1/* 2 * Tiny Code Generator for QEMU 3 * 4 * Copyright (c) 2008 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 25#include "elf.h" 26 27/* 28 * Standardize on the _CALL_FOO symbols used by GCC: 29 * Apple XCode does not define _CALL_DARWIN. 30 * Clang defines _CALL_ELF (64-bit) but not _CALL_SYSV or _CALL_AIX. 31 */ 32#if TCG_TARGET_REG_BITS == 64 33# ifdef _CALL_AIX 34 /* ok */ 35# elif defined(_CALL_ELF) && _CALL_ELF == 1 36# define _CALL_AIX 37# elif defined(_CALL_ELF) && _CALL_ELF == 2 38 /* ok */ 39# else 40# error "Unknown ABI" 41# endif 42#else 43# if defined(_CALL_SYSV) || defined(_CALL_DARWIN) 44 /* ok */ 45# elif defined(__APPLE__) 46# define _CALL_DARWIN 47# elif defined(__ELF__) 48# define _CALL_SYSV 49# else 50# error "Unknown ABI" 51# endif 52#endif 53 54#if TCG_TARGET_REG_BITS == 64 55# define TCG_TARGET_CALL_ARG_I32 TCG_CALL_ARG_EXTEND 56# define TCG_TARGET_CALL_RET_I128 TCG_CALL_RET_NORMAL 57#else 58# define TCG_TARGET_CALL_ARG_I32 TCG_CALL_ARG_NORMAL 59# define TCG_TARGET_CALL_RET_I128 TCG_CALL_RET_BY_REF 60#endif 61#ifdef _CALL_SYSV 62# define TCG_TARGET_CALL_ARG_I64 TCG_CALL_ARG_EVEN 63# define TCG_TARGET_CALL_ARG_I128 TCG_CALL_ARG_BY_REF 64#else 65# define TCG_TARGET_CALL_ARG_I64 TCG_CALL_ARG_NORMAL 66# define TCG_TARGET_CALL_ARG_I128 TCG_CALL_ARG_NORMAL 67#endif 68 69/* For some memory operations, we need a scratch that isn't R0. For the AIX 70 calling convention, we can re-use the TOC register since we'll be reloading 71 it at every call. Otherwise R12 will do nicely as neither a call-saved 72 register nor a parameter register. */ 73#ifdef _CALL_AIX 74# define TCG_REG_TMP1 TCG_REG_R2 75#else 76# define TCG_REG_TMP1 TCG_REG_R12 77#endif 78#define TCG_REG_TMP2 TCG_REG_R11 79 80#define TCG_VEC_TMP1 TCG_REG_V0 81#define TCG_VEC_TMP2 TCG_REG_V1 82 83#define TCG_REG_TB TCG_REG_R31 84#define USE_REG_TB (TCG_TARGET_REG_BITS == 64 && !have_isa_3_00) 85 86/* Shorthand for size of a pointer. Avoid promotion to unsigned. */ 87#define SZP ((int)sizeof(void *)) 88 89/* Shorthand for size of a register. */ 90#define SZR (TCG_TARGET_REG_BITS / 8) 91 92#define TCG_CT_CONST_S16 0x100 93#define TCG_CT_CONST_U16 0x200 94#define TCG_CT_CONST_S32 0x400 95#define TCG_CT_CONST_U32 0x800 96#define TCG_CT_CONST_ZERO 0x1000 97#define TCG_CT_CONST_MONE 0x2000 98#define TCG_CT_CONST_WSZ 0x4000 99#define TCG_CT_CONST_CMP 0x8000 100 101#define ALL_GENERAL_REGS 0xffffffffu 102#define ALL_VECTOR_REGS 0xffffffff00000000ull 103 104#ifndef R_PPC64_PCREL34 105#define R_PPC64_PCREL34 132 106#endif 107 108#define have_isel (cpuinfo & CPUINFO_ISEL) 109 110#define TCG_GUEST_BASE_REG TCG_REG_R30 111 112#ifdef CONFIG_DEBUG_TCG 113static const char tcg_target_reg_names[TCG_TARGET_NB_REGS][4] = { 114 "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7", 115 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15", 116 "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23", 117 "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31", 118 "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", 119 "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", 120 "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", 121 "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", 122}; 123#endif 124 125static const int tcg_target_reg_alloc_order[] = { 126 TCG_REG_R14, /* call saved registers */ 127 TCG_REG_R15, 128 TCG_REG_R16, 129 TCG_REG_R17, 130 TCG_REG_R18, 131 TCG_REG_R19, 132 TCG_REG_R20, 133 TCG_REG_R21, 134 TCG_REG_R22, 135 TCG_REG_R23, 136 TCG_REG_R24, 137 TCG_REG_R25, 138 TCG_REG_R26, 139 TCG_REG_R27, 140 TCG_REG_R28, 141 TCG_REG_R29, 142 TCG_REG_R30, 143 TCG_REG_R31, 144 TCG_REG_R12, /* call clobbered, non-arguments */ 145 TCG_REG_R11, 146 TCG_REG_R2, 147 TCG_REG_R13, 148 TCG_REG_R10, /* call clobbered, arguments */ 149 TCG_REG_R9, 150 TCG_REG_R8, 151 TCG_REG_R7, 152 TCG_REG_R6, 153 TCG_REG_R5, 154 TCG_REG_R4, 155 TCG_REG_R3, 156 157 /* V0 and V1 reserved as temporaries; V20 - V31 are call-saved */ 158 TCG_REG_V2, /* call clobbered, vectors */ 159 TCG_REG_V3, 160 TCG_REG_V4, 161 TCG_REG_V5, 162 TCG_REG_V6, 163 TCG_REG_V7, 164 TCG_REG_V8, 165 TCG_REG_V9, 166 TCG_REG_V10, 167 TCG_REG_V11, 168 TCG_REG_V12, 169 TCG_REG_V13, 170 TCG_REG_V14, 171 TCG_REG_V15, 172 TCG_REG_V16, 173 TCG_REG_V17, 174 TCG_REG_V18, 175 TCG_REG_V19, 176}; 177 178static const int tcg_target_call_iarg_regs[] = { 179 TCG_REG_R3, 180 TCG_REG_R4, 181 TCG_REG_R5, 182 TCG_REG_R6, 183 TCG_REG_R7, 184 TCG_REG_R8, 185 TCG_REG_R9, 186 TCG_REG_R10 187}; 188 189static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot) 190{ 191 tcg_debug_assert(kind == TCG_CALL_RET_NORMAL); 192 tcg_debug_assert(slot >= 0 && slot <= 1); 193 return TCG_REG_R3 + slot; 194} 195 196static const int tcg_target_callee_save_regs[] = { 197#ifdef _CALL_DARWIN 198 TCG_REG_R11, 199#endif 200 TCG_REG_R14, 201 TCG_REG_R15, 202 TCG_REG_R16, 203 TCG_REG_R17, 204 TCG_REG_R18, 205 TCG_REG_R19, 206 TCG_REG_R20, 207 TCG_REG_R21, 208 TCG_REG_R22, 209 TCG_REG_R23, 210 TCG_REG_R24, 211 TCG_REG_R25, 212 TCG_REG_R26, 213 TCG_REG_R27, /* currently used for the global env */ 214 TCG_REG_R28, 215 TCG_REG_R29, 216 TCG_REG_R30, 217 TCG_REG_R31 218}; 219 220/* For PPC, we use TB+4 instead of TB as the base. */ 221static inline ptrdiff_t ppc_tbrel_diff(TCGContext *s, const void *target) 222{ 223 return tcg_tbrel_diff(s, target) - 4; 224} 225 226static inline bool in_range_b(tcg_target_long target) 227{ 228 return target == sextract64(target, 0, 26); 229} 230 231static uint32_t reloc_pc24_val(const tcg_insn_unit *pc, 232 const tcg_insn_unit *target) 233{ 234 ptrdiff_t disp = tcg_ptr_byte_diff(target, pc); 235 tcg_debug_assert(in_range_b(disp)); 236 return disp & 0x3fffffc; 237} 238 239static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target) 240{ 241 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw); 242 ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx); 243 244 if (in_range_b(disp)) { 245 *src_rw = (*src_rw & ~0x3fffffc) | (disp & 0x3fffffc); 246 return true; 247 } 248 return false; 249} 250 251static uint16_t reloc_pc14_val(const tcg_insn_unit *pc, 252 const tcg_insn_unit *target) 253{ 254 ptrdiff_t disp = tcg_ptr_byte_diff(target, pc); 255 tcg_debug_assert(disp == (int16_t) disp); 256 return disp & 0xfffc; 257} 258 259static bool reloc_pc14(tcg_insn_unit *src_rw, const tcg_insn_unit *target) 260{ 261 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw); 262 ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx); 263 264 if (disp == (int16_t) disp) { 265 *src_rw = (*src_rw & ~0xfffc) | (disp & 0xfffc); 266 return true; 267 } 268 return false; 269} 270 271static bool reloc_pc34(tcg_insn_unit *src_rw, const tcg_insn_unit *target) 272{ 273 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw); 274 ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx); 275 276 if (disp == sextract64(disp, 0, 34)) { 277 src_rw[0] = (src_rw[0] & ~0x3ffff) | ((disp >> 16) & 0x3ffff); 278 src_rw[1] = (src_rw[1] & ~0xffff) | (disp & 0xffff); 279 return true; 280 } 281 return false; 282} 283 284static bool mask_operand(uint32_t c, int *mb, int *me); 285static bool mask64_operand(uint64_t c, int *mb, int *me); 286 287/* test if a constant matches the constraint */ 288static bool tcg_target_const_match(int64_t sval, int ct, 289 TCGType type, TCGCond cond, int vece) 290{ 291 uint64_t uval = sval; 292 int mb, me; 293 294 if (ct & TCG_CT_CONST) { 295 return 1; 296 } 297 298 if (type == TCG_TYPE_I32) { 299 uval = (uint32_t)sval; 300 sval = (int32_t)sval; 301 } 302 303 if (ct & TCG_CT_CONST_CMP) { 304 switch (cond) { 305 case TCG_COND_EQ: 306 case TCG_COND_NE: 307 ct |= TCG_CT_CONST_S16 | TCG_CT_CONST_U16; 308 break; 309 case TCG_COND_LT: 310 case TCG_COND_GE: 311 case TCG_COND_LE: 312 case TCG_COND_GT: 313 ct |= TCG_CT_CONST_S16; 314 break; 315 case TCG_COND_LTU: 316 case TCG_COND_GEU: 317 case TCG_COND_LEU: 318 case TCG_COND_GTU: 319 ct |= TCG_CT_CONST_U16; 320 break; 321 case TCG_COND_TSTEQ: 322 case TCG_COND_TSTNE: 323 if ((uval & ~0xffff) == 0 || (uval & ~0xffff0000ull) == 0) { 324 return 1; 325 } 326 if (uval == (uint32_t)uval && mask_operand(uval, &mb, &me)) { 327 return 1; 328 } 329 if (TCG_TARGET_REG_BITS == 64 && 330 mask64_operand(uval << clz64(uval), &mb, &me)) { 331 return 1; 332 } 333 return 0; 334 default: 335 g_assert_not_reached(); 336 } 337 } 338 339 if ((ct & TCG_CT_CONST_S16) && sval == (int16_t)sval) { 340 return 1; 341 } 342 if ((ct & TCG_CT_CONST_U16) && uval == (uint16_t)uval) { 343 return 1; 344 } 345 if ((ct & TCG_CT_CONST_S32) && sval == (int32_t)sval) { 346 return 1; 347 } 348 if ((ct & TCG_CT_CONST_U32) && uval == (uint32_t)uval) { 349 return 1; 350 } 351 if ((ct & TCG_CT_CONST_ZERO) && sval == 0) { 352 return 1; 353 } 354 if ((ct & TCG_CT_CONST_MONE) && sval == -1) { 355 return 1; 356 } 357 if ((ct & TCG_CT_CONST_WSZ) && sval == (type == TCG_TYPE_I32 ? 32 : 64)) { 358 return 1; 359 } 360 return 0; 361} 362 363#define OPCD(opc) ((opc)<<26) 364#define XO19(opc) (OPCD(19)|((opc)<<1)) 365#define MD30(opc) (OPCD(30)|((opc)<<2)) 366#define MDS30(opc) (OPCD(30)|((opc)<<1)) 367#define XO31(opc) (OPCD(31)|((opc)<<1)) 368#define XO58(opc) (OPCD(58)|(opc)) 369#define XO62(opc) (OPCD(62)|(opc)) 370#define VX4(opc) (OPCD(4)|(opc)) 371 372#define B OPCD( 18) 373#define BC OPCD( 16) 374 375#define LBZ OPCD( 34) 376#define LHZ OPCD( 40) 377#define LHA OPCD( 42) 378#define LWZ OPCD( 32) 379#define LWZUX XO31( 55) 380#define LD XO58( 0) 381#define LDX XO31( 21) 382#define LDU XO58( 1) 383#define LDUX XO31( 53) 384#define LWA XO58( 2) 385#define LWAX XO31(341) 386#define LQ OPCD( 56) 387 388#define STB OPCD( 38) 389#define STH OPCD( 44) 390#define STW OPCD( 36) 391#define STD XO62( 0) 392#define STDU XO62( 1) 393#define STDX XO31(149) 394#define STQ XO62( 2) 395 396#define PLWA OPCD( 41) 397#define PLD OPCD( 57) 398#define PLXSD OPCD( 42) 399#define PLXV OPCD(25 * 2 + 1) /* force tx=1 */ 400 401#define PSTD OPCD( 61) 402#define PSTXSD OPCD( 46) 403#define PSTXV OPCD(27 * 2 + 1) /* force sx=1 */ 404 405#define ADDIC OPCD( 12) 406#define ADDI OPCD( 14) 407#define ADDIS OPCD( 15) 408#define ORI OPCD( 24) 409#define ORIS OPCD( 25) 410#define XORI OPCD( 26) 411#define XORIS OPCD( 27) 412#define ANDI OPCD( 28) 413#define ANDIS OPCD( 29) 414#define MULLI OPCD( 7) 415#define CMPLI OPCD( 10) 416#define CMPI OPCD( 11) 417#define SUBFIC OPCD( 8) 418 419#define LWZU OPCD( 33) 420#define STWU OPCD( 37) 421 422#define RLWIMI OPCD( 20) 423#define RLWINM OPCD( 21) 424#define RLWNM OPCD( 23) 425 426#define RLDICL MD30( 0) 427#define RLDICR MD30( 1) 428#define RLDIMI MD30( 3) 429#define RLDCL MDS30( 8) 430 431#define BCLR XO19( 16) 432#define BCCTR XO19(528) 433#define CRAND XO19(257) 434#define CRANDC XO19(129) 435#define CRNAND XO19(225) 436#define CROR XO19(449) 437#define CRNOR XO19( 33) 438#define ADDPCIS XO19( 2) 439 440#define EXTSB XO31(954) 441#define EXTSH XO31(922) 442#define EXTSW XO31(986) 443#define ADD XO31(266) 444#define ADDE XO31(138) 445#define ADDME XO31(234) 446#define ADDZE XO31(202) 447#define ADDC XO31( 10) 448#define AND XO31( 28) 449#define SUBF XO31( 40) 450#define SUBFC XO31( 8) 451#define SUBFE XO31(136) 452#define SUBFME XO31(232) 453#define SUBFZE XO31(200) 454#define OR XO31(444) 455#define XOR XO31(316) 456#define MULLW XO31(235) 457#define MULHW XO31( 75) 458#define MULHWU XO31( 11) 459#define DIVW XO31(491) 460#define DIVWU XO31(459) 461#define MODSW XO31(779) 462#define MODUW XO31(267) 463#define CMP XO31( 0) 464#define CMPL XO31( 32) 465#define LHBRX XO31(790) 466#define LWBRX XO31(534) 467#define LDBRX XO31(532) 468#define STHBRX XO31(918) 469#define STWBRX XO31(662) 470#define STDBRX XO31(660) 471#define MFSPR XO31(339) 472#define MTSPR XO31(467) 473#define SRAWI XO31(824) 474#define NEG XO31(104) 475#define MFCR XO31( 19) 476#define MFOCRF (MFCR | (1u << 20)) 477#define NOR XO31(124) 478#define CNTLZW XO31( 26) 479#define CNTLZD XO31( 58) 480#define CNTTZW XO31(538) 481#define CNTTZD XO31(570) 482#define CNTPOPW XO31(378) 483#define CNTPOPD XO31(506) 484#define ANDC XO31( 60) 485#define ORC XO31(412) 486#define EQV XO31(284) 487#define NAND XO31(476) 488#define ISEL XO31( 15) 489 490#define MULLD XO31(233) 491#define MULHD XO31( 73) 492#define MULHDU XO31( 9) 493#define DIVD XO31(489) 494#define DIVDU XO31(457) 495#define MODSD XO31(777) 496#define MODUD XO31(265) 497 498#define LBZX XO31( 87) 499#define LHZX XO31(279) 500#define LHAX XO31(343) 501#define LWZX XO31( 23) 502#define STBX XO31(215) 503#define STHX XO31(407) 504#define STWX XO31(151) 505 506#define EIEIO XO31(854) 507#define HWSYNC XO31(598) 508#define LWSYNC (HWSYNC | (1u << 21)) 509 510#define SPR(a, b) ((((a)<<5)|(b))<<11) 511#define LR SPR(8, 0) 512#define CTR SPR(9, 0) 513 514#define SLW XO31( 24) 515#define SRW XO31(536) 516#define SRAW XO31(792) 517 518#define SLD XO31( 27) 519#define SRD XO31(539) 520#define SRAD XO31(794) 521#define SRADI XO31(413<<1) 522 523#define BRH XO31(219) 524#define BRW XO31(155) 525#define BRD XO31(187) 526 527#define TW XO31( 4) 528#define TRAP (TW | TO(31)) 529 530#define SETBC XO31(384) /* v3.10 */ 531#define SETBCR XO31(416) /* v3.10 */ 532#define SETNBC XO31(448) /* v3.10 */ 533#define SETNBCR XO31(480) /* v3.10 */ 534 535#define NOP ORI /* ori 0,0,0 */ 536 537#define LVX XO31(103) 538#define LVEBX XO31(7) 539#define LVEHX XO31(39) 540#define LVEWX XO31(71) 541#define LXSDX (XO31(588) | 1) /* v2.06, force tx=1 */ 542#define LXVDSX (XO31(332) | 1) /* v2.06, force tx=1 */ 543#define LXSIWZX (XO31(12) | 1) /* v2.07, force tx=1 */ 544#define LXV (OPCD(61) | 8 | 1) /* v3.00, force tx=1 */ 545#define LXSD (OPCD(57) | 2) /* v3.00 */ 546#define LXVWSX (XO31(364) | 1) /* v3.00, force tx=1 */ 547 548#define STVX XO31(231) 549#define STVEWX XO31(199) 550#define STXSDX (XO31(716) | 1) /* v2.06, force sx=1 */ 551#define STXSIWX (XO31(140) | 1) /* v2.07, force sx=1 */ 552#define STXV (OPCD(61) | 8 | 5) /* v3.00, force sx=1 */ 553#define STXSD (OPCD(61) | 2) /* v3.00 */ 554 555#define VADDSBS VX4(768) 556#define VADDUBS VX4(512) 557#define VADDUBM VX4(0) 558#define VADDSHS VX4(832) 559#define VADDUHS VX4(576) 560#define VADDUHM VX4(64) 561#define VADDSWS VX4(896) 562#define VADDUWS VX4(640) 563#define VADDUWM VX4(128) 564#define VADDUDM VX4(192) /* v2.07 */ 565 566#define VSUBSBS VX4(1792) 567#define VSUBUBS VX4(1536) 568#define VSUBUBM VX4(1024) 569#define VSUBSHS VX4(1856) 570#define VSUBUHS VX4(1600) 571#define VSUBUHM VX4(1088) 572#define VSUBSWS VX4(1920) 573#define VSUBUWS VX4(1664) 574#define VSUBUWM VX4(1152) 575#define VSUBUDM VX4(1216) /* v2.07 */ 576 577#define VNEGW (VX4(1538) | (6 << 16)) /* v3.00 */ 578#define VNEGD (VX4(1538) | (7 << 16)) /* v3.00 */ 579 580#define VMAXSB VX4(258) 581#define VMAXSH VX4(322) 582#define VMAXSW VX4(386) 583#define VMAXSD VX4(450) /* v2.07 */ 584#define VMAXUB VX4(2) 585#define VMAXUH VX4(66) 586#define VMAXUW VX4(130) 587#define VMAXUD VX4(194) /* v2.07 */ 588#define VMINSB VX4(770) 589#define VMINSH VX4(834) 590#define VMINSW VX4(898) 591#define VMINSD VX4(962) /* v2.07 */ 592#define VMINUB VX4(514) 593#define VMINUH VX4(578) 594#define VMINUW VX4(642) 595#define VMINUD VX4(706) /* v2.07 */ 596 597#define VCMPEQUB VX4(6) 598#define VCMPEQUH VX4(70) 599#define VCMPEQUW VX4(134) 600#define VCMPEQUD VX4(199) /* v2.07 */ 601#define VCMPGTSB VX4(774) 602#define VCMPGTSH VX4(838) 603#define VCMPGTSW VX4(902) 604#define VCMPGTSD VX4(967) /* v2.07 */ 605#define VCMPGTUB VX4(518) 606#define VCMPGTUH VX4(582) 607#define VCMPGTUW VX4(646) 608#define VCMPGTUD VX4(711) /* v2.07 */ 609#define VCMPNEB VX4(7) /* v3.00 */ 610#define VCMPNEH VX4(71) /* v3.00 */ 611#define VCMPNEW VX4(135) /* v3.00 */ 612 613#define VSLB VX4(260) 614#define VSLH VX4(324) 615#define VSLW VX4(388) 616#define VSLD VX4(1476) /* v2.07 */ 617#define VSRB VX4(516) 618#define VSRH VX4(580) 619#define VSRW VX4(644) 620#define VSRD VX4(1732) /* v2.07 */ 621#define VSRAB VX4(772) 622#define VSRAH VX4(836) 623#define VSRAW VX4(900) 624#define VSRAD VX4(964) /* v2.07 */ 625#define VRLB VX4(4) 626#define VRLH VX4(68) 627#define VRLW VX4(132) 628#define VRLD VX4(196) /* v2.07 */ 629 630#define VMULEUB VX4(520) 631#define VMULEUH VX4(584) 632#define VMULEUW VX4(648) /* v2.07 */ 633#define VMULOUB VX4(8) 634#define VMULOUH VX4(72) 635#define VMULOUW VX4(136) /* v2.07 */ 636#define VMULUWM VX4(137) /* v2.07 */ 637#define VMULLD VX4(457) /* v3.10 */ 638#define VMSUMUHM VX4(38) 639 640#define