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
92 #ifdef CONFIG_DEBUG_TCG
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 tcg_debug_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 tcg_debug_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
;
248 tcg_debug_assert(addend
== 0);
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 EIEIO XO31(854)
473 #define HWSYNC XO31(598)
474 #define LWSYNC (HWSYNC | (1u << 21))
476 #define SPR(a, b) ((((a)<<5)|(b))<<11)
478 #define CTR SPR(9, 0)
480 #define SLW XO31( 24)
481 #define SRW XO31(536)
482 #define SRAW XO31(792)
484 #define SLD XO31( 27)
485 #define SRD XO31(539)
486 #define SRAD XO31(794)
487 #define SRADI XO31(413<<1)
490 #define TRAP (TW | TO(31))
492 #define NOP ORI /* ori 0,0,0 */
494 #define RT(r) ((r)<<21)
495 #define RS(r) ((r)<<21)
496 #define RA(r) ((r)<<16)
497 #define RB(r) ((r)<<11)
498 #define TO(t) ((t)<<21)
499 #define SH(s) ((s)<<11)
500 #define MB(b) ((b)<<6)
501 #define ME(e) ((e)<<1)
502 #define BO(o) ((o)<<21)
503 #define MB64(b) ((b)<<5)
504 #define FXM(b) (1 << (19 - (b)))
508 #define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
509 #define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
510 #define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
511 #define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
513 #define BF(n) ((n)<<23)
514 #define BI(n, c) (((c)+((n)*4))<<16)
515 #define BT(n, c) (((c)+((n)*4))<<21)
516 #define BA(n, c) (((c)+((n)*4))<<16)
517 #define BB(n, c) (((c)+((n)*4))<<11)
518 #define BC_(n, c) (((c)+((n)*4))<<6)
520 #define BO_COND_TRUE BO(12)
521 #define BO_COND_FALSE BO( 4)
522 #define BO_ALWAYS BO(20)
531 static const uint32_t tcg_to_bc
[] = {
532 [TCG_COND_EQ
] = BC
| BI(7, CR_EQ
) | BO_COND_TRUE
,
533 [TCG_COND_NE
] = BC
| BI(7, CR_EQ
) | BO_COND_FALSE
,
534 [TCG_COND_LT
] = BC
| BI(7, CR_LT
) | BO_COND_TRUE
,
535 [TCG_COND_GE
] = BC
| BI(7, CR_LT
) | BO_COND_FALSE
,
536 [TCG_COND_LE
] = BC
| BI(7, CR_GT
) | BO_COND_FALSE
,
537 [TCG_COND_GT
] = BC
| BI(7, CR_GT
) | BO_COND_TRUE
,
538 [TCG_COND_LTU
] = BC
| BI(7, CR_LT
) | BO_COND_TRUE
,
539 [TCG_COND_GEU
] = BC
| BI(7, CR_LT
) | BO_COND_FALSE
,
540 [TCG_COND_LEU
] = BC
| BI(7, CR_GT
) | BO_COND_FALSE
,
541 [TCG_COND_GTU
] = BC
| BI(7, CR_GT
) | BO_COND_TRUE
,
544 /* The low bit here is set if the RA and RB fields must be inverted. */
545 static const uint32_t tcg_to_isel
[] = {
546 [TCG_COND_EQ
] = ISEL
| BC_(7, CR_EQ
),
547 [TCG_COND_NE
] = ISEL
| BC_(7, CR_EQ
) | 1,
548 [TCG_COND_LT
] = ISEL
| BC_(7, CR_LT
),
549 [TCG_COND_GE
] = ISEL
| BC_(7, CR_LT
) | 1,
550 [TCG_COND_LE
] = ISEL
| BC_(7, CR_GT
) | 1,
551 [TCG_COND_GT
] = ISEL
| BC_(7, CR_GT
),
552 [TCG_COND_LTU
] = ISEL
| BC_(7, CR_LT
),
553 [TCG_COND_GEU
] = ISEL
| BC_(7, CR_LT
) | 1,
554 [TCG_COND_LEU
] = ISEL
| BC_(7, CR_GT
) | 1,
555 [TCG_COND_GTU
] = ISEL
| BC_(7, CR_GT
),
558 static void tcg_out_mem_long(TCGContext
*s
, int opi
, int opx
, TCGReg rt
,
559 TCGReg base
, tcg_target_long offset
);
561 static void tcg_out_mov(TCGContext
*s
, TCGType type
, TCGReg ret
, TCGReg arg
)
563 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
565 tcg_out32(s
, OR
| SAB(arg
, ret
, arg
));
569 static inline void tcg_out_rld(TCGContext
*s
, int op
, TCGReg ra
, TCGReg rs
,
572 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64);
573 sh
= SH(sh
& 0x1f) | (((sh
>> 5) & 1) << 1);
574 mb
= MB64((mb
>> 5) | ((mb
<< 1) & 0x3f));
575 tcg_out32(s
, op
| RA(ra
) | RS(rs
) | sh
| mb
);
578 static inline void tcg_out_rlw(TCGContext
*s
, int op
, TCGReg ra
, TCGReg rs
,
579 int sh
, int mb
, int me
)
581 tcg_out32(s
, op
| RA(ra
) | RS(rs
) | SH(sh
) | MB(mb
) | ME(me
));
584 static inline void tcg_out_ext32u(TCGContext
*s
, TCGReg dst
, TCGReg src
)
586 tcg_out_rld(s
, RLDICL
, dst
, src
, 0, 32);
589 static inline void tcg_out_shli32(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
591 tcg_out_rlw(s
, RLWINM
, dst
, src
, c
, 0, 31 - c
);
594 static inline void tcg_out_shli64(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
596 tcg_out_rld(s
, RLDICR
, dst
, src
, c
, 63 - c
);
599 static inline void tcg_out_shri32(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
601 tcg_out_rlw(s
, RLWINM
, dst
, src
, 32 - c
, c
, 31);
604 static inline void tcg_out_shri64(TCGContext
*s
, TCGReg dst
, TCGReg src
, int c
)
606 tcg_out_rld(s
, RLDICL
, dst
, src
, 64 - c
, c
);
609 static void tcg_out_movi32(TCGContext
*s
, TCGReg ret
, int32_t arg
)
611 if (arg
== (int16_t) arg
) {
612 tcg_out32(s
, ADDI
| TAI(ret
, 0, arg
));
614 tcg_out32(s
, ADDIS
| TAI(ret
, 0, arg
>> 16));
616 tcg_out32(s
, ORI
| SAI(ret
, ret
, arg
));
621 static void tcg_out_movi(TCGContext
*s
, TCGType type
, TCGReg ret
,
624 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
625 if (type
== TCG_TYPE_I32
|| arg
== (int32_t)arg
) {
626 tcg_out_movi32(s
, ret
, arg
);
627 } else if (arg
== (uint32_t)arg
&& !(arg
& 0x8000)) {
628 tcg_out32(s
, ADDI
| TAI(ret
, 0, arg
));
629 tcg_out32(s
, ORIS
| SAI(ret
, ret
, arg
>> 16));
634 intptr_t diff
= arg
- (intptr_t)tb_ret_addr
;
635 if (diff
== (int32_t)diff
) {
636 tcg_out_mem_long(s
, ADDI
, ADD
, ret
, TCG_REG_RA
, diff
);
641 high
= arg
>> 31 >> 1;
642 tcg_out_movi32(s
, ret
, high
);
644 tcg_out_shli64(s
, ret
, ret
, 32);
646 if (arg
& 0xffff0000) {
647 tcg_out32(s
, ORIS
| SAI(ret
, ret
, arg
>> 16));
650 tcg_out32(s
, ORI
| SAI(ret
, ret
, arg
));
655 static bool mask_operand(uint32_t c
, int *mb
, int *me
)
659 /* Accept a bit pattern like:
663 Keep track of the transitions. */
664 if (c
== 0 || c
== -1) {
670 if (test
& (test
- 1)) {
675 *mb
= test
? clz32(test
& -test
) + 1 : 0;
679 static bool mask64_operand(uint64_t c
, int *mb
, int *me
)
688 /* Accept 1..10..0. */
694 /* Accept 0..01..1. */
695 if (lsb
== 1 && (c
& (c
+ 1)) == 0) {
696 *mb
= clz64(c
+ 1) + 1;
703 static void tcg_out_andi32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
)
707 if (mask_operand(c
, &mb
, &me
)) {
708 tcg_out_rlw(s
, RLWINM
, dst
, src
, 0, mb
, me
);
709 } else if ((c
& 0xffff) == c
) {
710 tcg_out32(s
, ANDI
| SAI(src
, dst
, c
));
712 } else if ((c
& 0xffff0000) == c
) {
713 tcg_out32(s
, ANDIS
| SAI(src
, dst
, c
>> 16));
716 tcg_out_movi(s
, TCG_TYPE_I32
, TCG_REG_R0
, c
);
717 tcg_out32(s
, AND
| SAB(src
, dst
, TCG_REG_R0
));
721 static void tcg_out_andi64(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint64_t c
)
725 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64);
726 if (mask64_operand(c
, &mb
, &me
)) {
728 tcg_out_rld(s
, RLDICR
, dst
, src
, 0, me
);
730 tcg_out_rld(s
, RLDICL
, dst
, src
, 0, mb
);
732 } else if ((c
& 0xffff) == c
) {
733 tcg_out32(s
, ANDI
| SAI(src
, dst
, c
));
735 } else if ((c
& 0xffff0000) == c
) {
736 tcg_out32(s
, ANDIS
| SAI(src
, dst
, c
>> 16));
739 tcg_out_movi(s
, TCG_TYPE_I64
, TCG_REG_R0
, c
);
740 tcg_out32(s
, AND
| SAB(src
, dst
, TCG_REG_R0
));
744 static void tcg_out_zori32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
,
745 int op_lo
, int op_hi
)
748 tcg_out32(s
, op_hi
| SAI(src
, dst
, c
>> 16));
752 tcg_out32(s
, op_lo
| SAI(src
, dst
, c
));
757 static void tcg_out_ori32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
)
759 tcg_out_zori32(s
, dst
, src
, c
, ORI
, ORIS
);
762 static void tcg_out_xori32(TCGContext
*s
, TCGReg dst
, TCGReg src
, uint32_t c
)
764 tcg_out_zori32(s
, dst
, src
, c
, XORI
, XORIS
);
767 static void tcg_out_b(TCGContext
*s
, int mask
, tcg_insn_unit
*target
)
769 ptrdiff_t disp
= tcg_pcrel_diff(s
, target
);
770 if (in_range_b(disp
)) {
771 tcg_out32(s
, B
| (disp
& 0x3fffffc) | mask
);
773 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_R0
, (uintptr_t)target
);
774 tcg_out32(s
, MTSPR
| RS(TCG_REG_R0
) | CTR
);
775 tcg_out32(s
, BCCTR
| BO_ALWAYS
| mask
);
779 static void tcg_out_mem_long(TCGContext
*s
, int opi
, int opx
