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
25 #include "tcg-be-ldst.h"
27 #if defined _CALL_DARWIN || defined __APPLE__
28 #define TCG_TARGET_CALL_DARWIN
31 # define TCG_TARGET_CALL_ALIGN_ARGS 1
34 /* For some memory operations, we need a scratch that isn't R0. For the AIX
35 calling convention, we can re-use the TOC register since we'll be reloading
36 it at every call. Otherwise R12 will do nicely as neither a call-saved
37 register nor a parameter register. */
39 # define TCG_REG_TMP1 TCG_REG_R2
41 # define TCG_REG_TMP1 TCG_REG_R12
44 /* For the 64-bit target, we don't like the 5 insn sequence needed to build
45 full 64-bit addresses. Better to have a base register to which we can
46 apply a 32-bit displacement.
48 There are generally three items of interest:
49 (1) helper functions in the main executable,
50 (2) TranslationBlock data structures,
51 (3) the return address in the epilogue.
53 For user-only, we USE_STATIC_CODE_GEN_BUFFER, so the code_gen_buffer
54 will be inside the main executable, and thus near enough to make a
55 pointer to the epilogue be within 2GB of all helper functions.
57 For softmmu, we'll let the kernel choose the address of code_gen_buffer,
58 and odds are it'll be somewhere close to the main malloc arena, and so
59 a pointer to the epilogue will be within 2GB of the TranslationBlocks.
61 For --enable-pie, everything will be kinda near everything else,
62 somewhere in high memory.
64 Thus we choose to keep the return address in a call-saved register. */
65 #define TCG_REG_RA TCG_REG_R31
66 #define USE_REG_RA (TCG_TARGET_REG_BITS == 64)
68 /* Shorthand for size of a pointer. Avoid promotion to unsigned. */
69 #define SZP ((int)sizeof(void *))
71 /* Shorthand for size of a register. */
72 #define SZR (TCG_TARGET_REG_BITS / 8)
74 #define TCG_CT_CONST_S16 0x100
75 #define TCG_CT_CONST_U16 0x200
76 #define TCG_CT_CONST_S32 0x400
77 #define TCG_CT_CONST_U32 0x800
78 #define TCG_CT_CONST_ZERO 0x1000
79 #define TCG_CT_CONST_MONE 0x2000
81 static tcg_insn_unit
*tb_ret_addr
;
84 static bool have_isa_2_06
;
85 #define HAVE_ISA_2_06 have_isa_2_06
86 #define HAVE_ISEL have_isa_2_06
88 #ifndef CONFIG_SOFTMMU
89 #define TCG_GUEST_BASE_REG 30
93 static const char * const tcg_target_reg_names
[TCG_TARGET_NB_REGS
] = {
129 static const int tcg_target_reg_alloc_order
[] = {
130 TCG_REG_R14
, /* call saved registers */
148 TCG_REG_R12
, /* call clobbered, non-arguments */
152 TCG_REG_R10
, /* call clobbered, arguments */
162 static const int tcg_target_call_iarg_regs
[] = {
173 static const int tcg_target_call_oarg_regs
[] = {
178 static const int tcg_target_callee_save_regs
[] = {
179 #ifdef TCG_TARGET_CALL_DARWIN
195 TCG_REG_R27
, /* currently used for the global env */
202 static inline bool in_range_b(tcg_target_long target
)
204 return target
== sextract64(target
, 0, 26);
207 static uint32_t reloc_pc24_val(tcg_insn_unit
*pc
, tcg_insn_unit
*target
)
209 ptrdiff_t disp
= tcg_ptr_byte_diff(target
, pc
);
210 assert(in_range_b(disp
));
211 return disp
& 0x3fffffc;
214 static void reloc_pc24(tcg_insn_unit
*pc
, tcg_insn_unit
*target
)
216 *pc
= (*pc
& ~0x3fffffc) | reloc_pc24_val(pc
, target
);
219 static uint16_t reloc_pc14_val(tcg_insn_unit
*pc
, tcg_insn_unit
*target
)
221 ptrdiff_t disp
= tcg_ptr_byte_diff(target
, pc
);
222 assert(disp
== (int16_t) disp
);
223 return disp
& 0xfffc;
226 static void reloc_pc14(tcg_insn_unit
*pc
, tcg_insn_unit
*target
)
228 *pc
= (*pc
& ~0xfffc) | reloc_pc14_val(pc
, target
);
231 static inline void tcg_out_b_noaddr(TCGContext
*s
, int insn
)
233 unsigned retrans
= *s
->code_ptr
& 0x3fffffc;
234 tcg_out32(s
, insn
| retrans
);
237 static inline void tcg_out_bc_noaddr(TCGContext
*s
, int insn
)
239 unsigned retrans
= *s
->code_ptr
& 0xfffc;
240 tcg_out32(s
, insn
| retrans
);
243 static void patch_reloc(tcg_insn_unit
*code_ptr
, int type
,
244 intptr_t value
, intptr_t addend
)
246 tcg_insn_unit
*target
= (tcg_insn_unit
*)value
;
251 reloc_pc14(code_ptr
, target
);
254 reloc_pc24(code_ptr
, target
);
261 /* parse target specific constraints */
262 static int target_parse_constraint(TCGArgConstraint
*ct
, const char **pct_str
)
268 case 'A': case 'B': case 'C': case 'D':
269 ct
->ct
|= TCG_CT_REG
;
270 tcg_regset_set_reg(ct
->u
.regs
, 3 + ct_str
[0] - 'A');
273 ct
->ct
|= TCG_CT_REG
;
274 tcg_regset_set32(ct
->u
.regs
, 0, 0xffffffff);
276 case 'L': /* qemu_ld constraint */
277 ct
->ct
|= TCG_CT_REG
;
278 tcg_regset_set32(ct
->u
.regs
, 0, 0xffffffff);
279 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R3
);
280 #ifdef CONFIG_SOFTMMU
281 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R4
);
282 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R5
);
285 case 'S': /* qemu_st constraint */
286 ct
->ct
|= TCG_CT_REG
;
287 tcg_regset_set32(ct
->u
.regs
, 0, 0xffffffff);
288 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R3
);
289 #ifdef CONFIG_SOFTMMU
290 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R4
);
291 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R5
);
292 tcg_regset_reset_reg(ct
->u
.regs
, TCG_REG_R6
);
296 ct
->ct
|= TCG_CT_CONST_S16
;
299 ct
->ct
|= TCG_CT_CONST_U16
;
302 ct
->ct
|= TCG_CT_CONST_MONE
;
305 ct
->ct
|= TCG_CT_CONST_S32
;
308 ct
->ct
|= TCG_CT_CONST_U32
;
311 ct
->ct
|= TCG_CT_CONST_ZERO
;
321 /* test if a constant matches the constraint */
322 static int tcg_target_const_match(tcg_target_long val
, TCGType type
,
323 const TCGArgConstraint
*arg_ct
)
326 if (ct
& TCG_CT_CONST
) {
330 /* The only 32-bit constraint we use aside from
331 TCG_CT_CONST is TCG_CT_CONST_S16. */
332 if (type
== TCG_TYPE_I32
) {
336 if ((ct
& TCG_CT_CONST_S16
) && val
== (int16_t)val
) {
338 } else if ((ct
& TCG_CT_CONST_U16
) && val
== (uint16_t)val
) {
340 } else if ((ct
& TCG_CT_CONST_S32
) && val
== (int32_t)val
) {
342 } else if ((ct
& TCG_CT_CONST_U32
) && val
== (uint32_t)val
) {
344 } else if ((ct
& TCG_CT_CONST_ZERO
) && val
== 0) {
346 } else if ((ct
& TCG_CT_CONST_MONE
) && val
== -1) {
352 #define OPCD(opc) ((opc)<<26)
353 #define XO19(opc) (OPCD(19)|((opc)<<1))
354 #define MD30(opc) (OPCD(30)|((opc)<<2))
355 #define MDS30(opc) (OPCD(30)|((opc)<<1))
356 #define XO31(opc) (OPCD(31)|((opc)<<1))
357 #define XO58(opc) (OPCD(58)|(opc))
358 #define XO62(opc) (OPCD(62)|(opc))
362 #define LBZ OPCD( 34)
363 #define LHZ OPCD( 40)
364 #define LHA OPCD( 42)
365 #define LWZ OPCD( 32)
366 #define STB OPCD( 38)
367 #define STH OPCD( 44)
368 #define STW OPCD( 36)
371 #define STDU XO62( 1)
372 #define STDX XO31(149)
375 #define LDX XO31( 21)
378 #define LWAX XO31(341)
380 #define ADDIC OPCD( 12)
381 #define ADDI OPCD( 14)
382 #define ADDIS OPCD( 15)
383 #define ORI OPCD( 24)
384 #define ORIS OPCD( 25)
385 #define XORI OPCD( 26)
386 #define XORIS OPCD( 27)
387 #define ANDI OPCD( 28)
388 #define ANDIS OPCD( 29)
389 #define MULLI OPCD( 7)
390 #define CMPLI OPCD( 10)
391 #define CMPI OPCD( 11)
392 #define SUBFIC OPCD( 8)
394 #define LWZU OPCD( 33)
395 #define STWU OPCD( 37)
397 #define RLWIMI OPCD( 20)
398 #define RLWINM OPCD( 21)
399 #define RLWNM OPCD( 23)
401 #define RLDICL MD30( 0)
402 #define RLDICR MD30( 1)
403 #define RLDIMI MD30( 3)
404 #define RLDCL MDS30( 8)
406 #define BCLR XO19( 16)
407 #define BCCTR XO19(528)
408 #define CRAND XO19(257)
409 #define CRANDC XO19(129)
410 #define CRNAND XO19(225)
411 #define CROR XO19(449)
412 #define CRNOR XO19( 33)
414 #define EXTSB XO31(954)
415 #define EXTSH XO31(922)
416 #define EXTSW XO31(986)
417 #define ADD XO31(266)
418 #define ADDE XO31(138)
419 #define ADDME XO31(234)
420 #define ADDZE XO31(202)
421 #define ADDC XO31( 10)
422 #define AND XO31( 28)
423 #define SUBF XO31( 40)
424 #define SUBFC XO31( 8)
425 #define SUBFE XO31(136)
426 #define SUBFME XO31(232)
427 #define SUBFZE XO31(200)
429 #define XOR XO31(316)
430 #define MULLW XO31(235)
431 #define MULHW XO31( 75)
432 #define MULHWU XO31( 11)
433 #define DIVW XO31(491)
434 #define DIVWU XO31(459)
436 #define CMPL XO31( 32)
437 #define LHBRX XO31(790)
438 #define LWBRX XO31(534)
439 #define LDBRX XO31(532)
440 #define STHBRX XO31(918)
441 #define STWBRX XO31(662)
442 #define STDBRX XO31(660)
443 #define MFSPR XO31(339)
444 #define MTSPR XO31(467)
445 #define SRAWI XO31(824)
446 #define NEG XO31(104)
447 #define MFCR XO31( 19)
448 #define MFOCRF (MFCR | (1u << 20))
449 #define NOR XO31(124)
450 #define CNTLZW XO31( 26)
451 #define CNTLZD XO31( 58)
452 #define ANDC XO31( 60)
453 #define ORC XO31(412)
454 #define EQV XO31(284)
455 #define NAND XO31(476)
456 #define ISEL XO31( 15)
458 #define MULLD XO31(233)
459 #define MULHD XO31( 73)
460 #define MULHDU XO31( 9)
461 #define DIVD XO31(489)
462 #define DIVDU XO31(457)
464 #define LBZX XO31( 87)
465 #define LHZX XO31(279)
466 #define LHAX XO31(343)
467 #define LWZX XO31( 23)
468 #define STBX XO31(215)
469 #define STHX XO31(407)
470 #define STWX XO31(151)
472 #define SPR(a, b) ((((a)<<5)|(b))<<11)
474 #define CTR SPR(9, 0)
476 #define SLW XO31( 24)
477 #define SRW XO31(536)
478 #define SRAW XO31(792)
480 #define SLD XO31( 27)
481 #define SRD XO31(539)
482 #define SRAD XO31(794)
483 #define SRADI XO31(413<<1)
486 #define TRAP (TW | TO(31))
488 #define NOP ORI /* ori 0,0,0 */
490 #define RT(r) ((r)<<21)
491 #define RS(r) ((r)<<21)
492 #define RA(r) ((r)<<16)
493 #define RB(r) ((r)<<11)
494 #define TO(t) ((t)<<21)
495 #define SH(s) ((s)<<11)
496 #define MB(b) ((b)<<6)
497 #define ME(e) ((e)<<1)
498 #define BO(o) ((o)<<21)
499 #define MB64(b) ((b)<<5)
500 #define FXM(b) (1 << (19 - (b)))
504 #define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
505 #define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
506 #define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
507 #define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
509 #define BF(n) ((n)<<23)
510 #define BI(n, c) (((c)+((n)*4))<<16)
511 #define BT(n, c) (((c)+((n)*4))<<21)
512 #define BA(n, c) (((c)+((n)*4))<<16)
513 #define BB(n, c) (((c)+((n)*4))<<11)
514 #define BC_(n, c) (((c)+((n)*4))<<6)
516 #define BO_COND_TRUE BO(12)
517 #define BO_COND_FALSE BO( 4)
518 #define BO_ALWAYS BO(20)
527 static const uint32_t tcg_to_bc
[] = {
528 [TCG_COND_EQ