VMRGHB VX4(12) 641#define VMRGHH VX4(76) 642#define VMRGHW VX4(140) 643#define VMRGLB VX4(268) 644#define VMRGLH VX4(332) 645#define VMRGLW VX4(396) 646 647#define VPKUHUM VX4(14) 648#define VPKUWUM VX4(78) 649 650#define VAND VX4(1028) 651#define VANDC VX4(1092) 652#define VNOR VX4(1284) 653#define VOR VX4(1156) 654#define VXOR VX4(1220) 655#define VEQV VX4(1668) /* v2.07 */ 656#define VNAND VX4(1412) /* v2.07 */ 657#define VORC VX4(1348) /* v2.07 */ 658 659#define VSPLTB VX4(524) 660#define VSPLTH VX4(588) 661#define VSPLTW VX4(652) 662#define VSPLTISB VX4(780) 663#define VSPLTISH VX4(844) 664#define VSPLTISW VX4(908) 665 666#define VSLDOI VX4(44) 667 668#define XXPERMDI (OPCD(60) | (10 << 3) | 7) /* v2.06, force ax=bx=tx=1 */ 669#define XXSEL (OPCD(60) | (3 << 4) | 0xf) /* v2.06, force ax=bx=cx=tx=1 */ 670#define XXSPLTIB (OPCD(60) | (360 << 1) | 1) /* v3.00, force tx=1 */ 671 672#define MFVSRD (XO31(51) | 1) /* v2.07, force sx=1 */ 673#define MFVSRWZ (XO31(115) | 1) /* v2.07, force sx=1 */ 674#define MTVSRD (XO31(179) | 1) /* v2.07, force tx=1 */ 675#define MTVSRWZ (XO31(243) | 1) /* v2.07, force tx=1 */ 676#define MTVSRDD (XO31(435) | 1) /* v3.00, force tx=1 */ 677#define MTVSRWS (XO31(403) | 1) /* v3.00, force tx=1 */ 678 679#define RT(r) ((r)<<21) 680#define RS(r) ((r)<<21) 681#define RA(r) ((r)<<16) 682#define RB(r) ((r)<<11) 683#define TO(t) ((t)<<21) 684#define SH(s) ((s)<<11) 685#define MB(b) ((b)<<6) 686#define ME(e) ((e)<<1) 687#define BO(o) ((o)<<21) 688#define MB64(b) ((b)<<5) 689#define FXM(b) (1 << (19 - (b))) 690 691#define VRT(r) (((r) & 31) << 21) 692#define VRA(r) (((r) & 31) << 16) 693#define VRB(r) (((r) & 31) << 11) 694#define VRC(r) (((r) & 31) << 6) 695 696#define LK 1 697 698#define TAB(t, a, b) (RT(t) | RA(a) | RB(b)) 699#define SAB(s, a, b) (RS(s) | RA(a) | RB(b)) 700#define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff)) 701#define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff)) 702 703#define BF(n) ((n)<<23) 704#define BI(n, c) (((c)+((n)*4))<<16) 705#define BT(n, c) (((c)+((n)*4))<<21) 706#define BA(n, c) (((c)+((n)*4))<<16) 707#define BB(n, c) (((c)+((n)*4))<<11) 708#define BC_(n, c) (((c)+((n)*4))<<6) 709 710#define BO_COND_TRUE BO(12) 711#define BO_COND_FALSE BO( 4) 712#define BO_ALWAYS BO(20) 713 714enum { 715 CR_LT, 716 CR_GT, 717 CR_EQ, 718 CR_SO 719}; 720 721static const uint32_t tcg_to_bc[16] = { 722 [TCG_COND_EQ] = BC | BI(0, CR_EQ) | BO_COND_TRUE, 723 [TCG_COND_NE] = BC | BI(0, CR_EQ) | BO_COND_FALSE, 724 [TCG_COND_TSTEQ] = BC | BI(0, CR_EQ) | BO_COND_TRUE, 725 [TCG_COND_TSTNE] = BC | BI(0, CR_EQ) | BO_COND_FALSE, 726 [TCG_COND_LT] = BC | BI(0, CR_LT) | BO_COND_TRUE, 727 [TCG_COND_GE] = BC | BI(0, CR_LT) | BO_COND_FALSE, 728 [TCG_COND_LE] = BC | BI(0, CR_GT) | BO_COND_FALSE, 729 [TCG_COND_GT] = BC | BI(0, CR_GT) | BO_COND_TRUE, 730 [TCG_COND_LTU] = BC | BI(0, CR_LT) | BO_COND_TRUE, 731 [TCG_COND_GEU] = BC | BI(0, CR_LT) | BO_COND_FALSE, 732 [TCG_COND_LEU] = BC | BI(0, CR_GT) | BO_COND_FALSE, 733 [TCG_COND_GTU] = BC | BI(0, CR_GT) | BO_COND_TRUE, 734}; 735 736/* The low bit here is set if the RA and RB fields must be inverted. */ 737static const uint32_t tcg_to_isel[16] = { 738 [TCG_COND_EQ] = ISEL | BC_(0, CR_EQ), 739 [TCG_COND_NE] = ISEL | BC_(0, CR_EQ) | 1, 740 [TCG_COND_TSTEQ] = ISEL | BC_(0, CR_EQ), 741 [TCG_COND_TSTNE] = ISEL | BC_(0, CR_EQ) | 1, 742 [TCG_COND_LT] = ISEL | BC_(0, CR_LT), 743 [TCG_COND_GE] = ISEL | BC_(0, CR_LT) | 1, 744 [TCG_COND_LE] = ISEL | BC_(0, CR_GT) | 1, 745 [TCG_COND_GT] = ISEL | BC_(0, CR_GT), 746 [TCG_COND_LTU] = ISEL | BC_(0, CR_LT), 747 [TCG_COND_GEU] = ISEL | BC_(0, CR_LT) | 1, 748 [TCG_COND_LEU] = ISEL | BC_(0, CR_GT) | 1, 749 [TCG_COND_GTU] = ISEL | BC_(0, CR_GT), 750}; 751 752static bool patch_reloc(tcg_insn_unit *code_ptr, int type, 753 intptr_t value, intptr_t addend) 754{ 755 const tcg_insn_unit *target; 756 int16_t lo; 757 int32_t hi; 758 759 value += addend; 760 target = (const tcg_insn_unit *)value; 761 762 switch (type) { 763 case R_PPC_REL14: 764 return reloc_pc14(code_ptr, target); 765 case R_PPC_REL24: 766 return reloc_pc24(code_ptr, target); 767 case R_PPC64_PCREL34: 768 return reloc_pc34(code_ptr, target); 769 case R_PPC_ADDR16: 770 /* 771 * We are (slightly) abusing this relocation type. In particular, 772 * assert that the low 2 bits are zero, and do not modify them. 773 * That way we can use this with LD et al that have opcode bits 774 * in the low 2 bits of the insn. 775 */ 776 if ((value & 3) || value != (int16_t)value) { 777 return false; 778 } 779 *code_ptr = (*code_ptr & ~0xfffc) | (value & 0xfffc); 780 break; 781 case R_PPC_ADDR32: 782 /* 783 * We are abusing this relocation type. Again, this points to 784 * a pair of insns, lis + load. This is an absolute address 785 * relocation for PPC32 so the lis cannot be removed. 786 */ 787 lo = value; 788 hi = value - lo; 789 if (hi + lo != value) { 790 return false; 791 } 792 code_ptr[0] = deposit32(code_ptr[0], 0, 16, hi >> 16); 793 code_ptr[1] = deposit32(code_ptr[1], 0, 16, lo); 794 break; 795 default: 796 g_assert_not_reached(); 797 } 798 return true; 799} 800 801/* Ensure that the prefixed instruction does not cross a 64-byte boundary. */ 802static bool tcg_out_need_prefix_align(TCGContext *s) 803{ 804 return ((uintptr_t)s->code_ptr & 0x3f) == 0x3c; 805} 806 807static void tcg_out_prefix_align(TCGContext *s) 808{ 809 if (tcg_out_need_prefix_align(s)) { 810 tcg_out32(s, NOP); 811 } 812} 813 814static ptrdiff_t tcg_pcrel_diff_for_prefix(TCGContext *s, const void *target) 815{ 816 return tcg_pcrel_diff(s, target) - (tcg_out_need_prefix_align(s) ? 4 : 0); 817} 818 819/* Output Type 00 Prefix - 8-Byte Load/Store Form (8LS:D) */ 820static void tcg_out_8ls_d(TCGContext *s, tcg_insn_unit opc, unsigned rt, 821 unsigned ra, tcg_target_long imm, bool r) 822{ 823 tcg_insn_unit p, i; 824 825 p = OPCD(1) | (r << 20) | ((imm >> 16) & 0x3ffff); 826 i = opc | TAI(rt, ra, imm); 827 828 tcg_out_prefix_align(s); 829 tcg_out32(s, p); 830 tcg_out32(s, i); 831} 832 833/* Output Type 10 Prefix - Modified Load/Store Form (MLS:D) */ 834static void tcg_out_mls_d(TCGContext *s, tcg_insn_unit opc, unsigned rt, 835 unsigned ra, tcg_target_long imm, bool r) 836{ 837 tcg_insn_unit p, i; 838 839 p = OPCD(1) | (2 << 24) | (r << 20) | ((imm >> 16) & 0x3ffff); 840 i = opc | TAI(rt, ra, imm); 841 842 tcg_out_prefix_align(s); 843 tcg_out32(s, p); 844 tcg_out32(s, i); 845} 846 847static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt, 848 TCGReg base, tcg_target_long offset); 849 850static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg) 851{ 852 if (ret == arg) { 853 return true; 854 } 855 switch (type) { 856 case TCG_TYPE_I64: 857 tcg_debug_assert(TCG_TARGET_REG_BITS == 64); 858 /* fallthru */ 859 case TCG_TYPE_I32: 860 if (ret < TCG_REG_V0) { 861 if (arg < TCG_REG_V0) { 862 tcg_out32(s, OR | SAB(arg, ret, arg)); 863 break; 864 } else if (have_isa_2_07) { 865 tcg_out32(s, (type == TCG_TYPE_I32 ? MFVSRWZ : MFVSRD) 866 | VRT(arg) | RA(ret)); 867 break; 868 } else { 869 /* Altivec does not support vector->integer moves. */ 870 return false; 871 } 872 } else if (arg < TCG_REG_V0) { 873 if (have_isa_2_07) { 874 tcg_out32(s, (type == TCG_TYPE_I32 ? MTVSRWZ : MTVSRD) 875 | VRT(ret) | RA(arg)); 876 break; 877 } else { 878 /* Altivec does not support integer->vector moves. */ 879 return false; 880 } 881 } 882 /* fallthru */ 883 case TCG_TYPE_V64: 884 case TCG_TYPE_V128: 885 tcg_debug_assert(ret >= TCG_REG_V0 && arg >= TCG_REG_V0); 886 tcg_out32(s, VOR | VRT(ret) | VRA(arg) | VRB(arg)); 887 break; 888 default: 889 g_assert_not_reached(); 890 } 891 return true; 892} 893 894static void tcg_out_rld_rc(TCGContext *s, int op, TCGReg ra, TCGReg rs, 895 int sh, int mb, bool rc) 896{ 897 tcg_debug_assert(TCG_TARGET_REG_BITS == 64); 898 sh = SH(sh & 0x1f) | (((sh >> 5) & 1) << 1); 899 mb = MB64((mb >> 5) | ((mb << 1) & 0x3f)); 900 tcg_out32(s, op | RA(ra) | RS(rs) | sh | mb | rc); 901} 902 903static void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs, 904 int sh, int mb) 905{ 906 tcg_out_rld_rc(s, op, ra, rs, sh, mb, false); 907} 908 909static void tcg_out_rlw_rc(TCGContext *s, int op, TCGReg ra, TCGReg rs, 910 int sh, int mb, int me, bool rc) 911{ 912 tcg_debug_assert((mb & 0x1f) == mb); 913 tcg_debug_assert((me & 0x1f) == me); 914 tcg_out32(s, op | RA(ra) | RS(rs) | SH(sh & 0x1f) | MB(mb) | ME(me) | rc); 915} 916 917static void tcg_out_rlw(TCGContext *s, int op, TCGReg ra, TCGReg rs, 918 int sh, int mb, int me) 919{ 920 tcg_out_rlw_rc(s, op, ra, rs, sh, mb, me, false); 921} 922 923static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dst, TCGReg src) 924{ 925 tcg_out32(s, EXTSB | RA(dst) | RS(src)); 926} 927 928static void tcg_out_ext8u(TCGContext *s, TCGReg dst, TCGReg src) 929{ 930 tcg_out32(s, ANDI | SAI(src, dst, 0xff)); 931} 932 933static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dst, TCGReg src) 934{ 935 tcg_out32(s, EXTSH | RA(dst) | RS(src)); 936} 937 938static void tcg_out_ext16u(TCGContext *s, TCGReg dst, TCGReg src) 939{ 940 tcg_out32(s, ANDI | SAI(src, dst, 0xffff)); 941} 942 943static void tcg_out_ext32s(TCGContext *s, TCGReg dst, TCGReg src) 944{ 945 tcg_debug_assert(TCG_TARGET_REG_BITS == 64); 946 tcg_out32(s, EXTSW | RA(dst) | RS(src)); 947} 948 949static void tcg_out_ext32u(TCGContext *s, TCGReg dst, TCGReg src) 950{ 951 tcg_debug_assert(TCG_TARGET_REG_BITS == 64); 952 tcg_out_rld(s, RLDICL, dst, src, 0, 32); 953} 954 955static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dst, TCGReg src) 956{ 957 tcg_out_ext32s(s, dst, src); 958} 959 960static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dst, TCGReg src) 961{ 962 tcg_out_ext32u(s, dst, src); 963} 964 965static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rn) 966{ 967 tcg_debug_assert(TCG_TARGET_REG_BITS == 64); 968 tcg_out_mov(s, TCG_TYPE_I32, rd, rn); 969} 970 971static inline void tcg_out_shli32(TCGContext *s, TCGReg dst, TCGReg src, int c) 972{ 973 tcg_out_rlw(s, RLWINM, dst, src, c, 0, 31 - c); 974} 975 976static inline void tcg_out_shli64(TCGContext *s, TCGReg dst, TCGReg src, int c) 977{ 978 tcg_out_rld(s, RLDICR, dst, src, c, 63 - c); 979} 980 981static inline void tcg_out_sari32(TCGContext *s, TCGReg dst, TCGReg src, int c) 982{ 983 /* Limit immediate shift count lest we create an illegal insn. */ 984 tcg_out32(s, SRAWI | RA(dst) | RS(src) | SH(c & 31)); 985} 986 987static inline void tcg_out_shri32(TCGContext *s, TCGReg dst, TCGReg src, int c) 988{ 989 tcg_out_rlw(s, RLWINM, dst, src, 32 - c, c, 31); 990} 991 992static inline void tcg_out_shri64(TCGContext *s, TCGReg dst, TCGReg src, int c) 993{ 994 tcg_out_rld(s, RLDICL, dst, src, 64 - c, c); 995} 996 997static inline void tcg_out_sari64(TCGContext *s, TCGReg dst, TCGReg src, int c) 998{ 999 tcg_out32(s, SRADI | RA(dst) | RS(src) | SH(c & 0x1f) | ((c >> 4) & 2)); 1000} 1001 1002static void tcg_out_addpcis(TCGContext *s, TCGReg dst, intptr_t imm) 1003{ 1004 uint32_t d0, d1, d2; 1005 1006 tcg_debug_assert((imm & 0xffff) == 0); 1007 tcg_debug_assert(imm == (int32_t)imm); 1008 1009 d2 = extract32(imm, 16, 1); 1010 d1 = extract32(imm, 17, 5); 1011 d0 = extract32(imm, 22, 10); 1012 tcg_out32(s, ADDPCIS | RT(dst) | (d1 << 16) | (d0 << 6) | d2); 1013} 1014 1015static void tcg_out_bswap16(TCGContext *s, TCGReg dst, TCGReg src, int flags) 1016{ 1017 TCGReg tmp = dst == src ? TCG_REG_R0 : dst; 1018 1019 if (have_isa_3_10) { 1020 tcg_out32(s, BRH | RA(dst) | RS(src)); 1021 if (flags & TCG_BSWAP_OS) { 1022 tcg_out_ext16s(s, TCG_TYPE_REG, dst, dst); 1023 } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) { 1024 tcg_out_ext16u(s, dst, dst); 1025 } 1026 return; 1027 } 1028 1029 /* 1030 * In the following, 1031 * dep(a, b, m) -> (a & ~m) | (b & m) 1032 * 1033 * Begin with: src = xxxxabcd 1034 */ 1035 /* tmp = rol32(src, 24) & 0x000000ff = 0000000c */ 1036 tcg_out_rlw(s, RLWINM, tmp, src, 24, 24, 31); 1037 /* tmp = dep(tmp, rol32(src, 8), 0x0000ff00) = 000000dc */ 1038 tcg_out_rlw(s, RLWIMI, tmp, src, 8, 16, 23); 1039 1040 if (flags & TCG_BSWAP_OS) { 1041 tcg_out_ext16s(s, TCG_TYPE_REG, dst, tmp); 1042 } else { 1043 tcg_out_mov(s, TCG_TYPE_REG, dst, tmp); 1044 } 1045} 1046 1047static void tcg_out_bswap32(TCGContext *s, TCGReg dst, TCGReg src, int flags) 1048{ 1049 TCGReg tmp = dst == src ? TCG_REG_R0 : dst; 1050 1051 if (have_isa_3_10) { 1052 tcg_out32(s, BRW | RA(dst) | RS(src)); 1053 if (flags & TCG_BSWAP_OS) { 1054 tcg_out_ext32s(s, dst, dst); 1055 } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) { 1056 tcg_out_ext32u(s, dst, dst); 1057 } 1058 return; 1059 } 1060 1061 /* 1062 * Stolen from gcc's builtin_bswap32. 1063 * In the following, 1064 * dep(a, b, m) -> (a & ~m) | (b & m) 1065 * 1066 * Begin with: src = xxxxabcd 1067 */ 1068 /* tmp = rol32(src, 8) & 0xffffffff = 0000bcda */ 1069 tcg_out_rlw(s, RLWINM, tmp, src, 8, 0, 31); 1070 /* tmp = dep(tmp, rol32(src, 24), 0xff000000) = 0000dcda */ 1071 tcg_out_rlw(s, RLWIMI, tmp, src, 24, 0, 7); 1072 /* tmp = dep(tmp, rol32(src, 24), 0x0000ff00) = 0000dcba */ 1073 tcg_out_rlw(s, RLWIMI, tmp, src, 24, 16, 23); 1074 1075 if (flags & TCG_BSWAP_OS) { 1076 tcg_out_ext32s(s, dst, tmp); 1077 } else { 1078 tcg_out_mov(s, TCG_TYPE_REG, dst, tmp); 1079 } 1080} 1081 1082static void tcg_out_bswap64(TCGContext *s, TCGReg dst, TCGReg src) 1083{ 1084 TCGReg t0 = dst == src ? TCG_REG_R0 : dst; 1085 TCGReg t1 = dst == src ? dst : TCG_REG_R0; 1086 1087 if (have_isa_3_10) { 1088 tcg_out32(s, BRD | RA(dst) | RS(src)); 1089 return; 1090 } 1091 1092 /* 1093 * In the following, 1094 * dep(a, b, m) -> (a & ~m) | (b & m) 1095 * 1096 * Begin with: src = abcdefgh 1097 */ 1098 /* t0 = rol32(src, 8) & 0xffffffff = 0000fghe */ 1099 tcg_out_rlw(s, RLWINM, t0, src, 8, 0, 31); 1100 /* t0 = dep(t0, rol32(src, 24), 0xff000000) = 0000hghe */ 1101 tcg_out_rlw(s, RLWIMI, t0, src, 24, 0, 7); 1102 /* t0 = dep(t0, rol32(src, 24), 0x0000ff00) = 0000hgfe */ 1103 tcg_out_rlw(s, RLWIMI, t0, src, 24, 16, 23); 1104 1105 /* t0 = rol64(t0, 32) = hgfe0000 */ 1106 tcg_out_rld(s, RLDICL, t0, t0, 32, 0); 1107 /* t1 = rol64(src, 32) = efghabcd */ 1108 tcg_out_rld(s, RLDICL, t1, src, 32, 0); 1109 1110 /* t0 = dep(t0, rol32(t1, 24), 0xffffffff) = hgfebcda */ 1111 tcg_out_rlw(s, RLWIMI, t0, t1, 8, 0, 31); 1112 /* t0 = dep(t0, rol32(t1, 24), 0xff000000) = hgfedcda */ 1113 tcg_out_rlw(s, RLWIMI, t0, t1, 24, 0, 7); 1114 /* t0 = dep(t0, rol32(t1, 24), 0x0000ff00) = hgfedcba */ 1115 tcg_out_rlw(s, RLWIMI, t0, t1, 24, 16, 23); 1116 1117 tcg_out_mov(s, TCG_TYPE_REG, dst, t0); 1118} 1119 1120/* Emit a move into ret of arg, if it can be done in one insn. */ 1121static bool tcg_out_movi_one(TCGContext *s, TCGReg ret, tcg_target_long arg) 1122{ 1123 if (arg == (int16_t)arg) { 1124 tcg_out32(s, ADDI | TAI(ret, 0, arg)); 1125 return true; 1126 } 1127 if (arg == (int32_t)arg && (arg & 0xffff) == 0) { 1128 tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16)); 1129 return true; 1130 } 1131 return false; 1132} 1133 1134static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret, 1135 tcg_target_long arg, bool in_prologue) 1136{ 1137 intptr_t tb_diff; 1138 tcg_target_long tmp; 1139 int shift; 1140 1141 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32); 1142 1143 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) { 1144 arg = (int32_t)arg; 1145 } 1146 1147 /* Load 16-bit immediates with one insn. */ 1148 if (tcg_out_movi_one(s, ret, arg)) { 1149 return; 1150 } 1151 1152 /* Load addresses within the TB with one insn. */ 1153 tb_diff = ppc_tbrel_diff(s, (void *)arg); 1154 if (!in_prologue && USE_REG_TB && tb_diff == (int16_t)tb_diff) { 1155 tcg_out32(s, ADDI | TAI(ret, TCG_REG_TB, tb_diff)); 1156 return; 1157 } 1158 1159 /* 1160 * Load values up to 34 bits, and pc-relative addresses, 1161 * with one prefixed insn. 1162 */ 1163 if (have_isa_3_10) { 1164 if (arg == sextract64(arg, 0, 34)) { 1165 /* pli ret,value = paddi ret,0,value,0 */ 1166 tcg_out_mls_d(s, ADDI, ret, 0, arg, 0); 1167 return; 1168 } 1169 1170 tmp = tcg_pcrel_diff_for_prefix(s, (void *)arg); 1171 if (tmp == sextract64(tmp, 0, 34)) { 1172 /* pla ret,value = paddi ret,0,value,1 */ 1173 tcg_out_mls_d(s, ADDI, ret, 0, tmp, 1); 1174 return; 1175 } 1176 } 1177 1178 /* Load 32-bit immediates with two insns. Note that we've already 1179 eliminated bare ADDIS, so we know both insns are required. */ 1180 if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) { 1181 tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16)); 1182 tcg_out32(s, ORI | SAI(ret, ret, arg)); 1183 return; 1184 } 1185 if (arg == (uint32_t)arg && !