2 * Tiny Code Generator for QEMU
4 * Copyright (c) 2008 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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
26 #include "tcg-pool.inc.c"
28 #if defined _CALL_DARWIN || defined __APPLE__
29 #define TCG_TARGET_CALL_DARWIN
32 # define TCG_TARGET_CALL_ALIGN_ARGS 1
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. */
40 # define TCG_REG_TMP1 TCG_REG_R2
42 # define TCG_REG_TMP1 TCG_REG_R12
45 #define TCG_REG_TB TCG_REG_R31
46 #define USE_REG_TB (TCG_TARGET_REG_BITS == 64)
48 /* Shorthand for size of a pointer. Avoid promotion to unsigned. */
49 #define SZP ((int)sizeof(void *))
51 /* Shorthand for size of a register. */
52 #define SZR (TCG_TARGET_REG_BITS / 8)
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
62 static tcg_insn_unit
*tb_ret_addr
;
67 #define HAVE_ISA_2_06 have_isa_2_06
68 #define HAVE_ISEL have_isa_2_06
70 #ifndef CONFIG_SOFTMMU
71 #define TCG_GUEST_BASE_REG 30
74 #ifdef CONFIG_DEBUG_TCG
75 static const char * const tcg_target_reg_names
[TCG_TARGET_NB_REGS
] = {
111 static const int tcg_target_reg_alloc_order
[] = {
112 TCG_REG_R14
, /* call saved registers */
130 TCG_REG_R12
, /* call clobbered, non-arguments */
134 TCG_REG_R10
, /* call clobbered, arguments */
144 static const int tcg_target_call_iarg_regs
[] = {
155 static const int tcg_target_call_oarg_regs
[] = {
160 static const int tcg_target_callee_save_regs
[] = {
161 #ifdef TCG_TARGET_CALL_DARWIN
177 TCG_REG_R27
, /* currently used for the global env */
184 static inline bool in_range_b(tcg_target_long target
)
186 return target
== sextract64(target
, 0, 26);
189 static uint32_t reloc_pc24_val(tcg_insn_unit
*pc
, tcg_insn_unit
*target
)
191 ptrdiff_t disp
= tcg_ptr_byte_diff(target
, pc
);
192 tcg_debug_assert(in_range_b(disp
));
193 return disp
& 0x3fffffc;
196 static void reloc_pc24(tcg_insn_unit
*pc
, tcg_insn_unit
*target
)
198 *pc
= (*pc
& ~0x3fffffc) | reloc_pc24_val(pc
, target
);
201 static uint16_t reloc_pc14_val(tcg_insn_unit
*pc
, tcg_insn_unit
*target
)
203 ptrdiff_t disp
= tcg_ptr_byte_diff(target
, pc
);
204 tcg_debug_assert(disp
== (int16_t) disp
);
205 return disp
& 0xfffc;
208 static void reloc_pc14(tcg_insn_unit
*pc
, tcg_insn_unit
*target
)
210 *pc
= (*pc
& ~0xfffc) | reloc_pc14_val(pc
, target
);
213 static inline void tcg_out_b_noaddr(TCGContext
*s
, int insn
)
215 unsigned retrans
= *s
->code_ptr
& 0x3fffffc;
216 tcg_out32(s
, insn
| retrans
);
219 static inline void tcg_out_bc_noaddr(TCGContext
*s
, int insn
)
221 unsigned retrans
= *s
->code_ptr
& 0xfffc;
222 tcg_out32(s
, insn
| retrans
);
225 static void patch_reloc(tcg_insn_unit
*code_ptr
, int type
,
226 intptr_t value
, intptr_t addend
)
228 tcg_insn_unit
*target
;
232 target
= (tcg_insn_unit
*)value
;
236 reloc_pc14(code_ptr
, target
);
239 reloc_pc24(code_ptr
, target
);
242 assert(value
== (int16_t)value
);
244 old
= deposit32(old
, 0, 16, value
);
252 /* parse target specific constraints */
253 static const char *target_parse_constraint(TCGArgConstraint
*ct
,
254 const char *ct_str
, TCGType type
)
257 case 'A': case 'B': case 'C': case 'D':
258 ct
->ct
|= TCG_CT_REG
;
259 tcg_regset_set_reg(ct
->u
.regs
, 3 + ct_str
[0] - 'A');
262 ct
->ct
|= TCG_CT_REG
;
263 ct
->u
.regs
= 0xffffffff;
265 case 'L': /* qemu_ld constraint */
266 ct
->ct
|= TCG_CT_REG
;
267 ct
->u
.regs
= 0xffffffff;
268 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R3
);
269 #ifdef CONFIG_SOFTMMU
270 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R4
);
271 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R5
);
274 case 'S': /* qemu_st constraint */
275 ct
->ct
|= TCG_CT_REG
;
276 ct
->u
.regs
= 0xffffffff;
277 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R3
);
278 #ifdef CONFIG_SOFTMMU
279 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R4
);
280 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R5
);
281 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R6
);
285 ct
->ct
|= TCG_CT_CONST_S16
;
288 ct
->ct
|= TCG_CT_CONST_U16
;
291 ct
->ct
|= TCG_CT_CONST_MONE
;
294 ct
->ct
|= TCG_CT_CONST_S32
;
297 ct
->ct
|= TCG_CT_CONST_U32
;
300 ct
->ct
|= TCG_CT_CONST_WSZ
;
303 ct
->ct
|= TCG_CT_CONST_ZERO
;
311 /* test if a constant matches the constraint */
312 static int tcg_target_const_match(tcg_target_long val
, TCGType type
,
313 const TCGArgConstraint
*arg_ct
)
316 if (ct
& TCG_CT_CONST
) {
320 /* The only 32-bit constraint we use aside from
321 TCG_CT_CONST is TCG_CT_CONST_S16. */
322 if (type
== TCG_TYPE_I32
) {
326 if ((ct
& TCG_CT_CONST_S16
) && val
== (int16_t)val
) {
328 } else if ((ct
& TCG_CT_CONST_U16
) && val
== (uint16_t)val
) {
330 } else if ((ct
& TCG_CT_CONST_S32
) && val
== (int32_t)val
) {
332 } else if ((ct
& TCG_CT_CONST_U32
) && val
== (uint32_t)val
) {
334 } else if ((ct
& TCG_CT_CONST_ZERO
) && val
== 0) {
336 } else if ((ct
& TCG_CT_CONST_MONE
) && val
== -1) {
338 } else if ((ct
& TCG_CT_CONST_WSZ
)
339 && val
== (type
== TCG_TYPE_I32
? 32 : 64)) {
345 #define OPCD(opc) ((opc)<<26)
346 #define XO19(opc) (OPCD(19)|((opc)<<1))
347 #define MD30(opc) (OPCD(30)|((opc)<<2))
348 #define MDS30(opc) (OPCD(30)|((opc)<<1))
349 #define XO31(opc) (OPCD(31)|((opc)<<1))
350 #define XO58(opc) (OPCD(58)|(opc))
351 #define XO62(opc) (OPCD(62)|(opc))
355 #define LBZ OPCD( 34)
356 #define LHZ OPCD( 40)
357 #define LHA OPCD( 42)
358 #define LWZ OPCD( 32)
359 #define STB OPCD( 38)
360 #define STH OPCD( 44)
361 #define STW OPCD( 36)
364 #define STDU XO62( 1)
365 #define STDX XO31(149)
368 #define LDX XO31( 21)
371 #define LWAX XO31(341)
373 #define ADDIC OPCD( 12)
374 #define ADDI OPCD( 14)
375 #define ADDIS OPCD( 15)
376 #define ORI OPCD( 24)
377 #define ORIS OPCD( 25)
378 #define XORI OPCD( 26)
379 #define XORIS OPCD( 27)
380 #define ANDI OPCD( 28)
381 #define ANDIS OPCD( 29)
382 #define MULLI OPCD( 7)
383 #define CMPLI OPCD( 10)
384 #define CMPI OPCD( 11)
385 #define SUBFIC OPCD( 8)
387 #define LWZU OPCD( 33)
388 #define STWU OPCD( 37)
390 #define RLWIMI OPCD( 20)
391 #define RLWINM OPCD( 21)
392 #define RLWNM OPCD( 23)
394 #define RLDICL MD30( 0)
395 #define RLDICR MD30( 1)
396 #define RLDIMI MD30( 3)
397 #define RLDCL MDS30( 8)
399 #define BCLR XO19( 16)
400 #define BCCTR XO19(528)
401 #define CRAND XO19(257)
402 #define CRANDC XO19(129)
403 #define CRNAND XO19(225)
404 #define CROR XO19(449)
405 #define CRNOR XO19( 33)
407 #define EXTSB XO31(954)
408 #define EXTSH XO31(922)
409 #define EXTSW XO31(986)
410 #define ADD XO31(266)
411 #define ADDE XO31(138)
412 #define ADDME XO31(234)
413 #define ADDZE XO31(202)
414 #define ADDC XO31( 10)
415 #define AND XO31( 28)
416 #define SUBF XO31( 40)
417 #define SUBFC XO31( 8)
418 #define SUBFE XO31(136)
419 #define SUBFME XO31(232)
420 #define SUBFZE XO31(200)
422 #define XOR XO31(316)
423 #define MULLW XO31(235)
424 #define MULHW XO31( 75)
425 #define MULHWU XO31( 11)
426 #define DIVW XO31(491)
427 #define DIVWU XO31(459)
429 #define CMPL XO31( 32)
430 #define LHBRX XO31(790)
431 #define LWBRX XO31(534)
432 #define LDBRX XO31(532)
433 #define STHBRX XO31(918)
434 #define STWBRX XO31(662)
435 #define STDBRX XO31(660)
436 #define MFSPR XO31(339)
437 #define MTSPR XO31(467)
438 #define SRAWI XO31(824)
439 #define NEG XO31(104)
440 #define MFCR XO31( 19)
441 #define MFOCRF (MFCR | (1u << 20))
442 #define NOR XO31(124)
443 #define CNTLZW XO31( 26)
444 #define CNTLZD XO31( 58)
445 #define CNTTZW XO31(538)
446 #define CNTTZD XO31(570)
447 #define CNTPOPW XO31(378)
448 #define CNTPOPD XO31(506)
449 #define ANDC XO31( 60)
450 #define ORC XO31(412)
451 #define EQV XO31(284)
452 #define NAND XO31(476)
453 #define ISEL XO31( 15)
455 #define MULLD XO31(233)
456 #define MULHD XO31( 73)
457 #define MULHDU XO31( 9)
458 #define DIVD XO31(489)
459 #define DIVDU XO31(457)
461 #define LBZX XO31( 87)
462 #define LHZX XO31(279)
463 #define LHAX XO31(343)
464 #define LWZX XO31( 23)
465 #define STBX XO31(215)
466 #define STHX XO31(407)
467 #define STWX XO31(151)
469 #define EIEIO XO31(854)
470 #define HWSYNC XO31(598)
471 #define LWSYNC (HWSYNC | (1u << 21))
473 #define SPR(a, b) ((((a)<<5)|(b))<<11)
475 #define CTR SPR(9, 0)
477 #define SLW XO31( 24)
478 #define SRW XO31(536)
479 #define SRAW XO31(792)
481 #define SLD XO31( 27)
482 #define SRD XO31(539)
483 #define SRAD XO31(794)
484 #define SRADI XO31(413<<1)
487 #define TRAP (TW | TO(31))
489 #define NOP ORI /* ori 0,0,0 */
491 #define RT(r) ((r)<<21)
492 #define RS(r) ((r)<<21)
493 #define RA(r) ((r)<<16)
494 #define RB(r) ((r)<<11)
495 #define TO(t) ((t)<<21)
496 #define SH(s) ((s)<<11)
497 #define MB(b) ((b)<<6)
498 #define ME(e) ((e)<<1)
499 #define BO(o) ((o)<<21)
500 #define MB64(b) ((b)<<5)
501 #define FXM(b) (1 << (19 - (b)))
505 #define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
506 #define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
507 #define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
508 #define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
510 #define BF(n) ((n)<<23)
511 #define BI(n, c) (((c)+((n)*4))<<16)
512 #define BT(n, c) (((c)+((n)*4))<<21)
513 #define BA(n, c) (((c)+((n)*4))<<16)
514 #define BB(n, c) (((c)+((n)*4))<<11)
515 #define BC_(n, c) (((c)+((n)*4))<<6)
517 #define BO_COND_TRUE BO(12)
518 #define BO_COND_FALSE BO( 4)
519 #define BO_ALWAYS BO(20)
528 static const uint32_t tcg_to_bc
[] = {
529 [TCG_COND_EQ
] = BC
| BI(7, CR_EQ
) | BO_COND_TRUE
,
530 [TCG_COND_NE
] = BC
| BI(7, CR_EQ
) | BO_COND_FALSE
,
531 [TCG_COND_LT
] = BC
| BI(7, CR_LT
) | BO_COND_TRUE
,
532 [TCG_COND_GE
] = BC
| BI(7, CR_LT
) | BO_COND_FALSE
,
533 [TCG_COND_LE
] = BC
| BI(7, CR_GT
) | BO_COND_FALSE
,
534 [TCG_COND_GT
] = BC
| BI(7, CR_GT
) | BO_COND_TRUE
,
535 [TCG_COND_LTU
] = BC
| BI(7, CR_LT
) | BO_COND_TRUE
,
536 [TCG_COND_GEU
] = BC
| BI(7, CR_LT
) | BO_COND_FALSE
,
537 [TCG_COND_LEU
] = BC
| BI(7, CR_GT
) | BO_COND_FALSE
,
538 [TCG_COND_GTU
] = BC
| BI(7, CR_GT
) | BO_COND_TRUE
,
541 /* The low bit here is set if the RA and RB fields must be inverted. */
542 static const uint32_t tcg_to_isel
[] = {
543 [TCG_COND_EQ
] = ISEL
| BC_(7, CR_EQ
),
544 [TCG_COND_NE
] = ISEL
| BC_(7, CR_EQ
) | 1,
545 [TCG_COND_LT
] = ISEL
| BC_(7, CR_LT
),
546 [TCG_COND_GE
] = ISEL
| BC_(7, CR_LT
) | 1,
547 [TCG_COND_LE
] = ISEL
| BC_(7, CR_GT
) | 1,
548 [TCG_COND_GT
] = ISEL
| BC_(7, CR_GT
),
549 [TCG_COND_LTU
] = ISEL
| BC_(7, CR_LT
),
550 [TCG_COND_GEU
] = ISEL
| BC_(7, CR_LT
) | 1,
551 [TCG_COND_LEU
] = ISEL
| BC_(7, CR_GT
) | 1,
552 [TCG_COND_GTU
] = ISEL
| BC_(7, CR_GT
),
555 static void tcg_out_mem_long(TCGContext
*s
, int opi
, int opx
, TCGReg rt
,
556 TCGReg base
, tcg_target_long offset
);
558 static void tcg_out_mov(TCGContext
