gic: provide defines for v2/v3 targetlist sizes
[qemu/ar7.git] / tcg / ppc / tcg-target.inc.c
blobeaf1bd9bfdb664a981e82b95cba6c21b7521b855
1 /*
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
22 * THE SOFTWARE.
25 #include "tcg-be-ldst.h"
27 #if defined _CALL_DARWIN || defined __APPLE__
28 #define TCG_TARGET_CALL_DARWIN
29 #endif
30 #ifdef _CALL_SYSV
31 # define TCG_TARGET_CALL_ALIGN_ARGS 1
32 #endif
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. */
38 #ifdef _CALL_AIX
39 # define TCG_REG_TMP1 TCG_REG_R2
40 #else
41 # define TCG_REG_TMP1 TCG_REG_R12
42 #endif
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;
83 #include "elf.h"
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
90 #endif
92 #ifdef CONFIG_DEBUG_TCG
93 static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
94 "r0",
95 "r1",
96 "r2",
97 "r3",
98 "r4",
99 "r5",
100 "r6",
101 "r7",
102 "r8",
103 "r9",
104 "r10",
105 "r11",
106 "r12",
107 "r13",
108 "r14",
109 "r15",
110 "r16",
111 "r17",
112 "r18",
113 "r19",
114 "r20",
115 "r21",
116 "r22",
117 "r23",
118 "r24",
119 "r25",
120 "r26",
121 "r27",
122 "r28",
123 "r29",
124 "r30",
125 "r31"
127 #endif
129 static const int tcg_target_reg_alloc_order[] = {
130 TCG_REG_R14, /* call saved registers */
131 TCG_REG_R15,
132 TCG_REG_R16,
133 TCG_REG_R17,
134 TCG_REG_R18,
135 TCG_REG_R19,
136 TCG_REG_R20,
137 TCG_REG_R21,
138 TCG_REG_R22,
139 TCG_REG_R23,
140 TCG_REG_R24,
141 TCG_REG_R25,
142 TCG_REG_R26,
143 TCG_REG_R27,
144 TCG_REG_R28,
145 TCG_REG_R29,
146 TCG_REG_R30,
147 TCG_REG_R31,
148 TCG_REG_R12, /* call clobbered, non-arguments */
149 TCG_REG_R11,
150 TCG_REG_R2,
151 TCG_REG_R13,
152 TCG_REG_R10, /* call clobbered, arguments */
153 TCG_REG_R9,
154 TCG_REG_R8,
155 TCG_REG_R7,
156 TCG_REG_R6,
157 TCG_REG_R5,
158 TCG_REG_R4,
159 TCG_REG_R3,
162 static const int tcg_target_call_iarg_regs[] = {
163 TCG_REG_R3,
164 TCG_REG_R4,
165 TCG_REG_R5,
166 TCG_REG_R6,
167 TCG_REG_R7,
168 TCG_REG_R8,
169 TCG_REG_R9,
170 TCG_REG_R10
173 static const int tcg_target_call_oarg_regs[] = {
174 TCG_REG_R3,
175 TCG_REG_R4
178 static const int tcg_target_callee_save_regs[] = {
179 #ifdef TCG_TARGET_CALL_DARWIN
180 TCG_REG_R11,
181 #endif
182 TCG_REG_R14,
183 TCG_REG_R15,
184 TCG_REG_R16,
185 TCG_REG_R17,
186 TCG_REG_R18,
187 TCG_REG_R19,
188 TCG_REG_R20,
189 TCG_REG_R21,
190 TCG_REG_R22,
191 TCG_REG_R23,
192 TCG_REG_R24,
193 TCG_REG_R25,
194 TCG_REG_R26,
195 TCG_REG_R27, /* currently used for the global env */
196 TCG_REG_R28,
197 TCG_REG_R29,
198 TCG_REG_R30,
199 TCG_REG_R31
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);
249 switch (type) {
250 case R_PPC_REL14:
251 reloc_pc14(code_ptr, target);
252 break;
253 case R_PPC_REL24:
254 reloc_pc24(code_ptr, target);
255 break;
256 default:
257 tcg_abort();
261 /* parse target specific constraints */
262 static int target_parse_constraint(TCGArgConstraint *ct, const char **pct_str)
264 const char *ct_str;
266 ct_str = *pct_str;
267 switch (ct_str[0]) {
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');
271 break;
272 case 'r':
273 ct->ct |= TCG_CT_REG;
274 tcg_regset_set32(ct->u.regs, 0, 0xffffffff);
275 break;
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);
283 #endif
284 break;
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);
293 #endif
294 break;
295 case 'I':
296 ct->ct |= TCG_CT_CONST_S16;
297 break;
298 case 'J':
299 ct->ct |= TCG_CT_CONST_U16;
300 break;
301 case 'M':
302 ct->ct |= TCG_CT_CONST_MONE;
303 break;
304 case 'T':
305 ct->ct |= TCG_CT_CONST_S32;
306 break;
307 case 'U':
308 ct->ct |= TCG_CT_CONST_U32;
309 break;
310 case 'Z':
311 ct->ct |= TCG_CT_CONST_ZERO;
312 break;
313 default:
314 return -1;
316 ct_str++;
317 *pct_str = ct_str;
318 return 0;
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)
325 int ct = arg_ct->ct;
326 if (ct & TCG_CT_CONST) {
327 return 1;
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) {
333 val = (int32_t)val;
336 if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
337 return 1;
338 } else if ((ct & TCG_CT_CONST_U16) && val == (uint16_t)val) {
339 return 1;
340 } else if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
341 return 1;
342 } else if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
343 return 1;
344 } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
345 return 1;
346 } else if ((ct & TCG_CT_CONST_MONE) && val == -1) {
347 return 1;
349 return 0;
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))
360 #define B OPCD( 18)
361 #define BC OPCD( 16)
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)
370 #define STD XO62( 0)
371 #define STDU XO62( 1)
372 #define STDX XO31(149)
374 #define LD XO58( 0)
375 #define LDX XO31( 21)
376 #define LDU XO58( 1)
377 #define LWA XO58( 2)
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)
428 #define OR XO31(444)
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)
435 #define CMP XO31( 0)
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)
473 #define LR SPR(8, 0)
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)
485 #define TW XO31( 4)
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)))
502 #define LK 1
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)
520 enum {
521 CR_LT,
522 CR_GT,
523 CR_EQ,
524 CR_SO
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);
560 if (ret != arg) {
561 tcg_out32(s, OR | SAB(arg, ret, arg));
565 static inline void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs,
566 int sh, int mb)
568 tcg_debug_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));
609 } else {
610 tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
611 if (arg & 0xffff) {
612 tcg_out32(s, ORI | SAI(ret, ret, arg));
617 static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,
618 tcg_target_long arg)
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));
626 } else {
627 int32_t high;
629 if (USE_REG_RA) {
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);
633 return;
637 high = arg >> 31 >> 1;
638 tcg_out_movi32(s, ret, high);
639 if (high) {
640 tcg_out_shli64(s, ret, ret, 32);
642 if (arg & 0xffff0000) {
643 tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
645 if (arg & 0xffff) {
646 tcg_out32(s, ORI | SAI(ret, ret, arg));
651 static bool mask_operand(uint32_t c, int *mb, int *me)
653 uint32_t lsb, test;
655 /* Accept a bit pattern like:
656 0....01....1
657 1....10....0
658 0..01..10..0
659 Keep track of the transitions. */
660 if (c == 0 || c == -1) {
661 return false;
663 test = c;
664 lsb = test & -test;
665 test += lsb;
666 if (test & (test - 1)) {
667 return false;
670 *me = clz32(lsb);
671 *mb = test ? clz32(test & -test) + 1 : 0;
672 return true;
675 static bool mask64_operand(uint64_t c, int *mb, int *me)
677 uint64_t lsb;
679 if (c == 0) {
680 return false;
683 lsb = c & -c;
684 /* Accept 1..10..0. */
685 if (c == -lsb) {
686 *mb = 0;
687 *me = clz64(lsb);
688 return true;
690 /* Accept 0..01..1. */
691 if (lsb == 1 && (c & (c + 1)) == 0) {
692 *mb = clz64(c + 1) + 1;
693 *me = 63;
694 return true;
696 return false;
699 static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
701 int mb, me;
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));
707 return;
708 } else if ((c & 0xffff0000) == c) {
709 tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
710 return;
711 } else {
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)
719 int mb, me;
721 tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
722 if (mask64_operand(c, &mb, &me)) {
723 if (mb == 0) {
724 tcg_out_rld(s, RLDICR, dst, src, 0, me);
725 } else {
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));
730 return;
731 } else if ((c & 0xffff0000) == c) {
732 tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
733 return;
734 } else {
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)
743 if (c >> 16) {
744 tcg_out32(s, op_hi | SAI(src, dst, c >> 16));
745 src = dst;
747 if (c & 0xffff) {
748 tcg_out32(s, op_lo | SAI(src, dst, c));
749 src = dst;
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);
768 } else {
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;
782 switch (opi) {
783 case LD: case LWA:
784 align = 3;
785 /* FALLTHRU */
786 default:
787 if (rt != TCG_REG_R0) {
788 rs = rt;
789 break;
791 break;
792 case STD:
793 align = 3;
794 /* FALLTHRU */
795 case STB: case STH: case STW:
796 is_store = true;
797 break;
800 /* For unaligned, or very large offsets, use the indexed form. */
801 if (offset & align || offset != (int32_t)offset) {
802 if (rs == base) {
803 rs = TCG_REG_R0;
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));
808 return;
811 l0 = (int16_t)offset;
812 offset = (offset - l0) >> 16;
813 l1 = (int16_t)offset;
815 if (l1 < 0 && orig >= 0) {
816 extra = 0x4000;
817 l1 = (int16_t)(offset - 0x4000);
819 if (l1) {
820 tcg_out32(s, ADDIS | TAI(rs, base, l1));
821 base = rs;
823 if (extra) {
824 tcg_out32(s, ADDIS | TAI(rs, base, extra));
825 base = rs;
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)
835 int opi, opx;