(arg & 0x8000)) { 1186 tcg_out32(s, ADDI | TAI(ret, 0, arg)); 1187 tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16)); 1188 return; 1189 } 1190 1191 /* Load masked 16-bit value. */ 1192 if (arg > 0 && (arg & 0x8000)) { 1193 tmp = arg | 0x7fff; 1194 if ((tmp & (tmp + 1)) == 0) { 1195 int mb = clz64(tmp + 1) + 1; 1196 tcg_out32(s, ADDI | TAI(ret, 0, arg)); 1197 tcg_out_rld(s, RLDICL, ret, ret, 0, mb); 1198 return; 1199 } 1200 } 1201 1202 /* Load common masks with 2 insns. */ 1203 shift = ctz64(arg); 1204 tmp = arg >> shift; 1205 if (tmp == (int16_t)tmp) { 1206 tcg_out32(s, ADDI | TAI(ret, 0, tmp)); 1207 tcg_out_shli64(s, ret, ret, shift); 1208 return; 1209 } 1210 shift = clz64(arg); 1211 if (tcg_out_movi_one(s, ret, arg << shift)) { 1212 tcg_out_shri64(s, ret, ret, shift); 1213 return; 1214 } 1215 1216 /* Load addresses within 2GB with 2 insns. */ 1217 if (have_isa_3_00) { 1218 intptr_t hi = tcg_pcrel_diff(s, (void *)arg) - 4; 1219 int16_t lo = hi; 1220 1221 hi -= lo; 1222 if (hi == (int32_t)hi) { 1223 tcg_out_addpcis(s, TCG_REG_TMP2, hi); 1224 tcg_out32(s, ADDI | TAI(ret, TCG_REG_TMP2, lo)); 1225 return; 1226 } 1227 } 1228 1229 /* Load addresses within 2GB of TB with 2 (or rarely 3) insns. */ 1230 if (!in_prologue && USE_REG_TB && tb_diff == (int32_t)tb_diff) { 1231 tcg_out_mem_long(s, ADDI, ADD, ret, TCG_REG_TB, tb_diff); 1232 return; 1233 } 1234 1235 /* Use the constant pool, if possible. */ 1236 if (!in_prologue && USE_REG_TB) { 1237 new_pool_label(s, arg, R_PPC_ADDR16, s->code_ptr, 1238 ppc_tbrel_diff(s, NULL)); 1239 tcg_out32(s, LD | TAI(ret, TCG_REG_TB, 0)); 1240 return; 1241 } 1242 if (have_isa_3_10) { 1243 tcg_out_8ls_d(s, PLD, ret, 0, 0, 1); 1244 new_pool_label(s, arg, R_PPC64_PCREL34, s->code_ptr - 2, 0); 1245 return; 1246 } 1247 if (have_isa_3_00) { 1248 tcg_out_addpcis(s, TCG_REG_TMP2, 0); 1249 new_pool_label(s, arg, R_PPC_REL14, s->code_ptr, 0); 1250 tcg_out32(s, LD | TAI(ret, TCG_REG_TMP2, 0)); 1251 return; 1252 } 1253 1254 tmp = arg >> 31 >> 1; 1255 tcg_out_movi(s, TCG_TYPE_I32, ret, tmp); 1256 if (tmp) { 1257 tcg_out_shli64(s, ret, ret, 32); 1258 } 1259 if (arg & 0xffff0000) { 1260 tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16)); 1261 } 1262 if (arg & 0xffff) { 1263 tcg_out32(s, ORI | SAI(ret, ret, arg)); 1264 } 1265} 1266 1267static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece, 1268 TCGReg ret, int64_t val) 1269{ 1270 uint32_t load_insn; 1271 int rel, low; 1272 intptr_t add; 1273 1274 switch (vece) { 1275 case MO_8: 1276 low = (int8_t)val; 1277 if (low >= -16 && low < 16) { 1278 tcg_out32(s, VSPLTISB | VRT(ret) | ((val & 31) << 16)); 1279 return; 1280 } 1281 if (have_isa_3_00) { 1282 tcg_out32(s, XXSPLTIB | VRT(ret) | ((val & 0xff) << 11)); 1283 return; 1284 } 1285 break; 1286 1287 case MO_16: 1288 low = (int16_t)val; 1289 if (low >= -16 && low < 16) { 1290 tcg_out32(s, VSPLTISH | VRT(ret) | ((val & 31) << 16)); 1291 return; 1292 } 1293 break; 1294 1295 case MO_32: 1296 low = (int32_t)val; 1297 if (low >= -16 && low < 16) { 1298 tcg_out32(s, VSPLTISW | VRT(ret) | ((val & 31) << 16)); 1299 return; 1300 } 1301 break; 1302 } 1303 1304 /* 1305 * Otherwise we must load the value from the constant pool. 1306 */ 1307 if (USE_REG_TB) { 1308 rel = R_PPC_ADDR16; 1309 add = ppc_tbrel_diff(s, NULL); 1310 } else if (have_isa_3_10) { 1311 if (type == TCG_TYPE_V64) { 1312 tcg_out_8ls_d(s, PLXSD, ret & 31, 0, 0, 1); 1313 new_pool_label(s, val, R_PPC64_PCREL34, s->code_ptr - 2, 0); 1314 } else { 1315 tcg_out_8ls_d(s, PLXV, ret & 31, 0, 0, 1); 1316 new_pool_l2(s, R_PPC64_PCREL34, s->code_ptr - 2, 0, val, val); 1317 } 1318 return; 1319 } else if (have_isa_3_00) { 1320 tcg_out_addpcis(s, TCG_REG_TMP1, 0); 1321 rel = R_PPC_REL14; 1322 add = 0; 1323 } else { 1324 rel = R_PPC_ADDR32; 1325 add = 0; 1326 } 1327 1328 if (have_vsx) { 1329 load_insn = type == TCG_TYPE_V64 ? LXSDX : LXVDSX; 1330 load_insn |= VRT(ret) | RB(TCG_REG_TMP1); 1331 if (TCG_TARGET_REG_BITS == 64) { 1332 new_pool_label(s, val, rel, s->code_ptr, add); 1333 } else { 1334 new_pool_l2(s, rel, s->code_ptr, add, val >> 32, val); 1335 } 1336 } else { 1337 load_insn = LVX | VRT(ret) | RB(TCG_REG_TMP1); 1338 if (TCG_TARGET_REG_BITS == 64) { 1339 new_pool_l2(s, rel, s->code_ptr, add, val, val); 1340 } else { 1341 new_pool_l4(s, rel, s->code_ptr, add, 1342 val >> 32, val, val >> 32, val); 1343 } 1344 } 1345 1346 if (USE_REG_TB) { 1347 tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, 0, 0)); 1348 load_insn |= RA(TCG_REG_TB); 1349 } else if (have_isa_3_00) { 1350 tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, TCG_REG_TMP1, 0)); 1351 } else { 1352 tcg_out32(s, ADDIS | TAI(TCG_REG_TMP1, 0, 0)); 1353 tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, TCG_REG_TMP1, 0)); 1354 } 1355 tcg_out32(s, load_insn); 1356} 1357 1358static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret, 1359 tcg_target_long arg) 1360{ 1361 switch (type) { 1362 case TCG_TYPE_I32: 1363 case TCG_TYPE_I64: 1364 tcg_debug_assert(ret < TCG_REG_V0); 1365 tcg_out_movi_int(s, type, ret, arg, false); 1366 break; 1367 1368 default: 1369 g_assert_not_reached(); 1370 } 1371} 1372 1373static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2) 1374{ 1375 return false; 1376} 1377 1378static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs, 1379 tcg_target_long imm) 1380{ 1381 /* This function is only used for passing structs by reference. */ 1382 g_assert_not_reached(); 1383} 1384 1385static bool mask_operand(uint32_t c, int *mb, int *me) 1386{ 1387 uint32_t lsb, test; 1388 1389 /* Accept a bit pattern like: 1390 0....01....1 1391 1....10....0 1392 0..01..10..0 1393 Keep track of the transitions. */ 1394 if (c == 0 || c == -1) { 1395 return false; 1396 } 1397 test = c; 1398 lsb = test & -test; 1399 test += lsb; 1400 if (test & (test - 1)) { 1401 return false; 1402 } 1403 1404 *me = clz32(lsb); 1405 *mb = test ? clz32(test & -test) + 1 : 0; 1406 return true; 1407} 1408 1409static bool mask64_operand(uint64_t c, int *mb, int *me) 1410{ 1411 uint64_t lsb; 1412 1413 if (c == 0) { 1414 return false; 1415 } 1416 1417 lsb = c & -c; 1418 /* Accept 1..10..0. */ 1419 if (c == -lsb) { 1420 *mb = 0; 1421 *me = clz64(lsb); 1422 return true; 1423 } 1424 /* Accept 0..01..1. */ 1425 if (lsb == 1 && (c & (c + 1)) == 0) { 1426 *mb = clz64(c + 1) + 1; 1427 *me = 63; 1428 return true; 1429 } 1430 return false; 1431} 1432 1433static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c) 1434{ 1435 int mb, me; 1436 1437 if (mask_operand(c, &mb, &me)) { 1438 tcg_out_rlw(s, RLWINM, dst, src, 0, mb, me); 1439 } else if ((c & 0xffff) == c) { 1440 tcg_out32(s, ANDI | SAI(src, dst, c)); 1441 return; 1442 } else if ((c & 0xffff0000) == c) { 1443 tcg_out32(s, ANDIS | SAI(src, dst, c >> 16)); 1444 return; 1445 } else { 1446 tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R0, c); 1447 tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0)); 1448 } 1449} 1450 1451static void tcg_out_andi64(TCGContext *s, TCGReg dst, TCGReg src, uint64_t c) 1452{ 1453 int mb, me; 1454 1455 tcg_debug_assert(TCG_TARGET_REG_BITS == 64); 1456 if (mask64_operand(c, &mb, &me)) { 1457 if (mb == 0) { 1458 tcg_out_rld(s, RLDICR, dst, src, 0, me); 1459 } else { 1460 tcg_out_rld(s, RLDICL, dst, src, 0, mb); 1461 } 1462 } else if ((c & 0xffff) == c) { 1463 tcg_out32(s, ANDI | SAI(src, dst, c)); 1464 return; 1465 } else if ((c & 0xffff0000) == c) { 1466 tcg_out32(s, ANDIS | SAI(src, dst, c >> 16)); 1467 return; 1468 } else { 1469 tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, c); 1470 tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0)); 1471 } 1472} 1473 1474static void tcg_out_zori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c, 1475 int op_lo, int op_hi) 1476{ 1477 if (c >> 16) { 1478 tcg_out32(s, op_hi | SAI(src, dst, c >> 16)); 1479 src = dst; 1480 } 1481 if (c & 0xffff) { 1482 tcg_out32(s, op_lo | SAI(src, dst, c)); 1483 src = dst; 1484 } 1485} 1486 1487static void tcg_out_ori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c) 1488{ 1489 tcg_out_zori32(s, dst, src, c, ORI, ORIS); 1490} 1491 1492static void tcg_out_xori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c) 1493{ 1494 tcg_out_zori32(s, dst, src, c, XORI, XORIS); 1495} 1496 1497static void tcg_out_b(TCGContext *s, int mask, const tcg_insn_unit *target) 1498{ 1499 ptrdiff_t disp = tcg_pcrel_diff(s, target); 1500 if (in_range_b(disp)) { 1501 tcg_out32(s, B | (disp & 0x3fffffc) | mask); 1502 } else { 1503 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, (uintptr_t)target); 1504 tcg_out32(s, MTSPR | RS(TCG_REG_R0) | CTR); 1505 tcg_out32(s, BCCTR | BO_ALWAYS | mask); 1506 } 1507} 1508 1509static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt, 1510 TCGReg base, tcg_target_long offset) 1511{ 1512 tcg_target_long orig = offset, l0, l1, extra = 0, align = 0; 1513 bool is_int_store = false; 1514 TCGReg rs = TCG_REG_TMP1; 1515 1516 switch (opi) { 1517 case LD: case LWA: 1518 align = 3; 1519 /* FALLTHRU */ 1520 default: 1521 if (rt > TCG_REG_R0 && rt < TCG_REG_V0) { 1522 rs = rt; 1523 break; 1524 } 1525 break; 1526 case LXSD: 1527 case STXSD: 1528 align = 3; 1529 break; 1530 case LXV: 1531 case STXV: 1532 align = 15; 1533 break; 1534 case STD: 1535 align = 3; 1536 /* FALLTHRU */ 1537 case STB: case STH: case STW: 1538 is_int_store = true; 1539 break; 1540 } 1541 1542 /* For unaligned or large offsets, use the prefixed form. */ 1543 if (have_isa_3_10 1544 && (offset != (int16_t)offset || (offset & align)) 1545 && offset == sextract64(offset, 0, 34)) { 1546 /* 1547 * Note that the MLS:D insns retain their un-prefixed opcode, 1548 * while the 8LS:D insns use a different opcode space. 1549 */ 1550 switch (opi) { 1551 case LBZ: 1552 case LHZ: 1553 case LHA: 1554 case LWZ: 1555 case STB: 1556 case STH: 1557 case STW: 1558 case ADDI: 1559 tcg_out_mls_d(s, opi, rt, base, offset, 0); 1560 return; 1561 case LWA: 1562 tcg_out_8ls_d(s, PLWA, rt, base, offset, 0); 1563 return; 1564 case LD: 1565 tcg_out_8ls_d(s, PLD, rt, base, offset, 0); 1566 return; 1567 case STD: 1568 tcg_out_8ls_d(s, PSTD, rt, base, offset, 0); 1569 return; 1570 case LXSD: 1571 tcg_out_8ls_d(s, PLXSD, rt & 31, base, offset, 0); 1572 return; 1573 case STXSD: 1574 tcg_out_8ls_d(s, PSTXSD, rt & 31, base, offset, 0); 1575 return; 1576 case LXV: 1577 tcg_out_8ls_d(s, PLXV, rt & 31, base, offset, 0); 1578 return; 1579 case STXV: 1580 tcg_out_8ls_d(s, PSTXV, rt & 31, base, offset, 0); 1581 return; 1582 } 1583 } 1584 1585 /* For unaligned, or very large offsets, use the indexed form. */ 1586 if (offset & align || offset != (int32_t)offset || opi == 0) { 1587 if (rs == base) { 1588 rs = TCG_REG_R0; 1589 } 1590 tcg_debug_assert(!is_int_store || rs != rt); 1591 tcg_out_movi(s, TCG_TYPE_PTR, rs, orig); 1592 tcg_out32(s, opx | TAB(rt & 31, base, rs)); 1593 return; 1594 } 1595 1596 l0 = (int16_t)offset; 1597 offset = (offset - l0) >> 16; 1598 l1 = (int16_t)offset; 1599 1600 if (l1 < 0 && orig >= 0) { 1601 extra = 0x4000; 1602 l1 = (int16_t)(offset - 0x4000); 1603 } 1604 if (l1) { 1605 tcg_out32(s, ADDIS | TAI(rs, base, l1)); 1606 base = rs; 1607 } 1608 if (extra) { 1609 tcg_out32(s, ADDIS | TAI(rs, base, extra)); 1610 base = rs; 1611 } 1612 if (opi != ADDI || base != rt || l0 != 0) { 1613 tcg_out32(s, opi | TAI(rt & 31, base, l0)); 1614 } 1615} 1616 1617static void tcg_out_vsldoi(TCGContext *s, TCGReg ret, 1618 TCGReg va, TCGReg vb, int shb) 1619{ 1620 tcg_out32(s, VSLDOI | VRT(ret) | VRA(va) | VRB(vb) | (shb << 6)); 1621} 1622 1623static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret, 1624 TCGReg base, intptr_t offset) 1625{ 1626 int shift; 1627 1628 switch (type) { 1629 case TCG_TYPE_I32: 1630 if (ret < TCG_REG_V0) { 1631 tcg_out_mem_long(s, LWZ, LWZX, ret, base, offset); 1632 break; 1633 } 1634 if (have_isa_2_07 && have_vsx) { 1635 tcg_out_mem_long(s, 0, LXSIWZX, ret, base, offset); 1636 break; 1637 } 1638 tcg_debug_assert((offset & 3) == 0); 1639 tcg_out_mem_long(s, 0, LVEWX, ret, base, offset); 1640 shift = (offset - 4) & 0xc; 1641 if (shift) { 1642 tcg_out_vsldoi(s, ret, ret, ret, shift); 1643 } 1644 break; 1645 case TCG_TYPE_I64: 1646 if (ret < TCG_REG_V0) { 1647 tcg_debug_assert(TCG_TARGET_REG_BITS == 64); 1648 tcg_out_mem_long(s, LD, LDX, ret, base, offset); 1649 break; 1650 } 1651 /* fallthru */ 1652 case TCG_TYPE_V64: 1653 tcg_debug_assert(ret >= TCG_REG_V0); 1654 if (have_vsx) { 1655 tcg_out_mem_long(s, have_isa_3_00 ? LXSD : 0, LXSDX, 1656 ret, base, offset); 1657 break; 1658 } 1659 tcg_debug_assert((offset & 7) == 0); 1660 tcg_out_mem_long(s, 0, LVX, ret, base, offset & -16); 1661 if (offset & 8) { 1662 tcg_out_vsldoi(s, ret, ret, ret, 8); 1663 } 1664 break; 1665 case TCG_TYPE_V128: 1666 tcg_debug_assert(ret >= TCG_REG_V0); 1667 tcg_debug_assert((offset & 15) == 0); 1668 tcg_out_mem_long(s, have_isa_3_00 ? LXV : 0, 1669 LVX, ret, base, offset); 1670 break; 1671 default: 1672 g_assert_not_reached(); 1673 } 1674} 1675 1676static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, 1677 TCGReg base, intptr_t offset) 1678{ 1679 int shift; 1680 1681 switch (type) { 1682 case TCG_TYPE_I32: 1683 if (arg < TCG_REG_V0) { 1684 tcg_out_mem_long(s, STW, STWX, arg, base, offset); 1685 break; 1686 } 1687 if (have_isa_2_07 && have_vsx) { 1688 tcg_out_mem_long(s, 0, STXSIWX, arg, base, offset); 1689 break; 1690 } 1691 assert((offset & 3) == 0); 1692 tcg_debug_assert((offset & 3) == 0); 1693 shift = (offset - 4) & 0xc; 1694 if (shift) { 1695 tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, shift); 1696 arg = TCG_VEC_TMP1; 1697 } 1698 tcg_out_mem_long(s, 0, STVEWX, arg, base, offset); 1699 break; 1700 case TCG_TYPE_I64: 1701 if (arg < TCG_REG_V0) { 1702 tcg_debug_assert(TCG_TARGET_REG_BITS == 64); 1703 tcg_out_mem_long(s, STD, STDX, arg, base, offset); 1704 break; 1705 } 1706 /* fallthru */ 1707 case TCG_TYPE_V64: 1708 tcg_debug_assert(arg >= TCG_REG_V0); 1709 if (have_vsx) { 1710 tcg_out_mem_long(s, have_isa_3_00 ? STXSD : 0, 1711 STXSDX, arg, base, offset); 1712 break; 1713 } 1714 tcg_debug_assert((offset & 7) == 0); 1715 if (offset & 8) { 1716 tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, 8); 1717 arg = TCG_VEC_TMP1; 1718 } 1719 tcg_out_mem_long(s, 0, STVEWX, arg, base, offset); 1720 tcg_out_mem_long(s, 0, STVEWX, arg, base, offset + 4); 1721 break; 1722 case TCG_TYPE_V128: 1723 tcg_debug_assert(arg >= TCG_REG_V0); 1724 tcg_out_mem_long(s, have_isa_3_00 ? STXV : 0, 1725 STVX, arg, base, offset); 1726 break; 1727 default: 1728 g_assert_not_reached(); 1729 } 1730} 1731 1732static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val, 1733 TCGReg base, intptr_t ofs) 1734{ 1735 return false; 1736} 1737 1738/* 1739 * Set dest non-zero if and only if (arg1 & arg2) is non-zero. 1740 * If RC, then also set RC0. 1741 */ 1742static void tcg_out_test(TCGContext *s, TCGReg dest, TCGReg arg1, TCGArg arg2, 1743 bool const_arg2, TCGType type, bool rc) 1744{ 1745 int mb, me; 1746 1747 if (!const_arg2) { 1748 tcg_out32(s, AND | SAB(arg1, dest, arg2) | rc); 1749 return; 1750 } 1751 1752 if (type == TCG_TYPE_I32) { 1753 arg2 = (uint32_t)arg2; 1754 } 1755 1756 if ((arg2 & ~0xffff) == 0) { 1757 tcg_out32(s, ANDI | SAI(arg1, dest, arg2)); 1758 return; 1759 } 1760 if ((arg2 & ~0xffff0000ull) == 0) { 1761 tcg_out32(s, ANDIS | SAI(arg1, dest, arg2 >> 16)); 1762 return; 1763 } 1764 if (arg2 == (uint32_t)arg2 && mask_operand(arg2, &mb, &me)) { 1765 tcg_out_rlw_rc(s, RLWINM, dest, arg1, 0, mb, me, rc); 1766 return; 1767 } 1768 if (TCG_TARGET_REG_BITS == 64) { 1769 int sh = clz64(arg2); 1770 if (mask64_operand(arg2 << sh, &mb, &me)) { 1771 tcg_out_rld_rc(s, RLDICR, dest, arg1, sh, me, rc); 1772 return; 1773 } 1774 } 1775 /* Constraints should satisfy this. */ 1776 g_assert_not_reached(); 1777} 1778 1779static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2, 1780 int const_arg2, int cr, TCGType type) 1781{ 1782 int imm; 1783 uint32_t op; 1784 1785 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32); 1786 1787 /* 1788 * Simplify the comparisons below wrt CMPI. 1789 * All of the tests are 16-bit, so a 32-bit sign extend always works. 