xref: /openbmc/qemu/tcg/ppc/tcg-target.c.inc (revision ef26632e)
1/*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2008 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25#include "elf.h"
26#include "../tcg-pool.c.inc"
27
28#if defined _CALL_DARWIN || defined __APPLE__
29#define TCG_TARGET_CALL_DARWIN
30#endif
31#ifdef _CALL_SYSV
32# define TCG_TARGET_CALL_ALIGN_ARGS   1
33#endif
34
35/* For some memory operations, we need a scratch that isn't R0.  For the AIX
36   calling convention, we can re-use the TOC register since we'll be reloading
37   it at every call.  Otherwise R12 will do nicely as neither a call-saved
38   register nor a parameter register.  */
39#ifdef _CALL_AIX
40# define TCG_REG_TMP1   TCG_REG_R2
41#else
42# define TCG_REG_TMP1   TCG_REG_R12
43#endif
44
45#define TCG_VEC_TMP1    TCG_REG_V0
46#define TCG_VEC_TMP2    TCG_REG_V1
47
48#define TCG_REG_TB     TCG_REG_R31
49#define USE_REG_TB     (TCG_TARGET_REG_BITS == 64)
50
51/* Shorthand for size of a pointer.  Avoid promotion to unsigned.  */
52#define SZP  ((int)sizeof(void *))
53
54/* Shorthand for size of a register.  */
55#define SZR  (TCG_TARGET_REG_BITS / 8)
56
57#define TCG_CT_CONST_S16  0x100
58#define TCG_CT_CONST_U16  0x200
59#define TCG_CT_CONST_S32  0x400
60#define TCG_CT_CONST_U32  0x800
61#define TCG_CT_CONST_ZERO 0x1000
62#define TCG_CT_CONST_MONE 0x2000
63#define TCG_CT_CONST_WSZ  0x4000
64
65TCGPowerISA have_isa;
66static bool have_isel;
67bool have_altivec;
68bool have_vsx;
69
70#ifndef CONFIG_SOFTMMU
71#define TCG_GUEST_BASE_REG 30
72#endif
73
74#ifdef CONFIG_DEBUG_TCG
75static const char tcg_target_reg_names[TCG_TARGET_NB_REGS][4] = {
76    "r0",  "r1",  "r2",  "r3",  "r4",  "r5",  "r6",  "r7",
77    "r8",  "r9",  "r10", "r11", "r12", "r13", "r14", "r15",
78    "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
79    "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31",
80    "v0",  "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7",
81    "v8",  "v9",  "v10", "v11", "v12", "v13", "v14", "v15",
82    "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
83    "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",
84};
85#endif
86
87static const int tcg_target_reg_alloc_order[] = {
88    TCG_REG_R14,  /* call saved registers */
89    TCG_REG_R15,
90    TCG_REG_R16,
91    TCG_REG_R17,
92    TCG_REG_R18,
93    TCG_REG_R19,
94    TCG_REG_R20,
95    TCG_REG_R21,
96    TCG_REG_R22,
97    TCG_REG_R23,
98    TCG_REG_R24,
99    TCG_REG_R25,
100    TCG_REG_R26,
101    TCG_REG_R27,
102    TCG_REG_R28,
103    TCG_REG_R29,
104    TCG_REG_R30,
105    TCG_REG_R31,
106    TCG_REG_R12,  /* call clobbered, non-arguments */
107    TCG_REG_R11,
108    TCG_REG_R2,
109    TCG_REG_R13,
110    TCG_REG_R10,  /* call clobbered, arguments */
111    TCG_REG_R9,
112    TCG_REG_R8,
113    TCG_REG_R7,
114    TCG_REG_R6,
115    TCG_REG_R5,
116    TCG_REG_R4,
117    TCG_REG_R3,
118
119    /* V0 and V1 reserved as temporaries; V20 - V31 are call-saved */
120    TCG_REG_V2,   /* call clobbered, vectors */
121    TCG_REG_V3,
122    TCG_REG_V4,
123    TCG_REG_V5,
124    TCG_REG_V6,
125    TCG_REG_V7,
126    TCG_REG_V8,
127    TCG_REG_V9,
128    TCG_REG_V10,
129    TCG_REG_V11,
130    TCG_REG_V12,
131    TCG_REG_V13,
132    TCG_REG_V14,
133    TCG_REG_V15,
134    TCG_REG_V16,
135    TCG_REG_V17,
136    TCG_REG_V18,
137    TCG_REG_V19,
138};
139
140static const int tcg_target_call_iarg_regs[] = {
141    TCG_REG_R3,
142    TCG_REG_R4,
143    TCG_REG_R5,
144    TCG_REG_R6,
145    TCG_REG_R7,
146    TCG_REG_R8,
147    TCG_REG_R9,
148    TCG_REG_R10
149};
150
151static const int tcg_target_call_oarg_regs[] = {
152    TCG_REG_R3,
153    TCG_REG_R4
154};
155
156static const int tcg_target_callee_save_regs[] = {
157#ifdef TCG_TARGET_CALL_DARWIN
158    TCG_REG_R11,
159#endif
160    TCG_REG_R14,
161    TCG_REG_R15,
162    TCG_REG_R16,
163    TCG_REG_R17,
164    TCG_REG_R18,
165    TCG_REG_R19,
166    TCG_REG_R20,
167    TCG_REG_R21,
168    TCG_REG_R22,
169    TCG_REG_R23,
170    TCG_REG_R24,
171    TCG_REG_R25,
172    TCG_REG_R26,
173    TCG_REG_R27, /* currently used for the global env */
174    TCG_REG_R28,
175    TCG_REG_R29,
176    TCG_REG_R30,
177    TCG_REG_R31
178};
179
180static inline bool in_range_b(tcg_target_long target)
181{
182    return target == sextract64(target, 0, 26);
183}
184
185static uint32_t reloc_pc24_val(const tcg_insn_unit *pc,
186			       const tcg_insn_unit *target)
187{
188    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
189    tcg_debug_assert(in_range_b(disp));
190    return disp & 0x3fffffc;
191}
192
193static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
194{
195    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
196    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
197
198    if (in_range_b(disp)) {
199        *src_rw = (*src_rw & ~0x3fffffc) | (disp & 0x3fffffc);
200        return true;
201    }
202    return false;
203}
204
205static uint16_t reloc_pc14_val(const tcg_insn_unit *pc,
206			       const tcg_insn_unit *target)
207{
208    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
209    tcg_debug_assert(disp == (int16_t) disp);
210    return disp & 0xfffc;
211}
212
213static bool reloc_pc14(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
214{
215    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
216    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
217
218    if (disp == (int16_t) disp) {
219        *src_rw = (*src_rw & ~0xfffc) | (disp & 0xfffc);
220        return true;
221    }
222    return false;
223}
224
225/* parse target specific constraints */
226static const char *target_parse_constraint(TCGArgConstraint *ct,
227                                           const char *ct_str, TCGType type)
228{
229    switch (*ct_str++) {
230    case 'A': case 'B': case 'C': case 'D':
231        tcg_regset_set_reg(ct->regs, 3 + ct_str[0] - 'A');
232        break;
233    case 'r':
234        ct->regs = 0xffffffff;
235        break;
236    case 'v':
237        ct->regs = 0xffffffff00000000ull;
238        break;
239    case 'L':                   /* qemu_ld constraint */
240        ct->regs = 0xffffffff;
241        tcg_regset_reset_reg(ct->regs, TCG_REG_R3);
242#ifdef CONFIG_SOFTMMU
243        tcg_regset_reset_reg(ct->regs, TCG_REG_R4);
244        tcg_regset_reset_reg(ct->regs, TCG_REG_R5);
245#endif
246        break;
247    case 'S':                   /* qemu_st constraint */
248        ct->regs = 0xffffffff;
249        tcg_regset_reset_reg(ct->regs, TCG_REG_R3);
250#ifdef CONFIG_SOFTMMU
251        tcg_regset_reset_reg(ct->regs, TCG_REG_R4);
252        tcg_regset_reset_reg(ct->regs, TCG_REG_R5);
253        tcg_regset_reset_reg(ct->regs, TCG_REG_R6);
254#endif
255        break;
256    case 'I':
257        ct->ct |= TCG_CT_CONST_S16;
258        break;
259    case 'J':
260        ct->ct |= TCG_CT_CONST_U16;
261        break;
262    case 'M':
263        ct->ct |= TCG_CT_CONST_MONE;
264        break;
265    case 'T':
266        ct->ct |= TCG_CT_CONST_S32;
267        break;
268    case 'U':
269        ct->ct |= TCG_CT_CONST_U32;
270        break;
271    case 'W':
272        ct->ct |= TCG_CT_CONST_WSZ;
273        break;
274    case 'Z':
275        ct->ct |= TCG_CT_CONST_ZERO;
276        break;
277    default:
278        return NULL;
279    }
280    return ct_str;
281}
282
283/* test if a constant matches the constraint */
284static int tcg_target_const_match(tcg_target_long val, TCGType type,
285                                  const TCGArgConstraint *arg_ct)
286{
287    int ct = arg_ct->ct;
288    if (ct & TCG_CT_CONST) {
289        return 1;
290    }
291
292    /* The only 32-bit constraint we use aside from
293       TCG_CT_CONST is TCG_CT_CONST_S16.  */
294    if (type == TCG_TYPE_I32) {
295        val = (int32_t)val;
296    }
297
298    if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
299        return 1;
300    } else if ((ct & TCG_CT_CONST_U16) && val == (uint16_t)val) {
301        return 1;
302    } else if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
303        return 1;
304    } else if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
305        return 1;
306    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
307        return 1;
308    } else if ((ct & TCG_CT_CONST_MONE) && val == -1) {
309        return 1;
310    } else if ((ct & TCG_CT_CONST_WSZ)
311               && val == (type == TCG_TYPE_I32 ? 32 : 64)) {
312        return 1;
313    }
314    return 0;
315}
316
317#define OPCD(opc) ((opc)<<26)
318#define XO19(opc) (OPCD(19)|((opc)<<1))
319#define MD30(opc) (OPCD(30)|((opc)<<2))
320#define MDS30(opc) (OPCD(30)|((opc)<<1))
321#define XO31(opc) (OPCD(31)|((opc)<<1))
322#define XO58(opc) (OPCD(58)|(opc))
323#define XO62(opc) (OPCD(62)|(opc))
324#define VX4(opc)  (OPCD(4)|(opc))
325
326#define B      OPCD( 18)
327#define BC     OPCD( 16)
328#define LBZ    OPCD( 34)
329#define LHZ    OPCD( 40)
330#define LHA    OPCD( 42)
331#define LWZ    OPCD( 32)
332#define LWZUX  XO31( 55)
333#define STB    OPCD( 38)
334#define STH    OPCD( 44)
335#define STW    OPCD( 36)
336
337#define STD    XO62(  0)
338#define STDU   XO62(  1)
339#define STDX   XO31(149)
340
341#define LD     XO58(  0)
342#define LDX    XO31( 21)
343#define LDU    XO58(  1)
344#define LDUX   XO31( 53)
345#define LWA    XO58(  2)
346#define LWAX   XO31(341)
347
348#define ADDIC  OPCD( 12)
349#define ADDI   OPCD( 14)
350#define ADDIS  OPCD( 15)
351#define ORI    OPCD( 24)
352#define ORIS   OPCD( 25)
353#define XORI   OPCD( 26)
354#define XORIS  OPCD( 27)
355#define ANDI   OPCD( 28)
356#define ANDIS  OPCD( 29)
357#define MULLI  OPCD(  7)
358#define CMPLI  OPCD( 10)
359#define CMPI   OPCD( 11)
360#define SUBFIC OPCD( 8)
361
362#define LWZU   OPCD( 33)
363#define STWU   OPCD( 37)
364
365#define RLWIMI OPCD( 20)
366#define RLWINM OPCD( 21)
367#define RLWNM  OPCD( 23)
368
369#define RLDICL MD30(  0)
370#define RLDICR MD30(  1)
371#define RLDIMI MD30(  3)
372#define RLDCL  MDS30( 8)
373
374#define BCLR   XO19( 16)
375#define BCCTR  XO19(528)
376#define CRAND  XO19(257)
377#define CRANDC XO19(129)
378#define CRNAND XO19(225)
379#define CROR   XO19(449)
380#define CRNOR  XO19( 33)
381
382#define EXTSB  XO31(954)
383#define EXTSH  XO31(922)
384#define EXTSW  XO31(986)
385#define ADD    XO31(266)
386#define ADDE   XO31(138)
387#define ADDME  XO31(234)
388#define ADDZE  XO31(202)
389#define ADDC   XO31( 10)
390#define AND    XO31( 28)
391#define SUBF   XO31( 40)
392#define SUBFC  XO31(  8)
393#define SUBFE  XO31(136)