837 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
838 if (type == TCG_TYPE_I32) {
839 opi = LWZ, opx = LWZX;
840 } else {
841 opi = LD, opx = LDX;
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)
849 int opi, opx;
851 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
852 if (type == TCG_TYPE_I32) {
853 opi = STW, opx = STWX;
854 } else {
855 opi = STD, opx = STDX;
857 tcg_out_mem_long(s, opi, opx, arg, arg1, arg2);
860 static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
861 TCGReg base, intptr_t ofs)
863 return false;
866 static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2,
867 int const_arg2, int cr, TCGType type)
869 int imm;
870 uint32_t op;
872 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
874 /* Simplify the comparisons below wrt CMPI. */
875 if (type == TCG_TYPE_I32) {
876 arg2 = (int32_t)arg2;
879 switch (cond) {
880 case TCG_COND_EQ:
881 case TCG_COND_NE:
882 if (const_arg2) {
883 if ((int16_t) arg2 == arg2) {
884 op = CMPI;
885 imm = 1;
886 break;
887 } else if ((uint16_t) arg2 == arg2) {
888 op = CMPLI;
889 imm = 1;
890 break;
893 op = CMPL;
894 imm = 0;
895 break;
897 case TCG_COND_LT:
898 case TCG_COND_GE:
899 case TCG_COND_LE:
900 case TCG_COND_GT:
901 if (const_arg2) {
902 if ((int16_t) arg2 == arg2) {
903 op = CMPI;
904 imm = 1;
905 break;
908 op = CMP;
909 imm = 0;
910 break;
912 case TCG_COND_LTU:
913 case TCG_COND_GEU:
914 case TCG_COND_LEU:
915 case TCG_COND_GTU:
916 if (const_arg2) {
917 if ((uint16_t) arg2 == arg2) {
918 op = CMPLI;
919 imm = 1;
920 break;
923 op = CMPL;
924 imm = 0;
925 break;
927 default:
928 tcg_abort();
930 op |= BF(cr) | ((type == TCG_TYPE_I64) << 21);
932 if (imm) {
933 tcg_out32(s, op | RA(arg1) | (arg2 & 0xffff));
934 } else {
935 if (const_arg2) {
936 tcg_out_movi(s, type, TCG_REG_R0, arg2);
937 arg2 = TCG_REG_R0;
939 tcg_out32(s, op | RA(arg1) | RB(arg2));
943 static void tcg_out_setcond_eq0(TCGContext *s, TCGType type,
944 TCGReg dst, TCGReg src)
946 if (type == TCG_TYPE_I32) {
947 tcg_out32(s, CNTLZW | RS(src) | RA(dst));
948 tcg_out_shri32(s, dst, dst, 5);
949 } else {
950 tcg_out32(s, CNTLZD | RS(src) | RA(dst));
951 tcg_out_shri64(s, dst, dst, 6);
955 static void tcg_out_setcond_ne0(TCGContext *s, TCGReg dst, TCGReg src)
957 /* X != 0 implies X + -1 generates a carry. Extra addition
958 trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C. */
959 if (dst != src) {
960 tcg_out32(s, ADDIC | TAI(dst, src, -1));
961 tcg_out32(s, SUBFE | TAB(dst, dst, src));
962 } else {
963 tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1));
964 tcg_out32(s, SUBFE | TAB(dst, TCG_REG_R0, src));
968 static TCGReg tcg_gen_setcond_xor(TCGContext *s, TCGReg arg1, TCGArg arg2,
969 bool const_arg2)
971 if (const_arg2) {
972 if ((uint32_t)arg2 == arg2) {
973 tcg_out_xori32(s, TCG_REG_R0, arg1, arg2);
974 } else {
975 tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, arg2);
976 tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, TCG_REG_R0));
978 } else {
979 tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, arg2));
981 return TCG_REG_R0;
984 static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
985 TCGArg arg0, TCGArg arg1, TCGArg arg2,
986 int const_arg2)
988 int crop, sh;
990 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
992 /* Ignore high bits of a potential constant arg2. */
993 if (type == TCG_TYPE_I32) {
994 arg2 = (uint32_t)arg2;
997 /* Handle common and trivial cases before handling anything else. */
998 if (arg2 == 0) {
999 switch (cond) {
1000 case TCG_COND_EQ:
1001 tcg_out_setcond_eq0(s, type, arg0, arg1);
1002 return;
1003 case TCG_COND_NE:
1004 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1005 tcg_out_ext32u(s, TCG_REG_R0, arg1);
1006 arg1 = TCG_REG_R0;
1008 tcg_out_setcond_ne0(s, arg0, arg1);
1009 return;
1010 case TCG_COND_GE:
1011 tcg_out32(s, NOR | SAB(arg1, arg0, arg1));
1012 arg1 = arg0;
1013 /* FALLTHRU */
1014 case TCG_COND_LT:
1015 /* Extract the sign bit. */
1016 if (type == TCG_TYPE_I32) {
1017 tcg_out_shri32(s, arg0, arg1, 31);
1018 } else {
1019 tcg_out_shri64(s, arg0, arg1, 63);
1021 return;
1022 default:
1023 break;
1027 /* If we have ISEL, we can implement everything with 3 or 4 insns.
1028 All other cases below are also at least 3 insns, so speed up the
1029 code generator by not considering them and always using ISEL. */
1030 if (HAVE_ISEL) {
1031 int isel, tab;
1033 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1035 isel = tcg_to_isel[cond];
1037 tcg_out_movi(s, type, arg0, 1);
1038 if (isel & 1) {
1039 /* arg0 = (bc ? 0 : 1) */
1040 tab = TAB(arg0, 0, arg0);
1041 isel &= ~1;
1042 } else {
1043 /* arg0 = (bc ? 1 : 0) */
1044 tcg_out_movi(s, type, TCG_REG_R0, 0);
1045 tab = TAB(arg0, arg0, TCG_REG_R0);
1047 tcg_out32(s, isel | tab);
1048 return;
1051 switch (cond) {
1052 case TCG_COND_EQ:
1053 arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1054 tcg_out_setcond_eq0(s, type, arg0, arg1);
1055 return;
1057 case TCG_COND_NE:
1058 arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1059 /* Discard the high bits only once, rather than both inputs. */
1060 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1061 tcg_out_ext32u(s, TCG_REG_R0, arg1);
1062 arg1 = TCG_REG_R0;
1064 tcg_out_setcond_ne0(s, arg0, arg1);
1065 return;
1067 case TCG_COND_GT:
1068 case TCG_COND_GTU:
1069 sh = 30;
1070 crop = 0;
1071 goto crtest;
1073 case TCG_COND_LT:
1074 case TCG_COND_LTU:
1075 sh = 29;
1076 crop = 0;
1077 goto crtest;
1079 case TCG_COND_GE:
1080 case TCG_COND_GEU:
1081 sh = 31;
1082 crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_LT) | BB(7, CR_LT);
1083 goto crtest;
1085 case TCG_COND_LE:
1086 case TCG_COND_LEU:
1087 sh = 31;
1088 crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_GT) | BB(7, CR_GT);
1089 crtest:
1090 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1091 if (crop) {
1092 tcg_out32(s, crop);
1094 tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1095 tcg_out_rlw(s, RLWINM, arg0, TCG_REG_R0, sh, 31, 31);
1096 break;
1098 default:
1099 tcg_abort();
1103 static void tcg_out_bc(TCGContext *s, int bc, TCGLabel *l)
1105 if (l->has_value) {
1106 tcg_out32(s, bc | reloc_pc14_val(s->code_ptr, l->u.value_ptr));
1107 } else {
1108 tcg_out_reloc(s, s->code_ptr, R_PPC_REL14, l, 0);
1109 tcg_out_bc_noaddr(s, bc);
1113 static void tcg_out_brcond(TCGContext *s, TCGCond cond,
1114 TCGArg arg1, TCGArg arg2, int const_arg2,
1115 TCGLabel *l, TCGType type)
1117 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1118 tcg_out_bc(s, tcg_to_bc[cond], l);
1121 static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
1122 TCGArg dest, TCGArg c1, TCGArg c2, TCGArg v1,
1123 TCGArg v2, bool const_c2)
1125 /* If for some reason both inputs are zero, don't produce bad code. */
1126 if (v1 == 0 && v2 == 0) {
1127 tcg_out_movi(s, type, dest, 0);
1128 return;
1131 tcg_out_cmp(s, cond, c1, c2, const_c2, 7, type);
1133 if (HAVE_ISEL) {
1134 int isel = tcg_to_isel[cond];
1136 /* Swap the V operands if the operation indicates inversion. */
1137 if (isel & 1) {
1138 int t = v1;
1139 v1 = v2;
1140 v2 = t;
1141 isel &= ~1;
1143 /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand. */
1144 if (v2 == 0) {
1145 tcg_out_movi(s, type, TCG_REG_R0, 0);
1147 tcg_out32(s, isel | TAB(dest, v1, v2));
1148 } else {
1149 if (dest == v2) {
1150 cond = tcg_invert_cond(cond);
1151 v2 = v1;
1152 } else if (dest != v1) {
1153 if (v1 == 0) {
1154 tcg_out_movi(s, type, dest, 0);
1155 } else {
1156 tcg_out_mov(s, type, dest, v1);
1159 /* Branch forward over one insn */
1160 tcg_out32(s, tcg_to_bc[cond] | 8);
1161 if (v2 == 0) {
1162 tcg_out_movi(s, type, dest, 0);
1163 } else {
1164 tcg_out_mov(s, type, dest, v2);
1169 static void tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1170 const int *const_args)
1172 static const struct { uint8_t bit1, bit2; } bits[] = {
1173 [TCG_COND_LT ] = { CR_LT, CR_LT },
1174 [TCG_COND_LE ] = { CR_LT, CR_GT },
1175 [TCG_COND_GT ] = { CR_GT, CR_GT },
1176 [TCG_COND_GE ] = { CR_GT, CR_LT },
1177 [TCG_COND_LTU] = { CR_LT, CR_LT },
1178 [TCG_COND_LEU] = { CR_LT, CR_GT },
1179 [TCG_COND_GTU] = { CR_GT, CR_GT },
1180 [TCG_COND_GEU] = { CR_GT, CR_LT },
1183 TCGCond cond = args[4], cond2;
1184 TCGArg al, ah, bl, bh;
1185 int blconst, bhconst;
1186 int op, bit1, bit2;
1188 al = args[0];
1189 ah = args[1];
1190 bl = args[2];
1191 bh = args[3];
1192 blconst = const_args[2];
1193 bhconst = const_args[3];
1195 switch (cond) {
1196 case TCG_COND_EQ:
1197 op = CRAND;
1198 goto do_equality;
1199 case TCG_COND_NE:
1200 op = CRNAND;
1201 do_equality:
1202 tcg_out_cmp(s, cond, al, bl, blconst, 6, TCG_TYPE_I32);
1203 tcg_out_cmp(s, cond, ah, bh, bhconst, 7, TCG_TYPE_I32);
1204 tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1205 break;
1207 case TCG_COND_LT:
1208 case TCG_COND_LE:
1209 case TCG_COND_GT:
1210 case TCG_COND_GE:
1211 case TCG_COND_LTU:
1212 case TCG_COND_LEU:
1213 case TCG_COND_GTU:
1214 case TCG_COND_GEU:
1215 bit1 = bits[cond].bit1;
1216 bit2 = bits[cond].bit2;