1790 */ 1791 if (type == TCG_TYPE_I32) { 1792 arg2 = (int32_t)arg2; 1793 } 1794 1795 switch (cond) { 1796 case TCG_COND_EQ: 1797 case TCG_COND_NE: 1798 if (const_arg2) { 1799 if ((int16_t) arg2 == arg2) { 1800 op = CMPI; 1801 imm = 1; 1802 break; 1803 } else if ((uint16_t) arg2 == arg2) { 1804 op = CMPLI; 1805 imm = 1; 1806 break; 1807 } 1808 } 1809 op = CMPL; 1810 imm = 0; 1811 break; 1812 1813 case TCG_COND_TSTEQ: 1814 case TCG_COND_TSTNE: 1815 tcg_debug_assert(cr == 0); 1816 tcg_out_test(s, TCG_REG_R0, arg1, arg2, const_arg2, type, true); 1817 return; 1818 1819 case TCG_COND_LT: 1820 case TCG_COND_GE: 1821 case TCG_COND_LE: 1822 case TCG_COND_GT: 1823 if (const_arg2) { 1824 if ((int16_t) arg2 == arg2) { 1825 op = CMPI; 1826 imm = 1; 1827 break; 1828 } 1829 } 1830 op = CMP; 1831 imm = 0; 1832 break; 1833 1834 case TCG_COND_LTU: 1835 case TCG_COND_GEU: 1836 case TCG_COND_LEU: 1837 case TCG_COND_GTU: 1838 if (const_arg2) { 1839 if ((uint16_t) arg2 == arg2) { 1840 op = CMPLI; 1841 imm = 1; 1842 break; 1843 } 1844 } 1845 op = CMPL; 1846 imm = 0; 1847 break; 1848 1849 default: 1850 g_assert_not_reached(); 1851 } 1852 op |= BF(cr) | ((type == TCG_TYPE_I64) << 21); 1853 1854 if (imm) { 1855 tcg_out32(s, op | RA(arg1) | (arg2 & 0xffff)); 1856 } else { 1857 if (const_arg2) { 1858 tcg_out_movi(s, type, TCG_REG_R0, arg2); 1859 arg2 = TCG_REG_R0; 1860 } 1861 tcg_out32(s, op | RA(arg1) | RB(arg2)); 1862 } 1863} 1864 1865static void tcg_out_setcond_eq0(TCGContext *s, TCGType type, 1866 TCGReg dst, TCGReg src, bool neg) 1867{ 1868 if (neg && (TCG_TARGET_REG_BITS == 32 || type == TCG_TYPE_I64)) { 1869 /* 1870 * X != 0 implies X + -1 generates a carry. 1871 * RT = (~X + X) + CA 1872 * = -1 + CA 1873 * = CA ? 0 : -1 1874 */ 1875 tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1)); 1876 tcg_out32(s, SUBFE | TAB(dst, src, src)); 1877 return; 1878 } 1879 1880 if (type == TCG_TYPE_I32) { 1881 tcg_out32(s, CNTLZW | RS(src) | RA(dst)); 1882 tcg_out_shri32(s, dst, dst, 5); 1883 } else { 1884 tcg_out32(s, CNTLZD | RS(src) | RA(dst)); 1885 tcg_out_shri64(s, dst, dst, 6); 1886 } 1887 if (neg) { 1888 tcg_out32(s, NEG | RT(dst) | RA(dst)); 1889 } 1890} 1891 1892static void tcg_out_setcond_ne0(TCGContext *s, TCGType type, 1893 TCGReg dst, TCGReg src, bool neg) 1894{ 1895 if (!neg && (TCG_TARGET_REG_BITS == 32 || type == TCG_TYPE_I64)) { 1896 /* 1897 * X != 0 implies X + -1 generates a carry. Extra addition 1898 * trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C. 1899 */ 1900 tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1)); 1901 tcg_out32(s, SUBFE | TAB(dst, TCG_REG_R0, src)); 1902 return; 1903 } 1904 tcg_out_setcond_eq0(s, type, dst, src, false); 1905 if (neg) { 1906 tcg_out32(s, ADDI | TAI(dst, dst, -1)); 1907 } else { 1908 tcg_out_xori32(s, dst, dst, 1); 1909 } 1910} 1911 1912static TCGReg tcg_gen_setcond_xor(TCGContext *s, TCGReg arg1, TCGArg arg2, 1913 bool const_arg2) 1914{ 1915 if (const_arg2) { 1916 if ((uint32_t)arg2 == arg2) { 1917 tcg_out_xori32(s, TCG_REG_R0, arg1, arg2); 1918 } else { 1919 tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, arg2); 1920 tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, TCG_REG_R0)); 1921 } 1922 } else { 1923 tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, arg2)); 1924 } 1925 return TCG_REG_R0; 1926} 1927 1928static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond, 1929 TCGArg arg0, TCGArg arg1, TCGArg arg2, 1930 int const_arg2, bool neg) 1931{ 1932 int sh; 1933 bool inv; 1934 1935 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32); 1936 1937 /* Ignore high bits of a potential constant arg2. */ 1938 if (type == TCG_TYPE_I32) { 1939 arg2 = (uint32_t)arg2; 1940 } 1941 1942 /* With SETBC/SETBCR, we can always implement with 2 insns. */ 1943 if (have_isa_3_10) { 1944 tcg_insn_unit bi, opc; 1945 1946 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 0, type); 1947 1948 /* Re-use tcg_to_bc for BI and BO_COND_{TRUE,FALSE}. */ 1949 bi = tcg_to_bc[cond] & (0x1f << 16); 1950 if (tcg_to_bc[cond] & BO(8)) { 1951 opc = neg ? SETNBC : SETBC; 1952 } else { 1953 opc = neg ? SETNBCR : SETBCR; 1954 } 1955 tcg_out32(s, opc | RT(arg0) | bi); 1956 return; 1957 } 1958 1959 /* Handle common and trivial cases before handling anything else. */ 1960 if (arg2 == 0) { 1961 switch (cond) { 1962 case TCG_COND_EQ: 1963 tcg_out_setcond_eq0(s, type, arg0, arg1, neg); 1964 return; 1965 case TCG_COND_NE: 1966 tcg_out_setcond_ne0(s, type, arg0, arg1, neg); 1967 return; 1968 case TCG_COND_GE: 1969 tcg_out32(s, NOR | SAB(arg1, arg0, arg1)); 1970 arg1 = arg0; 1971 /* FALLTHRU */ 1972 case TCG_COND_LT: 1973 /* Extract the sign bit. */ 1974 if (type == TCG_TYPE_I32) { 1975 if (neg) { 1976 tcg_out_sari32(s, arg0, arg1, 31); 1977 } else { 1978 tcg_out_shri32(s, arg0, arg1, 31); 1979 } 1980 } else { 1981 if (neg) { 1982 tcg_out_sari64(s, arg0, arg1, 63); 1983 } else { 1984 tcg_out_shri64(s, arg0, arg1, 63); 1985 } 1986 } 1987 return; 1988 default: 1989 break; 1990 } 1991 } 1992 1993 /* If we have ISEL, we can implement everything with 3 or 4 insns. 1994 All other cases below are also at least 3 insns, so speed up the 1995 code generator by not considering them and always using ISEL. */ 1996 if (have_isel) { 1997 int isel, tab; 1998 1999 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 0, type); 2000 2001 isel = tcg_to_isel[cond]; 2002 2003 tcg_out_movi(s, type, arg0, neg ? -1 : 1); 2004 if (isel & 1) { 2005 /* arg0 = (bc ? 0 : 1) */ 2006 tab = TAB(arg0, 0, arg0); 2007 isel &= ~1; 2008 } else { 2009 /* arg0 = (bc ? 1 : 0) */ 2010 tcg_out_movi(s, type, TCG_REG_R0, 0); 2011 tab = TAB(arg0, arg0, TCG_REG_R0); 2012 } 2013 tcg_out32(s, isel | tab); 2014 return; 2015 } 2016 2017 inv = false; 2018 switch (cond) { 2019 case TCG_COND_EQ: 2020 arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2); 2021 tcg_out_setcond_eq0(s, type, arg0, arg1, neg); 2022 break; 2023 2024 case TCG_COND_NE: 2025 arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2); 2026 tcg_out_setcond_ne0(s, type, arg0, arg1, neg); 2027 break; 2028 2029 case TCG_COND_TSTEQ: 2030 tcg_out_test(s, TCG_REG_R0, arg1, arg2, const_arg2, type, false); 2031 tcg_out_setcond_eq0(s, type, arg0, TCG_REG_R0, neg); 2032 break; 2033 2034 case TCG_COND_TSTNE: 2035 tcg_out_test(s, TCG_REG_R0, arg1, arg2, const_arg2, type, false); 2036 tcg_out_setcond_ne0(s, type, arg0, TCG_REG_R0, neg); 2037 break; 2038 2039 case TCG_COND_LE: 2040 case TCG_COND_LEU: 2041 inv = true; 2042 /* fall through */ 2043 case TCG_COND_GT: 2044 case TCG_COND_GTU: 2045 sh = 30; /* CR7 CR_GT */ 2046 goto crtest; 2047 2048 case TCG_COND_GE: 2049 case TCG_COND_GEU: 2050 inv = true; 2051 /* fall through */ 2052 case TCG_COND_LT: 2053 case TCG_COND_LTU: 2054 sh = 29; /* CR7 CR_LT */ 2055 goto crtest; 2056 2057 crtest: 2058 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type); 2059 tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7)); 2060 tcg_out_rlw(s, RLWINM, arg0, TCG_REG_R0, sh, 31, 31); 2061 if (neg && inv) { 2062 tcg_out32(s, ADDI | TAI(arg0, arg0, -1)); 2063 } else if (neg) { 2064 tcg_out32(s, NEG | RT(arg0) | RA(arg0)); 2065 } else if (inv) { 2066 tcg_out_xori32(s, arg0, arg0, 1); 2067 } 2068 break; 2069 2070 default: 2071 g_assert_not_reached(); 2072 } 2073} 2074 2075static void tcg_out_bc(TCGContext *s, TCGCond cond, int bd) 2076{ 2077 tcg_out32(s, tcg_to_bc[cond] | bd); 2078} 2079 2080static void tcg_out_bc_lab(TCGContext *s, TCGCond cond, TCGLabel *l) 2081{ 2082 int bd = 0; 2083 if (l->has_value) { 2084 bd = reloc_pc14_val(tcg_splitwx_to_rx(s->code_ptr), l->u.value_ptr); 2085 } else { 2086 tcg_out_reloc(s, s->code_ptr, R_PPC_REL14, l, 0); 2087 } 2088 tcg_out_bc(s, cond, bd); 2089} 2090 2091static void tcg_out_brcond(TCGContext *s, TCGCond cond, 2092 TCGArg arg1, TCGArg arg2, int const_arg2, 2093 TCGLabel *l, TCGType type) 2094{ 2095 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 0, type); 2096 tcg_out_bc_lab(s, cond, l); 2097} 2098 2099static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond, 2100 TCGArg dest, TCGArg c1, TCGArg c2, TCGArg v1, 2101 TCGArg v2, bool const_c2) 2102{ 2103 /* If for some reason both inputs are zero, don't produce bad code. */ 2104 if (v1 == 0 && v2 == 0) { 2105 tcg_out_movi(s, type, dest, 0); 2106 return; 2107 } 2108 2109 tcg_out_cmp(s, cond, c1, c2, const_c2, 0, type); 2110 2111 if (have_isel) { 2112 int isel = tcg_to_isel[cond]; 2113 2114 /* Swap the V operands if the operation indicates inversion. */ 2115 if (isel & 1) { 2116 int t = v1; 2117 v1 = v2; 2118 v2 = t; 2119 isel &= ~1; 2120 } 2121 /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand. */ 2122 if (v2 == 0) { 2123 tcg_out_movi(s, type, TCG_REG_R0, 0); 2124 } 2125 tcg_out32(s, isel | TAB(dest, v1, v2)); 2126 } else { 2127 if (dest == v2) { 2128 cond = tcg_invert_cond(cond); 2129 v2 = v1; 2130 } else if (dest != v1) { 2131 if (v1 == 0) { 2132 tcg_out_movi(s, type, dest, 0); 2133 } else { 2134 tcg_out_mov(s, type, dest, v1); 2135 } 2136 } 2137 /* Branch forward over one insn */ 2138 tcg_out_bc(s, cond, 8); 2139 if (v2 == 0) { 2140 tcg_out_movi(s, type, dest, 0); 2141 } else { 2142 tcg_out_mov(s, type, dest, v2); 2143 } 2144 } 2145} 2146 2147static void tcg_out_cntxz(TCGContext *s, TCGType type, uint32_t opc, 2148 TCGArg a0, TCGArg a1, TCGArg a2, bool const_a2) 2149{ 2150 if (const_a2 && a2 == (type == TCG_TYPE_I32 ? 32 : 64)) { 2151 tcg_out32(s, opc | RA(a0) | RS(a1)); 2152 } else { 2153 tcg_out_cmp(s, TCG_COND_EQ, a1, 0, 1, 0, type); 2154 /* Note that the only other valid constant for a2 is 0. */ 2155 if (have_isel) { 2156 tcg_out32(s, opc | RA(TCG_REG_R0) | RS(a1)); 2157 tcg_out32(s, tcg_to_isel[TCG_COND_EQ] | TAB(a0, a2, TCG_REG_R0)); 2158 } else if (!const_a2 && a0 == a2) { 2159 tcg_out_bc(s, TCG_COND_EQ, 8); 2160 tcg_out32(s, opc | RA(a0) | RS(a1)); 2161 } else { 2162 tcg_out32(s, opc | RA(a0) | RS(a1)); 2163 tcg_out_bc(s, TCG_COND_NE, 8); 2164 if (const_a2) { 2165 tcg_out_movi(s, type, a0, 0); 2166 } else { 2167 tcg_out_mov(s, type, a0, a2); 2168 } 2169 } 2170 } 2171} 2172 2173static void tcg_out_cmp2(TCGContext *s, const TCGArg *args, 2174 const int *const_args) 2175{ 2176 static const struct { uint8_t bit1, bit2; } bits[] = { 2177 [TCG_COND_LT ] = { CR_LT, CR_LT }, 2178 [TCG_COND_LE ] = { CR_LT, CR_GT }, 2179 [TCG_COND_GT ] = { CR_GT, CR_GT }, 2180 [TCG_COND_GE ] = { CR_GT, CR_LT }, 2181 [TCG_COND_LTU] = { CR_LT, CR_LT }, 2182 [TCG_COND_LEU] = { CR_LT, CR_GT }, 2183 [TCG_COND_GTU] = { CR_GT, CR_GT }, 2184 [TCG_COND_GEU] = { CR_GT, CR_LT }, 2185 }; 2186 2187 TCGCond cond = args[4], cond2; 2188 TCGArg al, ah, bl, bh; 2189 int blconst, bhconst; 2190 int op, bit1, bit2; 2191 2192 al = args[0]; 2193 ah = args[1]; 2194 bl = args[2]; 2195 bh = args[3]; 2196 blconst = const_args[2]; 2197 bhconst = const_args[3]; 2198 2199 switch (cond) { 2200 case TCG_COND_EQ: 2201 op = CRAND; 2202 goto do_equality; 2203 case TCG_COND_NE: 2204 op = CRNAND; 2205 do_equality: 2206 tcg_out_cmp(s, cond, al, bl, blconst, 6, TCG_TYPE_I32); 2207 tcg_out_cmp(s, cond, ah, bh, bhconst, 7, TCG_TYPE_I32); 2208 tcg_out32(s, op | BT(0, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ)); 2209 break; 2210 2211 case TCG_COND_TSTEQ: 2212 case TCG_COND_TSTNE: 2213 if (blconst) { 2214 tcg_out_andi32(s, TCG_REG_R0, al, bl); 2215 } else { 2216 tcg_out32(s, AND | SAB(al, TCG_REG_R0, bl)); 2217 } 2218 if (bhconst) { 2219 tcg_out_andi32(s, TCG_REG_TMP1, ah, bh); 2220 } else { 2221 tcg_out32(s, AND | SAB(ah, TCG_REG_TMP1, bh)); 2222 } 2223 tcg_out32(s, OR | SAB(TCG_REG_R0, TCG_REG_R0, TCG_REG_TMP1) | 1); 2224 break; 2225 2226 case TCG_COND_LT: 2227 case TCG_COND_LE: 2228 case TCG_COND_GT: 2229 case TCG_COND_GE: 2230 case TCG_COND_LTU: 2231 case TCG_COND_LEU: 2232 case TCG_COND_GTU: 2233 case TCG_COND_GEU: 2234 bit1 = bits[cond].bit1; 2235 bit2 = bits[cond].bit2; 2236 op = (bit1 != bit2 ? CRANDC : CRAND); 2237 cond2 = tcg_unsigned_cond(cond); 2238 2239 tcg_out_cmp(s, cond, ah, bh, bhconst, 6, TCG_TYPE_I32); 2240 tcg_out_cmp(s, cond2, al, bl, blconst, 7, TCG_TYPE_I32); 2241 tcg_out32(s, op | BT(0, CR_EQ) | BA(6, CR_EQ) | BB(7, bit2)); 2242 tcg_out32(s, CROR | BT(0, CR_EQ) | BA(6, bit1) | BB(0, CR_EQ)); 2243 break; 2244 2245 default: 2246 g_assert_not_reached(); 2247 } 2248} 2249 2250static void tcg_out_setcond2(TCGContext *s, const TCGArg *args, 2251 const int *const_args) 2252{ 2253 tcg_out_cmp2(s, args + 1, const_args + 1); 2254 tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(0)); 2255 tcg_out_rlw(s, RLWINM, args[0], TCG_REG_R0, CR_EQ + 0*4 + 1, 31, 31); 2256} 2257 2258static void tcg_out_brcond2(TCGContext *s, const TCGArg *args, 2259 const int *const_args) 2260{ 2261 tcg_out_cmp2(s, args, const_args); 2262 tcg_out_bc_lab(s, TCG_COND_EQ, arg_label(args[5])); 2263} 2264 2265static void tcg_out_mb(TCGContext *s, TCGArg a0) 2266{ 2267 uint32_t insn; 2268 2269 if (a0 & TCG_MO_ST_LD) { 2270 insn = HWSYNC; 2271 } else { 2272 insn = LWSYNC; 2273 } 2274 2275 tcg_out32(s, insn); 2276} 2277 2278static void tcg_out_call_int(TCGContext *s, int lk, 2279 const tcg_insn_unit *target) 2280{ 2281#ifdef _CALL_AIX 2282 /* Look through the descriptor. If the branch is in range, and we 2283 don't have to spend too much effort on building the toc. */ 2284 const void *tgt = ((const void * const *)target)[0]; 2285 uintptr_t toc = ((const uintptr_t *)target)[1]; 2286 intptr_t diff = tcg_pcrel_diff(s, tgt); 2287 2288 if (in_range_b(diff) && toc == (uint32_t)toc) { 2289 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, toc); 2290 tcg_out_b(s, lk, tgt); 2291 } else { 2292 /* Fold the low bits of the constant into the addresses below. */ 2293 intptr_t arg = (intptr_t)target; 2294 int ofs = (int16_t)arg; 2295 2296 if (ofs + 8 < 0x8000) { 2297 arg -= ofs; 2298 } else { 2299 ofs = 0; 2300 } 2301 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, arg); 2302 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_TMP1, ofs); 2303 tcg_out32(s, MTSPR | RA(TCG_REG_R0) | CTR); 2304 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_REG_TMP1, ofs + SZP); 2305 tcg_out32(s, BCCTR | BO_ALWAYS | lk); 2306 } 2307#elif defined(_CALL_ELF) && _CALL_ELF == 2 2308 intptr_t diff; 2309 2310 /* In the ELFv2 ABI, we have to set up r12 to contain the destination 2311 address, which the callee uses to compute its TOC address. */ 2312 /* FIXME: when the branch is in range, we could avoid r12 load if we 2313 knew that the destination uses the same TOC, and what its local 2314 entry point offset is. */ 2315 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R12, (intptr_t)target); 2316 2317 diff = tcg_pcrel_diff(s, target); 2318 if (in_range_b(diff)) { 2319 tcg_out_b(s, lk, target); 2320 } else { 2321 tcg_out32(s, MTSPR | RS(TCG_REG_R12) | CTR); 2322 tcg_out32(s, BCCTR | BO_ALWAYS | lk); 2323 } 2324#else 2325 tcg_out_b(s, lk, target); 2326#endif 2327} 2328 2329static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target, 2330 const TCGHelperInfo *info) 2331{ 2332 tcg_out_call_int(s, LK, target); 2333} 2334 2335static const uint32_t qemu_ldx_opc[(MO_SSIZE + MO_BSWAP) + 1] = { 2336 [MO_UB] = LBZX, 2337 [MO_UW] = LHZX, 2338 [MO_UL] = LWZX, 2339 [MO_UQ] = LDX, 2340 [MO_SW] = LHAX, 2341 [MO_SL] = LWAX, 2342 [MO_BSWAP | MO_UB] = LBZX, 2343 [MO_BSWAP | MO_UW] = LHBRX, 2344 [MO_BSWAP | MO_UL] = LWBRX, 2345 [MO_BSWAP | MO_UQ] = LDBRX, 2346}; 2347 2348static const uint32_t qemu_stx_opc[(MO_SIZE + MO_BSWAP) + 1] = { 2349 [MO_UB] = STBX, 2350 [MO_UW] = STHX, 2351 [MO_UL] = STWX, 2352 [MO_UQ] = STDX, 2353 [MO_BSWAP | MO_UB] = STBX, 2354 [MO_BSWAP | MO_UW] = STHBRX, 2355 [MO_BSWAP | MO_UL] = STWBRX, 2356 [MO_BSWAP | MO_UQ] = STDBRX, 2357}; 2358 2359static TCGReg ldst_ra_gen(TCGContext *s, const TCGLabelQemuLdst *l, int arg) 2360{ 2361 if (arg < 0) { 2362 arg = TCG_REG_TMP1; 2363 } 2364 tcg_out32(s, MFSPR | RT(arg) | LR); 2365 return arg; 2366} 2367 2368/* 2369 * For the purposes of ppc32 sorting 4 input registers into 4 argument 2370 * registers, there is an outside chance we would require 3 temps. 