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 #include "tcg-pool.inc.c"
27
28 #if defined _CALL_DARWIN || defined __APPLE__
29 #define TCG_TARGET_CALL_DARWIN
30 #endif
31 #ifdef _CALL_SYSV
32 # define TCG_TARGET_CALL_ALIGN_ARGS 1
33 #endif
34
35 /* For some memory operations, we need a scratch that isn't R0. For the AIX
36 calling convention, we can re-use the TOC register since we'll be reloading
37 it at every call. Otherwise R12 will do nicely as neither a call-saved
38 register nor a parameter register. */
39 #ifdef _CALL_AIX
40 # define TCG_REG_TMP1 TCG_REG_R2
41 #else
42 # define TCG_REG_TMP1 TCG_REG_R12
43 #endif
44
45 #define TCG_REG_TB TCG_REG_R31
46 #define USE_REG_TB (TCG_TARGET_REG_BITS == 64)
47
48 /* Shorthand for size of a pointer. Avoid promotion to unsigned. */
49 #define SZP ((int)sizeof(void *))
50
51 /* Shorthand for size of a register. */
52 #define SZR (TCG_TARGET_REG_BITS / 8)
53
54 #define TCG_CT_CONST_S16 0x100
55 #define TCG_CT_CONST_U16 0x200
56 #define TCG_CT_CONST_S32 0x400
57 #define TCG_CT_CONST_U32 0x800
58 #define TCG_CT_CONST_ZERO 0x1000
59 #define TCG_CT_CONST_MONE 0x2000
60 #define TCG_CT_CONST_WSZ 0x4000
61
62 static tcg_insn_unit *tb_ret_addr;
63
64 bool have_isa_2_06;
65 bool have_isa_3_00;
66
67 #define HAVE_ISA_2_06 have_isa_2_06
68 #define HAVE_ISEL have_isa_2_06
69
70 #ifndef CONFIG_SOFTMMU
71 #define TCG_GUEST_BASE_REG 30
72 #endif
73
74 #ifdef CONFIG_DEBUG_TCG
75 static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
76 "r0",
77 "r1",
78 "r2",
79 "r3",
80 "r4",
81 "r5",
82 "r6",
83 "r7",
84 "r8",
85 "r9",
86 "r10",
87 "r11",
88 "r12",
89 "r13",
90 "r14",
91 "r15",
92 "r16",
93 "r17",
94 "r18",
95 "r19",
96 "r20",
97 "r21",
98 "r22",
99 "r23",
100 "r24",
101 "r25",
102 "r26",
103 "r27",
104 "r28",
105 "r29",
106 "r30",
107 "r31"
108 };
109 #endif
110
111 static const int tcg_target_reg_alloc_order[] = {
112 TCG_REG_R14, /* call saved registers */
113 TCG_REG_R15,
114 TCG_REG_R16,
115 TCG_REG_R17,
116 TCG_REG_R18,
117 TCG_REG_R19,
118 TCG_REG_R20,
119 TCG_REG_R21,
120 TCG_REG_R22,
121 TCG_REG_R23,
122 TCG_REG_R24,
123 TCG_REG_R25,
124 TCG_REG_R26,
125 TCG_REG_R27,
126 TCG_REG_R28,
127 TCG_REG_R29,
128 TCG_REG_R30,
129 TCG_REG_R31,
130 TCG_REG_R12, /* call clobbered, non-arguments */
131 TCG_REG_R11,
132 TCG_REG_R2,
133 TCG_REG_R13,
134 TCG_REG_R10, /* call clobbered, arguments */
135 TCG_REG_R9,
136 TCG_REG_R8,
137 TCG_REG_R7,
138 TCG_REG_R6,
139 TCG_REG_R5,
140 TCG_REG_R4,
141 TCG_REG_R3,
142 };
143
144 static const int tcg_target_call_iarg_regs[] = {
145 TCG_REG_R3,
146 TCG_REG_R4,
147 TCG_REG_R5,
148 TCG_REG_R6,
149 TCG_REG_R7,
150 TCG_REG_R8,
151 TCG_REG_R9,
152 TCG_REG_R10
153 };
154
155 static const int tcg_target_call_oarg_regs[] = {
156 TCG_REG_R3,
157 TCG_REG_R4
158 };
159
160 static const int tcg_target_callee_save_regs[] = {
161 #ifdef TCG_TARGET_CALL_DARWIN
162 TCG_REG_R11,
163 #endif
164 TCG_REG_R14,
165 TCG_REG_R15,
166 TCG_REG_R16,
167 TCG_REG_R17,
168 TCG_REG_R18,
169 TCG_REG_R19,
170 TCG_REG_R20,
171 TCG_REG_R21,
172 TCG_REG_R22,
173 TCG_REG_R23,
174 TCG_REG_R24,
175 TCG_REG_R25,
176 TCG_REG_R26,
177 TCG_REG_R27, /* currently used for the global env */
178 TCG_REG_R28,
179 TCG_REG_R29,
180 TCG_REG_R30,
181 TCG_REG_R31
182 };
183
in_range_b(tcg_target_long target)184 static inline bool in_range_b(tcg_target_long target)
185 {
186 return target == sextract64(target, 0, 26);
187 }
188
reloc_pc24_val(tcg_insn_unit * pc,tcg_insn_unit * target)189 static uint32_t reloc_pc24_val(tcg_insn_unit *pc, tcg_insn_unit *target)
190 {
191 ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
192 tcg_debug_assert(in_range_b(disp));
193 return disp & 0x3fffffc;
194 }
195
reloc_pc24(tcg_insn_unit * pc,tcg_insn_unit * target)196 static void reloc_pc24(tcg_insn_unit *pc, tcg_insn_unit *target)
197 {
198 *pc = (*pc & ~0x3fffffc) | reloc_pc24_val(pc, target);
199 }
200
reloc_pc14_val(tcg_insn_unit * pc,tcg_insn_unit * target)201 static uint16_t reloc_pc14_val(tcg_insn_unit *pc, tcg_insn_unit *target)
202 {
203 ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
204 tcg_debug_assert(disp == (int16_t) disp);
205 return disp & 0xfffc;
206 }
207
reloc_pc14(tcg_insn_unit * pc,tcg_insn_unit * target)208 static void reloc_pc14(tcg_insn_unit *pc, tcg_insn_unit *target)
209 {
210 *pc = (*pc & ~0xfffc) | reloc_pc14_val(pc, target);
211 }
212
tcg_out_b_noaddr(TCGContext * s,int insn)213 static inline void tcg_out_b_noaddr(TCGContext *s, int insn)
214 {
215 unsigned retrans = *s->code_ptr & 0x3fffffc;
216 tcg_out32(s, insn | retrans);
217 }
218
tcg_out_bc_noaddr(TCGContext * s,int insn)219 static inline void tcg_out_bc_noaddr(TCGContext *s, int insn)
220 {
221 unsigned retrans = *s->code_ptr & 0xfffc;
222 tcg_out32(s, insn | retrans);
223 }
224
225 /* parse target specific constraints */
target_parse_constraint(TCGArgConstraint * ct,const char * ct_str,TCGType type)226 static const char *target_parse_constraint(TCGArgConstraint *ct,
227 const char *ct_str, TCGType type)
228 {
229 switch (*ct_str++) {
230 case 'A': case 'B': case 'C': case 'D':
231 ct->ct |= TCG_CT_REG;
232 tcg_regset_set_reg(ct->u.regs, 3 + ct_str[0] - 'A');
233 break;
234 case 'r':
235 ct->ct |= TCG_CT_REG;
236 ct->u.regs = 0xffffffff;
237 break;
238 case 'L': /* qemu_ld constraint */
239 ct->ct |= TCG_CT_REG;
240 ct->u.regs = 0xffffffff;
241 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3);
242 #ifdef CONFIG_SOFTMMU
243 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R4);
244 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R5);
245 #endif
246 break;
247 case 'S': /* qemu_st constraint */
248 ct->ct |= TCG_CT_REG;
249 ct->u.regs = 0xffffffff;
250 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3);
251 #ifdef CONFIG_SOFTMMU
252 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R4);
253 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R5);
254 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R6);
255 #endif
256 break;
257 case 'I':
258 ct->ct |= TCG_CT_CONST_S16;
259 break;
260 case 'J':
261 ct->ct |= TCG_CT_CONST_U16;
262 break;
263 case 'M':
264 ct->ct |= TCG_CT_CONST_MONE;
265 break;
266 case 'T':
267 ct->ct |= TCG_CT_CONST_S32;
268 break;
269 case 'U':
270 ct->ct |= TCG_CT_CONST_U32;
271 break;
272 case 'W':
273 ct->ct |= TCG_CT_CONST_WSZ;
274 break;
275 case 'Z':
276 ct->ct |= TCG_CT_CONST_ZERO;
277 break;
278 default:
279 return NULL;
280 }
281 return ct_str;
282 }
283
284 /* test if a constant matches the constraint */
tcg_target_const_match(tcg_target_long val,TCGType type,const TCGArgConstraint * arg_ct)285 static int tcg_target_const_match(tcg_target_long val, TCGType type,
286 const TCGArgConstraint *arg_ct)
287 {
288 int ct = arg_ct->ct;
289 if (ct & TCG_CT_CONST) {
290 return 1;
291 }
292
293 /* The only 32-bit constraint we use aside from
294 TCG_CT_CONST is TCG_CT_CONST_S16. */
295 if (type == TCG_TYPE_I32) {
296 val = (int32_t)val;
297 }
298
299 if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
300 return 1;
301 } else if ((ct & TCG_CT_CONST_U16) && val == (uint16_t)val) {
302 return 1;
303 } else if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
304 return 1;
305 } else if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
306 return 1;
307 } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
308 return 1;
309 } else if ((ct & TCG_CT_CONST_MONE) && val == -1) {
310 return 1;
311 } else if ((ct & TCG_CT_CONST_WSZ)
312 && val == (type == TCG_TYPE_I32 ? 32 : 64)) {
313 return 1;
314 }
315 return 0;
316 }
317
318 #define OPCD(opc) ((opc)<<26)
319 #define XO19(opc) (OPCD(19)|((opc)<<1))
320 #define MD30(opc) (OPCD(30)|((opc)<<2))
321 #define MDS30(opc) (OPCD(30)|((opc)<<1))
322 #define XO31(opc) (OPCD(31)|((opc)<<1))
323 #define XO58(opc) (OPCD(58)|(opc))
324 #define XO62(opc) (OPCD(62)|(opc))
325
326 #define B OPCD( 18)
327 #define BC OPCD( 16)
328 #define LBZ OPCD( 34)
329 #define LHZ OPCD( 40)
330 #define LHA OPCD( 42)
331 #define LWZ OPCD( 32)
332 #define STB OPCD( 38)
333 #define STH OPCD( 44)
334 #define STW OPCD( 36)
335
336 #define STD XO62( 0)
337 #define STDU XO62( 1)
338 #define STDX XO31(149)
339
340 #define LD XO58( 0)
341 #define LDX XO31( 21)
342 #define LDU XO58( 1)
343 #define LWA XO58( 2)
344 #define LWAX XO31(341)
345
346 #define ADDIC OPCD( 12)
347 #define ADDI OPCD( 14)
348 #define ADDIS OPCD( 15)
349 #define ORI OPCD( 24)
350 #define ORIS OPCD( 25)
351 #define XORI OPCD( 26)
352 #define XORIS OPCD( 27)
353 #define ANDI OPCD( 28)
354 #define ANDIS OPCD( 29)
355 #define MULLI OPCD( 7)
356 #define CMPLI OPCD( 10)
357 #define CMPI OPCD( 11)
358 #define SUBFIC OPCD( 8)
359
360 #define LWZU OPCD( 33)
361 #define STWU OPCD( 37)
362
363 #define RLWIMI OPCD( 20)
364 #define RLWINM OPCD( 21)
365 #define RLWNM OPCD( 23)
366
367 #define RLDICL MD30( 0)
368 #define RLDICR MD30( 1)
369 #define RLDIMI MD30( 3)
370 #define RLDCL MDS30( 8)
371
372 #define BCLR XO19( 16)
373 #define BCCTR XO19(528)
374 #define CRAND XO19(257)
375 #define CRANDC XO19(129)
376 #define CRNAND XO19(225)
377 #define CROR XO19(449)
378 #define CRNOR XO19( 33)
379
380 #define EXTSB XO31(954)
381 #define EXTSH XO31(922)
382 #define EXTSW XO31(986)
383 #define ADD XO31(266)
384 #define ADDE XO31(138)
385 #define ADDME XO31(234)
386 #define ADDZE XO31(202)
387 #define ADDC XO31( 10)
388 #define AND XO31( 28)
389 #define SUBF XO31( 40)
390 #define SUBFC XO31( 8)
391 #define SUBFE XO31(136)
392 #define SUBFME XO31(232)
393 #define SUBFZE XO31(200)
394 #define OR XO31(444)
395 #define XOR XO31(316)
396 #define MULLW XO31(235)
397 #define MULHW XO31( 75)
398 #define MULHWU XO31( 11)
399 #define DIVW XO31(491)
400 #define DIVWU XO31(459)
401 #define CMP XO31( 0)
402 #define CMPL XO31( 32)
403 #define LHBRX XO31(790)
404 #define LWBRX XO31(534)
405 #define LDBRX XO31(532)
406 #define STHBRX XO31(918)
407 #define STWBRX XO31(662)
408 #define STDBRX XO31(660)
409 #define MFSPR XO31(339)
410 #define MTSPR XO31(467)
411 #define SRAWI XO31(824)
412 #define NEG XO31(104)
413 #define MFCR XO31( 19)
414 #define MFOCRF (MFCR | (1u << 20))
415 #define NOR XO31(124)
416 #define CNTLZW XO31( 26)
417 #define CNTLZD XO31( 58)
