex_llsc_addr, hex_llsc_val_i64, src, 383 ctx->mem_idx, MO_64); 384 tcg_gen_movcond_i64(TCG_COND_EQ, tmp, tmp, hex_llsc_val_i64, 385 one, zero); 386 tcg_gen_extrl_i64_i32(pred, tmp); 387 tcg_temp_free_i64(tmp); 388 tcg_gen_br(done); 389 390 gen_set_label(fail); 391 tcg_gen_movi_tl(pred, 0); 392 393 gen_set_label(done); 394 tcg_gen_movi_tl(hex_llsc_addr, ~0); 395 } 396 397 static inline void gen_store32(TCGv vaddr, TCGv src, int width, int slot) 398 { 399 tcg_gen_mov_tl(hex_store_addr[slot], vaddr); 400 tcg_gen_movi_tl(hex_store_width[slot], width); 401 tcg_gen_mov_tl(hex_store_val32[slot], src); 402 } 403 404 static inline void gen_store1(TCGv_env cpu_env, TCGv vaddr, TCGv src, int slot) 405 { 406 gen_store32(vaddr, src, 1, slot); 407 } 408 409 static inline void gen_store1i(TCGv_env cpu_env, TCGv vaddr, int32_t src, int slot) 410 { 411 TCGv tmp = tcg_constant_tl(src); 412 gen_store1(cpu_env, vaddr, tmp, slot); 413 } 414 415 static inline void gen_store2(TCGv_env cpu_env, TCGv vaddr, TCGv src, int slot) 416 { 417 gen_store32(vaddr, src, 2, slot); 418 } 419 420 static inline void gen_store2i(TCGv_env cpu_env, TCGv vaddr, int32_t src, int slot) 421 { 422 TCGv tmp = tcg_constant_tl(src); 423 gen_store2(cpu_env, vaddr, tmp, slot); 424 } 425 426 static inline void gen_store4(TCGv_env cpu_env, TCGv vaddr, TCGv src, int slot) 427 { 428 gen_store32(vaddr, src, 4, slot); 429 } 430 431 static inline void gen_store4i(TCGv_env cpu_env, TCGv vaddr, int32_t src, int slot) 432 { 433 TCGv tmp = tcg_constant_tl(src); 434 gen_store4(cpu_env, vaddr, tmp, slot); 435 } 436 437 static inline void gen_store8(TCGv_env cpu_env, TCGv vaddr, TCGv_i64 src, int slot) 438 { 439 tcg_gen_mov_tl(hex_store_addr[slot], vaddr); 440 tcg_gen_movi_tl(hex_store_width[slot], 8); 441 tcg_gen_mov_i64(hex_store_val64[slot], src); 442 } 443 444 static inline void gen_store8i(TCGv_env cpu_env, TCGv vaddr, int64_t src, int slot) 445 { 446 TCGv_i64 tmp = tcg_constant_i64(src); 447 gen_store8(cpu_env, vaddr, tmp, slot); 448 } 449 450 static TCGv gen_8bitsof(TCGv result, TCGv value) 451 { 452 TCGv zero = tcg_constant_tl(0); 453 TCGv ones = tcg_constant_tl(0xff); 454 tcg_gen_movcond_tl(TCG_COND_NE, result, value, zero, ones, zero); 455 456 return result; 457 } 458 459 static intptr_t vreg_src_off(DisasContext *ctx, int num) 460 { 461 intptr_t offset = offsetof(CPUHexagonState, VRegs[num]); 462 463 if (test_bit(num, ctx->vregs_select)) { 464 offset = ctx_future_vreg_off(ctx, num, 1, false); 465 } 466 if (test_bit(num, ctx->vregs_updated_tmp)) { 467 offset = ctx_tmp_vreg_off(ctx, num, 1, false); 468 } 469 return offset; 470 } 471 472 static void gen_log_vreg_write(DisasContext *ctx, intptr_t srcoff, int num, 473 VRegWriteType