394#define SUBFME XO31(232)
395#define SUBFZE XO31(200)
396#define OR     XO31(444)
397#define XOR    XO31(316)
398#define MULLW  XO31(235)
399#define MULHW  XO31( 75)
400#define MULHWU XO31( 11)
401#define DIVW   XO31(491)
402#define DIVWU  XO31(459)
403#define CMP    XO31(  0)
404#define CMPL   XO31( 32)
405#define LHBRX  XO31(790)
406#define LWBRX  XO31(534)
407#define LDBRX  XO31(532)
408#define STHBRX XO31(918)
409#define STWBRX XO31(662)
410#define STDBRX XO31(660)
411#define MFSPR  XO31(339)
412#define MTSPR  XO31(467)
413#define SRAWI  XO31(824)
414#define NEG    XO31(104)
415#define MFCR   XO31( 19)
416#define MFOCRF (MFCR | (1u << 20))
417#define NOR    XO31(124)
418#define CNTLZW XO31( 26)
419#define CNTLZD XO31( 58)
420#define CNTTZW XO31(538)
421#define CNTTZD XO31(570)
422#define CNTPOPW XO31(378)
423#define CNTPOPD XO31(506)
424#define ANDC   XO31( 60)
425#define ORC    XO31(412)
426#define EQV    XO31(284)
427#define NAND   XO31(476)
428#define ISEL   XO31( 15)
429
430#define MULLD  XO31(233)
431#define MULHD  XO31( 73)
432#define MULHDU XO31(  9)
433#define DIVD   XO31(489)
434#define DIVDU  XO31(457)
435
436#define LBZX   XO31( 87)
437#define LHZX   XO31(279)
438#define LHAX   XO31(343)
439#define LWZX   XO31( 23)
440#define STBX   XO31(215)
441#define STHX   XO31(407)
442#define STWX   XO31(151)
443
444#define EIEIO  XO31(854)
445#define HWSYNC XO31(598)
446#define LWSYNC (HWSYNC | (1u << 21))
447
448#define SPR(a, b) ((((a)<<5)|(b))<<11)
449#define LR     SPR(8, 0)
450#define CTR    SPR(9, 0)
451
452#define SLW    XO31( 24)
453#define SRW    XO31(536)
454#define SRAW   XO31(792)
455
456#define SLD    XO31( 27)
457#define SRD    XO31(539)
458#define SRAD   XO31(794)
459#define SRADI  XO31(413<<1)
460
461#define TW     XO31( 4)
462#define TRAP   (TW | TO(31))
463
464#define NOP    ORI  /* ori 0,0,0 */
465
466#define LVX        XO31(103)
467#define LVEBX      XO31(7)
468#define LVEHX      XO31(39)
469#define LVEWX      XO31(71)
470#define LXSDX      (XO31(588) | 1)  /* v2.06, force tx=1 */
471#define LXVDSX     (XO31(332) | 1)  /* v2.06, force tx=1 */
472#define LXSIWZX    (XO31(12) | 1)   /* v2.07, force tx=1 */
473#define LXV        (OPCD(61) | 8 | 1)  /* v3.00, force tx=1 */
474#define LXSD       (OPCD(57) | 2)   /* v3.00 */
475#define LXVWSX     (XO31(364) | 1)  /* v3.00, force tx=1 */
476
477#define STVX       XO31(231)
478#define STVEWX     XO31(199)
479#define STXSDX     (XO31(716) | 1)  /* v2.06, force sx=1 */
480#define STXSIWX    (XO31(140) | 1)  /* v2.07, force sx=1 */
481#define STXV       (OPCD(61) | 8 | 5) /* v3.00, force sx=1 */
482#define STXSD      (OPCD(61) | 2)   /* v3.00 */
483
484#define VADDSBS    VX4(768)
485#define VADDUBS    VX4(512)
486#define VADDUBM    VX4(0)
487#define VADDSHS    VX4(832)
488#define VADDUHS    VX4(576)
489#define VADDUHM    VX4(64)
490#define VADDSWS    VX4(896)
491#define VADDUWS    VX4(640)
492#define VADDUWM    VX4(128)
493#define VADDUDM    VX4(192)       /* v2.07 */
494
495#define VSUBSBS    VX4(1792)
496#define VSUBUBS    VX4(1536)
497#define VSUBUBM    VX4(1024)
498#define VSUBSHS    VX4(1856)
499#define VSUBUHS    VX4(1600)
500#define VSUBUHM    VX4(1088)
501#define VSUBSWS    VX4(1920)
502#define VSUBUWS    VX4(1664)
503#define VSUBUWM    VX4(1152)
504#define VSUBUDM    VX4(1216)      /* v2.07 */
505
506#define VNEGW      (VX4(1538) | (6 << 16))  /* v3.00 */
507#define VNEGD      (VX4(1538) | (7 << 16))  /* v3.00 */
508
509#define VMAXSB     VX4(258)
510#define VMAXSH     VX4(322)
511#define VMAXSW     VX4(386)
512#define VMAXSD     VX4(450)       /* v2.07 */
513#define VMAXUB     VX4(2)
514#define VMAXUH     VX4(66)
515#define VMAXUW     VX4(130)
516#define VMAXUD     VX4(194)       /* v2.07 */
517#define VMINSB     VX4(770)
518#define VMINSH     VX4(834)
519#define VMINSW     VX4(898)
520#define VMINSD     VX4(962)       /* v2.07 */
521#define VMINUB     VX4(514)
522#define VMINUH     VX4(578)
523#define VMINUW     VX4(642)
524#define VMINUD     VX4(706)       /* v2.07 */
525
526#define VCMPEQUB   VX4(6)
527#define VCMPEQUH   VX4(70)
528#define VCMPEQUW   VX4(134)
529#define VCMPEQUD   VX4(199)       /* v2.07 */
530#define VCMPGTSB   VX4(774)
531#define VCMPGTSH   VX4(838)
532#define VCMPGTSW   VX4(902)
533#define VCMPGTSD   VX4(967)       /* v2.07 */
534#define VCMPGTUB   VX4(518)
535#define VCMPGTUH   VX4(582)
536#define VCMPGTUW   VX4(646)
537#define VCMPGTUD   VX4(711)       /* v2.07 */
538#define VCMPNEB    VX4(7)         /* v3.00 */
539#define VCMPNEH    VX4(71)        /* v3.00 */
540#define VCMPNEW    VX4(135)       /* v3.00 */
541
542#define VSLB       VX4(260)
543#define VSLH       VX4(324)
544#define VSLW       VX4(388)
545#define VSLD       VX4(1476)      /* v2.07 */
546#define VSRB       VX4(516)
547#define VSRH       VX4(580)
548#define VSRW       VX4(644)
549#define VSRD       VX4(1732)      /* v2.07 */
550#define VSRAB      VX4(772)
551#define VSRAH      VX4(836)
552#define VSRAW      VX4(900)
553#define VSRAD      VX4(964)       /* v2.07 */
554#define VRLB       VX4(4)
555#define VRLH       VX4(68)
556#define VRLW       VX4(132)
557#define VRLD       VX4(196)       /* v2.07 */
558
559#define VMULEUB    VX4(520)
560#define VMULEUH    VX4(584)
561#define VMULEUW    VX4(648)       /* v2.07 */
562#define VMULOUB    VX4(8)
563#define VMULOUH    VX4(72)
564#define VMULOUW    VX4(136)       /* v2.07 */
565#define VMULUWM    VX4(137)       /* v2.07 */
566#define VMULLD     VX4(457)       /* v3.10 */
567#define VMSUMUHM   VX4(38)
568
569#define VMRGHB     VX4(12)
570#define VMRGHH     VX4(76)
571#define VMRGHW     VX4(140)
572#define VMRGLB     VX4(268)
573#define VMRGLH     VX4(332)
574#define VMRGLW     VX4(396)
575
576#define VPKUHUM    VX4(14)
577#define VPKUWUM    VX4(78)
578
579#define VAND       VX4(1028)
580#define VANDC      VX4(1092)
581#define VNOR       VX4(1284)
582#define VOR        VX4(1156)
583#define VXOR       VX4(1220)
584#define VEQV       VX4(1668)      /* v2.07 */
585#define VNAND      VX4(1412)      /* v2.07 */
586#define VORC       VX4(1348)      /* v2.07 */
587
588#define VSPLTB     VX4(524)
589#define VSPLTH     VX4(588)
590#define VSPLTW     VX4(652)
591#define VSPLTISB   VX4(780)
592#define VSPLTISH   VX4(844)
593#define VSPLTISW   VX4(908)
594
595#define VSLDOI     VX4(44)
596
597#define XXPERMDI   (OPCD(60) | (10 << 3) | 7)  /* v2.06, force ax=bx=tx=1 */
598#define XXSEL      (OPCD(60) | (3 << 4) | 0xf) /* v2.06, force ax=bx=cx=tx=1 */
599#define XXSPLTIB   (OPCD(60) | (360 << 1) | 1) /* v3.00, force tx=1 */
600
601#define MFVSRD     (XO31(51) | 1)   /* v2.07, force sx=1 */
602#define MFVSRWZ    (XO31(115) | 1)  /* v2.07, force sx=1 */
603#define MTVSRD     (XO31(179) | 1)  /* v2.07, force tx=1 */
604#define MTVSRWZ    (XO31(243) | 1)  /* v2.07, force tx=1 */
605#define MTVSRDD    (XO31(435) | 1)  /* v3.00, force tx=1 */
606#define MTVSRWS    (XO31(403) | 1)  /* v3.00, force tx=1 */
607
608#define RT(r) ((r)<<21)
609#define RS(r) ((r)<<21)
610#define RA(r) ((r)<<16)
611#define RB(r) ((r)<<11)
612#define TO(t) ((t)<<21)
613#define SH(s) ((s)<<11)
614#define MB(b) ((b)<<6)
615#define ME(e) ((e)<<1)
616#define BO(o) ((o)<<21)
617#define MB64(b) ((b)<<5)
618#define FXM(b) (1 << (19 - (b)))
619
620#define VRT(r)  (((r) & 31) << 21)
621#define VRA(r)  (((r) & 31) << 16)
622#define VRB(r)  (((r) & 31) << 11)
623#define VRC(r)  (((r) & 31) <<  6)
624
625#define LK    1
626
627#define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
628#define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
629#define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
630#define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
631
632#define BF(n)    ((n)<<23)
633#define BI(n, c) (((c)+((n)*4))<<16)
634#define BT(n, c) (((c)+((n)*4))<<21)
635#define BA(n, c) (((c)+((n)*4))<<16)
636#define BB(n, c) (((c)+((n)*4))<<11)
637#define BC_(n, c) (((c)+((n)*4))<<6)
638
639#define BO_COND_TRUE  BO(12)
640#define BO_COND_FALSE BO( 4)
641#define BO_ALWAYS     BO(20)
642
643enum {
644    CR_LT,
645    CR_GT,
646    CR_EQ,
647    CR_SO
648};
649
650static const uint32_t tcg_to_bc[] = {
651    [TCG_COND_EQ]  = BC | BI(7, CR_EQ) | BO_COND_TRUE,
652    [TCG_COND_NE]  = BC | BI(7, CR_EQ) | BO_COND_FALSE,
653    [TCG_COND_LT]  = BC | BI(7, CR_LT) | BO_COND_TRUE,
654    [TCG_COND_GE]  = BC | BI(7, CR_LT) | BO_COND_FALSE,
655    [TCG_COND_LE]  = BC | BI(7, CR_GT) | BO_COND_FALSE,
656    [TCG_COND_GT]  = BC | BI(7, CR_GT) | BO_COND_TRUE,
657    [TCG_COND_LTU] = BC | BI(7, CR_LT) | BO_COND_TRUE,
658    [TCG_COND_GEU] = BC | BI(7, CR_LT) | BO_COND_FALSE,
659    [TCG_COND_LEU] = BC | BI(7, CR_GT) | BO_COND_FALSE,
660    [TCG_COND_GTU] = BC | BI(7, CR_GT) | BO_COND_TRUE,
661};
662
663/* The low bit here is set if the RA and RB fields must be inverted.  */
664static const uint32_t tcg_to_isel[] = {
665    [TCG_COND_EQ]  = ISEL | BC_(7, CR_EQ),
666    [TCG_COND_NE]  = ISEL | BC_(7, CR_EQ) | 1,
667    [TCG_COND_LT]  = ISEL | BC_(7, CR_LT),
668    [TCG_COND_GE]  = ISEL | BC_(7, CR_LT) | 1,
669    [TCG_COND_LE]  = ISEL | BC_(7, CR_GT) | 1,
670    [TCG_COND_GT]  = ISEL | BC_(7, CR_GT),
671    [TCG_COND_LTU] = ISEL | BC_(7, CR_LT),
672    [TCG_COND_GEU] = ISEL | BC_(7, CR_LT) | 1,
673    [TCG_COND_LEU] = ISEL | BC_(7, CR_GT) | 1,
674    [TCG_COND_GTU] = ISEL | BC_(7, CR_GT),
675};
676
677static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
678                        intptr_t value, intptr_t addend)
679{
680    const tcg_insn_unit *target;
681    int16_t lo;
682    int32_t hi;
683
684    value += addend;
685    target = (const tcg_insn_unit *)value;
686
687    switch (type) {
688    case R_PPC_REL14:
689        return reloc_pc14(code_ptr, target);
690    case R_PPC_REL24:
691        return reloc_pc24(code_ptr, target);
692    case R_PPC_ADDR16:
693        /*
694         * We are (slightly) abusing this relocation type.  In particular,
695         * assert that the low 2 bits are zero, and do not modify them.
696         * That way we can use this with LD et al that have opcode bits
697         * in the low 2 bits of the insn.
698         */
699        if ((value & 3) || value != (int16_t)value) {
700            return false;
701        }
702        *code_ptr = (*code_ptr & ~0xfffc) | (value & 0xfffc);
703        break;
704    case R_PPC_ADDR32:
705        /*
706         * We are abusing this relocation type.  Again, this points to
707         * a pair of insns, lis + load.  This is an absolute address
708         * relocation for PPC32 so the lis cannot be removed.