418 #define CNTTZW XO31(538)
419 #define CNTTZD XO31(570)
420 #define CNTPOPW XO31(378)
421 #define CNTPOPD XO31(506)
422 #define ANDC XO31( 60)
423 #define ORC XO31(412)
424 #define EQV XO31(284)
425 #define NAND XO31(476)
426 #define ISEL XO31( 15)
427
428 #define MULLD XO31(233)
429 #define MULHD XO31( 73)
430 #define MULHDU XO31( 9)
431 #define DIVD XO31(489)
432 #define DIVDU XO31(457)
433
434 #define LBZX XO31( 87)
435 #define LHZX XO31(279)
436 #define LHAX XO31(343)
437 #define LWZX XO31( 23)
438 #define STBX XO31(215)
439 #define STHX XO31(407)
440 #define STWX XO31(151)
441
442 #define EIEIO XO31(854)
443 #define HWSYNC XO31(598)
444 #define LWSYNC (HWSYNC | (1u << 21))
445
446 #define SPR(a, b) ((((a)<<5)|(b))<<11)
447 #define LR SPR(8, 0)
448 #define CTR SPR(9, 0)
449
450 #define SLW XO31( 24)
451 #define SRW XO31(536)
452 #define SRAW XO31(792)
453
454 #define SLD XO31( 27)
455 #define SRD XO31(539)
456 #define SRAD XO31(794)
457 #define SRADI XO31(413<<1)
458
459 #define TW XO31( 4)
460 #define TRAP (TW | TO(31))
461
462 #define NOP ORI /* ori 0,0,0 */
463
464 #define RT(r) ((r)<<21)
465 #define RS(r) ((r)<<21)
466 #define RA(r) ((r)<<16)
467 #define RB(r) ((r)<<11)
468 #define TO(t) ((t)<<21)
469 #define SH(s) ((s)<<11)
470 #define MB(b) ((b)<<6)
471 #define ME(e) ((e)<<1)
472 #define BO(o) ((o)<<21)
473 #define MB64(b) ((b)<<5)
474 #define FXM(b) (1 << (19 - (b)))
475
476 #define LK 1
477
478 #define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
479 #define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
480 #define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
481 #define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
482
483 #define BF(n) ((n)<<23)
484 #define BI(n, c) (((c)+((n)*4))<<16)
485 #define BT(n, c) (((c)+((n)*4))<<21)
486 #define BA(n, c) (((c)+((n)*4))<<16)
487 #define BB(n, c) (((c)+((n)*4))<<11)
488 #define BC_(n, c) (((c)+((n)*4))<<6)
489
490 #define BO_COND_TRUE BO(12)
491 #define BO_COND_FALSE BO( 4)
492 #define BO_ALWAYS BO(20)
493
494 enum {
495 CR_LT,
496 CR_GT,
497 CR_EQ,
498 CR_SO
499 };
500
501 static const uint32_t tcg_to_bc[] = {
502 [TCG_COND_EQ] = BC | BI(7, CR_EQ) | BO_COND_TRUE,
503 [TCG_COND_NE] = BC | BI(7, CR_EQ) | BO_COND_FALSE,
504 [TCG_COND_LT] = BC | BI(7, CR_LT) | BO_COND_TRUE,
505 [TCG_COND_GE] = BC | BI(7, CR_LT) | BO_COND_FALSE,
506 [TCG_COND_LE] = BC | BI(7, CR_GT) | BO_COND_FALSE,
507 [TCG_COND_GT] = BC | BI(7, CR_GT) | BO_COND_TRUE,
508 [TCG_COND_LTU] = BC | BI(7, CR_LT) | BO_COND_TRUE,
509 [TCG_COND_GEU] = BC | BI(7, CR_LT) | BO_COND_FALSE,
510 [TCG_COND_LEU] = BC | BI(7, CR_GT) | BO_COND_FALSE,
511 [TCG_COND_GTU] = BC | BI(7, CR_GT) | BO_COND_TRUE,
512 };
513
514 /* The low bit here is set if the RA and RB fields must be inverted. */
515 static const uint32_t tcg_to_isel[] = {
516 [TCG_COND_EQ] = ISEL | BC_(7, CR_EQ),
517 [TCG_COND_NE] = ISEL | BC_(7, CR_EQ) | 1,
518 [TCG_COND_LT] = ISEL | BC_(7, CR_LT),
519 [TCG_COND_GE] = ISEL | BC_(7, CR_LT) | 1,
520 [TCG_COND_LE] = ISEL | BC_(7, CR_GT) | 1,
521 [TCG_COND_GT] = ISEL | BC_(7, CR_GT),
522 [TCG_COND_LTU] = ISEL | BC_(7, CR_LT),
523 [TCG_COND_GEU] = ISEL | BC_(7, CR_LT) | 1,
524 [TCG_COND_LEU] = ISEL | BC_(7, CR_GT) | 1,
525 [TCG_COND_GTU] = ISEL | BC_(7, CR_GT),
526 };
527
patch_reloc(tcg_insn_unit * code_ptr,int type,intptr_t value,intptr_t addend)528 static void patch_reloc(tcg_insn_unit *code_ptr, int type,
529 intptr_t value, intptr_t addend)
530 {
531 tcg_insn_unit *target;
532 tcg_insn_unit old;
533
534 value += addend;
535 target = (tcg_insn_unit *)value;
536
537 switch (type) {
538 case R_PPC_REL14:
539 reloc_pc14(code_ptr, target);
540 break;
541 case R_PPC_REL24:
542 reloc_pc24(code_ptr, target);
543 break;
544 case R_PPC_ADDR16:
545 /* We are abusing this relocation type. This points to a pair
546 of insns, addis + load. If the displacement is small, we
547 can nop out the addis. */
548 if (value == (int16_t)value) {
549 code_ptr[0] = NOP;
550 old = deposit32(code_ptr[1], 0, 16, value);
551 code_ptr[1] = deposit32(old, 16, 5, TCG_REG_TB);
552 } else {
553 int16_t lo = value;
554 int hi = value - lo;
555 assert(hi + lo == value);
556 code_ptr[0] = deposit32(code_ptr[0], 0, 16, hi >> 16);
557 code_ptr[1] = deposit32(code_ptr[1], 0, 16, lo);
558 }
559 break;
560 default:
561 g_assert_not_reached();
562 }
563 }
564
565 static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
566 TCGReg base, tcg_target_long offset);
567
tcg_out_mov(TCGContext * s,TCGType type,TCGReg ret,TCGReg arg)568 static void tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
569 {
570 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
571 if (ret != arg) {
572 tcg_out32(s, OR | SAB(arg, ret, arg));
573 }
574 }
575
tcg_out_rld(TCGContext * s,int op,TCGReg ra,TCGReg rs,int sh,int mb)576 static inline void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs,
577 int sh, int mb)
578 {
579 tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
580 sh = SH(sh & 0x1f) | (((sh >> 5) & 1) << 1);
581 mb = MB64((mb >> 5) | ((mb << 1) & 0x3f));
582 tcg_out32(s, op | RA(ra) | RS(rs) | sh | mb);
583 }
584
tcg_out_rlw(TCGContext * s,int op,TCGReg ra,TCGReg rs,int sh,int mb,int me)585 static inline void tcg_out_rlw(TCGContext *s, int op, TCGReg ra, TCGReg rs,
586 int sh, int mb, int me)
587 {
588 tcg_out32(s, op | RA(ra) | RS(rs) | SH(sh) | MB(mb) | ME(me));
589 }
590
tcg_out_ext32u(TCGContext * s,TCGReg dst,TCGReg src)591 static inline void tcg_out_ext32u(TCGContext *s, TCGReg dst, TCGReg src)
592 {
593 tcg_out_rld(s, RLDICL, dst, src, 0, 32);
594 }
595
tcg_out_shli32(TCGContext * s,TCGReg dst,TCGReg src,int c)596 static inline void tcg_out_shli32(TCGContext *s, TCGReg dst, TCGReg src, int c)
597 {
598 tcg_out_rlw(s, RLWINM, dst, src, c, 0, 31 - c);
599 }
600
tcg_out_shli64(TCGContext * s,TCGReg dst,TCGReg src,int c)601 static inline void tcg_out_shli64(TCGContext *s, TCGReg dst, TCGReg src, int c)
602 {
603 tcg_out_rld(s, RLDICR, dst, src, c, 63 - c);
604 }
605
tcg_out_shri32(TCGContext * s,TCGReg dst,TCGReg src,int c)606 static inline void tcg_out_shri32(TCGContext *s, TCGReg dst, TCGReg src, int c)
607 {
608 tcg_out_rlw(s, RLWINM, dst, src, 32 - c, c, 31);
609 }
610
tcg_out_shri64(TCGContext * s,TCGReg dst,TCGReg src,int c)611 static inline void tcg_out_shri64(TCGContext *s, TCGReg dst, TCGReg src, int c)
612 {
613 tcg_out_rld(s, RLDICL, dst, src, 64 - c, c);
614 }
615
616 /* Emit a move into ret of arg, if it can be done in one insn. */
tcg_out_movi_one(TCGContext * s,TCGReg ret,tcg_target_long arg)617 static bool tcg_out_movi_one(TCGContext *s, TCGReg ret, tcg_target_long arg)
618 {
619 if (arg == (int16_t)arg) {
620 tcg_out32(s, ADDI | TAI(ret, 0, arg));
621 return true;
622 }
623 if (arg == (int32_t)arg && (arg & 0xffff) == 0) {
624 tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
625 return true;
626 }
627 return false;
628 }
629
tcg_out_movi_int(TCGContext * s,TCGType type,TCGReg ret,tcg_target_long arg,bool in_prologue)630 static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
631 tcg_target_long arg, bool in_prologue)
632 {
633 intptr_t tb_diff;
634 tcg_target_long tmp;
635 int shift;
636
637 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
638
639 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
640 arg = (int32_t)arg;
641 }
642
643 /* Load 16-bit immediates with one insn. */
644 if (tcg_out_movi_one(s, ret, arg)) {
645 return;
646 }
647
648 /* Load addresses within the TB with one insn. */
649 tb_diff = arg - (intptr_t)s->code_gen_ptr;
650 if (!in_prologue && USE_REG_TB && tb_diff == (int16_t)tb_diff) {
651 tcg_out32(s, ADDI | TAI(ret, TCG_REG_TB, tb_diff));
652 return;
653 }
654
655 /* Load 32-bit immediates with two insns. Note that we've already
656 eliminated bare ADDIS, so we know both insns are required. */
657 if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) {
658 tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
659 tcg_out32(s, ORI | SAI(ret, ret, arg));
660 return;
661 }
662 if (arg == (uint32_t)arg && !(arg & 0x8000)) {
663 tcg_out32(s, ADDI | TAI(ret, 0, arg));
664 tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
665 return;
666 }
667
668 /* Load masked 16-bit value. */
669 if (arg > 0 && (arg & 0x8000)) {
670 tmp = arg | 0x7fff;
671 if ((tmp & (tmp + 1)) == 0) {
672 int mb = clz64(tmp + 1) + 1;
673 tcg_out32(s, ADDI | TAI(ret, 0, arg));
674 tcg_out_rld(s, RLDICL, ret, ret, 0, mb);
675 return;
676 }
677 }
678
679 /* Load common masks with 2 insns. */
680 shift = ctz64(arg);
681 tmp = arg >> shift;
682 if (tmp == (int16_t)tmp) {
683 tcg_out32(s, ADDI | TAI(ret, 0, tmp));
684 tcg_out_shli64(s, ret, ret, shift);
685 return;
686 }
687 shift = clz64(arg);
688 if (tcg_out_movi_one(s, ret, arg << shift)) {
689 tcg_out_shri64(s, ret, ret, shift);
690 return;
691 }
692
693 /* Load addresses within 2GB of TB with 2 (or rarely 3) insns. */
694 if (!in_prologue && USE_REG_TB && tb_diff == (int32_t)tb_diff) {
695 tcg_out_mem_long(s, ADDI, ADD, ret, TCG_REG_TB, tb_diff);
696 return;
697 }
698
699 /* Use the constant pool, if possible. */
700 if (!in_prologue && USE_REG_TB) {
701 new_pool_label(s, arg, R_PPC_ADDR16, s->code_ptr,
702 -(intptr_t)s->code_gen_ptr);
703 tcg_out32(s, ADDIS | TAI(ret, TCG_REG_TB, 0));
704 tcg_out32(s, LD | TAI(ret, ret, 0));
705 return;
706 }
707
708 tmp = arg >> 31 >> 1;
709 tcg_out_movi(s, TCG_TYPE_I32, ret, tmp);
710 if (tmp) {
711 tcg_out_shli64(s, ret, ret, 32);
712 }
713 if (arg & 0xffff0000) {
714 tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
715 }
716 if (arg & 0xffff) {
717 tcg_out32(s, ORI | SAI(ret, ret, arg));
718 }
719 }
720
tcg_out_movi(TCGContext * s,TCGType type,TCGReg ret,tcg_target_long arg)721 static inline void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,
722 tcg_target_long arg)
723 {
724 tcg_out_movi_int(s, type, ret, arg, false);
725 }
726
mask_operand(uint32_t c,int * mb,int * me)727 static bool mask_operand(uint32_t c, int *mb, int *me)
728 {
729 uint32_t lsb, test;
730
731 /* Accept a bit pattern like:
732 0....01....1
733 1....10....0
734 0..01..10..0
735 Keep track of the transitions. */
736 if (c == 0 || c == -1) {