*s
, TCGType type
, TCGReg ret
, TCGReg arg
)
560 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
562 tcg_out32(s
, OR
| SAB(arg
, ret
, arg
));
566 static inline void tcg_out_rld(TCGContext
*s
, int op
, TCGReg ra
, TCGReg rs
,
569 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64);
570 sh
= SH(sh
& 0x1f) | (((sh
>> 5) & 1) << 1);
571 mb
= MB64((mb
>> 5) | ((mb
<< 1) & 0x3f));
572 tcg_out32(s
, op
| RA(ra
) | RS(rs
) | sh
| mb
);
575 static inline void tcg_out_rlw(TCGContext
*s
, int op
, TCGReg ra
, TCGReg rs
,
576 int sh
, int mb
, int me
)
578 tcg_out32(s
, op
| RA(ra
) | RS(rs
) | SH(sh
) | MB(mb
) | ME(me
));
581 static inline void tcg_out_ext32u(TCGContext
*s
, TCGReg dst
, TCGReg src
)
583 tcg_out_rld(s
, RLDICL
, dst
, src
, 0, 32);
586 static inline void tcg_out_shli32(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
588 tcg_out_rlw(s
, RLWINM
, dst
, src
, c
, 0, 31 - c
);
591 static inline void tcg_out_shli64(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
593 tcg_out_rld(s
, RLDICR
, dst
, src
, c
, 63 - c
);
596 static inline void tcg_out_shri32(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
598 tcg_out_rlw(s
, RLWINM
, dst
, src
, 32 - c
, c
, 31);
601 static inline void tcg_out_shri64(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
603 tcg_out_rld(s
, RLDICL
, dst
, src
, 64 - c
, c
);
606 /* Emit a move into ret of arg, if it can be done in one insn. */
607 static bool tcg_out_movi_one(TCGContext
*s
, TCGReg ret
, tcg_target_long arg
)
609 if (arg
== (int16_t)arg
) {
610 tcg_out32(s
, ADDI
| TAI(ret
, 0, arg
));
613 if (arg
== (int32_t)arg
&& (arg
& 0xffff) == 0) {
614 tcg_out32(s
, ADDIS
| TAI(ret
, 0, arg
>> 16));
620 static void tcg_out_movi_int(TCGContext
*s
, TCGType type
, TCGReg ret
,
621 tcg_target_long arg
, bool in_prologue
)
627 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
629 if (TCG_TARGET_REG_BITS
== 64 && type
== TCG_TYPE_I32
) {
633 /* Load 16-bit immediates with one insn. */
634 if (tcg_out_movi_one(s
, ret
, arg
)) {
638 /* Load addresses within the TB with one insn. */
639 tb_diff
= arg
- (intptr_t)s
->code_gen_ptr
;
640 if (!in_prologue
&& USE_REG_TB
&& tb_diff
== (int16_t)tb_diff
) {
641 tcg_out32(s
, ADDI
| TAI(ret
, TCG_REG_TB
, tb_diff
));
645 /* Load 32-bit immediates with two insns. Note that we've already
646 eliminated bare ADDIS, so we know both insns are required. */
647 if (TCG_TARGET_REG_BITS
== 32 || arg
== (int32_t)arg
) {
648 tcg_out32(s
, ADDIS
| TAI(ret
, 0, arg
>> 16));
649 tcg_out32(s
, ORI
| SAI(ret
, ret
, arg
));
652 if (arg
== (uint32_t)arg
&& !(arg
& 0x8000)) {
653 tcg_out32(s
, ADDI
| TAI(ret
, 0, arg
));
654 tcg_out32(s
, ORIS
| SAI(ret
, ret
, arg
>> 16));
658 /* Load masked 16-bit value. */
659 if (arg
> 0 && (arg
& 0x8000)) {
661 if ((tmp
& (tmp
+ 1)) == 0) {
662 int mb
= clz64(tmp
+ 1) + 1;
663 tcg_out32(s
, ADDI
| TAI(ret
, 0, arg
));
664 tcg_out_rld(s
, RLDICL
, ret
, ret
, 0, mb
);
669 /* Load common masks with 2 insns. */
672 if (tmp
== (int16_t)tmp
) {
673 tcg_out32(s
, ADDI
| TAI(ret
, 0, tmp
));
674 tcg_out_shli64(s
, ret
, ret
, shift
);
678 if (tcg_out_movi_one(s
, ret
, arg
<< shift
)) {
679 tcg_out_shri64(s
, ret
, ret
, shift
);
683 /* Load addresses within 2GB of TB with 2 (or rarely 3) insns. */
684 if (!in_prologue
&& USE_REG_TB
&& tb_diff
== (int32_t)tb_diff
) {
685 tcg_out_mem_long(s
, ADDI
, ADD
, ret
, TCG_REG_TB
, tb_diff
);
689 /* Use the constant pool, if possible. */
690 if (!in_prologue
&& USE_REG_TB
) {
691 new_pool_label(s
, arg
, R_PPC_ADDR16
, s
->code_ptr
,
692 -(intptr_t)s
->code_gen_ptr
);
693 tcg_out32(s
, LD
| TAI(ret
, TCG_REG_TB
, 0));
697 tmp
= arg
>> 31 >> 1;
698 tcg_out_movi(s
, TCG_TYPE_I32
, ret
, tmp
);
700 tcg_out_shli64(s
, ret
, ret
, 32);
702 if (arg
& 0xffff0000) {
703 tcg_out32(s
, ORIS
| SAI(ret
, ret
, arg
>> 16));
706 tcg_out32(s
, ORI
| SAI(ret
, ret
, arg
));
710 static inline void tcg_out_movi(TCGContext
*s
, TCGType type
, TCGReg ret
,
713 tcg_out_movi_int(s
, type
, ret
, arg
, false);
716 static bool mask_operand(uint32_t c
, int *mb
, int *me
)
720 /* Accept a bit pattern like:
724 Keep track of the transitions. */
725 if (c
== 0 || c
== -1) {
731 if (test
& (test
- 1)) {
736 *mb
= test
? clz32(test
& -test
) + 1 : 0;
740 static bool mask64_operand(uint64_t c
, int *mb
, int *me
)
749 /* Accept 1..10..0. */
755 /* Accept 0..01..1. */
756 if (lsb
== 1 && (c
& (c
+ 1)) == 0) {
757 *mb
= clz64(c
+ 1) + 1;
764 static void tcg_out_andi32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
)
768 if (mask_operand(c
, &mb
, &me
)) {
769 tcg_out_rlw(s
, RLWINM
, dst
, src
, 0, mb
, me
);
770 } else if ((c
& 0xffff) == c
) {
771 tcg_out32(s
, ANDI
| SAI(src
, dst
, c
));
773 } else if ((c
& 0xffff0000) == c
) {
774 tcg_out32(s
, ANDIS
| SAI(src
, dst
, c
>> 16));
777 tcg_out_movi(s
, TCG_TYPE_I32
, TCG_REG_R0
, c
);
778 tcg_out32(s
, AND
| SAB(src
, dst
, TCG_REG_R0
));
782 static void tcg_out_andi64(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint64_t c
)
786 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64);
787 if (mask64_operand(c
, &mb
, &me
)) {
789 tcg_out_rld(s
, RLDICR
, dst
, src
, 0, me
);
791 tcg_out_rld(s
, RLDICL
, dst
, src
, 0, mb
);
793 } else if ((c
& 0xffff) == c
) {
794 tcg_out32(s
, ANDI
| SAI(src
, dst
, c
));
796 } else if ((c
& 0xffff0000) == c
) {
797 tcg_out32(s
, ANDIS
| SAI(src
, dst
, c
>> 16));
800 tcg_out_movi(s
, TCG_TYPE_I64
, TCG_REG_R0
, c
);
801 tcg_out32(s
, AND
| SAB(src
, dst
, TCG_REG_R0
));
805 static void tcg_out_zori32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
,
806 int op_lo
, int op_hi
)
809 tcg_out32(s
, op_hi
| SAI(src
, dst
, c
>> 16));
813 tcg_out32(s
, op_lo
| SAI(src
, dst
, c
));
818 static void tcg_out_ori32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
)
820 tcg_out_zori32(s
, dst
, src
, c
, ORI
, ORIS
);
823 static void tcg_out_xori32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
)
825 tcg_out_zori32(s
, dst
, src
, c
, XORI
, XORIS
);
828 static void tcg_out_b(TCGContext
*s
, int mask
, tcg_insn_unit
*target
)
830 ptrdiff_t disp
= tcg_pcrel_diff(s
, target
);
831 if (in_range_b(disp
)) {
832 tcg_out32(s
, B
| (disp
& 0x3fffffc) | mask
);
834 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_R0
, (uintptr_t)target
);
835 tcg_out32(s
, MTSPR
| RS(TCG_REG_R0
) | CTR
);
836 tcg_out32(s
, BCCTR
| BO_ALWAYS
| mask
);
840 static void tcg_out_mem_long(TCGContext
*s
, int opi
, int opx
, TCGReg rt
,
841 TCGReg base
, tcg_target_long offset
)
843 tcg_target_long orig
= offset
, l0
, l1
, extra
= 0, align
= 0;
844 bool is_store
= false;
845 TCGReg rs
= TCG_REG_TMP1
;
852 if (rt
!= TCG_REG_R0
) {
860 case STB
: case STH
: case STW
:
865 /* For unaligned, or very large offsets, use the indexed form. */
866 if (offset
& align
|| offset
!= (int32_t)offset
) {
870 tcg_debug_assert(!is_store
|| rs
!= rt
);
871 tcg_out_movi(s
, TCG_TYPE_PTR
, rs
, orig
);
872 tcg_out32(s
, opx
| TAB(rt
, base
, rs
));
876 l0
= (int16_t)offset
;
877 offset
= (offset
- l0
) >> 16;
878 l1
= (int16_t)offset
;
880 if (l1
< 0 && orig
>= 0) {
882 l1
= (int16_t)(offset
- 0x4000);
885 tcg_out32(s
, ADDIS
| TAI(rs
, base
, l1
));
889 tcg_out32(s
, ADDIS
| TAI(rs
, base
, extra
));
892 if (opi
!= ADDI
|| base
!= rt
|| l0
!= 0) {
893 tcg_out32(s
, opi
| TAI(rt
, base
, l0
));
897 static inline void tcg_out_ld(TCGContext
*s
, TCGType type
, TCGReg ret
,
898 TCGReg arg1
, intptr_t arg2
)
902 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
903 if (type
== TCG_TYPE_I32
) {
904 opi
= LWZ
, opx
= LWZX
;
908 tcg_out_mem_long(s
, opi
, opx
, ret
, arg1
, arg2
);
911 static inline void tcg_out_st(TCGContext
*s
, TCGType type
, TCGReg arg
,
912 TCGReg arg1
, intptr_t arg2
)
916 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
917 if (type
== TCG_TYPE_I32
) {
918 opi
= STW
, opx
= STWX
;
920 opi
= STD
, opx
= STDX
;
922 tcg_out_mem_long(s
, opi
, opx
, arg
, arg1
, arg2
);
925 static inline bool tcg_out_sti(TCGContext
*s
, TCGType type
, TCGArg val
,
926 TCGReg base
, intptr_t ofs
)
931 static void tcg_out_cmp(TCGContext
*s
, int cond
, TCGArg arg1
, TCGArg arg2
,
932 int const_arg2
, int cr
, TCGType type
)
937 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
939 /* Simplify the comparisons below wrt CMPI. */
940 if (type
== TCG_TYPE_I32
) {
941 arg2
= (int32_t)arg2
;
948 if ((int16_t) arg2
== arg2
) {
952 } else if ((uint16_t) arg2
== arg2
) {
967 if ((int16_t) arg2
== arg2
) {
982 if ((uint16_t) arg2
== arg2
) {
995 op
|= BF(cr
) | ((type
== TCG_TYPE_I64
) << 21);
998 tcg_out32(s
, op
| RA(arg1
) | (arg2
& 0xffff));
1001 tcg_out_movi(s
, type
, TCG_REG_R0
, arg2
);
1004 tcg_out32(s
, op
| RA(arg1
) | RB(arg2
));
1008 static void tcg_out_setcond_eq0(TCGContext
*s
, TCGType type
,
1009 TCGReg dst
, TCGReg src
)
1011 if (type
== TCG_TYPE_I32
) {
1012 tcg_out32(s
, CNTLZW
| RS(src
) | RA(dst
));
1013 tcg_out_shri32(s
, dst
, dst
, 5);
1015 tcg_out32(s
, CNTLZD
| RS(src
) | RA(dst
));
1016 tcg_out_shri64(s
, dst
, dst
, 6);
1020 static void tcg_out_setcond_ne0(TCGContext
*s
, TCGReg dst
, TCGReg src
)
1022 /* X != 0 implies X + -1 generates a carry. Extra addition
1023 trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C. */
1025 tcg_out32(s
, ADDIC
| TAI(dst
, src
, -1));
1026 tcg_out32(s
, SUBFE
| TAB(dst
, dst
, src
));
1028 tcg_out32(s
, ADDIC
| TAI(TCG_REG_R0
, src
, -1));
1029 tcg_out32(s
, SUBFE
| TAB(dst
, TCG_REG_R0
, src
));
1033 static TCGReg
tcg_gen_setcond_xor(TCGContext
*s
, TCGReg arg1
, TCGArg arg2
,
1037 if ((uint32_t)arg2
== arg2
) {
1038 tcg_out_xori32(s
, TCG_REG_R0
, arg1
, arg2
);
1040 tcg_out_movi(s
, TCG_TYPE_I64
, TCG_REG_R0
, arg2
);
1041 tcg_out32(s
, XOR
| SAB(arg1
, TCG_REG_R0
, TCG_REG_R0
));
1044 tcg_out32(s
, XOR
| SAB(arg1
, TCG_REG_R0
, arg2
));
1049 static void tcg_out_setcond(TCGContext
*s
, TCGType type
, TCGCond cond
,
1050 TCGArg arg0
, TCGArg arg1
, TCGArg arg2
,
1055 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
1057 /* Ignore high bits of a potential constant arg2. */
1058 if (type
== TCG_TYPE_I32
) {
1059 arg2
= (uint32_t)arg2
;
1062 /* Handle common and trivial cases before handling anything else. */
1066 tcg_out_setcond_eq0(s
, type
, arg0
, arg1
);
1069 if (TCG_TARGET_REG_BITS
== 64 && type
== TCG_TYPE_I32
) {
1070 tcg_out_ext32u(s
, TCG_REG_R0
, arg1
);
1073 tcg_out_setcond_ne0(s
, arg0
, arg1
);
1076 tcg_out32(s
, NOR
| SAB(arg1
, arg0
, arg1
));
1080 /* Extract the sign bit. */
1081 if (type
== TCG_TYPE_I32
) {
1082 tcg_out_shri32(s
, arg0
, arg1
, 31);
1084 tcg_out_shri64(s
, arg0
, arg1
, 63);
1092 /* If we have ISEL, we can implement everything with 3 or 4 insns.