709         */
710        lo = value;
711        hi = value - lo;
712        if (hi + lo != value) {
713            return false;
714        }
715        code_ptr[0] = deposit32(code_ptr[0], 0, 16, hi >> 16);
716        code_ptr[1] = deposit32(code_ptr[1], 0, 16, lo);
717        break;
718    default:
719        g_assert_not_reached();
720    }
721    return true;
722}
723
724static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
725                             TCGReg base, tcg_target_long offset);
726
727static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
728{
729    if (ret == arg) {
730        return true;
731    }
732    switch (type) {
733    case TCG_TYPE_I64:
734        tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
735        /* fallthru */
736    case TCG_TYPE_I32:
737        if (ret < TCG_REG_V0) {
738            if (arg < TCG_REG_V0) {
739                tcg_out32(s, OR | SAB(arg, ret, arg));
740                break;
741            } else if (have_isa_2_07) {
742                tcg_out32(s, (type == TCG_TYPE_I32 ? MFVSRWZ : MFVSRD)
743                          | VRT(arg) | RA(ret));
744                break;
745            } else {
746                /* Altivec does not support vector->integer moves.  */
747                return false;
748            }
749        } else if (arg < TCG_REG_V0) {
750            if (have_isa_2_07) {
751                tcg_out32(s, (type == TCG_TYPE_I32 ? MTVSRWZ : MTVSRD)
752                          | VRT(ret) | RA(arg));
753                break;
754            } else {
755                /* Altivec does not support integer->vector moves.  */
756                return false;
757            }
758        }
759        /* fallthru */
760    case TCG_TYPE_V64:
761    case TCG_TYPE_V128:
762        tcg_debug_assert(ret >= TCG_REG_V0 && arg >= TCG_REG_V0);
763        tcg_out32(s, VOR | VRT(ret) | VRA(arg) | VRB(arg));
764        break;
765    default:
766        g_assert_not_reached();
767    }
768    return true;
769}
770
771static inline void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs,
772                               int sh, int mb)
773{
774    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
775    sh = SH(sh & 0x1f) | (((sh >> 5) & 1) << 1);
776    mb = MB64((mb >> 5) | ((mb << 1) & 0x3f));
777    tcg_out32(s, op | RA(ra) | RS(rs) | sh | mb);
778}
779
780static inline void tcg_out_rlw(TCGContext *s, int op, TCGReg ra, TCGReg rs,
781                               int sh, int mb, int me)
782{
783    tcg_out32(s, op | RA(ra) | RS(rs) | SH(sh) | MB(mb) | ME(me));
784}
785
786static inline void tcg_out_ext32u(TCGContext *s, TCGReg dst, TCGReg src)
787{
788    tcg_out_rld(s, RLDICL, dst, src, 0, 32);
789}
790
791static inline void tcg_out_shli32(TCGContext *s, TCGReg dst, TCGReg src, int c)
792{
793    tcg_out_rlw(s, RLWINM, dst, src, c, 0, 31 - c);
794}
795
796static inline void tcg_out_shli64(TCGContext *s, TCGReg dst, TCGReg src, int c)
797{
798    tcg_out_rld(s, RLDICR, dst, src, c, 63 - c);
799}
800
801static inline void tcg_out_shri32(TCGContext *s, TCGReg dst, TCGReg src, int c)
802{
803    tcg_out_rlw(s, RLWINM, dst, src, 32 - c, c, 31);
804}
805
806static inline void tcg_out_shri64(TCGContext *s, TCGReg dst, TCGReg src, int c)
807{
808    tcg_out_rld(s, RLDICL, dst, src, 64 - c, c);
809}
810
811/* Emit a move into ret of arg, if it can be done in one insn.  */
812static bool tcg_out_movi_one(TCGContext *s, TCGReg ret, tcg_target_long arg)
813{
814    if (arg == (int16_t)arg) {
815        tcg_out32(s, ADDI | TAI(ret, 0, arg));
816        return true;
817    }
818    if (arg == (int32_t)arg && (arg & 0xffff) == 0) {
819        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
820        return true;
821    }
822    return false;
823}
824
825static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
826                             tcg_target_long arg, bool in_prologue)
827{
828    intptr_t tb_diff;
829    tcg_target_long tmp;
830    int shift;
831
832    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
833
834    if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
835        arg = (int32_t)arg;
836    }
837
838    /* Load 16-bit immediates with one insn.  */
839    if (tcg_out_movi_one(s, ret, arg)) {
840        return;
841    }
842
843    /* Load addresses within the TB with one insn.  */
844    tb_diff = tcg_tbrel_diff(s, (void *)arg);
845    if (!in_prologue && USE_REG_TB && tb_diff == (int16_t)tb_diff) {
846        tcg_out32(s, ADDI | TAI(ret, TCG_REG_TB, tb_diff));
847        return;
848    }
849
850    /* Load 32-bit immediates with two insns.  Note that we've already
851       eliminated bare ADDIS, so we know both insns are required.  */
852    if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) {
853        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
854        tcg_out32(s, ORI | SAI(ret, ret, arg));
855        return;
856    }
857    if (arg == (uint32_t)arg && !(arg & 0x8000)) {
858        tcg_out32(s, ADDI | TAI(ret, 0, arg));
859        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
860        return;
861    }
862
863    /* Load masked 16-bit value.  */
864    if (arg > 0 && (arg & 0x8000)) {
865        tmp = arg | 0x7fff;
866        if ((tmp & (tmp + 1)) == 0) {
867            int mb = clz64(tmp + 1) + 1;
868            tcg_out32(s, ADDI | TAI(ret, 0, arg));
869            tcg_out_rld(s, RLDICL, ret, ret, 0, mb);
870            return;
871        }
872    }
873
874    /* Load common masks with 2 insns.  */
875    shift = ctz64(arg);
876    tmp = arg >> shift;
877    if (tmp == (int16_t)tmp) {
878        tcg_out32(s, ADDI | TAI(ret, 0, tmp));
879        tcg_out_shli64(s, ret, ret, shift);
880        return;
881    }
882    shift = clz64(arg);
883    if (tcg_out_movi_one(s, ret, arg << shift)) {
884        tcg_out_shri64(s, ret, ret, shift);
885        return;
886    }
887
888    /* Load addresses within 2GB of TB with 2 (or rarely 3) insns.  */
889    if (!in_prologue && USE_REG_TB && tb_diff == (int32_t)tb_diff) {
890        tcg_out_mem_long(s, ADDI, ADD, ret, TCG_REG_TB, tb_diff);
891        return;
892    }
893
894    /* Use the constant pool, if possible.  */
895    if (!in_prologue && USE_REG_TB) {
896        new_pool_label(s, arg, R_PPC_ADDR16, s->code_ptr,
897                       tcg_tbrel_diff(s, NULL));
898        tcg_out32(s, LD | TAI(ret, TCG_REG_TB, 0));
899        return;
900    }
901
902    tmp = arg >> 31 >> 1;
903    tcg_out_movi(s, TCG_TYPE_I32, ret, tmp);
904    if (tmp) {
905        tcg_out_shli64(s, ret, ret, 32);
906    }
907    if (arg & 0xffff0000) {
908        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
909    }
910    if (arg & 0xffff) {
911        tcg_out32(s, ORI | SAI(ret, ret, arg));
912    }
913}
914
915static void tcg_out_dupi_vec(TCGContext *s, TCGType type, TCGReg ret,
916                             tcg_target_long val)
917{
918    uint32_t load_insn;
919    int rel, low;
920    intptr_t add;
921
922    low = (int8_t)val;
923    if (low >= -16 && low < 16) {
924        if (val == (tcg_target_long)dup_const(MO_8, low)) {
925            tcg_out32(s, VSPLTISB | VRT(ret) | ((val & 31) << 16));
926            return;
927        }
928        if (val == (tcg_target_long)dup_const(MO_16, low)) {
929            tcg_out32(s, VSPLTISH | VRT(ret) | ((val & 31) << 16));
930            return;
931        }
932        if (val == (tcg_target_long)dup_const(MO_32, low)) {
933            tcg_out32(s, VSPLTISW | VRT(ret) | ((val & 31) << 16));
934            return;
935        }
936    }
937    if (have_isa_3_00 && val == (tcg_target_long)dup_const(MO_8, val)) {
938        tcg_out32(s, XXSPLTIB | VRT(ret) | ((val & 0xff) << 11));
939        return;
940    }
941
942    /*
943     * Otherwise we must load the value from the constant pool.
944     */
945    if (USE_REG_TB) {
946        rel = R_PPC_ADDR16;
947        add = tcg_tbrel_diff(s, NULL);
948    } else {
949        rel = R_PPC_ADDR32;
950        add = 0;
951    }
952
953    if (have_vsx) {
954        load_insn = type == TCG_TYPE_V64 ? LXSDX : LXVDSX;
955        load_insn |= VRT(ret) | RB(TCG_REG_TMP1);
956        if (TCG_TARGET_REG_BITS == 64) {
957            new_pool_label(s, val, rel, s->code_ptr, add);
958        } else {
959            new_pool_l2(s, rel, s->code_ptr, add, val, val);
960        }
961    } else {
962        load_insn = LVX | VRT(ret) | RB(TCG_REG_TMP1);
963        if (TCG_TARGET_REG_BITS == 64) {
964            new_pool_l2(s, rel, s->code_ptr, add, val, val);
965        } else {
966            new_pool_l4(s, rel, s->code_ptr, add, val, val, val, val);
967        }
968    }
969
970    if (USE_REG_TB) {
971        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, 0, 0));
972        load_insn |= RA(TCG_REG_TB);
973    } else {
974        tcg_out32(s, ADDIS | TAI(TCG_REG_TMP1, 0, 0));
975        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, TCG_REG_TMP1, 0));
976    }
977    tcg_out32(s, load_insn);
978}
979
980static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,
981                         tcg_target_long arg)
982{
983    switch (type) {
984    case TCG_TYPE_I32:
985    case TCG_TYPE_I64:
986        tcg_debug_assert(ret < TCG_REG_V0);
987        tcg_out_movi_int(s, type, ret, arg, false);
988        break;
989
990    case TCG_TYPE_V64:
991    case TCG_TYPE_V128:
992        tcg_debug_assert(ret >= TCG_REG_V0);
993        tcg_out_dupi_vec(s, type, ret, arg);
994        break;
995
996    default:
997        g_assert_not_reached();
998    }
999}
1000
1001static bool mask_operand(uint32_t c, int *mb, int *me)
1002{
1003    uint32_t lsb, test;
1004
1005    /* Accept a bit pattern like:
1006           0....01....1
1007           1....10....0
1008           0..01..10..0
1009       Keep track of the transitions.  */
1010    if (c == 0 || c == -1) {
1011        return false;
1012    }
1013    test = c;
1014    lsb = test & -test;
1015    test += lsb;
1016    if (test & (test - 1)) {
1017        return false;
1018    }
1019
1020    *me = clz32(lsb);
1021    *mb = test ? clz32(test & -test) + 1 : 0;
1022    return true;
1023}
1024
1025static bool mask64_operand(uint64_t c, int *mb, int *me)
1026{
1027    uint64_t lsb;
1028
1029    if (c == 0) {
1030        return false;
1031    }
1032
1033    lsb = c & -c;
1034    /* Accept 1..10..0.  */
1035    if (c == -lsb) {
1036        *mb = 0;
1037        *me = clz64(lsb);
1038        return true;
1039    }
1040    /* Accept 0..01..1.  */
1041    if (lsb == 1 && (c & (c + 1)) == 0) {
1042        *mb = clz64(c + 1) + 1;
1043        *me = 63;
1044        return true;
1045    }
1046    return false;
1047}
1048
1049static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1050{
1051    int mb, me;
1052
1053    if (mask_operand(c, &mb, &me)) {
1054        tcg_out_rlw(s, RLWINM, dst, src, 0, mb, me);
1055    } else if ((c & 0xffff) == c) {
1056        tcg_out32(s, ANDI | SAI(src, dst, c));
1057        return;
1058    } else if ((c & 0xffff0000) == c) {
1059        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1060        return;
1061    } else {
1062        tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R0, c);
1063        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1064    }
1065}
1066
1067static void tcg_out_andi64(TCGContext *s, TCGReg dst, TCGReg src, uint64_t c)
1068{
1069    int mb, me;
1070
1071    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1072    if (mask64_operand(c, &mb, &me)) {
1073        if (mb == 0) {
1074            tcg_out_rld(s, RLDICR, dst, src, 0, me);
1075        } else {
1076            tcg_out_rld(s, RLDICL, dst, src, 0, mb);
1077        }
1078    } else if ((c & 0xffff) == c) {
1079        tcg_out32(s, ANDI | SAI(src, dst, c));
1080        return;
1081    } else if ((c & 0xffff0000) == c) {
1082        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1083        return;
1084    } else {
1085        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, c);
1086        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1087    }
1088}
1089
1090static void tcg_out_zori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c,
1091                           int op_lo, int op_hi)