1217 op = (bit1 != bit2 ? CRANDC : CRAND);
1218 cond2 = tcg_unsigned_cond(cond);
1220 tcg_out_cmp(s, cond, ah, bh, bhconst, 6, TCG_TYPE_I32);
1221 tcg_out_cmp(s, cond2, al, bl, blconst, 7, TCG_TYPE_I32);
1222 tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, bit2));
1223 tcg_out32(s, CROR | BT(7, CR_EQ) | BA(6, bit1) | BB(7, CR_EQ));
1224 break;
1226 default:
1227 tcg_abort();
1231 static void tcg_out_setcond2(TCGContext *s, const TCGArg *args,
1232 const int *const_args)
1234 tcg_out_cmp2(s, args + 1, const_args + 1);
1235 tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1236 tcg_out_rlw(s, RLWINM, args[0], TCG_REG_R0, 31, 31, 31);
1239 static void tcg_out_brcond2 (TCGContext *s, const TCGArg *args,
1240 const int *const_args)
1242 tcg_out_cmp2(s, args, const_args);
1243 tcg_out_bc(s, BC | BI(7, CR_EQ) | BO_COND_TRUE, arg_label(args[5]));
1246 #ifdef __powerpc64__
1247 void ppc_tb_set_jmp_target(uintptr_t jmp_addr, uintptr_t addr)
1249 tcg_insn_unit i1, i2;
1250 uint64_t pair;
1251 intptr_t diff = addr - jmp_addr;
1253 if (in_range_b(diff)) {
1254 i1 = B | (diff & 0x3fffffc);
1255 i2 = NOP;
1256 } else if (USE_REG_RA) {
1257 intptr_t lo, hi;
1258 diff = addr - (uintptr_t)tb_ret_addr;
1259 lo = (int16_t)diff;
1260 hi = (int32_t)(diff - lo);
1261 tcg_debug_assert(diff == hi + lo);
1262 i1 = ADDIS | TAI(TCG_REG_TMP1, TCG_REG_RA, hi >> 16);
1263 i2 = ADDI | TAI(TCG_REG_TMP1, TCG_REG_TMP1, lo);
1264 } else {
1265 tcg_debug_assert(TCG_TARGET_REG_BITS == 32 || addr == (int32_t)addr);
1266 i1 = ADDIS | TAI(TCG_REG_TMP1, 0, addr >> 16);
1267 i2 = ORI | SAI(TCG_REG_TMP1, TCG_REG_TMP1, addr);
1269 #ifdef HOST_WORDS_BIGENDIAN
1270 pair = (uint64_t)i1 << 32 | i2;
1271 #else
1272 pair = (uint64_t)i2 << 32 | i1;
1273 #endif
1275 atomic_set((uint64_t *)jmp_addr, pair);
1276 flush_icache_range(jmp_addr, jmp_addr + 8);
1278 #else
1279 void ppc_tb_set_jmp_target(uintptr_t jmp_addr, uintptr_t addr)
1281 intptr_t diff = addr - jmp_addr;
1282 tcg_debug_assert(in_range_b(diff));
1283 atomic_set((uint32_t *)jmp_addr, B | (diff & 0x3fffffc));
1284 flush_icache_range(jmp_addr, jmp_addr + 4);
1286 #endif
1288 static void tcg_out_call(TCGContext *s, tcg_insn_unit *target)
1290 #ifdef _CALL_AIX
1291 /* Look through the descriptor. If the branch is in range, and we
1292 don't have to spend too much effort on building the toc. */
1293 void *tgt = ((void **)target)[0];
1294 uintptr_t toc = ((uintptr_t *)target)[1];
1295 intptr_t diff = tcg_pcrel_diff(s, tgt);
1297 if (in_range_b(diff) && toc == (uint32_t)toc) {
1298 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, toc);
1299 tcg_out_b(s, LK, tgt);
1300 } else {
1301 /* Fold the low bits of the constant into the addresses below. */
1302 intptr_t arg = (intptr_t)target;
1303 int ofs = (int16_t)arg;
1305 if (ofs + 8 < 0x8000) {
1306 arg -= ofs;
1307 } else {
1308 ofs = 0;
1310 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, arg);
1311 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_TMP1, ofs);
1312 tcg_out32(s, MTSPR | RA(TCG_REG_R0) | CTR);
1313 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_REG_TMP1, ofs + SZP);
1314 tcg_out32(s, BCCTR | BO_ALWAYS | LK);
1316 #elif defined(_CALL_ELF) && _CALL_ELF == 2
1317 intptr_t diff;
1319 /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1320 address, which the callee uses to compute its TOC address. */
1321 /* FIXME: when the branch is in range, we could avoid r12 load if we
1322 knew that the destination uses the same TOC, and what its local
1323 entry point offset is. */
1324 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R12, (intptr_t)target);
1326 diff = tcg_pcrel_diff(s, target);
1327 if (in_range_b(diff)) {
1328 tcg_out_b(s, LK, target);
1329 } else {
1330 tcg_out32(s, MTSPR | RS(TCG_REG_R12) | CTR);
1331 tcg_out32(s, BCCTR | BO_ALWAYS | LK);
1333 #else
1334 tcg_out_b(s, LK, target);
1335 #endif
1338 static const uint32_t qemu_ldx_opc[16] = {
1339 [MO_UB] = LBZX,
1340 [MO_UW] = LHZX,
1341 [MO_UL] = LWZX,
1342 [MO_Q] = LDX,
1343 [MO_SW] = LHAX,
1344 [MO_SL] = LWAX,
1345 [MO_BSWAP | MO_UB] = LBZX,
1346 [MO_BSWAP | MO_UW] = LHBRX,
1347 [MO_BSWAP | MO_UL] = LWBRX,
1348 [MO_BSWAP | MO_Q] = LDBRX,
1351 static const uint32_t qemu_stx_opc[16] = {
1352 [MO_UB] = STBX,
1353 [MO_UW] = STHX,
1354 [MO_UL] = STWX,
1355 [MO_Q] = STDX,
1356 [MO_BSWAP | MO_UB] = STBX,
1357 [MO_BSWAP | MO_UW] = STHBRX,
1358 [MO_BSWAP | MO_UL] = STWBRX,
1359 [MO_BSWAP | MO_Q] = STDBRX,
1362 static const uint32_t qemu_exts_opc[4] = {
1363 EXTSB, EXTSH, EXTSW, 0
1366 #if defined (CONFIG_SOFTMMU)
1367 /* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
1368 * int mmu_idx, uintptr_t ra)
1370 static void * const qemu_ld_helpers[16] = {
1371 [MO_UB] = helper_ret_ldub_mmu,
1372 [MO_LEUW] = helper_le_lduw_mmu,
1373 [MO_LEUL] = helper_le_ldul_mmu,
1374 [MO_LEQ] = helper_le_ldq_mmu,
1375 [MO_BEUW] = helper_be_lduw_mmu,
1376 [MO_BEUL] = helper_be_ldul_mmu,
1377 [MO_BEQ] = helper_be_ldq_mmu,
1380 /* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
1381 * uintxx_t val, int mmu_idx, uintptr_t ra)
1383 static void * const qemu_st_helpers[16] = {
1384 [MO_UB] = helper_ret_stb_mmu,
1385 [MO_LEUW] = helper_le_stw_mmu,
1386 [MO_LEUL] = helper_le_stl_mmu,
1387 [MO_LEQ] = helper_le_stq_mmu,
1388 [MO_BEUW] = helper_be_stw_mmu,
1389 [MO_BEUL] = helper_be_stl_mmu,
1390 [MO_BEQ] = helper_be_stq_mmu,
1393 /* Perform the TLB load and compare. Places the result of the comparison
1394 in CR7, loads the addend of the TLB into R3, and returns the register
1395 containing the guest address (zero-extended into R4). Clobbers R0 and R2. */
1397 static TCGReg tcg_out_tlb_read(TCGContext *s, TCGMemOp opc,
1398 TCGReg addrlo, TCGReg addrhi,
1399 int mem_index, bool is_read)
1401 int cmp_off
1402 = (is_read
1403 ? offsetof(CPUArchState, tlb_table[mem_index][0].addr_read)
1404 : offsetof(CPUArchState, tlb_table[mem_index][0].addr_write));
1405 int add_off = offsetof(CPUArchState, tlb_table[mem_index][0].addend);
1406 TCGReg base = TCG_AREG0;
1407 TCGMemOp s_bits = opc & MO_SIZE;
1408 int a_bits = get_alignment_bits(opc);
1410 /* Extract the page index, shifted into place for tlb index. */
1411 if (TCG_TARGET_REG_BITS == 64) {
1412 if (TARGET_LONG_BITS == 32) {
1413 /* Zero-extend the address into a place helpful for further use. */
1414 tcg_out_ext32u(s, TCG_REG_R4, addrlo);
1415 addrlo = TCG_REG_R4;
1416 } else {
1417 tcg_out_rld(s, RLDICL, TCG_REG_R3, addrlo,
1418 64 - TARGET_PAGE_BITS, 64 - CPU_TLB_BITS);
1422 /* Compensate for very large offsets. */
1423 if (add_off >= 0x8000) {
1424 /* Most target env are smaller than 32k; none are larger than 64k.
1425 Simplify the logic here merely to offset by 0x7ff0, giving us a
1426 range just shy of 64k. Check this assumption. */
1427 QEMU_BUILD_BUG_ON(offsetof(CPUArchState,
1428 tlb_table[NB_MMU_MODES - 1][1])
1429 > 0x7ff0 + 0x7fff);
1430 tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, base, 0x7ff0));
1431 base = TCG_REG_TMP1;
1432 cmp_off -= 0x7ff0;
1433 add_off -= 0x7ff0;
1436 /* Extraction and shifting, part 2. */
1437 if (TCG_TARGET_REG_BITS == 32 || TARGET_LONG_BITS == 32) {
1438 tcg_out_rlw(s, RLWINM, TCG_REG_R3, addrlo,
1439 32 - (TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS),
1440 32 - (CPU_TLB_BITS + CPU_TLB_ENTRY_BITS),
1441 31 - CPU_TLB_ENTRY_BITS);
1442 } else {
1443 tcg_out_shli64(s, TCG_REG_R3, TCG_REG_R3, CPU_TLB_ENTRY_BITS);
1446 tcg_out32(s, ADD | TAB(TCG_REG_R3, TCG_REG_R3, base));
1448 /* Load the tlb comparator. */
1449 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1450 tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R4, TCG_REG_R3, cmp_off);
1451 tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP1, TCG_REG_R3, cmp_off + 4);
1452 } else {
1453 tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP1, TCG_REG_R3, cmp_off);
1456 /* Load the TLB addend for use on the fast path. Do this asap
1457 to minimize any load use delay. */
1458 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_REG_R3, add_off);
1460 /* Clear the non-page, non-alignment bits from the address */
1461 if (TCG_TARGET_REG_BITS == 32 || TARGET_LONG_BITS == 32) {
1462 /* We don't support unaligned accesses on 32-bits, preserve
1463 * the bottom bits and thus trigger a comparison failure on
1464 * unaligned accesses
1466 if (a_bits < 0) {
1467 a_bits = s_bits;
1469 tcg_out_rlw(s, RLWINM, TCG_REG_R0, addrlo, 0,
1470 (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
1471 } else if (a_bits) {
1472 /* More than byte access, we need to handle alignment */
1473 if (a_bits > 0) {
1474 /* Alignment required by the front-end, same as 32-bits */
1475 tcg_out_rld(s, RLDICL, TCG_REG_R0, addrlo,
1476 64 - TARGET_PAGE_BITS, TARGET_PAGE_BITS - a_bits);
1477 tcg_out_rld(s, RLDICL, TCG_REG_R0, TCG_REG_R0, TARGET_PAGE_BITS, 0);
1478 } else {
1479 /* We support unaligned accesses, we need to make sure we fail
1480 * if we cross a page boundary. The trick is to add the
1481 * access_size-1 to the address before masking the low bits.