2371 */ 2372static const TCGLdstHelperParam ldst_helper_param = { 2373 .ra_gen = ldst_ra_gen, 2374 .ntmp = 3, 2375 .tmp = { TCG_REG_TMP1, TCG_REG_TMP2, TCG_REG_R0 } 2376}; 2377 2378static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb) 2379{ 2380 MemOp opc = get_memop(lb->oi); 2381 2382 if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) { 2383 return false; 2384 } 2385 2386 tcg_out_ld_helper_args(s, lb, &ldst_helper_param); 2387 tcg_out_call_int(s, LK, qemu_ld_helpers[opc & MO_SIZE]); 2388 tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param); 2389 2390 tcg_out_b(s, 0, lb->raddr); 2391 return true; 2392} 2393 2394static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb) 2395{ 2396 MemOp opc = get_memop(lb->oi); 2397 2398 if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) { 2399 return false; 2400 } 2401 2402 tcg_out_st_helper_args(s, lb, &ldst_helper_param); 2403 tcg_out_call_int(s, LK, qemu_st_helpers[opc & MO_SIZE]); 2404 2405 tcg_out_b(s, 0, lb->raddr); 2406 return true; 2407} 2408 2409typedef struct { 2410 TCGReg base; 2411 TCGReg index; 2412 TCGAtomAlign aa; 2413} HostAddress; 2414 2415bool tcg_target_has_memory_bswap(MemOp memop) 2416{ 2417 TCGAtomAlign aa; 2418 2419 if ((memop & MO_SIZE) <= MO_64) { 2420 return true; 2421 } 2422 2423 /* 2424 * Reject 16-byte memop with 16-byte atomicity, 2425 * but do allow a pair of 64-bit operations. 2426 */ 2427 aa = atom_and_align_for_opc(tcg_ctx, memop, MO_ATOM_IFALIGN, true); 2428 return aa.atom <= MO_64; 2429} 2430 2431/* We expect to use a 16-bit negative offset from ENV. */ 2432#define MIN_TLB_MASK_TABLE_OFS -32768 2433 2434/* 2435 * For system-mode, perform the TLB load and compare. 2436 * For user-mode, perform any required alignment tests. 2437 * In both cases, return a TCGLabelQemuLdst structure if the slow path 2438 * is required and fill in @h with the host address for the fast path. 2439 */ 2440static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h, 2441 TCGReg addrlo, TCGReg addrhi, 2442 MemOpIdx oi, bool is_ld) 2443{ 2444 TCGType addr_type = s->addr_type; 2445 TCGLabelQemuLdst *ldst = NULL; 2446 MemOp opc = get_memop(oi); 2447 MemOp a_bits, s_bits; 2448 2449 /* 2450 * Book II, Section 1.4, Single-Copy Atomicity, specifies: 2451 * 2452 * Before 3.0, "An access that is not atomic is performed as a set of 2453 * smaller disjoint atomic accesses. In general, the number and alignment 2454 * of these accesses are implementation-dependent." Thus MO_ATOM_IFALIGN. 2455 * 2456 * As of 3.0, "the non-atomic access is performed as described in 2457 * the corresponding list", which matches MO_ATOM_SUBALIGN. 2458 */ 2459 s_bits = opc & MO_SIZE; 2460 h->aa = atom_and_align_for_opc(s, opc, 2461 have_isa_3_00 ? MO_ATOM_SUBALIGN 2462 : MO_ATOM_IFALIGN, 2463 s_bits == MO_128); 2464 a_bits = h->aa.align; 2465 2466 if (tcg_use_softmmu) { 2467 int mem_index = get_mmuidx(oi); 2468 int cmp_off = is_ld ? offsetof(CPUTLBEntry, addr_read) 2469 : offsetof(CPUTLBEntry, addr_write); 2470 int fast_off = tlb_mask_table_ofs(s, mem_index); 2471 int mask_off = fast_off + offsetof(CPUTLBDescFast, mask); 2472 int table_off = fast_off + offsetof(CPUTLBDescFast, table); 2473 2474 ldst = new_ldst_label(s); 2475 ldst->is_ld = is_ld; 2476 ldst->oi = oi; 2477 ldst->addrlo_reg = addrlo; 2478 ldst->addrhi_reg = addrhi; 2479 2480 /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx]. */ 2481 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_AREG0, mask_off); 2482 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP2, TCG_AREG0, table_off); 2483 2484 /* Extract the page index, shifted into place for tlb index. */ 2485 if (TCG_TARGET_REG_BITS == 32) { 2486 tcg_out_shri32(s, TCG_REG_R0, addrlo, 2487 s->page_bits - CPU_TLB_ENTRY_BITS); 2488 } else { 2489 tcg_out_shri64(s, TCG_REG_R0, addrlo, 2490 s->page_bits - CPU_TLB_ENTRY_BITS); 2491 } 2492 tcg_out32(s, AND | SAB(TCG_REG_TMP1, TCG_REG_TMP1, TCG_REG_R0)); 2493 2494 /* 2495 * Load the (low part) TLB comparator into TMP2. 2496 * For 64-bit host, always load the entire 64-bit slot for simplicity. 2497 * We will ignore the high bits with tcg_out_cmp(..., addr_type). 2498 */ 2499 if (TCG_TARGET_REG_BITS == 64) { 2500 if (cmp_off == 0) { 2501 tcg_out32(s, LDUX | TAB(TCG_REG_TMP2, 2502 TCG_REG_TMP1, TCG_REG_TMP2)); 2503 } else { 2504 tcg_out32(s, ADD | TAB(TCG_REG_TMP1, 2505 TCG_REG_TMP1, TCG_REG_TMP2)); 2506 tcg_out_ld(s, TCG_TYPE_I64, TCG_REG_TMP2, 2507 TCG_REG_TMP1, cmp_off); 2508 } 2509 } else if (cmp_off == 0 && !HOST_BIG_ENDIAN) { 2510 tcg_out32(s, LWZUX | TAB(TCG_REG_TMP2, 2511 TCG_REG_TMP1, TCG_REG_TMP2)); 2512 } else { 2513 tcg_out32(s, ADD | TAB(TCG_REG_TMP1, TCG_REG_TMP1, TCG_REG_TMP2)); 2514 tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP2, TCG_REG_TMP1, 2515 cmp_off + 4 * HOST_BIG_ENDIAN); 2516 } 2517 2518 /* 2519 * Load the TLB addend for use on the fast path. 2520 * Do this asap to minimize any load use delay. 2521 */ 2522 if (TCG_TARGET_REG_BITS == 64 || addr_type == TCG_TYPE_I32) { 2523 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1, 2524 offsetof(CPUTLBEntry, addend)); 2525 } 2526 2527 /* Clear the non-page, non-alignment bits from the address in R0. */ 2528 if (TCG_TARGET_REG_BITS == 32) { 2529 /* 2530 * We don't support unaligned accesses on 32-bits. 2531 * Preserve the bottom bits and thus trigger a comparison 2532 * failure on unaligned accesses. 2533 */ 2534 if (a_bits < s_bits) { 2535 a_bits = s_bits; 2536 } 2537 tcg_out_rlw(s, RLWINM, TCG_REG_R0, addrlo, 0, 2538 (32 - a_bits) & 31, 31 - s->page_bits); 2539 } else { 2540 TCGReg t = addrlo; 2541 2542 /* 2543 * If the access is unaligned, we need to make sure we fail if we 2544 * cross a page boundary. The trick is to add the access size-1 2545 * to the address before masking the low bits. That will make the 2546 * address overflow to the next page if we cross a page boundary, 2547 * which will then force a mismatch of the TLB compare. 2548 */ 2549 if (a_bits < s_bits) { 2550 unsigned a_mask = (1 << a_bits) - 1; 2551 unsigned s_mask = (1 << s_bits) - 1; 2552 tcg_out32(s, ADDI | TAI(TCG_REG_R0, t, s_mask - a_mask)); 2553 t = TCG_REG_R0; 2554 } 2555 2556 /* Mask the address for the requested alignment. */ 2557 if (addr_type == TCG_TYPE_I32) { 2558 tcg_out_rlw(s, RLWINM, TCG_REG_R0, t, 0, 2559 (32 - a_bits) & 31, 31 - s->page_bits); 2560 } else if (a_bits == 0) { 2561 tcg_out_rld(s, RLDICR, TCG_REG_R0, t, 0, 63 - s->page_bits); 2562 } else { 2563 tcg_out_rld(s, RLDICL, TCG_REG_R0, t, 2564 64 - s->page_bits, s->page_bits - a_bits); 2565 tcg_out_rld(s, RLDICL, TCG_REG_R0, TCG_REG_R0, s->page_bits, 0); 2566 } 2567 } 2568 2569 if (TCG_TARGET_REG_BITS == 32 && addr_type != TCG_TYPE_I32) { 2570 /* Low part comparison into cr7. */ 2571 tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP2, 2572 0, 7, TCG_TYPE_I32); 2573 2574 /* Load the high part TLB comparator into TMP2. */ 2575 tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP2, TCG_REG_TMP1, 2576 cmp_off + 4 * !HOST_BIG_ENDIAN); 2577 2578 /* Load addend, deferred for this case. */ 2579 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1, 2580 offsetof(CPUTLBEntry, addend)); 2581 2582 /* High part comparison into cr6. */ 2583 tcg_out_cmp(s, TCG_COND_EQ, addrhi, TCG_REG_TMP2, 2584 0, 6, TCG_TYPE_I32); 2585 2586 /* Combine comparisons into cr0. */ 2587 tcg_out32(s, CRAND | BT(0, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ)); 2588 } else { 2589 /* Full comparison into cr0. */ 2590 tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP2, 2591 0, 0, addr_type); 2592 } 2593 2594 /* Load a pointer into the current opcode w/conditional branch-link. */ 2595 ldst->label_ptr[0] = s->code_ptr; 2596 tcg_out_bc(s, TCG_COND_NE, LK); 2597 2598 h->base = TCG_REG_TMP1; 2599 } else { 2600 if (a_bits) { 2601 ldst = new_ldst_label(s); 2602 ldst->is_ld = is_ld; 2603 ldst->oi = oi; 2604 ldst->addrlo_reg = addrlo; 2605 ldst->addrhi_reg = addrhi; 2606 2607 /* We are expecting a_bits to max out at 7, much lower than ANDI. */ 2608 tcg_debug_assert(a_bits < 16); 2609 tcg_out32(s, ANDI | SAI(addrlo, TCG_REG_R0, (1 << a_bits) - 1)); 2610 2611 ldst->label_ptr[0] = s->code_ptr; 2612 tcg_out32(s, BC | BI(0, CR_EQ) | BO_COND_FALSE | LK); 2613 } 2614 2615 h->base = guest_base ? TCG_GUEST_BASE_REG : 0; 2616 } 2617 2618 if (TCG_TARGET_REG_BITS == 64 && addr_type == TCG_TYPE_I32) { 2619 /* Zero-extend the guest address for use in the host address. */ 2620 tcg_out_ext32u(s, TCG_REG_TMP2, addrlo); 2621 h->index = TCG_REG_TMP2; 2622 } else { 2623 h->index = addrlo; 2624 } 2625 2626 return ldst; 2627} 2628 2629static void tcg_out_qemu_ld(TCGContext *s, TCGReg datalo, TCGReg datahi, 2630 TCGReg addrlo, TCGReg addrhi, 2631 MemOpIdx oi, TCGType data_type) 2632{ 2633 MemOp opc = get_memop(oi); 2634 TCGLabelQemuLdst *ldst; 2635 HostAddress h; 2636 2637 ldst = prepare_host_addr(s, &h, addrlo, addrhi, oi, true); 2638 2639 if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) { 2640 if (opc & MO_BSWAP) { 2641 tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4)); 2642 tcg_out32(s, LWBRX | TAB(datalo, h.base, h.index)); 2643 tcg_out32(s, LWBRX | TAB(datahi, h.base, TCG_REG_R0)); 2644 } else if (h.base != 0) { 2645 tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4)); 2646 tcg_out32(s, LWZX | TAB(datahi, h.base, h.index)); 2647 tcg_out32(s, LWZX | TAB(datalo, h.base, TCG_REG_R0)); 2648 } else if (h.index == datahi) { 2649 tcg_out32(s, LWZ | TAI(datalo, h.index, 4)); 2650 tcg_out32(s, LWZ | TAI(datahi, h.index, 0)); 2651 } else { 2652 tcg_out32(s, LWZ | TAI(datahi, h.index, 0)); 2653 tcg_out32(s, LWZ | TAI(datalo, h.index, 4)); 2654 } 2655 } else { 2656 uint32_t insn = qemu_ldx_opc[opc & (MO_BSWAP | MO_SSIZE)]; 2657 if (!have_isa_2_06 && insn == LDBRX) { 2658 tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4)); 2659 tcg_out32(s, LWBRX | TAB(datalo, h.base, h.index)); 2660 tcg_out32(s, LWBRX | TAB(TCG_REG_R0, h.base, TCG_REG_R0)); 2661 tcg_out_rld(s, RLDIMI, datalo, TCG_REG_R0, 32, 0); 2662 } else if (insn) { 2663 tcg_out32(s, insn | TAB(datalo, h.base, h.index)); 2664 } else { 2665 insn = qemu_ldx_opc[opc & (MO_SIZE | MO_BSWAP)]; 2666 tcg_out32(s, insn | TAB(datalo, h.base, h.index)); 2667 tcg_out_movext(s, TCG_TYPE_REG, datalo, 2668 TCG_TYPE_REG, opc & MO_SSIZE, datalo); 2669 } 2670 } 2671 2672 if (ldst) { 2673 ldst->type = data_type; 2674 ldst->datalo_reg = datalo; 2675 ldst->datahi_reg = datahi; 2676 ldst->raddr = tcg_splitwx_to_rx(s->code_ptr); 2677 } 2678} 2679 2680static void tcg_out_qemu_st(TCGContext *s, TCGReg datalo, TCGReg datahi, 2681 TCGReg addrlo, TCGReg addrhi, 2682 MemOpIdx oi, TCGType data_type) 2683{ 2684 MemOp opc = get_memop(oi); 2685 TCGLabelQemuLdst *ldst; 2686 HostAddress h; 2687 2688 ldst = prepare_host_addr(s, &h, addrlo, addrhi, oi, false); 2689 2690 if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) { 2691 if (opc & MO_BSWAP) { 2692 tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4)); 2693 tcg_out32(s, STWBRX | SAB(datalo, h.base, h.index)); 2694 tcg_out32(s, STWBRX | SAB(datahi, h.base, TCG_REG_R0)); 2695 } else if (h.base != 0) { 2696 tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4)); 2697 tcg_out32(s, STWX | SAB(datahi, h.base, h.index)); 2698 tcg_out32(s, STWX | SAB(datalo, h.base, TCG_REG_R0)); 2699 } else { 2700 tcg_out32(s, STW | TAI(datahi, h.index, 0)); 2701 tcg_out32(s, STW | TAI(datalo, h.index, 4)); 2702 } 2703 } else { 2704 uint32_t insn = qemu_stx_opc[opc & (MO_BSWAP | MO_SIZE)]; 2705 if (!have_isa_2_06 && insn == STDBRX) { 2706 tcg_out32(s, STWBRX | SAB(datalo, h.base, h.index)); 2707 tcg_out32(s, ADDI | TAI(TCG_REG_TMP2, h.index, 4)); 2708 tcg_out_shri64(s, TCG_REG_R0, datalo, 32); 2709 tcg_out32(s, STWBRX | SAB(TCG_REG_R0, h.base, TCG_REG_TMP2)); 2710 } else { 2711 tcg_out32(s, insn | SAB(datalo, h.base, h.index)); 2712 } 2713 } 2714 2715 if (ldst) { 2716 ldst->type = data_type; 2717 ldst->datalo_reg = datalo; 2718 ldst->datahi_reg = datahi; 2719 ldst->raddr = tcg_splitwx_to_rx(s->code_ptr); 2720 } 2721} 2722 2723static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi, 2724 TCGReg addr_reg, MemOpIdx oi, bool is_ld) 2725{ 2726 TCGLabelQemuLdst *ldst; 2727 HostAddress h; 2728 bool need_bswap; 2729 uint32_t insn; 2730 TCGReg index; 2731 2732 ldst = prepare_host_addr(s, &h, addr_reg, -1, oi, is_ld); 2733 2734 /* Compose the final address, as LQ/STQ have no indexing. */ 2735 index = h.index; 2736 if (h.base != 0) { 2737 index = TCG_REG_TMP1; 2738 tcg_out32(s, ADD | TAB(index, h.base, h.index)); 2739 } 2740 need_bswap = get_memop(oi) & MO_BSWAP; 2741 2742 if (h.aa.atom == MO_128) { 2743 tcg_debug_assert(!need_bswap); 2744 tcg_debug_assert(datalo & 1); 2745 tcg_debug_assert(datahi == datalo - 1); 2746 tcg_debug_assert(!is_ld || datahi != index); 2747 insn = is_ld ? LQ : STQ; 2748 tcg_out32(s, insn | TAI(datahi, index, 0)); 2749 } else { 2750 TCGReg d1, d2; 2751 2752 if (HOST_BIG_ENDIAN ^ need_bswap) { 2753 d1 = datahi, d2 = datalo; 2754 } else { 2755 d1 = datalo, d2 = datahi; 2756 } 2757 2758 if (need_bswap) { 2759 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, 8); 2760 insn = is_ld ? LDBRX : STDBRX; 2761 tcg_out32(s, insn | TAB(d1, 0, index)); 2762 tcg_out32(s, insn | TAB(d2, index, TCG_REG_R0)); 2763 } else { 2764 insn = is_ld ? LD : STD; 2765 tcg_out32(s, insn | TAI(d1, index, 0)); 2766 tcg_out32(s, insn | TAI(d2, index, 8)); 2767 } 2768 } 2769 2770 if (ldst) { 2771 ldst->type = TCG_TYPE_I128; 2772 ldst->datalo_reg = datalo; 2773 ldst->datahi_reg = datahi; 2774 ldst->raddr = tcg_splitwx_to_rx(s->code_ptr); 2775 } 2776} 2777 2778static void tcg_out_nop_fill(tcg_insn_unit *p, int count) 2779{ 2780 int i; 2781 for (i = 0; i < count; ++i) { 2782 p[i] = NOP; 2783 } 2784} 2785 2786/* Parameters for function call generation, used in tcg.c. */ 2787#define TCG_TARGET_STACK_ALIGN 16 2788 2789#ifdef _CALL_AIX 2790# define LINK_AREA_SIZE (6 * SZR) 2791# define LR_OFFSET (1 * SZR) 2792# define TCG_TARGET_CALL_STACK_OFFSET (LINK_AREA_SIZE + 8 * SZR) 2793#elif defined(_CALL_DARWIN) 2794# define LINK_AREA_SIZE (6 * SZR) 2795# define LR_OFFSET (2 * SZR) 2796#elif TCG_TARGET_REG_BITS == 64 2797# if defined(_CALL_ELF) && _CALL_ELF == 2 2798# define LINK_AREA_SIZE (4 * SZR) 2799# define LR_OFFSET (1 * SZR) 2800# endif 2801#else /* TCG_TARGET_REG_BITS == 32 */ 2802# if defined(_CALL_SYSV) 2803# define LINK_AREA_SIZE (2 * SZR) 2804# define LR_OFFSET (1 * SZR) 2805# endif 2806#endif 2807#ifndef LR_OFFSET 2808# error "Unhandled abi" 2809#endif 2810#ifndef TCG_TARGET_CALL_STACK_OFFSET 2811# define TCG_TARGET_CALL_STACK_OFFSET LINK_AREA_SIZE 2812#endif 2813 2814#define CPU_TEMP_BUF_SIZE (CPU_TEMP_BUF_NLONGS * (int)sizeof(long)) 2815#define REG_SAVE_SIZE ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR) 2816 2817#define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET \ 2818 + TCG_STATIC_CALL_ARGS_SIZE \ 2819 + CPU_TEMP_BUF_SIZE \ 2820 + REG_SAVE_SIZE \ 2821 + TCG_TARGET_STACK_ALIGN - 1) \ 2822 & -TCG_TARGET_STACK_ALIGN) 2823 2824#define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE) 2825 2826static void tcg_target_qemu_prologue(TCGContext *s) 2827{ 2828 int i; 2829 2830#ifdef _CALL_AIX 2831 const void **desc = (const void **)s->code_ptr; 2832 desc[0] = tcg_splitwx_to_rx(desc + 2); /* entry point */ 2833 desc[1] = 0; /* environment pointer */ 2834 s->code_ptr = (void *)(desc + 2); /* skip over descriptor */ 2835#endif 2836 2837 tcg_set_frame(s, TCG_REG_CALL_STACK, REG_SAVE_BOT - CPU_TEMP_BUF_SIZE, 2838 CPU_TEMP_BUF_SIZE); 2839 2840 /* Prologue */ 2841 tcg_out32(s, MFSPR | RT(TCG_REG_R0) | LR); 2842 tcg_out32(s, (SZR == 8 ? STDU : STWU) 2843 | SAI(TCG_REG_R1, TCG_REG_R1, -FRAME_SIZE)); 2844 2845 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) { 2846 tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i], 2847 TCG_REG_R1, REG_SAVE_BOT + i * SZR); 2848 } 2849 tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET); 2850 2851 if (!tcg_use_softmmu && guest_base) { 2852 tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true); 2853 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG); 2854 } 2855 2856 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]); 2857 tcg_out32(s, MTSPR | RS(tcg_target_call_iarg_regs[1]) | CTR); 2858 tcg_out32(s, BCCTR | BO_ALWAYS); 2859 2860 /* Epilogue */ 2861 tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr); 2862 2863 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET); 2864 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) { 2865 tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i], 2866 TCG_REG_R1, REG_SAVE_BOT + i * SZR); 2867 } 2868 tcg_out32(s, MTSPR | RS(TCG_REG_R0) | LR); 2869 tcg_out32(s, ADDI | TAI(TCG_REG_R1, TCG_REG_R1, FRAME_SIZE)); 2870 tcg_out32(s, BCLR | BO_ALWAYS); 2871} 2872 2873static void tcg_out_tb_start(TCGContext *s) 2874{ 2875 /* Load TCG_REG_TB. */ 2876 if (USE_REG_TB) { 2877 if (have_isa_3_00) { 2878 /* lnia REG_TB */ 2879 tcg_out_addpcis(s, TCG_REG_TB, 0); 2880 } else { 2881 /* bcl 20,31,$+4 (preferred form for getting nia) */ 