1092{
1093    if (c >> 16) {
1094        tcg_out32(s, op_hi | SAI(src, dst, c >> 16));
1095        src = dst;
1096    }
1097    if (c & 0xffff) {
1098        tcg_out32(s, op_lo | SAI(src, dst, c));
1099        src = dst;
1100    }
1101}
1102
1103static void tcg_out_ori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1104{
1105    tcg_out_zori32(s, dst, src, c, ORI, ORIS);
1106}
1107
1108static void tcg_out_xori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1109{
1110    tcg_out_zori32(s, dst, src, c, XORI, XORIS);
1111}
1112
1113static void tcg_out_b(TCGContext *s, int mask, const tcg_insn_unit *target)
1114{
1115    ptrdiff_t disp = tcg_pcrel_diff(s, target);
1116    if (in_range_b(disp)) {
1117        tcg_out32(s, B | (disp & 0x3fffffc) | mask);
1118    } else {
1119        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, (uintptr_t)target);
1120        tcg_out32(s, MTSPR | RS(TCG_REG_R0) | CTR);
1121        tcg_out32(s, BCCTR | BO_ALWAYS | mask);
1122    }
1123}
1124
1125static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
1126                             TCGReg base, tcg_target_long offset)
1127{
1128    tcg_target_long orig = offset, l0, l1, extra = 0, align = 0;
1129    bool is_int_store = false;
1130    TCGReg rs = TCG_REG_TMP1;
1131
1132    switch (opi) {
1133    case LD: case LWA:
1134        align = 3;
1135        /* FALLTHRU */
1136    default:
1137        if (rt > TCG_REG_R0 && rt < TCG_REG_V0) {
1138            rs = rt;
1139            break;
1140        }
1141        break;
1142    case LXSD:
1143    case STXSD:
1144        align = 3;
1145        break;
1146    case LXV:
1147    case STXV:
1148        align = 15;
1149        break;
1150    case STD:
1151        align = 3;
1152        /* FALLTHRU */
1153    case STB: case STH: case STW:
1154        is_int_store = true;
1155        break;
1156    }
1157
1158    /* For unaligned, or very large offsets, use the indexed form.  */
1159    if (offset & align || offset != (int32_t)offset || opi == 0) {
1160        if (rs == base) {
1161            rs = TCG_REG_R0;
1162        }
1163        tcg_debug_assert(!is_int_store || rs != rt);
1164        tcg_out_movi(s, TCG_TYPE_PTR, rs, orig);
1165        tcg_out32(s, opx | TAB(rt & 31, base, rs));
1166        return;
1167    }
1168
1169    l0 = (int16_t)offset;
1170    offset = (offset - l0) >> 16;
1171    l1 = (int16_t)offset;
1172
1173    if (l1 < 0 && orig >= 0) {
1174        extra = 0x4000;
1175        l1 = (int16_t)(offset - 0x4000);
1176    }
1177    if (l1) {
1178        tcg_out32(s, ADDIS | TAI(rs, base, l1));
1179        base = rs;
1180    }
1181    if (extra) {
1182        tcg_out32(s, ADDIS | TAI(rs, base, extra));
1183        base = rs;
1184    }
1185    if (opi != ADDI || base != rt || l0 != 0) {
1186        tcg_out32(s, opi | TAI(rt & 31, base, l0));
1187    }
1188}
1189
1190static void tcg_out_vsldoi(TCGContext *s, TCGReg ret,
1191                           TCGReg va, TCGReg vb, int shb)
1192{
1193    tcg_out32(s, VSLDOI | VRT(ret) | VRA(va) | VRB(vb) | (shb << 6));
1194}
1195
1196static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
1197                       TCGReg base, intptr_t offset)
1198{
1199    int shift;
1200
1201    switch (type) {
1202    case TCG_TYPE_I32:
1203        if (ret < TCG_REG_V0) {
1204            tcg_out_mem_long(s, LWZ, LWZX, ret, base, offset);
1205            break;
1206        }
1207        if (have_isa_2_07 && have_vsx) {
1208            tcg_out_mem_long(s, 0, LXSIWZX, ret, base, offset);
1209            break;
1210        }
1211        tcg_debug_assert((offset & 3) == 0);
1212        tcg_out_mem_long(s, 0, LVEWX, ret, base, offset);
1213        shift = (offset - 4) & 0xc;
1214        if (shift) {
1215            tcg_out_vsldoi(s, ret, ret, ret, shift);
1216        }
1217        break;
1218    case TCG_TYPE_I64:
1219        if (ret < TCG_REG_V0) {
1220            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1221            tcg_out_mem_long(s, LD, LDX, ret, base, offset);
1222            break;
1223        }
1224        /* fallthru */
1225    case TCG_TYPE_V64:
1226        tcg_debug_assert(ret >= TCG_REG_V0);
1227        if (have_vsx) {
1228            tcg_out_mem_long(s, have_isa_3_00 ? LXSD : 0, LXSDX,
1229                             ret, base, offset);
1230            break;
1231        }
1232        tcg_debug_assert((offset & 7) == 0);
1233        tcg_out_mem_long(s, 0, LVX, ret, base, offset & -16);
1234        if (offset & 8) {
1235            tcg_out_vsldoi(s, ret, ret, ret, 8);
1236        }
1237        break;
1238    case TCG_TYPE_V128:
1239        tcg_debug_assert(ret >= TCG_REG_V0);
1240        tcg_debug_assert((offset & 15) == 0);
1241        tcg_out_mem_long(s, have_isa_3_00 ? LXV : 0,
1242                         LVX, ret, base, offset);
1243        break;
1244    default:
1245        g_assert_not_reached();
1246    }
1247}
1248
1249static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
1250                              TCGReg base, intptr_t offset)
1251{
1252    int shift;
1253
1254    switch (type) {
1255    case TCG_TYPE_I32:
1256        if (arg < TCG_REG_V0) {
1257            tcg_out_mem_long(s, STW, STWX, arg, base, offset);
1258            break;
1259        }
1260        if (have_isa_2_07 && have_vsx) {
1261            tcg_out_mem_long(s, 0, STXSIWX, arg, base, offset);
1262            break;
1263        }
1264        assert((offset & 3) == 0);
1265        tcg_debug_assert((offset & 3) == 0);
1266        shift = (offset - 4) & 0xc;
1267        if (shift) {
1268            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, shift);
1269            arg = TCG_VEC_TMP1;
1270        }
1271        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1272        break;
1273    case TCG_TYPE_I64:
1274        if (arg < TCG_REG_V0) {
1275            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1276            tcg_out_mem_long(s, STD, STDX, arg, base, offset);
1277            break;
1278        }
1279        /* fallthru */
1280    case TCG_TYPE_V64:
1281        tcg_debug_assert(arg >= TCG_REG_V0);
1282        if (have_vsx) {
1283            tcg_out_mem_long(s, have_isa_3_00 ? STXSD : 0,
1284                             STXSDX, arg, base, offset);
1285            break;
1286        }
1287        tcg_debug_assert((offset & 7) == 0);
1288        if (offset & 8) {
1289            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, 8);
1290            arg = TCG_VEC_TMP1;
1291        }
1292        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1293        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset + 4);
1294        break;
1295    case TCG_TYPE_V128:
1296        tcg_debug_assert(arg >= TCG_REG_V0);
1297        tcg_out_mem_long(s, have_isa_3_00 ? STXV : 0,
1298                         STVX, arg, base, offset);
1299        break;
1300    default:
1301        g_assert_not_reached();
1302    }
1303}
1304
1305static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1306                               TCGReg base, intptr_t ofs)
1307{
1308    return false;
1309}
1310
1311static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2,
1312                        int const_arg2, int cr, TCGType type)
1313{
1314    int imm;
1315    uint32_t op;
1316
1317    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1318
1319    /* Simplify the comparisons below wrt CMPI.  */
1320    if (type == TCG_TYPE_I32) {
1321        arg2 = (int32_t)arg2;
1322    }
1323
1324    switch (cond) {
1325    case TCG_COND_EQ:
1326    case TCG_COND_NE:
1327        if (const_arg2) {
1328            if ((int16_t) arg2 == arg2) {
1329                op = CMPI;
1330                imm = 1;
1331                break;
1332            } else if ((uint16_t) arg2 == arg2) {
1333                op = CMPLI;
1334                imm = 1;
1335                break;
1336            }
1337        }
1338        op = CMPL;
1339        imm = 0;
1340        break;
1341
1342    case TCG_COND_LT:
1343    case TCG_COND_GE:
1344    case TCG_COND_LE:
1345    case TCG_COND_GT:
1346        if (const_arg2) {
1347            if ((int16_t) arg2 == arg2) {
1348                op = CMPI;
1349                imm = 1;
1350                break;
1351            }
1352        }
1353        op = CMP;
1354        imm = 0;
1355        break;
1356
1357    case TCG_COND_LTU:
1358    case TCG_COND_GEU:
1359    case TCG_COND_LEU:
1360    case TCG_COND_GTU:
1361        if (const_arg2) {
1362            if ((uint16_t) arg2 == arg2) {
1363                op = CMPLI;
1364                imm = 1;
1365                break;
1366            }
1367        }
1368        op = CMPL;
1369        imm = 0;
1370        break;
1371
1372    default:
1373        tcg_abort();
1374    }
1375    op |= BF(cr) | ((type == TCG_TYPE_I64) << 21);
1376
1377    if (imm) {
1378        tcg_out32(s, op | RA(arg1) | (arg2 & 0xffff));
1379    } else {
1380        if (const_arg2) {
1381            tcg_out_movi(s, type, TCG_REG_R0, arg2);
1382            arg2 = TCG_REG_R0;
1383        }
1384        tcg_out32(s, op | RA(arg1) | RB(arg2));
1385    }
1386}
1387
1388static void tcg_out_setcond_eq0(TCGContext *s, TCGType type,
1389                                TCGReg dst, TCGReg src)
1390{
1391    if (type == TCG_TYPE_I32) {
1392        tcg_out32(s, CNTLZW | RS(src) | RA(dst));
1393        tcg_out_shri32(s, dst, dst, 5);
1394    } else {
1395        tcg_out32(s, CNTLZD | RS(src) | RA(dst));
1396        tcg_out_shri64(s, dst, dst, 6);
1397    }
1398}
1399
1400static void tcg_out_setcond_ne0(TCGContext *s, TCGReg dst, TCGReg src)
1401{
1402    /* X != 0 implies X + -1 generates a carry.  Extra addition
1403       trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C.  */
1404    if (dst != src) {
1405        tcg_out32(s, ADDIC | TAI(dst, src, -1));
1406        tcg_out32(s, SUBFE | TAB(dst, dst, src));
1407    } else {
1408        tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1));
1409        tcg_out32(s, SUBFE | TAB(dst, TCG_REG_R0, src));
1410    }
1411}
1412
1413static TCGReg tcg_gen_setcond_xor(TCGContext *s, TCGReg arg1, TCGArg arg2,
1414                                  bool const_arg2)
1415{
1416    if (const_arg2) {
1417        if ((uint32_t)arg2 == arg2) {
1418            tcg_out_xori32(s, TCG_REG_R0, arg1, arg2);
1419        } else {
1420            tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, arg2);
1421            tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, TCG_REG_R0));
1422        }
1423    } else {
1424        tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, arg2));
1425    }
1426    return TCG_REG_R0;
1427}
1428
1429static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
1430                            TCGArg arg0, TCGArg arg1, TCGArg arg2,
1431                            int const_arg2)
1432{
1433    int crop, sh;
1434
1435    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1436
1437    /* Ignore high bits of a potential constant arg2.  */
1438    if (type == TCG_TYPE_I32) {
1439        arg2 = (uint32_t)arg2;
1440    }
1441
1442    /* Handle common and trivial cases before handling anything else.  */
1443    if (arg2 == 0) {
1444        switch (cond) {
1445        case TCG_COND_EQ:
1446            tcg_out_setcond_eq0(s, type, arg0, arg1);
1447            return;
1448        case TCG_COND_NE:
1449            if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1450                tcg_out_ext32u(s, TCG_REG_R0, arg1);
1451                arg1 = TCG_REG_R0;
1452            }
1453            tcg_out_setcond_ne0(s, arg0, arg1);
1454            return;
1455        case TCG_COND_GE:
1456            tcg_out32(s, NOR | SAB(arg1, arg0, arg1));
1457            arg1 = arg0;
1458            /* FALLTHRU */
1459        case TCG_COND_LT:
1460            /* Extract the sign bit.  */
1461            if (type == TCG_TYPE_I32) {
1462                tcg_out_shri32(s, arg0, arg1, 31);
1463            } else {
1464                tcg_out_shri64(s, arg0, arg1, 63);
1465            }
1466            return;
1467        default:
1468            break;
1469        }
1470    }
1471
1472    /* If we have ISEL, we can implement everything with 3 or 4 insns.