] = BC
| BI(7, CR_EQ
) | BO_COND_TRUE
,
529 [TCG_COND_NE
] = BC
| BI(7, CR_EQ
) | BO_COND_FALSE
,
530 [TCG_COND_LT
] = BC
| BI(7, CR_LT
) | BO_COND_TRUE
,
531 [TCG_COND_GE
] = BC
| BI(7, CR_LT
) | BO_COND_FALSE
,
532 [TCG_COND_LE
] = BC
| BI(7, CR_GT
) | BO_COND_FALSE
,
533 [TCG_COND_GT
] = BC
| BI(7, CR_GT
) | BO_COND_TRUE
,
534 [TCG_COND_LTU
] = BC
| BI(7, CR_LT
) | BO_COND_TRUE
,
535 [TCG_COND_GEU
] = BC
| BI(7, CR_LT
) | BO_COND_FALSE
,
536 [TCG_COND_LEU
] = BC
| BI(7, CR_GT
) | BO_COND_FALSE
,
537 [TCG_COND_GTU
] = BC
| BI(7, CR_GT
) | BO_COND_TRUE
,
540 /* The low bit here is set if the RA and RB fields must be inverted. */
541 static const uint32_t tcg_to_isel
[] = {
542 [TCG_COND_EQ
] = ISEL
| BC_(7, CR_EQ
),
543 [TCG_COND_NE
] = ISEL
| BC_(7, CR_EQ
) | 1,
544 [TCG_COND_LT
] = ISEL
| BC_(7, CR_LT
),
545 [TCG_COND_GE
] = ISEL
| BC_(7, CR_LT
) | 1,
546 [TCG_COND_LE
] = ISEL
| BC_(7, CR_GT
) | 1,
547 [TCG_COND_GT
] = ISEL
| BC_(7, CR_GT
),
548 [TCG_COND_LTU
] = ISEL
| BC_(7, CR_LT
),
549 [TCG_COND_GEU
] = ISEL
| BC_(7, CR_LT
) | 1,
550 [TCG_COND_LEU
] = ISEL
| BC_(7, CR_GT
) | 1,
551 [TCG_COND_GTU
] = ISEL
| BC_(7, CR_GT
),
554 static void tcg_out_mem_long(TCGContext
*s
, int opi
, int opx
, TCGReg rt
,
555 TCGReg base
, tcg_target_long offset
);
557 static void tcg_out_mov(TCGContext
*s
, TCGType type
, TCGReg ret
, TCGReg arg
)
559 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
561 tcg_out32(s
, OR
| SAB(arg
, ret
, arg
));
565 static inline void tcg_out_rld(TCGContext
*s
, int op
, TCGReg ra
, TCGReg rs
,
568 assert(TCG_TARGET_REG_BITS
== 64);
569 sh
= SH(sh
& 0x1f) | (((sh
>> 5) & 1) << 1);
570 mb
= MB64((mb
>> 5) | ((mb
<< 1) & 0x3f));
571 tcg_out32(s
, op
| RA(ra
) | RS(rs
) | sh
| mb
);
574 static inline void tcg_out_rlw(TCGContext
*s
, int op
, TCGReg ra
, TCGReg rs
,
575 int sh
, int mb
, int me
)
577 tcg_out32(s
, op
| RA(ra
) | RS(rs
) | SH(sh
) | MB(mb
) | ME(me
));
580 static inline void tcg_out_ext32u(TCGContext
*s
, TCGReg dst
, TCGReg src
)
582 tcg_out_rld(s
, RLDICL
, dst
, src
, 0, 32);
585 static inline void tcg_out_shli32(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
587 tcg_out_rlw(s
, RLWINM
, dst
, src
, c
, 0, 31 - c
);
590 static inline void tcg_out_shli64(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
592 tcg_out_rld(s
, RLDICR
, dst
, src
, c
, 63 - c
);
595 static inline void tcg_out_shri32(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
597 tcg_out_rlw(s
, RLWINM
, dst
, src
, 32 - c
, c
, 31);
600 static inline void tcg_out_shri64(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
602 tcg_out_rld(s
, RLDICL
, dst
, src
, 64 - c
, c
);
605 static void tcg_out_movi32(TCGContext
*s
, TCGReg ret
, int32_t arg
)
607 if (arg
== (int16_t) arg
) {
608 tcg_out32(s
, ADDI
| TAI(ret
, 0, arg
));
610 tcg_out32(s
, ADDIS
| TAI(ret
, 0, arg
>> 16));
612 tcg_out32(s
, ORI
| SAI(ret
, ret
, arg
));
617 static void tcg_out_movi(TCGContext
*s
, TCGType type
, TCGReg ret
,
620 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
621 if (type
== TCG_TYPE_I32
|| arg
== (int32_t)arg
) {
622 tcg_out_movi32(s
, ret
, arg
);
623 } else if (arg
== (uint32_t)arg
&& !(arg
& 0x8000)) {
624 tcg_out32(s
, ADDI
| TAI(ret
, 0, arg
));
625 tcg_out32(s
, ORIS
| SAI(ret
, ret
, arg
>> 16));
630 intptr_t diff
= arg
- (intptr_t)tb_ret_addr
;
631 if (diff
== (int32_t)diff
) {
632 tcg_out_mem_long(s
, ADDI
, ADD
, ret
, TCG_REG_RA
, diff
);
637 high
= arg
>> 31 >> 1;
638 tcg_out_movi32(s
, ret
, high
);
640 tcg_out_shli64(s
, ret
, ret
, 32);
642 if (arg
& 0xffff0000) {
643 tcg_out32(s
, ORIS
| SAI(ret
, ret
, arg
>> 16));
646 tcg_out32(s
, ORI
| SAI(ret
, ret
, arg
));
651 static bool mask_operand(uint32_t c
, int *mb
, int *me
)
655 /* Accept a bit pattern like:
659 Keep track of the transitions. */
660 if (c
== 0 || c
== -1) {
666 if (test
& (test
- 1)) {
671 *mb
= test
? clz32(test
& -test
) + 1 : 0;
675 static bool mask64_operand(uint64_t c
, int *mb
, int *me
)
684 /* Accept 1..10..0. */
690 /* Accept 0..01..1. */
691 if (lsb
== 1 && (c
& (c
+ 1)) == 0) {
692 *mb
= clz64(c
+ 1) + 1;
699 static void tcg_out_andi32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
)
703 if (mask_operand(c
, &mb
, &me
)) {
704 tcg_out_rlw(s
, RLWINM
, dst
, src
, 0, mb
, me
);
705 } else if ((c
& 0xffff) == c
) {
706 tcg_out32(s
, ANDI
| SAI(src
, dst
, c
));
708 } else if ((c
& 0xffff0000) == c
) {
709 tcg_out32(s
, ANDIS
| SAI(src
, dst
, c
>> 16));
712 tcg_out_movi(s
, TCG_TYPE_I32
, TCG_REG_R0
, c
);
713 tcg_out32(s
, AND
| SAB(src
, dst
, TCG_REG_R0
));
717 static void tcg_out_andi64(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint64_t c
)
721 assert(TCG_TARGET_REG_BITS
== 64);
722 if (mask64_operand(c
, &mb
, &me
)) {
724 tcg_out_rld(s
, RLDICR
, dst
, src
, 0, me
);
726 tcg_out_rld(s
, RLDICL
, dst
, src
, 0, mb
);
728 } else if ((c
& 0xffff) == c
) {
729 tcg_out32(s
, ANDI
| SAI(src
, dst
, c
));
731 } else if ((c
& 0xffff0000) == c
) {
732 tcg_out32(s
, ANDIS
| SAI(src
, dst
, c
>> 16));
735 tcg_out_movi(s
, TCG_TYPE_I64
, TCG_REG_R0
, c
);
736 tcg_out32(s
, AND
| SAB(src
, dst
, TCG_REG_R0
));
740 static void tcg_out_zori32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
,
741 int op_lo
, int op_hi
)
744 tcg_out32(s
, op_hi
| SAI(src
, dst
, c
>> 16));
748 tcg_out32(s
, op_lo
| SAI(src
, dst
, c
));
753 static void tcg_out_ori32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
)
755 tcg_out_zori32(s
, dst
, src
, c
, ORI
, ORIS
);
758 static void tcg_out_xori32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
)
760 tcg_out_zori32(s
, dst
, src
, c
, XORI
, XORIS
);
763 static void tcg_out_b(TCGContext
*s
, int mask
, tcg_insn_unit
*target
)
765 ptrdiff_t disp
= tcg_pcrel_diff(s
, target
);
766 if (in_range_b(disp
)) {
767 tcg_out32(s
, B
| (disp
& 0x3fffffc) | mask
);
769 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_R0
, (uintptr_t)target
);
770 tcg_out32(s
, MTSPR
| RS(TCG_REG_R0
) | CTR
);
771 tcg_out32(s
, BCCTR
| BO_ALWAYS
| mask
);
775 static void tcg_out_mem_long(TCGContext
*s
, int opi
, int opx
, TCGReg rt
,
776 TCGReg base
, tcg_target_long offset
)
778 tcg_target_long orig
= offset
, l0
, l1
, extra
= 0, align
= 0;
779 bool is_store
= false;
780 TCGReg rs
= TCG_REG_TMP1
;
787 if (rt
!= TCG_REG_R0
) {
795 case STB
: case STH
: case STW
:
800 /* For unaligned, or very large offsets, use the indexed form. */
801 if (offset
& align
|| offset
!= (int32_t)offset
) {
805 tcg_debug_assert(!is_store
|| rs
!= rt
);
806 tcg_out_movi(s
, TCG_TYPE_PTR
, rs
, orig
);
807 tcg_out32(s
, opx
| TAB(rt
, base
, rs
));
811 l0
= (int16_t)offset
;
812 offset
= (offset
- l0
) >> 16;
813 l1
= (int16_t)offset
;
815 if (l1
< 0 && orig
>= 0) {
817 l1
= (int16_t)(offset
- 0x4000);
820 tcg_out32(s
, ADDIS
| TAI(rs
, base
, l1
));
824 tcg_out32(s
, ADDIS
| TAI(rs
, base
, extra
));
827 if (opi
!= ADDI
|| base
!= rt
|| l0
!= 0) {
828 tcg_out32(s
, opi
| TAI(rt
, base
, l0
));
832 static inline void tcg_out_ld(TCGContext
*s
, TCGType type
, TCGReg ret
,
833 TCGReg arg1
, intptr_t arg2
)
837 assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
838 if (type
== TCG_TYPE_I32
) {
839 opi
= LWZ
, opx
= LWZX
;
843 tcg_out_mem_long(s
, opi
, opx
, ret
, arg1
, arg2
);
846 static inline void tcg_out_st(TCGContext
*s
, TCGType type
, TCGReg arg
,
847 TCGReg arg1
, intptr_t arg2
)
851 assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
852 if (type
== TCG_TYPE_I32
) {
853 opi
= STW
, opx
= STWX
;
855 opi
= STD
, opx
= STDX
;
857 tcg_out_mem_long(s
, opi
, opx
, arg
, arg1
, arg2
);
860 static void tcg_out_cmp(TCGContext
*s
, int cond
, TCGArg arg1
, TCGArg arg2
,
861 int const_arg2
, int cr
, TCGType type
)
866 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
868 /* Simplify the comparisons below wrt CMPI. */
869 if (type
== TCG_TYPE_I32
) {
870 arg2
= (int32_t)arg2
;
877 if ((int16_t) arg2
== arg2
) {
881 } else if ((uint16_t) arg2
== arg2
) {
896 if ((int16_t) arg2
== arg2
) {
911 if ((uint16_t) arg2
== arg2
) {
924 op
|= BF(cr
) | ((type
== TCG_TYPE_I64
) << 21);
927 tcg_out32(s
, op
| RA(arg1
) | (arg2
& 0xffff));
930 tcg_out_movi(s
, type
, TCG_REG_R0
, arg2
);
933 tcg_out32(s
, op
| RA(arg1
) | RB(arg2
));
937 static void tcg_out_setcond_eq0(TCGContext
*s
, TCGType type
,
938 TCGReg dst
, TCGReg src
)
940 if (type
== TCG_TYPE_I32
) {
941 tcg_out32(s
, CNTLZW
| RS(src
) | RA(dst
));
942 tcg_out_shri32(s
, dst
, dst
, 5);
944 tcg_out32(s
, CNTLZD
| RS(src
) | RA(dst
));
945 tcg_out_shri64(s
, dst
, dst
, 6);
949 static void tcg_out_setcond_ne0(TCGContext
*s
, TCGReg dst
, TCGReg src
)
951 /* X != 0 implies X + -1 generates a carry. Extra addition
952 trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C. */
954 tcg_out32(s
, ADDIC
| TAI(dst
, src
, -1));
955 tcg_out32(s
, SUBFE
| TAB(dst
, dst
, src
));
957 tcg_out32(s
, ADDIC
| TAI(TCG_REG_R0
, src
, -1));
958 tcg_out32(s
, SUBFE
| TAB(dst
, TCG_REG_R0
, src
));
962 static TCGReg
tcg_gen_setcond_xor(TCGContext
*s
, TCGReg arg1
, TCGArg arg2
,
966 if ((uint32_t)arg2
== arg2
) {
967 tcg_out_xori32(s
, TCG_REG_R0
, arg1
, arg2
);
969 tcg_out_movi(s
, TCG_TYPE_I64
, TCG_REG_R0
, arg2
);
970 tcg_out32(s
, XOR
| SAB(arg1
, TCG_REG_R0
, TCG_REG_R0
));
973 tcg_out32(s
, XOR
| SAB(arg1
, TCG_REG_R0
, arg2
));
978 static void tcg_out_setcond(TCGContext
*s
, TCGType type
, TCGCond cond
,
979 TCGArg arg0
, TCGArg arg1
, TCGArg arg2
,
984 assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
986 /* Ignore high bits of a potential constant arg2. */
987 if (type
== TCG_TYPE_I32
) {
988 arg2
= (uint32_t)arg2
;
991 /* Handle common and trivial cases before handling anything else. */
995 tcg_out_setcond_eq0(s
, type
, arg0
, arg1
);
998 if (TCG_TARGET_REG_BITS
== 64 && type
== TCG_TYPE_I32
) {
999 tcg_out_ext32u(s
, TCG_REG_R0
, arg1
);
1002 tcg_out_setcond_ne0(s
, arg0
, arg1
);
1005 tcg_out32(s
, NOR
| SAB(arg1
, arg0
, arg1
));
1009 /* Extract the sign bit. */
1010 if (type
== TCG_TYPE_I32
) {
1011 tcg_out_shri32(s
, arg0
, arg1
, 31);
1013 tcg_out_shri64(s
, arg0
, arg1
, 63);
1021 /* If we have ISEL, we can implement everything with 3 or 4 insns.