2882 tcg_out32(s, BC | BO_ALWAYS | BI(7, CR_SO) | 0x4 | LK); 2883 tcg_out32(s, MFSPR | RT(TCG_REG_TB) | LR); 2884 } 2885 } 2886} 2887 2888static void tcg_out_exit_tb(TCGContext *s, uintptr_t arg) 2889{ 2890 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, arg); 2891 tcg_out_b(s, 0, tcg_code_gen_epilogue); 2892} 2893 2894static void tcg_out_goto_tb(TCGContext *s, int which) 2895{ 2896 uintptr_t ptr = get_jmp_target_addr(s, which); 2897 int16_t lo; 2898 2899 /* Direct branch will be patched by tb_target_set_jmp_target. */ 2900 set_jmp_insn_offset(s, which); 2901 tcg_out32(s, NOP); 2902 2903 /* When branch is out of range, fall through to indirect. */ 2904 if (USE_REG_TB) { 2905 ptrdiff_t offset = ppc_tbrel_diff(s, (void *)ptr); 2906 tcg_out_mem_long(s, LD, LDX, TCG_REG_TMP1, TCG_REG_TB, offset); 2907 } else if (have_isa_3_10) { 2908 ptrdiff_t offset = tcg_pcrel_diff_for_prefix(s, (void *)ptr); 2909 tcg_out_8ls_d(s, PLD, TCG_REG_TMP1, 0, offset, 1); 2910 } else if (have_isa_3_00) { 2911 ptrdiff_t offset = tcg_pcrel_diff(s, (void *)ptr) - 4; 2912 lo = offset; 2913 tcg_out_addpcis(s, TCG_REG_TMP1, offset - lo); 2914 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1, lo); 2915 } else { 2916 lo = ptr; 2917 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, ptr - lo); 2918 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1, lo); 2919 } 2920 2921 tcg_out32(s, MTSPR | RS(TCG_REG_TMP1) | CTR); 2922 tcg_out32(s, BCCTR | BO_ALWAYS); 2923 set_jmp_reset_offset(s, which); 2924} 2925 2926void tb_target_set_jmp_target(const TranslationBlock *tb, int n, 2927 uintptr_t jmp_rx, uintptr_t jmp_rw) 2928{ 2929 uintptr_t addr = tb->jmp_target_addr[n]; 2930 intptr_t diff = addr - jmp_rx; 2931 tcg_insn_unit insn; 2932 2933 if (in_range_b(diff)) { 2934 insn = B | (diff & 0x3fffffc); 2935 } else { 2936 insn = NOP; 2937 } 2938 2939 qatomic_set((uint32_t *)jmp_rw, insn); 2940 flush_idcache_range(jmp_rx, jmp_rw, 4); 2941} 2942 2943static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type, 2944 const TCGArg args[TCG_MAX_OP_ARGS], 2945 const int const_args[TCG_MAX_OP_ARGS]) 2946{ 2947 TCGArg a0, a1, a2; 2948 2949 switch (opc) { 2950 case INDEX_op_goto_ptr: 2951 tcg_out32(s, MTSPR | RS(args[0]) | CTR); 2952 tcg_out32(s, ADDI | TAI(TCG_REG_R3, 0, 0)); 2953 tcg_out32(s, BCCTR | BO_ALWAYS); 2954 break; 2955 case INDEX_op_br: 2956 { 2957 TCGLabel *l = arg_label(args[0]); 2958 uint32_t insn = B; 2959 2960 if (l->has_value) { 2961 insn |= reloc_pc24_val(tcg_splitwx_to_rx(s->code_ptr), 2962 l->u.value_ptr); 2963 } else { 2964 tcg_out_reloc(s, s->code_ptr, R_PPC_REL24, l, 0); 2965 } 2966 tcg_out32(s, insn); 2967 } 2968 break; 2969 case INDEX_op_ld8u_i32: 2970 case INDEX_op_ld8u_i64: 2971 tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]); 2972 break; 2973 case INDEX_op_ld8s_i32: 2974 case INDEX_op_ld8s_i64: 2975 tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]); 2976 tcg_out_ext8s(s, TCG_TYPE_REG, args[0], args[0]); 2977 break; 2978 case INDEX_op_ld16u_i32: 2979 case INDEX_op_ld16u_i64: 2980 tcg_out_mem_long(s, LHZ, LHZX, args[0], args[1], args[2]); 2981 break; 2982 case INDEX_op_ld16s_i32: 2983 case INDEX_op_ld16s_i64: 2984 tcg_out_mem_long(s, LHA, LHAX, args[0], args[1], args[2]); 2985 break; 2986 case INDEX_op_ld_i32: 2987 case INDEX_op_ld32u_i64: 2988 tcg_out_mem_long(s, LWZ, LWZX, args[0], args[1], args[2]); 2989 break; 2990 case INDEX_op_ld32s_i64: 2991 tcg_out_mem_long(s, LWA, LWAX, args[0], args[1], args[2]); 2992 break; 2993 case INDEX_op_ld_i64: 2994 tcg_out_mem_long(s, LD, LDX, args[0], args[1], args[2]); 2995 break; 2996 case INDEX_op_st8_i32: 2997 case INDEX_op_st8_i64: 2998 tcg_out_mem_long(s, STB, STBX, args[0], args[1], args[2]); 2999 break; 3000 case INDEX_op_st16_i32: 3001 case INDEX_op_st16_i64: 3002 tcg_out_mem_long(s, STH, STHX, args[0], args[1], args[2]); 3003 break; 3004 case INDEX_op_st_i32: 3005 case INDEX_op_st32_i64: 3006 tcg_out_mem_long(s, STW, STWX, args[0], args[1], args[2]); 3007 break; 3008 case INDEX_op_st_i64: 3009 tcg_out_mem_long(s, STD, STDX, args[0], args[1], args[2]); 3010 break; 3011 3012 case INDEX_op_add_i32: 3013 a0 = args[0], a1 = args[1], a2 = args[2]; 3014 if (const_args[2]) { 3015 do_addi_32: 3016 tcg_out_mem_long(s, ADDI, ADD, a0, a1, (int32_t)a2); 3017 } else { 3018 tcg_out32(s, ADD | TAB(a0, a1, a2)); 3019 } 3020 break; 3021 case INDEX_op_sub_i32: 3022 a0 = args[0], a1 = args[1], a2 = args[2]; 3023 if (const_args[1]) { 3024 if (const_args[2]) { 3025 tcg_out_movi(s, TCG_TYPE_I32, a0, a1 - a2); 3026 } else { 3027 tcg_out32(s, SUBFIC | TAI(a0, a2, a1)); 3028 } 3029 } else if (const_args[2]) { 3030 a2 = -a2; 3031 goto do_addi_32; 3032 } else { 3033 tcg_out32(s, SUBF | TAB(a0, a2, a1)); 3034 } 3035 break; 3036 3037 case INDEX_op_and_i32: 3038 a0 = args[0], a1 = args[1], a2 = args[2]; 3039 if (const_args[2]) { 3040 tcg_out_andi32(s, a0, a1, a2); 3041 } else { 3042 tcg_out32(s, AND | SAB(a1, a0, a2)); 3043 } 3044 break; 3045 case INDEX_op_and_i64: 3046 a0 = args[0], a1 = args[1], a2 = args[2]; 3047 if (const_args[2]) { 3048 tcg_out_andi64(s, a0, a1, a2); 3049 } else { 3050 tcg_out32(s, AND | SAB(a1, a0, a2)); 3051 } 3052 break; 3053 case INDEX_op_or_i64: 3054 case INDEX_op_or_i32: 3055 a0 = args[0], a1 = args[1], a2 = args[2]; 3056 if (const_args[2]) { 3057 tcg_out_ori32(s, a0, a1, a2); 3058 } else { 3059 tcg_out32(s, OR | SAB(a1, a0, a2)); 3060 } 3061 break; 3062 case INDEX_op_xor_i64: 3063 case INDEX_op_xor_i32: 3064 a0 = args[0], a1 = args[1], a2 = args[2]; 3065 if (const_args[2]) { 3066 tcg_out_xori32(s, a0, a1, a2); 3067 } else { 3068 tcg_out32(s, XOR | SAB(a1, a0, a2)); 3069 } 3070 break; 3071 case INDEX_op_andc_i32: 3072 a0 = args[0], a1 = args[1], a2 = args[2]; 3073 if (const_args[2]) { 3074 tcg_out_andi32(s, a0, a1, ~a2); 3075 } else { 3076 tcg_out32(s, ANDC | SAB(a1, a0, a2)); 3077 } 3078 break; 3079 case INDEX_op_andc_i64: 3080 a0 = args[0], a1 = args[1], a2 = args[2]; 3081 if (const_args[2]) { 3082 tcg_out_andi64(s, a0, a1, ~a2); 3083 } else { 3084 tcg_out32(s, ANDC | SAB(a1, a0, a2)); 3085 } 3086 break; 3087 case INDEX_op_orc_i32: 3088 if (const_args[2]) { 3089 tcg_out_ori32(s, args[0], args[1], ~args[2]); 3090 break; 3091 } 3092 /* FALLTHRU */ 3093 case INDEX_op_orc_i64: 3094 tcg_out32(s, ORC | SAB(args[1], args[0], args[2])); 3095 break; 3096 case INDEX_op_eqv_i32: 3097 if (const_args[2]) { 3098 tcg_out_xori32(s, args[0], args[1], ~args[2]); 3099 break; 3100 } 3101 /* FALLTHRU */ 3102 case INDEX_op_eqv_i64: 3103 tcg_out32(s, EQV | SAB(args[1], args[0], args[2])); 3104 break; 3105 case INDEX_op_nand_i32: 3106 case INDEX_op_nand_i64: 3107 tcg_out32(s, NAND | SAB(args[1], args[0], args[2])); 3108 break; 3109 case INDEX_op_nor_i32: 3110 case INDEX_op_nor_i64: 3111 tcg_out32(s, NOR | SAB(args[1], args[0], args[2])); 3112 break; 3113 3114 case INDEX_op_clz_i32: 3115 tcg_out_cntxz(s, TCG_TYPE_I32, CNTLZW, args[0], args[1], 3116 args[2], const_args[2]); 3117 break; 3118 case INDEX_op_ctz_i32: 3119 tcg_out_cntxz(s, TCG_TYPE_I32, CNTTZW, args[0], args[1], 3120 args[2], const_args[2]); 3121 break; 3122 case INDEX_op_ctpop_i32: 3123 tcg_out32(s, CNTPOPW | SAB(args[1], args[0], 0)); 3124 break; 3125 3126 case INDEX_op_clz_i64: 3127 tcg_out_cntxz(s, TCG_TYPE_I64, CNTLZD, args[0], args[1], 3128 args[2], const_args[2]); 3129 break; 3130 case INDEX_op_ctz_i64: 3131 tcg_out_cntxz(s, TCG_TYPE_I64, CNTTZD, args[0], args[1], 3132 args[2], const_args[2]); 3133 break; 3134 case INDEX_op_ctpop_i64: 3135 tcg_out32(s, CNTPOPD | SAB(args[1], args[0], 0)); 3136 break; 3137 3138 case INDEX_op_mul_i32: 3139 a0 = args[0], a1 = args[1], a2 = args[2]; 3140 if (const_args[2]) { 3141 tcg_out32(s, MULLI | TAI(a0, a1, a2)); 3142 } else { 3143 tcg_out32(s, MULLW | TAB(a0, a1, a2)); 3144 } 3145 break; 3146 3147 case INDEX_op_div_i32: 3148 tcg_out32(s, DIVW | TAB(args[0], args[1], args[2])); 3149 break; 3150 3151 case INDEX_op_divu_i32: 3152 tcg_out32(s, DIVWU | TAB(args[0], args[1], args[2])); 3153 break; 3154 3155 case INDEX_op_rem_i32: 3156 tcg_out32(s, MODSW | TAB(args[0], args[1], args[2])); 3157 break; 3158 3159 case INDEX_op_remu_i32: 3160 tcg_out32(s, MODUW | TAB(args[0], args[1], args[2])); 3161 break; 3162 3163 case INDEX_op_shl_i32: 3164 if (const_args[2]) { 3165 /* Limit immediate shift count lest we create an illegal insn. */ 3166 tcg_out_shli32(s, args[0], args[1], args[2] & 31); 3167 } else { 3168 tcg_out32(s, SLW | SAB(args[1], args[0], args[2])); 3169 } 3170 break; 3171 case INDEX_op_shr_i32: 3172 if (const_args[2]) { 3173 /* Limit immediate shift count lest we create an illegal insn. */ 3174 tcg_out_shri32(s, args[0], args[1], args[2] & 31); 3175 } else { 3176 tcg_out32(s, SRW | SAB(args[1], args[0], args[2])); 3177 } 3178 break; 3179 case INDEX_op_sar_i32: 3180 if (const_args[2]) { 3181 tcg_out_sari32(s, args[0], args[1], args[2]); 3182 } else { 3183 tcg_out32(s, SRAW | SAB(args[1], args[0], args[2])); 3184 } 3185 break; 3186 case INDEX_op_rotl_i32: 3187 if (const_args[2]) { 3188 tcg_out_rlw(s, RLWINM, args[0], args[1], args[2], 0, 31); 3189 } else { 3190 tcg_out32(s, RLWNM | SAB(args[1], args[0], args[2]) 3191 | MB(0) | ME(31)); 3192 } 3193 break; 3194 case INDEX_op_rotr_i32: 3195 if (const_args[2]) { 3196 tcg_out_rlw(s, RLWINM, args[0], args[1], 32 - args[2], 0, 31); 3197 } else { 3198 tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 32)); 3199 tcg_out32(s, RLWNM | SAB(args[1], args[0], TCG_REG_R0) 3200 | MB(0) | ME(31)); 3201 } 3202 break; 3203 3204 case INDEX_op_brcond_i32: 3205 tcg_out_brcond(s, args[2], args[0], args[1], const_args[1], 3206 arg_label(args[3]), TCG_TYPE_I32); 3207 break; 3208 case INDEX_op_brcond_i64: 3209 tcg_out_brcond(s, args[2], args[0], args[1], const_args[1], 3210 arg_label(args[3]), TCG_TYPE_I64); 3211 break; 3212 case INDEX_op_brcond2_i32: 3213 tcg_out_brcond2(s, args, const_args); 3214 break; 3215 3216 case INDEX_op_neg_i32: 3217 case INDEX_op_neg_i64: 3218 tcg_out32(s, NEG | RT(args[0]) | RA(args[1])); 3219 break; 3220 3221 case INDEX_op_not_i32: 3222 case INDEX_op_not_i64: 3223 tcg_out32(s, NOR | SAB(args[1], args[0], args[1])); 3224 break; 3225 3226 case INDEX_op_add_i64: 3227 a0 = args[0], a1 = args[1], a2 = args[2]; 3228 if (const_args[2]) { 3229 do_addi_64: 3230 tcg_out_mem_long(s, ADDI, ADD, a0, a1, a2); 3231 } else { 3232 tcg_out32(s, ADD | TAB(a0, a1, a2)); 3233 } 3234 break; 3235 case INDEX_op_sub_i64: 3236 a0 = args[0], a1 = args[1], a2 = args[2]; 3237 if (const_args[1]) { 3238 if (const_args[2]) { 3239 tcg_out_movi(s, TCG_TYPE_I64, a0, a1 - a2); 3240 } else { 3241 tcg_out32(s, SUBFIC | TAI(a0, a2, a1)); 3242 } 3243 } else if (const_args[2]) { 3244 a2 = -a2; 3245 goto do_addi_64; 3246 } else { 3247 tcg_out32(s, SUBF | TAB(a0, a2, a1)); 3248 } 3249 break; 3250 3251 case INDEX_op_shl_i64: 3252 if (const_args[2]) { 3253 /* Limit immediate shift count lest we create an illegal insn. */ 3254 tcg_out_shli64(s, args[0], args[1], args[2] & 63); 3255 } else { 3256 tcg_out32(s, SLD | SAB(args[1], args[0], args[2])); 3257 } 3258 break; 3259 case INDEX_op_shr_i64: 3260 if (const_args[2]) { 3261 /* Limit immediate shift count lest we create an illegal insn. */ 3262 tcg_out_shri64(s, args[0], args[1], args[2] & 63); 3263 } else { 3264 tcg_out32(s, SRD | SAB(args[1], args[0], args[2])); 3265 } 3266 break; 3267 case INDEX_op_sar_i64: 3268 if (const_args[2]) { 3269 tcg_out_sari64(s, args[0], args[1], args[2]); 3270 } else { 3271 tcg_out32(s, SRAD | SAB(args[1], args[0], args[2])); 3272 } 3273 break; 3274 case INDEX_op_rotl_i64: 3275 if (const_args[2]) { 3276 tcg_out_rld(s, RLDICL, args[0], args[1], args[2], 0); 3277 } else { 3278 tcg_out32(s, RLDCL | SAB(args[1], args[0], args[2]) | MB64(0)); 3279 } 3280 break; 3281 case INDEX_op_rotr_i64: 3282 if (const_args[2]) { 3283 tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 0); 3284 } else { 3285 tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 64)); 3286 tcg_out32(s, RLDCL | SAB(args[1], args[0], TCG_REG_R0) | MB64(0)); 3287 } 3288 break; 3289 3290 case INDEX_op_mul_i64: 3291 a0 = args[0], a1 = args[1], a2 = args[2]; 3292 if (const_args[2]) { 3293 tcg_out32(s, MULLI | TAI(a0, a1, a2)); 3294 } else { 3295 tcg_out32(s, MULLD | TAB(a0, a1, a2)); 3296 } 3297 break; 3298 case INDEX_op_div_i64: 3299 tcg_out32(s, DIVD | TAB(args[0], args[1], args[2])); 3300 break; 3301 case INDEX_op_divu_i64: 3302 tcg_out32(s, DIVDU | TAB(args[0], args[1], args[2])); 3303 break; 3304 case INDEX_op_rem_i64: 3305 tcg_out32(s, MODSD | TAB(args[0], args[1], args[2])); 3306 break; 3307 case INDEX_op_remu_i64: 3308 tcg_out32(s, MODUD | TAB(args[0], args[1], args[2])); 3309 break; 3310 3311 case INDEX_op_qemu_ld_a64_i32: 3312 if (TCG_TARGET_REG_BITS == 32) { 3313 tcg_out_qemu_ld(s, args[0], -1, args[1], args[2], 3314 args[3], TCG_TYPE_I32); 3315 break; 3316 } 3317 /* fall through */ 3318 case INDEX_op_qemu_ld_a32_i32: 3319 tcg_out_qemu_ld(s, args[0], -1, args[1], -1, args[2], TCG_TYPE_I32); 3320 break; 3321 case INDEX_op_qemu_ld_a32_i64: 3322 if (TCG_TARGET_REG_BITS == 64) { 3323 tcg_out_qemu_ld(s, args[0], -1, args[1], -1, 3324 args[2], TCG_TYPE_I64); 3325 } else { 3326 tcg_out_qemu_ld(s, args[0], args[1], args[2], -1, 3327 args[3], TCG_TYPE_I64); 3328 } 3329 break; 3330 case INDEX_op_qemu_ld_a64_i64: 3331 if (TCG_TARGET_REG_BITS == 64) { 3332 tcg_out_qemu_ld(s, args[0], -1, args[1], -1, 3333 args[2], TCG_TYPE_I64); 3334 } else { 3335 tcg_out_qemu_ld(s, args[0], args[1], args[2], args[3], 3336 args[4], TCG_TYPE_I64); 3337 } 3338 break; 3339 case INDEX_op_qemu_ld_a32_i128: 3340 case INDEX_op_qemu_ld_a64_i128: 3341 tcg_debug_assert(TCG_TARGET_REG_BITS == 64); 3342 tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], true); 3343 break; 3344 3345 case INDEX_op_qemu_st_a64_i32: 3346 if (TCG_TARGET_REG_BITS == 32) { 3347 tcg_out_qemu_st(s, args[0], -1, args[1], args[2], 3348 args[3], TCG_TYPE_I32); 3349 break; 3350 } 3351 /* fall through */ 3352 case INDEX_op_qemu_st_a32_i32: 3353 tcg_out_qemu_st(s, args[0], -1, args[1], -1, args[2], TCG_TYPE_I32); 3354 break; 3355 case INDEX_op_qemu_st_a32_i64: 3356 if (TCG_TARGET_REG_BITS == 64) { 3357 tcg_out_qemu_st(s, args[0], -1, args[1], -1, 3358 args[2], TCG_TYPE_I64); 3359 } else { 3360 tcg_out_qemu_st(s, args[0], args[1], args[2], -1, 3361 args[3], TCG_TYPE_I64); 3362 } 3363 break; 3364 case INDEX_op_qemu_st_a64_i64: 3365 if (TCG_TARGET_REG_BITS == 64) { 3366 tcg_out_qemu_st(s, args[0], -1, args[1], -1, 3367 args[2], TCG_TYPE_I64); 3368 } else { 3369 tcg_out_qemu_st(s, args[0], args[1], args[2], args[3], 3370 args[4], TCG_TYPE_I64); 3371 } 3372 break; 3373 case INDEX_op_qemu_st_a32_i128: 3374 case INDEX_op_qemu_st_a64_i128: 3375 tcg_debug_assert(TCG_TARGET_REG_BITS == 64); 3376 tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], false); 3377 break; 3378 3379 case INDEX_op_setcond_i32: 3380 tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2], 3381 const_args[2], false); 3382 break; 3383 case INDEX_op_setcond_i64: 3384 tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2], 3385 const_args[2], false); 3386 break; 3387 case INDEX_op_negsetcond_i32: 3388 tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2], 3389 const_args[2], true); 3390 break; 3391 case INDEX_op_negsetcond_i64: 3392 tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2], 3393 const_args[2], true); 3394 break; 3395 case INDEX_op_setcond2_i32: 3396 tcg_out_setcond2(s, args, const_args); 3397 break; 3398 3399 case INDEX_op_bswap16_i32: 3400 case INDEX_op_bswap16_i64: 3401 tcg_out_bswap16(s, args[0], args[1], args[2]); 3402 break; 3403 case INDEX_op_bswap32_i32: 3404 tcg_out_bswap32(s, args[0], args[1], 0); 3405 break; 3406 case INDEX_op_bswap32_i64: 3407 tcg_out_bswap32(s, args[0], args[1], args[2]); 3408 break; 3409 case INDEX_op_bswap64_i64: 3410 tcg_out_bswap64(s, args[0], args[1]); 3411 break; 3412 3413 case INDEX_op_deposit_i32: 3414 if (const_args[2]) { 3415 uint32_t mask = ((2u << (args[4] - 1)) - 1) << args[3]; 3416 tcg_out_andi32(s, args[0], args[0], ~mask); 3417 } else { 3418 tcg_out_rlw(s, RLWIMI, args[0], args[2], args[3], 3419 32 - args[3] - args[4], 31 - args[3]); 3420 } 3421 break; 3422 case INDEX_op_deposit_i64: 3423 if (const_args[2]) { 3424 uint64_t mask = ((2ull << (args[4] - 1)) - 1) << args[3]; 3425 tcg_out_andi64(s, args[0], args[0], ~mask); 3426 } else { 3427 tcg_out_rld(s, RLDIMI, args[0], args[2], args[3], 3428 64 - args[3] - args[4]); 3429 } 3430 break; 3431 3432 case INDEX_op_extract_i32: 3433 if (args[2] == 0 && args[3] <= 16) { 3434 tcg_out32(s, ANDI | SAI(args[1], args[0], (1 << args[3]) - 1)); 3435 break; 3436 } 3437 tcg_out_rlw(s, RLWINM, args[0], args[1], 3438 32 - args[2], 32 - args[3], 31); 3439 break; 3440 case INDEX_op_extract_i64: 3441 if (args[2] == 0 && args[3] <= 16) { 3442 tcg_out32(s, ANDI | SAI(args[1], args[0], (1 << args[3]) - 1)); 3443 break; 3444 } 3445 tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 64 - args[3]); 3446 break; 3447 3448 case INDEX_op_sextract_i64: 3449 if (args[2] + args[3] == 32) { 3450 if (args[2] == 0) { 3451 tcg_out_ext32s(s, args[0], args[1]); 3452 } else { 3453 tcg_out_sari32(s, args[0], args[1], args[2]); 3454 } 3455 break; 3456 } 3457 /* FALLTHRU */ 3458 case INDEX_op_sextract_i32: 3459 if (args[2] == 0 && args[3] == 8) { 3460 tcg_out_ext8s(s, TCG_TYPE_I32, args[0], args[1]); 3461 } else if (args[2] == 0 && args[3] == 16) { 3462 tcg_out_ext16s(s, TCG_TYPE_I32, args[0], args[1]); 3463 } else { 3464 g_assert_not_reached(); 3465 } 3466 break; 3467 3468 case INDEX_op_movcond_i32: 3469 tcg_out_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], args[2], 3470 args[3], args[4], const_args[2]); 3471 break; 3472 case INDEX_op_movcond_i64: 3473 tcg_out_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], args[2], 3474 args[3], args[4], const_args[2]); 3475 break; 3476 3477#if TCG_TARGET_REG_BITS == 64 3478 case INDEX_op_add2_i64: 3479#else 3480 case INDEX_op_add2_i32: 3481#endif 3482 /* Note that the CA bit is defined based on the word size of the 3483 environment. So in 64-bit mode it's always carry-out of bit 63. 