737 return false;
738 }
739 test = c;
740 lsb = test & -test;
741 test += lsb;
742 if (test & (test - 1)) {
743 return false;
744 }
745
746 *me = clz32(lsb);
747 *mb = test ? clz32(test & -test) + 1 : 0;
748 return true;
749 }
750
mask64_operand(uint64_t c,int * mb,int * me)751 static bool mask64_operand(uint64_t c, int *mb, int *me)
752 {
753 uint64_t lsb;
754
755 if (c == 0) {
756 return false;
757 }
758
759 lsb = c & -c;
760 /* Accept 1..10..0. */
761 if (c == -lsb) {
762 *mb = 0;
763 *me = clz64(lsb);
764 return true;
765 }
766 /* Accept 0..01..1. */
767 if (lsb == 1 && (c & (c + 1)) == 0) {
768 *mb = clz64(c + 1) + 1;
769 *me = 63;
770 return true;
771 }
772 return false;
773 }
774
tcg_out_andi32(TCGContext * s,TCGReg dst,TCGReg src,uint32_t c)775 static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
776 {
777 int mb, me;
778
779 if (mask_operand(c, &mb, &me)) {
780 tcg_out_rlw(s, RLWINM, dst, src, 0, mb, me);
781 } else if ((c & 0xffff) == c) {
782 tcg_out32(s, ANDI | SAI(src, dst, c));
783 return;
784 } else if ((c & 0xffff0000) == c) {
785 tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
786 return;
787 } else {
788 tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R0, c);
789 tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
790 }
791 }
792
tcg_out_andi64(TCGContext * s,TCGReg dst,TCGReg src,uint64_t c)793 static void tcg_out_andi64(TCGContext *s, TCGReg dst, TCGReg src, uint64_t c)
794 {
795 int mb, me;
796
797 tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
798 if (mask64_operand(c, &mb, &me)) {
799 if (mb == 0) {
800 tcg_out_rld(s, RLDICR, dst, src, 0, me);
801 } else {
802 tcg_out_rld(s, RLDICL, dst, src, 0, mb);
803 }
804 } else if ((c & 0xffff) == c) {
805 tcg_out32(s, ANDI | SAI(src, dst, c));
806 return;
807 } else if ((c & 0xffff0000) == c) {
808 tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
809 return;
810 } else {
811 tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, c);
812 tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
813 }
814 }
815
tcg_out_zori32(TCGContext * s,TCGReg dst,TCGReg src,uint32_t c,int op_lo,int op_hi)816 static void tcg_out_zori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c,
817 int op_lo, int op_hi)
818 {
819 if (c >> 16) {
820 tcg_out32(s, op_hi | SAI(src, dst, c >> 16));
821 src = dst;
822 }
823 if (c & 0xffff) {
824 tcg_out32(s, op_lo | SAI(src, dst, c));
825 src = dst;
826 }
827 }
828
tcg_out_ori32(TCGContext * s,TCGReg dst,TCGReg src,uint32_t c)829 static void tcg_out_ori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
830 {
831 tcg_out_zori32(s, dst, src, c, ORI, ORIS);
832 }
833
tcg_out_xori32(TCGContext * s,TCGReg dst,TCGReg src,uint32_t c)834 static void tcg_out_xori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
835 {
836 tcg_out_zori32(s, dst, src, c, XORI, XORIS);
837 }
838
tcg_out_b(TCGContext * s,int mask,tcg_insn_unit * target)839 static void tcg_out_b(TCGContext *s, int mask, tcg_insn_unit *target)
840 {
841 ptrdiff_t disp = tcg_pcrel_diff(s, target);
842 if (in_range_b(disp)) {
843 tcg_out32(s, B | (disp & 0x3fffffc) | mask);
844 } else {
845 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, (uintptr_t)target);
846 tcg_out32(s, MTSPR | RS(TCG_REG_R0) | CTR);
847 tcg_out32(s, BCCTR | BO_ALWAYS | mask);
848 }
849 }
850
tcg_out_mem_long(TCGContext * s,int opi,int opx,TCGReg rt,TCGReg base,tcg_target_long offset)851 static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
852 TCGReg base, tcg_target_long offset)
853 {
854 tcg_target_long orig = offset, l0, l1, extra = 0, align = 0;
855 bool is_store = false;
856 TCGReg rs = TCG_REG_TMP1;
857
858 switch (opi) {
859 case LD: case LWA:
860 align = 3;
861 /* FALLTHRU */
862 default:
863 if (rt != TCG_REG_R0) {
864 rs = rt;
865 break;
866 }
867 break;
868 case STD:
869 align = 3;
870 /* FALLTHRU */
871 case STB: case STH: case STW:
872 is_store = true;
873 break;
874 }
875
876 /* For unaligned, or very large offsets, use the indexed form. */
877 if (offset & align || offset != (int32_t)offset) {
878 if (rs == base) {
879 rs = TCG_REG_R0;
880 }
881 tcg_debug_assert(!is_store || rs != rt);
882 tcg_out_movi(s, TCG_TYPE_PTR, rs, orig);
883 tcg_out32(s, opx | TAB(rt, base, rs));
884 return;
885 }
886
887 l0 = (int16_t)offset;
888 offset = (offset - l0) >> 16;
889 l1 = (int16_t)offset;
890
891 if (l1 < 0 && orig >= 0) {
892 extra = 0x4000;
893 l1 = (int16_t)(offset - 0x4000);
894 }
895 if (l1) {
896 tcg_out32(s, ADDIS | TAI(rs, base, l1));
897 base = rs;
898 }
899 if (extra) {
900 tcg_out32(s, ADDIS | TAI(rs, base, extra));
901 base = rs;
902 }
903 if (opi != ADDI || base != rt || l0 != 0) {
904 tcg_out32(s, opi | TAI(rt, base, l0));
905 }
906 }
907
tcg_out_ld(TCGContext * s,TCGType type,TCGReg ret,TCGReg arg1,intptr_t arg2)908 static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
909 TCGReg arg1, intptr_t arg2)
910 {
911 int opi, opx;
912
913 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
914 if (type == TCG_TYPE_I32) {
915 opi = LWZ, opx = LWZX;
916 } else {
917 opi = LD, opx = LDX;
918 }
919 tcg_out_mem_long(s, opi, opx, ret, arg1, arg2);
920 }
921
tcg_out_st(TCGContext * s,TCGType type,TCGReg arg,TCGReg arg1,intptr_t arg2)922 static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
923 TCGReg arg1, intptr_t arg2)
924 {
925 int opi, opx;
926
927 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
928 if (type == TCG_TYPE_I32) {
929 opi = STW, opx = STWX;
930 } else {
931 opi = STD, opx = STDX;
932 }
933 tcg_out_mem_long(s, opi, opx, arg, arg1, arg2);
934 }
935
tcg_out_sti(TCGContext * s,TCGType type,TCGArg val,TCGReg base,intptr_t ofs)936 static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
937 TCGReg base, intptr_t ofs)
938 {
939 return false;
940 }
941
tcg_out_cmp(TCGContext * s,int cond,TCGArg arg1,TCGArg arg2,int const_arg2,int cr,TCGType type)942 static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2,
943 int const_arg2, int cr, TCGType type)
944 {
945 int imm;
946 uint32_t op;
947
948 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
949
950 /* Simplify the comparisons below wrt CMPI. */
951 if (type == TCG_TYPE_I32) {
952 arg2 = (int32_t)arg2;
953 }
954
955 switch (cond) {
956 case TCG_COND_EQ:
957 case TCG_COND_NE:
958 if (const_arg2) {
959 if ((int16_t) arg2 == arg2) {
960 op = CMPI;
961 imm = 1;
962 break;
963 } else if ((uint16_t) arg2 == arg2) {
964 op = CMPLI;
965 imm = 1;
966 break;
967 }
968 }
969 op = CMPL;
970 imm = 0;
971 break;
972
973 case TCG_COND_LT:
974 case TCG_COND_GE:
975 case TCG_COND_LE:
976 case TCG_COND_GT:
977 if (const_arg2) {
978 if ((int16_t) arg2 == arg2) {
979 op = CMPI;
980 imm = 1;
981 break;
982 }
983 }
984 op = CMP;
985 imm = 0;
986 break;
987
988 case TCG_COND_LTU:
989 case TCG_COND_GEU:
990 case TCG_COND_LEU:
991 case TCG_COND_GTU:
992 if (const_arg2) {
993 if ((uint16_t) arg2 == arg2) {
994 op = CMPLI;
995 imm = 1;
996 break;
997 }
998 }
999 op = CMPL;
1000 imm = 0;
1001 break;
1002
1003 default:
1004 tcg_abort();
1005 }
1006 op |= BF(cr) | ((type == TCG_TYPE_I64) << 21);
1007
1008 if (imm) {
1009 tcg_out32(s, op | RA(arg1) | (arg2 & 0xffff));
1010 } else {
1011 if (const_arg2) {
1012 tcg_out_movi(s, type, TCG_REG_R0, arg2);
1013 arg2 = TCG_REG_R0;
1014 }
1015 tcg_out32(s, op | RA(arg1) | RB(arg2));
1016 }
1017 }
1018
tcg_out_setcond_eq0(TCGContext * s,TCGType type,TCGReg dst,TCGReg src)1019 static void tcg_out_setcond_eq0(TCGContext *s, TCGType type,
1020 TCGReg dst, TCGReg src)
1021 {
1022 if (type == TCG_TYPE_I32) {
1023 tcg_out32(s, CNTLZW | RS(src) | RA(dst));
1024 tcg_out_shri32(s, dst, dst, 5);
1025 } else {
1026 tcg_out32(s, CNTLZD | RS(src) | RA(dst));
1027 tcg_out_shri64(s, dst, dst, 6);
1028 }
1029 }
1030
tcg_out_setcond_ne0(TCGContext * s,TCGReg dst,TCGReg src)1031 static void tcg_out_setcond_ne0(TCGContext *s, TCGReg dst, TCGReg src)
1032 {
1033 /* X != 0 implies X + -1 generates a carry. Extra addition
1034 trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C. */
1035 if (dst != src) {
1036 tcg_out32(s, ADDIC | TAI(dst, src, -1));
1037 tcg_out32(s, SUBFE | TAB(dst, dst, src));
1038 } else {
1039 tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1));
1040 tcg_out32(s, SUBFE | TAB(dst, TCG_REG_R0, src));
1041 }
1042 }
1043
tcg_gen_setcond_xor(TCGContext * s,TCGReg arg1,TCGArg arg2,bool const_arg2)1044 static TCGReg tcg_gen_setcond_xor(TCGContext *s, TCGReg arg1, TCGArg arg2,
1045 bool const_arg2)
1046 {
1047 if (const_arg2) {
1048 if ((uint32_t)arg2 == arg2) {
1049 tcg_out_xori32(s, TCG_REG_R0, arg1, arg2);
1050 } else {
1051 tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, arg2);
1052 tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, TCG_REG_R0));
1053 }
1054 } else {
1055 tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, arg2));
1056 }
1057 return TCG_REG_R0;
1058 }
1059
tcg_out_setcond(TCGContext * s,TCGType type,TCGCond cond,TCGArg arg0,TCGArg arg1,TCGArg arg2,int const_arg2)1060 static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
1061 TCGArg arg0, TCGArg arg1, TCGArg arg2,
1062 int const_arg2)
1063 {
1064 int crop, sh;
1065
1066 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1067
1068 /* Ignore high bits of a potential constant arg2. */
1069 if (type == TCG_TYPE_I32) {
1070 arg2 = (uint32_t)arg2;
1071 }
1072
1073 /* Handle common and trivial cases before handling anything else. */
1074 if (arg2 == 0) {
1075 switch (cond) {
1076 case TCG_COND_EQ:
1077 tcg_out_setcond_eq0(s, type, arg0, arg1);
1078 return;
1079 case TCG_COND_NE:
1080 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1081 tcg_out_ext32u(s, TCG_REG_R0, arg1);
1082 arg1 = TCG_REG_R0;
1083 }
1084 tcg_out_setcond_ne0(s, arg0, arg1);
1085 return;
1086 case TCG_COND_GE:
1087 tcg_out32(s, NOR | SAB(arg1, arg0, arg1));
1088 arg1 = arg0;
1089 /* FALLTHRU */
1090 case TCG_COND_LT:
1091 /* Extract the sign bit. */
1092 if (type == TCG_TYPE_I32) {
1093 tcg_out_shri32(s, arg0, arg1, 31);
1094 } else {
1095 tcg_out_shri64(s, arg0, arg1, 63);
1096 }
1097 return;
1098 default:
1099 break;
1100 }
1101 }
1102
1103 /* If we have ISEL, we can implement everything with 3 or 4 insns.