1482 * That will make the address overflow to the next page if we
1483 * cross a page boundary which will then force a mismatch of
1484 * the TLB compare since the next page cannot possibly be in
1485 * the same TLB index.
1487 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, (1 << s_bits) - 1));
1488 tcg_out_rld(s, RLDICR, TCG_REG_R0, TCG_REG_R0,
1489 0, 63 - TARGET_PAGE_BITS);
1491 } else {
1492 /* Byte access, just chop off the bits below the page index */
1493 tcg_out_rld(s, RLDICR, TCG_REG_R0, addrlo, 0, 63 - TARGET_PAGE_BITS);
1496 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1497 tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
1498 0, 7, TCG_TYPE_I32);
1499 tcg_out_cmp(s, TCG_COND_EQ, addrhi, TCG_REG_R4, 0, 6, TCG_TYPE_I32);
1500 tcg_out32(s, CRAND | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1501 } else {
1502 tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
1503 0, 7, TCG_TYPE_TL);
1506 return addrlo;
1509 /* Record the context of a call to the out of line helper code for the slow
1510 path for a load or store, so that we can later generate the correct
1511 helper code. */
1512 static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
1513 TCGReg datalo_reg, TCGReg datahi_reg,
1514 TCGReg addrlo_reg, TCGReg addrhi_reg,
1515 tcg_insn_unit *raddr, tcg_insn_unit *lptr)
1517 TCGLabelQemuLdst *label = new_ldst_label(s);
1519 label->is_ld = is_ld;
1520 label->oi = oi;
1521 label->datalo_reg = datalo_reg;
1522 label->datahi_reg = datahi_reg;
1523 label->addrlo_reg = addrlo_reg;
1524 label->addrhi_reg = addrhi_reg;
1525 label->raddr = raddr;
1526 label->label_ptr[0] = lptr;
1529 static void tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1531 TCGMemOpIdx oi = lb->oi;
1532 TCGMemOp opc = get_memop(oi);
1533 TCGReg hi, lo, arg = TCG_REG_R3;
1535 reloc_pc14(lb->label_ptr[0], s->code_ptr);
1537 tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
1539 lo = lb->addrlo_reg;
1540 hi = lb->addrhi_reg;
1541 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1542 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1543 arg |= 1;
1544 #endif
1545 tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
1546 tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
1547 } else {
1548 /* If the address needed to be zero-extended, we'll have already
1549 placed it in R4. The only remaining case is 64-bit guest. */
1550 tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
1553 tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
1554 tcg_out32(s, MFSPR | RT(arg) | LR);
1556 tcg_out_call(s, qemu_ld_helpers[opc & (MO_BSWAP | MO_SIZE)]);
1558 lo = lb->datalo_reg;
1559 hi = lb->datahi_reg;
1560 if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
1561 tcg_out_mov(s, TCG_TYPE_I32, lo, TCG_REG_R4);
1562 tcg_out_mov(s, TCG_TYPE_I32, hi, TCG_REG_R3);
1563 } else if (opc & MO_SIGN) {
1564 uint32_t insn = qemu_exts_opc[opc & MO_SIZE];
1565 tcg_out32(s, insn | RA(lo) | RS(TCG_REG_R3));
1566 } else {
1567 tcg_out_mov(s, TCG_TYPE_REG, lo, TCG_REG_R3);
1570 tcg_out_b(s, 0, lb->raddr);
1573 static void tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1575 TCGMemOpIdx oi = lb->oi;
1576 TCGMemOp opc = get_memop(oi);
1577 TCGMemOp s_bits = opc & MO_SIZE;
1578 TCGReg hi, lo, arg = TCG_REG_R3;
1580 reloc_pc14(lb->label_ptr[0], s->code_ptr);
1582 tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
1584 lo = lb->addrlo_reg;
1585 hi = lb->addrhi_reg;
1586 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1587 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1588 arg |= 1;
1589 #endif
1590 tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
1591 tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
1592 } else {
1593 /* If the address needed to be zero-extended, we'll have already
1594 placed it in R4. The only remaining case is 64-bit guest. */
1595 tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
1598 lo = lb->datalo_reg;
1599 hi = lb->datahi_reg;
1600 if (TCG_TARGET_REG_BITS == 32) {
1601 switch (s_bits) {
1602 case MO_64:
1603 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1604 arg |= 1;
1605 #endif
1606 tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
1607 /* FALLTHRU */
1608 case MO_32:
1609 tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
1610 break;
1611 default:
1612 tcg_out_rlw(s, RLWINM, arg++, lo, 0, 32 - (8 << s_bits), 31);
1613 break;
1615 } else {
1616 if (s_bits == MO_64) {
1617 tcg_out_mov(s, TCG_TYPE_I64, arg++, lo);
1618 } else {
1619 tcg_out_rld(s, RLDICL, arg++, lo, 0, 64 - (8 << s_bits));
1623 tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
1624 tcg_out32(s, MFSPR | RT(arg) | LR);
1626 tcg_out_call(s, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
1628 tcg_out_b(s, 0, lb->raddr);
1630 #endif /* SOFTMMU */
1632 static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64)
1634 TCGReg datalo, datahi, addrlo, rbase;
1635 TCGReg addrhi __attribute__((unused));
1636 TCGMemOpIdx oi;
1637 TCGMemOp opc, s_bits;
1638 #ifdef CONFIG_SOFTMMU
1639 int mem_index;
1640 tcg_insn_unit *label_ptr;
1641 #endif
1643 datalo = *args++;
1644 datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
1645 addrlo = *args++;
1646 addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
1647 oi = *args++;
1648 opc = get_memop(oi);
1649 s_bits = opc & MO_SIZE;
1651 #ifdef CONFIG_SOFTMMU
1652 mem_index = get_mmuidx(oi);
1653 addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, true);
1655 /* Load a pointer into the current opcode w/conditional branch-link. */
1656 label_ptr = s->code_ptr;
1657 tcg_out_bc_noaddr(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
1659 rbase = TCG_REG_R3;
1660 #else /* !CONFIG_SOFTMMU */
1661 rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
1662 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
1663 tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
1664 addrlo = TCG_REG_TMP1;
1666 #endif
1668 if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
1669 if (opc & MO_BSWAP) {
1670 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1671 tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
1672 tcg_out32(s, LWBRX | TAB(datahi, rbase, TCG_REG_R0));
1673 } else if (rbase != 0) {
1674 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1675 tcg_out32(s, LWZX | TAB(datahi, rbase, addrlo));
1676 tcg_out32(s, LWZX | TAB(datalo, rbase, TCG_REG_R0));
1677 } else if (addrlo == datahi) {
1678 tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
1679 tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
1680 } else {
1681 tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
1682 tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
1684 } else {
1685 uint32_t insn = qemu_ldx_opc[opc & (MO_BSWAP | MO_SSIZE)];
1686 if (!HAVE_ISA_2_06 && insn == LDBRX) {
1687 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1688 tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
1689 tcg_out32(s, LWBRX | TAB(TCG_REG_R0, rbase, TCG_REG_R0));
1690 tcg_out_rld(s, RLDIMI, datalo, TCG_REG_R0, 32, 0);
1691 } else if (insn) {
1692 tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
1693 } else {
1694 insn = qemu_ldx_opc[opc & (MO_SIZE | MO_BSWAP)];
1695 tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
1696 insn = qemu_exts_opc[s_bits];
1697 tcg_out32(s, insn | RA(datalo) | RS(datalo));
1701 #ifdef CONFIG_SOFTMMU
1702 add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
1703 s->code_ptr, label_ptr);
1704 #endif
1707 static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
1709 TCGReg datalo, datahi, addrlo, rbase;
1710 TCGReg addrhi __attribute__((unused));
1711 TCGMemOpIdx oi;
1712 TCGMemOp opc, s_bits;
1713 #ifdef CONFIG_SOFTMMU
1714 int mem_index;
1715 tcg_insn_unit *label_ptr;
1716 #endif
1718 datalo = *args++;
1719 datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
1720 addrlo = *args++;
1721 addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
1722 oi = *args++;
1723 opc = get_memop(oi);
1724 s_bits = opc & MO_SIZE;
1726 #ifdef CONFIG_SOFTMMU
1727 mem_index = get_mmuidx(oi);
1728 addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, false);