1022 All other cases below are also at least 3 insns, so speed up the
1023 code generator by not considering them and always using ISEL. */
1027 tcg_out_cmp(s
, cond
, arg1
, arg2
, const_arg2
, 7, type
);
1029 isel
= tcg_to_isel
[cond
];
1031 tcg_out_movi(s
, type
, arg0
, 1);
1033 /* arg0 = (bc ? 0 : 1) */
1034 tab
= TAB(arg0
, 0, arg0
);
1037 /* arg0 = (bc ? 1 : 0) */
1038 tcg_out_movi(s
, type
, TCG_REG_R0
, 0);
1039 tab
= TAB(arg0
, arg0
, TCG_REG_R0
);
1041 tcg_out32(s
, isel
| tab
);
1047 arg1
= tcg_gen_setcond_xor(s
, arg1
, arg2
, const_arg2
);
1048 tcg_out_setcond_eq0(s
, type
, arg0
, arg1
);
1052 arg1
= tcg_gen_setcond_xor(s
, arg1
, arg2
, const_arg2
);
1053 /* Discard the high bits only once, rather than both inputs. */
1054 if (TCG_TARGET_REG_BITS
== 64 && type
== TCG_TYPE_I32
) {
1055 tcg_out_ext32u(s
, TCG_REG_R0
, arg1
);
1058 tcg_out_setcond_ne0(s
, arg0
, arg1
);
1076 crop
= CRNOR
| BT(7, CR_EQ
) | BA(7, CR_LT
) | BB(7, CR_LT
);
1082 crop
= CRNOR
| BT(7, CR_EQ
) | BA(7, CR_GT
) | BB(7, CR_GT
);
1084 tcg_out_cmp(s
, cond
, arg1
, arg2
, const_arg2
, 7, type
);
1088 tcg_out32(s
, MFOCRF
| RT(TCG_REG_R0
) | FXM(7));
1089 tcg_out_rlw(s
, RLWINM
, arg0
, TCG_REG_R0
, sh
, 31, 31);
1097 static void tcg_out_bc(TCGContext
*s
, int bc
, TCGLabel
*l
)
1100 tcg_out32(s
, bc
| reloc_pc14_val(s
->code_ptr
, l
->u
.value_ptr
));
1102 tcg_out_reloc(s
, s
->code_ptr
, R_PPC_REL14
, l
, 0);
1103 tcg_out_bc_noaddr(s
, bc
);
1107 static void tcg_out_brcond(TCGContext
*s
, TCGCond cond
,
1108 TCGArg arg1
, TCGArg arg2
, int const_arg2
,
1109 TCGLabel
*l
, TCGType type
)
1111 tcg_out_cmp(s
, cond
, arg1
, arg2
, const_arg2
, 7, type
);
1112 tcg_out_bc(s
, tcg_to_bc
[cond
], l
);
1115 static void tcg_out_movcond(TCGContext
*s
, TCGType type
, TCGCond cond
,
1116 TCGArg dest
, TCGArg c1
, TCGArg c2
, TCGArg v1
,
1117 TCGArg v2
, bool const_c2
)
1119 /* If for some reason both inputs are zero, don't produce bad code. */
1120 if (v1
== 0 && v2
== 0) {
1121 tcg_out_movi(s
, type
, dest
, 0);
1125 tcg_out_cmp(s
, cond
, c1
, c2
, const_c2
, 7, type
);
1128 int isel
= tcg_to_isel
[cond
];
1130 /* Swap the V operands if the operation indicates inversion. */
1137 /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand. */
1139 tcg_out_movi(s
, type
, TCG_REG_R0
, 0);
1141 tcg_out32(s
, isel
| TAB(dest
, v1
, v2
));
1144 cond
= tcg_invert_cond(cond
);
1146 } else if (dest
!= v1
) {
1148 tcg_out_movi(s
, type
, dest
, 0);
1150 tcg_out_mov(s
, type
, dest
, v1
);
1153 /* Branch forward over one insn */
1154 tcg_out32(s
, tcg_to_bc
[cond
] | 8);
1156 tcg_out_movi(s
, type
, dest
, 0);
1158 tcg_out_mov(s
, type
, dest
, v2
);
1163 static void tcg_out_cmp2(TCGContext
*s
, const TCGArg
*args
,
1164 const int *const_args
)
1166 static const struct { uint8_t bit1
, bit2
; } bits
[] = {
1167 [TCG_COND_LT
] = { CR_LT
, CR_LT
},
1168 [TCG_COND_LE
] = { CR_LT
, CR_GT
},
1169 [TCG_COND_GT
] = { CR_GT
, CR_GT
},
1170 [TCG_COND_GE
] = { CR_GT
, CR_LT
},
1171 [TCG_COND_LTU
] = { CR_LT
, CR_LT
},
1172 [TCG_COND_LEU
] = { CR_LT
, CR_GT
},
1173 [TCG_COND_GTU
] = { CR_GT
, CR_GT
},
1174 [TCG_COND_GEU
] = { CR_GT
, CR_LT
},
1177 TCGCond cond
= args
[4], cond2
;
1178 TCGArg al
, ah
, bl
, bh
;
1179 int blconst
, bhconst
;
1186 blconst
= const_args
[2];
1187 bhconst
= const_args
[3];
1196 tcg_out_cmp(s
, cond
, al
, bl
, blconst
, 6, TCG_TYPE_I32
);
1197 tcg_out_cmp(s
, cond
, ah
, bh
, bhconst
, 7, TCG_TYPE_I32
);
1198 tcg_out32(s
, op
| BT(7, CR_EQ
) | BA(6, CR_EQ
) | BB(7, CR_EQ
));
1209 bit1
= bits
[cond
].bit1
;
1210 bit2
= bits
[cond
].bit2
;
1211 op
= (bit1
!= bit2
? CRANDC
: CRAND
);
1212 cond2
= tcg_unsigned_cond(cond
);
1214 tcg_out_cmp(s
, cond
, ah
, bh
, bhconst
, 6, TCG_TYPE_I32
);
1215 tcg_out_cmp(s
, cond2
, al
, bl
, blconst
, 7, TCG_TYPE_I32
);
1216 tcg_out32(s
, op
| BT(7, CR_EQ
) | BA(6, CR_EQ
) | BB(7, bit2
));
1217 tcg_out32(s
, CROR
| BT(7, CR_EQ
) | BA(6, bit1
) | BB(7, CR_EQ
));
1225 static void tcg_out_setcond2(TCGContext
*s
, const TCGArg
*args
,
1226 const int *const_args
)
1228 tcg_out_cmp2(s
, args
+ 1, const_args
+ 1);
1229 tcg_out32(s
, MFOCRF
| RT(TCG_REG_R0
) | FXM(7));
1230 tcg_out_rlw(s
, RLWINM
, args
[0], TCG_REG_R0
, 31, 31, 31);
1233 static void tcg_out_brcond2 (TCGContext
*s
, const TCGArg
*args
,
1234 const int *const_args
)
1236 tcg_out_cmp2(s
, args
, const_args
);
1237 tcg_out_bc(s
, BC
| BI(7, CR_EQ
) | BO_COND_TRUE
, arg_label(args
[5]));
1240 void ppc_tb_set_jmp_target(uintptr_t jmp_addr
, uintptr_t addr
)
1242 tcg_insn_unit i1
, i2
;
1244 intptr_t diff
= addr
- jmp_addr
;
1246 if (in_range_b(diff
)) {
1247 i1
= B
| (diff
& 0x3fffffc);
1249 } else if (USE_REG_RA
) {
1251 diff
= addr
- (uintptr_t)tb_ret_addr
;
1253 hi
= (int32_t)(diff
- lo
);
1254 assert(diff
== hi
+ lo
);
1255 i1
= ADDIS
| TAI(TCG_REG_TMP1
, TCG_REG_RA
, hi
>> 16);
1256 i2
= ADDI
| TAI(TCG_REG_TMP1
, TCG_REG_TMP1
, lo
);
1258 assert(TCG_TARGET_REG_BITS
== 32 || addr
== (int32_t)addr
);
1259 i1
= ADDIS
| TAI(TCG_REG_TMP1
, 0, addr
>> 16);
1260 i2
= ORI
| SAI(TCG_REG_TMP1
, TCG_REG_TMP1
, addr
);
1262 #ifdef HOST_WORDS_BIGENDIAN
1263 pair
= (uint64_t)i1
<< 32 | i2
;
1265 pair
= (uint64_t)i2
<< 32 | i1
;
1268 /* ??? __atomic_store_8, presuming there's some way to do that
1269 for 32-bit, otherwise this is good enough for 64-bit. */
1270 *(uint64_t *)jmp_addr
= pair
;
1271 flush_icache_range(jmp_addr
, jmp_addr
+ 8);
1274 static void tcg_out_call(TCGContext
*s
, tcg_insn_unit
*target
)
1277 /* Look through the descriptor. If the branch is in range, and we
1278 don't have to spend too much effort on building the toc. */
1279 void *tgt
= ((void **)target
)[0];
1280 uintptr_t toc
= ((uintptr_t *)target
)[1];
1281 intptr_t diff
= tcg_pcrel_diff(s
, tgt
);
1283 if (in_range_b(diff
) && toc
== (uint32_t)toc
) {
1284 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_TMP1
, toc
);
1285 tcg_out_b(s
, LK
, tgt
);
1287 /* Fold the low bits of the constant into the addresses below. */
1288 intptr_t arg
= (intptr_t)target
;
1289 int ofs
= (int16_t)arg
;
1291 if (ofs
+ 8 < 0x8000) {
1296 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_TMP1
, arg
);
1297 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R0
, TCG_REG_TMP1
, ofs
);
1298 tcg_out32(s
, MTSPR
| RA(TCG_REG_R0
) | CTR
);
1299 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R2
, TCG_REG_TMP1
, ofs
+ SZP
);
1300 tcg_out32(s
, BCCTR
| BO_ALWAYS
| LK
);
1302 #elif defined(_CALL_ELF) && _CALL_ELF == 2
1305 /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1306 address, which the callee uses to compute its TOC address. */
1307 /* FIXME: when the branch is in range, we could avoid r12 load if we
1308 knew that the destination uses the same TOC, and what its local
1309 entry point offset is. */
1310 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_R12
, (intptr_t)target
);
1312 diff
= tcg_pcrel_diff(s
, target
);
1313 if (in_range_b(diff
)) {
1314 tcg_out_b(s
, LK
, target
);
1316 tcg_out32(s
, MTSPR
| RS(TCG_REG_R12
) | CTR
);
1317 tcg_out32(s
, BCCTR
| BO_ALWAYS
| LK
);
1320 tcg_out_b(s
, LK
, target
);
1324 static const uint32_t qemu_ldx_opc
[16] = {
1331 [MO_BSWAP
| MO_UB
] = LBZX
,
1332 [MO_BSWAP
| MO_UW
] = LHBRX
,
1333 [MO_BSWAP
| MO_UL
] = LWBRX
,
1334 [MO_BSWAP
| MO_Q
] = LDBRX
,
1337 static const uint32_t qemu_stx_opc
[16] = {
1342 [MO_BSWAP
| MO_UB
] = STBX
,
1343 [MO_BSWAP
| MO_UW
] = STHBRX
,
1344 [MO_BSWAP
| MO_UL
] = STWBRX
,
1345 [MO_BSWAP
| MO_Q
] = STDBRX
,
1348 static const uint32_t qemu_exts_opc
[4] = {
1349 EXTSB
, EXTSH
, EXTSW
, 0
1352 #if defined (CONFIG_SOFTMMU)
1353 /* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
1354 * int mmu_idx, uintptr_t ra)
1356 static void * const qemu_ld_helpers
[16] = {
1357 [MO_UB
] = helper_ret_ldub_mmu
,
1358 [MO_LEUW
] = helper_le_lduw_mmu
,
1359 [MO_LEUL
] = helper_le_ldul_mmu
,
1360 [MO_LEQ
] = helper_le_ldq_mmu
,
1361 [MO_BEUW
] = helper_be_lduw_mmu
,
1362 [MO_BEUL
] = helper_be_ldul_mmu
,
1363 [MO_BEQ
] = helper_be_ldq_mmu
,
1366 /* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
1367 * uintxx_t val, int mmu_idx, uintptr_t ra)
1369 static void * const qemu_st_helpers
[16] = {
1370 [MO_UB
] = helper_ret_stb_mmu
,
1371 [MO_LEUW
] = helper_le_stw_mmu
,
1372 [MO_LEUL
] = helper_le_stl_mmu
,
1373 [MO_LEQ
] = helper_le_stq_mmu
,
1374 [MO_BEUW
] = helper_be_stw_mmu
,