1104 All other cases below are also at least 3 insns, so speed up the
1105 code generator by not considering them and always using ISEL. */
1106 if (HAVE_ISEL) {
1107 int isel, tab;
1108
1109 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1110
1111 isel = tcg_to_isel[cond];
1112
1113 tcg_out_movi(s, type, arg0, 1);
1114 if (isel & 1) {
1115 /* arg0 = (bc ? 0 : 1) */
1116 tab = TAB(arg0, 0, arg0);
1117 isel &= ~1;
1118 } else {
1119 /* arg0 = (bc ? 1 : 0) */
1120 tcg_out_movi(s, type, TCG_REG_R0, 0);
1121 tab = TAB(arg0, arg0, TCG_REG_R0);
1122 }
1123 tcg_out32(s, isel | tab);
1124 return;
1125 }
1126
1127 switch (cond) {
1128 case TCG_COND_EQ:
1129 arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1130 tcg_out_setcond_eq0(s, type, arg0, arg1);
1131 return;
1132
1133 case TCG_COND_NE:
1134 arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1135 /* Discard the high bits only once, rather than both inputs. */
1136 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1137 tcg_out_ext32u(s, TCG_REG_R0, arg1);
1138 arg1 = TCG_REG_R0;
1139 }
1140 tcg_out_setcond_ne0(s, arg0, arg1);
1141 return;
1142
1143 case TCG_COND_GT:
1144 case TCG_COND_GTU:
1145 sh = 30;
1146 crop = 0;
1147 goto crtest;
1148
1149 case TCG_COND_LT:
1150 case TCG_COND_LTU:
1151 sh = 29;
1152 crop = 0;
1153 goto crtest;
1154
1155 case TCG_COND_GE:
1156 case TCG_COND_GEU:
1157 sh = 31;
1158 crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_LT) | BB(7, CR_LT);
1159 goto crtest;
1160
1161 case TCG_COND_LE:
1162 case TCG_COND_LEU:
1163 sh = 31;
1164 crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_GT) | BB(7, CR_GT);
1165 crtest:
1166 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1167 if (crop) {
1168 tcg_out32(s, crop);
1169 }
1170 tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1171 tcg_out_rlw(s, RLWINM, arg0, TCG_REG_R0, sh, 31, 31);
1172 break;
1173
1174 default:
1175 tcg_abort();
1176 }
1177 }
1178
tcg_out_bc(TCGContext * s,int bc,TCGLabel * l)1179 static void tcg_out_bc(TCGContext *s, int bc, TCGLabel *l)
1180 {
1181 if (l->has_value) {
1182 tcg_out32(s, bc | reloc_pc14_val(s->code_ptr, l->u.value_ptr));
1183 } else {
1184 tcg_out_reloc(s, s->code_ptr, R_PPC_REL14, l, 0);
1185 tcg_out_bc_noaddr(s, bc);
1186 }
1187 }
1188
tcg_out_brcond(TCGContext * s,TCGCond cond,TCGArg arg1,TCGArg arg2,int const_arg2,TCGLabel * l,TCGType type)1189 static void tcg_out_brcond(TCGContext *s, TCGCond cond,
1190 TCGArg arg1, TCGArg arg2, int const_arg2,
1191 TCGLabel *l, TCGType type)
1192 {
1193 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1194 tcg_out_bc(s, tcg_to_bc[cond], l);
1195 }
1196
tcg_out_movcond(TCGContext * s,TCGType type,TCGCond cond,TCGArg dest,TCGArg c1,TCGArg c2,TCGArg v1,TCGArg v2,bool const_c2)1197 static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
1198 TCGArg dest, TCGArg c1, TCGArg c2, TCGArg v1,
1199 TCGArg v2, bool const_c2)
1200 {
1201 /* If for some reason both inputs are zero, don't produce bad code. */
1202 if (v1 == 0 && v2 == 0) {
1203 tcg_out_movi(s, type, dest, 0);
1204 return;
1205 }
1206
1207 tcg_out_cmp(s, cond, c1, c2, const_c2, 7, type);
1208
1209 if (HAVE_ISEL) {
1210 int isel = tcg_to_isel[cond];
1211
1212 /* Swap the V operands if the operation indicates inversion. */
1213 if (isel & 1) {
1214 int t = v1;
1215 v1 = v2;
1216 v2 = t;
1217 isel &= ~1;
1218 }
1219 /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand. */
1220 if (v2 == 0) {
1221 tcg_out_movi(s, type, TCG_REG_R0, 0);
1222 }
1223 tcg_out32(s, isel | TAB(dest, v1, v2));
1224 } else {
1225 if (dest == v2) {
1226 cond = tcg_invert_cond(cond);
1227 v2 = v1;
1228 } else if (dest != v1) {
1229 if (v1 == 0) {
1230 tcg_out_movi(s, type, dest, 0);
1231 } else {
1232 tcg_out_mov(s, type, dest, v1);
1233 }
1234 }
1235 /* Branch forward over one insn */
1236 tcg_out32(s, tcg_to_bc[cond] | 8);
1237 if (v2 == 0) {
1238 tcg_out_movi(s, type, dest, 0);
1239 } else {
1240 tcg_out_mov(s, type, dest, v2);
1241 }
1242 }
1243 }
1244
tcg_out_cntxz(TCGContext * s,TCGType type,uint32_t opc,TCGArg a0,TCGArg a1,TCGArg a2,bool const_a2)1245 static void tcg_out_cntxz(TCGContext *s, TCGType type, uint32_t opc,
1246 TCGArg a0, TCGArg a1, TCGArg a2, bool const_a2)
1247 {
1248 if (const_a2 && a2 == (type == TCG_TYPE_I32 ? 32 : 64)) {
1249 tcg_out32(s, opc | RA(a0) | RS(a1));
1250 } else {
1251 tcg_out_cmp(s, TCG_COND_EQ, a1, 0, 1, 7, type);
1252 /* Note that the only other valid constant for a2 is 0. */
1253 if (HAVE_ISEL) {
1254 tcg_out32(s, opc | RA(TCG_REG_R0) | RS(a1));
1255 tcg_out32(s, tcg_to_isel[TCG_COND_EQ] | TAB(a0, a2, TCG_REG_R0));
1256 } else if (!const_a2 && a0 == a2) {
1257 tcg_out32(s, tcg_to_bc[TCG_COND_EQ] | 8);
1258 tcg_out32(s, opc | RA(a0) | RS(a1));
1259 } else {
1260 tcg_out32(s, opc | RA(a0) | RS(a1));
1261 tcg_out32(s, tcg_to_bc[TCG_COND_NE] | 8);
1262 if (const_a2) {
1263 tcg_out_movi(s, type, a0, 0);
1264 } else {
1265 tcg_out_mov(s, type, a0, a2);
1266 }
1267 }
1268 }
1269 }
1270
tcg_out_cmp2(TCGContext * s,const TCGArg * args,const int * const_args)1271 static void tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1272 const int *const_args)
1273 {
1274 static const struct { uint8_t bit1, bit2; } bits[] = {
1275 [TCG_COND_LT ] = { CR_LT, CR_LT },
1276 [TCG_COND_LE ] = { CR_LT, CR_GT },
1277 [TCG_COND_GT ] = { CR_GT, CR_GT },
1278 [TCG_COND_GE ] = { CR_GT, CR_LT },
1279 [TCG_COND_LTU] = { CR_LT, CR_LT },
1280 [TCG_COND_LEU] = { CR_LT, CR_GT },
1281 [TCG_COND_GTU] = { CR_GT, CR_GT },
1282 [TCG_COND_GEU] = { CR_GT, CR_LT },
1283 };
1284
1285 TCGCond cond = args[4], cond2;
1286 TCGArg al, ah, bl, bh;
1287 int blconst, bhconst;
1288 int op, bit1, bit2;
1289
1290 al = args[0];
1291 ah = args[1];
1292 bl = args[2];
1293 bh = args[3];
1294 blconst = const_args[2];
1295 bhconst = const_args[3];
1296
1297 switch (cond) {
1298 case TCG_COND_EQ:
1299 op = CRAND;
1300 goto do_equality;
1301 case TCG_COND_NE:
1302 op = CRNAND;
1303 do_equality:
1304 tcg_out_cmp(s, cond, al, bl, blconst, 6, TCG_TYPE_I32);
1305 tcg_out_cmp(s, cond, ah, bh, bhconst, 7, TCG_TYPE_I32);
1306 tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1307 break;
1308
1309 case TCG_COND_LT:
1310 case TCG_COND_LE:
1311 case TCG_COND_GT:
1312 case TCG_COND_GE:
1313 case TCG_COND_LTU:
1314 case TCG_COND_LEU:
1315 case TCG_COND_GTU:
1316 case TCG_COND_GEU:
1317 bit1 = bits[cond].bit1;
1318 bit2 = bits[cond].bit2;
1319 op = (bit1 != bit2 ? CRANDC : CRAND);
1320 cond2 = tcg_unsigned_cond(cond);
1321
1322 tcg_out_cmp(s, cond, ah, bh, bhconst, 6, TCG_TYPE_I32);
1323 tcg_out_cmp(s, cond2, al, bl, blconst, 7, TCG_TYPE_I32);
1324 tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, bit2));
1325 tcg_out32(s, CROR | BT(7, CR_EQ) | BA(6, bit1) | BB(7, CR_EQ));
1326 break;
1327
1328 default:
1329 tcg_abort();
1330 }
1331 }
1332
tcg_out_setcond2(TCGContext * s,const TCGArg * args,const int * const_args)1333 static void tcg_out_setcond2(TCGContext *s, const TCGArg *args,
1334 const int *const_args)
1335 {
1336 tcg_out_cmp2(s, args + 1, const_args + 1);
1337 tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1338 tcg_out_rlw(s, RLWINM, args[0], TCG_REG_R0, 31, 31, 31);
1339 }
1340
tcg_out_brcond2(TCGContext * s,const TCGArg * args,const int * const_args)1341 static void tcg_out_brcond2 (TCGContext *s, const TCGArg *args,
1342 const int *const_args)
1343 {
1344 tcg_out_cmp2(s, args, const_args);
1345 tcg_out_bc(s, BC | BI(7, CR_EQ) | BO_COND_TRUE, arg_label(args[5]));
1346 }
1347
tcg_out_mb(TCGContext * s,TCGArg a0)1348 static void tcg_out_mb(TCGContext *s, TCGArg a0)
1349 {
1350 uint32_t insn = HWSYNC;
1351 a0 &= TCG_MO_ALL;
1352 if (a0 == TCG_MO_LD_LD) {
1353 insn = LWSYNC;
1354 } else if (a0 == TCG_MO_ST_ST) {
1355 insn = EIEIO;
1356 }
1357 tcg_out32(s, insn);
1358 }
1359
tb_target_set_jmp_target(uintptr_t tc_ptr,uintptr_t jmp_addr,uintptr_t addr)1360 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
1361 uintptr_t addr)
1362 {
1363 if (TCG_TARGET_REG_BITS == 64) {
1364 tcg_insn_unit i1, i2;
1365 intptr_t tb_diff = addr - tc_ptr;
1366 intptr_t br_diff = addr - (jmp_addr + 4);
1367 uint64_t pair;
1368
1369 /* This does not exercise the range of the branch, but we do
1370 still need to be able to load the new value of TCG_REG_TB.
1371 But this does still happen quite often. */
1372 if (tb_diff == (int16_t)tb_diff) {
1373 i1 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, tb_diff);
1374 i2 = B | (br_diff & 0x3fffffc);
1375 } else {
1376 intptr_t lo = (int16_t)tb_diff;
1377 intptr_t hi = (int32_t)(tb_diff - lo);
1378 assert(tb_diff == hi + lo);
1379 i1 = ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, hi >> 16);
1380 i2 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, lo);
1381 }
1382 #ifdef HOST_WORDS_BIGENDIAN
1383 pair = (uint64_t)i1 << 32 | i2;
1384 #else
1385 pair = (uint64_t)i2 << 32 | i1;
1386 #endif
1387
1388 /* As per the enclosing if, this is ppc64. Avoid the _Static_assert
1389 within atomic_set that would fail to build a ppc32 host. */
1390 atomic_set__nocheck((uint64_t *)jmp_addr, pair);
1391 flush_icache_range(jmp_addr, jmp_addr + 8);
1392 } else {
1393 intptr_t diff = addr - jmp_addr;
1394 tcg_debug_assert(in_range_b(diff));
1395 atomic_set((uint32_t *)jmp_addr, B | (diff & 0x3fffffc));
1396 flush_icache_range(jmp_addr, jmp_addr + 4);
1397 }
1398 }
1399
tcg_out_call(TCGContext * s,tcg_insn_unit * target)1400 static void tcg_out_call(TCGContext *s, tcg_insn_unit *target)
1401 {
1402 #ifdef _CALL_AIX
1403 /* Look through the descriptor. If the branch is in range, and we
1404 don't have to spend too much effort on building the toc. */
1405 void *tgt = ((void **)target)[0];
1406 uintptr_t toc = ((uintptr_t *)target)[1];
1407 intptr_t diff = tcg_pcrel_diff(s, tgt);
1408
1409 if (in_range_b(diff) && toc == (uint32_t)toc) {
1410 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, toc);
1411 tcg_out_b(s, LK, tgt);
1412 } else {
1413 /* Fold the low bits of the constant into the addresses below. */
1414 intptr_t arg = (intptr_t)target;
1415 int ofs = (int16_t)arg;
1416
1417 if (ofs + 8 < 0x8000) {
1418 arg -= ofs;
1419 } else {
1420 ofs = 0;
1421 }
1422 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, arg);
1423 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_TMP1, ofs);
1424 tcg_out32(s, MTSPR | RA(TCG_REG_R0) | CTR);
1425 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_REG_TMP1, ofs + SZP);
1426 tcg_out32(s, BCCTR | BO_ALWAYS | LK);
1427 }
1428 #elif defined(_CALL_ELF) && _CALL_ELF == 2
1429 intptr_t diff;
1430
1431 /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1432 address, which the callee uses to compute its TOC address. */
1433 /* FIXME: when the branch is in range, we could avoid r12 load if we
1434 knew that the destination uses the same TOC, and what its local
1435 entry point offset is. */
1436 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R12, (intptr_t)target);
1437
1438 diff = tcg_pcrel_diff(s, target);
1439 if (in_range_b(diff)) {
1440 tcg_out_b(s, LK, target);
1441 } else {
1442 tcg_out32(s, MTSPR | RS(TCG_REG_R12) | CTR);
1443 tcg_out32(s, BCCTR | BO_ALWAYS | LK);
1444 }
1445 #else
1446 tcg_out_b(s, LK, target);
1447 #endif
1448 }
1449
1450 static const uint32_t qemu_ldx_opc[16] = {
1451 [MO_UB] = LBZX,
1452 [MO_UW] = LHZX,
1453 [MO_UL] = LWZX,
1454 [MO_Q] = LDX,
1455 [MO_SW] = LHAX,
1456 [MO_SL] = LWAX,
1457 [MO_BSWAP | MO_UB] = LBZX,
1458 [MO_BSWAP | MO_UW] = LHBRX,
1459 [MO_BSWAP | MO_UL] = LWBRX,
1460 [MO_BSWAP | MO_Q] = LDBRX,
1461 };
1462
1463 static const uint32_t qemu_stx_opc[16] = {
1464 [MO_UB] = STBX,
1465 [MO_UW] = STHX,
1466 [MO_UL] = STWX,
1467 [MO_Q] = STDX,
1468 [MO_BSWAP | MO_UB] = STBX,
1469 [MO_BSWAP | MO_UW] = STHBRX,
1470 [MO_BSWAP | MO_UL] = STWBRX,
1471 [MO_BSWAP | MO_Q] = STDBRX,
1472 };
1473
1474 static const uint32_t qemu_exts_opc[4] = {
1475 EXTSB, EXTSH, EXTSW, 0
1476 };
1477
1478 #if defined (CONFIG_SOFTMMU)
1479 #include "tcg-ldst.inc.c"
1480
1481 /* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
1482 * int mmu_idx, uintptr_t ra)
1483 */
1484 static void * const qemu_ld_helpers[16] = {
1485 [MO_UB] = helper_ret_ldub_mmu,
1486 [MO_LEUW] = helper_le_lduw_mmu,
1487 [MO_LEUL] = helper_le_ldul_mmu,
1488 [MO_LEQ] = helper_le_ldq_mmu,
1489 [MO_BEUW] = helper_be_lduw_mmu,
1490 [MO_BEUL] = helper_be_ldul_mmu,
1491 [MO_BEQ] = helper_be_ldq_mmu,
1492 };
1493
1494 /* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
1495 * uintxx_t val, int mmu_idx, uintptr_t ra)
1496 */
1497 static void * const qemu_st_helpers[16] = {
1498 [MO_UB] = helper_ret_stb_mmu,
1499 [MO_LEUW] = helper_le_stw_mmu,
1500 [MO_LEUL] = helper_le_stl_mmu,
1501 [MO_LEQ] = helper_le_stq_mmu,
1502 [MO_BEUW] = helper_be_stw_mmu,
1503 [MO_BEUL] = helper_be_stl_mmu,
1504 [MO_BEQ] = helper_be_stq_mmu,
1505 };
1506
1507 /* Perform the TLB load and compare. Places the result of the comparison
1508 in CR7, loads the addend of the TLB into R3, and returns the register
1509 containing the guest address (zero-extended into R4). Clobbers R0 and R2. */
1510
tcg_out_tlb_read(TCGContext * s,TCGMemOp opc,TCGReg addrlo,TCGReg addrhi,int mem_index,bool is_read)1511 static TCGReg tcg_out_tlb_read(TCGContext *s, TCGMemOp opc,
1512 TCGReg addrlo, TCGReg addrhi,
1513 int mem_index, bool is_read)
1514 {
1515 int cmp_off
1516 = (is_read
1517 ? offsetof(CPUArchState, tlb_table[mem_index][0].addr_read)
1518 : offsetof(CPUArchState, tlb_table[mem_index][0].addr_write));
1519 int add_off = offsetof(CPUArchState, tlb_table[mem_index][0].addend);
1520 TCGReg base = TCG_AREG0;
1521 unsigned s_bits = opc & MO_SIZE;
1522 unsigned a_bits = get_alignment_bits(opc);
1523
1524 /* Extract the page index, shifted into place for tlb index. */
1525 if (TCG_TARGET_REG_BITS == 64) {
1526 if (TARGET_LONG_BITS == 32) {
1527 /* Zero-extend the address into a place helpful for further use. */
1528 tcg_out_ext32u(s, TCG_REG_R4, addrlo);
1529 addrlo = TCG_REG_R4;
1530 } else {
1531 tcg_out_rld(s, RLDICL, TCG_REG_R3, addrlo,
1532 64 - TARGET_PAGE_BITS, 64 - CPU_TLB_BITS);
1533 }
1534 }
1535
1536 /* Compensate for very large offsets. */
1537 if (add_off >= 0x8000) {
1538 int low = (int16_t)cmp_off;
1539 int high = cmp_off - low;
1540 assert((high & 0xffff) == 0);
1541 assert(cmp_off - high == (int16_t)(cmp_off - high));
1542 assert(add_off - high == (int16_t)(add_off - high));
1543 tcg_out32(s, ADDIS | TAI(TCG_REG_TMP1, base, high >> 16));
1544 base = TCG_REG_TMP1;
1545 cmp_off -= high;
1546 add_off -= high;
1547 }
1548
1549 /* Extraction and shifting, part 2. */
1550 if (TCG_TARGET_REG_BITS == 32 || TARGET_LONG_BITS == 32) {
1551 tcg_out_rlw(s, RLWINM, TCG_REG_R3, addrlo,
1552 32 - (TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS),
1553 32 - (CPU_TLB_BITS + CPU_TLB_ENTRY_BITS),
1554 31 - CPU_TLB_ENTRY_BITS);
1555 } else {
1556 tcg_out_shli64(s, TCG_REG_R3, TCG_REG_R3, CPU_TLB_ENTRY_BITS);
1557 }
1558
1559 tcg_out32(s, ADD | TAB(TCG_REG_R3, TCG_REG_R3, base));
1560
1561 /* Load the tlb comparator. */
1562 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1563 tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R4, TCG_REG_R3, cmp_off);
1564 tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP1, TCG_REG_R3, cmp_off + 4);
1565 } else {
1566 tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP1, TCG_REG_R3, cmp_off);
1567 }
1568
1569 /* Load the TLB addend for use on the fast path. Do this asap
1570 to minimize any load use delay. */
1571 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_REG_R3, add_off);
1572
1573 /* Clear the non-page, non-alignment bits from the address */
1574 if (TCG_TARGET_REG_BITS == 32) {
1575 /* We don't support unaligned accesses on 32-bits.