1473       All other cases below are also at least 3 insns, so speed up the
1474       code generator by not considering them and always using ISEL.  */
1475    if (have_isel) {
1476        int isel, tab;
1477
1478        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1479
1480        isel = tcg_to_isel[cond];
1481
1482        tcg_out_movi(s, type, arg0, 1);
1483        if (isel & 1) {
1484            /* arg0 = (bc ? 0 : 1) */
1485            tab = TAB(arg0, 0, arg0);
1486            isel &= ~1;
1487        } else {
1488            /* arg0 = (bc ? 1 : 0) */
1489            tcg_out_movi(s, type, TCG_REG_R0, 0);
1490            tab = TAB(arg0, arg0, TCG_REG_R0);
1491        }
1492        tcg_out32(s, isel | tab);
1493        return;
1494    }
1495
1496    switch (cond) {
1497    case TCG_COND_EQ:
1498        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1499        tcg_out_setcond_eq0(s, type, arg0, arg1);
1500        return;
1501
1502    case TCG_COND_NE:
1503        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1504        /* Discard the high bits only once, rather than both inputs.  */
1505        if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1506            tcg_out_ext32u(s, TCG_REG_R0, arg1);
1507            arg1 = TCG_REG_R0;
1508        }
1509        tcg_out_setcond_ne0(s, arg0, arg1);
1510        return;
1511
1512    case TCG_COND_GT:
1513    case TCG_COND_GTU:
1514        sh = 30;
1515        crop = 0;
1516        goto crtest;
1517
1518    case TCG_COND_LT:
1519    case TCG_COND_LTU:
1520        sh = 29;
1521        crop = 0;
1522        goto crtest;
1523
1524    case TCG_COND_GE:
1525    case TCG_COND_GEU:
1526        sh = 31;
1527        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_LT) | BB(7, CR_LT);
1528        goto crtest;
1529
1530    case TCG_COND_LE:
1531    case TCG_COND_LEU:
1532        sh = 31;
1533        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_GT) | BB(7, CR_GT);
1534    crtest:
1535        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1536        if (crop) {
1537            tcg_out32(s, crop);
1538        }
1539        tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1540        tcg_out_rlw(s, RLWINM, arg0, TCG_REG_R0, sh, 31, 31);
1541        break;
1542
1543    default:
1544        tcg_abort();
1545    }
1546}
1547
1548static void tcg_out_bc(TCGContext *s, int bc, TCGLabel *l)
1549{
1550    if (l->has_value) {
1551        bc |= reloc_pc14_val(tcg_splitwx_to_rx(s->code_ptr), l->u.value_ptr);
1552    } else {
1553        tcg_out_reloc(s, s->code_ptr, R_PPC_REL14, l, 0);
1554    }
1555    tcg_out32(s, bc);
1556}
1557
1558static void tcg_out_brcond(TCGContext *s, TCGCond cond,
1559                           TCGArg arg1, TCGArg arg2, int const_arg2,
1560                           TCGLabel *l, TCGType type)
1561{
1562    tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1563    tcg_out_bc(s, tcg_to_bc[cond], l);
1564}
1565
1566static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
1567                            TCGArg dest, TCGArg c1, TCGArg c2, TCGArg v1,
1568                            TCGArg v2, bool const_c2)
1569{
1570    /* If for some reason both inputs are zero, don't produce bad code.  */
1571    if (v1 == 0 && v2 == 0) {
1572        tcg_out_movi(s, type, dest, 0);
1573        return;
1574    }
1575
1576    tcg_out_cmp(s, cond, c1, c2, const_c2, 7, type);
1577
1578    if (have_isel) {
1579        int isel = tcg_to_isel[cond];
1580
1581        /* Swap the V operands if the operation indicates inversion.  */
1582        if (isel & 1) {
1583            int t = v1;
1584            v1 = v2;
1585            v2 = t;
1586            isel &= ~1;
1587        }
1588        /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand.  */
1589        if (v2 == 0) {
1590            tcg_out_movi(s, type, TCG_REG_R0, 0);
1591        }
1592        tcg_out32(s, isel | TAB(dest, v1, v2));
1593    } else {
1594        if (dest == v2) {
1595            cond = tcg_invert_cond(cond);
1596            v2 = v1;
1597        } else if (dest != v1) {
1598            if (v1 == 0) {
1599                tcg_out_movi(s, type, dest, 0);
1600            } else {
1601                tcg_out_mov(s, type, dest, v1);
1602            }
1603        }
1604        /* Branch forward over one insn */
1605        tcg_out32(s, tcg_to_bc[cond] | 8);
1606        if (v2 == 0) {
1607            tcg_out_movi(s, type, dest, 0);
1608        } else {
1609            tcg_out_mov(s, type, dest, v2);
1610        }
1611    }
1612}
1613
1614static void tcg_out_cntxz(TCGContext *s, TCGType type, uint32_t opc,
1615                          TCGArg a0, TCGArg a1, TCGArg a2, bool const_a2)
1616{
1617    if (const_a2 && a2 == (type == TCG_TYPE_I32 ? 32 : 64)) {
1618        tcg_out32(s, opc | RA(a0) | RS(a1));
1619    } else {
1620        tcg_out_cmp(s, TCG_COND_EQ, a1, 0, 1, 7, type);
1621        /* Note that the only other valid constant for a2 is 0.  */
1622        if (have_isel) {
1623            tcg_out32(s, opc | RA(TCG_REG_R0) | RS(a1));
1624            tcg_out32(s, tcg_to_isel[TCG_COND_EQ] | TAB(a0, a2, TCG_REG_R0));
1625        } else if (!const_a2 && a0 == a2) {
1626            tcg_out32(s, tcg_to_bc[TCG_COND_EQ] | 8);
1627            tcg_out32(s, opc | RA(a0) | RS(a1));
1628        } else {
1629            tcg_out32(s, opc | RA(a0) | RS(a1));
1630            tcg_out32(s, tcg_to_bc[TCG_COND_NE] | 8);
1631            if (const_a2) {
1632                tcg_out_movi(s, type, a0, 0);
1633            } else {
1634                tcg_out_mov(s, type, a0, a2);
1635            }
1636        }
1637    }
1638}
1639
1640static void tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1641                         const int *const_args)
1642{
1643    static const struct { uint8_t bit1, bit2; } bits[] = {
1644        [TCG_COND_LT ] = { CR_LT, CR_LT },
1645        [TCG_COND_LE ] = { CR_LT, CR_GT },
1646        [TCG_COND_GT ] = { CR_GT, CR_GT },
1647        [TCG_COND_GE ] = { CR_GT, CR_LT },
1648        [TCG_COND_LTU] = { CR_LT, CR_LT },
1649        [TCG_COND_LEU] = { CR_LT, CR_GT },
1650        [TCG_COND_GTU] = { CR_GT, CR_GT },
1651        [TCG_COND_GEU] = { CR_GT, CR_LT },
1652    };
1653
1654    TCGCond cond = args[4], cond2;
1655    TCGArg al, ah, bl, bh;
1656    int blconst, bhconst;
1657    int op, bit1, bit2;
1658
1659    al = args[0];
1660    ah = args[1];
1661    bl = args[2];
1662    bh = args[3];
1663    blconst = const_args[2];
1664    bhconst = const_args[3];
1665
1666    switch (cond) {
1667    case TCG_COND_EQ:
1668        op = CRAND;
1669        goto do_equality;
1670    case TCG_COND_NE:
1671        op = CRNAND;
1672    do_equality:
1673        tcg_out_cmp(s, cond, al, bl, blconst, 6, TCG_TYPE_I32);
1674        tcg_out_cmp(s, cond, ah, bh, bhconst, 7, TCG_TYPE_I32);
1675        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1676        break;
1677
1678    case TCG_COND_LT:
1679    case TCG_COND_LE:
1680    case TCG_COND_GT:
1681    case TCG_COND_GE:
1682    case TCG_COND_LTU:
1683    case TCG_COND_LEU:
1684    case TCG_COND_GTU:
1685    case TCG_COND_GEU:
1686        bit1 = bits[cond].bit1;
1687        bit2 = bits[cond].bit2;
1688        op = (bit1 != bit2 ? CRANDC : CRAND);
1689        cond2 = tcg_unsigned_cond(cond);
1690
1691        tcg_out_cmp(s, cond, ah, bh, bhconst, 6, TCG_TYPE_I32);
1692        tcg_out_cmp(s, cond2, al, bl, blconst, 7, TCG_TYPE_I32);
1693        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, bit2));
1694        tcg_out32(s, CROR | BT(7, CR_EQ) | BA(6, bit1) | BB(7, CR_EQ));
1695        break;
1696
1697    default:
1698        tcg_abort();
1699    }
1700}
1701
1702static void tcg_out_setcond2(TCGContext *s, const TCGArg *args,
1703                             const int *const_args)
1704{
1705    tcg_out_cmp2(s, args + 1, const_args + 1);
1706    tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1707    tcg_out_rlw(s, RLWINM, args[0], TCG_REG_R0, 31, 31, 31);
1708}
1709
1710static void tcg_out_brcond2 (TCGContext *s, const TCGArg *args,
1711                             const int *const_args)
1712{
1713    tcg_out_cmp2(s, args, const_args);
1714    tcg_out_bc(s, BC | BI(7, CR_EQ) | BO_COND_TRUE, arg_label(args[5]));
1715}
1716
1717static void tcg_out_mb(TCGContext *s, TCGArg a0)
1718{
1719    uint32_t insn = HWSYNC;
1720    a0 &= TCG_MO_ALL;
1721    if (a0 == TCG_MO_LD_LD) {
1722        insn = LWSYNC;
1723    } else if (a0 == TCG_MO_ST_ST) {
1724        insn = EIEIO;
1725    }
1726    tcg_out32(s, insn);
1727}
1728
1729void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_rx,
1730                              uintptr_t jmp_rw, uintptr_t addr)
1731{
1732    if (TCG_TARGET_REG_BITS == 64) {
1733        tcg_insn_unit i1, i2;
1734        intptr_t tb_diff = addr - tc_ptr;
1735        intptr_t br_diff = addr - (jmp_rx + 4);
1736        uint64_t pair;
1737
1738        /* This does not exercise the range of the branch, but we do
1739           still need to be able to load the new value of TCG_REG_TB.
1740           But this does still happen quite often.  */
1741        if (tb_diff == (int16_t)tb_diff) {
1742            i1 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, tb_diff);
1743            i2 = B | (br_diff & 0x3fffffc);
1744        } else {
1745            intptr_t lo = (int16_t)tb_diff;
1746            intptr_t hi = (int32_t)(tb_diff - lo);
1747            assert(tb_diff == hi + lo);
1748            i1 = ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, hi >> 16);
1749            i2 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, lo);
1750        }
1751#ifdef HOST_WORDS_BIGENDIAN
1752        pair = (uint64_t)i1 << 32 | i2;
1753#else
1754        pair = (uint64_t)i2 << 32 | i1;
1755#endif
1756
1757        /* As per the enclosing if, this is ppc64.  Avoid the _Static_assert
1758           within qatomic_set that would fail to build a ppc32 host.  */
1759        qatomic_set__nocheck((uint64_t *)jmp_rw, pair);
1760        flush_idcache_range(jmp_rx, jmp_rw, 8);
1761    } else {
1762        intptr_t diff = addr - jmp_rx;
1763        tcg_debug_assert(in_range_b(diff));
1764        qatomic_set((uint32_t *)jmp_rw, B | (diff & 0x3fffffc));
1765        flush_idcache_range(jmp_rx, jmp_rw, 4);
1766    }
1767}
1768
1769static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target)
1770{
1771#ifdef _CALL_AIX
1772    /* Look through the descriptor.  If the branch is in range, and we
1773       don't have to spend too much effort on building the toc.  */
1774    const void *tgt = ((const void * const *)target)[0];
1775    uintptr_t toc = ((const uintptr_t *)target)[1];
1776    intptr_t diff = tcg_pcrel_diff(s, tgt);
1777
1778    if (in_range_b(diff) && toc == (uint32_t)toc) {
1779        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, toc);
1780        tcg_out_b(s, LK, tgt);
1781    } else {
1782        /* Fold the low bits of the constant into the addresses below.  */
1783        intptr_t arg = (intptr_t)target;
1784        int ofs = (int16_t)arg;
1785
1786        if (ofs + 8 < 0x8000) {
1787            arg -= ofs;
1788        } else {
1789            ofs = 0;
1790        }
1791        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, arg);
1792        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_TMP1, ofs);
1793        tcg_out32(s, MTSPR | RA(TCG_REG_R0) | CTR);
1794        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_REG_TMP1, ofs + SZP);
1795        tcg_out32(s, BCCTR | BO_ALWAYS | LK);
1796    }
1797#elif defined(_CALL_ELF) && _CALL_ELF == 2
1798    intptr_t diff;
1799
1800    /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1801       address, which the callee uses to compute its TOC address.  */
1802    /* FIXME: when the branch is in range, we could avoid r12 load if we
1803       knew that the destination uses the same TOC, and what its local
1804       entry point offset is.  */
1805    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R12, (intptr_t)target);
1806
1807    diff = tcg_pcrel_diff(s, target);
1808    if (in_range_b(diff)) {
1809        tcg_out_b(s, LK, target);
1810    } else {
1811        tcg_out32(s, MTSPR | RS(TCG_REG_R12) | CTR);
1812        tcg_out32(s, BCCTR | BO_ALWAYS | LK);
1813    }
1814#else
1815    tcg_out_b(s, LK, target);
1816#endif
1817}
1818
1819static const uint32_t qemu_ldx_opc[16] = {
1820    [MO_UB] = LBZX,
1821    [MO_UW] = LHZX,
1822    [MO_UL] = LWZX,
1823    [MO_Q]  = LDX,
1824    [MO_SW] = LHAX,
1825    [MO_SL] = LWAX,
1826    [MO_BSWAP | MO_UB] = LBZX,
1827    [MO_BSWAP | MO_UW] = LHBRX,
1828    [MO_BSWAP | MO_UL] = LWBRX,
1829    [MO_BSWAP | MO_Q]  = LDBRX,
1830};
1831
1832static const uint32_t qemu_stx_opc[16] = {
1833    [MO_UB] = STBX,
1834    [MO_UW] = STHX,
1835    [MO_UL] = STWX,
1836    [MO_Q]  = STDX,
1837    [MO_BSWAP | MO_UB] = STBX,
1838    [MO_BSWAP | MO_UW] = STHBRX,
1839    [MO_BSWAP | MO_UL] = STWBRX,
1840    [MO_BSWAP | MO_Q]  = STDBRX,
1841};
1842
1843static const uint32_t qemu_exts_opc[4] = {
1844    EXTSB, EXTSH, EXTSW, 0
1845};
1846
1847#if defined (CONFIG_SOFTMMU)
1848#include "../tcg-ldst.c.inc"
1849
1850/* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
1851 *                                 int mmu_idx, uintptr_t ra)
1852 */
1853static void * const qemu_ld_helpers[16] = {
1854    [MO_UB]   = helper_ret_ldub_mmu,
1855    [MO_LEUW] = helper_le_lduw_mmu,
1856    [MO_LEUL] = helper_le_ldul_mmu,
1857    [MO_LEQ]  = helper_le_ldq_mmu,
1858    [MO_BEUW] = helper_be_lduw_mmu,
1859    [MO_BEUL] = helper_be_ldul_mmu,
1860    [MO_BEQ]  = helper_be_ldq_mmu,
1861};
1862
1863/* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
1864 *                                 uintxx_t val, int mmu_idx, uintptr_t ra)
1865 */
1866static void * const qemu_st_helpers[16] = {
1867    [MO_UB]   = helper_ret_stb_mmu,
1868    [MO_LEUW] = helper_le_stw_mmu,
1869    [MO_LEUL] = helper_le_stl_mmu,
1870    [MO_LEQ]  = helper_le_stq_mmu,
1871    [MO_BEUW] = helper_be_stw_mmu,
1872    [MO_BEUL] = helper_be_stl_mmu,
1873    [MO_BEQ]  = helper_be_stq_mmu,
1874};
1875
1876/* We expect to use a 16-bit negative offset from ENV.  */
1877QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
1878QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -32768);
1879
1880/* Perform the TLB load and compare.  Places the result of the comparison
1881   in CR7, loads the addend of the TLB into R3, and returns the register
1882   containing the guest address (zero-extended into R4).  Clobbers R0 and R2. */
1883
1884static TCGReg tcg_out_tlb_read(TCGContext *s, MemOp opc,
1885                               TCGReg addrlo, TCGReg addrhi,
1886                               int mem_index, bool is_read)
1887{
1888    int cmp_off
1889        = (is_read
1890           ? offsetof(CPUTLBEntry, addr_read)
1891           : offsetof(CPUTLBEntry, addr_write));
1892    int fast_off = TLB_MASK_TABLE_OFS(mem_index);
1893    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1894    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1895    unsigned s_bits = opc & MO_SIZE;
1896    unsigned a_bits = get_alignment_bits(opc);
1897
1898    /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx].  */
1899    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_AREG0, mask_off);
1900    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R4, TCG_AREG0, table_off);
1901
1902    /* Extract the page index, shifted into place for tlb index.  */
1903    if (TCG_TARGET_REG_BITS == 32) {
1904        tcg_out_shri32(s, TCG_REG_TMP1, addrlo,
1905                       TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
1906    } else {
1907        tcg_out_shri64(s, TCG_REG_TMP1, addrlo,
1908                       TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
1909    }
1910    tcg_out32(s, AND | SAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_TMP1));
1911
1912    /* Load the TLB comparator.  */
1913    if (cmp_off == 0 && TCG_TARGET_REG_BITS >= TARGET_LONG_BITS) {
1914        uint32_t lxu = (TCG_TARGET_REG_BITS == 32 || TARGET_LONG_BITS == 32
1915                        ? LWZUX : LDUX);
1916        tcg_out32(s, lxu | TAB(TCG_REG_TMP1, TCG_REG_R3, TCG_REG_R4));
1917    } else {
1918        tcg_out32(s, ADD | TAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_R4));
1919        if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1920            tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP1, TCG_REG_R3, cmp_off + 4);
1921            tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R4, TCG_REG_R3, cmp_off);
1922        } else {
1923            tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP1, TCG_REG_R3, cmp_off);
1924        }
1925    }
1926
1927    /* Load the TLB addend for use on the fast path.  Do this asap
1928       to minimize any load use delay.  */
1929    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_REG_R3,
1930               offsetof(CPUTLBEntry, addend));
1931
1932    /* Clear the non-page, non-alignment bits from the address */
1933    if (TCG_TARGET_REG_BITS == 32) {
1934        /* We don't support unaligned accesses on 32-bits.