1093 All other cases below are also at least 3 insns, so speed up the
1094 code generator by not considering them and always using ISEL. */
1098 tcg_out_cmp(s
, cond
, arg1
, arg2
, const_arg2
, 7, type
);
1100 isel
= tcg_to_isel
[cond
];
1102 tcg_out_movi(s
, type
, arg0
, 1);
1104 /* arg0 = (bc ? 0 : 1) */
1105 tab
= TAB(arg0
, 0, arg0
);
1108 /* arg0 = (bc ? 1 : 0) */
1109 tcg_out_movi(s
, type
, TCG_REG_R0
, 0);
1110 tab
= TAB(arg0
, arg0
, TCG_REG_R0
);
1112 tcg_out32(s
, isel
| tab
);
1118 arg1
= tcg_gen_setcond_xor(s
, arg1
, arg2
, const_arg2
);
1119 tcg_out_setcond_eq0(s
, type
, arg0
, arg1
);
1123 arg1
= tcg_gen_setcond_xor(s
, arg1
, arg2
, const_arg2
);
1124 /* Discard the high bits only once, rather than both inputs. */
1125 if (TCG_TARGET_REG_BITS
== 64 && type
== TCG_TYPE_I32
) {
1126 tcg_out_ext32u(s
, TCG_REG_R0
, arg1
);
1129 tcg_out_setcond_ne0(s
, arg0
, arg1
);
1147 crop
= CRNOR
| BT(7, CR_EQ
) | BA(7, CR_LT
) | BB(7, CR_LT
);
1153 crop
= CRNOR
| BT(7, CR_EQ
) | BA(7, CR_GT
) | BB(7, CR_GT
);
1155 tcg_out_cmp(s
, cond
, arg1
, arg2
, const_arg2
, 7, type
);
1159 tcg_out32(s
, MFOCRF
| RT(TCG_REG_R0
) | FXM(7));
1160 tcg_out_rlw(s
, RLWINM
, arg0
, TCG_REG_R0
, sh
, 31, 31);
1168 static void tcg_out_bc(TCGContext
*s
, int bc
, TCGLabel
*l
)
1171 tcg_out32(s
, bc
| reloc_pc14_val(s
->code_ptr
, l
->u
.value_ptr
));
1173 tcg_out_reloc(s
, s
->code_ptr
, R_PPC_REL14
, l
, 0);
1174 tcg_out_bc_noaddr(s
, bc
);
1178 static void tcg_out_brcond(TCGContext
*s
, TCGCond cond
,
1179 TCGArg arg1
, TCGArg arg2
, int const_arg2
,
1180 TCGLabel
*l
, TCGType type
)
1182 tcg_out_cmp(s
, cond
, arg1
, arg2
, const_arg2
, 7, type
);
1183 tcg_out_bc(s
, tcg_to_bc
[cond
], l
);
1186 static void tcg_out_movcond(TCGContext
*s
, TCGType type
, TCGCond cond
,
1187 TCGArg dest
, TCGArg c1
, TCGArg c2
, TCGArg v1
,
1188 TCGArg v2
, bool const_c2
)
1190 /* If for some reason both inputs are zero, don't produce bad code. */
1191 if (v1
== 0 && v2
== 0) {
1192 tcg_out_movi(s
, type
, dest
, 0);
1196 tcg_out_cmp(s
, cond
, c1
, c2
, const_c2
, 7, type
);
1199 int isel
= tcg_to_isel
[cond
];
1201 /* Swap the V operands if the operation indicates inversion. */
1208 /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand. */
1210 tcg_out_movi(s
, type
, TCG_REG_R0
, 0);
1212 tcg_out32(s
, isel
| TAB(dest
, v1
, v2
));
1215 cond
= tcg_invert_cond(cond
);
1217 } else if (dest
!= v1
) {
1219 tcg_out_movi(s
, type
, dest
, 0);
1221 tcg_out_mov(s
, type
, dest
, v1
);
1224 /* Branch forward over one insn */
1225 tcg_out32(s
, tcg_to_bc
[cond
] | 8);
1227 tcg_out_movi(s
, type
, dest
, 0);
1229 tcg_out_mov(s
, type
, dest
, v2
);
1234 static void tcg_out_cntxz(TCGContext
*s
, TCGType type
, uint32_t opc
,
1235 TCGArg a0
, TCGArg a1
, TCGArg a2
, bool const_a2
)
1237 if (const_a2
&& a2
== (type
== TCG_TYPE_I32
? 32 : 64)) {
1238 tcg_out32(s
, opc
| RA(a0
) | RS(a1
));
1240 tcg_out_cmp(s
, TCG_COND_EQ
, a1
, 0, 1, 7, type
);
1241 /* Note that the only other valid constant for a2 is 0. */
1243 tcg_out32(s
, opc
| RA(TCG_REG_R0
) | RS(a1
));
1244 tcg_out32(s
, tcg_to_isel
[TCG_COND_EQ
] | TAB(a0
, a2
, TCG_REG_R0
));
1245 } else if (!const_a2
&& a0
== a2
) {
1246 tcg_out32(s
, tcg_to_bc
[TCG_COND_EQ
] | 8);
1247 tcg_out32(s
, opc
| RA(a0
) | RS(a1
));
1249 tcg_out32(s
, opc
| RA(a0
) | RS(a1
));
1250 tcg_out32(s
, tcg_to_bc
[TCG_COND_NE
] | 8);
1252 tcg_out_movi(s
, type
, a0
, 0);
1254 tcg_out_mov(s
, type
, a0
, a2
);
1260 static void tcg_out_cmp2(TCGContext
*s
, const TCGArg
*args
,
1261 const int *const_args
)
1263 static const struct { uint8_t bit1
, bit2
; } bits
[] = {
1264 [TCG_COND_LT
] = { CR_LT
, CR_LT
},
1265 [TCG_COND_LE
] = { CR_LT
, CR_GT
},
1266 [TCG_COND_GT
] = { CR_GT
, CR_GT
},
1267 [TCG_COND_GE
] = { CR_GT
, CR_LT
},
1268 [TCG_COND_LTU
] = { CR_LT
, CR_LT
},
1269 [TCG_COND_LEU
] = { CR_LT
, CR_GT
},
1270 [TCG_COND_GTU
] = { CR_GT
, CR_GT
},
1271 [TCG_COND_GEU
] = { CR_GT
, CR_LT
},
1274 TCGCond cond
= args
[4], cond2
;
1275 TCGArg al
, ah
, bl
, bh
;
1276 int blconst
, bhconst
;
1283 blconst
= const_args
[2];
1284 bhconst
= const_args
[3];
1293 tcg_out_cmp(s
, cond
, al
, bl
, blconst
, 6, TCG_TYPE_I32
);
1294 tcg_out_cmp(s
, cond
, ah
, bh
, bhconst
, 7, TCG_TYPE_I32
);
1295 tcg_out32(s
, op
| BT(7, CR_EQ
) | BA(6, CR_EQ
) | BB(7, CR_EQ
));
1306 bit1
= bits
[cond
].bit1
;
1307 bit2
= bits
[cond
].bit2
;
1308 op
= (bit1
!= bit2
? CRANDC
: CRAND
);
1309 cond2
= tcg_unsigned_cond(cond
);
1311 tcg_out_cmp(s
, cond
, ah
, bh
, bhconst
, 6, TCG_TYPE_I32
);
1312 tcg_out_cmp(s
, cond2
, al
, bl
, blconst
, 7, TCG_TYPE_I32
);
1313 tcg_out32(s
, op
| BT(7, CR_EQ
) | BA(6, CR_EQ
) | BB(7, bit2
));
1314 tcg_out32(s
, CROR
| BT(7, CR_EQ
) | BA(6, bit1
) | BB(7, CR_EQ
));
1322 static void tcg_out_setcond2(TCGContext
*s
, const TCGArg
*args
,
1323 const int *const_args
)
1325 tcg_out_cmp2(s
, args
+ 1, const_args
+ 1);
1326 tcg_out32(s
, MFOCRF
| RT(TCG_REG_R0
) | FXM(7));
1327 tcg_out_rlw(s
, RLWINM
, args
[0], TCG_REG_R0
, 31, 31, 31);
1330 static void tcg_out_brcond2 (TCGContext
*s
, const TCGArg
*args
,
1331 const int *const_args
)
1333 tcg_out_cmp2(s
, args
, const_args
);
1334 tcg_out_bc(s
, BC
| BI(7, CR_EQ
) | BO_COND_TRUE
, arg_label(args
[5]));
1337 static void tcg_out_mb(TCGContext
*s
, TCGArg a0
)
1339 uint32_t insn
= HWSYNC
;
1341 if (a0
== TCG_MO_LD_LD
) {
1343 } else if (a0
== TCG_MO_ST_ST
) {
1349 void tb_target_set_jmp_target(uintptr_t tc_ptr
, uintptr_t jmp_addr
,
1352 if (TCG_TARGET_REG_BITS
== 64) {
1353 tcg_insn_unit i1
, i2
;
1354 intptr_t tb_diff
= addr
- tc_ptr
;
1355 intptr_t br_diff
= addr
- (jmp_addr
+ 4);
1358 /* This does not exercise the range of the branch, but we do
1359 still need to be able to load the new value of TCG_REG_TB.