1576 * Preserve the bottom bits and thus trigger a comparison
1577 * failure on unaligned accesses.
1578 */
1579 if (a_bits < s_bits) {
1580 a_bits = s_bits;
1581 }
1582 tcg_out_rlw(s, RLWINM, TCG_REG_R0, addrlo, 0,
1583 (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
1584 } else {
1585 TCGReg t = addrlo;
1586
1587 /* If the access is unaligned, we need to make sure we fail if we
1588 * cross a page boundary. The trick is to add the access size-1
1589 * to the address before masking the low bits. That will make the
1590 * address overflow to the next page if we cross a page boundary,
1591 * which will then force a mismatch of the TLB compare.
1592 */
1593 if (a_bits < s_bits) {
1594 unsigned a_mask = (1 << a_bits) - 1;
1595 unsigned s_mask = (1 << s_bits) - 1;
1596 tcg_out32(s, ADDI | TAI(TCG_REG_R0, t, s_mask - a_mask));
1597 t = TCG_REG_R0;
1598 }
1599
1600 /* Mask the address for the requested alignment. */
1601 if (TARGET_LONG_BITS == 32) {
1602 tcg_out_rlw(s, RLWINM, TCG_REG_R0, t, 0,
1603 (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
1604 } else if (a_bits == 0) {
1605 tcg_out_rld(s, RLDICR, TCG_REG_R0, t, 0, 63 - TARGET_PAGE_BITS);
1606 } else {
1607 tcg_out_rld(s, RLDICL, TCG_REG_R0, t,
1608 64 - TARGET_PAGE_BITS, TARGET_PAGE_BITS - a_bits);
1609 tcg_out_rld(s, RLDICL, TCG_REG_R0, TCG_REG_R0, TARGET_PAGE_BITS, 0);
1610 }
1611 }
1612
1613 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1614 tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
1615 0, 7, TCG_TYPE_I32);
1616 tcg_out_cmp(s, TCG_COND_EQ, addrhi, TCG_REG_R4, 0, 6, TCG_TYPE_I32);
1617 tcg_out32(s, CRAND | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1618 } else {
1619 tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
1620 0, 7, TCG_TYPE_TL);
1621 }
1622
1623 return addrlo;
1624 }
1625
1626 /* Record the context of a call to the out of line helper code for the slow
1627 path for a load or store, so that we can later generate the correct
1628 helper code. */
add_qemu_ldst_label(TCGContext * s,bool is_ld,TCGMemOpIdx oi,TCGReg datalo_reg,TCGReg datahi_reg,TCGReg addrlo_reg,TCGReg addrhi_reg,tcg_insn_unit * raddr,tcg_insn_unit * lptr)1629 static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
1630 TCGReg datalo_reg, TCGReg datahi_reg,
1631 TCGReg addrlo_reg, TCGReg addrhi_reg,
1632 tcg_insn_unit *raddr, tcg_insn_unit *lptr)
1633 {
1634 TCGLabelQemuLdst *label = new_ldst_label(s);
1635
1636 label->is_ld = is_ld;
1637 label->oi = oi;
1638 label->datalo_reg = datalo_reg;
1639 label->datahi_reg = datahi_reg;
1640 label->addrlo_reg = addrlo_reg;
1641 label->addrhi_reg = addrhi_reg;
1642 label->raddr = raddr;
1643 label->label_ptr[0] = lptr;
1644 }
1645
tcg_out_qemu_ld_slow_path(TCGContext * s,TCGLabelQemuLdst * lb)1646 static void tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1647 {
1648 TCGMemOpIdx oi = lb->oi;
1649 TCGMemOp opc = get_memop(oi);
1650 TCGReg hi, lo, arg = TCG_REG_R3;
1651
1652 reloc_pc14(lb->label_ptr[0], s->code_ptr);
1653
1654 tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
1655
1656 lo = lb->addrlo_reg;
1657 hi = lb->addrhi_reg;
1658 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1659 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1660 arg |= 1;
1661 #endif
1662 tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
1663 tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
1664 } else {
1665 /* If the address needed to be zero-extended, we'll have already
1666 placed it in R4. The only remaining case is 64-bit guest. */
1667 tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
1668 }
1669
1670 tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
1671 tcg_out32(s, MFSPR | RT(arg) | LR);
1672
1673 tcg_out_call(s, qemu_ld_helpers[opc & (MO_BSWAP | MO_SIZE)]);
1674
1675 lo = lb->datalo_reg;
1676 hi = lb->datahi_reg;
1677 if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
1678 tcg_out_mov(s, TCG_TYPE_I32, lo, TCG_REG_R4);
1679 tcg_out_mov(s, TCG_TYPE_I32, hi, TCG_REG_R3);
1680 } else if (opc & MO_SIGN) {
1681 uint32_t insn = qemu_exts_opc[opc & MO_SIZE];
1682 tcg_out32(s, insn | RA(lo) | RS(TCG_REG_R3));
1683 } else {
1684 tcg_out_mov(s, TCG_TYPE_REG, lo, TCG_REG_R3);
1685 }
1686
1687 tcg_out_b(s, 0, lb->raddr);
1688 }
1689
tcg_out_qemu_st_slow_path(TCGContext * s,TCGLabelQemuLdst * lb)1690 static void tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1691 {
1692 TCGMemOpIdx oi = lb->oi;
1693 TCGMemOp opc = get_memop(oi);
1694 TCGMemOp s_bits = opc & MO_SIZE;
1695 TCGReg hi, lo, arg = TCG_REG_R3;
1696
1697 reloc_pc14(lb->label_ptr[0], s->code_ptr);
1698
1699 tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
1700
1701 lo = lb->addrlo_reg;
1702 hi = lb->addrhi_reg;
1703 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1704 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1705 arg |= 1;
1706 #endif
1707 tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
1708 tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
1709 } else {
1710 /* If the address needed to be zero-extended, we'll have already
1711 placed it in R4. The only remaining case is 64-bit guest. */
1712 tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
1713 }
1714
1715 lo = lb->datalo_reg;
1716 hi = lb->datahi_reg;
1717 if (TCG_TARGET_REG_BITS == 32) {
1718 switch (s_bits) {
1719 case MO_64:
1720 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1721 arg |= 1;
1722 #endif
1723 tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
1724 /* FALLTHRU */
1725 case MO_32:
1726 tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
1727 break;
1728 default:
1729 tcg_out_rlw(s, RLWINM, arg++, lo, 0, 32 - (8 << s_bits), 31);
1730 break;
1731 }
1732 } else {
1733 if (s_bits == MO_64) {
1734 tcg_out_mov(s, TCG_TYPE_I64, arg++, lo);
1735 } else {
1736 tcg_out_rld(s, RLDICL, arg++, lo, 0, 64 - (8 << s_bits));
1737 }
1738 }
1739
1740 tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
1741 tcg_out32(s, MFSPR | RT(arg) | LR);
1742
1743 tcg_out_call(s, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
1744
1745 tcg_out_b(s, 0, lb->raddr);
1746 }
1747 #endif /* SOFTMMU */
1748
tcg_out_qemu_ld(TCGContext * s,const TCGArg * args,bool is_64)1749 static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64)
1750 {
1751 TCGReg datalo, datahi, addrlo, rbase;
1752 TCGReg addrhi __attribute__((unused));
1753 TCGMemOpIdx oi;
1754 TCGMemOp opc, s_bits;
1755 #ifdef CONFIG_SOFTMMU
1756 int mem_index;
1757 tcg_insn_unit *label_ptr;
1758 #endif
1759
1760 datalo = *args++;
1761 datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
1762 addrlo = *args++;
1763 addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
1764 oi = *args++;
1765 opc = get_memop(oi);
1766 s_bits = opc & MO_SIZE;
1767
1768 #ifdef CONFIG_SOFTMMU
1769 mem_index = get_mmuidx(oi);
1770 addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, true);
1771
1772 /* Load a pointer into the current opcode w/conditional branch-link. */
1773 label_ptr = s->code_ptr;
1774 tcg_out_bc_noaddr(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
1775
1776 rbase = TCG_REG_R3;
1777 #else /* !CONFIG_SOFTMMU */
1778 rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
1779 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
1780 tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
1781 addrlo = TCG_REG_TMP1;
1782 }
1783 #endif
1784
1785 if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
1786 if (opc & MO_BSWAP) {
1787 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1788 tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
1789 tcg_out32(s, LWBRX | TAB(datahi, rbase, TCG_REG_R0));
1790 } else if (rbase != 0) {
1791 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1792 tcg_out32(s, LWZX | TAB(datahi, rbase, addrlo));
1793 tcg_out32(s, LWZX | TAB(datalo, rbase, TCG_REG_R0));
1794 } else if (addrlo == datahi) {
1795 tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
1796 tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
1797 } else {
1798 tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
1799 tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
1800 }
1801 } else {
1802 uint32_t insn = qemu_ldx_opc[opc & (MO_BSWAP | MO_SSIZE)];
1803 if (!HAVE_ISA_2_06 && insn == LDBRX) {
1804 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1805 tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
1806 tcg_out32(s, LWBRX | TAB(TCG_REG_R0, rbase, TCG_REG_R0));
1807 tcg_out_rld(s, RLDIMI, datalo, TCG_REG_R0, 32, 0);
1808 } else if (insn) {
1809 tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
1810 } else {
1811 insn = qemu_ldx_opc[opc & (MO_SIZE | MO_BSWAP)];
1812 tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
1813 insn = qemu_exts_opc[s_bits];
1814 tcg_out32(s, insn | RA(datalo) | RS(datalo));
1815 }
1816 }
1817
1818 #ifdef CONFIG_SOFTMMU
1819 add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
1820 s->code_ptr, label_ptr);
1821 #endif
1822 }
1823
tcg_out_qemu_st(TCGContext * s,const TCGArg * args,bool is_64)1824 static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
1825 {
1826 TCGReg datalo, datahi, addrlo, rbase;
1827 TCGReg addrhi __attribute__((unused));
1828 TCGMemOpIdx oi;
1829 TCGMemOp opc, s_bits;
1830 #ifdef CONFIG_SOFTMMU
1831 int mem_index;
1832 tcg_insn_unit *label_ptr;
1833 #endif
1834
1835 datalo = *args++;
1836 datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
1837 addrlo = *args++;
1838 addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
1839 oi = *args++;
1840 opc = get_memop(oi);
1841 s_bits = opc & MO_SIZE;
1842
1843 #ifdef CONFIG_SOFTMMU
1844 mem_index = get_mmuidx(oi);
1845 addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, false);
1846
1847 /* Load a pointer into the current opcode w/conditional branch-link. */
1848 label_ptr = s->code_ptr;
1849 tcg_out_bc_noaddr(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
1850
1851 rbase = TCG_REG_R3;
1852 #else /* !CONFIG_SOFTMMU */
1853 rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
1854 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
1855 tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
1856 addrlo = TCG_REG_TMP1;
1857 }
1858 #endif
1859
1860 if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
1861 if (opc & MO_BSWAP) {
1862 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1863 tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
1864 tcg_out32(s, STWBRX | SAB(datahi, rbase, TCG_REG_R0));
1865 } else if (rbase != 0) {