1375 [MO_BEUL
] = helper_be_stl_mmu
,
1376 [MO_BEQ
] = helper_be_stq_mmu
,
1379 /* Perform the TLB load and compare. Places the result of the comparison
1380 in CR7, loads the addend of the TLB into R3, and returns the register
1381 containing the guest address (zero-extended into R4). Clobbers R0 and R2. */
1383 static TCGReg
tcg_out_tlb_read(TCGContext
*s
, TCGMemOp opc
,
1384 TCGReg addrlo
, TCGReg addrhi
,
1385 int mem_index
, bool is_read
)
1389 ? offsetof(CPUArchState
, tlb_table
[mem_index
][0].addr_read
)
1390 : offsetof(CPUArchState
, tlb_table
[mem_index
][0].addr_write
));
1391 int add_off
= offsetof(CPUArchState
, tlb_table
[mem_index
][0].addend
);
1392 TCGReg base
= TCG_AREG0
;
1393 TCGMemOp s_bits
= opc
& MO_SIZE
;
1395 /* Extract the page index, shifted into place for tlb index. */
1396 if (TCG_TARGET_REG_BITS
== 64) {
1397 if (TARGET_LONG_BITS
== 32) {
1398 /* Zero-extend the address into a place helpful for further use. */
1399 tcg_out_ext32u(s
, TCG_REG_R4
, addrlo
);
1400 addrlo
= TCG_REG_R4
;
1402 tcg_out_rld(s
, RLDICL
, TCG_REG_R3
, addrlo
,
1403 64 - TARGET_PAGE_BITS
, 64 - CPU_TLB_BITS
);
1407 /* Compensate for very large offsets. */
1408 if (add_off
>= 0x8000) {
1409 /* Most target env are smaller than 32k; none are larger than 64k.
1410 Simplify the logic here merely to offset by 0x7ff0, giving us a
1411 range just shy of 64k. Check this assumption. */
1412 QEMU_BUILD_BUG_ON(offsetof(CPUArchState
,
1413 tlb_table
[NB_MMU_MODES
- 1][1])
1415 tcg_out32(s
, ADDI
| TAI(TCG_REG_TMP1
, base
, 0x7ff0));
1416 base
= TCG_REG_TMP1
;
1421 /* Extraction and shifting, part 2. */
1422 if (TCG_TARGET_REG_BITS
== 32 || TARGET_LONG_BITS
== 32) {
1423 tcg_out_rlw(s
, RLWINM
, TCG_REG_R3
, addrlo
,
1424 32 - (TARGET_PAGE_BITS
- CPU_TLB_ENTRY_BITS
),
1425 32 - (CPU_TLB_BITS
+ CPU_TLB_ENTRY_BITS
),
1426 31 - CPU_TLB_ENTRY_BITS
);
1428 tcg_out_shli64(s
, TCG_REG_R3
, TCG_REG_R3
, CPU_TLB_ENTRY_BITS
);
1431 tcg_out32(s
, ADD
| TAB(TCG_REG_R3
, TCG_REG_R3
, base
));
1433 /* Load the tlb comparator. */
1434 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1435 tcg_out_ld(s
, TCG_TYPE_I32
, TCG_REG_R4
, TCG_REG_R3
, cmp_off
);
1436 tcg_out_ld(s
, TCG_TYPE_I32
, TCG_REG_TMP1
, TCG_REG_R3
, cmp_off
+ 4);
1438 tcg_out_ld(s
, TCG_TYPE_TL
, TCG_REG_TMP1
, TCG_REG_R3
, cmp_off
);
1441 /* Load the TLB addend for use on the fast path. Do this asap
1442 to minimize any load use delay. */
1443 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R3
, TCG_REG_R3
, add_off
);
1445 /* Clear the non-page, non-alignment bits from the address */
1446 if (TCG_TARGET_REG_BITS
== 32 || TARGET_LONG_BITS
== 32) {
1447 /* We don't support unaligned accesses on 32-bits, preserve
1448 * the bottom bits and thus trigger a comparison failure on
1449 * unaligned accesses
1451 tcg_out_rlw(s
, RLWINM
, TCG_REG_R0
, addrlo
, 0,
1452 (32 - s_bits
) & 31, 31 - TARGET_PAGE_BITS
);
1453 } else if (s_bits
) {
1454 /* > byte access, we need to handle alignment */
1455 if ((opc
& MO_AMASK
) == MO_ALIGN
) {
1456 /* Alignment required by the front-end, same as 32-bits */
1457 tcg_out_rld(s
, RLDICL
, TCG_REG_R0
, addrlo
,
1458 64 - TARGET_PAGE_BITS
, TARGET_PAGE_BITS
- s_bits
);
1459 tcg_out_rld(s
, RLDICL
, TCG_REG_R0
, TCG_REG_R0
, TARGET_PAGE_BITS
, 0);
1461 /* We support unaligned accesses, we need to make sure we fail
1462 * if we cross a page boundary. The trick is to add the
1463 * access_size-1 to the address before masking the low bits.
1464 * That will make the address overflow to the next page if we
1465 * cross a page boundary which will then force a mismatch of
1466 * the TLB compare since the next page cannot possibly be in
1467 * the same TLB index.
1469 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, (1 << s_bits
) - 1));
1470 tcg_out_rld(s
, RLDICR
, TCG_REG_R0
, TCG_REG_R0
,
1471 0, 63 - TARGET_PAGE_BITS
);
1474 /* Byte access, just chop off the bits below the page index */
1475 tcg_out_rld(s
, RLDICR
, TCG_REG_R0
, addrlo
, 0, 63 - TARGET_PAGE_BITS
);
1478 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1479 tcg_out_cmp(s
, TCG_COND_EQ
, TCG_REG_R0
, TCG_REG_TMP1
,
1480 0, 7, TCG_TYPE_I32
);
1481 tcg_out_cmp(s
, TCG_COND_EQ
, addrhi
, TCG_REG_R4
, 0, 6, TCG_TYPE_I32
);
1482 tcg_out32(s
, CRAND
| BT(7, CR_EQ
) | BA(6, CR_EQ
) | BB(7, CR_EQ
));
1484 tcg_out_cmp(s
, TCG_COND_EQ
, TCG_REG_R0
, TCG_REG_TMP1
,
1491 /* Record the context of a call to the out of line helper code for the slow
1492 path for a load or store, so that we can later generate the correct
1494 static void add_qemu_ldst_label(TCGContext
*s
, bool is_ld
, TCGMemOpIdx oi
,
1495 TCGReg datalo_reg
, TCGReg datahi_reg
,
1496 TCGReg addrlo_reg
, TCGReg addrhi_reg
,
1497 tcg_insn_unit
*raddr
, tcg_insn_unit
*lptr
)
1499 TCGLabelQemuLdst
*label
= new_ldst_label(s
);
1501 label
->is_ld
= is_ld
;
1503 label
->datalo_reg
= datalo_reg
;
1504 label
->datahi_reg
= datahi_reg
;
1505 label
->addrlo_reg
= addrlo_reg
;
1506 label
->addrhi_reg
= addrhi_reg
;
1507 label
->raddr
= raddr
;
1508 label
->label_ptr
[0] = lptr
;
1511 static void tcg_out_qemu_ld_slow_path(TCGContext
*s
, TCGLabelQemuLdst
*lb
)
1513 TCGMemOpIdx oi
= lb
->oi
;
1514 TCGMemOp opc
= get_memop(oi
);
1515 TCGReg hi
, lo
, arg
= TCG_REG_R3
;
1517 reloc_pc14(lb
->label_ptr
[0], s
->code_ptr
);
1519 tcg_out_mov(s
, TCG_TYPE_PTR
, arg
++, TCG_AREG0
);
1521 lo
= lb
->addrlo_reg
;
1522 hi
= lb
->addrhi_reg
;
1523 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1524 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1527 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, hi
);
1528 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, lo
);
1530 /* If the address needed to be zero-extended, we'll have already
1531 placed it in R4. The only remaining case is 64-bit guest. */
1532 tcg_out_mov(s
, TCG_TYPE_TL
, arg
++, lo
);
1535 tcg_out_movi(s
, TCG_TYPE_I32
, arg
++, oi
);
1536 tcg_out32(s
, MFSPR
| RT(arg
) | LR
);
1538 tcg_out_call(s
, qemu_ld_helpers
[opc
& (MO_BSWAP
| MO_SIZE
)]);
1540 lo
= lb
->datalo_reg
;
1541 hi
= lb
->datahi_reg
;
1542 if (TCG_TARGET_REG_BITS
== 32 && (opc
& MO_SIZE
) == MO_64
) {
1543 tcg_out_mov(s
, TCG_TYPE_I32
, lo
, TCG_REG_R4
);
1544 tcg_out_mov(s
, TCG_TYPE_I32
, hi
, TCG_REG_R3
);
1545 } else if (opc
& MO_SIGN
) {
1546 uint32_t insn
= qemu_exts_opc
[opc
& MO_SIZE
];
1547 tcg_out32(s
, insn
| RA(lo
) | RS(TCG_REG_R3
));
1549 tcg_out_mov(s
, TCG_TYPE_REG
, lo
, TCG_REG_R3
);
1552 tcg_out_b(s
, 0, lb
->raddr
);
1555 static void tcg_out_qemu_st_slow_path(TCGContext
*s
, TCGLabelQemuLdst
*lb
)
1557 TCGMemOpIdx oi
= lb
->oi
;
1558 TCGMemOp opc
= get_memop(oi
);
1559 TCGMemOp s_bits
= opc
& MO_SIZE
;
1560 TCGReg hi
, lo
, arg
= TCG_REG_R3
;
1562 reloc_pc14(lb
->label_ptr
[0], s
->code_ptr
);
1564 tcg_out_mov(s
, TCG_TYPE_PTR
, arg
++, TCG_AREG0
);
1566 lo
= lb
->addrlo_reg
;
1567 hi
= lb
->addrhi_reg
;
1568 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1569 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1572 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, hi
);
1573 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, lo
);
1575 /* If the address needed to be zero-extended, we'll have already
1576 placed it in R4. The only remaining case is 64-bit guest. */
1577 tcg_out_mov(s
, TCG_TYPE_TL
, arg
++, lo
);
1580 lo
= lb
->datalo_reg
;
1581 hi
= lb
->datahi_reg
;
1582 if (TCG_TARGET_REG_BITS
== 32) {
1585 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1588 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, hi
);
1591 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, lo
);
1594 tcg_out_rlw(s
, RLWINM
, arg
++, lo
, 0, 32 - (8 << s_bits
), 31);
1598 if (s_bits
== MO_64
) {
1599 tcg_out_mov(s
, TCG_TYPE_I64
, arg
++, lo
);
1601 tcg_out_rld(s
, RLDICL
, arg
++, lo
, 0, 64 - (8 << s_bits
));
1605 tcg_out_movi(s
, TCG_TYPE_I32
, arg
++, oi
);
1606 tcg_out32(s
, MFSPR
| RT(arg
) | LR
);
1608 tcg_out_call(s
, qemu_st_helpers
[opc
& (MO_BSWAP
| MO_SIZE
)]);
1610 tcg_out_b(s
, 0, lb
->raddr
);
1612 #endif /* SOFTMMU */
1614 static void tcg_out_qemu_ld(TCGContext
*s
, const TCGArg
*args
, bool is_64
)
1616 TCGReg datalo
, datahi
, addrlo
, rbase
;
1617 TCGReg addrhi
__attribute__((unused
));
1619 TCGMemOp opc
, s_bits
;
1620 #ifdef CONFIG_SOFTMMU
1622 tcg_insn_unit
*label_ptr
;
1626 datahi
= (TCG_TARGET_REG_BITS
== 32 && is_64