1935         * Preserve the bottom bits and thus trigger a comparison
1936         * failure on unaligned accesses.
1937         */
1938        if (a_bits < s_bits) {
1939            a_bits = s_bits;
1940        }
1941        tcg_out_rlw(s, RLWINM, TCG_REG_R0, addrlo, 0,
1942                    (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
1943    } else {
1944        TCGReg t = addrlo;
1945
1946        /* If the access is unaligned, we need to make sure we fail if we
1947         * cross a page boundary.  The trick is to add the access size-1
1948         * to the address before masking the low bits.  That will make the
1949         * address overflow to the next page if we cross a page boundary,
1950         * which will then force a mismatch of the TLB compare.
1951         */
1952        if (a_bits < s_bits) {
1953            unsigned a_mask = (1 << a_bits) - 1;
1954            unsigned s_mask = (1 << s_bits) - 1;
1955            tcg_out32(s, ADDI | TAI(TCG_REG_R0, t, s_mask - a_mask));
1956            t = TCG_REG_R0;
1957        }
1958
1959        /* Mask the address for the requested alignment.  */
1960        if (TARGET_LONG_BITS == 32) {
1961            tcg_out_rlw(s, RLWINM, TCG_REG_R0, t, 0,
1962                        (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
1963            /* Zero-extend the address for use in the final address.  */
1964            tcg_out_ext32u(s, TCG_REG_R4, addrlo);
1965            addrlo = TCG_REG_R4;
1966        } else if (a_bits == 0) {
1967            tcg_out_rld(s, RLDICR, TCG_REG_R0, t, 0, 63 - TARGET_PAGE_BITS);
1968        } else {
1969            tcg_out_rld(s, RLDICL, TCG_REG_R0, t,
1970                        64 - TARGET_PAGE_BITS, TARGET_PAGE_BITS - a_bits);
1971            tcg_out_rld(s, RLDICL, TCG_REG_R0, TCG_REG_R0, TARGET_PAGE_BITS, 0);
1972        }
1973    }
1974
1975    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1976        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
1977                    0, 7, TCG_TYPE_I32);
1978        tcg_out_cmp(s, TCG_COND_EQ, addrhi, TCG_REG_R4, 0, 6, TCG_TYPE_I32);
1979        tcg_out32(s, CRAND | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1980    } else {
1981        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
1982                    0, 7, TCG_TYPE_TL);
1983    }
1984
1985    return addrlo;
1986}
1987
1988/* Record the context of a call to the out of line helper code for the slow
1989   path for a load or store, so that we can later generate the correct
1990   helper code.  */
1991static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
1992                                TCGReg datalo_reg, TCGReg datahi_reg,
1993                                TCGReg addrlo_reg, TCGReg addrhi_reg,
1994                                tcg_insn_unit *raddr, tcg_insn_unit *lptr)
1995{
1996    TCGLabelQemuLdst *label = new_ldst_label(s);
1997
1998    label->is_ld = is_ld;
1999    label->oi = oi;
2000    label->datalo_reg = datalo_reg;
2001    label->datahi_reg = datahi_reg;
2002    label->addrlo_reg = addrlo_reg;
2003    label->addrhi_reg = addrhi_reg;
2004    label->raddr = tcg_splitwx_to_rx(raddr);
2005    label->label_ptr[0] = lptr;
2006}
2007
2008static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2009{
2010    TCGMemOpIdx oi = lb->oi;
2011    MemOp opc = get_memop(oi);
2012    TCGReg hi, lo, arg = TCG_REG_R3;
2013
2014    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2015        return false;
2016    }
2017
2018    tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
2019
2020    lo = lb->addrlo_reg;
2021    hi = lb->addrhi_reg;
2022    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2023#ifdef TCG_TARGET_CALL_ALIGN_ARGS
2024        arg |= 1;
2025#endif
2026        tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2027        tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2028    } else {
2029        /* If the address needed to be zero-extended, we'll have already
2030           placed it in R4.  The only remaining case is 64-bit guest.  */
2031        tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
2032    }
2033
2034    tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
2035    tcg_out32(s, MFSPR | RT(arg) | LR);
2036
2037    tcg_out_call(s, qemu_ld_helpers[opc & (MO_BSWAP | MO_SIZE)]);
2038
2039    lo = lb->datalo_reg;
2040    hi = lb->datahi_reg;
2041    if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
2042        tcg_out_mov(s, TCG_TYPE_I32, lo, TCG_REG_R4);
2043        tcg_out_mov(s, TCG_TYPE_I32, hi, TCG_REG_R3);
2044    } else if (opc & MO_SIGN) {
2045        uint32_t insn = qemu_exts_opc[opc & MO_SIZE];
2046        tcg_out32(s, insn | RA(lo) | RS(TCG_REG_R3));
2047    } else {
2048        tcg_out_mov(s, TCG_TYPE_REG, lo, TCG_REG_R3);
2049    }
2050
2051    tcg_out_b(s, 0, lb->raddr);
2052    return true;
2053}
2054
2055static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2056{
2057    TCGMemOpIdx oi = lb->oi;
2058    MemOp opc = get_memop(oi);
2059    MemOp s_bits = opc & MO_SIZE;
2060    TCGReg hi, lo, arg = TCG_REG_R3;
2061
2062    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2063        return false;
2064    }
2065
2066    tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
2067
2068    lo = lb->addrlo_reg;
2069    hi = lb->addrhi_reg;
2070    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2071#ifdef TCG_TARGET_CALL_ALIGN_ARGS
2072        arg |= 1;
2073#endif
2074        tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2075        tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2076    } else {
2077        /* If the address needed to be zero-extended, we'll have already
2078           placed it in R4.  The only remaining case is 64-bit guest.  */
2079        tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
2080    }
2081
2082    lo = lb->datalo_reg;
2083    hi = lb->datahi_reg;
2084    if (TCG_TARGET_REG_BITS == 32) {
2085        switch (s_bits) {
2086        case MO_64:
2087#ifdef TCG_TARGET_CALL_ALIGN_ARGS
2088            arg |= 1;
2089#endif
2090            tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2091            /* FALLTHRU */
2092        case MO_32:
2093            tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2094            break;
2095        default:
2096            tcg_out_rlw(s, RLWINM, arg++, lo, 0, 32 - (8 << s_bits), 31);
2097            break;
2098        }
2099    } else {
2100        if (s_bits == MO_64) {
2101            tcg_out_mov(s, TCG_TYPE_I64, arg++, lo);
2102        } else {
2103            tcg_out_rld(s, RLDICL, arg++, lo, 0, 64 - (8 << s_bits));
2104        }
2105    }
2106
2107    tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
2108    tcg_out32(s, MFSPR | RT(arg) | LR);
2109
2110    tcg_out_call(s, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
2111
2112    tcg_out_b(s, 0, lb->raddr);
2113    return true;
2114}
2115#endif /* SOFTMMU */
2116
2117static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64)
2118{
2119    TCGReg datalo, datahi, addrlo, rbase;
2120    TCGReg addrhi __attribute__((unused));
2121    TCGMemOpIdx oi;
2122    MemOp opc, s_bits;
2123#ifdef CONFIG_SOFTMMU
2124    int mem_index;
2125    tcg_insn_unit *label_ptr;
2126#endif
2127
2128    datalo = *args++;
2129    datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
2130    addrlo = *args++;
2131    addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
2132    oi = *args++;
2133    opc = get_memop(oi);
2134    s_bits = opc & MO_SIZE;
2135
2136#ifdef CONFIG_SOFTMMU
2137    mem_index = get_mmuidx(oi);
2138    addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, true);
2139
2140    /* Load a pointer into the current opcode w/conditional branch-link. */
2141    label_ptr = s->code_ptr;
2142    tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
2143
2144    rbase = TCG_REG_R3;
2145#else  /* !CONFIG_SOFTMMU */
2146    rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
2147    if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
2148        tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
2149        addrlo = TCG_REG_TMP1;
2150    }
2151#endif
2152
2153    if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
2154        if (opc & MO_BSWAP) {
2155            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2156            tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
2157            tcg_out32(s, LWBRX | TAB(datahi, rbase, TCG_REG_R0));
2158        } else if (rbase != 0) {
2159            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2160            tcg_out32(s, LWZX | TAB(datahi, rbase, addrlo));
2161            tcg_out32(s, LWZX | TAB(datalo, rbase, TCG_REG_R0));
2162        } else if (addrlo == datahi) {
2163            tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
2164            tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
2165        } else {
2166            tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
2167            tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
2168        }
2169    } else {
2170        uint32_t insn = qemu_ldx_opc[opc & (MO_BSWAP | MO_SSIZE)];
2171        if (!have_isa_2_06 && insn == LDBRX) {
2172            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2173            tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
2174            tcg_out32(s, LWBRX | TAB(TCG_REG_R0, rbase, TCG_REG_R0));
2175            tcg_out_rld(s, RLDIMI, datalo, TCG_REG_R0, 32, 0);
2176        } else if (insn) {
2177            tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
2178        } else {
2179            insn = qemu_ldx_opc[opc & (MO_SIZE | MO_BSWAP)];
2180            tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
2181            insn = qemu_exts_opc[s_bits];
2182            tcg_out32(s, insn | RA(datalo) | RS(datalo));
2183        }
2184    }
2185
2186#ifdef CONFIG_SOFTMMU
2187    add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
2188                        s->code_ptr, label_ptr);
2189#endif
2190}
2191
2192static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
2193{
2194    TCGReg datalo, datahi, addrlo, rbase;
2195    TCGReg addrhi __attribute__((unused));
2196    TCGMemOpIdx oi;
2197    MemOp opc, s_bits;
2198#ifdef CONFIG_SOFTMMU
2199    int mem_index;
2200    tcg_insn_unit *label_ptr;
2201#endif
2202
2203    datalo = *args++;
2204    datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
2205    addrlo = *args++;
2206    addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
2207    oi = *args++;
2208    opc = get_memop(oi);
2209    s_bits = opc & MO_SIZE;
2210
2211#ifdef CONFIG_SOFTMMU
2212    mem_index = get_mmuidx(oi);
2213    addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, false);
2214
2215    /* Load a pointer into the current opcode w/conditional branch-link. */
2216    label_ptr = s->code_ptr;
2217    tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
2218
2219    rbase = TCG_REG_R3;
2220#else  /* !CONFIG_SOFTMMU */
2221    rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
2222    if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
2223        tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
2224        addrlo = TCG_REG_TMP1;
2225    }
2226#endif
2227
2228    if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
2229        if (opc & MO_BSWAP) {
2230            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2231            tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
2232            tcg_out32(s, STWBRX | SAB(datahi, rbase, TCG_REG_R0));
2233        } else if (rbase != 0) {
2234            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2235            tcg_out32(s, STWX | SAB(datahi, rbase, addrlo));
2236            tcg_out32(s, STWX | SAB(datalo, rbase, TCG_REG_R0));
2237        } else {
2238            tcg_out32(s, STW | TAI(datahi, addrlo, 0));
2239            tcg_out32(s, STW | TAI(datalo, addrlo, 4));
2240        }
2241    } else {
2242        uint32_t insn = qemu_stx_opc[opc & (MO_BSWAP | MO_SIZE)];
2243        if (!have_isa_2_06 && insn == STDBRX) {
2244            tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
2245            tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, addrlo, 4));
2246            tcg_out_shri64(s, TCG_REG_R0, datalo, 32);
2247            tcg_out32(s, STWBRX | SAB(TCG_REG_R0, rbase, TCG_REG_TMP1));
2248        } else {
2249            tcg_out32(s, insn | SAB(datalo, rbase, addrlo));
2250        }
2251    }
2252
2253#ifdef CONFIG_SOFTMMU
2254    add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
2255                        s->code_ptr, label_ptr);
2256#endif
2257}
2258
2259static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
2260{
2261    int i;
2262    for (i = 0; i < count; ++i) {
2263        p[i] = NOP;
2264    }