, TCGReg rt
,
780 TCGReg base
, tcg_target_long offset
)
782 tcg_target_long orig
= offset
, l0
, l1
, extra
= 0, align
= 0;
783 bool is_store
= false;
784 TCGReg rs
= TCG_REG_TMP1
;
791 if (rt
!= TCG_REG_R0
) {
799 case STB
: case STH
: case STW
:
804 /* For unaligned, or very large offsets, use the indexed form. */
805 if (offset
& align
|| offset
!= (int32_t)offset
) {
809 tcg_debug_assert(!is_store
|| rs
!= rt
);
810 tcg_out_movi(s
, TCG_TYPE_PTR
, rs
, orig
);
811 tcg_out32(s
, opx
| TAB(rt
, base
, rs
));
815 l0
= (int16_t)offset
;
816 offset
= (offset
- l0
) >> 16;
817 l1
= (int16_t)offset
;
819 if (l1
< 0 && orig
>= 0) {
821 l1
= (int16_t)(offset
- 0x4000);
824 tcg_out32(s
, ADDIS
| TAI(rs
, base
, l1
));
828 tcg_out32(s
, ADDIS
| TAI(rs
, base
, extra
));
831 if (opi
!= ADDI
|| base
!= rt
|| l0
!= 0) {
832 tcg_out32(s
, opi
| TAI(rt
, base
, l0
));
836 static inline void tcg_out_ld(TCGContext
*s
, TCGType type
, TCGReg ret
,
837 TCGReg arg1
, intptr_t arg2
)
841 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
842 if (type
== TCG_TYPE_I32
) {
843 opi
= LWZ
, opx
= LWZX
;
847 tcg_out_mem_long(s
, opi
, opx
, ret
, arg1
, arg2
);
850 static inline void tcg_out_st(TCGContext
*s
, TCGType type
, TCGReg arg
,
851 TCGReg arg1
, intptr_t arg2
)
855 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
856 if (type
== TCG_TYPE_I32
) {
857 opi
= STW
, opx
= STWX
;
859 opi
= STD
, opx
= STDX
;
861 tcg_out_mem_long(s
, opi
, opx
, arg
, arg1
, arg2
);
864 static inline bool tcg_out_sti(TCGContext
*s
, TCGType type
, TCGArg val
,
865 TCGReg base
, intptr_t ofs
)
870 static void tcg_out_cmp(TCGContext
*s
, int cond
, TCGArg arg1
, TCGArg arg2
,
871 int const_arg2
, int cr
, TCGType type
)
876 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
878 /* Simplify the comparisons below wrt CMPI. */
879 if (type
== TCG_TYPE_I32
) {
880 arg2
= (int32_t)arg2
;
887 if ((int16_t) arg2
== arg2
) {
891 } else if ((uint16_t) arg2
== arg2
) {
906 if ((int16_t) arg2
== arg2
) {
921 if ((uint16_t) arg2
== arg2
) {
934 op
|= BF(cr
) | ((type
== TCG_TYPE_I64
) << 21);
937 tcg_out32(s
, op
| RA(arg1
) | (arg2
& 0xffff));
940 tcg_out_movi(s
, type
, TCG_REG_R0
, arg2
);
943 tcg_out32(s
, op
| RA(arg1
) | RB(arg2
));
947 static void tcg_out_setcond_eq0(TCGContext
*s
, TCGType type
,
948 TCGReg dst
, TCGReg src
)
950 if (type
== TCG_TYPE_I32
) {
951 tcg_out32(s
, CNTLZW
| RS(src
) | RA(dst
));
952 tcg_out_shri32(s
, dst
, dst
, 5);
954 tcg_out32(s
, CNTLZD
| RS(src
) | RA(dst
));
955 tcg_out_shri64(s
, dst
, dst
, 6);
959 static void tcg_out_setcond_ne0(TCGContext
*s
, TCGReg dst
, TCGReg src
)
961 /* X != 0 implies X + -1 generates a carry. Extra addition
962 trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C. */
964 tcg_out32(s
, ADDIC
| TAI(dst
, src
, -1));
965 tcg_out32(s
, SUBFE
| TAB(dst
, dst
, src
));
967 tcg_out32(s
, ADDIC
| TAI(TCG_REG_R0
, src
, -1));
968 tcg_out32(s
, SUBFE
| TAB(dst
, TCG_REG_R0
, src
));
972 static TCGReg
tcg_gen_setcond_xor(TCGContext
*s
, TCGReg arg1
, TCGArg arg2
,
976 if ((uint32_t)arg2
== arg2
) {
977 tcg_out_xori32(s
, TCG_REG_R0
, arg1
, arg2
);
979 tcg_out_movi(s
, TCG_TYPE_I64
, TCG_REG_R0
, arg2
);
980 tcg_out32(s
, XOR
| SAB(arg1
, TCG_REG_R0
, TCG_REG_R0
));
983 tcg_out32(s
, XOR
| SAB(arg1
, TCG_REG_R0
, arg2
));
988 static void tcg_out_setcond(TCGContext
*s
, TCGType type
, TCGCond cond
,
989 TCGArg arg0
, TCGArg arg1
, TCGArg arg2
,
994 tcg_debug_assert(TCG_TARGET_REG_BITS
== 64 || type
== TCG_TYPE_I32
);
996 /* Ignore high bits of a potential constant arg2. */
997 if (type
== TCG_TYPE_I32
) {
998 arg2
= (uint32_t)arg2
;
1001 /* Handle common and trivial cases before handling anything else. */
1005 tcg_out_setcond_eq0(s
, type
, arg0
, arg1
);
1008 if (TCG_TARGET_REG_BITS
== 64 && type
== TCG_TYPE_I32
) {
1009 tcg_out_ext32u(s
, TCG_REG_R0
, arg1
);
1012 tcg_out_setcond_ne0(s
, arg0
, arg1
);
1015 tcg_out32(s
, NOR
| SAB(arg1
, arg0
, arg1
));
1019 /* Extract the sign bit. */
1020 if (type
== TCG_TYPE_I32
) {
1021 tcg_out_shri32(s
, arg0
, arg1
, 31);
1023 tcg_out_shri64(s
, arg0
, arg1
, 63);
1031 /* If we have ISEL, we can implement everything with 3 or 4 insns.
1032 All other cases below are also at least 3 insns, so speed up the
1033 code generator by not considering them and always using ISEL. */
1037 tcg_out_cmp(s
, cond
, arg1
, arg2
, const_arg2
, 7, type
);
1039 isel
= tcg_to_isel
[cond
];
1041 tcg_out_movi(s
, type
, arg0
, 1);
1043 /* arg0 = (bc ? 0 : 1) */
1044 tab
= TAB(arg0
, 0, arg0
);
1047 /* arg0 = (bc ? 1 : 0) */
1048 tcg_out_movi(s
, type
, TCG_REG_R0
, 0);
1049 tab
= TAB(arg0
, arg0
, TCG_REG_R0
);
1051 tcg_out32(s
, isel
| tab
);
1057 arg1
= tcg_gen_setcond_xor(s
, arg1
, arg2
, const_arg2
);
1058 tcg_out_setcond_eq0(s
, type
, arg0
, arg1
);
1062 arg1
= tcg_gen_setcond_xor(s
, arg1
, arg2
, const_arg2
);
1063 /* Discard the high bits only once, rather than both inputs. */
1064 if (TCG_TARGET_REG_BITS
== 64 && type
== TCG_TYPE_I32
) {
1065 tcg_out_ext32u(s
, TCG_REG_R0
, arg1
);
1068 tcg_out_setcond_ne0(s
, arg0
, arg1
);
1086 crop
= CRNOR
| BT(7, CR_EQ
) | BA(7, CR_LT
) | BB(7, CR_LT
);
1092 crop
= CRNOR
| BT(7, CR_EQ
) | BA(7, CR_GT
) | BB(7, CR_GT
);
1094 tcg_out_cmp(s
, cond
, arg1
, arg2
, const_arg2
, 7, type
);
1098 tcg_out32(s
, MFOCRF
| RT(TCG_REG_R0
) | FXM(7));
1099 tcg_out_rlw(s
, RLWINM
, arg0
, TCG_REG_R0
, sh
, 31, 31);
1107 static void tcg_out_bc(TCGContext
*s
, int bc
, TCGLabel
*l
)
1110 tcg_out32(s
, bc
| reloc_pc14_val(s
->code_ptr
, l
->u
.value_ptr
));
1112 tcg_out_reloc(s
, s
->code_ptr
, R_PPC_REL14
, l
, 0);
1113 tcg_out_bc_noaddr(s
, bc
);
1117 static void tcg_out_brcond(TCGContext
*s
, TCGCond cond
,
1118 TCGArg arg1
, TCGArg arg2
, int const_arg2
,
1119 TCGLabel
*l
, TCGType type
)
1121 tcg_out_cmp(s
, cond
, arg1
, arg2
, const_arg2
, 7, type
);
1122 tcg_out_bc(s
, tcg_to_bc
[cond
], l
);
1125 static void tcg_out_movcond(TCGContext
*s
, TCGType type
, TCGCond cond
,
1126 TCGArg dest
, TCGArg c1
, TCGArg c2
, TCGArg v1
,
1127 TCGArg v2
, bool const_c2
)
1129 /* If for some reason both inputs are zero, don't produce bad code. */
1130 if (v1
== 0 && v2
== 0) {
1131 tcg_out_movi(s
, type
, dest
, 0);
1135 tcg_out_cmp(s
, cond
, c1
, c2
, const_c2
, 7, type
);
1138 int isel
= tcg_to_isel
[cond
];
1140 /* Swap the V operands if the operation indicates inversion. */
1147 /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand. */
1149 tcg_out_movi(s
, type
, TCG_REG_R0
, 0);
1151 tcg_out32(s
, isel
| TAB(dest
, v1
, v2
));
1154 cond
= tcg_invert_cond(cond
);
1156 } else if (dest
!= v1
) {
1158 tcg_out_movi(s
, type
, dest
, 0);
1160 tcg_out_mov(s
, type
, dest
, v1
);
1163 /* Branch forward over one insn */
1164 tcg_out32(s
, tcg_to_bc
[cond
] | 8);
1166 tcg_out_movi(s
, type
, dest
, 0);
1168 tcg_out_mov(s
, type
, dest
, v2
);
1173 static void tcg_out_cmp2(TCGContext
*s
, const TCGArg
*args
,
1174 const int *const_args
)
1176 static const struct { uint8_t bit1
, bit2
; } bits
[] = {
1177 [TCG_COND_LT
] = { CR_LT
, CR_LT
},
1178 [TCG_COND_LE
] = { CR_LT
, CR_GT
},
1179 [TCG_COND_GT
] = { CR_GT
, CR_GT
},
1180 [TCG_COND_GE
] = { CR_GT
, CR_LT
},
1181 [TCG_COND_LTU
] = { CR_LT
, CR_LT
},
1182 [TCG_COND_LEU
] = { CR_LT
, CR_GT
},
1183 [TCG_COND_GTU
] = { CR_GT
, CR_GT
},
1184 [TCG_COND_GEU
] = { CR_GT
, CR_LT
},
1187 TCGCond cond
= args
[4], cond2
;
1188 TCGArg al
, ah
, bl
, bh
;
1189 int blconst
, bhconst
;
1196 blconst
= const_args
[2];
1197 bhconst
= const_args
[3];
1206 tcg_out_cmp(s
, cond
, al
, bl
, blconst
, 6, TCG_TYPE_I32
);
1207 tcg_out_cmp(s
, cond
, ah
, bh
, bhconst
, 7, TCG_TYPE_I32
);
1208 tcg_out32(s
, op
| BT(7, CR_EQ
) | BA(6, CR_EQ
) | BB(7, CR_EQ
));
1219 bit1
= bits
[cond
].bit1
;
1220 bit2
= bits
[cond
].bit2
;
1221 op
= (bit1
!= bit2
? CRANDC
: CRAND
);
1222 cond2
= tcg_unsigned_cond(cond
);