? *args
++ : 0);
1628 addrhi
= (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
? *args
++ : 0);
1630 opc
= get_memop(oi
);
1631 s_bits
= opc
& MO_SIZE
;
1633 #ifdef CONFIG_SOFTMMU
1634 mem_index
= get_mmuidx(oi
);
1635 addrlo
= tcg_out_tlb_read(s
, opc
, addrlo
, addrhi
, mem_index
, true);
1637 /* Load a pointer into the current opcode w/conditional branch-link. */
1638 label_ptr
= s
->code_ptr
;
1639 tcg_out_bc_noaddr(s
, BC
| BI(7, CR_EQ
) | BO_COND_FALSE
| LK
);
1642 #else /* !CONFIG_SOFTMMU */
1643 rbase
= guest_base
? TCG_GUEST_BASE_REG
: 0;
1644 if (TCG_TARGET_REG_BITS
> TARGET_LONG_BITS
) {
1645 tcg_out_ext32u(s
, TCG_REG_TMP1
, addrlo
);
1646 addrlo
= TCG_REG_TMP1
;
1650 if (TCG_TARGET_REG_BITS
== 32 && s_bits
== MO_64
) {
1651 if (opc
& MO_BSWAP
) {
1652 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1653 tcg_out32(s
, LWBRX
| TAB(datalo
, rbase
, addrlo
));
1654 tcg_out32(s
, LWBRX
| TAB(datahi
, rbase
, TCG_REG_R0
));
1655 } else if (rbase
!= 0) {
1656 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1657 tcg_out32(s
, LWZX
| TAB(datahi
, rbase
, addrlo
));
1658 tcg_out32(s
, LWZX
| TAB(datalo
, rbase
, TCG_REG_R0
));
1659 } else if (addrlo
== datahi
) {
1660 tcg_out32(s
, LWZ
| TAI(datalo
, addrlo
, 4));
1661 tcg_out32(s
, LWZ
| TAI(datahi
, addrlo
, 0));
1663 tcg_out32(s
, LWZ
| TAI(datahi
, addrlo
, 0));
1664 tcg_out32(s
, LWZ
| TAI(datalo
, addrlo
, 4));
1667 uint32_t insn
= qemu_ldx_opc
[opc
& (MO_BSWAP
| MO_SSIZE
)];
1668 if (!HAVE_ISA_2_06
&& insn
== LDBRX
) {
1669 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1670 tcg_out32(s
, LWBRX
| TAB(datalo
, rbase
, addrlo
));
1671 tcg_out32(s
, LWBRX
| TAB(TCG_REG_R0
, rbase
, TCG_REG_R0
));
1672 tcg_out_rld(s
, RLDIMI
, datalo
, TCG_REG_R0
, 32, 0);
1674 tcg_out32(s
, insn
| TAB(datalo
, rbase
, addrlo
));
1676 insn
= qemu_ldx_opc
[opc
& (MO_SIZE
| MO_BSWAP
)];
1677 tcg_out32(s
, insn
| TAB(datalo
, rbase
, addrlo
));
1678 insn
= qemu_exts_opc
[s_bits
];
1679 tcg_out32(s
, insn
| RA(datalo
) | RS(datalo
));
1683 #ifdef CONFIG_SOFTMMU
1684 add_qemu_ldst_label(s
, true, oi
, datalo
, datahi
, addrlo
, addrhi
,
1685 s
->code_ptr
, label_ptr
);
1689 static void tcg_out_qemu_st(TCGContext
*s
, const TCGArg
*args
, bool is_64
)
1691 TCGReg datalo
, datahi
, addrlo
, rbase
;
1692 TCGReg addrhi
__attribute__((unused
));
1694 TCGMemOp opc
, s_bits
;
1695 #ifdef CONFIG_SOFTMMU
1697 tcg_insn_unit
*label_ptr
;
1701 datahi
= (TCG_TARGET_REG_BITS
== 32 && is_64
? *args
++ : 0);
1703 addrhi
= (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
? *args
++ : 0);
1705 opc
= get_memop(oi
);
1706 s_bits
= opc
& MO_SIZE
;
1708 #ifdef CONFIG_SOFTMMU
1709 mem_index
= get_mmuidx(oi
);
1710 addrlo
= tcg_out_tlb_read(s
, opc
, addrlo
, addrhi
, mem_index
, false);
1712 /* Load a pointer into the current opcode w/conditional branch-link. */
1713 label_ptr
= s
->code_ptr
;
1714 tcg_out_bc_noaddr(s
, BC
| BI(7, CR_EQ
) | BO_COND_FALSE
| LK
);
1717 #else /* !CONFIG_SOFTMMU */
1718 rbase
= guest_base
? TCG_GUEST_BASE_REG
: 0;
1719 if (TCG_TARGET_REG_BITS
> TARGET_LONG_BITS
) {
1720 tcg_out_ext32u(s
, TCG_REG_TMP1
, addrlo
);
1721 addrlo
= TCG_REG_TMP1
;
1725 if (TCG_TARGET_REG_BITS
== 32 && s_bits
== MO_64
) {
1726 if (opc
& MO_BSWAP
) {
1727 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1728 tcg_out32(s
, STWBRX
| SAB(datalo
, rbase
, addrlo
));
1729 tcg_out32(s
, STWBRX
| SAB(datahi
, rbase
, TCG_REG_R0
));
1730 } else if (rbase
!= 0) {
1731 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1732 tcg_out32(s
, STWX
| SAB(datahi
, rbase
, addrlo
));
1733 tcg_out32(s
, STWX
| SAB(datalo
, rbase
, TCG_REG_R0
));
1735 tcg_out32(s
, STW
| TAI(datahi
, addrlo
, 0));
1736 tcg_out32(s
, STW
| TAI(datalo
, addrlo
, 4));
1739 uint32_t insn
= qemu_stx_opc
[opc
& (MO_BSWAP
| MO_SIZE
)];
1740 if (!HAVE_ISA_2_06
&& insn
== STDBRX
) {
1741 tcg_out32(s
, STWBRX
| SAB(datalo
, rbase
, addrlo
));
1742 tcg_out32(s
, ADDI
| TAI(TCG_REG_TMP1
, addrlo
, 4));
1743 tcg_out_shri64(s
, TCG_REG_R0
, datalo
, 32);
1744 tcg_out32(s
, STWBRX
| SAB(TCG_REG_R0
, rbase
, TCG_REG_TMP1
));
1746 tcg_out32(s
, insn
| SAB(datalo
, rbase
, addrlo
));
1750 #ifdef CONFIG_SOFTMMU
1751 add_qemu_ldst_label(s
, false, oi
, datalo
, datahi
, addrlo
, addrhi
,
1752 s
->code_ptr
, label_ptr
);
1756 /* Parameters for function call generation, used in tcg.c. */
1757 #define TCG_TARGET_STACK_ALIGN 16
1758 #define TCG_TARGET_EXTEND_ARGS 1
1761 # define LINK_AREA_SIZE (6 * SZR)
1762 # define LR_OFFSET (1 * SZR)
1763 # define TCG_TARGET_CALL_STACK_OFFSET (LINK_AREA_SIZE + 8 * SZR)
1764 #elif defined(TCG_TARGET_CALL_DARWIN)
1765 # define LINK_AREA_SIZE (6 * SZR)
1766 # define LR_OFFSET (2 * SZR)
1767 #elif TCG_TARGET_REG_BITS == 64
1768 # if defined(_CALL_ELF) && _CALL_ELF == 2
1769 # define LINK_AREA_SIZE (4 * SZR)
1770 # define LR_OFFSET (1 * SZR)
1772 #else /* TCG_TARGET_REG_BITS == 32 */
1773 # if defined(_CALL_SYSV)
1774 # define LINK_AREA_SIZE (2 * SZR)
1775 # define LR_OFFSET (1 * SZR)
1779 # error "Unhandled abi"
1781 #ifndef TCG_TARGET_CALL_STACK_OFFSET
1782 # define TCG_TARGET_CALL_STACK_OFFSET LINK_AREA_SIZE
1785 #define CPU_TEMP_BUF_SIZE (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
1786 #define REG_SAVE_SIZE ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
1788 #define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET \
1789 + TCG_STATIC_CALL_ARGS_SIZE \
1790 + CPU_TEMP_BUF_SIZE \
1792 + TCG_TARGET_STACK_ALIGN - 1) \
1793 & -TCG_TARGET_STACK_ALIGN)
1795 #define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
1797 static void tcg_target_qemu_prologue(TCGContext
*s
)
1802 void **desc
= (void **)s
->code_ptr
;
1803 desc
[0] = desc
+ 2; /* entry point */
1804 desc
[1] = 0; /* environment pointer */
1805 s
->code_ptr
= (void *)(desc
+ 2); /* skip over descriptor */
1808 tcg_set_frame(s
, TCG_REG_CALL_STACK
, REG_SAVE_BOT
- CPU_TEMP_BUF_SIZE
,
1812 tcg_out32(s
, MFSPR
| RT(TCG_REG_R0
) | LR
);
1813 tcg_out32(s
, (SZR
== 8 ? STDU
: STWU
)
1814 | SAI(TCG_REG_R1
, TCG_REG_R1
, -FRAME_SIZE
));
1816 for (i
= 0; i
< ARRAY_SIZE(tcg_target_callee_save_regs
); ++i
) {
1817 tcg_out_st(s
, TCG_TYPE_REG
, tcg_target_callee_save_regs
[i
],
1818 TCG_REG_R1
, REG_SAVE_BOT
+ i
* SZR
);
1820 tcg_out_st(s
, TCG_TYPE_PTR
, TCG_REG_R0
, TCG_REG_R1
, FRAME_SIZE
+LR_OFFSET
);
1822 #ifndef CONFIG_SOFTMMU
1824 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_GUEST_BASE_REG
, guest_base
);
1825 tcg_regset_set_reg(s
->reserved_regs
, TCG_GUEST_BASE_REG
);
1829 tcg_out_mov(s
, TCG_TYPE_PTR
, TCG_AREG0
, tcg_target_call_iarg_regs
[0]);
1830 tcg_out32(s
, MTSPR
| RS(tcg_target_call_iarg_regs
[1]) | CTR
);
1834 /* Make the caller load the value as the TOC into R2. */
1835 tb_ret_addr
= s
->code_ptr
+ 2;
1836 desc
[1] = tb_ret_addr
;
1837 tcg_out_mov(s
, TCG_TYPE_PTR
, TCG_REG_RA
, TCG_REG_R2
);
1838 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1839 #elif defined(_CALL_ELF) && _CALL_ELF == 2
1840 /* Compute from the incoming R12 value. */
1841 tb_ret_addr
= s
->code_ptr
+ 2;
1842 tcg_out32(s
, ADDI
| TAI(TCG_REG_RA
, TCG_REG_R12
,
1843 tcg_ptr_byte_diff(tb_ret_addr
, s
->code_buf
)));
1844 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1846 /* Reserve max 5 insns for the constant load. */
1847 tb_ret_addr
= s
->code_ptr
+ 6;
1848 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_RA
, (intptr_t)tb_ret_addr
);
1849 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1850 while (s
->code_ptr
< tb_ret_addr
) {
1855 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1856 tb_ret_addr
= s
->code_ptr
;
1860 assert(tb_ret_addr
== s
->code_ptr
);
1862 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R0
, TCG_REG_R1
, FRAME_SIZE
+LR_OFFSET
);
1863 for (i
= 0; i
< ARRAY_SIZE(tcg_target_callee_save_regs
); ++i
) {
1864 tcg_out_ld(s
, TCG_TYPE_REG
, tcg_target_callee_save_regs
[i
],
1865 TCG_REG_R1
, REG_SAVE_BOT
+ i
* SZR
);
1867 tcg_out32(s
, MTSPR
| RS(TCG_REG_R0
) | LR
);
1868 tcg_out32(s
, ADDI
| TAI(TCG_REG_R1
, TCG_REG_R1
, FRAME_SIZE
));
1869 tcg_out32(s
, BCLR
| BO_ALWAYS
);
1872 static void tcg_out_op(TCGContext
*s
, TCGOpcode opc
, const TCGArg
*args
,
1873 const int *const_args
)
1879 case INDEX_op_exit_tb
:
1881 ptrdiff_t disp
= tcg_pcrel_diff(s
, tb_ret_addr
);
1883 /* Use a direct branch if we can, otherwise use the value in RA.