2265}
2266
2267/* Parameters for function call generation, used in tcg.c.  */
2268#define TCG_TARGET_STACK_ALIGN       16
2269#define TCG_TARGET_EXTEND_ARGS       1
2270
2271#ifdef _CALL_AIX
2272# define LINK_AREA_SIZE                (6 * SZR)
2273# define LR_OFFSET                     (1 * SZR)
2274# define TCG_TARGET_CALL_STACK_OFFSET  (LINK_AREA_SIZE + 8 * SZR)
2275#elif defined(TCG_TARGET_CALL_DARWIN)
2276# define LINK_AREA_SIZE                (6 * SZR)
2277# define LR_OFFSET                     (2 * SZR)
2278#elif TCG_TARGET_REG_BITS == 64
2279# if defined(_CALL_ELF) && _CALL_ELF == 2
2280#  define LINK_AREA_SIZE               (4 * SZR)
2281#  define LR_OFFSET                    (1 * SZR)
2282# endif
2283#else /* TCG_TARGET_REG_BITS == 32 */
2284# if defined(_CALL_SYSV)
2285#  define LINK_AREA_SIZE               (2 * SZR)
2286#  define LR_OFFSET                    (1 * SZR)
2287# endif
2288#endif
2289#ifndef LR_OFFSET
2290# error "Unhandled abi"
2291#endif
2292#ifndef TCG_TARGET_CALL_STACK_OFFSET
2293# define TCG_TARGET_CALL_STACK_OFFSET  LINK_AREA_SIZE
2294#endif
2295
2296#define CPU_TEMP_BUF_SIZE  (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
2297#define REG_SAVE_SIZE      ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
2298
2299#define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET   \
2300                     + TCG_STATIC_CALL_ARGS_SIZE    \
2301                     + CPU_TEMP_BUF_SIZE            \
2302                     + REG_SAVE_SIZE                \
2303                     + TCG_TARGET_STACK_ALIGN - 1)  \
2304                    & -TCG_TARGET_STACK_ALIGN)
2305
2306#define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
2307
2308static void tcg_target_qemu_prologue(TCGContext *s)
2309{
2310    int i;
2311
2312#ifdef _CALL_AIX
2313    const void **desc = (const void **)s->code_ptr;
2314    desc[0] = tcg_splitwx_to_rx(desc + 2);  /* entry point */
2315    desc[1] = 0;                            /* environment pointer */
2316    s->code_ptr = (void *)(desc + 2);       /* skip over descriptor */
2317#endif
2318
2319    tcg_set_frame(s, TCG_REG_CALL_STACK, REG_SAVE_BOT - CPU_TEMP_BUF_SIZE,
2320                  CPU_TEMP_BUF_SIZE);
2321
2322    /* Prologue */
2323    tcg_out32(s, MFSPR | RT(TCG_REG_R0) | LR);
2324    tcg_out32(s, (SZR == 8 ? STDU : STWU)
2325              | SAI(TCG_REG_R1, TCG_REG_R1, -FRAME_SIZE));
2326
2327    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2328        tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2329                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2330    }
2331    tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2332
2333#ifndef CONFIG_SOFTMMU
2334    if (guest_base) {
2335        tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true);
2336        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2337    }
2338#endif
2339
2340    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2341    tcg_out32(s, MTSPR | RS(tcg_target_call_iarg_regs[1]) | CTR);
2342    if (USE_REG_TB) {
2343        tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, tcg_target_call_iarg_regs[1]);
2344    }
2345    tcg_out32(s, BCCTR | BO_ALWAYS);
2346
2347    /* Epilogue */
2348    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
2349
2350    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2351    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2352        tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2353                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2354    }
2355    tcg_out32(s, MTSPR | RS(TCG_REG_R0) | LR);
2356    tcg_out32(s, ADDI | TAI(TCG_REG_R1, TCG_REG_R1, FRAME_SIZE));
2357    tcg_out32(s, BCLR | BO_ALWAYS);
2358}
2359
2360static void tcg_out_op(TCGContext *s, TCGOpcode opc, const TCGArg *args,
2361                       const int *const_args)
2362{
2363    TCGArg a0, a1, a2;
2364    int c;
2365
2366    switch (opc) {
2367    case INDEX_op_exit_tb:
2368        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, args[0]);
2369        tcg_out_b(s, 0, tcg_code_gen_epilogue);
2370        break;
2371    case INDEX_op_goto_tb:
2372        if (s->tb_jmp_insn_offset) {
2373            /* Direct jump. */
2374            if (TCG_TARGET_REG_BITS == 64) {
2375                /* Ensure the next insns are 8-byte aligned. */
2376                if ((uintptr_t)s->code_ptr & 7) {
2377                    tcg_out32(s, NOP);
2378                }
2379                s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2380                tcg_out32(s, ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2381                tcg_out32(s, ADDI | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2382            } else {
2383                s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2384                tcg_out32(s, B);
2385                s->tb_jmp_reset_offset[args[0]] = tcg_current_code_size(s);
2386                break;
2387            }
2388        } else {
2389            /* Indirect jump. */
2390            tcg_debug_assert(s->tb_jmp_insn_offset == NULL);
2391            tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, 0,
2392                       (intptr_t)(s->tb_jmp_insn_offset + args[0]));
2393        }
2394        tcg_out32(s, MTSPR | RS(TCG_REG_TB) | CTR);
2395        tcg_out32(s, BCCTR | BO_ALWAYS);
2396        set_jmp_reset_offset(s, args[0]);
2397        if (USE_REG_TB) {
2398            /* For the unlinked case, need to reset TCG_REG_TB.  */
2399            tcg_out_mem_long(s, ADDI, ADD, TCG_REG_TB, TCG_REG_TB,
2400                             -tcg_current_code_size(s));
2401        }
2402        break;
2403    case INDEX_op_goto_ptr:
2404        tcg_out32(s, MTSPR | RS(args[0]) | CTR);
2405        if (USE_REG_TB) {
2406            tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, args[0]);
2407        }
2408        tcg_out32(s, ADDI | TAI(TCG_REG_R3, 0, 0));
2409        tcg_out32(s, BCCTR | BO_ALWAYS);
2410        break;
2411    case INDEX_op_br:
2412        {
2413            TCGLabel *l = arg_label(args[0]);
2414            uint32_t insn = B;
2415
2416            if (l->has_value) {
2417                insn |= reloc_pc24_val(tcg_splitwx_to_rx(s->code_ptr),
2418                                       l->u.value_ptr);
2419            } else {
2420                tcg_out_reloc(s, s->code_ptr, R_PPC_REL24, l, 0);
2421            }
2422            tcg_out32(s, insn);
2423        }
2424        break;
2425    case INDEX_op_ld8u_i32:
2426    case INDEX_op_ld8u_i64:
2427        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2428        break;
2429    case INDEX_op_ld8s_i32:
2430    case INDEX_op_ld8s_i64:
2431        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2432        tcg_out32(s, EXTSB | RS(args[0]) | RA(args[0]));
2433        break;
2434    case INDEX_op_ld16u_i32:
2435    case INDEX_op_ld16u_i64:
2436        tcg_out_mem_long(s, LHZ, LHZX, args[0], args[1], args[2]);
2437        break;
2438    case INDEX_op_ld16s_i32:
2439    case INDEX_op_ld16s_i64:
2440        tcg_out_mem_long(s, LHA, LHAX, args[0], args[1], args[2]);
2441        break;
2442    case INDEX_op_ld_i32:
2443    case INDEX_op_ld32u_i64:
2444        tcg_out_mem_long(s, LWZ, LWZX, args[0], args[1], args[2]);
2445        break;
2446    case INDEX_op_ld32s_i64:
2447        tcg_out_mem_long(s, LWA, LWAX, args[0], args[1], args[2]);
2448        break;
2449    case INDEX_op_ld_i64:
2450        tcg_out_mem_long(s, LD, LDX, args[0], args[1], args[2]);
2451        break;
2452    case INDEX_op_st8_i32:
2453    case INDEX_op_st8_i64:
2454        tcg_out_mem_long(s, STB, STBX, args[0], args[1], args[2]);
2455        break;
2456    case INDEX_op_st16_i32:
2457    case INDEX_op_st16_i64:
2458        tcg_out_mem_long(s, STH, STHX, args[0], args[1], args[2]);
2459        break;
2460    case INDEX_op_st_i32:
2461    case INDEX_op_st32_i64:
2462        tcg_out_mem_long(s, STW, STWX, args[0], args[1], args[2]);
2463        break;
2464    case INDEX_op_st_i64:
2465        tcg_out_mem_long(s, STD, STDX, args[0], args[1], args[2]);
2466        break;
2467
2468    case INDEX_op_add_i32:
2469        a0 = args[0], a1 = args[1], a2 = args[2];
2470        if (const_args[2]) {
2471        do_addi_32:
2472            tcg_out_mem_long(s, ADDI, ADD, a0, a1, (int32_t)a2);
2473        } else {
2474            tcg_out32(s, ADD | TAB(a0, a1, a2));
2475        }
2476        break;
2477    case INDEX_op_sub_i32:
2478        a0 = args[0], a1 = args[1], a2 = args[2];
2479        if (const_args[1]) {
2480            if (const_args[2]) {
2481                tcg_out_movi(s, TCG_TYPE_I32, a0, a1 - a2);
2482            } else {
2483                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2484            }
2485        } else if (const_args[2]) {
2486            a2 = -a2;
2487            goto do_addi_32;
2488        } else {
2489            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2490        }
2491        break;
2492
2493    case INDEX_op_and_i32:
2494        a0 = args[0], a1 = args[1], a2 = args[2];
2495        if (const_args[2]) {
2496            tcg_out_andi32(s, a0, a1, a2);
2497        } else {
2498            tcg_out32(s, AND | SAB(a1, a0, a2));
2499        }
2500        break;
2501    case INDEX_op_and_i64:
2502        a0 = args[0], a1 = args[1], a2 = args[2];
2503        if (const_args[2]) {
2504            tcg_out_andi64(s, a0, a1, a2);
2505        } else {
2506            tcg_out32(s, AND | SAB(a1, a0, a2));
2507        }
2508        break;
2509    case INDEX_op_or_i64:
2510    case INDEX_op_or_i32:
2511        a0 = args[0], a1 = args[1], a2 = args[2];
2512        if (const_args[2]) {
2513            tcg_out_ori32(s, a0, a1, a2);
2514        } else {
2515            tcg_out32(s, OR | SAB(a1, a0, a2));
2516        }
2517        break;
2518    case INDEX_op_xor_i64:
2519    case INDEX_op_xor_i32:
2520        a0 = args[0], a1 = args[1], a2 = args[2];
2521        if (const_args[2]) {
2522            tcg_out_xori32(s, a0, a1, a2);
2523        } else {
2524            tcg_out32(s, XOR | SAB(a1, a0, a2));
2525        }
2526        break;
2527    case INDEX_op_andc_i32:
2528        a0 = args[0], a1 = args[1], a2 = args[2];
2529        if (const_args[2]) {
2530            tcg_out_andi32(s, a0, a1, ~a2);
2531        } else {
2532            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2533        }
2534        break;
2535    case INDEX_op_andc_i64:
2536        a0 = args[0], a1 = args[1], a2 = args[2];
2537        if (const_args[2]) {
2538            tcg_out_andi64(s, a0, a1, ~a2);
2539        } else {
2540            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2541        }
2542        break;
2543    case INDEX_op_orc_i32:
2544        if (const_args[2]) {
2545            tcg_out_ori32(s, args[0], args[1], ~args[2]);
2546            break;
2547        }
2548        /* FALLTHRU */
2549    case INDEX_op_orc_i64:
2550        tcg_out32(s, ORC | SAB(args[1], args[0], args[2]));
2551        break;
2552    case INDEX_op_eqv_i32:
2553        if (const_args[2]) {
2554            tcg_out_xori32(s, args[0], args[1], ~args[2]);
2555            break;
2556        }
2557        /* FALLTHRU */
2558    case INDEX_op_eqv_i64:
2559        tcg_out32(s, EQV | SAB(args[1], args[0], args[2]));
2560        break;
2561    case INDEX_op_nand_i32:
2562    case INDEX_op_nand_i64:
2563        tcg_out32(s, NAND | SAB(args[1], args[0], args[2]));
2564        break;
2565    case INDEX_op_nor_i32:
2566    case INDEX_op_nor_i64:
2567        tcg_out32(s, NOR | SAB(args[1], args[0], args[2]));
2568        break;
2569
2570    case INDEX_op_clz_i32:
2571        tcg_out_cntxz(s, TCG_TYPE_I32, CNTLZW, args[0], args[1],
2572                      args[2], const_args[2]);
2573        break;
2574    case INDEX_op_ctz_i32:
2575        tcg_out_cntxz(s, TCG_TYPE_I32, CNTTZW, args[0], args[1],
2576                      args[2], const_args[2]);
2577        break;
2578    case INDEX_op_ctpop_i32:
2579        tcg_out32(s, CNTPOPW | SAB(args[1], args[0], 0));
2580        break;
2581
2582    case INDEX_op_clz_i64:
2583        tcg_out_cntxz(s, TCG_TYPE_I64, CNTLZD, args[0], args[1],
2584                      args[2], const_args[2]);
2585        break;
2586    case INDEX_op_ctz_i64:
2587        tcg_out_cntxz(s, TCG_TYPE_I64, CNTTZD, args[0], args[1],
2588                      args[2], const_args[2]);
2589        break;
2590    case INDEX_op_ctpop_i64:
2591        tcg_out32(s, CNTPOPD | SAB(args[1], args[0], 0));
2592        break;
2593
2594    case INDEX_op_mul_i32:
2595        a0 = args[0], a1 = args[1], a2 = args[2];
2596        if (const_args[2]) {
2597            tcg_out32(s, MULLI | TAI(a0, a1, a2));
2598        } else {
2599            tcg_out32(s, MULLW | TAB(a0, a1, a2));
2600        }
2601        break;
2602