1360 But this does still happen quite often. */
1361 if (tb_diff
== (int16_t)tb_diff
) {
1362 i1
= ADDI
| TAI(TCG_REG_TB
, TCG_REG_TB
, tb_diff
);
1363 i2
= B
| (br_diff
& 0x3fffffc);
1365 intptr_t lo
= (int16_t)tb_diff
;
1366 intptr_t hi
= (int32_t)(tb_diff
- lo
);
1367 assert(tb_diff
== hi
+ lo
);
1368 i1
= ADDIS
| TAI(TCG_REG_TB
, TCG_REG_TB
, hi
>> 16);
1369 i2
= ADDI
| TAI(TCG_REG_TB
, TCG_REG_TB
, lo
);
1371 #ifdef HOST_WORDS_BIGENDIAN
1372 pair
= (uint64_t)i1
<< 32 | i2
;
1374 pair
= (uint64_t)i2
<< 32 | i1
;
1377 /* As per the enclosing if, this is ppc64. Avoid the _Static_assert
1378 within atomic_set that would fail to build a ppc32 host. */
1379 atomic_set__nocheck((uint64_t *)jmp_addr
, pair
);
1380 flush_icache_range(jmp_addr
, jmp_addr
+ 8);
1382 intptr_t diff
= addr
- jmp_addr
;
1383 tcg_debug_assert(in_range_b(diff
));
1384 atomic_set((uint32_t *)jmp_addr
, B
| (diff
& 0x3fffffc));
1385 flush_icache_range(jmp_addr
, jmp_addr
+ 4);
1389 static void tcg_out_call(TCGContext
*s
, tcg_insn_unit
*target
)
1392 /* Look through the descriptor. If the branch is in range, and we
1393 don't have to spend too much effort on building the toc. */
1394 void *tgt
= ((void **)target
)[0];
1395 uintptr_t toc
= ((uintptr_t *)target
)[1];
1396 intptr_t diff
= tcg_pcrel_diff(s
, tgt
);
1398 if (in_range_b(diff
) && toc
== (uint32_t)toc
) {
1399 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_TMP1
, toc
);
1400 tcg_out_b(s
, LK
, tgt
);
1402 /* Fold the low bits of the constant into the addresses below. */
1403 intptr_t arg
= (intptr_t)target
;
1404 int ofs
= (int16_t)arg
;
1406 if (ofs
+ 8 < 0x8000) {
1411 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_TMP1
, arg
);
1412 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R0
, TCG_REG_TMP1
, ofs
);
1413 tcg_out32(s
, MTSPR
| RA(TCG_REG_R0
) | CTR
);
1414 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R2
, TCG_REG_TMP1
, ofs
+ SZP
);
1415 tcg_out32(s
, BCCTR
| BO_ALWAYS
| LK
);
1417 #elif defined(_CALL_ELF) && _CALL_ELF == 2
1420 /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1421 address, which the callee uses to compute its TOC address. */
1422 /* FIXME: when the branch is in range, we could avoid r12 load if we
1423 knew that the destination uses the same TOC, and what its local
1424 entry point offset is. */
1425 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_R12
, (intptr_t)target
);
1427 diff
= tcg_pcrel_diff(s
, target
);
1428 if (in_range_b(diff
)) {
1429 tcg_out_b(s
, LK
, target
);
1431 tcg_out32(s
, MTSPR
| RS(TCG_REG_R12
) | CTR
);
1432 tcg_out32(s
, BCCTR
| BO_ALWAYS
| LK
);
1435 tcg_out_b(s
, LK
, target
);
1439 static const uint32_t qemu_ldx_opc
[16] = {
1446 [MO_BSWAP
| MO_UB
] = LBZX
,
1447 [MO_BSWAP
| MO_UW
] = LHBRX
,
1448 [MO_BSWAP
| MO_UL
] = LWBRX
,
1449 [MO_BSWAP
| MO_Q
] = LDBRX
,
1452 static const uint32_t qemu_stx_opc
[16] = {
1457 [MO_BSWAP
| MO_UB
] = STBX
,
1458 [MO_BSWAP
| MO_UW
] = STHBRX
,
1459 [MO_BSWAP
| MO_UL
] = STWBRX
,
1460 [MO_BSWAP
| MO_Q
] = STDBRX
,
1463 static const uint32_t qemu_exts_opc
[4] = {
1464 EXTSB
, EXTSH
, EXTSW
, 0
1467 #if defined (CONFIG_SOFTMMU)
1468 #include "tcg-ldst.inc.c"
1470 /* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
1471 * int mmu_idx, uintptr_t ra)
1473 static void * const qemu_ld_helpers
[16] = {
1474 [MO_UB
] = helper_ret_ldub_mmu
,
1475 [MO_LEUW
] = helper_le_lduw_mmu
,
1476 [MO_LEUL
] = helper_le_ldul_mmu
,
1477 [MO_LEQ
] = helper_le_ldq_mmu
,
1478 [MO_BEUW
] = helper_be_lduw_mmu
,
1479 [MO_BEUL
] = helper_be_ldul_mmu
,
1480 [MO_BEQ
] = helper_be_ldq_mmu
,
1483 /* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
1484 * uintxx_t val, int mmu_idx, uintptr_t ra)
1486 static void * const qemu_st_helpers
[16] = {
1487 [MO_UB
] = helper_ret_stb_mmu
,
1488 [MO_LEUW
] = helper_le_stw_mmu
,
1489 [MO_LEUL
] = helper_le_stl_mmu
,
1490 [MO_LEQ
] = helper_le_stq_mmu
,
1491 [MO_BEUW
] = helper_be_stw_mmu
,
1492 [MO_BEUL
] = helper_be_stl_mmu
,
1493 [MO_BEQ
] = helper_be_stq_mmu
,
1496 /* Perform the TLB load and compare. Places the result of the comparison
1497 in CR7, loads the addend of the TLB into R3, and returns the register
1498 containing the guest address (zero-extended into R4). Clobbers R0 and R2. */
1500 static TCGReg
tcg_out_tlb_read(TCGContext
*s
, TCGMemOp opc
,
1501 TCGReg addrlo
, TCGReg addrhi
,
1502 int mem_index
, bool is_read
)
1506 ? offsetof(CPUArchState
, tlb_table
[mem_index
][0].addr_read
)
1507 : offsetof(CPUArchState
, tlb_table
[mem_index
][0].addr_write
));
1508 int add_off
= offsetof(CPUArchState
, tlb_table
[mem_index
][0].addend
);
1509 TCGReg base
= TCG_AREG0
;
1510 unsigned s_bits
= opc
& MO_SIZE
;
1511 unsigned a_bits
= get_alignment_bits(opc
);
1513 /* Extract the page index, shifted into place for tlb index. */
1514 if (TCG_TARGET_REG_BITS
== 64) {
1515 if (TARGET_LONG_BITS
== 32) {
1516 /* Zero-extend the address into a place helpful for further use. */
1517 tcg_out_ext32u(s
, TCG_REG_R4
, addrlo
);
1518 addrlo
= TCG_REG_R4
;
1520 tcg_out_rld(s
, RLDICL
, TCG_REG_R3
, addrlo
,
1521 64 - TARGET_PAGE_BITS
, 64 - CPU_TLB_BITS
);
1525 /* Compensate for very large offsets. */
1526 if (add_off
>= 0x8000) {
1527 int low
= (int16_t)cmp_off
;
1528 int high
= cmp_off
- low
;
1529 assert((high
& 0xffff) == 0);
1530 assert(cmp_off
- high
== (int16_t)(cmp_off
- high
));
1531 assert(add_off
- high
== (int16_t)(add_off
- high
));
1532 tcg_out32(s
, ADDIS
| TAI(TCG_REG_TMP1
, base
, high
>> 16));
1533 base
= TCG_REG_TMP1
;
1538 /* Extraction and shifting, part 2. */
1539 if (TCG_TARGET_REG_BITS
== 32 || TARGET_LONG_BITS
== 32) {
1540 tcg_out_rlw(s
, RLWINM
, TCG_REG_R3
, addrlo
,
1541 32 - (TARGET_PAGE_BITS
- CPU_TLB_ENTRY_BITS
),
1542 32 - (CPU_TLB_BITS
+ CPU_TLB_ENTRY_BITS
),
1543 31 - CPU_TLB_ENTRY_BITS
);
1545 tcg_out_shli64(s
, TCG_REG_R3
, TCG_REG_R3
, CPU_TLB_ENTRY_BITS
);
1548 tcg_out32(s
, ADD
| TAB(TCG_REG_R3
, TCG_REG_R3
, base
));
1550 /* Load the tlb comparator. */
1551 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1552 tcg_out_ld(s
, TCG_TYPE_I32
, TCG_REG_R4
, TCG_REG_R3
, cmp_off
);
1553 tcg_out_ld(s
, TCG_TYPE_I32
, TCG_REG_TMP1
, TCG_REG_R3
, cmp_off
+ 4);
1555 tcg_out_ld(s
, TCG_TYPE_TL
, TCG_REG_TMP1
, TCG_REG_R3
, cmp_off
);
1558 /* Load the TLB addend for use on the fast path. Do this asap
1559 to minimize any load use delay. */
1560 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R3
, TCG_REG_R3
, add_off
);
1562 /* Clear the non-page, non-alignment bits from the address */
1563 if (TCG_TARGET_REG_BITS
== 32) {
1564 /* We don't support unaligned accesses on 32-bits.
1565 * Preserve the bottom bits and thus trigger a comparison
1566 * failure on unaligned accesses.
1568 if (a_bits
< s_bits
) {
1571 tcg_out_rlw(s
, RLWINM
, TCG_REG_R0
, addrlo
, 0,
1572 (32 - a_bits
) & 31, 31 - TARGET_PAGE_BITS
);
1576 /* If the access is unaligned, we need to make sure we fail if we
1577 * cross a page boundary. The trick is to add the access size-1
1578 * to the address before masking the low bits. That will make the
1579 * address overflow to the next page if we cross a page boundary,
1580 * which will then force a mismatch of the TLB compare.