1224 tcg_out_cmp(s
, cond
, ah
, bh
, bhconst
, 6, TCG_TYPE_I32
);
1225 tcg_out_cmp(s
, cond2
, al
, bl
, blconst
, 7, TCG_TYPE_I32
);
1226 tcg_out32(s
, op
| BT(7, CR_EQ
) | BA(6, CR_EQ
) | BB(7, bit2
));
1227 tcg_out32(s
, CROR
| BT(7, CR_EQ
) | BA(6, bit1
) | BB(7, CR_EQ
));
1235 static void tcg_out_setcond2(TCGContext
*s
, const TCGArg
*args
,
1236 const int *const_args
)
1238 tcg_out_cmp2(s
, args
+ 1, const_args
+ 1);
1239 tcg_out32(s
, MFOCRF
| RT(TCG_REG_R0
) | FXM(7));
1240 tcg_out_rlw(s
, RLWINM
, args
[0], TCG_REG_R0
, 31, 31, 31);
1243 static void tcg_out_brcond2 (TCGContext
*s
, const TCGArg
*args
,
1244 const int *const_args
)
1246 tcg_out_cmp2(s
, args
, const_args
);
1247 tcg_out_bc(s
, BC
| BI(7, CR_EQ
) | BO_COND_TRUE
, arg_label(args
[5]));
1250 static void tcg_out_mb(TCGContext
*s
, TCGArg a0
)
1252 uint32_t insn
= HWSYNC
;
1254 if (a0
== TCG_MO_LD_LD
) {
1256 } else if (a0
== TCG_MO_ST_ST
) {
1262 #ifdef __powerpc64__
1263 void ppc_tb_set_jmp_target(uintptr_t jmp_addr
, uintptr_t addr
)
1265 tcg_insn_unit i1
, i2
;
1267 intptr_t diff
= addr
- jmp_addr
;
1269 if (in_range_b(diff
)) {
1270 i1
= B
| (diff
& 0x3fffffc);
1272 } else if (USE_REG_RA
) {
1274 diff
= addr
- (uintptr_t)tb_ret_addr
;
1276 hi
= (int32_t)(diff
- lo
);
1277 tcg_debug_assert(diff
== hi
+ lo
);
1278 i1
= ADDIS
| TAI(TCG_REG_TMP1
, TCG_REG_RA
, hi
>> 16);
1279 i2
= ADDI
| TAI(TCG_REG_TMP1
, TCG_REG_TMP1
, lo
);
1281 tcg_debug_assert(TCG_TARGET_REG_BITS
== 32 || addr
== (int32_t)addr
);
1282 i1
= ADDIS
| TAI(TCG_REG_TMP1
, 0, addr
>> 16);
1283 i2
= ORI
| SAI(TCG_REG_TMP1
, TCG_REG_TMP1
, addr
);
1285 #ifdef HOST_WORDS_BIGENDIAN
1286 pair
= (uint64_t)i1
<< 32 | i2
;
1288 pair
= (uint64_t)i2
<< 32 | i1
;
1291 atomic_set((uint64_t *)jmp_addr
, pair
);
1292 flush_icache_range(jmp_addr
, jmp_addr
+ 8);
1295 void ppc_tb_set_jmp_target(uintptr_t jmp_addr
, uintptr_t addr
)
1297 intptr_t diff
= addr
- jmp_addr
;
1298 tcg_debug_assert(in_range_b(diff
));
1299 atomic_set((uint32_t *)jmp_addr
, B
| (diff
& 0x3fffffc));
1300 flush_icache_range(jmp_addr
, jmp_addr
+ 4);
1304 static void tcg_out_call(TCGContext
*s
, tcg_insn_unit
*target
)
1307 /* Look through the descriptor. If the branch is in range, and we
1308 don't have to spend too much effort on building the toc. */
1309 void *tgt
= ((void **)target
)[0];
1310 uintptr_t toc
= ((uintptr_t *)target
)[1];
1311 intptr_t diff
= tcg_pcrel_diff(s
, tgt
);
1313 if (in_range_b(diff
) && toc
== (uint32_t)toc
) {
1314 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_TMP1
, toc
);
1315 tcg_out_b(s
, LK
, tgt
);
1317 /* Fold the low bits of the constant into the addresses below. */
1318 intptr_t arg
= (intptr_t)target
;
1319 int ofs
= (int16_t)arg
;
1321 if (ofs
+ 8 < 0x8000) {
1326 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_TMP1
, arg
);
1327 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R0
, TCG_REG_TMP1
, ofs
);
1328 tcg_out32(s
, MTSPR
| RA(TCG_REG_R0
) | CTR
);
1329 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R2
, TCG_REG_TMP1
, ofs
+ SZP
);
1330 tcg_out32(s
, BCCTR
| BO_ALWAYS
| LK
);
1332 #elif defined(_CALL_ELF) && _CALL_ELF == 2
1335 /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1336 address, which the callee uses to compute its TOC address. */
1337 /* FIXME: when the branch is in range, we could avoid r12 load if we
1338 knew that the destination uses the same TOC, and what its local
1339 entry point offset is. */
1340 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_R12
, (intptr_t)target
);
1342 diff
= tcg_pcrel_diff(s
, target
);
1343 if (in_range_b(diff
)) {
1344 tcg_out_b(s
, LK
, target
);
1346 tcg_out32(s
, MTSPR
| RS(TCG_REG_R12
) | CTR
);
1347 tcg_out32(s
, BCCTR
| BO_ALWAYS
| LK
);
1350 tcg_out_b(s
, LK
, target
);
1354 static const uint32_t qemu_ldx_opc
[16] = {
1361 [MO_BSWAP
| MO_UB
] = LBZX
,
1362 [MO_BSWAP
| MO_UW
] = LHBRX
,
1363 [MO_BSWAP
| MO_UL
] = LWBRX
,
1364 [MO_BSWAP
| MO_Q
] = LDBRX
,
1367 static const uint32_t qemu_stx_opc
[16] = {
1372 [MO_BSWAP
| MO_UB
] = STBX
,
1373 [MO_BSWAP
| MO_UW
] = STHBRX
,
1374 [MO_BSWAP
| MO_UL
] = STWBRX
,
1375 [MO_BSWAP
| MO_Q
] = STDBRX
,
1378 static const uint32_t qemu_exts_opc
[4] = {
1379 EXTSB
, EXTSH
, EXTSW
, 0
1382 #if defined (CONFIG_SOFTMMU)
1383 /* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
1384 * int mmu_idx, uintptr_t ra)
1386 static void * const qemu_ld_helpers
[16] = {
1387 [MO_UB
] = helper_ret_ldub_mmu
,
1388 [MO_LEUW
] = helper_le_lduw_mmu
,
1389 [MO_LEUL
] = helper_le_ldul_mmu
,
1390 [MO_LEQ
] = helper_le_ldq_mmu
,
1391 [MO_BEUW
] = helper_be_lduw_mmu
,
1392 [MO_BEUL
] = helper_be_ldul_mmu
,
1393 [MO_BEQ
] = helper_be_ldq_mmu
,
1396 /* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
1397 * uintxx_t val, int mmu_idx, uintptr_t ra)
1399 static void * const qemu_st_helpers
[16] = {
1400 [MO_UB
] = helper_ret_stb_mmu
,
1401 [MO_LEUW
] = helper_le_stw_mmu
,
1402 [MO_LEUL
] = helper_le_stl_mmu
,
1403 [MO_LEQ
] = helper_le_stq_mmu
,
1404 [MO_BEUW
] = helper_be_stw_mmu
,
1405 [MO_BEUL
] = helper_be_stl_mmu
,
1406 [MO_BEQ
] = helper_be_stq_mmu
,
1409 /* Perform the TLB load and compare. Places the result of the comparison
1410 in CR7, loads the addend of the TLB into R3, and returns the register
1411 containing the guest address (zero-extended into R4). Clobbers R0 and R2. */
1413 static TCGReg
tcg_out_tlb_read(TCGContext
*s
, TCGMemOp opc
,
1414 TCGReg addrlo
, TCGReg addrhi
,
1415 int mem_index
, bool is_read
)
1419 ? offsetof(CPUArchState
, tlb_table
[mem_index
][0].addr_read
)
1420 : offsetof(CPUArchState
, tlb_table
[mem_index
][0].addr_write
));
1421 int add_off
= offsetof(CPUArchState
, tlb_table
[mem_index
][0].addend
);
1422 TCGReg base
= TCG_AREG0
;
1423 unsigned s_bits
= opc
& MO_SIZE
;
1424 unsigned a_bits
= get_alignment_bits(opc
);
1426 /* Extract the page index, shifted into place for tlb index. */
1427 if (TCG_TARGET_REG_BITS
== 64) {
1428 if (TARGET_LONG_BITS
== 32) {
1429 /* Zero-extend the address into a place helpful for further use. */
1430 tcg_out_ext32u(s
, TCG_REG_R4
, addrlo
);
1431 addrlo
= TCG_REG_R4
;
1433 tcg_out_rld(s
, RLDICL
, TCG_REG_R3
, addrlo
,
1434 64 - TARGET_PAGE_BITS
, 64 - CPU_TLB_BITS
);
1438 /* Compensate for very large offsets. */
1439 if (add_off
>= 0x8000) {
1440 /* Most target env are smaller than 32k; none are larger than 64k.
1441 Simplify the logic here merely to offset by 0x7ff0, giving us a
1442 range just shy of 64k. Check this assumption. */
1443 QEMU_BUILD_BUG_ON(offsetof(CPUArchState
,
1444 tlb_table
[NB_MMU_MODES
- 1][1])
1446 tcg_out32(s
, ADDI
| TAI(TCG_REG_TMP1
, base
, 0x7ff0));
1447 base
= TCG_REG_TMP1
;
1452 /* Extraction and shifting, part 2. */
1453 if (TCG_TARGET_REG_BITS
== 32 || TARGET_LONG_BITS
== 32) {
1454 tcg_out_rlw(s
, RLWINM
, TCG_REG_R3
, addrlo
,
1455 32 - (TARGET_PAGE_BITS
- CPU_TLB_ENTRY_BITS
),
1456 32 - (CPU_TLB_BITS
+ CPU_TLB_ENTRY_BITS
),
1457 31 - CPU_TLB_ENTRY_BITS
);
1459 tcg_out_shli64(s
, TCG_REG_R3
, TCG_REG_R3
, CPU_TLB_ENTRY_BITS
);
1462 tcg_out32(s
, ADD
| TAB(TCG_REG_R3
, TCG_REG_R3
, base
));
1464 /* Load the tlb comparator. */
1465 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1466 tcg_out_ld(s
, TCG_TYPE_I32
, TCG_REG_R4
, TCG_REG_R3
, cmp_off
);
1467 tcg_out_ld(s
, TCG_TYPE_I32
, TCG_REG_TMP1
, TCG_REG_R3
, cmp_off
+ 4);
1469 tcg_out_ld(s
, TCG_TYPE_TL
, TCG_REG_TMP1
, TCG_REG_R3
, cmp_off
);
1472 /* Load the TLB addend for use on the fast path. Do this asap
1473 to minimize any load use delay. */
1474 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R3
, TCG_REG_R3
, add_off
);
1476 /* Clear the non-page, non-alignment bits from the address */
1477 if (TCG_TARGET_REG_BITS
== 32) {
1478 /* We don't support unaligned accesses on 32-bits.
1479 * Preserve the bottom bits and thus trigger a comparison
1480 * failure on unaligned accesses.