1884 Note that the direct branch is always backward, thus we need
1885 to account for the possibility of 5 insns from the movi. */
1886 if (!in_range_b(disp
- 20)) {
1887 tcg_out32(s
, MTSPR
| RS(TCG_REG_RA
) | CTR
);
1888 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_R3
, args
[0]);
1889 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1893 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_R3
, args
[0]);
1894 tcg_out_b(s
, 0, tb_ret_addr
);
1896 case INDEX_op_goto_tb
:
1897 tcg_debug_assert(s
->tb_jmp_offset
);
1898 /* Direct jump. Ensure the next insns are 8-byte aligned. */
1899 if ((uintptr_t)s
->code_ptr
& 7) {
1902 s
->tb_jmp_offset
[args
[0]] = tcg_current_code_size(s
);
1903 /* To be replaced by either a branch+nop or a load into TMP1. */
1905 tcg_out32(s
, MTSPR
| RS(TCG_REG_TMP1
) | CTR
);
1906 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1907 s
->tb_next_offset
[args
[0]] = tcg_current_code_size(s
);
1911 TCGLabel
*l
= arg_label(args
[0]);
1914 tcg_out_b(s
, 0, l
->u
.value_ptr
);
1916 tcg_out_reloc(s
, s
->code_ptr
, R_PPC_REL24
, l
, 0);
1917 tcg_out_b_noaddr(s
, B
);
1921 case INDEX_op_ld8u_i32
:
1922 case INDEX_op_ld8u_i64
:
1923 tcg_out_mem_long(s
, LBZ
, LBZX
, args
[0], args
[1], args
[2]);
1925 case INDEX_op_ld8s_i32
:
1926 case INDEX_op_ld8s_i64
:
1927 tcg_out_mem_long(s
, LBZ
, LBZX
, args
[0], args
[1], args
[2]);
1928 tcg_out32(s
, EXTSB
| RS(args
[0]) | RA(args
[0]));
1930 case INDEX_op_ld16u_i32
:
1931 case INDEX_op_ld16u_i64
:
1932 tcg_out_mem_long(s
, LHZ
, LHZX
, args
[0], args
[1], args
[2]);
1934 case INDEX_op_ld16s_i32
:
1935 case INDEX_op_ld16s_i64
:
1936 tcg_out_mem_long(s
, LHA
, LHAX
, args
[0], args
[1], args
[2]);
1938 case INDEX_op_ld_i32
:
1939 case INDEX_op_ld32u_i64
:
1940 tcg_out_mem_long(s
, LWZ
, LWZX
, args
[0], args
[1], args
[2]);
1942 case INDEX_op_ld32s_i64
:
1943 tcg_out_mem_long(s
, LWA
, LWAX
, args
[0], args
[1], args
[2]);
1945 case INDEX_op_ld_i64
:
1946 tcg_out_mem_long(s
, LD
, LDX
, args
[0], args
[1], args
[2]);
1948 case INDEX_op_st8_i32
:
1949 case INDEX_op_st8_i64
:
1950 tcg_out_mem_long(s
, STB
, STBX
, args
[0], args
[1], args
[2]);
1952 case INDEX_op_st16_i32
:
1953 case INDEX_op_st16_i64
:
1954 tcg_out_mem_long(s
, STH
, STHX
, args
[0], args
[1], args
[2]);
1956 case INDEX_op_st_i32
:
1957 case INDEX_op_st32_i64
:
1958 tcg_out_mem_long(s
, STW
, STWX
, args
[0], args
[1], args
[2]);
1960 case INDEX_op_st_i64
:
1961 tcg_out_mem_long(s
, STD
, STDX
, args
[0], args
[1], args
[2]);
1964 case INDEX_op_add_i32
:
1965 a0
= args
[0], a1
= args
[1], a2
= args
[2];
1966 if (const_args
[2]) {
1968 tcg_out_mem_long(s
, ADDI
, ADD
, a0
, a1
, (int32_t)a2
);
1970 tcg_out32(s
, ADD
| TAB(a0
, a1
, a2
));
1973 case INDEX_op_sub_i32
:
1974 a0
= args
[0], a1
= args
[1], a2
= args
[2];
1975 if (const_args
[1]) {
1976 if (const_args
[2]) {
1977 tcg_out_movi(s
, TCG_TYPE_I32
, a0
, a1
- a2
);
1979 tcg_out32(s
, SUBFIC
| TAI(a0
, a2
, a1
));
1981 } else if (const_args
[2]) {
1985 tcg_out32(s
, SUBF
| TAB(a0
, a2
, a1
));
1989 case INDEX_op_and_i32
:
1990 a0
= args
[0], a1
= args
[1], a2
= args
[2];
1991 if (const_args
[2]) {
1992 tcg_out_andi32(s
, a0
, a1
, a2
);
1994 tcg_out32(s
, AND
| SAB(a1
, a0
, a2
));
1997 case INDEX_op_and_i64
:
1998 a0
= args
[0], a1
= args
[1], a2
= args
[2];
1999 if (const_args
[2]) {
2000 tcg_out_andi64(s
, a0
, a1
, a2
);
2002 tcg_out32(s
, AND
| SAB(a1
, a0
, a2
));
2005 case INDEX_op_or_i64
:
2006 case INDEX_op_or_i32
:
2007 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2008 if (const_args
[2]) {
2009 tcg_out_ori32(s
, a0
, a1
, a2
);
2011 tcg_out32(s
, OR
| SAB(a1
, a0
, a2
));
2014 case INDEX_op_xor_i64
:
2015 case INDEX_op_xor_i32
:
2016 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2017 if (const_args
[2]) {
2018 tcg_out_xori32(s
, a0
, a1
, a2
);
2020 tcg_out32(s
, XOR
| SAB(a1
, a0
, a2
));
2023 case INDEX_op_andc_i32
:
2024 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2025 if (const_args
[2]) {
2026 tcg_out_andi32(s
, a0
, a1
, ~a2
);
2028 tcg_out32(s
, ANDC
| SAB(a1
, a0
, a2
));
2031 case INDEX_op_andc_i64
:
2032 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2033 if (const_args
[2]) {
2034 tcg_out_andi64(s
, a0
, a1
, ~a2
);
2036 tcg_out32(s
, ANDC
| SAB(a1
, a0
, a2
));
2039 case INDEX_op_orc_i32
:
2040 if (const_args
[2]) {
2041 tcg_out_ori32(s
, args
[0], args
[1], ~args
[2]);
2045 case INDEX_op_orc_i64
:
2046 tcg_out32(s
, ORC
| SAB(args
[1], args
[0], args
[2]));
2048 case INDEX_op_eqv_i32
:
2049 if (const_args
[2]) {
2050 tcg_out_xori32(s
, args
[0], args
[1], ~args
[2]);
2054 case INDEX_op_eqv_i64
:
2055 tcg_out32(s
, EQV
| SAB(args
[1], args
[0], args
[2]));
2057 case INDEX_op_nand_i32
:
2058 case INDEX_op_nand_i64
:
2059 tcg_out32(s
, NAND
| SAB(args
[1], args
[0], args
[2]));
2061 case INDEX_op_nor_i32
:
2062 case INDEX_op_nor_i64
:
2063 tcg_out32(s
, NOR
| SAB(args
[1], args
[0], args
[2]));
2066 case INDEX_op_mul_i32
:
2067 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2068 if (const_args
[2]) {
2069 tcg_out32(s
, MULLI
| TAI(a0
, a1
, a2
));
2071 tcg_out32(s
, MULLW
| TAB(a0
, a1
, a2
));
2075 case INDEX_op_div_i32
:
2076 tcg_out32(s
, DIVW
| TAB(args
[0], args
[1], args
[2]));
2079 case INDEX_op_divu_i32
:
2080 tcg_out32(s
, DIVWU
| TAB(args
[0], args
[1], args
[2]));
2083 case INDEX_op_shl_i32
:
2084 if (const_args
[2]) {
2085 tcg_out_shli32(s
, args
[0], args
[1], args
[2]);
2087 tcg_out32(s
, SLW
| SAB(args
[1], args
[0], args
[2]));
2090 case INDEX_op_shr_i32
:
2091 if (const_args
[2]) {
2092 tcg_out_shri32(s
, args
[0], args
[1], args
[2]);
2094 tcg_out32(s
, SRW
| SAB(args
[1], args
[0], args
[2]));
2097 case INDEX_op_sar_i32
:
2098 if (const_args
[2]) {
2099 tcg_out32(s
, SRAWI
| RS(args
[1]) | RA(args
[0]) | SH(args
[2]));
2101 tcg_out32(s
, SRAW
| SAB(args
[1], args
[0], args
[2]));
2104 case INDEX_op_rotl_i32
:
2105 if (const_args
[2]) {
2106 tcg_out_rlw(s
, RLWINM
, args
[0], args
[1], args
[2], 0, 31);
2108 tcg_out32(s
, RLWNM
| SAB(args
[1], args
[0], args
[2])
2112 case INDEX_op_rotr_i32
:
2113 if (const_args
[2]) {
2114 tcg_out_rlw(s
, RLWINM
, args
[0], args
[1], 32 - args
[2], 0, 31);
2116 tcg_out32(s
, SUBFIC
| TAI(TCG_REG_R0
, args
[2], 32));
2117 tcg_out32(s
, RLWNM
| SAB(args
[1], args
[0], TCG_REG_R0
)
2122 case INDEX_op_brcond_i32
:
2123 tcg_out_brcond(s
, args
[2], args
[0], args
[1], const_args
[1],
2124 arg_label(args
[3]), TCG_TYPE_I32
);
2126 case INDEX_op_brcond_i64
:
2127 tcg_out_brcond(s
, args
[2], args
[0], args
[1], const_args
[1],
2128 arg_label(args
[3]), TCG_TYPE_I64
);
2130 case INDEX_op_brcond2_i32
:
2131 tcg_out_brcond2(s
, args
, const_args
);
2134 case INDEX_op_neg_i32
:
2135 case INDEX_op_neg_i64
:
2136 tcg_out32(s
, NEG
| RT(args
[0]) | RA(args
[1]));
2139 case INDEX_op_not_i32
:
2140 case INDEX_op_not_i64
:
2141 tcg_out32(s
, NOR
| SAB(args
[1], args
[0], args
[1]));
2144 case INDEX_op_add_i64
:
2145 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2146 if (const_args
[2]) {
2148 tcg_out_mem_long(s
, ADDI
, ADD
, a0
, a1
, a2
);
2150 tcg_out32(s
, ADD
| TAB(a0
, a1
, a2
));
2153 case INDEX_op_sub_i64
:
2154 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2155 if (const_args
[1]) {
2156 if (const_args
[2]) {
2157 tcg_out_movi(s
, TCG_TYPE_I64
, a0
, a1
- a2
);
2159 tcg_out32(s
, SUBFIC
| TAI(a0
, a2
, a1
));
2161 } else if (const_args
[2]) {
2165 tcg_out32(s
, SUBF
| TAB(a0
, a2
, a1
));
2169 case INDEX_op_shl_i64
:
2170 if (const_args
[2]) {
2171 tcg_out_shli64(s
, args
[0], args
[1], args
[2]);
2173 tcg_out32(s
, SLD
| SAB(args
[1], args
[0], args
[2]));
2176 case INDEX_op_shr_i64
:
2177 if (const_args
[2]) {
2178 tcg_out_shri64(s
, args
[0], args
[1], args
[2]);
2180 tcg_out32(s
, SRD
| SAB(args
[1], args
[0], args
[2]));
2183 case INDEX_op_sar_i64
:
2184 if (const_args
[2]) {
2185 int sh
= SH(args
[2] & 0x1f) | (((args
[2] >> 5) & 1) << 1);
2186 tcg_out32(s
, SRADI
| RA(args
[0]) | RS(args
[1]) | sh
);
2188 tcg_out32(s
, SRAD
| SAB(args
[1], args
[0], args
[2]));
2191 case INDEX_op_rotl_i64
:
2192 if (const_args
[2]) {
2193 tcg_out_rld(s
, RLDICL
, args
[0], args
[1], args
[2], 0);
2195 tcg_out32(s
, RLDCL
| SAB(args
[1], args
[0], args
[2]) | MB64(0));
2198 case INDEX_op_rotr_i64
:
2199 if (const_args
[2]) {
2200 tcg_out_rld(s
, RLDICL
, args
[0], args