3484 The fallback code using deposit works just as well for 32-bit. */ 3485 a0 = args[0], a1 = args[1]; 3486 if (a0 == args[3] || (!const_args[5] && a0 == args[5])) { 3487 a0 = TCG_REG_R0; 3488 } 3489 if (const_args[4]) { 3490 tcg_out32(s, ADDIC | TAI(a0, args[2], args[4])); 3491 } else { 3492 tcg_out32(s, ADDC | TAB(a0, args[2], args[4])); 3493 } 3494 if (const_args[5]) { 3495 tcg_out32(s, (args[5] ? ADDME : ADDZE) | RT(a1) | RA(args[3])); 3496 } else { 3497 tcg_out32(s, ADDE | TAB(a1, args[3], args[5])); 3498 } 3499 if (a0 != args[0]) { 3500 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0); 3501 } 3502 break; 3503 3504#if TCG_TARGET_REG_BITS == 64 3505 case INDEX_op_sub2_i64: 3506#else 3507 case INDEX_op_sub2_i32: 3508#endif 3509 a0 = args[0], a1 = args[1]; 3510 if (a0 == args[5] || (!const_args[3] && a0 == args[3])) { 3511 a0 = TCG_REG_R0; 3512 } 3513 if (const_args[2]) { 3514 tcg_out32(s, SUBFIC | TAI(a0, args[4], args[2])); 3515 } else { 3516 tcg_out32(s, SUBFC | TAB(a0, args[4], args[2])); 3517 } 3518 if (const_args[3]) { 3519 tcg_out32(s, (args[3] ? SUBFME : SUBFZE) | RT(a1) | RA(args[5])); 3520 } else { 3521 tcg_out32(s, SUBFE | TAB(a1, args[5], args[3])); 3522 } 3523 if (a0 != args[0]) { 3524 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0); 3525 } 3526 break; 3527 3528 case INDEX_op_muluh_i32: 3529 tcg_out32(s, MULHWU | TAB(args[0], args[1], args[2])); 3530 break; 3531 case INDEX_op_mulsh_i32: 3532 tcg_out32(s, MULHW | TAB(args[0], args[1], args[2])); 3533 break; 3534 case INDEX_op_muluh_i64: 3535 tcg_out32(s, MULHDU | TAB(args[0], args[1], args[2])); 3536 break; 3537 case INDEX_op_mulsh_i64: 3538 tcg_out32(s, MULHD | TAB(args[0], args[1], args[2])); 3539 break; 3540 3541 case INDEX_op_mb: 3542 tcg_out_mb(s, args[0]); 3543 break; 3544 3545 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */ 3546 case INDEX_op_mov_i64: 3547 case INDEX_op_call: /* Always emitted via tcg_out_call. */ 3548 case INDEX_op_exit_tb: /* Always emitted via tcg_out_exit_tb. */ 3549 case INDEX_op_goto_tb: /* Always emitted via tcg_out_goto_tb. */ 3550 case INDEX_op_ext8s_i32: /* Always emitted via tcg_reg_alloc_op. */ 3551 case INDEX_op_ext8s_i64: 3552 case INDEX_op_ext8u_i32: 3553 case INDEX_op_ext8u_i64: 3554 case INDEX_op_ext16s_i32: 3555 case INDEX_op_ext16s_i64: 3556 case INDEX_op_ext16u_i32: 3557 case INDEX_op_ext16u_i64: 3558 case INDEX_op_ext32s_i64: 3559 case INDEX_op_ext32u_i64: 3560 case INDEX_op_ext_i32_i64: 3561 case INDEX_op_extu_i32_i64: 3562 case INDEX_op_extrl_i64_i32: 3563 default: 3564 g_assert_not_reached(); 3565 } 3566} 3567 3568int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece) 3569{ 3570 switch (opc) { 3571 case INDEX_op_and_vec: 3572 case INDEX_op_or_vec: 3573 case INDEX_op_xor_vec: 3574 case INDEX_op_andc_vec: 3575 case INDEX_op_not_vec: 3576 case INDEX_op_nor_vec: 3577 case INDEX_op_eqv_vec: 3578 case INDEX_op_nand_vec: 3579 return 1; 3580 case INDEX_op_orc_vec: 3581 return have_isa_2_07; 3582 case INDEX_op_add_vec: 3583 case INDEX_op_sub_vec: 3584 case INDEX_op_smax_vec: 3585 case INDEX_op_smin_vec: 3586 case INDEX_op_umax_vec: 3587 case INDEX_op_umin_vec: 3588 case INDEX_op_shlv_vec: 3589 case INDEX_op_shrv_vec: 3590 case INDEX_op_sarv_vec: 3591 case INDEX_op_rotlv_vec: 3592 return vece <= MO_32 || have_isa_2_07; 3593 case INDEX_op_ssadd_vec: 3594 case INDEX_op_sssub_vec: 3595 case INDEX_op_usadd_vec: 3596 case INDEX_op_ussub_vec: 3597 return vece <= MO_32; 3598 case INDEX_op_shli_vec: 3599 case INDEX_op_shri_vec: 3600 case INDEX_op_sari_vec: 3601 case INDEX_op_rotli_vec: 3602 return vece <= MO_32 || have_isa_2_07 ? -1 : 0; 3603 case INDEX_op_cmp_vec: 3604 case INDEX_op_cmpsel_vec: 3605 return vece <= MO_32 || have_isa_2_07 ? 1 : 0; 3606 case INDEX_op_neg_vec: 3607 return vece >= MO_32 && have_isa_3_00; 3608 case INDEX_op_mul_vec: 3609 switch (vece) { 3610 case MO_8: 3611 case MO_16: 3612 return -1; 3613 case MO_32: 3614 return have_isa_2_07 ? 1 : -1; 3615 case MO_64: 3616 return have_isa_3_10; 3617 } 3618 return 0; 3619 case INDEX_op_bitsel_vec: 3620 return have_vsx; 3621 case INDEX_op_rotrv_vec: 3622 return -1; 3623 default: 3624 return 0; 3625 } 3626} 3627 3628static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece, 3629 TCGReg dst, TCGReg src) 3630{ 3631 tcg_debug_assert(dst >= TCG_REG_V0); 3632 3633 /* Splat from integer reg allowed via constraints for v3.00. */ 3634 if (src < TCG_REG_V0) { 3635 tcg_debug_assert(have_isa_3_00); 3636 switch (vece) { 3637 case MO_64: 3638 tcg_out32(s, MTVSRDD | VRT(dst) | RA(src) | RB(src)); 3639 return true; 3640 case MO_32: 3641 tcg_out32(s, MTVSRWS | VRT(dst) | RA(src)); 3642 return true; 3643 default: 3644 /* Fail, so that we fall back on either dupm or mov+dup. */ 3645 return false; 3646 } 3647 } 3648 3649 /* 3650 * Recall we use (or emulate) VSX integer loads, so the integer is 3651 * right justified within the left (zero-index) double-word. 3652 */ 3653 switch (vece) { 3654 case MO_8: 3655 tcg_out32(s, VSPLTB | VRT(dst) | VRB(src) | (7 << 16)); 3656 break; 3657 case MO_16: 3658 tcg_out32(s, VSPLTH | VRT(dst) | VRB(src) | (3 << 16)); 3659 break; 3660 case MO_32: 3661 tcg_out32(s, VSPLTW | VRT(dst) | VRB(src) | (1 << 16)); 3662 break; 3663 case MO_64: 3664 if (have_vsx) { 3665 tcg_out32(s, XXPERMDI | VRT(dst) | VRA(src) | VRB(src)); 3666 break; 3667 } 3668 tcg_out_vsldoi(s, TCG_VEC_TMP1, src, src, 8); 3669 tcg_out_vsldoi(s, dst, TCG_VEC_TMP1, src, 8); 3670 break; 3671 default: 3672 g_assert_not_reached(); 3673 } 3674 return true; 3675} 3676 3677static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece, 3678 TCGReg out, TCGReg base, intptr_t offset) 3679{ 3680 int elt; 3681 3682 tcg_debug_assert(out >= TCG_REG_V0); 3683 switch (vece) { 3684 case MO_8: 3685 if (have_isa_3_00) { 3686 tcg_out_mem_long(s, LXV, LVX, out, base, offset & -16); 3687 } else { 3688 tcg_out_mem_long(s, 0, LVEBX, out, base, offset); 3689 } 3690 elt = extract32(offset, 0, 4); 3691#if !HOST_BIG_ENDIAN 3692 elt ^= 15; 3693#endif 3694 tcg_out32(s, VSPLTB | VRT(out) | VRB(out) | (elt << 16)); 3695 break; 3696 case MO_16: 3697 tcg_debug_assert((offset & 1) == 0); 3698 if (have_isa_3_00) { 3699 tcg_out_mem_long(s, LXV | 8, LVX, out, base, offset & -16); 3700 } else { 3701 tcg_out_mem_long(s, 0, LVEHX, out, base, offset); 3702 } 3703 elt = extract32(offset, 1, 3); 3704#if !HOST_BIG_ENDIAN 3705 elt ^= 7; 3706#endif 3707 tcg_out32(s, VSPLTH | VRT(out) | VRB(out) | (elt << 16)); 3708 break; 3709 case MO_32: 3710 if (have_isa_3_00) { 3711 tcg_out_mem_long(s, 0, LXVWSX, out, base, offset); 3712 break; 3713 } 3714 tcg_debug_assert((offset & 3) == 0); 3715 tcg_out_mem_long(s, 0, LVEWX, out, base, offset); 3716 elt = extract32(offset, 2, 2); 3717#if !HOST_BIG_ENDIAN 3718 elt ^= 3; 3719#endif 3720 tcg_out32(s, VSPLTW | VRT(out) | VRB(out) | (elt << 16)); 3721 break; 3722 case MO_64: 3723 if (have_vsx) { 3724 tcg_out_mem_long(s, 0, LXVDSX, out, base, offset); 3725 break; 3726 } 3727 tcg_debug_assert((offset & 7) == 0); 3728 tcg_out_mem_long(s, 0, LVX, out, base, offset & -16); 3729 tcg_out_vsldoi(s, TCG_VEC_TMP1, out, out, 8); 3730 elt = extract32(offset, 3, 1); 3731#if !HOST_BIG_ENDIAN 3732 elt = !elt; 3733#endif 3734 if (elt) { 3735 tcg_out_vsldoi(s, out, out, TCG_VEC_TMP1, 8); 3736 } else { 3737 tcg_out_vsldoi(s, out, TCG_VEC_TMP1, out, 8); 3738 } 3739 break; 3740 default: 3741 g_assert_not_reached(); 3742 } 3743 return true; 3744} 3745 3746static void tcg_out_not_vec(TCGContext *s, TCGReg a0, TCGReg a1) 3747{ 3748 tcg_out32(s, VNOR | VRT(a0) | VRA(a1) | VRB(a1)); 3749} 3750 3751static void tcg_out_or_vec(TCGContext *s, TCGReg a0, TCGReg a1, TCGReg a2) 3752{ 3753 tcg_out32(s, VOR | VRT(a0) | VRA(a1) | VRB(a2)); 3754} 3755 3756static void tcg_out_orc_vec(TCGContext *s, TCGReg a0, TCGReg a1, TCGReg a2) 3757{ 3758 tcg_out32(s, VORC | VRT(a0) | VRA(a1) | VRB(a2)); 3759} 3760 3761static void tcg_out_and_vec(TCGContext *s, TCGReg a0, TCGReg a1, TCGReg a2) 3762{ 3763 tcg_out32(s, VAND | VRT(a0) | VRA(a1) | VRB(a2)); 3764} 3765 3766static void tcg_out_andc_vec(TCGContext *s, TCGReg a0, TCGReg a1, TCGReg a2) 3767{ 3768 tcg_out32(s, VANDC | VRT(a0) | VRA(a1) | VRB(a2)); 3769} 3770 3771static void tcg_out_bitsel_vec(TCGContext *s, TCGReg d, 3772 TCGReg c, TCGReg t, TCGReg f) 3773{ 3774 if (TCG_TARGET_HAS_bitsel_vec) { 3775 tcg_out32(s, XXSEL | VRT(d) | VRC(c) | VRB(t) | VRA(f)); 3776 } else { 3777 tcg_out_and_vec(s, TCG_VEC_TMP2, t, c); 3778 tcg_out_andc_vec(s, d, f, c); 3779 tcg_out_or_vec(s, d, d, TCG_VEC_TMP2); 3780 } 3781} 3782 3783static bool tcg_out_cmp_vec_noinv(TCGContext *s, unsigned vece, TCGReg a0, 3784 TCGReg a1, TCGReg a2, TCGCond cond) 3785{ 3786 static const uint32_t 3787 eq_op[4] = { VCMPEQUB, VCMPEQUH, VCMPEQUW, VCMPEQUD }, 3788 ne_op[4] = { VCMPNEB, VCMPNEH, VCMPNEW, 0 }, 3789 gts_op[4] = { VCMPGTSB, VCMPGTSH, VCMPGTSW, VCMPGTSD }, 3790 gtu_op[4] = { VCMPGTUB, VCMPGTUH, VCMPGTUW, VCMPGTUD }; 3791 uint32_t insn; 3792 3793 bool need_swap = false, need_inv = false; 3794 3795 tcg_debug_assert(vece <= MO_32 || have_isa_2_07); 3796 3797 switch (cond) { 3798 case TCG_COND_EQ: 3799 case TCG_COND_GT: 3800 case TCG_COND_GTU: 3801 break; 3802 case TCG_COND_NE: 3803 if (have_isa_3_00 && vece <= MO_32) { 3804 break; 3805 } 3806 /* fall through */ 3807 case TCG_COND_LE: 3808 case TCG_COND_LEU: 3809 need_inv = true; 3810 break; 3811 case TCG_COND_LT: 3812 case TCG_COND_LTU: 3813 need_swap = true; 3814 break; 3815 case TCG_COND_GE: 3816 case TCG_COND_GEU: 3817 need_swap = need_inv = true; 3818 break; 3819 default: 3820 g_assert_not_reached(); 3821 } 3822 3823 if (need_inv) { 3824 cond = tcg_invert_cond(cond); 3825 } 3826 if (need_swap) { 3827 TCGReg swap = a1; 3828 a1 = a2; 3829 a2 = swap; 3830 cond = tcg_swap_cond(cond); 3831 } 3832 3833 switch (cond) { 3834 case TCG_COND_EQ: 3835 insn = eq_op[vece]; 3836 break; 3837 case TCG_COND_NE: 3838 insn = ne_op[vece]; 3839 break; 3840 case TCG_COND_GT: 3841 insn = gts_op[vece]; 3842 break; 3843 case TCG_COND_GTU: 3844 insn = gtu_op[vece]; 3845 break; 3846 default: 3847 g_assert_not_reached(); 3848 } 3849 tcg_out32(s, insn | VRT(a0) | VRA(a1) | VRB(a2)); 3850 3851 return need_inv; 3852} 3853 3854static void tcg_out_cmp_vec(TCGContext *s, unsigned vece, TCGReg a0, 3855 TCGReg a1, TCGReg a2, TCGCond cond) 3856{ 3857 if (tcg_out_cmp_vec_noinv(s, vece, a0, a1, a2, cond)) { 3858 tcg_out_not_vec(s, a0, a0); 3859 } 3860} 3861 3862static void tcg_out_cmpsel_vec(TCGContext *s, unsigned vece, TCGReg a0, 3863 TCGReg c1, TCGReg c2, TCGArg v3, int const_v3, 3864 TCGReg v4, TCGCond cond) 3865{ 3866 bool inv = tcg_out_cmp_vec_noinv(s, vece, TCG_VEC_TMP1, c1, c2, cond); 3867 3868 if (!const_v3) { 3869 if (inv) { 3870 tcg_out_bitsel_vec(s, a0, TCG_VEC_TMP1, v4, v3); 3871 } else { 3872 tcg_out_bitsel_vec(s, a0, TCG_VEC_TMP1, v3, v4); 3873 } 3874 } else if (v3) { 3875 if (inv) { 3876 tcg_out_orc_vec(s, a0, v4, TCG_VEC_TMP1); 3877 } else { 3878 tcg_out_or_vec(s, a0, v4, TCG_VEC_TMP1); 3879 } 3880 } else { 3881 if (inv) { 3882 tcg_out_and_vec(s, a0, v4, TCG_VEC_TMP1); 3883 } else { 3884 tcg_out_andc_vec(s, a0, v4, TCG_VEC_TMP1); 3885 } 3886 } 3887} 3888 3889static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc, 3890 unsigned vecl, unsigned vece, 3891 const TCGArg args[TCG_MAX_OP_ARGS], 3892 const int const_args[TCG_MAX_OP_ARGS]) 3893{ 3894 static const uint32_t 3895 add_op[4] = { VADDUBM, VADDUHM, VADDUWM, VADDUDM }, 3896 sub_op[4] = { VSUBUBM, VSUBUHM, VSUBUWM, VSUBUDM }, 3897 mul_op[4] = { 0, 0, VMULUWM, VMULLD }, 3898 neg_op[4] = { 0, 0, VNEGW, VNEGD }, 3899 ssadd_op[4] = { VADDSBS, VADDSHS, VADDSWS, 0 }, 3900 usadd_op[4] = { VADDUBS, VADDUHS, VADDUWS, 0 }, 3901 sssub_op[4] = { VSUBSBS, VSUBSHS, VSUBSWS, 0 }, 3902 ussub_op[4] = { VSUBUBS, VSUBUHS, VSUBUWS, 0 }, 3903 umin_op[4] = { VMINUB, VMINUH, VMINUW, VMINUD }, 3904 smin_op[4] = { VMINSB, VMINSH, VMINSW, VMINSD }, 3905 umax_op[4] = { VMAXUB, VMAXUH, VMAXUW, VMAXUD }, 3906 smax_op[4] = { VMAXSB, VMAXSH, VMAXSW, VMAXSD }, 3907 shlv_op[4] = { VSLB, VSLH, VSLW, VSLD }, 3908 shrv_op[4] = { VSRB, VSRH, VSRW, VSRD }, 3909 sarv_op[4] = { VSRAB, VSRAH, VSRAW, VSRAD }, 3910 mrgh_op[4] = { VMRGHB, VMRGHH, VMRGHW, 0 }, 3911 mrgl_op[4] = { VMRGLB, VMRGLH, VMRGLW, 0 }, 3912 muleu_op[4] = { VMULEUB, VMULEUH, VMULEUW, 0 }, 3913 mulou_op[4] = { VMULOUB, VMULOUH, VMULOUW, 0 }, 3914 pkum_op[4] = { VPKUHUM, VPKUWUM, 0, 0 }, 3915 rotl_op[4] = { VRLB, VRLH, VRLW, VRLD }; 3916 3917 TCGType type = vecl + TCG_TYPE_V64; 3918 TCGArg a0 = args[0], a1 = args[1], a2 = args[2]; 3919 uint32_t insn; 3920 3921 switch (opc) { 3922 case INDEX_op_ld_vec: 3923 tcg_out_ld(s, type, a0, a1, a2); 3924 return; 3925 case INDEX_op_st_vec: 3926 tcg_out_st(s, type, a0, a1, a2); 3927 return; 3928 case INDEX_op_dupm_vec: 3929 tcg_out_dupm_vec(s, type, vece, a0, a1, a2); 3930 return; 3931 3932 case INDEX_op_add_vec: 3933 insn = add_op[vece]; 3934 break; 3935 case INDEX_op_sub_vec: 3936 insn = sub_op[vece]; 3937 break; 3938 case INDEX_op_neg_vec: 3939 insn = neg_op[vece]; 3940 a2 = a1; 3941 a1 = 0; 3942 break; 3943 case INDEX_op_mul_vec: 3944 insn = mul_op[vece]; 3945 break; 3946 case INDEX_op_ssadd_vec: 3947 insn = ssadd_op[vece]; 3948 break; 3949 case INDEX_op_sssub_vec: 3950 insn = sssub_op[vece]; 3951 break; 3952 case INDEX_op_usadd_vec: 3953 insn = usadd_op[vece]; 3954 break; 3955 case INDEX_op_ussub_vec: 3956 insn = ussub_op[vece]; 3957 break; 3958 case INDEX_op_smin_vec: 3959 insn = smin_op[vece]; 3960 break; 3961 case INDEX_op_umin_vec: 3962 insn = umin_op[vece]; 3963 break; 3964 case INDEX_op_smax_vec: 3965 insn = smax_op[vece]; 3966 break; 3967 case INDEX_op_umax_vec: 3968 insn = umax_op[vece]; 3969 break; 3970 case INDEX_op_shlv_vec: 3971 insn = shlv_op[vece]; 3972 break; 3973 case INDEX_op_shrv_vec: 3974 insn = shrv_op[vece]; 3975 break; 3976 case INDEX_op_sarv_vec: 3977 insn = sarv_op[vece]; 3978 break; 3979 case INDEX_op_and_vec: 3980 tcg_out_and_vec(s, a0, a1, a2); 3981 return; 3982 case INDEX_op_or_vec: 3983 tcg_out_or_vec(s, a0, a1, a2); 3984 return; 3985 case INDEX_op_xor_vec: 3986 insn = VXOR; 3987 break; 3988 case INDEX_op_andc_vec: 3989 tcg_out_andc_vec(s, a0, a1, a2); 3990 return; 3991 case INDEX_op_not_vec: 3992 tcg_out_not_vec(s, a0, a1); 3993 return; 3994 case INDEX_op_orc_vec: 3995 tcg_out_orc_vec(s, a0, a1, a2); 3996 return; 3997 case INDEX_op_nand_vec: 3998 insn = VNAND; 3999 break; 4000 case INDEX_op_nor_vec: 4001 insn = VNOR; 4002 break; 4003 case INDEX_op_eqv_vec: 4004 insn = VEQV; 4005 break; 4006 4007 case INDEX_op_cmp_vec: 4008 tcg_out_cmp_vec(s, vece, a0, a1, a2, args[3]); 4009 return; 4010 case INDEX_op_cmpsel_vec: 4011 tcg_out_cmpsel_vec(s, vece, a0, a1, a2, 4012 args[3], const_args[3], args[4], args[5]); 4013 return; 4014 case INDEX_op_bitsel_vec: 4015 tcg_out_bitsel_vec(s, a0, a1, a2, args[3]); 4016 return; 4017 4018 case INDEX_op_dup2_vec: 4019 assert(TCG_TARGET_REG_BITS == 32); 4020 /* With inputs a1 = xLxx, a2 = xHxx */ 4021 tcg_out32(s, VMRGHW | VRT(a0) | VRA(a2) | VRB(a1)); /* a0 = xxHL */ 4022 tcg_out_vsldoi(s, TCG_VEC_TMP1, a0, a0, 8); /* tmp = HLxx */ 4023 tcg_out_vsldoi(s, a0, a0, TCG_VEC_TMP1, 8); /* a0 = HLHL */ 4024 return; 4025 4026 case INDEX_op_ppc_mrgh_vec: 4027 insn = mrgh_op[vece]; 4028 break; 4029 case INDEX_op_ppc_mrgl_vec: 4030 insn = mrgl_op[vece]; 4031 break; 4032 case INDEX_op_ppc_muleu_vec: 4033 insn = muleu_op[vece]; 4034 break; 4035 case INDEX_op_ppc_mulou_vec: 4036 insn = mulou_op[vece]; 4037 break; 4038 case INDEX_op_ppc_pkum_vec: 4039 insn = pkum_op[vece]; 4040 break; 4041 case INDEX_op_rotlv_vec: 4042 insn = rotl_op[vece]; 4043 break; 4044 case INDEX_op_ppc_msum_vec: 4045 tcg_debug_assert(vece == MO_16); 4046 tcg_out32(s, VMSUMUHM | VRT(a0) | VRA(a1) | VRB(a2) | VRC(args[3])); 4047 return; 4048 4049 case INDEX_op_mov_vec: /* Always emitted via tcg_out_mov. */ 4050 case INDEX_op_dup_vec: /* Always emitted via tcg_out_dup_vec. */ 4051 default: 4052 g_assert_not_reached(); 4053 } 4054 4055 tcg_debug_assert(insn != 0); 4056 tcg_out32(s, insn | VRT(a0) | VRA(a1) | VRB(a2)); 4057} 4058 4059static void expand_vec_shi(TCGType type, unsigned vece, TCGv_vec v0, 4060 TCGv_vec v1, TCGArg imm, TCGOpcode opci) 4061{ 4062 TCGv_vec t1; 4063 4064 if (vece == MO_32) { 4065 /* 4066 * Only 5 bits are significant, and VSPLTISB can represent -16..15. 