1482 if (a_bits
< s_bits
) {
1485 tcg_out_rlw(s
, RLWINM
, TCG_REG_R0
, addrlo
, 0,
1486 (32 - a_bits
) & 31, 31 - TARGET_PAGE_BITS
);
1490 /* If the access is unaligned, we need to make sure we fail if we
1491 * cross a page boundary. The trick is to add the access size-1
1492 * to the address before masking the low bits. That will make the
1493 * address overflow to the next page if we cross a page boundary,
1494 * which will then force a mismatch of the TLB compare.
1496 if (a_bits
< s_bits
) {
1497 unsigned a_mask
= (1 << a_bits
) - 1;
1498 unsigned s_mask
= (1 << s_bits
) - 1;
1499 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, t
, s_mask
- a_mask
));
1503 /* Mask the address for the requested alignment. */
1504 if (TARGET_LONG_BITS
== 32) {
1505 tcg_out_rlw(s
, RLWINM
, TCG_REG_R0
, t
, 0,
1506 (32 - a_bits
) & 31, 31 - TARGET_PAGE_BITS
);
1507 } else if (a_bits
== 0) {
1508 tcg_out_rld(s
, RLDICR
, TCG_REG_R0
, t
, 0, 63 - TARGET_PAGE_BITS
);
1510 tcg_out_rld(s
, RLDICL
, TCG_REG_R0
, t
,
1511 64 - TARGET_PAGE_BITS
, TARGET_PAGE_BITS
- a_bits
);
1512 tcg_out_rld(s
, RLDICL
, TCG_REG_R0
, TCG_REG_R0
, TARGET_PAGE_BITS
, 0);
1516 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1517 tcg_out_cmp(s
, TCG_COND_EQ
, TCG_REG_R0
, TCG_REG_TMP1
,
1518 0, 7, TCG_TYPE_I32
);
1519 tcg_out_cmp(s
, TCG_COND_EQ
, addrhi
, TCG_REG_R4
, 0, 6, TCG_TYPE_I32
);
1520 tcg_out32(s
, CRAND
| BT(7, CR_EQ
) | BA(6, CR_EQ
) | BB(7, CR_EQ
));
1522 tcg_out_cmp(s
, TCG_COND_EQ
, TCG_REG_R0
, TCG_REG_TMP1
,
1529 /* Record the context of a call to the out of line helper code for the slow
1530 path for a load or store, so that we can later generate the correct
1532 static void add_qemu_ldst_label(TCGContext
*s
, bool is_ld
, TCGMemOpIdx oi
,
1533 TCGReg datalo_reg
, TCGReg datahi_reg
,
1534 TCGReg addrlo_reg
, TCGReg addrhi_reg
,
1535 tcg_insn_unit
*raddr
, tcg_insn_unit
*lptr
)
1537 TCGLabelQemuLdst
*label
= new_ldst_label(s
);
1539 label
->is_ld
= is_ld
;
1541 label
->datalo_reg
= datalo_reg
;
1542 label
->datahi_reg
= datahi_reg
;
1543 label
->addrlo_reg
= addrlo_reg
;
1544 label
->addrhi_reg
= addrhi_reg
;
1545 label
->raddr
= raddr
;
1546 label
->label_ptr
[0] = lptr
;
1549 static void tcg_out_qemu_ld_slow_path(TCGContext
*s
, TCGLabelQemuLdst
*lb
)
1551 TCGMemOpIdx oi
= lb
->oi
;
1552 TCGMemOp opc
= get_memop(oi
);
1553 TCGReg hi
, lo
, arg
= TCG_REG_R3
;
1555 reloc_pc14(lb
->label_ptr
[0], s
->code_ptr
);
1557 tcg_out_mov(s
, TCG_TYPE_PTR
, arg
++, TCG_AREG0
);
1559 lo
= lb
->addrlo_reg
;
1560 hi
= lb
->addrhi_reg
;
1561 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1562 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1565 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, hi
);
1566 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, lo
);
1568 /* If the address needed to be zero-extended, we'll have already
1569 placed it in R4. The only remaining case is 64-bit guest. */
1570 tcg_out_mov(s
, TCG_TYPE_TL
, arg
++, lo
);
1573 tcg_out_movi(s
, TCG_TYPE_I32
, arg
++, oi
);
1574 tcg_out32(s
, MFSPR
| RT(arg
) | LR
);
1576 tcg_out_call(s
, qemu_ld_helpers
[opc
& (MO_BSWAP
| MO_SIZE
)]);
1578 lo
= lb
->datalo_reg
;
1579 hi
= lb
->datahi_reg
;
1580 if (TCG_TARGET_REG_BITS
== 32 && (opc
& MO_SIZE
) == MO_64
) {
1581 tcg_out_mov(s
, TCG_TYPE_I32
, lo
, TCG_REG_R4
);
1582 tcg_out_mov(s
, TCG_TYPE_I32
, hi
, TCG_REG_R3
);
1583 } else if (opc
& MO_SIGN
) {
1584 uint32_t insn
= qemu_exts_opc
[opc
& MO_SIZE
];
1585 tcg_out32(s
, insn
| RA(lo
) | RS(TCG_REG_R3
));
1587 tcg_out_mov(s
, TCG_TYPE_REG
, lo
, TCG_REG_R3
);
1590 tcg_out_b(s
, 0, lb
->raddr
);
1593 static void tcg_out_qemu_st_slow_path(TCGContext
*s
, TCGLabelQemuLdst
*lb
)
1595 TCGMemOpIdx oi
= lb
->oi
;
1596 TCGMemOp opc
= get_memop(oi
);
1597 TCGMemOp s_bits
= opc
& MO_SIZE
;
1598 TCGReg hi
, lo
, arg
= TCG_REG_R3
;
1600 reloc_pc14(lb
->label_ptr
[0], s
->code_ptr
);
1602 tcg_out_mov(s
, TCG_TYPE_PTR
, arg
++, TCG_AREG0
);
1604 lo
= lb
->addrlo_reg
;
1605 hi
= lb
->addrhi_reg
;
1606 if (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
) {
1607 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1610 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, hi
);
1611 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, lo
);
1613 /* If the address needed to be zero-extended, we'll have already
1614 placed it in R4. The only remaining case is 64-bit guest. */
1615 tcg_out_mov(s
, TCG_TYPE_TL
, arg
++, lo
);
1618 lo
= lb
->datalo_reg
;
1619 hi
= lb
->datahi_reg
;
1620 if (TCG_TARGET_REG_BITS
== 32) {
1623 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1626 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, hi
);
1629 tcg_out_mov(s
, TCG_TYPE_I32
, arg
++, lo
);
1632 tcg_out_rlw(s
, RLWINM
, arg
++, lo
, 0, 32 - (8 << s_bits
), 31);
1636 if (s_bits
== MO_64
) {
1637 tcg_out_mov(s
, TCG_TYPE_I64
, arg
++, lo
);
1639 tcg_out_rld(s
, RLDICL
, arg
++, lo
, 0, 64 - (8 << s_bits
));
1643 tcg_out_movi(s
, TCG_TYPE_I32
, arg
++, oi
);
1644 tcg_out32(s
, MFSPR
| RT(arg
) | LR
);
1646 tcg_out_call(s
, qemu_st_helpers
[opc
& (MO_BSWAP
| MO_SIZE
)]);
1648 tcg_out_b(s
, 0, lb
->raddr
);
1650 #endif /* SOFTMMU */
1652 static void tcg_out_qemu_ld(TCGContext
*s
, const TCGArg
*args
, bool is_64
)
1654 TCGReg datalo
, datahi
, addrlo
, rbase
;
1655 TCGReg addrhi
__attribute__((unused
));
1657 TCGMemOp opc
, s_bits
;
1658 #ifdef CONFIG_SOFTMMU
1660 tcg_insn_unit
*label_ptr
;
1664 datahi
= (TCG_TARGET_REG_BITS
== 32 && is_64
? *args
++ : 0);
1666 addrhi
= (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
? *args
++ : 0);
1668 opc
= get_memop(oi
);
1669 s_bits
= opc
& MO_SIZE
;
1671 #ifdef CONFIG_SOFTMMU
1672 mem_index
= get_mmuidx(oi
);
1673 addrlo
= tcg_out_tlb_read(s
, opc
, addrlo
, addrhi
, mem_index
, true);
1675 /* Load a pointer into the current opcode w/conditional branch-link. */
1676 label_ptr
= s
->code_ptr
;
1677 tcg_out_bc_noaddr(s
, BC
| BI(7, CR_EQ
) | BO_COND_FALSE
| LK
);
1680 #else /* !CONFIG_SOFTMMU */
1681 rbase
= guest_base
? TCG_GUEST_BASE_REG
: 0;
1682 if (TCG_TARGET_REG_BITS
> TARGET_LONG_BITS
) {
1683 tcg_out_ext32u(s
, TCG_REG_TMP1
, addrlo
);
1684 addrlo
= TCG_REG_TMP1
;
1688 if (TCG_TARGET_REG_BITS
== 32 && s_bits
== MO_64
) {
1689 if (opc
& MO_BSWAP
) {
1690 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1691 tcg_out32(s
, LWBRX
| TAB(datalo
, rbase
, addrlo
));
1692 tcg_out32(s
, LWBRX
| TAB(datahi
, rbase
, TCG_REG_R0
));
1693 } else if (rbase
!= 0) {
1694 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1695 tcg_out32(s
, LWZX
| TAB(datahi
, rbase
, addrlo
));
1696 tcg_out32(s
, LWZX
| TAB(datalo
, rbase
, TCG_REG_R0
));
1697 } else if (addrlo
== datahi
) {
1698 tcg_out32(s
, LWZ
| TAI(datalo
, addrlo
, 4));
1699 tcg_out32(s
, LWZ
| TAI(datahi
, addrlo
, 0));
1701 tcg_out32(s
, LWZ
| TAI(datahi
, addrlo
, 0));
1702 tcg_out32(s
, LWZ
| TAI(datalo
, addrlo
, 4));