1730 /* Load a pointer into the current opcode w/conditional branch-link. */
1731 label_ptr = s->code_ptr;
1732 tcg_out_bc_noaddr(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
1734 rbase = TCG_REG_R3;
1735 #else /* !CONFIG_SOFTMMU */
1736 rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
1737 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
1738 tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
1739 addrlo = TCG_REG_TMP1;
1741 #endif
1743 if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
1744 if (opc & MO_BSWAP) {
1745 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1746 tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
1747 tcg_out32(s, STWBRX | SAB(datahi, rbase, TCG_REG_R0));
1748 } else if (rbase != 0) {
1749 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1750 tcg_out32(s, STWX | SAB(datahi, rbase, addrlo));
1751 tcg_out32(s, STWX | SAB(datalo, rbase, TCG_REG_R0));
1752 } else {
1753 tcg_out32(s, STW | TAI(datahi, addrlo, 0));
1754 tcg_out32(s, STW | TAI(datalo, addrlo, 4));
1756 } else {
1757 uint32_t insn = qemu_stx_opc[opc & (MO_BSWAP | MO_SIZE)];
1758 if (!HAVE_ISA_2_06 && insn == STDBRX) {
1759 tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
1760 tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, addrlo, 4));
1761 tcg_out_shri64(s, TCG_REG_R0, datalo, 32);
1762 tcg_out32(s, STWBRX | SAB(TCG_REG_R0, rbase, TCG_REG_TMP1));
1763 } else {
1764 tcg_out32(s, insn | SAB(datalo, rbase, addrlo));
1768 #ifdef CONFIG_SOFTMMU
1769 add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
1770 s->code_ptr, label_ptr);
1771 #endif
1774 /* Parameters for function call generation, used in tcg.c. */
1775 #define TCG_TARGET_STACK_ALIGN 16
1776 #define TCG_TARGET_EXTEND_ARGS 1
1778 #ifdef _CALL_AIX
1779 # define LINK_AREA_SIZE (6 * SZR)
1780 # define LR_OFFSET (1 * SZR)
1781 # define TCG_TARGET_CALL_STACK_OFFSET (LINK_AREA_SIZE + 8 * SZR)
1782 #elif defined(TCG_TARGET_CALL_DARWIN)
1783 # define LINK_AREA_SIZE (6 * SZR)
1784 # define LR_OFFSET (2 * SZR)
1785 #elif TCG_TARGET_REG_BITS == 64
1786 # if defined(_CALL_ELF) && _CALL_ELF == 2
1787 # define LINK_AREA_SIZE (4 * SZR)
1788 # define LR_OFFSET (1 * SZR)
1789 # endif
1790 #else /* TCG_TARGET_REG_BITS == 32 */
1791 # if defined(_CALL_SYSV)
1792 # define LINK_AREA_SIZE (2 * SZR)
1793 # define LR_OFFSET (1 * SZR)
1794 # endif
1795 #endif
1796 #ifndef LR_OFFSET
1797 # error "Unhandled abi"
1798 #endif
1799 #ifndef TCG_TARGET_CALL_STACK_OFFSET
1800 # define TCG_TARGET_CALL_STACK_OFFSET LINK_AREA_SIZE
1801 #endif
1803 #define CPU_TEMP_BUF_SIZE (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
1804 #define REG_SAVE_SIZE ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
1806 #define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET \
1807 + TCG_STATIC_CALL_ARGS_SIZE \
1808 + CPU_TEMP_BUF_SIZE \
1809 + REG_SAVE_SIZE \
1810 + TCG_TARGET_STACK_ALIGN - 1) \
1811 & -TCG_TARGET_STACK_ALIGN)
1813 #define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
1815 static void tcg_target_qemu_prologue(TCGContext *s)
1817 int i;
1819 #ifdef _CALL_AIX
1820 void **desc = (void **)s->code_ptr;
1821 desc[0] = desc + 2; /* entry point */
1822 desc[1] = 0; /* environment pointer */
1823 s->code_ptr = (void *)(desc + 2); /* skip over descriptor */
1824 #endif
1826 tcg_set_frame(s, TCG_REG_CALL_STACK, REG_SAVE_BOT - CPU_TEMP_BUF_SIZE,
1827 CPU_TEMP_BUF_SIZE);
1829 /* Prologue */
1830 tcg_out32(s, MFSPR | RT(TCG_REG_R0) | LR);
1831 tcg_out32(s, (SZR == 8 ? STDU : STWU)
1832 | SAI(TCG_REG_R1, TCG_REG_R1, -FRAME_SIZE));
1834 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
1835 tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
1836 TCG_REG_R1, REG_SAVE_BOT + i * SZR);
1838 tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
1840 #ifndef CONFIG_SOFTMMU
1841 if (guest_base) {
1842 tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
1843 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
1845 #endif
1847 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
1848 tcg_out32(s, MTSPR | RS(tcg_target_call_iarg_regs[1]) | CTR);
1850 if (USE_REG_RA) {
1851 #ifdef _CALL_AIX
1852 /* Make the caller load the value as the TOC into R2. */
1853 tb_ret_addr = s->code_ptr + 2;
1854 desc[1] = tb_ret_addr;
1855 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_RA, TCG_REG_R2);
1856 tcg_out32(s, BCCTR | BO_ALWAYS);
1857 #elif defined(_CALL_ELF) && _CALL_ELF == 2
1858 /* Compute from the incoming R12 value. */
1859 tb_ret_addr = s->code_ptr + 2;
1860 tcg_out32(s, ADDI | TAI(TCG_REG_RA, TCG_REG_R12,
1861 tcg_ptr_byte_diff(tb_ret_addr, s->code_buf)));
1862 tcg_out32(s, BCCTR | BO_ALWAYS);
1863 #else
1864 /* Reserve max 5 insns for the constant load. */
1865 tb_ret_addr = s->code_ptr + 6;
1866 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_RA, (intptr_t)tb_ret_addr);
1867 tcg_out32(s, BCCTR | BO_ALWAYS);
1868 while (s->code_ptr < tb_ret_addr) {
1869 tcg_out32(s, NOP);
1871 #endif
1872 } else {
1873 tcg_out32(s, BCCTR | BO_ALWAYS);
1874 tb_ret_addr = s->code_ptr;
1877 /* Epilogue */
1878 tcg_debug_assert(tb_ret_addr == s->code_ptr);
1880 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
1881 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
1882 tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
1883 TCG_REG_R1, REG_SAVE_BOT + i * SZR);
1885 tcg_out32(s, MTSPR | RS(TCG_REG_R0) | LR);
1886 tcg_out32(s, ADDI | TAI(TCG_REG_R1, TCG_REG_R1, FRAME_SIZE));
1887 tcg_out32(s, BCLR | BO_ALWAYS);
1890 static void tcg_out_op(TCGContext *s, TCGOpcode opc, const TCGArg *args,
1891 const int *const_args)
1893 TCGArg a0, a1, a2;
1894 int c;
1896 switch (opc) {
1897 case INDEX_op_exit_tb:
1898 if (USE_REG_RA) {
1899 ptrdiff_t disp = tcg_pcrel_diff(s, tb_ret_addr);
1901 /* Use a direct branch if we can, otherwise use the value in RA.
1902 Note that the direct branch is always backward, thus we need
1903 to account for the possibility of 5 insns from the movi. */
1904 if (!in_range_b(disp - 20)) {
1905 tcg_out32(s, MTSPR | RS(TCG_REG_RA) | CTR);
1906 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, args[0]);
1907 tcg_out32(s, BCCTR | BO_ALWAYS);
1908 break;
1911 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, args[0]);
1912 tcg_out_b(s, 0, tb_ret_addr);
1913 break;
1914 case INDEX_op_goto_tb:
1915 tcg_debug_assert(s->tb_jmp_insn_offset);
1916 /* Direct jump. */
1917 #ifdef __powerpc64__
1918 /* Ensure the next insns are 8-byte aligned. */
1919 if ((uintptr_t)s->code_ptr & 7) {
1920 tcg_out32(s, NOP);
1922 s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
1923 /* To be replaced by either a branch+nop or a load into TMP1. */
1924 s->code_ptr += 2;
1925 tcg_out32(s, MTSPR | RS(TCG_REG_TMP1) | CTR);
1926 tcg_out32(s, BCCTR | BO_ALWAYS);
1927 #else
1928 /* To be replaced by a branch. */
1929 s->code_ptr++;
1930 #endif
1931 s->tb_jmp_reset_offset[args[0]] = tcg_current_code_size(s);
1932 break;
1933 case INDEX_op_br:
1935 TCGLabel *l = arg_label(args[0]);
1937 if (l->has_value) {
1938 tcg_out_b(s, 0, l->u.value_ptr);
1939 } else {
1940 tcg_out_reloc(s, s->code_ptr, R_PPC_REL24, l, 0);
1941 tcg_out_b_noaddr(s, B);
1944 break;
1945 case INDEX_op_ld8u_i32:
1946 case INDEX_op_ld8u_i64:
1947 tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
1948 break;
1949 case INDEX_op_ld8s_i32:
1950 case INDEX_op_ld8s_i64:
1951 tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
1952 tcg_out32(s, EXTSB | RS(args[0]) | RA(args[0]));
1953 break;
1954 case INDEX_op_ld16u_i32:
1955 case INDEX_op_ld16u_i64:
1956 tcg_out_mem_long(s, LHZ, LHZX, args[0], args[1], args[2]);
1957 break;
1958 case INDEX_op_ld16s_i32:
1959 case INDEX_op_ld16s_i64:
1960 tcg_out_mem_long(s, LHA, LHAX, args[0], args[1], args[2]);
1961 break;
1962 case INDEX_op_ld_i32:
1963 case INDEX_op_ld32u_i64:
1964 tcg_out_mem_long(s, LWZ, LWZX, args[0], args[1], args[2]);
1965 break;
1966 case INDEX_op_ld32s_i64:
1967 tcg_out_mem_long(s, LWA, LWAX, args[0], args[1], args[2]);
1968 break;
1969 case INDEX_op_ld_i64:
1970 tcg_out_mem_long(s, LD, LDX, args[0], args[1], args[2]);
1971 break;
1972 case INDEX_op_st8_i32:
1973 case INDEX_op_st8_i64:
1974 tcg_out_mem_long(s, STB, STBX, args[0], args[1], args[2]);