1582 if (a_bits
< s_bits
) {
1583 unsigned a_mask
= (1 << a_bits
) - 1;
1584 unsigned s_mask
= (1 << s_bits
) - 1;
1585 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, t
, s_mask
- a_mask
));
1589 /* Mask the address for the requested alignment. */
1590 if (TARGET_LONG_BITS
== 32) {
1591 tcg_out_rlw(s
, RLWINM
, TCG_REG_R0
, t
, 0,
1592 (32 - a_bits
) & 31, 31 - TARGET_PAGE_BITS
);
1593 } else if (a_bits
== 0) {
1594 tcg_out_rld(s
, RLDICR
, TCG_REG_R0
, t
, 0, 63 - TARGET_PAGE_BITS
);
1596 tcg_out_rld(s
, RLDICL
, TCG_REG_R0
, t
,
1597 64 - TARGET_PAGE_BITS
, TARGET_PAGE_BITS
- a_bits
);
1598 tcg_out_rld(s
, RLDICL
, TCG_REG_R0
, TCG_REG_R0
, TARGET_PAGE_BITS
, 0);
1602 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1603 tcg_out_cmp(s
, TCG_COND_EQ
, TCG_REG_R0
, TCG_REG_TMP1
,
1604 0, 7, TCG_TYPE_I32
);
1605 tcg_out_cmp(s
, TCG_COND_EQ
, addrhi
, TCG_REG_R4
, 0, 6, TCG_TYPE_I32
);
1606 tcg_out32(s
, CRAND
| BT(7, CR_EQ
) | BA(6, CR_EQ
) | BB(7, CR_EQ
));
1608 tcg_out_cmp(s
, TCG_COND_EQ
, TCG_REG_R0
, TCG_REG_TMP1
,
1615 /* Record the context of a call to the out of line helper code for the slow
1616 path for a load or store, so that we can later generate the correct
1618 static void add_qemu_ldst_label(TCGContext
*s
, bool is_ld
, TCGMemOpIdx oi
,
1619 TCGReg datalo_reg
, TCGReg datahi_reg
,
1620 TCGReg addrlo_reg
, TCGReg addrhi_reg
,
1621 tcg_insn_unit
*raddr
, tcg_insn_unit
*lptr
)
1623 TCGLabelQemuLdst
*label
= new_ldst_label(s
);
1625 label
->is_ld
= is_ld
;
1627 label
->datalo_reg
= datalo_reg
;
1628 label
->datahi_reg
= datahi_reg
;
1629 label
->addrlo_reg
= addrlo_reg
;
1630 label
->addrhi_reg
= addrhi_reg
;
1631 label
->raddr
= raddr
;
1632 label
->label_ptr
[0] = lptr
;
1635 static void tcg_out_qemu_ld_slow_path(TCGContext
*s
, TCGLabelQemuLdst
*lb
)
1637 TCGMemOpIdx oi
= lb
->oi
;
1638 TCGMemOp opc
= get_memop(oi
);
1639 TCGReg hi
, lo
, arg
= TCG_REG_R3
;
1641 reloc_pc14(lb
->label_ptr
[0], s
->code_ptr
);
1643 tcg_out_mov(s
, TCG_TYPE_PTR
, arg
++, TCG_AREG0
);
1645 lo
= lb
->addrlo_reg
;
1646 hi
= lb
->addrhi_reg
;
1647 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1648 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1651 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, hi
);
1652 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, lo
);
1654 /* If the address needed to be zero-extended, we'll have already
1655 placed it in R4. The only remaining case is 64-bit guest. */
1656 tcg_out_mov(s
, TCG_TYPE_TL
, arg
++, lo
);
1659 tcg_out_movi(s
, TCG_TYPE_I32
, arg
++, oi
);
1660 tcg_out32(s
, MFSPR
| RT(arg
) | LR
);
1662 tcg_out_call(s
, qemu_ld_helpers
[opc
& (MO_BSWAP
| MO_SIZE
)]);
1664 lo
= lb
->datalo_reg
;
1665 hi
= lb
->datahi_reg
;
1666 if (TCG_TARGET_REG_BITS
== 32 && (opc
& MO_SIZE
) == MO_64
) {
1667 tcg_out_mov(s
, TCG_TYPE_I32
, lo
, TCG_REG_R4
);
1668 tcg_out_mov(s
, TCG_TYPE_I32
, hi
, TCG_REG_R3
);
1669 } else if (opc
& MO_SIGN
) {
1670 uint32_t insn
= qemu_exts_opc
[opc
& MO_SIZE
];
1671 tcg_out32(s
, insn
| RA(lo
) | RS(TCG_REG_R3
));
1673 tcg_out_mov(s
, TCG_TYPE_REG
, lo
, TCG_REG_R3
);
1676 tcg_out_b(s
, 0, lb
->raddr
);
1679 static void tcg_out_qemu_st_slow_path(TCGContext
*s
, TCGLabelQemuLdst
*lb
)
1681 TCGMemOpIdx oi
= lb
->oi
;
1682 TCGMemOp opc
= get_memop(oi
);
1683 TCGMemOp s_bits
= opc
& MO_SIZE
;
1684 TCGReg hi
, lo
, arg
= TCG_REG_R3
;
1686 reloc_pc14(lb
->label_ptr
[0], s
->code_ptr
);
1688 tcg_out_mov(s
, TCG_TYPE_PTR
, arg
++, TCG_AREG0
);
1690 lo
= lb
->addrlo_reg
;
1691 hi
= lb
->addrhi_reg
;
1692 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1693 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1696 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, hi
);
1697 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, lo
);
1699 /* If the address needed to be zero-extended, we'll have already
1700 placed it in R4. The only remaining case is 64-bit guest. */
1701 tcg_out_mov(s
, TCG_TYPE_TL
, arg
++, lo
);
1704 lo
= lb
->datalo_reg
;
1705 hi
= lb
->datahi_reg
;
1706 if (TCG_TARGET_REG_BITS
== 32) {
1709 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1712 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, hi
);
1715 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, lo
);
1718 tcg_out_rlw(s
, RLWINM
, arg
++, lo
, 0, 32 - (8 << s_bits
), 31);
1722 if (s_bits
== MO_64
) {
1723 tcg_out_mov(s
, TCG_TYPE_I64
, arg
++, lo
);
1725 tcg_out_rld(s
, RLDICL
, arg
++, lo
, 0, 64 - (8 << s_bits
));
1729 tcg_out_movi(s
, TCG_TYPE_I32
, arg
++, oi
);
1730 tcg_out32(s
, MFSPR
| RT(arg
) | LR
);
1732 tcg_out_call(s
, qemu_st_helpers
[opc
& (MO_BSWAP
| MO_SIZE
)]);
1734 tcg_out_b(s
, 0, lb
->raddr
);
1736 #endif /* SOFTMMU */
1738 static void tcg_out_qemu_ld(TCGContext
*s
, const TCGArg
*args
, bool is_64
)
1740 TCGReg datalo
, datahi
, addrlo
, rbase
;
1741 TCGReg addrhi
__attribute__((unused
));
1743 TCGMemOp opc
, s_bits
;
1744 #ifdef CONFIG_SOFTMMU
1746 tcg_insn_unit
*label_ptr
;
1750 datahi
= (TCG_TARGET_REG_BITS
== 32 && is_64
? *args
++ : 0);
1752 addrhi
= (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
? *args
++ : 0);
1754 opc
= get_memop(oi
);
1755 s_bits
= opc
& MO_SIZE
;
1757 #ifdef CONFIG_SOFTMMU
1758 mem_index
= get_mmuidx(oi
);
1759 addrlo
= tcg_out_tlb_read(s
, opc
, addrlo
, addrhi
, mem_index
, true);
1761 /* Load a pointer into the current opcode w/conditional branch-link. */
1762 label_ptr
= s
->code_ptr
;
1763 tcg_out_bc_noaddr(s
, BC
| BI(7, CR_EQ
) | BO_COND_FALSE
| LK
);
1766 #else /* !CONFIG_SOFTMMU */
1767 rbase
= guest_base
? TCG_GUEST_BASE_REG
: 0;
1768 if (TCG_TARGET_REG_BITS
> TARGET_LONG_BITS
) {
1769 tcg_out_ext32u(s
, TCG_REG_TMP1
, addrlo
);
1770 addrlo
= TCG_REG_TMP1
;
1774 if (TCG_TARGET_REG_BITS
== 32 && s_bits
== MO_64
) {
1775 if (opc
& MO_BSWAP
) {
1776 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1777 tcg_out32(s
, LWBRX
| TAB(datalo
, rbase
, addrlo
));
1778 tcg_out32(s
, LWBRX
| TAB(datahi
, rbase
, TCG_REG_R0
));
1779 } else if (rbase
!= 0) {
1780 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1781 tcg_out32(s
, LWZX
| TAB(datahi
, rbase
, addrlo
));
1782 tcg_out32(s
, LWZX
| TAB(datalo
, rbase
, TCG_REG_R0
));
1783 } else if (addrlo
== datahi
) {
1784 tcg_out32(s
, LWZ
| TAI(datalo
, addrlo
, 4));
1785 tcg_out32(s
, LWZ
| TAI(datahi
, addrlo
, 0));
1787 tcg_out32(s
, LWZ
| TAI(datahi
, addrlo
, 0));
1788 tcg_out32(s
, LWZ
| TAI(datalo
, addrlo
, 4));
1791 uint32_t insn
= qemu_ldx_opc
[opc
& (MO_BSWAP
| MO_SSIZE
)];
1792 if (!HAVE_ISA_2_06
&& insn
== LDBRX
) {
1793 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1794 tcg_out32(s
, LWBRX
| TAB(datalo
, rbase
, addrlo
));
1795 tcg_out32(s
, LWBRX
| TAB(TCG_REG_R0
, rbase
, TCG_REG_R0
));
1796 tcg_out_rld(s
, RLDIMI
, datalo
, TCG_REG_R0
, 32, 0);
1798 tcg_out32(s
, insn
| TAB(datalo
, rbase
, addrlo
));
1800 insn
= qemu_ldx_opc
[opc
& (MO_SIZE
| MO_BSWAP
)];
1801 tcg_out32(s
, insn
| TAB(datalo
, rbase
, addrlo
));
1802 insn
= qemu_exts_opc
[s_bits
];
1803 tcg_out32(s
, insn
| RA(datalo
) | RS(datalo
));
1807 #ifdef CONFIG_SOFTMMU
1808 add_qemu_ldst_label(s
, true, oi
, datalo
, datahi
, addrlo
, addrhi
,
1809 s
->code_ptr
, label_ptr
);
1813 static void tcg_out_qemu_st(TCGContext
*s
, const TCGArg
*args
, bool is_64
)
1815 TCGReg datalo
, datahi
, addrlo
, rbase
;
1816 TCGReg addrhi
__attribute__((unused
));
1818 TCGMemOp opc
, s_bits
;
1819 #ifdef CONFIG_SOFTMMU
1821 tcg_insn_unit
*label_ptr
;
1825 datahi
= (TCG_TARGET_REG_BITS
== 32 && is_64
? *args
++ : 0);
1827 addrhi
= (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
? *args
++ : 0);
1829 opc
= get_memop(oi
);
1830 s_bits
= opc
& MO_SIZE
;
1832 #ifdef CONFIG_SOFTMMU
1833 mem_index
= get_mmuidx(oi
);
1834 addrlo
= tcg_out_tlb_read(s
, opc
, addrlo
, addrhi
, mem_index
, false);
1836 /* Load a pointer into the current opcode w/conditional branch-link. */
1837 label_ptr
= s
->code_ptr
;
1838 tcg_out_bc_noaddr(s
, BC
| BI(7, CR_EQ
) | BO_COND_FALSE
| LK
);
1841 #else /* !CONFIG_SOFTMMU */
1842 rbase
= guest_base
? TCG_GUEST_BASE_REG
: 0;
1843 if (TCG_TARGET_REG_BITS
> TARGET_LONG_BITS
) {
1844 tcg_out_ext32u(s
, TCG_REG_TMP1
, addrlo
);
1845 addrlo
= TCG_REG_TMP1
;
1849 if (TCG_TARGET_REG_BITS
== 32 && s_bits
== MO_64
) {
1850 if (opc
& MO_BSWAP
) {
1851 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1852 tcg_out32(s
, STWBRX
| SAB(datalo
, rbase
, addrlo
));
1853 tcg_out32(s
, STWBRX
| SAB(datahi
, rbase
, TCG_REG_R0
));
1854 } else if (rbase
!= 0) {
1855 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1856 tcg_out32(s
, STWX
| SAB(datahi
, rbase
, addrlo
));
1857 tcg_out32(s
, STWX
| SAB(datalo
, rbase
, TCG_REG_R0
));
1859 tcg_out32(s
, STW
| TAI(datahi
, addrlo
, 0));
1860 tcg_out32(s
, STW
| TAI(datalo
, addrlo
, 4));
1863 uint32_t insn
= qemu_stx_opc
[opc
& (MO_BSWAP
| MO_SIZE
)];
1864 if (!HAVE_ISA_2_06
&& insn
== STDBRX
) {
1865 tcg_out32(s
, STWBRX
| SAB(datalo
, rbase
, addrlo
));
1866 tcg_out32(s
, ADDI
| TAI(TCG_REG_TMP1
, addrlo
, 4));
1867 tcg_out_shri64(s
, TCG_REG_R0
, datalo
, 32);
1868 tcg_out32(s
, STWBRX
| SAB(TCG_REG_R0