1705 uint32_t insn
= qemu_ldx_opc
[opc
& (MO_BSWAP
| MO_SSIZE
)];
1706 if (!HAVE_ISA_2_06
&& insn
== LDBRX
) {
1707 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1708 tcg_out32(s
, LWBRX
| TAB(datalo
, rbase
, addrlo
));
1709 tcg_out32(s
, LWBRX
| TAB(TCG_REG_R0
, rbase
, TCG_REG_R0
));
1710 tcg_out_rld(s
, RLDIMI
, datalo
, TCG_REG_R0
, 32, 0);
1712 tcg_out32(s
, insn
| TAB(datalo
, rbase
, addrlo
));
1714 insn
= qemu_ldx_opc
[opc
& (MO_SIZE
| MO_BSWAP
)];
1715 tcg_out32(s
, insn
| TAB(datalo
, rbase
, addrlo
));
1716 insn
= qemu_exts_opc
[s_bits
];
1717 tcg_out32(s
, insn
| RA(datalo
) | RS(datalo
));
1721 #ifdef CONFIG_SOFTMMU
1722 add_qemu_ldst_label(s
, true, oi
, datalo
, datahi
, addrlo
, addrhi
,
1723 s
->code_ptr
, label_ptr
);
1727 static void tcg_out_qemu_st(TCGContext
*s
, const TCGArg
*args
, bool is_64
)
1729 TCGReg datalo
, datahi
, addrlo
, rbase
;
1730 TCGReg addrhi
__attribute__((unused
));
1732 TCGMemOp opc
, s_bits
;
1733 #ifdef CONFIG_SOFTMMU
1735 tcg_insn_unit
*label_ptr
;
1739 datahi
= (TCG_TARGET_REG_BITS
== 32 && is_64
? *args
++ : 0);
1741 addrhi
= (TCG_TARGET_REG_BITS
< TARGET_LONG_BITS
? *args
++ : 0);
1743 opc
= get_memop(oi
);
1744 s_bits
= opc
& MO_SIZE
;
1746 #ifdef CONFIG_SOFTMMU
1747 mem_index
= get_mmuidx(oi
);
1748 addrlo
= tcg_out_tlb_read(s
, opc
, addrlo
, addrhi
, mem_index
, false);
1750 /* Load a pointer into the current opcode w/conditional branch-link. */
1751 label_ptr
= s
->code_ptr
;
1752 tcg_out_bc_noaddr(s
, BC
| BI(7, CR_EQ
) | BO_COND_FALSE
| LK
);
1755 #else /* !CONFIG_SOFTMMU */
1756 rbase
= guest_base
? TCG_GUEST_BASE_REG
: 0;
1757 if (TCG_TARGET_REG_BITS
> TARGET_LONG_BITS
) {
1758 tcg_out_ext32u(s
, TCG_REG_TMP1
, addrlo
);
1759 addrlo
= TCG_REG_TMP1
;
1763 if (TCG_TARGET_REG_BITS
== 32 && s_bits
== MO_64
) {
1764 if (opc
& MO_BSWAP
) {
1765 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1766 tcg_out32(s
, STWBRX
| SAB(datalo
, rbase
, addrlo
));
1767 tcg_out32(s
, STWBRX
| SAB(datahi
, rbase
, TCG_REG_R0
));
1768 } else if (rbase
!= 0) {
1769 tcg_out32(s
, ADDI
| TAI(TCG_REG_R0
, addrlo
, 4));
1770 tcg_out32(s
, STWX
| SAB(datahi
, rbase
, addrlo
));
1771 tcg_out32(s
, STWX
| SAB(datalo
, rbase
, TCG_REG_R0
));
1773 tcg_out32(s
, STW
| TAI(datahi
, addrlo
, 0));
1774 tcg_out32(s
, STW
| TAI(datalo
, addrlo
, 4));
1777 uint32_t insn
= qemu_stx_opc
[opc
& (MO_BSWAP
| MO_SIZE
)];
1778 if (!HAVE_ISA_2_06
&& insn
== STDBRX
) {
1779 tcg_out32(s
, STWBRX
| SAB(datalo
, rbase
, addrlo
));
1780 tcg_out32(s
, ADDI
| TAI(TCG_REG_TMP1
, addrlo
, 4));
1781 tcg_out_shri64(s
, TCG_REG_R0
, datalo
, 32);
1782 tcg_out32(s
, STWBRX
| SAB(TCG_REG_R0
, rbase
, TCG_REG_TMP1
));
1784 tcg_out32(s
, insn
| SAB(datalo
, rbase
, addrlo
));
1788 #ifdef CONFIG_SOFTMMU
1789 add_qemu_ldst_label(s
, false, oi
, datalo
, datahi
, addrlo
, addrhi
,
1790 s
->code_ptr
, label_ptr
);
1794 /* Parameters for function call generation, used in tcg.c. */
1795 #define TCG_TARGET_STACK_ALIGN 16
1796 #define TCG_TARGET_EXTEND_ARGS 1
1799 # define LINK_AREA_SIZE (6 * SZR)
1800 # define LR_OFFSET (1 * SZR)
1801 # define TCG_TARGET_CALL_STACK_OFFSET (LINK_AREA_SIZE + 8 * SZR)
1802 #elif defined(TCG_TARGET_CALL_DARWIN)
1803 # define LINK_AREA_SIZE (6 * SZR)
1804 # define LR_OFFSET (2 * SZR)
1805 #elif TCG_TARGET_REG_BITS == 64
1806 # if defined(_CALL_ELF) && _CALL_ELF == 2
1807 # define LINK_AREA_SIZE (4 * SZR)
1808 # define LR_OFFSET (1 * SZR)
1810 #else /* TCG_TARGET_REG_BITS == 32 */
1811 # if defined(_CALL_SYSV)
1812 # define LINK_AREA_SIZE (2 * SZR)
1813 # define LR_OFFSET (1 * SZR)
1817 # error "Unhandled abi"
1819 #ifndef TCG_TARGET_CALL_STACK_OFFSET
1820 # define TCG_TARGET_CALL_STACK_OFFSET LINK_AREA_SIZE
1823 #define CPU_TEMP_BUF_SIZE (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
1824 #define REG_SAVE_SIZE ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
1826 #define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET \
1827 + TCG_STATIC_CALL_ARGS_SIZE \
1828 + CPU_TEMP_BUF_SIZE \
1830 + TCG_TARGET_STACK_ALIGN - 1) \
1831 & -TCG_TARGET_STACK_ALIGN)
1833 #define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
1835 static void tcg_target_qemu_prologue(TCGContext
*s
)
1840 void **desc
= (void **)s
->code_ptr
;
1841 desc
[0] = desc
+ 2; /* entry point */
1842 desc
[1] = 0; /* environment pointer */
1843 s
->code_ptr
= (void *)(desc
+ 2); /* skip over descriptor */
1846 tcg_set_frame(s
, TCG_REG_CALL_STACK
, REG_SAVE_BOT
- CPU_TEMP_BUF_SIZE
,
1850 tcg_out32(s
, MFSPR
| RT(TCG_REG_R0
) | LR
);
1851 tcg_out32(s
, (SZR
== 8 ? STDU
: STWU
)
1852 | SAI(TCG_REG_R1
, TCG_REG_R1
, -FRAME_SIZE
));
1854 for (i
= 0; i
< ARRAY_SIZE(tcg_target_callee_save_regs
); ++i
) {
1855 tcg_out_st(s
, TCG_TYPE_REG
, tcg_target_callee_save_regs
[i
],
1856 TCG_REG_R1
, REG_SAVE_BOT
+ i
* SZR
);
1858 tcg_out_st(s
, TCG_TYPE_PTR
, TCG_REG_R0
, TCG_REG_R1
, FRAME_SIZE
+LR_OFFSET
);
1860 #ifndef CONFIG_SOFTMMU
1862 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_GUEST_BASE_REG
, guest_base
);
1863 tcg_regset_set_reg(s
->reserved_regs
, TCG_GUEST_BASE_REG
);
1867 tcg_out_mov(s
, TCG_TYPE_PTR
, TCG_AREG0
, tcg_target_call_iarg_regs
[0]);
1868 tcg_out32(s
, MTSPR
| RS(tcg_target_call_iarg_regs
[1]) | CTR
);
1872 /* Make the caller load the value as the TOC into R2. */
1873 tb_ret_addr
= s
->code_ptr
+ 2;
1874 desc
[1] = tb_ret_addr
;
1875 tcg_out_mov(s
, TCG_TYPE_PTR
, TCG_REG_RA
, TCG_REG_R2
);
1876 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1877 #elif defined(_CALL_ELF) && _CALL_ELF == 2
1878 /* Compute from the incoming R12 value. */
1879 tb_ret_addr
= s
->code_ptr
+ 2;
1880 tcg_out32(s
, ADDI
| TAI(TCG_REG_RA
, TCG_REG_R12
,
1881 tcg_ptr_byte_diff(tb_ret_addr
, s
->code_buf
)));
1882 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1884 /* Reserve max 5 insns for the constant load. */
1885 tb_ret_addr
= s
->code_ptr
+ 6;
1886 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_RA
, (intptr_t)tb_ret_addr
);
1887 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1888 while (s
->code_ptr
< tb_ret_addr
) {
1893 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1894 tb_ret_addr
= s
->code_ptr
;
1898 tcg_debug_assert(tb_ret_addr
== s
->code_ptr
);
1900 tcg_out_ld(s
, TCG_TYPE_PTR
, TCG_REG_R0
, TCG_REG_R1
, FRAME_SIZE
+LR_OFFSET
);
1901 for (i
= 0; i
< ARRAY_SIZE(tcg_target_callee_save_regs
); ++i
) {
1902 tcg_out_ld(s
, TCG_TYPE_REG
, tcg_target_callee_save_regs
[i
],
1903 TCG_REG_R1
, REG_SAVE_BOT
+ i
* SZR
);
1905 tcg_out32(s
, MTSPR
| RS(TCG_REG_R0
) | LR
);
1906 tcg_out32(s
, ADDI
| TAI(TCG_REG_R1
, TCG_REG_R1
, FRAME_SIZE
));
1907 tcg_out32(s
, BCLR
| BO_ALWAYS
);
1910 static void tcg_out_op(TCGContext
*s
, TCGOpcode opc
, const TCGArg
*args
,
1911 const int *const_args
)
1917 case INDEX_op_exit_tb
:
1919 ptrdiff_t disp
= tcg_pcrel_diff(s
, tb_ret_addr
);
1921 /* Use a direct branch if we can, otherwise use the value in RA.