1975 break;
1976 case INDEX_op_st16_i32:
1977 case INDEX_op_st16_i64:
1978 tcg_out_mem_long(s, STH, STHX, args[0], args[1], args[2]);
1979 break;
1980 case INDEX_op_st_i32:
1981 case INDEX_op_st32_i64:
1982 tcg_out_mem_long(s, STW, STWX, args[0], args[1], args[2]);
1983 break;
1984 case INDEX_op_st_i64:
1985 tcg_out_mem_long(s, STD, STDX, args[0], args[1], args[2]);
1986 break;
1988 case INDEX_op_add_i32:
1989 a0 = args[0], a1 = args[1], a2 = args[2];
1990 if (const_args[2]) {
1991 do_addi_32:
1992 tcg_out_mem_long(s, ADDI, ADD, a0, a1, (int32_t)a2);
1993 } else {
1994 tcg_out32(s, ADD | TAB(a0, a1, a2));
1996 break;
1997 case INDEX_op_sub_i32:
1998 a0 = args[0], a1 = args[1], a2 = args[2];
1999 if (const_args[1]) {
2000 if (const_args[2]) {
2001 tcg_out_movi(s, TCG_TYPE_I32, a0, a1 - a2);
2002 } else {
2003 tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2005 } else if (const_args[2]) {
2006 a2 = -a2;
2007 goto do_addi_32;
2008 } else {
2009 tcg_out32(s, SUBF | TAB(a0, a2, a1));
2011 break;
2013 case INDEX_op_and_i32:
2014 a0 = args[0], a1 = args[1], a2 = args[2];
2015 if (const_args[2]) {
2016 tcg_out_andi32(s, a0, a1, a2);
2017 } else {
2018 tcg_out32(s, AND | SAB(a1, a0, a2));
2020 break;
2021 case INDEX_op_and_i64:
2022 a0 = args[0], a1 = args[1], a2 = args[2];
2023 if (const_args[2]) {
2024 tcg_out_andi64(s, a0, a1, a2);
2025 } else {
2026 tcg_out32(s, AND | SAB(a1, a0, a2));
2028 break;
2029 case INDEX_op_or_i64:
2030 case INDEX_op_or_i32:
2031 a0 = args[0], a1 = args[1], a2 = args[2];
2032 if (const_args[2]) {
2033 tcg_out_ori32(s, a0, a1, a2);
2034 } else {
2035 tcg_out32(s, OR | SAB(a1, a0, a2));
2037 break;
2038 case INDEX_op_xor_i64:
2039 case INDEX_op_xor_i32:
2040 a0 = args[0], a1 = args[1], a2 = args[2];
2041 if (const_args[2]) {
2042 tcg_out_xori32(s, a0, a1, a2);
2043 } else {
2044 tcg_out32(s, XOR | SAB(a1, a0, a2));
2046 break;
2047 case INDEX_op_andc_i32:
2048 a0 = args[0], a1 = args[1], a2 = args[2];
2049 if (const_args[2]) {
2050 tcg_out_andi32(s, a0, a1, ~a2);
2051 } else {
2052 tcg_out32(s, ANDC | SAB(a1, a0, a2));
2054 break;
2055 case INDEX_op_andc_i64:
2056 a0 = args[0], a1 = args[1], a2 = args[2];
2057 if (const_args[2]) {
2058 tcg_out_andi64(s, a0, a1, ~a2);
2059 } else {
2060 tcg_out32(s, ANDC | SAB(a1, a0, a2));
2062 break;
2063 case INDEX_op_orc_i32:
2064 if (const_args[2]) {
2065 tcg_out_ori32(s, args[0], args[1], ~args[2]);
2066 break;
2068 /* FALLTHRU */
2069 case INDEX_op_orc_i64:
2070 tcg_out32(s, ORC | SAB(args[1], args[0], args[2]));
2071 break;
2072 case INDEX_op_eqv_i32:
2073 if (const_args[2]) {
2074 tcg_out_xori32(s, args[0], args[1], ~args[2]);
2075 break;
2077 /* FALLTHRU */
2078 case INDEX_op_eqv_i64:
2079 tcg_out32(s, EQV | SAB(args[1], args[0], args[2]));
2080 break;
2081 case INDEX_op_nand_i32:
2082 case INDEX_op_nand_i64:
2083 tcg_out32(s, NAND | SAB(args[1], args[0], args[2]));
2084 break;
2085 case INDEX_op_nor_i32:
2086 case INDEX_op_nor_i64:
2087 tcg_out32(s, NOR | SAB(args[1], args[0], args[2]));
2088 break;
2090 case INDEX_op_mul_i32:
2091 a0 = args[0], a1 = args[1], a2 = args[2];
2092 if (const_args[2]) {
2093 tcg_out32(s, MULLI | TAI(a0, a1, a2));
2094 } else {
2095 tcg_out32(s, MULLW | TAB(a0, a1, a2));
2097 break;
2099 case INDEX_op_div_i32:
2100 tcg_out32(s, DIVW | TAB(args[0], args[1], args[2]));
2101 break;
2103 case INDEX_op_divu_i32:
2104 tcg_out32(s, DIVWU | TAB(args[0], args[1], args[2]));
2105 break;
2107 case INDEX_op_shl_i32:
2108 if (const_args[2]) {
2109 tcg_out_shli32(s, args[0], args[1], args[2]);
2110 } else {
2111 tcg_out32(s, SLW | SAB(args[1], args[0], args[2]));
2113 break;
2114 case INDEX_op_shr_i32:
2115 if (const_args[2]) {
2116 tcg_out_shri32(s, args[0], args[1], args[2]);
2117 } else {
2118 tcg_out32(s, SRW | SAB(args[1], args[0], args[2]));
2120 break;
2121 case INDEX_op_sar_i32:
2122 if (const_args[2]) {
2123 tcg_out32(s, SRAWI | RS(args[1]) | RA(args[0]) | SH(args[2]));
2124 } else {
2125 tcg_out32(s, SRAW | SAB(args[1], args[0], args[2]));
2127 break;
2128 case INDEX_op_rotl_i32:
2129 if (const_args[2]) {
2130 tcg_out_rlw(s, RLWINM, args[0], args[1], args[2], 0, 31);
2131 } else {
2132 tcg_out32(s, RLWNM | SAB(args[1], args[0], args[2])
2133 | MB(0) | ME(31));
2135 break;
2136 case INDEX_op_rotr_i32:
2137 if (const_args[2]) {
2138 tcg_out_rlw(s, RLWINM, args[0], args[1], 32 - args[2], 0, 31);
2139 } else {
2140 tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 32));
2141 tcg_out32(s, RLWNM | SAB(args[1], args[0], TCG_REG_R0)
2142 | MB(0) | ME(31));
2144 break;
2146 case INDEX_op_brcond_i32:
2147 tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2148 arg_label(args[3]), TCG_TYPE_I32);
2149 break;
2150 case INDEX_op_brcond_i64:
2151 tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2152 arg_label(args[3]), TCG_TYPE_I64);
2153 break;
2154 case INDEX_op_brcond2_i32:
2155 tcg_out_brcond2(s, args, const_args);
2156 break;
2158 case INDEX_op_neg_i32:
2159 case INDEX_op_neg_i64:
2160 tcg_out32(s, NEG | RT(args[0]) | RA(args[1]));
2161 break;
2163 case INDEX_op_not_i32:
2164 case INDEX_op_not_i64:
2165 tcg_out32(s, NOR | SAB(args[1], args[0], args[1]));
2166 break;
2168 case INDEX_op_add_i64:
2169 a0 = args[0], a1 = args[1], a2 = args[2];
2170 if (const_args[2]) {
2171 do_addi_64:
2172 tcg_out_mem_long(s, ADDI, ADD, a0, a1, a2);
2173 } else {
2174 tcg_out32(s, ADD | TAB(a0, a1, a2));
2176 break;
2177 case INDEX_op_sub_i64:
2178 a0 = args[0], a1 = args[1], a2 = args[2];
2179 if (const_args[1]) {
2180 if (const_args[2]) {
2181 tcg_out_movi(s, TCG_TYPE_I64, a0, a1 - a2);
2182 } else {
2183 tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2185 } else if (const_args[2]) {
2186 a2 = -a2;
2187 goto do_addi_64;
2188 } else {
2189 tcg_out32(s, SUBF | TAB(a0, a2, a1));
2191 break;
2193 case INDEX_op_shl_i64:
2194 if (const_args[2]) {
2195 tcg_out_shli64(s, args[0], args[1], args[2]);
2196 } else {
2197 tcg_out32(s, SLD | SAB(args[1], args[0], args[2]));
2199 break;
2200 case INDEX_op_shr_i64:
2201 if (const_args[2]) {
2202 tcg_out_shri64(s, args[0], args[1], args[2]);
2203 } else {
2204 tcg_out32(s, SRD | SAB(args[1], args[0], args[2]));
2206 break;
2207 case INDEX_op_sar_i64:
2208 if (const_args[2]) {
2209 int sh = SH(args[2] & 0x1f) | (((args[2] >> 5) & 1) << 1);
2210 tcg_out32(s, SRADI | RA(args[0]) | RS(args[1]) | sh);
2211 } else {
2212 tcg_out32(s, SRAD | SAB(args[1], args[0], args[2]));
2214 break;
2215 case INDEX_op_rotl_i64:
2216 if (const_args[2]) {
2217 tcg_out_rld(s, RLDICL, args[0], args[1], args[2], 0);
2218 } else {
2219 tcg_out32(s, RLDCL | SAB(args[1], args[0], args[2]) | MB64(0));
2221 break;
2222 case INDEX_op_rotr_i64:
2223 if (const_args[2]) {
2224 tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 0);
2225 } else {
2226 tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 64));
2227 tcg_out32(s, RLDCL | SAB(args[1], args[0], TCG_REG_R0) | MB64(0));
2229 break;
2231 case INDEX_op_mul_i64:
2232 a0 = args[0], a1 = args[1], a2 = args[2];
2233 if (const_args[2]) {
2234 tcg_out32(s, MULLI | TAI(a0, a1, a2));
2235 } else {
2236 tcg_out32(s, MULLD | TAB(a0, a1, a2));
2238 break;
2239 case INDEX_op_div_i64:
2240 tcg_out32(s, DIVD | TAB(args[0], args[1], args[2]));
2241 break;
2242 case INDEX_op_divu_i64:
2243 tcg_out32(s, DIVDU | TAB(args[0], args[1], args[2]));
2244 break;
2246 case INDEX_op_qemu_ld_i32:
2247 tcg_out_qemu_ld(s, args, false);
2248 break;
2249 case INDEX_op_qemu_ld_i64:
2250 tcg_out_qemu_ld(s, args, true);
2251 break;
2252 case INDEX_op_qemu_st_i32:
2253 tcg_out_qemu_st(s, args, false);
2254 break;
2255 case INDEX_op_qemu_st_i64:
2256 tcg_out_qemu_st(s, args, true);
2257 break;
2259 case INDEX_op_ext8s_i32:
2260 case INDEX_op_ext8s_i64:
2261 c = EXTSB;
2262 goto gen_ext;
2263 case INDEX_op_ext16s_i32:
2264 case INDEX_op_ext16s_i64:
2265 c = EXTSH;
2266 goto gen_ext;
2267 case INDEX_op_ext_i32_i64:
2268 case INDEX_op_ext32s_i64:
2269 c = EXTSW;
2270 goto gen_ext;
2271 gen_ext:
2272 tcg_out32(s, c | RS(args[1]) | RA(args[0]));
2273 break;
2274 case INDEX_op_extu_i32_i64:
2275 tcg_out_ext32u(s, args[0], args[1]);
2276 break;
2278 case INDEX_op_setcond_i32:
2279 tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2],
2280 const_args[2]);
2281 break;
2282 case INDEX_op_setcond_i64:
2283 tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2],
2284 const_args[2]);
2285 break;
2286 case INDEX_op_setcond2_i32:
2287 tcg_out_setcond2(s, args, const_args);
2288 break;
2290 case INDEX_op_bswap16_i32:
2291 case INDEX_op_bswap16_i64:
2292 a0 = args[0], a1 = args[1];
2293 /* a1 = abcd */
2294 if (a0 != a1) {
2295 /* a0 = (a1 r<< 24) & 0xff # 000c */
2296 tcg_out_rlw(s, RLWINM, a0, a1, 24, 24, 31);
2297 /* a0 = (a0 & ~0xff00) | (a1 r<< 8) & 0xff00 # 00dc */
2298 tcg_out_rlw(s, RLWIMI, a0, a1, 8, 16, 23);
2299 } else {
2300 /* r0 = (a1 r<< 8) & 0xff00 # 00d0 */
2301 tcg_out_rlw(s, RLWINM, TCG_REG_R0, a1, 8, 16, 23);
2302 /* a0 = (a1 r<< 24) & 0xff # 000c */
2303 tcg_out_rlw(s, RLWINM, a0, a1, 24, 24, 31);
2304 /* a0 = a0 | r0 # 00dc */
2305 tcg_out32(s, OR | SAB(TCG_REG_R0, a0, a0));
2307 break;
2309 case INDEX_op_bswap32_i32:
2310 case INDEX_op_bswap32_i64:
2311 /* Stolen from gcc's builtin_bswap32 */
2312 a1 = args[1];
2313 a0 = args[0] == a1 ? TCG_REG_R0 : args[0];
2315 /* a1 = args[1] # abcd */
2316 /* a0 = rotate_left (a1, 8) # bcda */
2317 tcg_out_rlw(s, RLWINM, a0, a1, 8, 0, 31);
2318 /* a0 = (a0 & ~0xff000000) | ((a1 r<< 24) & 0xff000000) # dcda */
2319 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 0, 7);
2320 /* a0 = (a0 & ~0x0000ff00) | ((a1 r<< 24) & 0x0000ff00) # dcba */
2321 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 16, 23);
2323 if (a0 == TCG_REG_R0) {
2324 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2326 break;
2328 case INDEX_op_bswap64_i64:
2329 a0 = args[0], a1 = args[1], a2 = TCG_REG_R0;
2330 if (a0 == a1) {
2331 a0 = TCG_REG_R0;
2332 a2 = a1;
2335 /* a1 = # abcd efgh */
2336 /* a0 = rl32(a1, 8) # 0000 fghe */
2337 tcg_out_rlw(s, RLWINM, a0, a1, 8, 0, 31);
2338 /* a0 = dep(a0, rl32(a1, 24), 0xff000000) # 0000 hghe */
2339 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 0, 7);
2340 /* a0 = dep(a0, rl32(a1, 24), 0x0000ff00) # 0000 hgfe */
2341 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 16, 23);
2343 /* a0 = rl64(a0, 32) # hgfe 0000 */
2344 /* a2 = rl64(a1, 32) # efgh abcd */
2345 tcg_out_rld(s, RLDICL, a0, a0, 32, 0);
2346 tcg_out_rld(s, RLDICL, a2, a1, 32, 0);
2348 /* a0 = dep(a0, rl32(a2, 8), 0xffffffff) # hgfe bcda */
2349 tcg_out_rlw(s, RLWIMI, a0, a2, 8, 0, 31);
2350 /* a0 = dep(a0, rl32(a2, 24), 0xff000000) # hgfe dcda */
2351 tcg_out_rlw(s, RLWIMI, a0, a2, 24, 0, 7);
2352 /* a0 = dep(a0, rl32(a2, 24), 0x0000ff00) # hgfe dcba */
2353 tcg_out_rlw(s, RLWIMI, a0, a2, 24, 16, 23);
2355 if (a0 == 0) {
2356 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2358 break;
2360 case INDEX_op_deposit_i32:
2361 if (const_args[2]) {
2362 uint32_t mask = ((2u << (args[4] - 1)) - 1) << args[3];
2363 tcg_out_andi32(s, args[0], args[0], ~mask);
2364 } else {
2365 tcg_out_rlw(s, RLWIMI, args[0], args[2], args[3],
2366 32 - args[3] - args[4], 31 - args[3]);
2368 break;
2369 case INDEX_op_deposit_i64:
2370 if (const_args[2]) {
2371 uint64_t mask = ((2ull << (args[4] - 1)) - 1) << args[3];
2372 tcg_out_andi64(s, args[0], args[0], ~mask);
2373 } else {
2374 tcg_out_rld(s, RLDIMI, args[0], args[2], args[3],
2375 64 - args[3] - args[4]);
2377 break;
2379 case INDEX_op_movcond_i32:
2380 tcg_out_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], args[2],
2381 args[3], args[4], const_args[2]);
2382 break;
2383 case INDEX_op_movcond_i64:
2384 tcg_out_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], args[2],
2385 args[3], args[4], const_args[2]);
2386 break;
2388 #if TCG_TARGET_REG_BITS == 64
2389 case INDEX_op_add2_i64:
2390 #else
2391 case INDEX_op_add2_i32:
2392 #endif
2393 /* Note that the CA bit is defined based on the word size of the
2394 environment. So in 64-bit mode it's always carry-out of bit 63.
2395 The fallback code using deposit works just as well for 32-bit. */
2396 a0 = args[0], a1 = args[1];
2397 if (a0 == args[3] || (!const_args[5] && a0 == args[5])) {
2398 a0 = TCG_REG_R0;
2400 if (const_args[4]) {
2401 tcg_out32(s, ADDIC | TAI(a0, args[2], args[4]));
2402 } else {
2403 tcg_out32(s, ADDC | TAB(a0, args[2], args[4]));
2405 if (const_args[5]) {
2406 tcg_out32(s, (args[5] ? ADDME : ADDZE) | RT(a1) | RA(args[3]));
2407 } else {
2408 tcg_out32(s, ADDE | TAB(a1, args[3], args[5]));
2410 if (a0 != args[0]) {
2411 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2413 break;
2415 #if TCG_TARGET_REG_BITS == 64
2416 case INDEX_op_sub2_i64:
2417 #else
2418 case INDEX_op_sub2_i32:
2419 #endif
2420 a0 = args[0], a1 = args[1];
2421 if (a0 == args[5] || (!const_args[3] && a0 == args[3])) {
2422 a0 = TCG_REG_R0;
2424 if (const_args[2]) {
2425 tcg_out32(s, SUBFIC | TAI(a0, args[4], args[2]));
2426 } else {
2427 tcg_out32(s, SUBFC | TAB(a0, args[4], args[2]));
2429 if (const_args[3]) {
2430 tcg_out32(s, (args[3] ? SUBFME : SUBFZE) | RT(a1) | RA(args[5]));
2431 } else {
2432 tcg_out32(s, SUBFE | TAB(a1, args[5], args[3]));
2434 if (a0 != args[0]) {
2435 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2437 break;
2439 case INDEX_op_muluh_i32:
2440 tcg_out32(s, MULHWU | TAB(args[0], args[1], args[2]));
2441 break;
2442 case INDEX_op_mulsh_i32:
2443 tcg_out32(s, MULHW | TAB(args[0], args[1], args[2]));
2444 break;
2445 case INDEX_op_muluh_i64:
2446 tcg_out32(s, MULHDU | TAB(args[0], args[1], args[2]));
2447 break;
2448 case INDEX_op_mulsh_i64:
2449 tcg_out32(s, MULHD | TAB(args[0], args[1], args[2]));
2450 break;
2452 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */
2453 case INDEX_op_mov_i64:
2454 case INDEX_op_movi_i32: /* Always emitted via tcg_out_movi. */
2455 case INDEX_op_movi_i64:
2456 case INDEX_op_call: /* Always emitted via tcg_out_call. */
2457 default:
2458 tcg_abort();
2462 static const TCGTargetOpDef ppc_op_defs[] = {
2463 { INDEX_op_exit_tb, { } },
2464 { INDEX_op_goto_tb, { } },
2465 { INDEX_op_br, { } },
2467 { INDEX_op_ld8u_i32, { "r", "r" } },
2468 { INDEX_op_ld8s_i32, { "r", "r" } },
2469 { INDEX_op_ld16u_i32, { "r", "r" } },
2470 { INDEX_op_ld16s_i32, { "r", "r" } },
2471 { INDEX_op_ld_i32, { "r", "r" } },
2473 { INDEX_op_st8_i32, { "r", "r" } },
2474 { INDEX_op_st16_i32, { "r", "r" } },
2475 { INDEX_op_st_i32, { "r", "r" } },
2477 { INDEX_op_add_i32, { "r", "r", "ri" } },
2478 { INDEX_op_mul_i32, { "r", "r", "rI" } },
2479 { INDEX_op_div_i32, { "r", "r", "r" } },
2480 { INDEX_op_divu_i32, { "r", "r", "r" } },
2481 { INDEX_op_sub_i32, { "r", "rI", "ri" } },
2482 { INDEX_op_and_i32, { "r", "r", "ri" } },
2483 { INDEX_op_or_i32, { "r", "r", "ri" } },
2484 { INDEX_op_xor_i32, { "r", "r", "ri" } },
2485 { INDEX_op_andc_i32, { "r", "r", "ri" } },
2486 { INDEX_op_orc_i32, { "r", "r", "ri" } },
2487 { INDEX_op_eqv_i32, { "r", "r", "ri" } },
2488 { INDEX_op_nand_i32, { "r", "r", "r" } },
2489 { INDEX_op_nor_i32, { "r", "r", "r" } },
2491 { INDEX_op_shl_i32, { "r", "r", "ri" } },
2492 { INDEX_op_shr_i32, { "r", "r", "ri" } },
2493 { INDEX_op_sar_i32, { "r", "r", "ri" } },
2494 { INDEX_op_rotl_i32, { "r", "r", "ri" } },
2495 { INDEX_op_rotr_i32, { "r", "r", "ri" } },
2497 { INDEX_op_neg_i32, { "r", "r" } },
2498 { INDEX_op_not_i32, { "r", "r" } },
2499 { INDEX_op_ext8s_i32, { "r", "r" } },
2500 { INDEX_op_ext16s_i32, { "r", "r" } },
2501 { INDEX_op_bswap16_i32, { "r", "r" } },
2502 { INDEX_op_bswap32_i32, { "r", "r" } },
2504 { INDEX_op_brcond_i32, { "r", "ri" } },
2505 { INDEX_op_setcond_i32, { "r", "r", "ri" } },
2506 { INDEX_op_movcond_i32, { "r", "r", "ri", "rZ", "rZ" } },
2508 { INDEX_op_deposit_i32, { "r", "0", "rZ" } },
2510 { INDEX_op_muluh_i32, { "r", "r", "r" } },
2511 { INDEX_op_mulsh_i32, { "r", "r", "r" } },
2513 #if TCG_TARGET_REG_BITS == 64
2514 { INDEX_op_ld8u_i64, { "r", "r" } },
2515 { INDEX_op_ld8s_i64, { "r", "r" } },
2516 { INDEX_op_ld16u_i64, { "r", "r" } },
2517 { INDEX_op_ld16s_i64, { "r", "r" } },
2518 { INDEX_op_ld32u_i64, { "r", "r" } },
2519 { INDEX_op_ld32s_i64, { "r", "r" } },
2520 { INDEX_op_ld_i64, { "r", "r" } },
2522 { INDEX_op_st8_i64, { "r", "r" } },
2523 { INDEX_op_st16_i64, { "r", "r" } },
2524 { INDEX_op_st32_i64, { "r", "r" } },