1866 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1867 tcg_out32(s, STWX | SAB(datahi, rbase, addrlo));
1868 tcg_out32(s, STWX | SAB(datalo, rbase, TCG_REG_R0));
1869 } else {
1870 tcg_out32(s, STW | TAI(datahi, addrlo, 0));
1871 tcg_out32(s, STW | TAI(datalo, addrlo, 4));
1872 }
1873 } else {
1874 uint32_t insn = qemu_stx_opc[opc & (MO_BSWAP | MO_SIZE)];
1875 if (!HAVE_ISA_2_06 && insn == STDBRX) {
1876 tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
1877 tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, addrlo, 4));
1878 tcg_out_shri64(s, TCG_REG_R0, datalo, 32);
1879 tcg_out32(s, STWBRX | SAB(TCG_REG_R0, rbase, TCG_REG_TMP1));
1880 } else {
1881 tcg_out32(s, insn | SAB(datalo, rbase, addrlo));
1882 }
1883 }
1884
1885 #ifdef CONFIG_SOFTMMU
1886 add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
1887 s->code_ptr, label_ptr);
1888 #endif
1889 }
1890
tcg_out_nop_fill(tcg_insn_unit * p,int count)1891 static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
1892 {
1893 int i;
1894 for (i = 0; i < count; ++i) {
1895 p[i] = NOP;
1896 }
1897 }
1898
1899 /* Parameters for function call generation, used in tcg.c. */
1900 #define TCG_TARGET_STACK_ALIGN 16
1901 #define TCG_TARGET_EXTEND_ARGS 1
1902
1903 #ifdef _CALL_AIX
1904 # define LINK_AREA_SIZE (6 * SZR)
1905 # define LR_OFFSET (1 * SZR)
1906 # define TCG_TARGET_CALL_STACK_OFFSET (LINK_AREA_SIZE + 8 * SZR)
1907 #elif defined(TCG_TARGET_CALL_DARWIN)
1908 # define LINK_AREA_SIZE (6 * SZR)
1909 # define LR_OFFSET (2 * SZR)
1910 #elif TCG_TARGET_REG_BITS == 64
1911 # if defined(_CALL_ELF) && _CALL_ELF == 2
1912 # define LINK_AREA_SIZE (4 * SZR)
1913 # define LR_OFFSET (1 * SZR)
1914 # endif
1915 #else /* TCG_TARGET_REG_BITS == 32 */
1916 # if defined(_CALL_SYSV)
1917 # define LINK_AREA_SIZE (2 * SZR)
1918 # define LR_OFFSET (1 * SZR)
1919 # endif
1920 #endif
1921 #ifndef LR_OFFSET
1922 # error "Unhandled abi"
1923 #endif
1924 #ifndef TCG_TARGET_CALL_STACK_OFFSET
1925 # define TCG_TARGET_CALL_STACK_OFFSET LINK_AREA_SIZE
1926 #endif
1927
1928 #define CPU_TEMP_BUF_SIZE (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
1929 #define REG_SAVE_SIZE ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
1930
1931 #define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET \
1932 + TCG_STATIC_CALL_ARGS_SIZE \
1933 + CPU_TEMP_BUF_SIZE \
1934 + REG_SAVE_SIZE \
1935 + TCG_TARGET_STACK_ALIGN - 1) \
1936 & -TCG_TARGET_STACK_ALIGN)
1937
1938 #define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
1939
tcg_target_qemu_prologue(TCGContext * s)1940 static void tcg_target_qemu_prologue(TCGContext *s)
1941 {
1942 int i;
1943
1944 #ifdef _CALL_AIX
1945 void **desc = (void **)s->code_ptr;
1946 desc[0] = desc + 2; /* entry point */
1947 desc[1] = 0; /* environment pointer */
1948 s->code_ptr = (void *)(desc + 2); /* skip over descriptor */
1949 #endif
1950
1951 tcg_set_frame(s, TCG_REG_CALL_STACK, REG_SAVE_BOT - CPU_TEMP_BUF_SIZE,
1952 CPU_TEMP_BUF_SIZE);
1953
1954 /* Prologue */
1955 tcg_out32(s, MFSPR | RT(TCG_REG_R0) | LR);
1956 tcg_out32(s, (SZR == 8 ? STDU : STWU)
1957 | SAI(TCG_REG_R1, TCG_REG_R1, -FRAME_SIZE));
1958
1959 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
1960 tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
1961 TCG_REG_R1, REG_SAVE_BOT + i * SZR);
1962 }
1963 tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
1964
1965 #ifndef CONFIG_SOFTMMU
1966 if (guest_base) {
1967 tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true);
1968 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
1969 }
1970 #endif
1971
1972 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
1973 tcg_out32(s, MTSPR | RS(tcg_target_call_iarg_regs[1]) | CTR);
1974 if (USE_REG_TB) {
1975 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, tcg_target_call_iarg_regs[1]);
1976 }
1977 tcg_out32(s, BCCTR | BO_ALWAYS);
1978
1979 /* Epilogue */
1980 s->code_gen_epilogue = tb_ret_addr = s->code_ptr;
1981
1982 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
1983 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
1984 tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
1985 TCG_REG_R1, REG_SAVE_BOT + i * SZR);
1986 }
1987 tcg_out32(s, MTSPR | RS(TCG_REG_R0) | LR);
1988 tcg_out32(s, ADDI | TAI(TCG_REG_R1, TCG_REG_R1, FRAME_SIZE));
1989 tcg_out32(s, BCLR | BO_ALWAYS);
1990 }
1991
tcg_out_op(TCGContext * s,TCGOpcode opc,const TCGArg * args,const int * const_args)1992 static void tcg_out_op(TCGContext *s, TCGOpcode opc, const TCGArg *args,
1993 const int *const_args)
1994 {
1995 TCGArg a0, a1, a2;
1996 int c;
1997
1998 switch (opc) {
1999 case INDEX_op_exit_tb:
2000 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, args[0]);
2001 tcg_out_b(s, 0, tb_ret_addr);
2002 break;
2003 case INDEX_op_goto_tb:
2004 if (s->tb_jmp_insn_offset) {
2005 /* Direct jump. */
2006 if (TCG_TARGET_REG_BITS == 64) {
2007 /* Ensure the next insns are 8-byte aligned. */
2008 if ((uintptr_t)s->code_ptr & 7) {
2009 tcg_out32(s, NOP);
2010 }
2011 s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2012 tcg_out32(s, ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2013 tcg_out32(s, ADDI | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2014 } else {
2015 s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2016 tcg_out32(s, B);
2017 s->tb_jmp_reset_offset[args[0]] = tcg_current_code_size(s);
2018 break;
2019 }
2020 } else {
2021 /* Indirect jump. */
2022 tcg_debug_assert(s->tb_jmp_insn_offset == NULL);
2023 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, 0,
2024 (intptr_t)(s->tb_jmp_insn_offset + args[0]));
2025 }
2026 tcg_out32(s, MTSPR | RS(TCG_REG_TB) | CTR);
2027 tcg_out32(s, BCCTR | BO_ALWAYS);
2028 set_jmp_reset_offset(s, args[0]);
2029 if (USE_REG_TB) {
2030 /* For the unlinked case, need to reset TCG_REG_TB. */
2031 c = -tcg_current_code_size(s);
2032 assert(c == (int16_t)c);
2033 tcg_out32(s, ADDI | TAI(TCG_REG_TB, TCG_REG_TB, c));
2034 }
2035 break;
2036 case INDEX_op_goto_ptr:
2037 tcg_out32(s, MTSPR | RS(args[0]) | CTR);
2038 if (USE_REG_TB) {
2039 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, args[0]);
2040 }
2041 tcg_out32(s, ADDI | TAI(TCG_REG_R3, 0, 0));
2042 tcg_out32(s, BCCTR | BO_ALWAYS);
2043 break;
2044 case INDEX_op_br:
2045 {
2046 TCGLabel *l = arg_label(args[0]);
2047
2048 if (l->has_value) {
2049 tcg_out_b(s, 0, l->u.value_ptr);
2050 } else {
2051 tcg_out_reloc(s, s->code_ptr, R_PPC_REL24, l, 0);
2052 tcg_out_b_noaddr(s, B);
2053 }
2054 }
2055 break;
2056 case INDEX_op_ld8u_i32:
2057 case INDEX_op_ld8u_i64:
2058 tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2059 break;
2060 case INDEX_op_ld8s_i32:
2061 case INDEX_op_ld8s_i64:
2062 tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2063 tcg_out32(s, EXTSB | RS(args[0]) | RA(args[0]));
2064 break;
2065 case INDEX_op_ld16u_i32:
2066 case INDEX_op_ld16u_i64:
2067 tcg_out_mem_long(s, LHZ, LHZX, args[0], args[1], args[2]);
2068 break;
2069 case INDEX_op_ld16s_i32:
2070 case INDEX_op_ld16s_i64:
2071 tcg_out_mem_long(s, LHA, LHAX, args[0], args[1], args[2]);
2072 break;
2073 case INDEX_op_ld_i32:
2074 case INDEX_op_ld32u_i64:
2075 tcg_out_mem_long(s, LWZ, LWZX, args[0], args[1], args[2]);
2076 break;
2077 case INDEX_op_ld32s_i64:
2078 tcg_out_mem_long(s, LWA, LWAX, args[0], args[1], args[2]);
2079 break;
2080 case INDEX_op_ld_i64:
2081 tcg_out_mem_long(s, LD, LDX, args[0], args[1], args[2]);
2082 break;
2083 case INDEX_op_st8_i32:
2084 case INDEX_op_st8_i64:
2085 tcg_out_mem_long(s, STB, STBX, args[0], args[1], args[2]);
2086 break;
2087 case INDEX_op_st16_i32:
2088 case INDEX_op_st16_i64:
2089 tcg_out_mem_long(s, STH, STHX, args[0], args[1], args[2]);
2090 break;
2091 case INDEX_op_st_i32:
2092 case INDEX_op_st32_i64:
2093 tcg_out_mem_long(s, STW, STWX, args[0], args[1], args[2]);
2094 break;
2095 case INDEX_op_st_i64:
2096 tcg_out_mem_long(s, STD, STDX, args[0], args[1], args[2]);
2097 break;
2098
2099 case INDEX_op_add_i32:
2100 a0 = args[0], a1 = args[1], a2 = args[2];
2101 if (const_args[2]) {
2102 do_addi_32:
2103 tcg_out_mem_long(s, ADDI, ADD, a0, a1, (int32_t)a2);
2104 } else {
2105 tcg_out32(s, ADD | TAB(a0, a1, a2));
2106 }
2107 break;
2108 case INDEX_op_sub_i32:
2109 a0 = args[0], a1 = args[1], a2 = args[2];
2110 if (const_args[1]) {
2111 if (const_args[2]) {
2112 tcg_out_movi(s, TCG_TYPE_I32, a0, a1 - a2);
2113 } else {
2114 tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2115 }
2116 } else if (const_args[2]) {
2117 a2 = -a2;
2118 goto do_addi_32;
2119 } else {
2120 tcg_out32(s, SUBF | TAB(a0, a2, a1));
2121 }
2122 break;
2123
2124 case INDEX_op_and_i32:
2125 a0 = args[0], a1 = args[1], a2 = args[2];
2126 if (const_args[2]) {
2127 tcg_out_andi32(s, a0, a1, a2);
2128 } else {
2129 tcg_out32(s, AND | SAB(a1, a0, a2));
2130 }
2131 break;
2132 case INDEX_op_and_i64:
2133 a0 = args[0], a1 = args[1], a2 = args[2];
2134 if (const_args[2]) {
2135 tcg_out_andi64(s, a0, a1, a2);
2136 } else {
2137 tcg_out32(s, AND | SAB(a1, a0, a2));
2138 }
2139 break;
2140 case INDEX_op_or_i64:
2141 case INDEX_op_or_i32:
2142 a0 = args[0], a1 = args[1], a2 = args[2];
2143 if (const_args[2]) {
2144 tcg_out_ori32(s, a0, a1, a2);
2145 } else {
2146 tcg_out32(s, OR | SAB(a1, a0, a2));
2147 }
2148 break;
2149 case INDEX_op_xor_i64:
2150 case INDEX_op_xor_i32:
2151 a0 = args[0], a1 = args[1], a2 = args[2];
2152 if (const_args[2]) {
2153 tcg_out_xori32(s, a0, a1, a2);
2154 } else {
2155 tcg_out32(s, XOR | SAB(a1, a0, a2));
2156 }
2157 break;
2158 case INDEX_op_andc_i32:
2159 a0 = args[0], a1 = args[1], a2 = args[2];
2160 if (const_args[2]) {
2161 tcg_out_andi32(s, a0, a1, ~a2);
2162 } else {
2163 tcg_out32(s, ANDC | SAB(a1, a0, a2));
2164 }
2165 break;
2166 case INDEX_op_andc_i64:
2167 a0 = args[0], a1 = args[1], a2 = args[2];
2168 if (const_args[2]) {
2169 tcg_out_andi64(s, a0, a1, ~a2);
2170 } else {
2171 tcg_out32(s, ANDC | SAB(a1, a0, a2));
2172 }
2173 break;
2174 case INDEX_op_orc_i32:
2175 if (const_args[2]) {
2176 tcg_out_ori32(s, args[0], args[1], ~args[2]);
2177 break;
2178 }
2179 /* FALLTHRU */
2180 case INDEX_op_orc_i64:
2181 tcg_out32(s, ORC | SAB(args[1], args[0], args[2]));
2182 break;
2183 case INDEX_op_eqv_i32:
2184 if (const_args[2]) {
2185 tcg_out_xori32(s, args[0], args[1], ~args[2]);
2186 break;
2187 }
2188 /* FALLTHRU */
2189 case INDEX_op_eqv_i64:
2190 tcg_out32(s, EQV | SAB(args[1], args[0], args[2]));
2191 break;
2192 case INDEX_op_nand_i32:
2193 case INDEX_op_nand_i64:
2194 tcg_out32(s, NAND | SAB(args[1], args[0], args[2]));
2195 break;
2196 case INDEX_op_nor_i32:
2197 case INDEX_op_nor_i64:
2198 tcg_out32(s, NOR | SAB(args[1], args[0], args[2]));
2199 break;
2200
2201 case INDEX_op_clz_i32:
2202 tcg_out_cntxz(s, TCG_TYPE_I32, CNTLZW, args[0], args[1],
2203 args[2], const_args[2]);
2204 break;
2205 case INDEX_op_ctz_i32:
2206 tcg_out_cntxz(s, TCG_TYPE_I32, CNTTZW, args[0], args[1],
2207 args[2], const_args[2]);
2208 break;
2209 case INDEX_op_ctpop_i32:
2210 tcg_out32(s, CNTPOPW | SAB(args[1], args[0], 0));
2211 break;
2212
2213 case INDEX_op_clz_i64:
2214 tcg_out_cntxz(s, TCG_TYPE_I64, CNTLZD, args[0], args[1],
2215 args[2], const_args[2]);
2216 break;
2217 case INDEX_op_ctz_i64:
2218 tcg_out_cntxz(s, TCG_TYPE_I64, CNTTZD, args[0], args[1],
2219 args[2], const_args[2]);
2220 break;
2221 case INDEX_op_ctpop_i64:
2222 tcg_out32(s, CNTPOPD | SAB(args[1], args[0], 0));
2223 break;
2224
2225 case INDEX_op_mul_i32:
2226 a0 = args[0], a1 = args[1], a2 = args[2];
2227 if (const_args[2]) {
2228 tcg_out32(s, MULLI | TAI(a0, a1, a2));
2229 } else {
2230 tcg_out32(s, MULLW | TAB(a0, a1, a2));
2231 }
2232 break;
2233
2234 case INDEX_op_div_i32:
2235 tcg_out32(s, DIVW | TAB(args[0], args[1], args[2]));
2236 break;
2237
2238 case INDEX_op_divu_i32:
2239 tcg_out32(s, DIVWU | TAB(args[0], args[1], args[2]));
2240 break;
2241
2242 case INDEX_op_shl_i32:
2243 if (const_args[2]) {
2244 tcg_out_shli32(s, args[0], args[1], args[2]);
2245 } else {
2246 tcg_out32(s, SLW | SAB(args[1], args[0], args[2]));
2247 }
2248 break;
2249 case INDEX_op_shr_i32:
2250 if (const_args[2]) {
2251 tcg_out_shri32(s, args[0], args[1], args[2]);
2252 } else {
2253 tcg_out32(s, SRW | SAB(args[1], args[0], args[2]));
2254 }
2255 break;
2256 case INDEX_op_sar_i32:
2257 if (const_args[2]) {
2258 tcg_out32(s, SRAWI | RS(args[1]) | RA(args[0]) | SH(args[2]));
2259 } else {
2260 tcg_out32(s, SRAW | SAB(args[1], args[0], args[2]));
2261 }
2262 break;
2263 case INDEX_op_rotl_i32:
2264 if (const_args[2]) {
2265 tcg_out_rlw(s, RLWINM, args[0], args[1], args[2], 0, 31);
2266 } else {
2267 tcg_out32(s, RLWNM | SAB(args[1], args[0], args[2])
2268 | MB(0) | ME(31));
2269 }
2270 break;
2271 case INDEX_op_rotr_i32:
2272 if (const_args[2]) {
2273 tcg_out_rlw(s, RLWINM, args[0], args[1], 32 - args[2], 0, 31);
2274 } else {
2275 tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 32));
2276 tcg_out32(s, RLWNM | SAB(args[1], args[0], TCG_REG_R0)
2277 | MB(0) | ME(31));
2278 }
2279 break;
2280
2281 case INDEX_op_brcond_i32:
2282 tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2283 arg_label(args[3]), TCG_TYPE_I32);
2284 break;
2285 case INDEX_op_brcond_i64:
2286 tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2287 arg_label(args[3]), TCG_TYPE_I64);
2288 break;
2289 case INDEX_op_brcond2_i32:
2290 tcg_out_brcond2(s, args, const_args);
2291 break;
2292
2293 case INDEX_op_neg_i32:
2294 case INDEX_op_neg_i64:
2295 tcg_out32(s, NEG | RT(args[0]) | RA(args[1]));
2296 break;
2297
2298 case INDEX_op_not_i32:
2299 case INDEX_op_not_i64:
2300 tcg_out32(s, NOR | SAB(args[1], args[0], args[1]));
2301 break;
2302
2303 case INDEX_op_add_i64:
2304 a0 = args[0], a1 = args[1], a2 = args[2];
2305 if (const_args[2]) {
2306 do_addi_64:
2307 tcg_out_mem_long(s, ADDI, ADD, a0, a1, a2);
2308 } else {
2309 tcg_out32(s, ADD | TAB(a0, a1, a2));
2310 }
2311 break;
2312 case INDEX_op_sub_i64:
2313 a0 = args[0], a1 = args[1], a2 = args[2];
2314 if (const_args[1]) {
2315 if (const_args[2]) {
2316 tcg_out_movi(s, TCG_TYPE_I64, a0, a1 - a2);
2317 } else {
2318 tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2319 }
2320 } else if (const_args[2]) {
2321 a2 = -a2;
2322 goto do_addi_64;
2323 } else {
2324 tcg_out32(s, SUBF | TAB(a0, a2, a1));
2325 }
2326 break;
2327
2328 case INDEX_op_shl_i64:
2329 if (const_args[2]) {
2330 tcg_out_shli64(s, args[0], args[1], args[2]);
2331 } else {
2332 tcg_out32(s, SLD | SAB(args[1], args[0], args[2]));
2333 }
2334 break;
2335 case INDEX_op_shr_i64:
2336 if (const_args[2]) {
2337 tcg_out_shri64(s, args[0], args[1], args[2]);
2338 } else {
2339 tcg_out32(s, SRD | SAB(args[1], args[0], args[2]));
2340 }
2341 break;
2342 case INDEX_op_sar_i64:
2343 if (const_args[2]) {
2344 int sh = SH(args[2] & 0x1f) | (((args[2] >> 5) & 1) << 1);
2345 tcg_out32(s, SRADI | RA(args[0]) | RS(args[1]) | sh);
2346 } else {
2347 tcg_out32(s, SRAD | SAB(args[1], args[0], args[2]));
2348 }
2349 break;
2350 case INDEX_op_rotl_i64:
2351 if (const_args[2]) {
2352 tcg_out_rld(s, RLDICL, args[0], args[1], args[2], 0);
2353 } else {
2354 tcg_out32(s, RLDCL | SAB(args[1], args[0], args[2]) | MB64(0));
2355 }
2356 break;
2357 case INDEX_op_rotr_i64:
2358 if (const_args[2]) {
2359 tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 0);
2360 } else {
2361 tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 64));
2362 tcg_out32(s, RLDCL | SAB(args[1], args[0], TCG_REG_R0) | MB64(0));
2363 }
2364 break;
2365
2366 case INDEX_op_mul_i64:
2367 a0 = args[0], a1 = args[1], a2 = args[2];
2368 if (const_args[2]) {
2369 tcg_out32(s, MULLI | TAI(a0, a1, a2));
2370 } else {
2371 tcg_out32(s, MULLD | TAB(a0, a1, a2));
2372 }
2373 break;
2374 case INDEX_op_div_i64:
2375 tcg_out32(s, DIVD | TAB(args[0], args[1], args[2]));
2376 break;
2377 case INDEX_op_divu_i64:
2378 tcg_out32(s, DIVDU | TAB(args[0], args[1], args[2]));
2379 break;
2380
2381 case INDEX_op_qemu_ld_i32:
2382 tcg_out_qemu_ld(s, args, false);
2383 break;
2384 case INDEX_op_qemu_ld_i64:
2385 tcg_out_qemu_ld(s, args, true);
2386 break;
2387 case INDEX_op_qemu_st_i32:
2388 tcg_out_qemu_st(s, args, false);
2389 break;
2390 case INDEX_op_qemu_st_i64:
2391 tcg_out_qemu_st(s, args, true);
2392 break;
2393
2394 case INDEX_op_ext8s_i32:
2395 case INDEX_op_ext8s_i64:
2396 c = EXTSB;
2397 goto gen_ext;
2398 case INDEX_op_ext16s_i32:
2399 case INDEX_op_ext16s_i64:
2400 c = EXTSH;
2401 goto gen_ext;
2402 case INDEX_op_ext_i32_i64:
2403 case INDEX_op_ext32s_i64:
2404 c = EXTSW;
2405 goto gen_ext;
2406 gen_ext:
2407 tcg_out32(s, c | RS(args[1]) | RA(args[0]));
2408 break;
2409 case INDEX_op_extu_i32_i64:
2410 tcg_out_ext32u(s, args[0], args[1]);
2411 break;
2412
2413 case INDEX_op_setcond_i32:
2414 tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2],
2415 const_args[2]);
2416 break;
2417 case INDEX_op_setcond_i64:
2418 tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2],
2419 const_args[2]);
2420 break;
2421 case INDEX_op_setcond2_i32:
2422 tcg_out_setcond2(s, args, const_args);
2423 break;
2424
2425 case INDEX_op_bswap16_i32:
2426 case INDEX_op_bswap16_i64:
2427 a0 = args[0], a1 = args[1];
2428 /* a1 = abcd */
2429 if (a0 != a1) {
2430 /* a0 = (a1 r<< 24) & 0xff # 000c */
2431 tcg_out_rlw(s, RLWINM, a0, a1, 24, 24, 31);
2432 /* a0 = (a0 & ~0xff00) | (a1 r<< 8) & 0xff00 # 00dc */
2433 tcg_out_rlw(s, RLWIMI, a0, a1, 8, 16, 23);
2434 } else {
2435 /* r0 = (a1 r<< 8) & 0xff00 # 00d0 */
2436 tcg_out_rlw(s, RLWINM, TCG_REG_R0, a1, 8, 16, 23);
2437 /* a0 = (a1 r<< 24) & 0xff # 000c */
2438 tcg_out_rlw(s, RLWINM, a0, a1, 24, 24, 31);
2439 /* a0 = a0 | r0 # 00dc */
2440 tcg_out32(s, OR | SAB(TCG_REG_R0, a0, a0));
2441 }
2442 break;
2443
2444 case INDEX_op_bswap32_i32:
2445 case INDEX_op_bswap32_i64:
2446 /* Stolen from gcc's builtin_bswap32 */
2447 a1 = args[1];
2448 a0 = args[0] == a1 ? TCG_REG_R0 : args[0];
2449
2450 /* a1 = args[1] # abcd */
2451 /* a0 = rotate_left (a1, 8) # bcda */
2452 tcg_out_rlw(s, RLWINM, a0, a1, 8, 0, 31);
2453 /* a0 = (a0 & ~0xff000000) | ((a1 r<< 24) & 0xff000000) # dcda */
2454 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 0, 7);
2455 /* a0 = (a0 & ~0x0000ff00) | ((a1 r<< 24) & 0x0000ff00) # dcba */
2456 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 16, 23);
2457
2458 if (a0 == TCG_REG_R0) {
2459 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2460 }
2461 break;
2462
2463 case INDEX_op_bswap64_i64:
2464 a0 = args[0], a1 = args[1], a2 = TCG_REG_R0;
2465 if (a0 == a1) {
2466 a0 = TCG_REG_R0;
2467 a2 = a1;
2468 }
2469
2470 /* a1 = # abcd efgh */
2471 /* a0 = rl32(a1, 8) # 0000 fghe */
2472 tcg_out_rlw(s, RLWINM, a0, a1, 8, 0, 31);
2473 /* a0 = dep(a0, rl32(a1, 24), 0xff000000) # 0000 hghe */
2474 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 0, 7);
2475 /* a0 = dep(a0, rl32(a1, 24), 0x0000ff00) # 0000 hgfe */
2476 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 16, 23);
2477
2478 /* a0 = rl64(a0, 32) # hgfe 0000 */
2479 /* a2 = rl64(a1, 32) # efgh abcd */
2480 tcg_out_rld(s, RLDICL, a0, a0, 32, 0);
2481 tcg_out_rld(s, RLDICL, a2, a1, 32, 0);
2482
2483 /* a0 = dep(a0, rl32(a2, 8), 0xffffffff) # hgfe bcda */
2484 tcg_out_rlw(s, RLWIMI, a0, a2, 8, 0, 31);
2485 /* a0 = dep(a0, rl32(a2, 24), 0xff000000) # hgfe dcda */
2486 tcg_out_rlw(s, RLWIMI, a0, a2, 24, 0, 7);
2487 /* a0 = dep(a0, rl32(a2, 24), 0x0000ff00) # hgfe dcba */
2488 tcg_out_rlw(s, RLWIMI, a0, a2, 24, 16, 23);
2489
2490 if (a0 == 0) {
2491 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2492 }
2493 break;
2494
2495 case INDEX_op_deposit_i32:
2496 if (const_args[2]) {
2497 uint32_t mask = ((2u << (args[4] - 1)) - 1) << args[3];
2498 tcg_out_andi32(s, args[0], args[0], ~mask);
2499 } else {
2500 tcg_out_rlw(s, RLWIMI, args[0], args[2], args[3],
2501 32 - args[3] - args[4], 31 - args[3]);
2502 }
2503 break;
2504 case INDEX_op_deposit_i64:
2505 if (const_args[2]) {
2506 uint64_t mask = ((2ull << (args[4] - 1)) - 1) << args[3];
2507 tcg_out_andi64(s, args[0], args[0], ~mask);
2508 } else {
2509 tcg_out_rld(s, RLDIMI, args[0], args[2], args[3],
2510 64 - args[3] - args[4]);
2511 }
2512 break;
2513
2514 case INDEX_op_extract_i32:
2515 tcg_out_rlw(s, RLWINM, args[0], args[1],
2516 32 - args[2], 32 - args[3], 31);
2517 break;
2518 case INDEX_op_extract_i64:
2519 tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 64 - args[3]);
2520 break;
2521
2522 case INDEX_op_movcond_i32:
2523 tcg_out_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], args[2],
2524 args[3], args[4], const_args[2]);
2525 break;
2526 case INDEX_op_movcond_i64:
2527 tcg_out_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], args[2],
2528 args[3], args[4], const_args[2]);
2529 break;
2530
2531 #if TCG_TARGET_REG_BITS == 64
2532 case INDEX_op_add2_i64:
2533 #else
2534 case INDEX_op_add2_i32:
2535 #endif
2536 /* Note that the CA bit is defined based on the word size of the
2537 environment. So in 64-bit mode it's always carry-out of bit 63.