1922 Note that the direct branch is always backward, thus we need
1923 to account for the possibility of 5 insns from the movi. */
1924 if (!in_range_b(disp
- 20)) {
1925 tcg_out32(s
, MTSPR
| RS(TCG_REG_RA
) | CTR
);
1926 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_R3
, args
[0]);
1927 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1931 tcg_out_movi(s
, TCG_TYPE_PTR
, TCG_REG_R3
, args
[0]);
1932 tcg_out_b(s
, 0, tb_ret_addr
);
1934 case INDEX_op_goto_tb
:
1935 tcg_debug_assert(s
->tb_jmp_insn_offset
);
1937 #ifdef __powerpc64__
1938 /* Ensure the next insns are 8-byte aligned. */
1939 if ((uintptr_t)s
->code_ptr
& 7) {
1942 s
->tb_jmp_insn_offset
[args
[0]] = tcg_current_code_size(s
);
1943 /* To be replaced by either a branch+nop or a load into TMP1. */
1945 tcg_out32(s
, MTSPR
| RS(TCG_REG_TMP1
) | CTR
);
1946 tcg_out32(s
, BCCTR
| BO_ALWAYS
);
1948 /* To be replaced by a branch. */
1951 s
->tb_jmp_reset_offset
[args
[0]] = tcg_current_code_size(s
);
1955 TCGLabel
*l
= arg_label(args
[0]);
1958 tcg_out_b(s
, 0, l
->u
.value_ptr
);
1960 tcg_out_reloc(s
, s
->code_ptr
, R_PPC_REL24
, l
, 0);
1961 tcg_out_b_noaddr(s
, B
);
1965 case INDEX_op_ld8u_i32
:
1966 case INDEX_op_ld8u_i64
:
1967 tcg_out_mem_long(s
, LBZ
, LBZX
, args
[0], args
[1], args
[2]);
1969 case INDEX_op_ld8s_i32
:
1970 case INDEX_op_ld8s_i64
:
1971 tcg_out_mem_long(s
, LBZ
, LBZX
, args
[0], args
[1], args
[2]);
1972 tcg_out32(s
, EXTSB
| RS(args
[0]) | RA(args
[0]));
1974 case INDEX_op_ld16u_i32
:
1975 case INDEX_op_ld16u_i64
:
1976 tcg_out_mem_long(s
, LHZ
, LHZX
, args
[0], args
[1], args
[2]);
1978 case INDEX_op_ld16s_i32
:
1979 case INDEX_op_ld16s_i64
:
1980 tcg_out_mem_long(s
, LHA
, LHAX
, args
[0], args
[1], args
[2]);
1982 case INDEX_op_ld_i32
:
1983 case INDEX_op_ld32u_i64
:
1984 tcg_out_mem_long(s
, LWZ
, LWZX
, args
[0], args
[1], args
[2]);
1986 case INDEX_op_ld32s_i64
:
1987 tcg_out_mem_long(s
, LWA
, LWAX
, args
[0], args
[1], args
[2]);
1989 case INDEX_op_ld_i64
:
1990 tcg_out_mem_long(s
, LD
, LDX
, args
[0], args
[1], args
[2]);
1992 case INDEX_op_st8_i32
:
1993 case INDEX_op_st8_i64
:
1994 tcg_out_mem_long(s
, STB
, STBX
, args
[0], args
[1], args
[2]);
1996 case INDEX_op_st16_i32
:
1997 case INDEX_op_st16_i64
:
1998 tcg_out_mem_long(s
, STH
, STHX
, args
[0], args
[1], args
[2]);
2000 case INDEX_op_st_i32
:
2001 case INDEX_op_st32_i64
:
2002 tcg_out_mem_long(s
, STW
, STWX
, args
[0], args
[1], args
[2]);
2004 case INDEX_op_st_i64
:
2005 tcg_out_mem_long(s
, STD
, STDX
, args
[0], args
[1], args
[2]);
2008 case INDEX_op_add_i32
:
2009 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2010 if (const_args
[2]) {
2012 tcg_out_mem_long(s
, ADDI
, ADD
, a0
, a1
, (int32_t)a2
);
2014 tcg_out32(s
, ADD
| TAB(a0
, a1
, a2
));
2017 case INDEX_op_sub_i32
:
2018 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2019 if (const_args
[1]) {
2020 if (const_args
[2]) {
2021 tcg_out_movi(s
, TCG_TYPE_I32
, a0
, a1
- a2
);
2023 tcg_out32(s
, SUBFIC
| TAI(a0
, a2
, a1
));
2025 } else if (const_args
[2]) {
2029 tcg_out32(s
, SUBF
| TAB(a0
, a2
, a1
));
2033 case INDEX_op_and_i32
:
2034 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2035 if (const_args
[2]) {
2036 tcg_out_andi32(s
, a0
, a1
, a2
);
2038 tcg_out32(s
, AND
| SAB(a1
, a0
, a2
));
2041 case INDEX_op_and_i64
:
2042 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2043 if (const_args
[2]) {
2044 tcg_out_andi64(s
, a0
, a1
, a2
);
2046 tcg_out32(s
, AND
| SAB(a1
, a0
, a2
));
2049 case INDEX_op_or_i64
:
2050 case INDEX_op_or_i32
:
2051 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2052 if (const_args
[2]) {
2053 tcg_out_ori32(s
, a0
, a1
, a2
);
2055 tcg_out32(s
, OR
| SAB(a1
, a0
, a2
));
2058 case INDEX_op_xor_i64
:
2059 case INDEX_op_xor_i32
:
2060 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2061 if (const_args
[2]) {
2062 tcg_out_xori32(s
, a0
, a1
, a2
);
2064 tcg_out32(s
, XOR
| SAB(a1
, a0
, a2
));
2067 case INDEX_op_andc_i32
:
2068 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2069 if (const_args
[2]) {
2070 tcg_out_andi32(s
, a0
, a1
, ~a2
);
2072 tcg_out32(s
, ANDC
| SAB(a1
, a0
, a2
));
2075 case INDEX_op_andc_i64
:
2076 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2077 if (const_args
[2]) {
2078 tcg_out_andi64(s
, a0
, a1
, ~a2
);
2080 tcg_out32(s
, ANDC
| SAB(a1
, a0
, a2
));
2083 case INDEX_op_orc_i32
:
2084 if (const_args
[2]) {
2085 tcg_out_ori32(s
, args
[0], args
[1], ~args
[2]);
2089 case INDEX_op_orc_i64
:
2090 tcg_out32(s
, ORC
| SAB(args
[1], args
[0], args
[2]));
2092 case INDEX_op_eqv_i32
:
2093 if (const_args
[2]) {
2094 tcg_out_xori32(s
, args
[0], args
[1], ~args
[2]);
2098 case INDEX_op_eqv_i64
:
2099 tcg_out32(s
, EQV
| SAB(args
[1], args
[0], args
[2]));
2101 case INDEX_op_nand_i32
:
2102 case INDEX_op_nand_i64
:
2103 tcg_out32(s
, NAND
| SAB(args
[1], args
[0], args
[2]));
2105 case INDEX_op_nor_i32
:
2106 case INDEX_op_nor_i64
:
2107 tcg_out32(s
, NOR
| SAB(args
[1], args
[0], args
[2]));
2110 case INDEX_op_mul_i32
:
2111 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2112 if (const_args
[2]) {
2113 tcg_out32(s
, MULLI
| TAI(a0
, a1
, a2
));
2115 tcg_out32(s
, MULLW
| TAB(a0
, a1
, a2
));
2119 case INDEX_op_div_i32
:
2120 tcg_out32(s
, DIVW
| TAB(args
[0], args
[1], args
[2]));
2123 case INDEX_op_divu_i32
:
2124 tcg_out32(s
, DIVWU
| TAB(args
[0], args
[1], args
[2]));
2127 case INDEX_op_shl_i32
:
2128 if (const_args
[2]) {
2129 tcg_out_shli32(s
, args
[0], args
[1], args
[2]);
2131 tcg_out32(s
, SLW
| SAB(args
[1], args
[0], args
[2]));
2134 case INDEX_op_shr_i32
:
2135 if (const_args
[2]) {
2136 tcg_out_shri32(s
, args
[0], args
[1], args
[2]);
2138 tcg_out32(s
, SRW
| SAB(args
[1], args
[0], args
[2]));
2141 case INDEX_op_sar_i32
:
2142 if (const_args
[2]) {
2143 tcg_out32(s
, SRAWI
| RS(args
[1]) | RA(args
[0]) | SH(args
[2]));
2145 tcg_out32(s
, SRAW
| SAB(args
[1], args
[0], args
[2]));
2148 case INDEX_op_rotl_i32
:
2149 if (const_args
[2]) {
2150 tcg_out_rlw(s
, RLWINM
, args
[0], args
[1], args
[2], 0, 31);
2152 tcg_out32(s
, RLWNM
| SAB(args
[1], args
[0], args
[2])
2156 case INDEX_op_rotr_i32
:
2157 if (const_args
[2]) {
2158 tcg_out_rlw(s
, RLWINM
, args
[0], args
[1], 32 - args
[2], 0, 31);
2160 tcg_out32(s
, SUBFIC
| TAI(TCG_REG_R0
, args
[2], 32));
2161 tcg_out32(s
, RLWNM
| SAB(args
[1], args
[0], TCG_REG_R0
)
2166 case INDEX_op_brcond_i32
:
2167 tcg_out_brcond(s
, args
[2], args
[0], args
[1], const_args
[1],
2168 arg_label(args
[3]), TCG_TYPE_I32
);
2170 case INDEX_op_brcond_i64
:
2171 tcg_out_brcond(s
, args
[2], args
[0], args
[1], const_args
[1],
2172 arg_label(args
[3]), TCG_TYPE_I64
);
2174 case INDEX_op_brcond2_i32
:
2175 tcg_out_brcond2(s
, args
, const_args
);
2178 case INDEX_op_neg_i32
:
2179 case INDEX_op_neg_i64
:
2180 tcg_out32(s
, NEG
| RT(args
[0]) | RA(args
[1]));
2183 case INDEX_op_not_i32
:
2184 case INDEX_op_not_i64
:
2185 tcg_out32(s
, NOR
| SAB(args
[1], args
[0], args
[1]));
2188 case INDEX_op_add_i64
:
2189 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2190 if (const_args
[2]) {
2192 tcg_out_mem_long(s
, ADDI
, ADD
, a0
, a1
, a2
);
2194 tcg_out32(s
, ADD
| TAB(a0
, a1
, a2
));
2197 case INDEX_op_sub_i64
:
2198 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2199 if (const_args
[1]) {
2200 if (const_args
[2]) {
2201 tcg_out_movi(s
, TCG_TYPE_I64
, a0
, a1
- a2
);
2203 tcg_out32(s
, SUBFIC
| TAI(a0
, a2
, a1
));
2205 } else if (const_args
[2]) {
2209 tcg_out32(s
, SUBF
| TAB(a0
, a2
, a1
));
2213 case INDEX_op_shl_i64
:
2214 if (const_args
[2]) {
2215 tcg_out_shli64(s
, args
[0], args
[1], args
[2]);
2217 tcg_out32(s
, SLD
| SAB(args
[1], args
[0], args
[2]));
2220 case INDEX_op_shr_i64
:
2221 if (const_args
[2]) {
2222 tcg_out_shri64(s
, args
[0], args
[1], args
[2]);
2224 tcg_out32(s
, SRD
| SAB(args
[1], args
[0], args
[2]));
2227 case INDEX_op_sar_i64
:
2228 if (const_args
[2]) {
2229 int sh
= SH(args
[2] & 0x1f) | (((args
[2] >> 5) & 1) << 1);
2230 tcg_out32(s
, SRADI
| RA(args
[0]) | RS(args
[1]) | sh
);
2232 tcg_out32(s
, SRAD
| SAB(args
[1], args
[0], args
[2]));
2235 case INDEX_op_rotl_i64
:
2236 if (const_args
[2]) {
2237 tcg_out_rld(s
, RLDICL
, args
[0], args
[1], args
[2], 0);
2239 tcg_out32(s
, RLDCL
| SAB(args
[1], args
[0], args
[2]) | MB64(0));
2242 case INDEX_op_rotr_i64
:
2243 if (const_args
[2]) {