2525 { INDEX_op_st_i64, { "r", "r" } },
2527 { INDEX_op_add_i64, { "r", "r", "rT" } },
2528 { INDEX_op_sub_i64, { "r", "rI", "rT" } },
2529 { INDEX_op_and_i64, { "r", "r", "ri" } },
2530 { INDEX_op_or_i64, { "r", "r", "rU" } },
2531 { INDEX_op_xor_i64, { "r", "r", "rU" } },
2532 { INDEX_op_andc_i64, { "r", "r", "ri" } },
2533 { INDEX_op_orc_i64, { "r", "r", "r" } },
2534 { INDEX_op_eqv_i64, { "r", "r", "r" } },
2535 { INDEX_op_nand_i64, { "r", "r", "r" } },
2536 { INDEX_op_nor_i64, { "r", "r", "r" } },
2538 { INDEX_op_shl_i64, { "r", "r", "ri" } },
2539 { INDEX_op_shr_i64, { "r", "r", "ri" } },
2540 { INDEX_op_sar_i64, { "r", "r", "ri" } },
2541 { INDEX_op_rotl_i64, { "r", "r", "ri" } },
2542 { INDEX_op_rotr_i64, { "r", "r", "ri" } },
2544 { INDEX_op_mul_i64, { "r", "r", "rI" } },
2545 { INDEX_op_div_i64, { "r", "r", "r" } },
2546 { INDEX_op_divu_i64, { "r", "r", "r" } },
2548 { INDEX_op_neg_i64, { "r", "r" } },
2549 { INDEX_op_not_i64, { "r", "r" } },
2550 { INDEX_op_ext8s_i64, { "r", "r" } },
2551 { INDEX_op_ext16s_i64, { "r", "r" } },
2552 { INDEX_op_ext32s_i64, { "r", "r" } },
2553 { INDEX_op_ext_i32_i64, { "r", "r" } },
2554 { INDEX_op_extu_i32_i64, { "r", "r" } },
2555 { INDEX_op_bswap16_i64, { "r", "r" } },
2556 { INDEX_op_bswap32_i64, { "r", "r" } },
2557 { INDEX_op_bswap64_i64, { "r", "r" } },
2559 { INDEX_op_brcond_i64, { "r", "ri" } },
2560 { INDEX_op_setcond_i64, { "r", "r", "ri" } },
2561 { INDEX_op_movcond_i64, { "r", "r", "ri", "rZ", "rZ" } },
2563 { INDEX_op_deposit_i64, { "r", "0", "rZ" } },
2565 { INDEX_op_mulsh_i64, { "r", "r", "r" } },
2566 { INDEX_op_muluh_i64, { "r", "r", "r" } },
2567 #endif
2569 #if TCG_TARGET_REG_BITS == 32
2570 { INDEX_op_brcond2_i32, { "r", "r", "ri", "ri" } },
2571 { INDEX_op_setcond2_i32, { "r", "r", "r", "ri", "ri" } },
2572 #endif
2574 #if TCG_TARGET_REG_BITS == 64
2575 { INDEX_op_add2_i64, { "r", "r", "r", "r", "rI", "rZM" } },
2576 { INDEX_op_sub2_i64, { "r", "r", "rI", "rZM", "r", "r" } },
2577 #else
2578 { INDEX_op_add2_i32, { "r", "r", "r", "r", "rI", "rZM" } },
2579 { INDEX_op_sub2_i32, { "r", "r", "rI", "rZM", "r", "r" } },
2580 #endif
2582 #if TCG_TARGET_REG_BITS == 64
2583 { INDEX_op_qemu_ld_i32, { "r", "L" } },
2584 { INDEX_op_qemu_st_i32, { "S", "S" } },
2585 { INDEX_op_qemu_ld_i64, { "r", "L" } },
2586 { INDEX_op_qemu_st_i64, { "S", "S" } },
2587 #elif TARGET_LONG_BITS == 32
2588 { INDEX_op_qemu_ld_i32, { "r", "L" } },
2589 { INDEX_op_qemu_st_i32, { "S", "S" } },
2590 { INDEX_op_qemu_ld_i64, { "L", "L", "L" } },
2591 { INDEX_op_qemu_st_i64, { "S", "S", "S" } },
2592 #else
2593 { INDEX_op_qemu_ld_i32, { "r", "L", "L" } },
2594 { INDEX_op_qemu_st_i32, { "S", "S", "S" } },
2595 { INDEX_op_qemu_ld_i64, { "L", "L", "L", "L" } },
2596 { INDEX_op_qemu_st_i64, { "S", "S", "S", "S" } },
2597 #endif
2599 { -1 },
2602 static void tcg_target_init(TCGContext *s)
2604 unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2605 if (hwcap & PPC_FEATURE_ARCH_2_06) {
2606 have_isa_2_06 = true;
2609 tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I32], 0, 0xffffffff);
2610 tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I64], 0, 0xffffffff);
2611 tcg_regset_set32(tcg_target_call_clobber_regs, 0,
2612 (1 << TCG_REG_R0) |
2613 (1 << TCG_REG_R2) |
2614 (1 << TCG_REG_R3) |
2615 (1 << TCG_REG_R4) |
2616 (1 << TCG_REG_R5) |
2617 (1 << TCG_REG_R6) |
2618 (1 << TCG_REG_R7) |
2619 (1 << TCG_REG_R8) |
2620 (1 << TCG_REG_R9) |
2621 (1 << TCG_REG_R10) |
2622 (1 << TCG_REG_R11) |
2623 (1 << TCG_REG_R12));
2625 tcg_regset_clear(s->reserved_regs);
2626 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); /* tcg temp */
2627 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R1); /* stack pointer */
2628 #if defined(_CALL_SYSV)
2629 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R2); /* toc pointer */
2630 #endif
2631 #if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
2632 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R13); /* thread pointer */
2633 #endif
2634 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1); /* mem temp */
2635 if (USE_REG_RA) {
2636 tcg_regset_set_reg(s->reserved_regs, TCG_REG_RA); /* return addr */
2639 tcg_add_target_add_op_defs(ppc_op_defs);
2642 #ifdef __ELF__
2643 typedef struct {
2644 DebugFrameCIE cie;
2645 DebugFrameFDEHeader fde;
2646 uint8_t fde_def_cfa[4];
2647 uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2 + 3];
2648 } DebugFrame;
2650 /* We're expecting a 2 byte uleb128 encoded value. */
2651 QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
2653 #if TCG_TARGET_REG_BITS == 64
2654 # define ELF_HOST_MACHINE EM_PPC64
2655 #else
2656 # define ELF_HOST_MACHINE EM_PPC
2657 #endif
2659 static DebugFrame debug_frame = {
2660 .cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
2661 .cie.id = -1,
2662 .cie.version = 1,
2663 .cie.code_align = 1,
2664 .cie.data_align = (-SZR & 0x7f), /* sleb128 -SZR */
2665 .cie.return_column = 65,
2667 /* Total FDE size does not include the "len" member. */
2668 .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset),
2670 .fde_def_cfa = {
2671 12, TCG_REG_R1, /* DW_CFA_def_cfa r1, ... */
2672 (FRAME_SIZE & 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */
2673 (FRAME_SIZE >> 7)
2675 .fde_reg_ofs = {
2676 /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
2677 0x11, 65, (LR_OFFSET / -SZR) & 0x7f,
2681 void tcg_register_jit(void *buf, size_t buf_size)
2683 uint8_t *p = &debug_frame.fde_reg_ofs[3];
2684 int i;
2686 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i, p += 2) {
2687 p[0] = 0x80 + tcg_target_callee_save_regs[i];
2688 p[1] = (FRAME_SIZE - (REG_SAVE_BOT + i * SZR)) / SZR;
2691 debug_frame.fde.func_start = (uintptr_t)buf;
2692 debug_frame.fde.func_len = buf_size;
2694 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
2696 #endif /* __ELF__ */
2698 static size_t dcache_bsize = 16;
2699 static size_t icache_bsize = 16;
2701 void flush_icache_range(uintptr_t start, uintptr_t stop)
2703 uintptr_t p, start1, stop1;
2704 size_t dsize = dcache_bsize;
2705 size_t isize = icache_bsize;
2707 start1 = start & ~(dsize - 1);
2708 stop1 = (stop + dsize - 1) & ~(dsize - 1);
2709 for (p = start1; p < stop1; p += dsize) {
2710 asm volatile ("dcbst 0,%0" : : "r"(p) : "memory");
2712 asm volatile ("sync" : : : "memory");
2714 start &= start & ~(isize - 1);
2715 stop1 = (stop + isize - 1) & ~(isize - 1);
2716 for (p = start1; p < stop1; p += isize) {
2717 asm volatile ("icbi 0,%0" : : "r"(p) : "memory");
2719 asm volatile ("sync" : : : "memory");
2720 asm volatile ("isync" : : : "memory");
2723 #if defined _AIX
2724 #include <sys/systemcfg.h>
2726 static void __attribute__((constructor)) tcg_cache_init(void)
2728 icache_bsize = _system_configuration.icache_line;
2729 dcache_bsize = _system_configuration.dcache_line;
2732 #elif defined __linux__
2733 static void __attribute__((constructor)) tcg_cache_init(void)
2735 unsigned long dsize = qemu_getauxval(AT_DCACHEBSIZE);
2736 unsigned long isize = qemu_getauxval(AT_ICACHEBSIZE);
2738 if (dsize == 0 || isize == 0) {
2739 if (dsize == 0) {
2740 fprintf(stderr, "getauxval AT_DCACHEBSIZE failed\n");
2742 if (isize == 0) {
2743 fprintf(stderr, "getauxval AT_ICACHEBSIZE failed\n");
2745 exit(1);
2747 dcache_bsize = dsize;
2748 icache_bsize = isize;
2751 #elif defined __APPLE__
2752 #include <sys/sysctl.h>
2754 static void __attribute__((constructor)) tcg_cache_init(void)
2756 size_t len;
2757 unsigned cacheline;
2758 int name[2] = { CTL_HW, HW_CACHELINE };
2760 len = sizeof(cacheline);
2761 if (sysctl(name, 2, &cacheline, &len, NULL, 0)) {
2762 perror("sysctl CTL_HW HW_CACHELINE failed");
2763 exit(1);
2765 dcache_bsize = cacheline;
2766 icache_bsize = cacheline;
2769 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2770 #include <sys/sysctl.h>
2772 static void __attribute__((constructor)) tcg_cache_init(void)
2774 size_t len = 4;
2775 unsigned cacheline;
2777 if (sysctlbyname ("machdep.cacheline_size", &cacheline, &len, NULL, 0)) {
2778 fprintf(stderr, "sysctlbyname machdep.cacheline_size failed: %s\n",
2779 strerror(errno));
2780 exit(1);
2782 dcache_bsize = cacheline;
2783 icache_bsize = cacheline;
2785 #endif