, rbase
, TCG_REG_TMP1
));
1870 tcg_out32(s
, insn
| SAB(datalo
, rbase
, addrlo
));
1874 #ifdef CONFIG_SOFTMMU
1875 add_qemu_ldst_label(s
, false, oi
, datalo
, datahi
, addrlo
, addrhi
,
1876 s
->code_ptr
, label_ptr
);
1880 static void tcg_out_nop_fill(tcg_insn_unit
*p
, int count
)
1883 for (i
= 0; i
< count
; ++i
) {
1888 /* Parameters for function call generation, used in tcg.c. */
1889 #define TCG_TARGET_STACK_ALIGN 16
1890 #define TCG_TARGET_EXTEND_ARGS 1
1893 # define LINK_AREA_SIZE (6 * SZR)
1894 # define LR_OFFSET (1 * SZR)
1895 # define TCG_TARGET_CALL_STACK_OFFSET (LINK_AREA_SIZE + 8 * SZR)
1896 #elif defined(TCG_TARGET_CALL_DARWIN)
1897 # define LINK_AREA_SIZE (6 * SZR)
1898 # define LR_OFFSET (2 * SZR)
1899 #elif TCG_TARGET_REG_BITS == 64
1900 # if defined(_CALL_ELF) && _CALL_ELF == 2
1901 # define LINK_AREA_SIZE (4 * SZR)
1902 # define LR_OFFSET (1 * SZR)
1904 #else /* TCG_TARGET_REG_BITS == 32 */
1905 # if defined(_CALL_SYSV)
1906 # define LINK_AREA_SIZE (2 * SZR)
1907 # define LR_OFFSET (1 * SZR)
1911 # error "Unhandled abi"
1913 #ifndef TCG_TARGET_CALL_STACK_OFFSET
1914 # define TCG_TARGET_CALL_STACK_OFFSET LINK_AREA_SIZE
1917 #define CPU_TEMP_BUF_SIZE (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
1918 #define REG_SAVE_SIZE ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
1920 #define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET \
1921 + TCG_STATIC_CALL_ARGS_SIZE \
1922 + CPU_TEMP_BUF_SIZE \
1924 + TCG_TARGET_STACK_ALIGN - 1) \
1925 & -TCG_TARGET_STACK_ALIGN)
1927 #define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
1929 static void tcg_target_qemu_prologue(TCGContext
*s
)
1934 void **desc
= (void **)s
->code_ptr
;
1935 desc
[0] = desc
+ 2; /* entry point */
1936 desc
[1] = 0; /* environment pointer */
1937 s
->code_ptr
= (void *)(desc
+ 2); /* skip over descriptor */
1940 tcg_set_frame(s
, TCG_REG_CALL_STACK
, REG_SAVE_BOT
- CPU_TEMP_BUF_SIZE
,
1944 tcg_out32(s
, MFSPR
| RT(TCG_REG_R0
) | LR
);
1945 tcg_out32(s
, (SZR
== 8 ? STDU
: STWU
)
1946 | SAI(TCG_REG_R1
, TCG_REG_R1
, -FRAME_SIZE
));
1948 for (i
= 0; i
< ARRAY_SIZE(tcg_target_callee_save_regs
); ++i
) {
1949 tcg_out_st(s
, TCG_TYPE_REG
, tcg_target_callee_save_regs
[i
],
1950 TCG_REG_R1
, REG_SAVE_BOT
+ i
* SZR
);
1952 tcg_out_st(s
, TCG_TYPE_PTR
, TCG_REG_R0
, TCG_REG_R1
, FRAME_SIZE
+LR_OFFSET
);
1954 #ifndef CONFIG_SOFTMMU
1956 tcg_out_movi_int(s
, TCG_TYPE_PTR
, TCG_GUEST_BASE_REG
, guest_base
, true);
1957 tcg_regset_set_reg(s
->reserved_regs
, TCG_GUEST_BASE_REG
);
1961 tcg_out_mov(s
, TCG_TYPE_PTR
, TCG_AREG0
, tcg_target_call_iarg_regs
[0]);
1962 tcg_out32(s
, MTSPR
| RS(tcg_target_call_iarg_regs
[1]) | CTR
);
1964 tcg_out_mov(s
, TCG_TYPE_PTR
, TCG_REG_TB
, tcg_target_call_iarg_regs
[1]);
1966 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1969 s
->code_gen_epilogue
= tb_ret_addr
= s
->code_ptr
;
1971 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R0
, TCG_REG_R1
, FRAME_SIZE
+LR_OFFSET
);
1972 for (i
= 0; i
< ARRAY_SIZE(tcg_target_callee_save_regs
); ++i
) {
1973 tcg_out_ld(s
, TCG_TYPE_REG
, tcg_target_callee_save_regs
[i
],
1974 TCG_REG_R1
, REG_SAVE_BOT
+ i
* SZR
);
1976 tcg_out32(s
, MTSPR
| RS(TCG_REG_R0
) | LR
);
1977 tcg_out32(s
, ADDI
| TAI(TCG_REG_R1
, TCG_REG_R1
, FRAME_SIZE
));
1978 tcg_out32(s
, BCLR
| BO_ALWAYS
);
1981 static void tcg_out_op(TCGContext
*s
, TCGOpcode opc
, const TCGArg
*args
,
1982 const int *const_args
)
1988 case INDEX_op_exit_tb
:
1989 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_R3
, args
[0]);
1990 tcg_out_b(s
, 0, tb_ret_addr
);
1992 case INDEX_op_goto_tb
:
1993 if (s
->tb_jmp_insn_offset
) {
1995 if (TCG_TARGET_REG_BITS
== 64) {
1996 /* Ensure the next insns are 8-byte aligned. */
1997 if ((uintptr_t)s
->code_ptr
& 7) {
2000 s
->tb_jmp_insn_offset
[args
[0]] = tcg_current_code_size(s
);
2001 tcg_out32(s
, ADDIS
| TAI(TCG_REG_TB
, TCG_REG_TB
, 0));
2002 tcg_out32(s
, ADDI
| TAI(TCG_REG_TB
, TCG_REG_TB
, 0));
2004 s
->tb_jmp_insn_offset
[args
[0]] = tcg_current_code_size(s
);
2006 s
->tb_jmp_reset_offset
[args
[0]] = tcg_current_code_size(s
);
2010 /* Indirect jump. */
2011 tcg_debug_assert(s
->tb_jmp_insn_offset
== NULL
);
2012 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_TB
, 0,
2013 (intptr_t)(s
->tb_jmp_insn_offset
+ args
[0]));
2015 tcg_out32(s
, MTSPR
| RS(TCG_REG_TB
) | CTR
);
2016 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
2017 s
->tb_jmp_reset_offset
[args
[0]] = c
= tcg_current_code_size(s
);
2019 /* For the unlinked case, need to reset TCG_REG_TB. */
2021 assert(c
== (int16_t)c
);
2022 tcg_out32(s
, ADDI
| TAI(TCG_REG_TB
, TCG_REG_TB
, c
));
2025 case INDEX_op_goto_ptr
:
2026 tcg_out32(s
, MTSPR
| RS(args
[0]) | CTR
);
2028 tcg_out_mov(s
, TCG_TYPE_PTR
, TCG_REG_TB
, args
[0]);
2030 tcg_out32(s
, ADDI
| TAI(TCG_REG_R3
, 0, 0));
2031 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
2035 TCGLabel
*l
= arg_label(args
[0]);
2038 tcg_out_b(s
, 0, l
->u
.value_ptr
);
2040 tcg_out_reloc(s
, s
->code_ptr
, R_PPC_REL24
, l
, 0);
2041 tcg_out_b_noaddr(s
, B
);
2045 case INDEX_op_ld8u_i32
:
2046 case INDEX_op_ld8u_i64
:
2047 tcg_out_mem_long(s
, LBZ
, LBZX
, args
[0], args
[1], args
[2]);
2049 case INDEX_op_ld8s_i32
:
2050 case INDEX_op_ld8s_i64
:
2051 tcg_out_mem_long(s
, LBZ
, LBZX
, args
[0], args
[1], args
[2]);
2052 tcg_out32(s
, EXTSB
| RS(args
[0]) | RA(args
[0]));
2054 case INDEX_op_ld16u_i32
:
2055 case INDEX_op_ld16u_i64
:
2056 tcg_out_mem_long(s
, LHZ
, LHZX
, args
[0], args
[1], args
[2]);
2058 case INDEX_op_ld16s_i32
:
2059 case INDEX_op_ld16s_i64
:
2060 tcg_out_mem_long(s
, LHA
, LHAX
, args
[0], args
[1], args
[2]);
2062 case INDEX_op_ld_i32
:
2063 case INDEX_op_ld32u_i64
:
2064 tcg_out_mem_long(s
, LWZ
, LWZX
, args
[0], args
[1], args
[2]);
2066 case INDEX_op_ld32s_i64
:
2067 tcg_out_mem_long(s
, LWA
, LWAX
, args
[0], args
[1], args
[2]);
2069 case INDEX_op_ld_i64
:
2070 tcg_out_mem_long(s
, LD
, LDX
, args
[0], args
[1], args
[2]);
2072 case INDEX_op_st8_i32
:
2073 case INDEX_op_st8_i64
:
2074 tcg_out_mem_long(s
, STB
, STBX
, args
[0], args
[1], args
[2]);
2076 case INDEX_op_st16_i32
:
2077 case INDEX_op_st16_i64
:
2078 tcg_out_mem_long(s
, STH
, STHX
, args
[0], args
[1], args
[2]);
2080 case INDEX_op_st_i32
:
2081 case INDEX_op_st32_i64
:
2082 tcg_out_mem_long(s
, STW
, STWX
, args
[0], args
[1], args
[2]);
2084 case INDEX_op_st_i64
:
2085 tcg_out_mem_long(s
, STD
, STDX
, args
[0], args
[1], args
[2]);
2088 case INDEX_op_add_i32
:
2089 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2090 if (const_args
[2]) {
2092 tcg_out_mem_long(s
, ADDI
, ADD
, a0
, a1
, (int32_t)a2
);
2094 tcg_out32(s
, ADD
| TAB(a0
, a1
, a2
));
2097 case INDEX_op_sub_i32
:
2098 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2099 if (const_args
[1]) {
2100 if (const_args
[2]) {
2101 tcg_out_movi(s
, TCG_TYPE_I32
, a0
, a1
- a2
);
2103 tcg_out32(s
, SUBFIC
| TAI(a0
, a2
, a1
));
2105 } else if (const_args
[2]) {
2109 tcg_out32(s
, SUBF
| TAB(a0
, a2
, a1
));
2113 case INDEX_op_and_i32
:
2114 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2115 if (const_args
[2]) {
2116 tcg_out_andi32(s
, a0
, a1
, a2
);
2118 tcg_out32(s
, AND
| SAB(a1
, a0
, a2
));
2121 case INDEX_op_and_i64
:
2122 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2123 if (const_args
[2]) {
2124 tcg_out_andi64(s
, a0
, a1
, a2
);
2126 tcg_out32(s
, AND
| SAB(a1
, a0
, a2
));
2129 case INDEX_op_or_i64
:
2130 case INDEX_op_or_i32
:
2131 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2132 if (const_args
[2]) {
2133 tcg_out_ori32(s
, a0
, a1
, a2
);
2135 tcg_out32(s
, OR
| SAB(a1
, a0
, a2
));
2138 case INDEX_op_xor_i64
:
2139 case INDEX_op_xor_i32
:
2140 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2141 if (const_args
[2]) {
2142 tcg_out_xori32(s
, a0
, a1
, a2
);
2144 tcg_out32(s
, XOR
| SAB(a1
, a0
, a2
));
2147 case INDEX_op_andc_i32
:
2148 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2149 if (const_args
[2]) {
2150 tcg_out_andi32(s
, a0
, a1
, ~a2
);
2152 tcg_out32(s
, ANDC
| SAB(a1
, a0
, a2
));
2155 case INDEX_op_andc_i64
:
2156 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2157 if (const_args
[2]) {
2158 tcg_out_andi64(s
, a0
, a1
, ~a2
);
2160 tcg_out32(s
, ANDC
| SAB(a1
, a0
, a2
));
2163 case INDEX_op_orc_i32
:
2164 if (const_args
[2]) {
2165 tcg_out_ori32(s
, args
[0], args
[1], ~args
[2]);
2169 case INDEX_op_orc_i64
:
2170 tcg_out32(s
, ORC
| SAB(args
[1], args
[0], args
[2]));
2172 case INDEX_op_eqv_i32
:
2173 if (const_args
[2]) {
2174 tcg_out_xori32(s
, args
[0], args
[1], ~args
[2]);
2178 case INDEX_op_eqv_i64
:
2179 tcg_out32(s
, EQV
| SAB(args
[1], args
[0], args
[2]));
2181 case INDEX_op_nand_i32
:
2182 case INDEX_op_nand_i64
:
2183 tcg_out32(s
, NAND
| SAB(args
[1], args
[0], args
[2]));
2185 case INDEX_op_nor_i32
:
2186 case INDEX_op_nor_i64
:
2187 tcg_out32(s
, NOR
| SAB(args
[1], args
[0], args
[2]));
2190 case INDEX_op_clz_i32
:
2191 tcg_out_cntxz(s
, TCG_TYPE_I32
, CNTLZW
, args
[0], args
[1],
2192 args
[2], const_args
[2]);
2194 case INDEX_op_ctz_i32
:
2195 tcg_out_cntxz(s