2244 tcg_out_rld(s
, RLDICL
, args
[0], args
[1], 64 - args
[2], 0);
2246 tcg_out32(s
, SUBFIC
| TAI(TCG_REG_R0
, args
[2], 64));
2247 tcg_out32(s
, RLDCL
| SAB(args
[1], args
[0], TCG_REG_R0
) | MB64(0));
2251 case INDEX_op_mul_i64
:
2252 a0
= args
[0], a1
= args
[1], a2
= args
[2];
2253 if (const_args
[2]) {
2254 tcg_out32(s
, MULLI
| TAI(a0
, a1
, a2
));
2256 tcg_out32(s
, MULLD
| TAB(a0
, a1
, a2
));
2259 case INDEX_op_div_i64
:
2260 tcg_out32(s
, DIVD
| TAB(args
[0], args
[1], args
[2]));
2262 case INDEX_op_divu_i64
:
2263 tcg_out32(s
, DIVDU
| TAB(args
[0], args
[1], args
[2]));
2266 case INDEX_op_qemu_ld_i32
:
2267 tcg_out_qemu_ld(s
, args
, false);
2269 case INDEX_op_qemu_ld_i64
:
2270 tcg_out_qemu_ld(s
, args
, true);
2272 case INDEX_op_qemu_st_i32
:
2273 tcg_out_qemu_st(s
, args
, false);
2275 case INDEX_op_qemu_st_i64
:
2276 tcg_out_qemu_st(s
, args
, true);
2279 case INDEX_op_ext8s_i32
:
2280 case INDEX_op_ext8s_i64
:
2283 case INDEX_op_ext16s_i32
:
2284 case INDEX_op_ext16s_i64
:
2287 case INDEX_op_ext_i32_i64
:
2288 case INDEX_op_ext32s_i64
:
2292 tcg_out32(s
, c
| RS(args
[1]) | RA(args
[0]));
2294 case INDEX_op_extu_i32_i64
:
2295 tcg_out_ext32u(s
, args
[0], args
[1]);
2298 case INDEX_op_setcond_i32
:
2299 tcg_out_setcond(s
, TCG_TYPE_I32
, args
[3], args
[0], args
[1], args
[2],
2302 case INDEX_op_setcond_i64
:
2303 tcg_out_setcond(s
, TCG_TYPE_I64
, args
[3], args
[0], args
[1], args
[2],
2306 case INDEX_op_setcond2_i32
:
2307 tcg_out_setcond2(s
, args
, const_args
);
2310 case INDEX_op_bswap16_i32
:
2311 case INDEX_op_bswap16_i64
:
2312 a0
= args
[0], a1
= args
[1];
2315 /* a0 = (a1 r<< 24) & 0xff # 000c */
2316 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 24, 24, 31);
2317 /* a0 = (a0 & ~0xff00) | (a1 r<< 8) & 0xff00 # 00dc */
2318 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 8, 16, 23);
2320 /* r0 = (a1 r<< 8) & 0xff00 # 00d0 */
2321 tcg_out_rlw(s
, RLWINM
, TCG_REG_R0
, a1
, 8, 16, 23);
2322 /* a0 = (a1 r<< 24) & 0xff # 000c */
2323 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 24, 24, 31);
2324 /* a0 = a0 | r0 # 00dc */
2325 tcg_out32(s
, OR
| SAB(TCG_REG_R0
, a0
, a0
));
2329 case INDEX_op_bswap32_i32
:
2330 case INDEX_op_bswap32_i64
:
2331 /* Stolen from gcc's builtin_bswap32 */
2333 a0
= args
[0] == a1
? TCG_REG_R0
: args
[0];
2335 /* a1 = args[1] # abcd */
2336 /* a0 = rotate_left (a1, 8) # bcda */
2337 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 8, 0, 31);
2338 /* a0 = (a0 & ~0xff000000) | ((a1 r<< 24) & 0xff000000) # dcda */
2339 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 0, 7);
2340 /* a0 = (a0 & ~0x0000ff00) | ((a1 r<< 24) & 0x0000ff00) # dcba */
2341 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 16, 23);
2343 if (a0
== TCG_REG_R0
) {
2344 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2348 case INDEX_op_bswap64_i64
:
2349 a0
= args
[0], a1
= args
[1], a2
= TCG_REG_R0
;
2355 /* a1 = # abcd efgh */
2356 /* a0 = rl32(a1, 8) # 0000 fghe */
2357 tcg_out_rlw(s
, RLWINM
, a0
, a1
, 8, 0, 31);
2358 /* a0 = dep(a0, rl32(a1, 24), 0xff000000) # 0000 hghe */
2359 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 0, 7);
2360 /* a0 = dep(a0, rl32(a1, 24), 0x0000ff00) # 0000 hgfe */
2361 tcg_out_rlw(s
, RLWIMI
, a0
, a1
, 24, 16, 23);
2363 /* a0 = rl64(a0, 32) # hgfe 0000 */
2364 /* a2 = rl64(a1, 32) # efgh abcd */
2365 tcg_out_rld(s
, RLDICL
, a0
, a0
, 32, 0);
2366 tcg_out_rld(s
, RLDICL
, a2
, a1
, 32, 0);
2368 /* a0 = dep(a0, rl32(a2, 8), 0xffffffff) # hgfe bcda */
2369 tcg_out_rlw(s
, RLWIMI
, a0
, a2
, 8, 0, 31);
2370 /* a0 = dep(a0, rl32(a2, 24), 0xff000000) # hgfe dcda */
2371 tcg_out_rlw(s
, RLWIMI
, a0
, a2
, 24, 0, 7);
2372 /* a0 = dep(a0, rl32(a2, 24), 0x0000ff00) # hgfe dcba */
2373 tcg_out_rlw(s
, RLWIMI
, a0
, a2
, 24, 16, 23);
2376 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2380 case INDEX_op_deposit_i32
:
2381 if (const_args
[2]) {
2382 uint32_t mask
= ((2u << (args
[4] - 1)) - 1) << args
[3];
2383 tcg_out_andi32(s
, args
[0], args
[0], ~mask
);
2385 tcg_out_rlw(s
, RLWIMI
, args
[0], args
[2], args
[3],
2386 32 - args
[3] - args
[4], 31 - args
[3]);
2389 case INDEX_op_deposit_i64
:
2390 if (const_args
[2]) {
2391 uint64_t mask
= ((2ull << (args
[4] - 1)) - 1) << args
[3];
2392 tcg_out_andi64(s
, args
[0], args
[0], ~mask
);
2394 tcg_out_rld(s
, RLDIMI
, args
[0], args
[2], args
[3],
2395 64 - args
[3] - args
[4]);
2399 case INDEX_op_movcond_i32
:
2400 tcg_out_movcond(s
, TCG_TYPE_I32
, args
[5], args
[0], args
[1], args
[2],
2401 args
[3], args
[4], const_args
[2]);
2403 case INDEX_op_movcond_i64
:
2404 tcg_out_movcond(s
, TCG_TYPE_I64
, args
[5], args
[0], args
[1], args
[2],
2405 args
[3], args
[4], const_args
[2]);
2408 #if TCG_TARGET_REG_BITS == 64
2409 case INDEX_op_add2_i64
:
2411 case INDEX_op_add2_i32
:
2413 /* Note that the CA bit is defined based on the word size of the
2414 environment. So in 64-bit mode it's always carry-out of bit 63.
2415 The fallback code using deposit works just as well for 32-bit. */
2416 a0
= args
[0], a1
= args
[1];
2417 if (a0
== args
[3] || (!const_args
[5] && a0
== args
[5])) {
2420 if (const_args
[4]) {
2421 tcg_out32(s
, ADDIC
| TAI(a0
, args
[2], args
[4]));
2423 tcg_out32(s
, ADDC
| TAB(a0
, args
[2], args
[4]));
2425 if (const_args
[5]) {
2426 tcg_out32(s
, (args
[5] ? ADDME
: ADDZE
) | RT(a1
) | RA(args
[3]));
2428 tcg_out32(s
, ADDE
| TAB(a1
, args
[3], args
[5]));
2430 if (a0
!= args
[0]) {
2431 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2435 #if TCG_TARGET_REG_BITS == 64
2436 case INDEX_op_sub2_i64
:
2438 case INDEX_op_sub2_i32
:
2440 a0
= args
[0], a1
= args
[1];
2441 if (a0
== args
[5] || (!const_args
[3] && a0
== args
[3])) {
2444 if (const_args
[2]) {
2445 tcg_out32(s
, SUBFIC
| TAI(a0
, args
[4], args
[2]));
2447 tcg_out32(s
, SUBFC
| TAB(a0
, args
[4], args
[2]));
2449 if (const_args
[3]) {
2450 tcg_out32(s
, (args
[3] ? SUBFME
: SUBFZE
) | RT(a1
) | RA(args
[5]));
2452 tcg_out32(s
, SUBFE
| TAB(a1
, args
[5], args
[3]));
2454 if (a0
!= args
[0]) {
2455 tcg_out_mov(s
, TCG_TYPE_REG
, args
[0], a0
);
2459 case INDEX_op_muluh_i32
:
2460 tcg_out32(s
, MULHWU
| TAB(args
[0], args
[1], args
[2]));
2462 case INDEX_op_mulsh_i32
:
2463 tcg_out32(s
, MULHW
| TAB(args
[0], args
[1], args
[2]));
2465 case INDEX_op_muluh_i64
:
2466 tcg_out32(s
, MULHDU
| TAB(args
[0], args
[1], args
[2]));
2468 case INDEX_op_mulsh_i64
:
2469 tcg_out32(s
, MULHD
| TAB(args
[0], args
[1], args
[2]));
2473 tcg_out_mb(s
, args
[0]);
2476 case INDEX_op_mov_i32
: /* Always emitted via tcg_out_mov. */
2477 case INDEX_op_mov_i64
:
2478 case INDEX_op_movi_i32
: /* Always emitted via tcg_out_movi. */
2479 case INDEX_op_movi_i64
:
2480 case INDEX_op_call
: /* Always emitted via tcg_out_call. */
2486 static const TCGTargetOpDef ppc_op_defs
[] = {
2487 { INDEX_op_exit_tb
, { } },
2488 { INDEX_op_goto_tb
, { } },
2489 { INDEX_op_br
, { } },
2491 { INDEX_op_ld8u_i32
, { "r", "r" } },
2492 { INDEX_op_ld8s_i32
, { "r", "r" } },
2493 { INDEX_op_ld16u_i32
, { "r", "r" } },
2494 { INDEX_op_ld16s_i32
, { "r", "r" } },
2495 { INDEX_op_ld_i32
, { "r", "r" } },
2497 { INDEX_op_st8_i32
, { "r", "r" } },
2498 { INDEX_op_st16_i32
, { "r", "r" } },
2499 { INDEX_op_st_i32
, { "r", "r" } },
2501 { INDEX_op_add_i32
, { "r", "r", "ri" } },
2502 { INDEX_op_mul_i32
, { "r", "r", "rI" } },
2503 { INDEX_op_div_i32
, { "r", "r", "r" } },
2504 { INDEX_op_divu_i32
, { "r", "r", "r" } },
2505 { INDEX_op_sub_i32
, { "r", "rI", "ri" } },
2506 { INDEX_op_and_i32
, { "r", "r", "ri" } },
2507 { INDEX_op_or_i32
, { "r", "r", "ri" } },
2508 { INDEX_op_xor_i32
, { "r", "r", "ri" } },
2509 { INDEX_op_andc_i32
, { "r", "r", "ri" } },
2510 { INDEX_op_orc_i32
, { "r", "r", "ri" } },
2511 { INDEX_op_eqv_i32
, { "r", "r", "ri" } },
2512 { INDEX_op_nand_i32
, { "r", "r", "r" } },
2513 { INDEX_op_nor_i32
, { "r", "r", "r" } },
2515 { INDEX_op_shl_i32
, { "r", "r", "ri" } },
2516 { INDEX_op_shr_i32
, { "r", "r", "ri" } },
2517 { INDEX_op_sar_i32
, { "r", "r", "ri" } },
2518 { INDEX_op_rotl_i32
, { "r", "r", "ri" } },
2519 { INDEX_op_rotr_i32
, { "r", "r", "ri" } },
2521 { INDEX_op_neg_i32
, { "r", "r" } },
2522 { INDEX_op_not_i32
, { "r", "r" } },
2523 { INDEX_op_ext8s_i32
, { "r", "r" } },
2524 { INDEX_op_ext16s_i32
, { "r", "r" } },
2525 { INDEX_op_bswap16_i32
, { "r", "r" } },
2526 { INDEX_op_bswap32_i32