type, int slot_num, 474 bool is_predicated) 475 { 476 TCGLabel *label_end = NULL; 477 intptr_t dstoff; 478 479 if (is_predicated) { 480 TCGv cancelled = tcg_temp_local_new(); 481 label_end = gen_new_label(); 482 483 /* Don't do anything if the slot was cancelled */ 484 tcg_gen_extract_tl(cancelled, hex_slot_cancelled, slot_num, 1); 485 tcg_gen_brcondi_tl(TCG_COND_NE, cancelled, 0, label_end); 486 tcg_temp_free(cancelled); 487 } 488 489 if (type != EXT_TMP) { 490 dstoff = ctx_future_vreg_off(ctx, num, 1, true); 491 tcg_gen_gvec_mov(MO_64, dstoff, srcoff, 492 sizeof(MMVector), sizeof(MMVector)); 493 tcg_gen_ori_tl(hex_VRegs_updated, hex_VRegs_updated, 1 << num); 494 } else { 495 dstoff = ctx_tmp_vreg_off(ctx, num, 1, false); 496 tcg_gen_gvec_mov(MO_64, dstoff, srcoff, 497 sizeof(MMVector), sizeof(MMVector)); 498 } 499 500 if (is_predicated) { 501 gen_set_label(label_end); 502 } 503 } 504 505 static void gen_log_vreg_write_pair(DisasContext *ctx, intptr_t srcoff, int num, 506 VRegWriteType type, int slot_num, 507 bool is_predicated) 508 { 509 gen_log_vreg_write(ctx, srcoff, num ^ 0, type, slot_num, is_predicated); 510 srcoff += sizeof(MMVector); 511 gen_log_vreg_write(ctx, srcoff, num ^ 1, type, slot_num, is_predicated); 512 } 513 514 static void gen_log_qreg_write(intptr_t srcoff, int num, int vnew, 515 int slot_num, bool is_predicated) 516 { 517 TCGLabel *label_end = NULL; 518 intptr_t dstoff; 519 520 if (is_predicated) { 521 TCGv cancelled = tcg_temp_local_new(); 522 label_end = gen_new_label(); 523 524 /* Don't do anything if the slot was cancelled */ 525 tcg_gen_extract_tl(cancelled, hex_slot_cancelled, slot_num, 1); 526 tcg_gen_brcondi_tl(TCG_COND_NE, cancelled, 0, label_end); 527 tcg_temp_free(cancelled); 528 } 529 530 dstoff = offsetof(CPUHexagonState, future_QRegs[num]); 531 tcg_gen_gvec_mov(MO_64, dstoff, srcoff, sizeof(MMQReg), sizeof(MMQReg)); 532 533 if (is_predicated) { 534 tcg_gen_ori_tl(hex_QRegs_updated, hex_QRegs_updated, 1 << num); 535 gen_set_label(label_end); 536 } 537 } 538 539 static void gen_vreg_load(DisasContext *ctx, intptr_t dstoff, TCGv src, 540 bool aligned) 541 { 542 TCGv_i64 tmp = tcg_temp_new_i64(); 543 if (aligned) { 544 tcg_gen_andi_tl(src, src, ~((int32_t)sizeof(MMVector) - 1)); 545 } 546 for (int i = 0; i < sizeof(MMVector) / 8; i++) { 547 tcg_gen_qemu_ld64(tmp, src, ctx->mem_idx); 548 tcg_gen_addi_tl(src, src, 8); 549 tcg_gen_st_i64(tmp, cpu_env, dstoff + i * 8); 550 } 551 tcg_temp_free_i64(tmp); 552 } 553 554 static void gen_vreg_store(DisasContext *ctx, Insn *insn, Packet *pkt, 555 TCGv EA, intptr_t srcoff, int slot, bool aligned) 556 { 557 intptr_t dstoff = offsetof(CPUHexagonState, vstore[slot].data); 558 intptr_t maskoff = offsetof(CPUHexagonState, vstore[slot].mask); 559 560 if (is_gather_store_insn(insn, pkt)) { 561 TCGv sl = tcg_constant_tl(slot); 562 gen_helper_gather_store(cpu_env, EA, sl); 563 return; 564 } 565 566 tcg_gen_movi_tl(hex_vstore_pending[slot], 1); 567 if (aligned) { 568 tcg_gen_andi_tl(hex_vstore_addr[slot], EA, 569 ~((int32_t)sizeof(MMVector) - 1)); 570 } else { 571 tcg_gen_mov_tl(hex_vstore_addr[slot], EA); 572 } 573 tcg_gen_movi_tl(hex_vstore_size[slot], sizeof(MMVector)); 574 575 /* Copy the data to the vstore buffer */ 576 tcg_gen_gvec_mov(MO_64, dstoff, srcoff, sizeof(MMVector), sizeof(MMVector)); 577 /* Set the mask to all 1's */ 578 tcg_gen_gvec_dup_imm(MO_64, maskoff, sizeof(MMQReg), sizeof(MMQReg), ~0LL); 579 } 580 581 static void gen_vreg_masked_store(DisasContext *ctx, TCGv EA, intptr_t srcoff, 582 intptr_t bitsoff, int slot, bool invert) 583 { 584 intptr_t dstoff = offsetof(CPUHexagonState, vstore[slot].data); 585 intptr_t maskoff = offsetof(CPUHexagonState, vstore[slot].mask); 586 587 tcg_gen_movi_tl(hex_vstore_pending[slot], 1); 588 tcg_gen_andi_tl(hex_vstore_addr[slot], EA, 589 ~((int32_t)sizeof(MMVector) - 1)); 590 tcg_gen_movi_tl(hex_vstore_size[slot], sizeof(MMVector)); 591 592 /* Copy the data to the vstore buffer */ 593 tcg_gen_gvec_mov(MO_64, dstoff, srcoff, sizeof(MMVector), sizeof(MMVector)); 594 /* Copy the mask */ 595 tcg_gen_gvec_mov(MO_64, maskoff, bitsoff, sizeof(MMQReg), sizeof(MMQReg)); 596 if (invert) { 597 tcg_gen_gvec_not(MO_64, maskoff, maskoff, 598 sizeof(MMQReg), sizeof(MMQReg)); 599 } 600 } 601 602 static void vec_to_qvec(size_t size, intptr_t dstoff, intptr_t srcoff) 603 { 604 TCGv_i64 tmp = tcg_temp_new_i64(); 605 TCGv_i64 word = tcg_temp_new_i64(); 606 TCGv_i64 bits = tcg_temp_new_i64(); 607 TCGv_i64 mask = tcg_temp_new_i64(); 608 TCGv_i64 zero = tcg_constant_i64(0); 609 TCGv_i64 ones = tcg_constant_i64(~0); 610 611 for (int i = 0; i < sizeof(MMVector) / 8; i++) { 612 tcg_gen_ld_i64(tmp, cpu_env, srcoff + i * 8); 613 tcg_gen_movi_i64(mask, 0); 614 615 for (int j = 0; j < 8; j += size) { 616 tcg_gen_extract_i64(word, tmp, j * 8, size * 8); 617 tcg_gen_movcond_i64(TCG_COND_NE, bits, word, zero, ones, zero); 618 tcg_gen_deposit_i64(mask, mask, bits, j, size); 619 } 620 621 tcg_gen_st8_i64(mask, cpu_env, dstoff + i); 622 } 623 tcg_temp_free_i64(tmp); 624 tcg_temp_free_i64(word); 625 tcg_temp_free_i64(bits); 626 tcg_temp_free_i64(mask); 627 } 628 629 static void probe_noshuf_load(TCGv va, int s, int mi) 630 { 631 TCGv size = tcg_constant_tl(s); 632 TCGv mem_idx = tcg_constant_tl(mi); 633 gen_helper_probe_noshuf_load(cpu_env, va, size, mem_idx); 634 } 635 636 #include "tcg_funcs_generated.c.inc" 637 #include "tcg_func_table_generated.c.inc" 638