, TCG_TYPE_I32
, CNTTZW
, args
[0], args
[1],
2196 args
[2], const_args
[2]);
2198 case INDEX_op_ctpop_i32
:
2199 tcg_out32(s
, CNTPOPW
| SAB(args
[1], args
[0], 0));
2202 case INDEX_op_clz_i64
:
2203 tcg_out_cntxz(s
, TCG_TYPE_I64
, CNTLZD
, args
[0], args
[1],
2204 args
[2], const_args
[2]);
2206 case INDEX_op_ctz_i64
:
2207 tcg_out_cntxz(s
, TCG_TYPE_I64
, CNTTZD
, args
[0], args
[1],
2208 args
[2], const_args
[2]);
2210 case INDEX_op_ctpop_i64
:
2211 tcg_out32(s
, CNTPOPD
| SAB(args
[1], args
[0], 0));
2214 case INDEX_op_mul_i32
:
2215 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2216 if (const_args
[2]) {
2217 tcg_out32(s
, MULLI
| TAI(a0
, a1
, a2
));
2219 tcg_out32(s
, MULLW
| TAB(a0
, a1
, a2
));
2223 case INDEX_op_div_i32
:
2224 tcg_out32(s
, DIVW
| TAB(args
[0], args
[1], args
[2]));
2227 case INDEX_op_divu_i32
:
2228 tcg_out32(s
, DIVWU
| TAB(args
[0], args
[1], args
[2]));
2231 case INDEX_op_shl_i32
:
2232 if (const_args
[2]) {
2233 tcg_out_shli32(s
, args
[0], args
[1], args
[2]);
2235 tcg_out32(s
, SLW
| SAB(args
[1], args
[0], args
[2]));
2238 case INDEX_op_shr_i32
:
2239 if (const_args
[2]) {
2240 tcg_out_shri32(s
, args
[0], args
[1], args
[2]);
2242 tcg_out32(s
, SRW
| SAB(args
[1], args
[0], args
[2]));
2245 case INDEX_op_sar_i32
:
2246 if (const_args
[2]) {
2247 tcg_out32(s
, SRAWI
| RS(args
[1]) | RA(args
[0]) | SH(args
[2]));
2249 tcg_out32(s
, SRAW
| SAB(args
[1], args
[0], args
[2]));
2252 case INDEX_op_rotl_i32
:
2253 if (const_args
[2]) {
2254 tcg_out_rlw(s
, RLWINM
, args
[0], args
[1], args
[2], 0, 31);
2256 tcg_out32(s
, RLWNM
| SAB(args
[1], args
[0], args
[2])
2260 case INDEX_op_rotr_i32
:
2261 if (const_args
[2]) {
2262 tcg_out_rlw(s
, RLWINM
, args
[0], args
[1], 32 - args
[2], 0, 31);
2264 tcg_out32(s
, SUBFIC
| TAI(TCG_REG_R0
, args
[2], 32));
2265 tcg_out32(s
, RLWNM
| SAB(args
[1], args
[0], TCG_REG_R0
)
2270 case INDEX_op_brcond_i32
:
2271 tcg_out_brcond(s
, args
[2], args
[0], args
[1], const_args
[1],
2272 arg_label(args
[3]), TCG_TYPE_I32
);
2274 case INDEX_op_brcond_i64
:
2275 tcg_out_brcond(s
, args
[2], args
[0], args
[1], const_args
[1],
2276 arg_label(args
[3]), TCG_TYPE_I64
);
2278 case INDEX_op_brcond2_i32
:
2279 tcg_out_brcond2(s
, args
, const_args
);
2282 case INDEX_op_neg_i32
:
2283 case INDEX_op_neg_i64
:
2284 tcg_out32(s
, NEG
| RT(args
[0]) | RA(args
[1]));
2287 case INDEX_op_not_i32
:
2288 case INDEX_op_not_i64
:
2289 tcg_out32(s
, NOR
| SAB(args
[1], args
[0], args
[1]));
2292 case INDEX_op_add_i64
:
2293 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2294 if (const_args
[2]) {
2296 tcg_out_mem_long(s
, ADDI
, ADD
, a0
, a1
, a2
);
2298 tcg_out32(s
, ADD
| TAB(a0
, a1
, a2
));
2301 case INDEX_op_sub_i64
:
2302 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2303 if (const_args
[1]) {
2304 if (const_args
[2]) {
2305 tcg_out_movi(s
, TCG_TYPE_I64
, a0
, a1
- a2
);
2307 tcg_out32(s
, SUBFIC
| TAI(a0
, a2
, a1
));
2309 } else if (const_args
[2]) {
2313 tcg_out32(s
, SUBF
| TAB(a0
, a2
, a1
));
2317 case INDEX_op_shl_i64
:
2318 if (const_args
[2]) {
2319 tcg_out_shli64(s
, args
[0], args
[1], args
[2]);
2321 tcg_out32(s
, SLD
| SAB(args
[1], args
[0], args
[2]));
2324 case INDEX_op_shr_i64
:
2325 if (const_args
[2]) {
2326 tcg_out_shri64(s
, args
[0], args
[1], args
[2]);
2328 tcg_out32(s
, SRD
| SAB(args
[1], args
[0], args
[2]));
2331 case INDEX_op_sar_i64
:
2332 if (const_args
[2]) {
2333 int sh
= SH(args
[2] & 0x1f) | (((args
[2] >> 5) & 1) << 1);
2334 tcg_out32(s
, SRADI
| RA(args
[0]) | RS(args
[1]) | sh
);
2336 tcg_out32(s
, SRAD
| SAB(args
[1], args
[0], args
[2]));
2339 case INDEX_op_rotl_i64
:
2340 if (const_args
[2]) {
2341 tcg_out_rld(s
, RLDICL
, args
[0], args
[1], args
[2], 0);
2343 tcg_out32(s
, RLDCL
| SAB(args
[1], args
[0], args
[2]) | MB64(0));
2346 case INDEX_op_rotr_i64
:
2347 if (const_args
[2]) {
2348 tcg_out_rld(s
, RLDICL
, args
[0], args
[1], 64 - args
[2], 0);
2350 tcg_out32(s
, SUBFIC
| TAI(TCG_REG_R0
, args
[2], 64));
2351 tcg_out32(s
, RLDCL
| SAB(args
[1], args
[0], TCG_REG_R0
) | MB64(0));
2355 case INDEX_op_mul_i64
:
2356 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2357 if (const_args
[2]) {
2358 tcg_out32(s
, MULLI
| TAI(a0
, a1
, a2
));
2360 tcg_out32(s
, MULLD
| TAB(a0
, a1
, a2
));
2363 case INDEX_op_div_i64
:
2364 tcg_out32(s
, DIVD
| TAB(args
[0], args
[1], args
[2]));
2366 case INDEX_op_divu_i64
:
2367 tcg_out32(s
, DIVDU
| TAB(args
[0], args
[1], args
[2]));
2370 case INDEX_op_qemu_ld_i32
:
2371 tcg_out_qemu_ld(s
, args
, false);
2373 case INDEX_op_qemu_ld_i64
:
2374 tcg_out_qemu_ld(s
, args
, true);
2376 case INDEX_op_qemu_st_i32
:
2377 tcg_out_qemu_st(s
, args
, false);
2379 case INDEX_op_qemu_st_i64
:
2380 tcg_out_qemu_st(s
, args
, true);
2383 case INDEX_op_ext8s_i32
:
2384 case INDEX_op_ext8s_i64
:
2387 case INDEX_op_ext16s_i32
:
2388 case INDEX_op_ext16s_i64
:
2391 case INDEX_op_ext_i32_i64
:
2392 case INDEX_op_ext32s_i64
:
2396 tcg_out32(s
, c
| RS(args
[1]) | RA(args
[0]));
2398 case INDEX_op_extu_i32_i64
:
2399 tcg_out_ext32u(s
, args
[0], args
[1]);
2402 case INDEX_op_setcond_i32
:
2403 tcg_out_setcond(s
, TCG_TYPE_I32
, args
[3], args
[0], args
[1], args
[2],
2406 case INDEX_op_setcond_i64
:
2407 tcg_out_setcond(s
, TCG_TYPE_I64
, args
[3], args
[0], args
[1], args
[2],
2410 case INDEX_op_setcond2_i32
:
2411 tcg_out_setcond2(s
, args
, const_args
);
2414 case INDEX_op_bswap16_i32
:
2415 case INDEX_op_bswap16_i64
:
2416 a0
= args
[0], a1
= args
[1];
2419 /* a0 = (a1 r<< 24) & 0xff # 000c */
2420 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 24, 24, 31);
2421 /* a0 = (a0 & ~0xff00) | (a1 r<< 8) & 0xff00 # 00dc */
2422 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 8, 16, 23);
2424 /* r0 = (a1 r<< 8) & 0xff00 # 00d0 */
2425 tcg_out_rlw(s
, RLWINM
, TCG_REG_R0
, a1
, 8, 16, 23);
2426 /* a0 = (a1 r<< 24) & 0xff # 000c */
2427 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 24, 24, 31);
2428 /* a0 = a0 | r0 # 00dc */
2429 tcg_out32(s
, OR
| SAB(TCG_REG_R0
, a0
, a0
));
2433 case INDEX_op_bswap32_i32
:
2434 case INDEX_op_bswap32_i64
:
2435 /* Stolen from gcc's builtin_bswap32 */
2437 a0
= args
[0] == a1
? TCG_REG_R0
: args
[0];
2439 /* a1 = args[1] # abcd */
2440 /* a0 = rotate_left (a1, 8) # bcda */
2441 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 8, 0, 31);
2442 /* a0 = (a0 & ~0xff000000) | ((a1 r<< 24) & 0xff000000) # dcda */
2443 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 0, 7);
2444 /* a0 = (a0 & ~0x0000ff00) | ((a1 r<< 24) & 0x0000ff00) # dcba */
2445 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 16, 23);
2447 if (a0
== TCG_REG_R0
) {
2448 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2452 case INDEX_op_bswap64_i64
:
2453 a0
= args
[0], a1
= args
[1], a2
= TCG_REG_R0
;
2459 /* a1 = # abcd efgh */
2460 /* a0 = rl32(a1, 8) # 0000 fghe */
2461 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 8, 0, 31);
2462 /* a0 = dep(a0, rl32(a1, 24), 0xff000000) # 0000 hghe */
2463 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 0, 7);
2464 /* a0 = dep(a0, rl32(a1, 24), 0x0000ff00) # 0000 hgfe */
2465 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 16, 23);
2467 /* a0 = rl64(a0, 32) # hgfe 0000 */
2468 /* a2 = rl64(a1, 32) # efgh abcd */
2469 tcg_out_rld(s
, RLDICL
, a0
, a0
, 32, 0);
2470 tcg_out_rld(s
, RLDICL
, a2
, a1
, 32, 0);
2472 /* a0 = dep(a0, rl32(a2, 8), 0xffffffff) # hgfe bcda */
2473 tcg_out_rlw(s
, RLWIMI
, a0
, a2
, 8, 0, 31);
2474 /* a0 = dep(a0, rl32(a2, 24), 0xff000000) # hgfe dcda */
2475 tcg_out_rlw(s
, RLWIMI
, a0
, a2
, 24, 0, 7);
2476 /* a0 = dep(a0, rl32(a2, 24), 0x0000ff00) # hgfe dcba */
2477 tcg_out_rlw(s
, RLWIMI
, a0
, a2
, 24, 16, 23);
2480 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2484 case INDEX_op_deposit_i32
:
2485 if (const_args
[2]) {
2486 uint32_t mask
= ((2u << (args
[4] - 1)) - 1) << args
[3];
2487 tcg_out_andi32(s
, args
[0], args
[0], ~mask
);
2489 tcg_out_rlw(s
, RLWIMI
, args
[0], args
[2], args
[3],
2490 32 - args
[3] - args
[4], 31 - args
[3]);
2493 case INDEX_op_deposit_i64
:
2494 if (const_args
[2]) {
2495 uint64_t mask
= ((2ull << (args
[4] - 1)) - 1) << args
[3];
2496 tcg_out_andi64(s
, args
[0], args
[0], ~mask
);
2498 tcg_out_rld(s
, RLDIMI
, args
[0], args
[2], args
[3],
2499 64 - args
[3] - args
[4]);
2503 case INDEX_op_extract_i32
:
2504 tcg_out_rlw(s
, RLWINM
, args
[0], args
[1],
2505 32 - args
[2], 32 - args
[3], 31);
2507 case INDEX_op_extract_i64
:
2508 tcg_out_rld(s
, RLDICL
, args
[0], args
[1], 64 - args
[2], 64 - args
[3]);
2511 case INDEX_op_movcond_i32
:
2512 tcg_out_movcond(s
, TCG_TYPE_I32
, args
[5], args
[0], args
[1], args
[2],
2513 args
[3], args
[4], const_args
[2]);
2515 case INDEX_op_movcond_i64
:
2516 tcg_out_movcond(s
, TCG_TYPE_I64
, args
[5], args
[0], args
[1], args
[2],
2517 args
[3], args
[4], const_args
[2]);
2520 #if TCG_TARGET_REG_BITS == 64
2521 case INDEX_op_add2_i64
:
2523 case INDEX_op_add2_i32
:
2525 /* Note that the CA bit is defined based on the word size of the
2526 environment. So in 64-bit mode it's always carry-out of bit 63.