, { "r", "r" } },
2528 { INDEX_op_brcond_i32
, { "r", "ri" } },
2529 { INDEX_op_setcond_i32
, { "r", "r", "ri" } },
2530 { INDEX_op_movcond_i32
, { "r", "r", "ri", "rZ", "rZ" } },
2532 { INDEX_op_deposit_i32
, { "r", "0", "rZ" } },
2534 { INDEX_op_muluh_i32
, { "r", "r", "r" } },
2535 { INDEX_op_mulsh_i32
, { "r", "r", "r" } },
2537 #if TCG_TARGET_REG_BITS == 64
2538 { INDEX_op_ld8u_i64
, { "r", "r" } },
2539 { INDEX_op_ld8s_i64
, { "r", "r" } },
2540 { INDEX_op_ld16u_i64
, { "r", "r" } },
2541 { INDEX_op_ld16s_i64
, { "r", "r" } },
2542 { INDEX_op_ld32u_i64
, { "r", "r" } },
2543 { INDEX_op_ld32s_i64
, { "r", "r" } },
2544 { INDEX_op_ld_i64
, { "r", "r" } },
2546 { INDEX_op_st8_i64
, { "r", "r" } },
2547 { INDEX_op_st16_i64
, { "r", "r" } },
2548 { INDEX_op_st32_i64
, { "r", "r" } },
2549 { INDEX_op_st_i64
, { "r", "r" } },
2551 { INDEX_op_add_i64
, { "r", "r", "rT" } },
2552 { INDEX_op_sub_i64
, { "r", "rI", "rT" } },
2553 { INDEX_op_and_i64
, { "r", "r", "ri" } },
2554 { INDEX_op_or_i64
, { "r", "r", "rU" } },
2555 { INDEX_op_xor_i64
, { "r", "r", "rU" } },
2556 { INDEX_op_andc_i64
, { "r", "r", "ri" } },
2557 { INDEX_op_orc_i64
, { "r", "r", "r" } },
2558 { INDEX_op_eqv_i64
, { "r", "r", "r" } },
2559 { INDEX_op_nand_i64
, { "r", "r", "r" } },
2560 { INDEX_op_nor_i64
, { "r", "r", "r" } },
2562 { INDEX_op_shl_i64
, { "r", "r", "ri" } },
2563 { INDEX_op_shr_i64
, { "r", "r", "ri" } },
2564 { INDEX_op_sar_i64
, { "r", "r", "ri" } },
2565 { INDEX_op_rotl_i64
, { "r", "r", "ri" } },
2566 { INDEX_op_rotr_i64
, { "r", "r", "ri" } },
2568 { INDEX_op_mul_i64
, { "r", "r", "rI" } },
2569 { INDEX_op_div_i64
, { "r", "r", "r" } },
2570 { INDEX_op_divu_i64
, { "r", "r", "r" } },
2572 { INDEX_op_neg_i64
, { "r", "r" } },
2573 { INDEX_op_not_i64
, { "r", "r" } },
2574 { INDEX_op_ext8s_i64
, { "r", "r" } },
2575 { INDEX_op_ext16s_i64
, { "r", "r" } },
2576 { INDEX_op_ext32s_i64
, { "r", "r" } },
2577 { INDEX_op_ext_i32_i64
, { "r", "r" } },
2578 { INDEX_op_extu_i32_i64
, { "r", "r" } },
2579 { INDEX_op_bswap16_i64
, { "r", "r" } },
2580 { INDEX_op_bswap32_i64
, { "r", "r" } },
2581 { INDEX_op_bswap64_i64
, { "r", "r" } },
2583 { INDEX_op_brcond_i64
, { "r", "ri" } },
2584 { INDEX_op_setcond_i64
, { "r", "r", "ri" } },
2585 { INDEX_op_movcond_i64
, { "r", "r", "ri", "rZ", "rZ" } },
2587 { INDEX_op_deposit_i64
, { "r", "0", "rZ" } },
2589 { INDEX_op_mulsh_i64
, { "r", "r", "r" } },
2590 { INDEX_op_muluh_i64
, { "r", "r", "r" } },
2593 #if TCG_TARGET_REG_BITS == 32
2594 { INDEX_op_brcond2_i32
, { "r", "r", "ri", "ri" } },
2595 { INDEX_op_setcond2_i32
, { "r", "r", "r", "ri", "ri" } },
2598 #if TCG_TARGET_REG_BITS == 64
2599 { INDEX_op_add2_i64
, { "r", "r", "r", "r", "rI", "rZM" } },
2600 { INDEX_op_sub2_i64
, { "r", "r", "rI", "rZM", "r", "r" } },
2602 { INDEX_op_add2_i32
, { "r", "r", "r", "r", "rI", "rZM" } },
2603 { INDEX_op_sub2_i32
, { "r", "r", "rI", "rZM", "r", "r" } },
2606 #if TCG_TARGET_REG_BITS == 64
2607 { INDEX_op_qemu_ld_i32
, { "r", "L" } },
2608 { INDEX_op_qemu_st_i32
, { "S", "S" } },
2609 { INDEX_op_qemu_ld_i64
, { "r", "L" } },
2610 { INDEX_op_qemu_st_i64
, { "S", "S" } },
2611 #elif TARGET_LONG_BITS == 32
2612 { INDEX_op_qemu_ld_i32
, { "r", "L" } },
2613 { INDEX_op_qemu_st_i32
, { "S", "S" } },
2614 { INDEX_op_qemu_ld_i64
, { "L", "L", "L" } },
2615 { INDEX_op_qemu_st_i64
, { "S", "S", "S" } },
2617 { INDEX_op_qemu_ld_i32
, { "r", "L", "L" } },
2618 { INDEX_op_qemu_st_i32
, { "S", "S", "S" } },
2619 { INDEX_op_qemu_ld_i64
, { "L", "L", "L", "L" } },
2620 { INDEX_op_qemu_st_i64
, { "S", "S", "S", "S" } },
2623 { INDEX_op_mb
, { } },
2627 static void tcg_target_init(TCGContext
*s
)
2629 unsigned long hwcap
= qemu_getauxval(AT_HWCAP
);
2630 if (hwcap
& PPC_FEATURE_ARCH_2_06
) {
2631 have_isa_2_06
= true;
2634 tcg_regset_set32(tcg_target_available_regs
[TCG_TYPE_I32
], 0, 0xffffffff);
2635 tcg_regset_set32(tcg_target_available_regs
[TCG_TYPE_I64
], 0, 0xffffffff);
2636 tcg_regset_set32(tcg_target_call_clobber_regs
, 0,
2646 (1 << TCG_REG_R10
) |
2647 (1 << TCG_REG_R11
) |
2648 (1 << TCG_REG_R12
));
2650 tcg_regset_clear(s
->reserved_regs
);
2651 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R0
); /* tcg temp */
2652 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R1
); /* stack pointer */
2653 #if defined(_CALL_SYSV)
2654 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R2
); /* toc pointer */
2656 #if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
2657 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_R13
); /* thread pointer */
2659 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_TMP1
); /* mem temp */
2661 tcg_regset_set_reg(s
->reserved_regs
, TCG_REG_RA
); /* return addr */
2664 tcg_add_target_add_op_defs(ppc_op_defs
);
2670 DebugFrameFDEHeader fde
;
2671 uint8_t fde_def_cfa
[4];
2672 uint8_t fde_reg_ofs
[ARRAY_SIZE(tcg_target_callee_save_regs
) * 2 + 3];
2675 /* We're expecting a 2 byte uleb128 encoded value. */
2676 QEMU_BUILD_BUG_ON(FRAME_SIZE
>= (1 << 14));
2678 #if TCG_TARGET_REG_BITS == 64
2679 # define ELF_HOST_MACHINE EM_PPC64
2681 # define ELF_HOST_MACHINE EM_PPC
2684 static DebugFrame debug_frame
= {
2685 .cie
.len
= sizeof(DebugFrameCIE
)-4, /* length after .len member */
2688 .cie
.code_align
= 1,
2689 .cie
.data_align
= (-SZR
& 0x7f), /* sleb128 -SZR */
2690 .cie
.return_column
= 65,
2692 /* Total FDE size does not include the "len" member. */
2693 .fde
.len
= sizeof(DebugFrame
) - offsetof(DebugFrame
, fde
.cie_offset
),
2696 12, TCG_REG_R1
, /* DW_CFA_def_cfa r1, ... */
2697 (FRAME_SIZE
& 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */
2701 /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
2702 0x11, 65, (LR_OFFSET
/ -SZR
) & 0x7f,
2706 void tcg_register_jit(void *buf
, size_t buf_size
)
2708 uint8_t *p
= &debug_frame
.fde_reg_ofs
[3];
2711 for (i
= 0; i
< ARRAY_SIZE(tcg_target_callee_save_regs
); ++i
, p
+= 2) {
2712 p
[0] = 0x80 + tcg_target_callee_save_regs
[i
];
2713 p
[1] = (FRAME_SIZE
- (REG_SAVE_BOT
+ i
* SZR
)) / SZR
;
2716 debug_frame
.fde
.func_start
= (uintptr_t)buf
;
2717 debug_frame
.fde
.func_len
= buf_size
;
2719 tcg_register_jit_int(buf
, buf_size
, &debug_frame
, sizeof(debug_frame
));
2721 #endif /* __ELF__ */
2723 static size_t dcache_bsize
= 16;
2724 static size_t icache_bsize
= 16;
2726 void flush_icache_range(uintptr_t start
, uintptr_t stop
)
2728 uintptr_t p
, start1
, stop1
;
2729 size_t dsize
= dcache_bsize
;
2730 size_t isize
= icache_bsize
;
2732 start1
= start
& ~(dsize
- 1);
2733 stop1
= (stop
+ dsize
- 1) & ~(dsize
- 1);
2734 for (p
= start1
; p
< stop1
; p
+= dsize
) {
2735 asm volatile ("dcbst 0,%0" : : "r"(p
) : "memory");
2737 asm volatile ("sync" : : : "memory");
2739 start
&= start
& ~(isize
- 1);
2740 stop1
= (stop
+ isize
- 1) & ~(isize
- 1);
2741 for (p
= start1
; p
< stop1
; p
+= isize
) {
2742 asm volatile ("icbi 0,%0" : : "r"(p
) : "memory");
2744 asm volatile ("sync" : : : "memory");
2745 asm volatile ("isync" : : : "memory");
2749 #include <sys/systemcfg.h>
2751 static void __attribute__((constructor
)) tcg_cache_init(void)
2753 icache_bsize
= _system_configuration
.icache_line
;
2754 dcache_bsize
= _system_configuration
.dcache_line
;
2757 #elif defined __linux__
2758 static void __attribute__((constructor
)) tcg_cache_init(void)
2760 unsigned long dsize
= qemu_getauxval(AT_DCACHEBSIZE
);
2761 unsigned long isize
= qemu_getauxval(AT_ICACHEBSIZE
);
2763 if (dsize
== 0 || isize
== 0) {
2765 fprintf(stderr
, "getauxval AT_DCACHEBSIZE failed\n");
2768 fprintf(stderr
, "getauxval AT_ICACHEBSIZE failed\n");
2772 dcache_bsize
= dsize
;
2773 icache_bsize
= isize
;
2776 #elif defined __APPLE__
2777 #include <sys/sysctl.h>
2779 static void __attribute__((constructor
)) tcg_cache_init(void)
2783 int name
[2] = { CTL_HW
, HW_CACHELINE
};
2785 len
= sizeof(cacheline
);
2786 if (sysctl(name
, 2, &cacheline
, &len
, NULL
, 0)) {
2787 perror("sysctl CTL_HW HW_CACHELINE failed");
2790 dcache_bsize
= cacheline
;
2791 icache_bsize
= cacheline
;
2794 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2795 #include <sys/sysctl.h>
2797 static void __attribute__((constructor
)) tcg_cache_init(void)
2802 if (sysctlbyname ("machdep.cacheline_size", &cacheline
, &len
, NULL
, 0)) {
2803 fprintf(stderr
, "sysctlbyname machdep.cacheline_size failed: %s\n",
2807 dcache_bsize
= cacheline
;
2808 icache_bsize
= cacheline
;