4067 * So using negative numbers gets us the 4th bit easily. 4068 */ 4069 imm = sextract32(imm, 0, 5); 4070 } else { 4071 imm &= (8 << vece) - 1; 4072 } 4073 4074 /* Splat w/bytes for xxspltib when 2.07 allows MO_64. */ 4075 t1 = tcg_constant_vec(type, MO_8, imm); 4076 vec_gen_3(opci, type, vece, tcgv_vec_arg(v0), 4077 tcgv_vec_arg(v1), tcgv_vec_arg(t1)); 4078} 4079 4080static void expand_vec_mul(TCGType type, unsigned vece, TCGv_vec v0, 4081 TCGv_vec v1, TCGv_vec v2) 4082{ 4083 TCGv_vec t1 = tcg_temp_new_vec(type); 4084 TCGv_vec t2 = tcg_temp_new_vec(type); 4085 TCGv_vec c0, c16; 4086 4087 switch (vece) { 4088 case MO_8: 4089 case MO_16: 4090 vec_gen_3(INDEX_op_ppc_muleu_vec, type, vece, tcgv_vec_arg(t1), 4091 tcgv_vec_arg(v1), tcgv_vec_arg(v2)); 4092 vec_gen_3(INDEX_op_ppc_mulou_vec, type, vece, tcgv_vec_arg(t2), 4093 tcgv_vec_arg(v1), tcgv_vec_arg(v2)); 4094 vec_gen_3(INDEX_op_ppc_mrgh_vec, type, vece + 1, tcgv_vec_arg(v0), 4095 tcgv_vec_arg(t1), tcgv_vec_arg(t2)); 4096 vec_gen_3(INDEX_op_ppc_mrgl_vec, type, vece + 1, tcgv_vec_arg(t1), 4097 tcgv_vec_arg(t1), tcgv_vec_arg(t2)); 4098 vec_gen_3(INDEX_op_ppc_pkum_vec, type, vece, tcgv_vec_arg(v0), 4099 tcgv_vec_arg(v0), tcgv_vec_arg(t1)); 4100 break; 4101 4102 case MO_32: 4103 tcg_debug_assert(!have_isa_2_07); 4104 /* 4105 * Only 5 bits are significant, and VSPLTISB can represent -16..15. 4106 * So using -16 is a quick way to represent 16. 4107 */ 4108 c16 = tcg_constant_vec(type, MO_8, -16); 4109 c0 = tcg_constant_vec(type, MO_8, 0); 4110 4111 vec_gen_3(INDEX_op_rotlv_vec, type, MO_32, tcgv_vec_arg(t1), 4112 tcgv_vec_arg(v2), tcgv_vec_arg(c16)); 4113 vec_gen_3(INDEX_op_ppc_mulou_vec, type, MO_16, tcgv_vec_arg(t2), 4114 tcgv_vec_arg(v1), tcgv_vec_arg(v2)); 4115 vec_gen_4(INDEX_op_ppc_msum_vec, type, MO_16, tcgv_vec_arg(t1), 4116 tcgv_vec_arg(v1), tcgv_vec_arg(t1), tcgv_vec_arg(c0)); 4117 vec_gen_3(INDEX_op_shlv_vec, type, MO_32, tcgv_vec_arg(t1), 4118 tcgv_vec_arg(t1), tcgv_vec_arg(c16)); 4119 tcg_gen_add_vec(MO_32, v0, t1, t2); 4120 break; 4121 4122 default: 4123 g_assert_not_reached(); 4124 } 4125 tcg_temp_free_vec(t1); 4126 tcg_temp_free_vec(t2); 4127} 4128 4129void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece, 4130 TCGArg a0, ...) 4131{ 4132 va_list va; 4133 TCGv_vec v0, v1, v2, t0; 4134 TCGArg a2; 4135 4136 va_start(va, a0); 4137 v0 = temp_tcgv_vec(arg_temp(a0)); 4138 v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg))); 4139 a2 = va_arg(va, TCGArg); 4140 4141 switch (opc) { 4142 case INDEX_op_shli_vec: 4143 expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shlv_vec); 4144 break; 4145 case INDEX_op_shri_vec: 4146 expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shrv_vec); 4147 break; 4148 case INDEX_op_sari_vec: 4149 expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_sarv_vec); 4150 break; 4151 case INDEX_op_rotli_vec: 4152 expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_rotlv_vec); 4153 break; 4154 case INDEX_op_mul_vec: 4155 v2 = temp_tcgv_vec(arg_temp(a2)); 4156 expand_vec_mul(type, vece, v0, v1, v2); 4157 break; 4158 case INDEX_op_rotlv_vec: 4159 v2 = temp_tcgv_vec(arg_temp(a2)); 4160 t0 = tcg_temp_new_vec(type); 4161 tcg_gen_neg_vec(vece, t0, v2); 4162 tcg_gen_rotlv_vec(vece, v0, v1, t0); 4163 tcg_temp_free_vec(t0); 4164 break; 4165 default: 4166 g_assert_not_reached(); 4167 } 4168 va_end(va); 4169} 4170 4171static TCGConstraintSetIndex 4172tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags) 4173{ 4174 switch (op) { 4175 case INDEX_op_goto_ptr: 4176 return C_O0_I1(r); 4177 4178 case INDEX_op_ld8u_i32: 4179 case INDEX_op_ld8s_i32: 4180 case INDEX_op_ld16u_i32: 4181 case INDEX_op_ld16s_i32: 4182 case INDEX_op_ld_i32: 4183 case INDEX_op_ctpop_i32: 4184 case INDEX_op_neg_i32: 4185 case INDEX_op_not_i32: 4186 case INDEX_op_ext8s_i32: 4187 case INDEX_op_ext16s_i32: 4188 case INDEX_op_bswap16_i32: 4189 case INDEX_op_bswap32_i32: 4190 case INDEX_op_extract_i32: 4191 case INDEX_op_sextract_i32: 4192 case INDEX_op_ld8u_i64: 4193 case INDEX_op_ld8s_i64: 4194 case INDEX_op_ld16u_i64: 4195 case INDEX_op_ld16s_i64: 4196 case INDEX_op_ld32u_i64: 4197 case INDEX_op_ld32s_i64: 4198 case INDEX_op_ld_i64: 4199 case INDEX_op_ctpop_i64: 4200 case INDEX_op_neg_i64: 4201 case INDEX_op_not_i64: 4202 case INDEX_op_ext8s_i64: 4203 case INDEX_op_ext16s_i64: 4204 case INDEX_op_ext32s_i64: 4205 case INDEX_op_ext_i32_i64: 4206 case INDEX_op_extu_i32_i64: 4207 case INDEX_op_bswap16_i64: 4208 case INDEX_op_bswap32_i64: 4209 case INDEX_op_bswap64_i64: 4210 case INDEX_op_extract_i64: 4211 case INDEX_op_sextract_i64: 4212 return C_O1_I1(r, r); 4213 4214 case INDEX_op_st8_i32: 4215 case INDEX_op_st16_i32: 4216 case INDEX_op_st_i32: 4217 case INDEX_op_st8_i64: 4218 case INDEX_op_st16_i64: 4219 case INDEX_op_st32_i64: 4220 case INDEX_op_st_i64: 4221 return C_O0_I2(r, r); 4222 4223 case INDEX_op_add_i32: 4224 case INDEX_op_and_i32: 4225 case INDEX_op_or_i32: 4226 case INDEX_op_xor_i32: 4227 case INDEX_op_andc_i32: 4228 case INDEX_op_orc_i32: 4229 case INDEX_op_eqv_i32: 4230 case INDEX_op_shl_i32: 4231 case INDEX_op_shr_i32: 4232 case INDEX_op_sar_i32: 4233 case INDEX_op_rotl_i32: 4234 case INDEX_op_rotr_i32: 4235 case INDEX_op_and_i64: 4236 case INDEX_op_andc_i64: 4237 case INDEX_op_shl_i64: 4238 case INDEX_op_shr_i64: 4239 case INDEX_op_sar_i64: 4240 case INDEX_op_rotl_i64: 4241 case INDEX_op_rotr_i64: 4242 return C_O1_I2(r, r, ri); 4243 4244 case INDEX_op_mul_i32: 4245 case INDEX_op_mul_i64: 4246 return C_O1_I2(r, r, rI); 4247 4248 case INDEX_op_div_i32: 4249 case INDEX_op_divu_i32: 4250 case INDEX_op_rem_i32: 4251 case INDEX_op_remu_i32: 4252 case INDEX_op_nand_i32: 4253 case INDEX_op_nor_i32: 4254 case INDEX_op_muluh_i32: 4255 case INDEX_op_mulsh_i32: 4256 case INDEX_op_orc_i64: 4257 case INDEX_op_eqv_i64: 4258 case INDEX_op_nand_i64: 4259 case INDEX_op_nor_i64: 4260 case INDEX_op_div_i64: 4261 case INDEX_op_divu_i64: 4262 case INDEX_op_rem_i64: 4263 case INDEX_op_remu_i64: 4264 case INDEX_op_mulsh_i64: 4265 case INDEX_op_muluh_i64: 4266 return C_O1_I2(r, r, r); 4267 4268 case INDEX_op_sub_i32: 4269 return C_O1_I2(r, rI, ri); 4270 case INDEX_op_add_i64: 4271 return C_O1_I2(r, r, rT); 4272 case INDEX_op_or_i64: 4273 case INDEX_op_xor_i64: 4274 return C_O1_I2(r, r, rU); 4275 case INDEX_op_sub_i64: 4276 return C_O1_I2(r, rI, rT); 4277 case INDEX_op_clz_i32: 4278 case INDEX_op_ctz_i32: 4279 case INDEX_op_clz_i64: 4280 case INDEX_op_ctz_i64: 4281 return C_O1_I2(r, r, rZW); 4282 4283 case INDEX_op_brcond_i32: 4284 case INDEX_op_brcond_i64: 4285 return C_O0_I2(r, rC); 4286 case INDEX_op_setcond_i32: 4287 case INDEX_op_setcond_i64: 4288 case INDEX_op_negsetcond_i32: 4289 case INDEX_op_negsetcond_i64: 4290 return C_O1_I2(r, r, rC); 4291 case INDEX_op_movcond_i32: 4292 case INDEX_op_movcond_i64: 4293 return C_O1_I4(r, r, rC, rZ, rZ); 4294 4295 case INDEX_op_deposit_i32: 4296 case INDEX_op_deposit_i64: 4297 return C_O1_I2(r, 0, rZ); 4298 case INDEX_op_brcond2_i32: 4299 return C_O0_I4(r, r, ri, ri); 4300 case INDEX_op_setcond2_i32: 4301 return C_O1_I4(r, r, r, ri, ri); 4302 case INDEX_op_add2_i64: 4303 case INDEX_op_add2_i32: 4304 return C_O2_I4(r, r, r, r, rI, rZM); 4305 case INDEX_op_sub2_i64: 4306 case INDEX_op_sub2_i32: 4307 return C_O2_I4(r, r, rI, rZM, r, r); 4308 4309 case INDEX_op_qemu_ld_a32_i32: 4310 return C_O1_I1(r, r); 4311 case INDEX_op_qemu_ld_a64_i32: 4312 return TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, r) : C_O1_I2(r, r, r); 4313 case INDEX_op_qemu_ld_a32_i64: 4314 return TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, r) : C_O2_I1(r, r, r); 4315 case INDEX_op_qemu_ld_a64_i64: 4316 return TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, r) : C_O2_I2(r, r, r, r); 4317 4318 case INDEX_op_qemu_st_a32_i32: 4319 return C_O0_I2(r, r); 4320 case INDEX_op_qemu_st_a64_i32: 4321 return TCG_TARGET_REG_BITS == 64 ? C_O0_I2(r, r) : C_O0_I3(r, r, r); 4322 case INDEX_op_qemu_st_a32_i64: 4323 return TCG_TARGET_REG_BITS == 64 ? C_O0_I2(r, r) : C_O0_I3(r, r, r); 4324 case INDEX_op_qemu_st_a64_i64: 4325 return TCG_TARGET_REG_BITS == 64 ? C_O0_I2(r, r) : C_O0_I4(r, r, r, r); 4326 4327 case INDEX_op_qemu_ld_a32_i128: 4328 case INDEX_op_qemu_ld_a64_i128: 4329 return C_N1O1_I1(o, m, r); 4330 case INDEX_op_qemu_st_a32_i128: 4331 case INDEX_op_qemu_st_a64_i128: 4332 return C_O0_I3(o, m, r); 4333 4334 case INDEX_op_add_vec: 4335 case INDEX_op_sub_vec: 4336 case INDEX_op_mul_vec: 4337 case INDEX_op_and_vec: 4338 case INDEX_op_or_vec: 4339 case INDEX_op_xor_vec: 4340 case INDEX_op_andc_vec: 4341 case INDEX_op_orc_vec: 4342 case INDEX_op_nor_vec: 4343 case INDEX_op_eqv_vec: 4344 case INDEX_op_nand_vec: 4345 case INDEX_op_cmp_vec: 4346 case INDEX_op_ssadd_vec: 4347 case INDEX_op_sssub_vec: 4348 case INDEX_op_usadd_vec: 4349 case INDEX_op_ussub_vec: 4350 case INDEX_op_smax_vec: 4351 case INDEX_op_smin_vec: 4352 case INDEX_op_umax_vec: 4353 case INDEX_op_umin_vec: 4354 case INDEX_op_shlv_vec: 4355 case INDEX_op_shrv_vec: 4356 case INDEX_op_sarv_vec: 4357 case INDEX_op_rotlv_vec: 4358 case INDEX_op_rotrv_vec: 4359 case INDEX_op_ppc_mrgh_vec: 4360 case INDEX_op_ppc_mrgl_vec: 4361 case INDEX_op_ppc_muleu_vec: 4362 case INDEX_op_ppc_mulou_vec: 4363 case INDEX_op_ppc_pkum_vec: 4364 case INDEX_op_dup2_vec: 4365 return C_O1_I2(v, v, v); 4366 4367 case INDEX_op_not_vec: 4368 case INDEX_op_neg_vec: 4369 return C_O1_I1(v, v); 4370 4371 case INDEX_op_dup_vec: 4372 return have_isa_3_00 ? C_O1_I1(v, vr) : C_O1_I1(v, v); 4373 4374 case INDEX_op_ld_vec: 4375 case INDEX_op_dupm_vec: 4376 return C_O1_I1(v, r); 4377 4378 case INDEX_op_st_vec: 4379 return C_O0_I2(v, r); 4380 4381 case INDEX_op_bitsel_vec: 4382 case INDEX_op_ppc_msum_vec: 4383 return C_O1_I3(v, v, v, v); 4384 case INDEX_op_cmpsel_vec: 4385 return C_O1_I4(v, v, v, vZM, v); 4386 4387 default: 4388 return C_NotImplemented; 4389 } 4390} 4391 4392static void tcg_target_init(TCGContext *s) 4393{ 4394 tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff; 4395 tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff; 4396 if (have_altivec) { 4397 tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull; 4398 tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull; 4399 } 4400 4401 tcg_target_call_clobber_regs = 0; 4402 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0); 4403 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2); 4404 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3); 4405 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4); 4406 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5); 4407 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6); 4408 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R7); 4409 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R8); 4410 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R9); 4411 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R10); 4412 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R11); 4413 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12); 4414 4415 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0); 4416 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1); 4417 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2); 4418 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3); 4419 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4); 4420 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5); 4421 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6); 4422 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7); 4423 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V8); 4424 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V9); 4425 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V10); 4426 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V11); 4427 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V12); 4428 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V13); 4429 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V14); 4430 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V15); 4431 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16); 4432 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17); 4433 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18); 4434 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19); 4435 4436 s->reserved_regs = 0; 4437 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); /* tcg temp */ 4438 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R1); /* stack pointer */ 4439#if defined(_CALL_SYSV) 4440 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R2); /* toc pointer */ 4441#endif 4442#if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64 4443 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R13); /* thread pointer */ 4444#endif 4445 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1); 4446 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP2); 4447 tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP1); 4448 tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP2); 4449 if (USE_REG_TB) { 4450 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB); /* tb->tc_ptr */ 4451 } 4452} 4453 4454#ifdef __ELF__ 4455typedef struct { 4456 DebugFrameCIE cie; 4457 DebugFrameFDEHeader fde; 4458 uint8_t fde_def_cfa[4]; 4459 uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2 + 3]; 4460} DebugFrame; 4461 4462/* We're expecting a 2 byte uleb128 encoded value. */ 4463QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14)); 4464 4465#if TCG_TARGET_REG_BITS == 64 4466# define ELF_HOST_MACHINE EM_PPC64 4467#else 4468# define ELF_HOST_MACHINE EM_PPC 4469#endif 4470 4471static DebugFrame debug_frame = { 4472 .cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */ 4473 .cie.id = -1, 4474 .cie.version = 1, 4475 .cie.code_align = 1, 4476 .cie.data_align = (-SZR & 0x7f), /* sleb128 -SZR */ 4477 .cie.return_column = 65, 4478 4479 /* Total FDE size does not include the "len" member. */ 4480 .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset), 4481 4482 .fde_def_cfa = { 4483 12, TCG_REG_R1, /* DW_CFA_def_cfa r1, ... */ 4484 (FRAME_SIZE & 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */ 4485 (FRAME_SIZE >> 7) 4486 }, 4487 .fde_reg_ofs = { 4488 /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */ 4489 0x11, 65, (LR_OFFSET / -SZR) & 0x7f, 4490 } 4491}; 4492 4493void tcg_register_jit(const void *buf, size_t buf_size) 4494{ 4495 uint8_t *p = &debug_frame.fde_reg_ofs[3]; 4496 int i; 4497 4498 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i, p += 2) { 4499 p[0] = 0x80 + tcg_target_callee_save_regs[i]; 4500 p[1] = (FRAME_SIZE - (REG_SAVE_BOT + i * SZR)) / SZR; 4501 } 4502 4503 debug_frame.fde.func_start = (uintptr_t)buf; 4504 debug_frame.fde.func_len = buf_size; 4505 4506 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame)); 4507} 4508#endif /* __ELF__ */ 4509#undef VMULEUB 4510#undef VMULEUH 4511#undef VMULEUW 4512#undef VMULOUB 4513#undef VMULOUH 4514#undef VMULOUW 4515#undef VMSUMUHM 4516