2538 The fallback code using deposit works just as well for 32-bit. */
2539 a0 = args[0], a1 = args[1];
2540 if (a0 == args[3] || (!const_args[5] && a0 == args[5])) {
2541 a0 = TCG_REG_R0;
2542 }
2543 if (const_args[4]) {
2544 tcg_out32(s, ADDIC | TAI(a0, args[2], args[4]));
2545 } else {
2546 tcg_out32(s, ADDC | TAB(a0, args[2], args[4]));
2547 }
2548 if (const_args[5]) {
2549 tcg_out32(s, (args[5] ? ADDME : ADDZE) | RT(a1) | RA(args[3]));
2550 } else {
2551 tcg_out32(s, ADDE | TAB(a1, args[3], args[5]));
2552 }
2553 if (a0 != args[0]) {
2554 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2555 }
2556 break;
2557
2558 #if TCG_TARGET_REG_BITS == 64
2559 case INDEX_op_sub2_i64:
2560 #else
2561 case INDEX_op_sub2_i32:
2562 #endif
2563 a0 = args[0], a1 = args[1];
2564 if (a0 == args[5] || (!const_args[3] && a0 == args[3])) {
2565 a0 = TCG_REG_R0;
2566 }
2567 if (const_args[2]) {
2568 tcg_out32(s, SUBFIC | TAI(a0, args[4], args[2]));
2569 } else {
2570 tcg_out32(s, SUBFC | TAB(a0, args[4], args[2]));
2571 }
2572 if (const_args[3]) {
2573 tcg_out32(s, (args[3] ? SUBFME : SUBFZE) | RT(a1) | RA(args[5]));
2574 } else {
2575 tcg_out32(s, SUBFE | TAB(a1, args[5], args[3]));
2576 }
2577 if (a0 != args[0]) {
2578 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2579 }
2580 break;
2581
2582 case INDEX_op_muluh_i32:
2583 tcg_out32(s, MULHWU | TAB(args[0], args[1], args[2]));
2584 break;
2585 case INDEX_op_mulsh_i32:
2586 tcg_out32(s, MULHW | TAB(args[0], args[1], args[2]));
2587 break;
2588 case INDEX_op_muluh_i64:
2589 tcg_out32(s, MULHDU | TAB(args[0], args[1], args[2]));
2590 break;
2591 case INDEX_op_mulsh_i64:
2592 tcg_out32(s, MULHD | TAB(args[0], args[1], args[2]));
2593 break;
2594
2595 case INDEX_op_mb:
2596 tcg_out_mb(s, args[0]);
2597 break;
2598
2599 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */
2600 case INDEX_op_mov_i64:
2601 case INDEX_op_movi_i32: /* Always emitted via tcg_out_movi. */
2602 case INDEX_op_movi_i64:
2603 case INDEX_op_call: /* Always emitted via tcg_out_call. */
2604 default:
2605 tcg_abort();
2606 }
2607 }
2608
tcg_target_op_def(TCGOpcode op)2609 static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op)
2610 {
2611 static const TCGTargetOpDef r = { .args_ct_str = { "r" } };
2612 static const TCGTargetOpDef r_r = { .args_ct_str = { "r", "r" } };
2613 static const TCGTargetOpDef r_L = { .args_ct_str = { "r", "L" } };
2614 static const TCGTargetOpDef S_S = { .args_ct_str = { "S", "S" } };
2615 static const TCGTargetOpDef r_ri = { .args_ct_str = { "r", "ri" } };
2616 static const TCGTargetOpDef r_r_r = { .args_ct_str = { "r", "r", "r" } };
2617 static const TCGTargetOpDef r_L_L = { .args_ct_str = { "r", "L", "L" } };
2618 static const TCGTargetOpDef L_L_L = { .args_ct_str = { "L", "L", "L" } };
2619 static const TCGTargetOpDef S_S_S = { .args_ct_str = { "S", "S", "S" } };
2620 static const TCGTargetOpDef r_r_ri = { .args_ct_str = { "r", "r", "ri" } };
2621 static const TCGTargetOpDef r_r_rI = { .args_ct_str = { "r", "r", "rI" } };
2622 static const TCGTargetOpDef r_r_rT = { .args_ct_str = { "r", "r", "rT" } };
2623 static const TCGTargetOpDef r_r_rU = { .args_ct_str = { "r", "r", "rU" } };
2624 static const TCGTargetOpDef r_rI_ri
2625 = { .args_ct_str = { "r", "rI", "ri" } };
2626 static const TCGTargetOpDef r_rI_rT
2627 = { .args_ct_str = { "r", "rI", "rT" } };
2628 static const TCGTargetOpDef r_r_rZW
2629 = { .args_ct_str = { "r", "r", "rZW" } };
2630 static const TCGTargetOpDef L_L_L_L
2631 = { .args_ct_str = { "L", "L", "L", "L" } };
2632 static const TCGTargetOpDef S_S_S_S
2633 = { .args_ct_str = { "S", "S", "S", "S" } };
2634 static const TCGTargetOpDef movc
2635 = { .args_ct_str = { "r", "r", "ri", "rZ", "rZ" } };
2636 static const TCGTargetOpDef dep
2637 = { .args_ct_str = { "r", "0", "rZ" } };
2638 static const TCGTargetOpDef br2
2639 = { .args_ct_str = { "r", "r", "ri", "ri" } };
2640 static const TCGTargetOpDef setc2
2641 = { .args_ct_str = { "r", "r", "r", "ri", "ri" } };
2642 static const TCGTargetOpDef add2
2643 = { .args_ct_str = { "r", "r", "r", "r", "rI", "rZM" } };
2644 static const TCGTargetOpDef sub2
2645 = { .args_ct_str = { "r", "r", "rI", "rZM", "r", "r" } };
2646
2647 switch (op) {
2648 case INDEX_op_goto_ptr:
2649 return &r;
2650
2651 case INDEX_op_ld8u_i32:
2652 case INDEX_op_ld8s_i32:
2653 case INDEX_op_ld16u_i32:
2654 case INDEX_op_ld16s_i32:
2655 case INDEX_op_ld_i32:
2656 case INDEX_op_st8_i32:
2657 case INDEX_op_st16_i32:
2658 case INDEX_op_st_i32:
2659 case INDEX_op_ctpop_i32:
2660 case INDEX_op_neg_i32:
2661 case INDEX_op_not_i32:
2662 case INDEX_op_ext8s_i32:
2663 case INDEX_op_ext16s_i32:
2664 case INDEX_op_bswap16_i32:
2665 case INDEX_op_bswap32_i32:
2666 case INDEX_op_extract_i32:
2667 case INDEX_op_ld8u_i64:
2668 case INDEX_op_ld8s_i64:
2669 case INDEX_op_ld16u_i64:
2670 case INDEX_op_ld16s_i64:
2671 case INDEX_op_ld32u_i64:
2672 case INDEX_op_ld32s_i64:
2673 case INDEX_op_ld_i64:
2674 case INDEX_op_st8_i64:
2675 case INDEX_op_st16_i64:
2676 case INDEX_op_st32_i64:
2677 case INDEX_op_st_i64:
2678 case INDEX_op_ctpop_i64:
2679 case INDEX_op_neg_i64:
2680 case INDEX_op_not_i64:
2681 case INDEX_op_ext8s_i64:
2682 case INDEX_op_ext16s_i64:
2683 case INDEX_op_ext32s_i64:
2684 case INDEX_op_ext_i32_i64:
2685 case INDEX_op_extu_i32_i64:
2686 case INDEX_op_bswap16_i64:
2687 case INDEX_op_bswap32_i64:
2688 case INDEX_op_bswap64_i64:
2689 case INDEX_op_extract_i64:
2690 return &r_r;
2691
2692 case INDEX_op_add_i32:
2693 case INDEX_op_and_i32:
2694 case INDEX_op_or_i32:
2695 case INDEX_op_xor_i32:
2696 case INDEX_op_andc_i32:
2697 case INDEX_op_orc_i32:
2698 case INDEX_op_eqv_i32:
2699 case INDEX_op_shl_i32:
2700 case INDEX_op_shr_i32:
2701 case INDEX_op_sar_i32:
2702 case INDEX_op_rotl_i32:
2703 case INDEX_op_rotr_i32:
2704 case INDEX_op_setcond_i32:
2705 case INDEX_op_and_i64:
2706 case INDEX_op_andc_i64:
2707 case INDEX_op_shl_i64:
2708 case INDEX_op_shr_i64:
2709 case INDEX_op_sar_i64:
2710 case INDEX_op_rotl_i64:
2711 case INDEX_op_rotr_i64:
2712 case INDEX_op_setcond_i64:
2713 return &r_r_ri;
2714 case INDEX_op_mul_i32:
2715 case INDEX_op_mul_i64:
2716 return &r_r_rI;
2717 case INDEX_op_div_i32:
2718 case INDEX_op_divu_i32:
2719 case INDEX_op_nand_i32:
2720 case INDEX_op_nor_i32:
2721 case INDEX_op_muluh_i32:
2722 case INDEX_op_mulsh_i32:
2723 case INDEX_op_orc_i64:
2724 case INDEX_op_eqv_i64:
2725 case INDEX_op_nand_i64:
2726 case INDEX_op_nor_i64:
2727 case INDEX_op_div_i64:
2728 case INDEX_op_divu_i64:
2729 case INDEX_op_mulsh_i64:
2730 case INDEX_op_muluh_i64:
2731 return &r_r_r;
2732 case INDEX_op_sub_i32:
2733 return &r_rI_ri;
2734 case INDEX_op_add_i64:
2735 return &r_r_rT;
2736 case INDEX_op_or_i64:
2737 case INDEX_op_xor_i64:
2738 return &r_r_rU;
2739 case INDEX_op_sub_i64:
2740 return &r_rI_rT;
2741 case INDEX_op_clz_i32:
2742 case INDEX_op_ctz_i32:
2743 case INDEX_op_clz_i64:
2744 case INDEX_op_ctz_i64:
2745 return &r_r_rZW;
2746
2747 case INDEX_op_brcond_i32:
2748 case INDEX_op_brcond_i64:
2749 return &r_ri;
2750
2751 case INDEX_op_movcond_i32:
2752 case INDEX_op_movcond_i64:
2753 return &movc;
2754 case INDEX_op_deposit_i32:
2755 case INDEX_op_deposit_i64:
2756 return &dep;
2757 case INDEX_op_brcond2_i32:
2758 return &br2;
2759 case INDEX_op_setcond2_i32:
2760 return &setc2;
2761 case INDEX_op_add2_i64:
2762 case INDEX_op_add2_i32:
2763 return &add2;
2764 case INDEX_op_sub2_i64:
2765 case INDEX_op_sub2_i32:
2766 return &sub2;
2767
2768 case INDEX_op_qemu_ld_i32:
2769 return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
2770 ? &r_L : &r_L_L);
2771 case INDEX_op_qemu_st_i32:
2772 return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
2773 ? &S_S : &S_S_S);
2774 case INDEX_op_qemu_ld_i64:
2775 return (TCG_TARGET_REG_BITS == 64 ? &r_L
2776 : TARGET_LONG_BITS == 32 ? &L_L_L : &L_L_L_L);
2777 case INDEX_op_qemu_st_i64:
2778 return (TCG_TARGET_REG_BITS == 64 ? &S_S
2779 : TARGET_LONG_BITS == 32 ? &S_S_S : &S_S_S_S);
2780
2781 default:
2782 return NULL;
2783 }
2784 }
2785
tcg_target_init(TCGContext * s)2786 static void tcg_target_init(TCGContext *s)
2787 {
2788 unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2789 unsigned long hwcap2 = qemu_getauxval(AT_HWCAP2);
2790
2791 if (hwcap & PPC_FEATURE_ARCH_2_06) {
2792 have_isa_2_06 = true;
2793 }
2794 #ifdef PPC_FEATURE2_ARCH_3_00
2795 if (hwcap2 & PPC_FEATURE2_ARCH_3_00) {
2796 have_isa_3_00 = true;
2797 }
2798 #endif
2799
2800 tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
2801 tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
2802
2803 tcg_target_call_clobber_regs = 0;
2804 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
2805 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
2806 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
2807 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
2808 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
2809 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
2810 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R7);
2811 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R8);
2812 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R9);
2813 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R10);
2814 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R11);
2815 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
2816
2817 s->reserved_regs = 0;
2818 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); /* tcg temp */
2819 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R1); /* stack pointer */
2820 #if defined(_CALL_SYSV)
2821 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R2); /* toc pointer */
2822 #endif
2823 #if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
2824 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R13); /* thread pointer */
2825 #endif
2826 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1); /* mem temp */
2827 if (USE_REG_TB) {
2828 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB); /* tb->tc_ptr */
2829 }
2830 }
2831
2832 #ifdef __ELF__
2833 typedef struct {
2834 DebugFrameCIE cie;
2835 DebugFrameFDEHeader fde;
2836 uint8_t fde_def_cfa[4];
2837 uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2 + 3];
2838 } DebugFrame;
2839
2840 /* We're expecting a 2 byte uleb128 encoded value. */
2841 QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
2842
2843 #if TCG_TARGET_REG_BITS == 64
2844 # define ELF_HOST_MACHINE EM_PPC64
2845 #else
2846 # define ELF_HOST_MACHINE EM_PPC
2847 #endif
2848
2849 static DebugFrame debug_frame = {
2850 .cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
2851 .cie.id = -1,
2852 .cie.version = 1,
2853 .cie.code_align = 1,
2854 .cie.data_align = (-SZR & 0x7f), /* sleb128 -SZR */
2855 .cie.return_column = 65,
2856
2857 /* Total FDE size does not include the "len" member. */
2858 .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset),
2859
2860 .fde_def_cfa = {
2861 12, TCG_REG_R1, /* DW_CFA_def_cfa r1, ... */
2862 (FRAME_SIZE & 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */
2863 (FRAME_SIZE >> 7)
2864 },
2865 .fde_reg_ofs = {
2866 /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
2867 0x11, 65, (LR_OFFSET / -SZR) & 0x7f,
2868 }
2869 };
2870
tcg_register_jit(void * buf,size_t buf_size)2871 void tcg_register_jit(void *buf, size_t buf_size)
2872 {
2873 uint8_t *p = &debug_frame.fde_reg_ofs[3];
2874 int i;
2875
2876 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i, p += 2) {
2877 p[0] = 0x80 + tcg_target_callee_save_regs[i];
2878 p[1] = (FRAME_SIZE - (REG_SAVE_BOT + i * SZR)) / SZR;
2879 }
2880
2881 debug_frame.fde.func_start = (uintptr_t)buf;
2882 debug_frame.fde.func_len = buf_size;
2883
2884 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
2885 }
2886 #endif /* __ELF__ */
2887
flush_icache_range(uintptr_t start,uintptr_t stop)2888 void flush_icache_range(uintptr_t start, uintptr_t stop)
2889 {
2890 uintptr_t p, start1, stop1;
2891 size_t dsize = qemu_dcache_linesize;
2892 size_t isize = qemu_icache_linesize;
2893
2894 start1 = start & ~(dsize - 1);
2895 stop1 = (stop + dsize - 1) & ~(dsize - 1);
2896 for (p = start1; p < stop1; p += dsize) {
2897 asm volatile ("dcbst 0,%0" : : "r"(p) : "memory");
2898 }
2899 asm volatile ("sync" : : : "memory");
2900
2901 start &= start & ~(isize - 1);
2902 stop1 = (stop + isize - 1) & ~(isize - 1);
2903 for (p = start1; p < stop1; p += isize) {
2904 asm volatile ("icbi 0,%0" : : "r"(p) : "memory");
2905 }
2906 asm volatile ("sync" : : : "memory");
2907 asm volatile ("isync" : : : "memory");
2908 }
2909