2527 The fallback code using deposit works just as well for 32-bit. */
2528 a0
= args
[0], a1
= args
[1];
2529 if (a0
== args
[3] || (!const_args
[5] && a0
== args
[5])) {
2532 if (const_args
[4]) {
2533 tcg_out32(s
, ADDIC
| TAI(a0
, args
[2], args
[4]));
2535 tcg_out32(s
, ADDC
| TAB(a0
, args
[2], args
[4]));
2537 if (const_args
[5]) {
2538 tcg_out32(s
, (args
[5] ? ADDME
: ADDZE
) | RT(a1
) | RA(args
[3]));
2540 tcg_out32(s
, ADDE
| TAB(a1
, args
[3], args
[5]));
2542 if (a0
!= args
[0]) {
2543 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2547 #if TCG_TARGET_REG_BITS == 64
2548 case INDEX_op_sub2_i64
:
2550 case INDEX_op_sub2_i32
:
2552 a0
= args
[0], a1
= args
[1];
2553 if (a0
== args
[5] || (!const_args
[3] && a0
== args
[3])) {
2556 if (const_args
[2]) {
2557 tcg_out32(s
, SUBFIC
| TAI(a0
, args
[4], args
[2]));
2559 tcg_out32(s
, SUBFC
| TAB(a0
, args
[4], args
[2]));
2561 if (const_args
[3]) {
2562 tcg_out32(s
, (args
[3] ? SUBFME
: SUBFZE
) | RT(a1
) | RA(args
[5]));
2564 tcg_out32(s
, SUBFE
| TAB(a1
, args
[5], args
[3]));
2566 if (a0
!= args
[0]) {
2567 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2571 case INDEX_op_muluh_i32
:
2572 tcg_out32(s
, MULHWU
| TAB(args
[0], args
[1], args
[2]));
2574 case INDEX_op_mulsh_i32
:
2575 tcg_out32(s
, MULHW
| TAB(args
[0], args
[1], args
[2]));
2577 case INDEX_op_muluh_i64
:
2578 tcg_out32(s
, MULHDU
| TAB(args
[0], args
[1], args
[2]));
2580 case INDEX_op_mulsh_i64
:
2581 tcg_out32(s
, MULHD
| TAB(args
[0], args
[1], args
[2]));
2585 tcg_out_mb(s
, args
[0]);
2588 case INDEX_op_mov_i32
: /* Always emitted via tcg_out_mov. */
2589 case INDEX_op_mov_i64
:
2590 case INDEX_op_movi_i32
: /* Always emitted via tcg_out_movi. */
2591 case INDEX_op_movi_i64
:
2592 case INDEX_op_call
: /* Always emitted via tcg_out_call. */
2598 static const TCGTargetOpDef
*tcg_target_op_def(TCGOpcode op
)
2600 static const TCGTargetOpDef r
= { .args_ct_str
= { "r" } };
2601 static const TCGTargetOpDef r_r
= { .args_ct_str
= { "r", "r" } };
2602 static const TCGTargetOpDef r_L
= { .args_ct_str
= { "r", "L" } };
2603 static const TCGTargetOpDef S_S
= { .args_ct_str
= { "S", "S" } };
2604 static const TCGTargetOpDef r_ri
= { .args_ct_str
= { "r", "ri" } };
2605 static const TCGTargetOpDef r_r_r
= { .args_ct_str
= { "r", "r", "r" } };
2606 static const TCGTargetOpDef r_L_L
= { .args_ct_str
= { "r", "L", "L" } };
2607 static const TCGTargetOpDef L_L_L
= { .args_ct_str
= { "L", "L", "L" } };
2608 static const TCGTargetOpDef S_S_S
= { .args_ct_str
= { "S", "S", "S" } };
2609 static const TCGTargetOpDef r_r_ri
= { .args_ct_str
= { "r", "r", "ri" } };
2610 static const TCGTargetOpDef r_r_rI
= { .args_ct_str
= { "r", "r", "rI" } };
2611 static const TCGTargetOpDef r_r_rT
= { .args_ct_str
= { "r", "r", "rT" } };
2612 static const TCGTargetOpDef r_r_rU
= { .args_ct_str
= { "r", "r", "rU" } };
2613 static const TCGTargetOpDef r_rI_ri
2614 = { .args_ct_str
= { "r", "rI", "ri" } };
2615 static const TCGTargetOpDef r_rI_rT
2616 = { .args_ct_str
= { "r", "rI", "rT" } };
2617 static const TCGTargetOpDef r_r_rZW
2618 = { .args_ct_str
= { "r", "r", "rZW" } };
2619 static const TCGTargetOpDef L_L_L_L
2620 = { .args_ct_str
= { "L", "L", "L", "L" } };
2621 static const TCGTargetOpDef S_S_S_S
2622 = { .args_ct_str
= { "S", "S", "S", "S" } };
2623 static const TCGTargetOpDef movc
2624 = { .args_ct_str
= { "r", "r", "ri", "rZ", "rZ" } };
2625 static const TCGTargetOpDef dep
2626 = { .args_ct_str
= { "r", "0", "rZ" } };
2627 static const TCGTargetOpDef br2
2628 = { .args_ct_str
= { "r", "r", "ri", "ri" } };
2629 static const TCGTargetOpDef setc2
2630 = { .args_ct_str
= { "r", "r", "r", "ri", "ri" } };
2631 static const TCGTargetOpDef add2
2632 = { .args_ct_str
= { "r", "r", "r", "r", "rI", "rZM" } };
2633 static const TCGTargetOpDef sub2
2634 = { .args_ct_str
= { "r", "r", "rI", "rZM", "r", "r" } };
2637 case INDEX_op_goto_ptr
:
2640 case INDEX_op_ld8u_i32
:
2641 case INDEX_op_ld8s_i32
:
2642 case INDEX_op_ld16u_i32
:
2643 case INDEX_op_ld16s_i32
:
2644 case INDEX_op_ld_i32
:
2645 case INDEX_op_st8_i32
:
2646 case INDEX_op_st16_i32
:
2647 case INDEX_op_st_i32
:
2648 case INDEX_op_ctpop_i32
:
2649 case INDEX_op_neg_i32
:
2650 case INDEX_op_not_i32
:
2651 case INDEX_op_ext8s_i32
:
2652 case INDEX_op_ext16s_i32
:
2653 case INDEX_op_bswap16_i32
:
2654 case INDEX_op_bswap32_i32
:
2655 case INDEX_op_extract_i32
:
2656 case INDEX_op_ld8u_i64
:
2657 case INDEX_op_ld8s_i64
:
2658 case INDEX_op_ld16u_i64
:
2659 case INDEX_op_ld16s_i64
:
2660 case INDEX_op_ld32u_i64
:
2661 case INDEX_op_ld32s_i64
:
2662 case INDEX_op_ld_i64
:
2663 case INDEX_op_st8_i64
:
2664 case INDEX_op_st16_i64
:
2665 case INDEX_op_st32_i64
:
2666 case INDEX_op_st_i64
:
2667 case INDEX_op_ctpop_i64
:
2668 case INDEX_op_neg_i64
:
2669 case INDEX_op_not_i64
:
2670 case INDEX_op_ext8s_i64
:
2671 case INDEX_op_ext16s_i64
:
2672 case INDEX_op_ext32s_i64
:
2673 case INDEX_op_ext_i32_i64
:
2674 case INDEX_op_extu_i32_i64
:
2675 case INDEX_op_bswap16_i64
:
2676 case INDEX_op_bswap32_i64
:
2677 case INDEX_op_bswap64_i64
:
2678 case INDEX_op_extract_i64
:
2681 case INDEX_op_add_i32
:
2682 case INDEX_op_and_i32
:
2683 case INDEX_op_or_i32
:
2684 case INDEX_op_xor_i32
:
2685 case INDEX_op_andc_i32
:
2686 case INDEX_op_orc_i32
:
2687 case INDEX_op_eqv_i32
:
2688 case INDEX_op_shl_i32
:
2689 case INDEX_op_shr_i32
:
2690 case INDEX_op_sar_i32
:
2691 case INDEX_op_rotl_i32
:
2692 case INDEX_op_rotr_i32
:
2693 case INDEX_op_setcond_i32
:
2694 case INDEX_op_and_i64
:
2695 case INDEX_op_andc_i64
:
2696 case INDEX_op_shl_i64
:
2697 case INDEX_op_shr_i64
:
2698 case INDEX_op_sar_i64
:
2699 case INDEX_op_rotl_i64
:
2700 case INDEX_op_rotr_i64
:
2701 case INDEX_op_setcond_i64
:
2703 case INDEX_op_mul_i32
:
2704 case INDEX_op_mul_i64
:
2706 case INDEX_op_div_i32
:
2707 case INDEX_op_divu_i32
:
2708 case INDEX_op_nand_i32
:
2709 case INDEX_op_nor_i32
:
2710 case INDEX_op_muluh_i32
:
2711 case INDEX_op_mulsh_i32
:
2712 case INDEX_op_orc_i64
:
2713 case INDEX_op_eqv_i64
:
2714 case INDEX_op_nand_i64
:
2715 case INDEX_op_nor_i64
:
2716 case INDEX_op_div_i64
:
2717 case INDEX_op_divu_i64
:
2718 case INDEX_op_mulsh_i64
:
2719 case INDEX_op_muluh_i64
:
2721 case INDEX_op_sub_i32
:
2723 case INDEX_op_add_i64
:
2725 case INDEX_op_or_i64
:
2726 case INDEX_op_xor_i64
:
2728 case INDEX_op_sub_i64
:
2730 case INDEX_op_clz_i32
:
2731 case INDEX_op_ctz_i32
:
2732 case INDEX_op_clz_i64
:
2733 case INDEX_op_ctz_i64
:
2736 case INDEX_op_brcond_i32
:
2737 case INDEX_op_brcond_i64
:
2740 case INDEX_op_movcond_i32
:
2741 case INDEX_op_movcond_i64
:
2743 case INDEX_op_deposit_i32
:
2744 case INDEX_op_deposit_i64
:
2746 case INDEX_op_brcond2_i32
:
2748 case INDEX_op_setcond2_i32
:
2750 case INDEX_op_add2_i64
:
2751 case INDEX_op_add2_i32
:
2753 case INDEX_op_sub2_i64
:
2754 case INDEX_op_sub2_i32
:
2757 case INDEX_op_qemu_ld_i32
:
2758 return (TCG_TARGET_REG_BITS
== 64 || TARGET_LONG_BITS
== 32
2760 case INDEX_op_qemu_st_i32
:
2761 return (TCG_TARGET_REG_BITS
== 64 || TARGET_LONG_BITS
== 32
2763 case INDEX_op_qemu_ld_i64
:
2764 return (TCG_TARGET_REG_BITS
== 64 ? &r_L
2765 : TARGET_LONG_BITS
== 32 ? &L_L_L
: &L_L_L_L
);
2766 case INDEX_op_qemu_st_i64
:
2767 return (TCG_TARGET_REG_BITS
== 64 ? &S_S
2768 : TARGET_LONG_BITS
== 32 ? &S_S_S
: &S_S_S_S
);
2775 static void tcg_target_init(TCGContext
*s
)
2777 unsigned long hwcap
= qemu_getauxval(AT_HWCAP
);
2778 unsigned long hwcap2
= qemu_getauxval(AT_HWCAP2
);
2780 if (hwcap
& PPC_FEATURE_ARCH_2_06
) {
2781 have_isa_2_06
= true;
2783 #ifdef PPC_FEATURE2_ARCH_3_00
2784 if (hwcap2
& PPC_FEATURE2_ARCH_3_00
) {
2785 have_isa_3_00
= true;
2789 tcg_target_available_regs
[TCG_TYPE_I32
] = 0xffffffff;
2790 tcg_target_available_regs
[TCG_TYPE_I64
] = 0xffffffff;
2792 tcg_target_call_clobber_regs
= 0;
2793 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R0
);
2794 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R2
);
2795 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R3
);
2796 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R4
);
2797 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R5
);
2798 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R6
);
2799 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R7
);
2800 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R8
);
2801 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R9
);
2802 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R10
);
2803 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R11
);
2804 tcg_regset_set_reg(tcg_target_call_clobber_regs
, TCG_REG_R12
);
2806 s
->reserved_regs
= 0;
2807 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R0
); /* tcg temp */
2808 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R1
); /* stack pointer */
2809 #if defined(_CALL_SYSV)
2810 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R2
); /* toc pointer */
2812 #if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
2813 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R13
); /* thread pointer */
2815 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_TMP1
); /* mem temp */
2817 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_TB
); /* tb->tc_ptr */
2824 DebugFrameFDEHeader fde
;
2825 uint8_t fde_def_cfa
[4];
2826 uint8_t fde_reg_ofs
[ARRAY_SIZE(tcg_target_callee_save_regs
) * 2 + 3];
2829 /* We're expecting a 2 byte uleb128 encoded value. */
2830 QEMU_BUILD_BUG_ON(FRAME_SIZE
>= (1 << 14));
2832 #if TCG_TARGET_REG_BITS == 64
2833 # define ELF_HOST_MACHINE EM_PPC64
2835 # define ELF_HOST_MACHINE EM_PPC
2838 static DebugFrame debug_frame
= {
2839 .cie
.len
= sizeof(DebugFrameCIE
)-4, /* length after .len member */
2842 .cie
.code_align
= 1,
2843 .cie
.data_align
= (-SZR
& 0x7f), /* sleb128 -SZR */
2844 .cie
.return_column
= 65,
2846 /* Total FDE size does not include the "len" member. */
2847 .fde
.len
= sizeof(DebugFrame
) - offsetof(DebugFrame
, fde
.cie_offset
),
2850 12, TCG_REG_R1
, /* DW_CFA_def_cfa r1, ... */
2851 (FRAME_SIZE
& 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */
2855 /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
2856 0x11, 65, (LR_OFFSET
/ -SZR
) & 0x7f,
2860 void tcg_register_jit(void *buf
, size_t buf_size
)
2862 uint8_t *p
= &debug_frame
.fde_reg_ofs
[3];
2865 for (i
= 0; i
< ARRAY_SIZE(tcg_target_callee_save_regs
); ++i
, p
+= 2) {
2866 p
[0] = 0x80 + tcg_target_callee_save_regs
[i
];
2867 p
[1] = (FRAME_SIZE
- (REG_SAVE_BOT
+ i
* SZR
)) / SZR
;
2870 debug_frame
.fde
.func_start
= (uintptr_t)buf
;
2871 debug_frame
.fde
.func_len
= buf_size
;
2873 tcg_register_jit_int(buf
, buf_size
, &debug_frame
, sizeof(debug_frame
));
2875 #endif /* __ELF__ */
2877 void flush_icache_range(uintptr_t start
, uintptr_t stop
)
2879 uintptr_t p
, start1
, stop1
;
2880 size_t dsize
= qemu_dcache_linesize
;
2881 size_t isize
= qemu_icache_linesize
;
2883 start1
= start
& ~(dsize
- 1);
2884 stop1
= (stop
+ dsize
- 1) & ~(dsize
- 1);
2885 for (p
= start1
; p
< stop1
; p
+= dsize
) {
2886 asm volatile ("dcbst 0,%0" : : "r"(p
) : "memory");
2888 asm volatile ("sync" : : : "memory");
2890 start
&= start
& ~(isize
- 1);
2891 stop1
= (stop
+ isize
- 1) & ~(isize
- 1);
2892 for (p
= start1
; p
< stop1
; p
+= isize
) {
2893 asm volatile ("icbi 0,%0" : : "r"(p
) : "memory");
2895 asm volatile ("sync" : : : "memory");
2896 asm volatile ("isync" : : : "memory");