[1], 64 - args
[2], 0);
2202 tcg_out32(s
, SUBFIC
| TAI(TCG_REG_R0
, args
[2], 64));
2203 tcg_out32(s
, RLDCL
| SAB(args
[1], args
[0], TCG_REG_R0
) | MB64(0));
2207 case INDEX_op_mul_i64
:
2208 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2209 if (const_args
[2]) {
2210 tcg_out32(s
, MULLI
| TAI(a0
, a1
, a2
));
2212 tcg_out32(s
, MULLD
| TAB(a0
, a1
, a2
));
2215 case INDEX_op_div_i64
:
2216 tcg_out32(s
, DIVD
| TAB(args
[0], args
[1], args
[2]));
2218 case INDEX_op_divu_i64
:
2219 tcg_out32(s
, DIVDU
| TAB(args
[0], args
[1], args
[2]));
2222 case INDEX_op_qemu_ld_i32
:
2223 tcg_out_qemu_ld(s
, args
, false);
2225 case INDEX_op_qemu_ld_i64
:
2226 tcg_out_qemu_ld(s
, args
, true);
2228 case INDEX_op_qemu_st_i32
:
2229 tcg_out_qemu_st(s
, args
, false);
2231 case INDEX_op_qemu_st_i64
:
2232 tcg_out_qemu_st(s
, args
, true);
2235 case INDEX_op_ext8s_i32
:
2236 case INDEX_op_ext8s_i64
:
2239 case INDEX_op_ext16s_i32
:
2240 case INDEX_op_ext16s_i64
:
2243 case INDEX_op_ext_i32_i64
:
2244 case INDEX_op_ext32s_i64
:
2248 tcg_out32(s
, c
| RS(args
[1]) | RA(args
[0]));
2250 case INDEX_op_extu_i32_i64
:
2251 tcg_out_ext32u(s
, args
[0], args
[1]);
2254 case INDEX_op_setcond_i32
:
2255 tcg_out_setcond(s
, TCG_TYPE_I32
, args
[3], args
[0], args
[1], args
[2],
2258 case INDEX_op_setcond_i64
:
2259 tcg_out_setcond(s
, TCG_TYPE_I64
, args
[3], args
[0], args
[1], args
[2],
2262 case INDEX_op_setcond2_i32
:
2263 tcg_out_setcond2(s
, args
, const_args
);
2266 case INDEX_op_bswap16_i32
:
2267 case INDEX_op_bswap16_i64
:
2268 a0
= args
[0], a1
= args
[1];
2271 /* a0 = (a1 r<< 24) & 0xff # 000c */
2272 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 24, 24, 31);
2273 /* a0 = (a0 & ~0xff00) | (a1 r<< 8) & 0xff00 # 00dc */
2274 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 8, 16, 23);
2276 /* r0 = (a1 r<< 8) & 0xff00 # 00d0 */
2277 tcg_out_rlw(s
, RLWINM
, TCG_REG_R0
, a1
, 8, 16, 23);
2278 /* a0 = (a1 r<< 24) & 0xff # 000c */
2279 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 24, 24, 31);
2280 /* a0 = a0 | r0 # 00dc */
2281 tcg_out32(s
, OR
| SAB(TCG_REG_R0
, a0
, a0
));
2285 case INDEX_op_bswap32_i32
:
2286 case INDEX_op_bswap32_i64
:
2287 /* Stolen from gcc's builtin_bswap32 */
2289 a0
= args
[0] == a1
? TCG_REG_R0
: args
[0];
2291 /* a1 = args[1] # abcd */
2292 /* a0 = rotate_left (a1, 8) # bcda */
2293 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 8, 0, 31);
2294 /* a0 = (a0 & ~0xff000000) | ((a1 r<< 24) & 0xff000000) # dcda */
2295 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 0, 7);
2296 /* a0 = (a0 & ~0x0000ff00) | ((a1 r<< 24) & 0x0000ff00) # dcba */
2297 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 16, 23);
2299 if (a0
== TCG_REG_R0
) {
2300 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2304 case INDEX_op_bswap64_i64
:
2305 a0
= args
[0], a1
= args
[1], a2
= TCG_REG_R0
;
2311 /* a1 = # abcd efgh */
2312 /* a0 = rl32(a1, 8) # 0000 fghe */
2313 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 8, 0, 31);
2314 /* a0 = dep(a0, rl32(a1, 24), 0xff000000) # 0000 hghe */
2315 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 0, 7);
2316 /* a0 = dep(a0, rl32(a1, 24), 0x0000ff00) # 0000 hgfe */
2317 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 16, 23);
2319 /* a0 = rl64(a0, 32) # hgfe 0000 */
2320 /* a2 = rl64(a1, 32) # efgh abcd */
2321 tcg_out_rld(s
, RLDICL
, a0
, a0
, 32, 0);
2322 tcg_out_rld(s
, RLDICL
, a2
, a1
, 32, 0);
2324 /* a0 = dep(a0, rl32(a2, 8), 0xffffffff) # hgfe bcda */
2325 tcg_out_rlw(s
, RLWIMI
, a0
, a2
, 8, 0, 31);
2326 /* a0 = dep(a0, rl32(a2, 24), 0xff000000) # hgfe dcda */
2327 tcg_out_rlw(s
, RLWIMI
, a0
, a2
, 24, 0, 7);
2328 /* a0 = dep(a0, rl32(a2, 24), 0x0000ff00) # hgfe dcba */
2329 tcg_out_rlw(s
, RLWIMI
, a0
, a2
, 24, 16, 23);
2332 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2336 case INDEX_op_deposit_i32
:
2337 if (const_args
[2]) {
2338 uint32_t mask
= ((2u << (args
[4] - 1)) - 1) << args
[3];
2339 tcg_out_andi32(s
, args
[0], args
[0], ~mask
);
2341 tcg_out_rlw(s
, RLWIMI
, args
[0], args
[2], args
[3],
2342 32 - args
[3] - args
[4], 31 - args
[3]);
2345 case INDEX_op_deposit_i64
:
2346 if (const_args
[2]) {
2347 uint64_t mask
= ((2ull << (args
[4] - 1)) - 1) << args
[3];
2348 tcg_out_andi64(s
, args
[0], args
[0], ~mask
);
2350 tcg_out_rld(s
, RLDIMI
, args
[0], args
[2], args
[3],
2351 64 - args
[3] - args
[4]);
2355 case INDEX_op_movcond_i32
:
2356 tcg_out_movcond(s
, TCG_TYPE_I32
, args
[5], args
[0], args
[1], args
[2],
2357 args
[3], args
[4], const_args
[2]);
2359 case INDEX_op_movcond_i64
:
2360 tcg_out_movcond(s
, TCG_TYPE_I64
, args
[5], args
[0], args
[1], args
[2],
2361 args
[3], args
[4], const_args
[2]);
2364 #if TCG_TARGET_REG_BITS == 64
2365 case INDEX_op_add2_i64
:
2367 case INDEX_op_add2_i32
:
2369 /* Note that the CA bit is defined based on the word size of the
2370 environment. So in 64-bit mode it's always carry-out of bit 63.
2371 The fallback code using deposit works just as well for 32-bit. */
2372 a0
= args
[0], a1
= args
[1];
2373 if (a0
== args
[3] || (!const_args
[5] && a0
== args
[5])) {
2376 if (const_args
[4]) {
2377 tcg_out32(s
, ADDIC
| TAI(a0
, args
[2], args
[4]));
2379 tcg_out32(s
, ADDC
| TAB(a0
, args
[2], args
[4]));
2381 if (const_args
[5]) {
2382 tcg_out32(s
, (args
[5] ? ADDME
: ADDZE
) | RT(a1
) | RA(args
[3]));
2384 tcg_out32(s
, ADDE
| TAB(a1
, args
[3], args
[5]));
2386 if (a0
!= args
[0]) {
2387 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2391 #if TCG_TARGET_REG_BITS == 64
2392 case INDEX_op_sub2_i64
:
2394 case INDEX_op_sub2_i32
:
2396 a0
= args
[0], a1
= args
[1];
2397 if (a0
== args
[5] || (!const_args
[3] && a0
== args
[3])) {
2400 if (const_args
[2]) {
2401 tcg_out32(s
, SUBFIC
| TAI(a0
, args
[4], args
[2]));
2403 tcg_out32(s
, SUBFC
| TAB(a0
, args
[4], args
[2]));
2405 if (const_args
[3]) {
2406 tcg_out32(s
, (args
[3] ? SUBFME
: SUBFZE
) | RT(a1
) | RA(args
[5]));
2408 tcg_out32(s
, SUBFE
| TAB(a1
, args
[5], args
[3]));
2410 if (a0
!= args
[0]) {
2411 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2415 case INDEX_op_muluh_i32
:
2416 tcg_out32(s
, MULHWU
| TAB(args
[0], args
[1], args
[2]));
2418 case INDEX_op_mulsh_i32
:
2419 tcg_out32(s
, MULHW
| TAB(args
[0], args
[1], args
[2]));
2421 case INDEX_op_muluh_i64
:
2422 tcg_out32(s
, MULHDU
| TAB(args
[0], args
[1], args
[2]));
2424 case INDEX_op_mulsh_i64
:
2425 tcg_out32(s
, MULHD
| TAB(args
[0], args
[1], args
[2]));
2428 case INDEX_op_mov_i32
: /* Always emitted via tcg_out_mov. */
2429 case INDEX_op_mov_i64
:
2430 case INDEX_op_movi_i32
: /* Always emitted via tcg_out_movi. */
2431 case INDEX_op_movi_i64
:
2432 case INDEX_op_call
: /* Always emitted via tcg_out_call. */
2438 static const TCGTargetOpDef ppc_op_defs
[] = {
2439 { INDEX_op_exit_tb
, { } },
2440 { INDEX_op_goto_tb
, { } },
2441 { INDEX_op_br
, { } },
2443 { INDEX_op_ld8u_i32
, { "r", "r" } },
2444 { INDEX_op_ld8s_i32
, { "r", "r" } },
2445 { INDEX_op_ld16u_i32
, { "r", "r" } },
2446 { INDEX_op_ld16s_i32
, { "r", "r" } },
2447 { INDEX_op_ld_i32
, { "r", "r" } },
2449 { INDEX_op_st8_i32
, { "r", "r" } },
2450 { INDEX_op_st16_i32
, { "r", "r" } },
2451 { INDEX_op_st_i32
, { "r", "r" } },
2453 { INDEX_op_add_i32
, { "r", "r", "ri" } },
2454 { INDEX_op_mul_i32
, { "r", "r", "rI" } },
2455 { INDEX_op_div_i32
, { "r", "r", "r" } },
2456 { INDEX_op_divu_i32
, { "r", "r", "r" } },
2457 { INDEX_op_sub_i32
, { "r", "rI", "ri" } },
2458 { INDEX_op_and_i32
, { "r", "r", "ri" } },
2459 { INDEX_op_or_i32
, { "r", "r", "ri" } },
2460 { INDEX_op_xor_i32
, { "r", "r", "ri" } },
2461 { INDEX_op_andc_i32
, { "r", "r", "ri" } },
2462 { INDEX_op_orc_i32
, { "r", "r", "ri" } },
2463 { INDEX_op_eqv_i32
, { "r", "r", "ri" } },
2464 { INDEX_op_nand_i32
, { "r", "r", "r" } },
2465 { INDEX_op_nor_i32
, { "r", "r", "r" } },
2467 { INDEX_op_shl_i32
, { "r", "r", "ri" } },
2468 { INDEX_op_shr_i32
, { "r", "r", "ri" } },
2469 { INDEX_op_sar_i32
, { "r", "r", "ri" } },
2470 { INDEX_op_rotl_i32
, { "r", "r", "ri" } },
2471 { INDEX_op_rotr_i32
, { "r", "r", "ri" } },
2473 { INDEX_op_neg_i32
, { "r", "r" } },
2474 { INDEX_op_not_i32
, { "r", "r" } },
2475 { INDEX_op_ext8s_i32
, { "r", "r" } },
2476 { INDEX_op_ext16s_i32
, { "r", "r" } },
2477 { INDEX_op_bswap16_i32
, { "r", "r" } },
2478 { INDEX_op_bswap32_i32
, { "r", "r" } },
2480 { INDEX_op_brcond_i32
, { "r", "ri" } },
2481 { INDEX_op_setcond_i32
, { "r", "r", "ri" } },
2482 { INDEX_op_movcond_i32
, { "r", "r", "ri", "rZ", "rZ" } },
2484 { INDEX_op_deposit_i32
, { "r", "0", "rZ" } },
2486 { INDEX_op_muluh_i32
, { "r", "r", "r" } },
2487 { INDEX_op_mulsh_i32
, { "r", "r", "r" } },
2489 #if TCG_TARGET_REG_BITS == 64
2490 { INDEX_op_ld8u_i64
, { "r", "r" } },
2491 { INDEX_op_ld8s_i64
, { "r", "r" } },
2492 { INDEX_op_ld16u_i64
, { "r", "r" } },
2493 { INDEX_op_ld16s_i64
, { "r", "r" } },
2494 { INDEX_op_ld32u_i64
, { "r", "r" } },
2495 { INDEX_op_ld32s_i64
, { "r", "r" } },
2496 { INDEX_op_ld_i64
, { "r", "r" } },
2498 { INDEX_op_st8_i64
, { "r", "r" } },
2499 { INDEX_op_st16_i64
, { "r", "r" } },
2500 { INDEX_op_st32_i64
, { "r", "r" } },
2501 { INDEX_op_st_i64
, { "r", "r" } },
2503 { INDEX_op_add_i64
, { "r", "r", "rT" } },
2504 { INDEX_op_sub_i64
, { "r", "rI", "rT" } },
2505 { INDEX_op_and_i64
, { "r", "r", "ri" } },
2506 { INDEX_op_or_i64
, { "r", "r", "rU" } },
2507 { INDEX_op_xor_i64
, { "r", "r", "rU" } },
2508 { INDEX_op_andc_i64
, { "r", "r", "ri" } },
2509 { INDEX_op_orc_i64
, { "r", "r", "r" } },
2510 { INDEX_op_eqv_i64
, { "r", "r", "r" } },
2511 { INDEX_op_nand_i64
, { "r", "r", "r" } },
2512 { INDEX_op_nor_i64
, { "r", "r", "r" } },
2514 { INDEX_op_shl_i64
, { "r", "r", "ri" } },
2515 { INDEX_op_shr_i64
, { "r", "r", "ri" } },
2516 { INDEX_op_sar_i64
, { "r", "r", "ri" } },
2517 { INDEX_op_rotl_i64
, { "r", "r", "ri" } },
2518 { INDEX_op_rotr_i64
, { "r", "r", "ri" } },
2520 { INDEX_op_mul_i64
, { "r", "r", "rI" } },
2521 { INDEX_op_div_i64
, { "r", "r", "r" } },
2522 { INDEX_op_divu_i64
, { "r", "r", "r" } },
2524 { INDEX_op_neg_i64
, { "r", "r" } },
2525 { INDEX_op_not_i64
, { "r", "r" } },
2526 { INDEX_op_ext8s_i64
, { "r", "r" } },
2527 { INDEX_op_ext16s_i64
, { "r", "r" } },
2528 { INDEX_op_ext32s_i64
, { "r", "r" } },
2529 { INDEX_op_ext_i32_i64
, { "r", "r" } },
2530 { INDEX_op_extu_i32_i64
, { "r", "r" } },
2531 { INDEX_op_bswap16_i64
, { "r", "r" } },
2532 { INDEX_op_bswap32_i64
, { "r", "r" } },
2533 { INDEX_op_bswap64_i64
, { "r", "r" } },
2535 { INDEX_op_brcond_i64
, { "r", "ri" } },
2536 { INDEX_op_setcond_i64
, { "r", "r", "ri" } },
2537 { INDEX_op_movcond_i64
, { "r", "r", "ri", "rZ", "rZ" } },
2539 { INDEX_op_deposit_i64
, { "r", "0", "rZ" } },
2541 { INDEX_op_mulsh_i64
, { "r", "r", "r" } },
2542 { INDEX_op_muluh_i64
, { "r", "r", "r" } },
2545 #if TCG_TARGET_REG_BITS == 32
2546 { INDEX_op_brcond2_i32
, { "r", "r", "ri", "ri" } },
2547 { INDEX_op_setcond2_i32
, { "r", "r", "r", "ri", "ri" } },
2550 #if TCG_TARGET_REG_BITS == 64
2551 { INDEX_op_add2_i64
, { "r", "r", "r", "r", "rI", "rZM" } },
2552 { INDEX_op_sub2_i64
, { "r", "r", "rI", "rZM", "r", "r" } },
2554 { INDEX_op_add2_i32
, { "r", "r", "r", "r", "rI", "rZM" } },
2555 { INDEX_op_sub2_i32
, { "r", "r", "rI", "rZM", "r", "r" } },
2558 #if TCG_TARGET_REG_BITS == 64
2559 { INDEX_op_qemu_ld_i32
, { "r", "L" } },
2560 { INDEX_op_qemu_st_i32
, { "S", "S" } },
2561 { INDEX_op_qemu_ld_i64
, { "r", "L" } },
2562 { INDEX_op_qemu_st_i64
, { "S", "S" } },
2563 #elif TARGET_LONG_BITS == 32
2564 { INDEX_op_qemu_ld_i32
, { "r", "L" } },
2565 { INDEX_op_qemu_st_i32
, { "S", "S" } },
2566 { INDEX_op_qemu_ld_i64
, { "L", "L", "L" } },
2567 { INDEX_op_qemu_st_i64
, { "S", "S", "S" } },
2569 { INDEX_op_qemu_ld_i32
, { "r", "L", "L" } },
2570 { INDEX_op_qemu_st_i32
, { "S", "S", "S" } },
2571 { INDEX_op_qemu_ld_i64
, { "L", "L", "L", "L" } },
2572 { INDEX_op_qemu_st_i64
, { "S", "S", "S", "S" } },
2578 static void tcg_target_init(TCGContext
*s
)
2580 unsigned long hwcap
= qemu_getauxval(AT_HWCAP
);
2581 if (hwcap
& PPC_FEATURE_ARCH_2_06
) {
2582 have_isa_2_06
= true;
2585 tcg_regset_set32(tcg_target_available_regs
[TCG_TYPE_I32
], 0, 0xffffffff);
2586 tcg_regset_set32(tcg_target_available_regs
[TCG_TYPE_I64
], 0, 0xffffffff);
2587 tcg_regset_set32(tcg_target_call_clobber_regs
, 0,
2597 (1 << TCG_REG_R10
) |
2598 (1 << TCG_REG_R11
) |
2599 (1 << TCG_REG_R12
));
2601 tcg_regset_clear(s
->reserved_regs
);
2602 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R0
); /* tcg temp */
2603 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R1
); /* stack pointer */
2604 #if defined(_CALL_SYSV)
2605 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R2
); /* toc pointer */
2607 #if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
2608 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R13
); /* thread pointer */
2610 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_TMP1
); /* mem temp */
2612 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_RA
); /* return addr */
2615 tcg_add_target_add_op_defs(ppc_op_defs
);
2621 DebugFrameFDEHeader fde
;
2622 uint8_t fde_def_cfa
[4];
2623 uint8_t fde_reg_ofs
[ARRAY_SIZE(tcg_target_callee_save_regs
) * 2 + 3];
2626 /* We're expecting a 2 byte uleb128 encoded value. */
2627 QEMU_BUILD_BUG_ON(FRAME_SIZE
>= (1 << 14));
2629 #if TCG_TARGET_REG_BITS == 64
2630 # define ELF_HOST_MACHINE EM_PPC64
2632 # define ELF_HOST_MACHINE EM_PPC
2635 static DebugFrame debug_frame
= {
2636 .cie
.len
= sizeof(DebugFrameCIE
)-4, /* length after .len member */
2639 .cie
.code_align
= 1,
2640 .cie
.data_align
= (-SZR
& 0x7f), /* sleb128 -SZR */
2641 .cie
.return_column
= 65,
2643 /* Total FDE size does not include the "len" member. */
2644 .fde
.len
= sizeof(DebugFrame
) - offsetof(DebugFrame
, fde
.cie_offset
),
2647 12, TCG_REG_R1
, /* DW_CFA_def_cfa r1, ... */
2648 (FRAME_SIZE
& 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */
2652 /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
2653 0x11, 65, (LR_OFFSET
/ -SZR
) & 0x7f,
2657 void tcg_register_jit(void *buf
, size_t buf_size
)
2659 uint8_t *p
= &debug_frame
.fde_reg_ofs
[3];
2662 for (i
= 0; i
< ARRAY_SIZE(tcg_target_callee_save_regs
); ++i
, p
+= 2) {
2663 p
[0] = 0x80 + tcg_target_callee_save_regs
[i
];
2664 p
[1] = (FRAME_SIZE
- (REG_SAVE_BOT
+ i
* SZR
)) / SZR
;
2667 debug_frame
.fde
.func_start
= (uintptr_t)buf
;
2668 debug_frame
.fde
.func_len
= buf_size
;
2670 tcg_register_jit_int(buf
, buf_size
, &debug_frame
, sizeof(debug_frame
));
2672 #endif /* __ELF__ */
2674 static size_t dcache_bsize
= 16;
2675 static size_t icache_bsize
= 16;
2677 void flush_icache_range(uintptr_t start
, uintptr_t stop
)
2679 uintptr_t p
, start1
, stop1
;
2680 size_t dsize
= dcache_bsize
;
2681 size_t isize
= icache_bsize
;
2683 start1
= start
& ~(dsize
- 1);
2684 stop1
= (stop
+ dsize
- 1) & ~(dsize
- 1);
2685 for (p
= start1
; p
< stop1
; p
+= dsize
) {
2686 asm volatile ("dcbst 0,%0" : : "r"(p
) : "memory");
2688 asm volatile ("sync" : : : "memory");
2690 start
&= start
& ~(isize
- 1);
2691 stop1
= (stop
+ isize
- 1) & ~(isize
- 1);
2692 for (p
= start1
; p
< stop1
; p
+= isize
) {
2693 asm volatile ("icbi 0,%0" : : "r"(p
) : "memory");
2695 asm volatile ("sync" : : : "memory");
2696 asm volatile ("isync" : : : "memory");
2700 #include <sys/systemcfg.h>
2702 static void __attribute__((constructor
)) tcg_cache_init(void)
2704 icache_bsize
= _system_configuration
.icache_line
;
2705 dcache_bsize
= _system_configuration
.dcache_line
;
2708 #elif defined __linux__
2709 static void __attribute__((constructor
)) tcg_cache_init(void)
2711 unsigned long dsize
= qemu_getauxval(AT_DCACHEBSIZE
);
2712 unsigned long isize
= qemu_getauxval(AT_ICACHEBSIZE
);
2714 if (dsize
== 0 || isize
== 0) {
2716 fprintf(stderr
, "getauxval AT_DCACHEBSIZE failed\n");
2719 fprintf(stderr
, "getauxval AT_ICACHEBSIZE failed\n");
2723 dcache_bsize
= dsize
;
2724 icache_bsize
= isize
;
2727 #elif defined __APPLE__
2728 #include <sys/sysctl.h>
2730 static void __attribute__((constructor
)) tcg_cache_init(void)
2734 int name
[2] = { CTL_HW
, HW_CACHELINE
};
2736 len
= sizeof(cacheline
);
2737 if (sysctl(name
, 2, &cacheline
, &len
, NULL
, 0)) {
2738 perror("sysctl CTL_HW HW_CACHELINE failed");
2741 dcache_bsize
= cacheline
;
2742 icache_bsize
= cacheline
;
2745 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2746 #include <sys/sysctl.h>
2748 static void __attribute__((constructor
)) tcg_cache_init(void)
2753 if (sysctlbyname ("machdep.cacheline_size", &cacheline
, &len
, NULL
, 0)) {
2754 fprintf(stderr
, "sysctlbyname machdep.cacheline_size failed: %s\n",
2758 dcache_bsize
= cacheline
;
2759 icache_bsize
= cacheline
;