2603    case INDEX_op_div_i32:
2604        tcg_out32(s, DIVW | TAB(args[0], args[1], args[2]));
2605        break;
2606
2607    case INDEX_op_divu_i32:
2608        tcg_out32(s, DIVWU | TAB(args[0], args[1], args[2]));
2609        break;
2610
2611    case INDEX_op_shl_i32:
2612        if (const_args[2]) {
2613            /* Limit immediate shift count lest we create an illegal insn.  */
2614            tcg_out_shli32(s, args[0], args[1], args[2] & 31);
2615        } else {
2616            tcg_out32(s, SLW | SAB(args[1], args[0], args[2]));
2617        }
2618        break;
2619    case INDEX_op_shr_i32:
2620        if (const_args[2]) {
2621            /* Limit immediate shift count lest we create an illegal insn.  */
2622            tcg_out_shri32(s, args[0], args[1], args[2] & 31);
2623        } else {
2624            tcg_out32(s, SRW | SAB(args[1], args[0], args[2]));
2625        }
2626        break;
2627    case INDEX_op_sar_i32:
2628        if (const_args[2]) {
2629            /* Limit immediate shift count lest we create an illegal insn.  */
2630            tcg_out32(s, SRAWI | RS(args[1]) | RA(args[0]) | SH(args[2] & 31));
2631        } else {
2632            tcg_out32(s, SRAW | SAB(args[1], args[0], args[2]));
2633        }
2634        break;
2635    case INDEX_op_rotl_i32:
2636        if (const_args[2]) {
2637            tcg_out_rlw(s, RLWINM, args[0], args[1], args[2], 0, 31);
2638        } else {
2639            tcg_out32(s, RLWNM | SAB(args[1], args[0], args[2])
2640                         | MB(0) | ME(31));
2641        }
2642        break;
2643    case INDEX_op_rotr_i32:
2644        if (const_args[2]) {
2645            tcg_out_rlw(s, RLWINM, args[0], args[1], 32 - args[2], 0, 31);
2646        } else {
2647            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 32));
2648            tcg_out32(s, RLWNM | SAB(args[1], args[0], TCG_REG_R0)
2649                         | MB(0) | ME(31));
2650        }
2651        break;
2652
2653    case INDEX_op_brcond_i32:
2654        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2655                       arg_label(args[3]), TCG_TYPE_I32);
2656        break;
2657    case INDEX_op_brcond_i64:
2658        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2659                       arg_label(args[3]), TCG_TYPE_I64);
2660        break;
2661    case INDEX_op_brcond2_i32:
2662        tcg_out_brcond2(s, args, const_args);
2663        break;
2664
2665    case INDEX_op_neg_i32:
2666    case INDEX_op_neg_i64:
2667        tcg_out32(s, NEG | RT(args[0]) | RA(args[1]));
2668        break;
2669
2670    case INDEX_op_not_i32:
2671    case INDEX_op_not_i64:
2672        tcg_out32(s, NOR | SAB(args[1], args[0], args[1]));
2673        break;
2674
2675    case INDEX_op_add_i64:
2676        a0 = args[0], a1 = args[1], a2 = args[2];
2677        if (const_args[2]) {
2678        do_addi_64:
2679            tcg_out_mem_long(s, ADDI, ADD, a0, a1, a2);
2680        } else {
2681            tcg_out32(s, ADD | TAB(a0, a1, a2));
2682        }
2683        break;
2684    case INDEX_op_sub_i64:
2685        a0 = args[0], a1 = args[1], a2 = args[2];
2686        if (const_args[1]) {
2687            if (const_args[2]) {
2688                tcg_out_movi(s, TCG_TYPE_I64, a0, a1 - a2);
2689            } else {
2690                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2691            }
2692        } else if (const_args[2]) {
2693            a2 = -a2;
2694            goto do_addi_64;
2695        } else {
2696            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2697        }
2698        break;
2699
2700    case INDEX_op_shl_i64:
2701        if (const_args[2]) {
2702            /* Limit immediate shift count lest we create an illegal insn.  */
2703            tcg_out_shli64(s, args[0], args[1], args[2] & 63);
2704        } else {
2705            tcg_out32(s, SLD | SAB(args[1], args[0], args[2]));
2706        }
2707        break;
2708    case INDEX_op_shr_i64:
2709        if (const_args[2]) {
2710            /* Limit immediate shift count lest we create an illegal insn.  */
2711            tcg_out_shri64(s, args[0], args[1], args[2] & 63);
2712        } else {
2713            tcg_out32(s, SRD | SAB(args[1], args[0], args[2]));
2714        }
2715        break;
2716    case INDEX_op_sar_i64:
2717        if (const_args[2]) {
2718            int sh = SH(args[2] & 0x1f) | (((args[2] >> 5) & 1) << 1);
2719            tcg_out32(s, SRADI | RA(args[0]) | RS(args[1]) | sh);
2720        } else {
2721            tcg_out32(s, SRAD | SAB(args[1], args[0], args[2]));
2722        }
2723        break;
2724    case INDEX_op_rotl_i64:
2725        if (const_args[2]) {
2726            tcg_out_rld(s, RLDICL, args[0], args[1], args[2], 0);
2727        } else {
2728            tcg_out32(s, RLDCL | SAB(args[1], args[0], args[2]) | MB64(0));
2729        }
2730        break;
2731    case INDEX_op_rotr_i64:
2732        if (const_args[2]) {
2733            tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 0);
2734        } else {
2735            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 64));
2736            tcg_out32(s, RLDCL | SAB(args[1], args[0], TCG_REG_R0) | MB64(0));
2737        }
2738        break;
2739
2740    case INDEX_op_mul_i64:
2741        a0 = args[0], a1 = args[1], a2 = args[2];
2742        if (const_args[2]) {
2743            tcg_out32(s, MULLI | TAI(a0, a1, a2));
2744        } else {
2745            tcg_out32(s, MULLD | TAB(a0, a1, a2));
2746        }
2747        break;
2748    case INDEX_op_div_i64:
2749        tcg_out32(s, DIVD | TAB(args[0], args[1], args[2]));
2750        break;
2751    case INDEX_op_divu_i64:
2752        tcg_out32(s, DIVDU | TAB(args[0], args[1], args[2]));
2753        break;
2754
2755    case INDEX_op_qemu_ld_i32:
2756        tcg_out_qemu_ld(s, args, false);
2757        break;
2758    case INDEX_op_qemu_ld_i64:
2759        tcg_out_qemu_ld(s, args, true);
2760        break;
2761    case INDEX_op_qemu_st_i32:
2762        tcg_out_qemu_st(s, args, false);
2763        break;
2764    case INDEX_op_qemu_st_i64:
2765        tcg_out_qemu_st(s, args, true);
2766        break;
2767
2768    case INDEX_op_ext8s_i32:
2769    case INDEX_op_ext8s_i64:
2770        c = EXTSB;
2771        goto gen_ext;
2772    case INDEX_op_ext16s_i32:
2773    case INDEX_op_ext16s_i64:
2774        c = EXTSH;
2775        goto gen_ext;
2776    case INDEX_op_ext_i32_i64:
2777    case INDEX_op_ext32s_i64:
2778        c = EXTSW;
2779        goto gen_ext;
2780    gen_ext:
2781        tcg_out32(s, c | RS(args[1]) | RA(args[0]));
2782        break;
2783    case INDEX_op_extu_i32_i64:
2784        tcg_out_ext32u(s, args[0], args[1]);
2785        break;
2786
2787    case INDEX_op_setcond_i32:
2788        tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2],
2789                        const_args[2]);
2790        break;
2791    case INDEX_op_setcond_i64:
2792        tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2],
2793                        const_args[2]);
2794        break;
2795    case INDEX_op_setcond2_i32:
2796        tcg_out_setcond2(s, args, const_args);
2797        break;
2798
2799    case INDEX_op_bswap16_i32:
2800    case INDEX_op_bswap16_i64:
2801        a0 = args[0], a1 = args[1];
2802        /* a1 = abcd */
2803        if (a0 != a1) {
2804            /* a0 = (a1 r<< 24) & 0xff # 000c */
2805            tcg_out_rlw(s, RLWINM, a0, a1, 24, 24, 31);
2806            /* a0 = (a0 & ~0xff00) | (a1 r<< 8) & 0xff00 # 00dc */
2807            tcg_out_rlw(s, RLWIMI, a0, a1, 8, 16, 23);
2808        } else {
2809            /* r0 = (a1 r<< 8) & 0xff00 # 00d0 */
2810            tcg_out_rlw(s, RLWINM, TCG_REG_R0, a1, 8, 16, 23);
2811            /* a0 = (a1 r<< 24) & 0xff # 000c */
2812            tcg_out_rlw(s, RLWINM, a0, a1, 24, 24, 31);
2813            /* a0 = a0 | r0 # 00dc */
2814            tcg_out32(s, OR | SAB(TCG_REG_R0, a0, a0));
2815        }
2816        break;
2817
2818    case INDEX_op_bswap32_i32:
2819    case INDEX_op_bswap32_i64:
2820        /* Stolen from gcc's builtin_bswap32 */
2821        a1 = args[1];
2822        a0 = args[0] == a1 ? TCG_REG_R0 : args[0];
2823
2824        /* a1 = args[1] # abcd */
2825        /* a0 = rotate_left (a1, 8) # bcda */
2826        tcg_out_rlw(s, RLWINM, a0, a1, 8, 0, 31);
2827        /* a0 = (a0 & ~0xff000000) | ((a1 r<< 24) & 0xff000000) # dcda */
2828        tcg_out_rlw(s, RLWIMI, a0, a1, 24, 0, 7);
2829        /* a0 = (a0 & ~0x0000ff00) | ((a1 r<< 24) & 0x0000ff00) # dcba */
2830        tcg_out_rlw(s, RLWIMI, a0, a1, 24, 16, 23);
2831
2832        if (a0 == TCG_REG_R0) {
2833            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2834        }
2835        break;
2836
2837    case INDEX_op_bswap64_i64:
2838        a0 = args[0], a1 = args[1], a2 = TCG_REG_R0;
2839        if (a0 == a1) {
2840            a0 = TCG_REG_R0;
2841            a2 = a1;
2842        }
2843
2844        /* a1 = # abcd efgh */
2845        /* a0 = rl32(a1, 8) # 0000 fghe */
2846        tcg_out_rlw(s, RLWINM, a0, a1, 8, 0, 31);
2847        /* a0 = dep(a0, rl32(a1, 24), 0xff000000) # 0000 hghe */
2848        tcg_out_rlw(s, RLWIMI, a0, a1, 24, 0, 7);
2849        /* a0 = dep(a0, rl32(a1, 24), 0x0000ff00) # 0000 hgfe */
2850        tcg_out_rlw(s, RLWIMI, a0, a1, 24, 16, 23);
2851
2852        /* a0 = rl64(a0, 32) # hgfe 0000 */
2853        /* a2 = rl64(a1, 32) # efgh abcd */
2854        tcg_out_rld(s, RLDICL, a0, a0, 32, 0);
2855        tcg_out_rld(s, RLDICL, a2, a1, 32, 0);
2856
2857        /* a0 = dep(a0, rl32(a2, 8), 0xffffffff)  # hgfe bcda */
2858        tcg_out_rlw(s, RLWIMI, a0, a2, 8, 0, 31);
2859        /* a0 = dep(a0, rl32(a2, 24), 0xff000000) # hgfe dcda */
2860        tcg_out_rlw(s, RLWIMI, a0, a2, 24, 0, 7);
2861        /* a0 = dep(a0, rl32(a2, 24), 0x0000ff00) # hgfe dcba */
2862        tcg_out_rlw(s, RLWIMI, a0, a2, 24, 16, 23);
2863
2864        if (a0 == 0) {
2865            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2866        }
2867        break;
2868
2869    case INDEX_op_deposit_i32:
2870        if (const_args[2]) {
2871            uint32_t mask = ((2u << (args[4] - 1)) - 1) << args[3];
2872            tcg_out_andi32(s, args[0], args[0], ~mask);
2873        } else {
2874            tcg_out_rlw(s, RLWIMI, args[0], args[2], args[3],
2875                        32 - args[3] - args[4], 31 - args[3]);
2876        }
2877        break;
2878    case INDEX_op_deposit_i64:
2879        if (const_args[2]) {
2880            uint64_t mask = ((2ull << (args[4] - 1)) - 1) << args[3];
2881            tcg_out_andi64(s, args[0], args[0], ~mask);
2882        } else {
2883            tcg_out_rld(s, RLDIMI, args[0], args[2], args[3],
2884                        64 - args[3] - args[4]);
2885        }
2886        break;
2887
2888    case INDEX_op_extract_i32:
2889        tcg_out_rlw(s, RLWINM, args[0], args[1],
2890                    32 - args[2], 32 - args[3], 31);
2891        break;
2892    case INDEX_op_extract_i64:
2893        tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 64 - args[3]);
2894        break;
2895
2896    case INDEX_op_movcond_i32:
2897        tcg_out_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], args[2],
2898                        args[3], args[4], const_args[2]);
2899        break;
2900    case INDEX_op_movcond_i64:
2901        tcg_out_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], args[2],
2902                        args[3], args[4], const_args[2]);
2903        break;
2904
2905#if TCG_TARGET_REG_BITS == 64
2906    case INDEX_op_add2_i64:
2907#else
2908    case INDEX_op_add2_i32:
2909#endif
2910        /* Note that the CA bit is defined based on the word size of the
2911           environment.  So in 64-bit mode it's always carry-out of bit 63.
2912           The fallback code using deposit works just as well for 32-bit.  */
2913        a0 = args[0], a1 = args[1];
2914        if (a0 == args[3] || (!const_args[5] && a0 == args[5])) {
2915            a0 = TCG_REG_R0;
2916        }
2917        if (const_args[4]) {
2918            tcg_out32(s, ADDIC | TAI(a0, args[2], args[4]));
2919        } else {
2920            tcg_out32(s, ADDC | TAB(a0, args[2], args[4]));
2921        }
2922        if (const_args[5]) {
2923            tcg_out32(s, (args[5] ? ADDME : ADDZE) | RT(a1) | RA(args[3]));
2924        } else {
2925            tcg_out32(s, ADDE | TAB(a1, args[3], args[5]));
2926        }
2927        if (a0 != args[0]) {
2928            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2929        }
2930        break;
2931
2932#if TCG_TARGET_REG_BITS == 64
2933    case INDEX_op_sub2_i64:
2934#else
2935    case INDEX_op_sub2_i32:
2936#endif
2937        a0 = args[0], a1 = args[1];
2938        if (a0 == args[5] || (!const_args[3] && a0 == args[3])) {
2939            a0 = TCG_REG_R0;
2940        }
2941        if (const_args[2]) {
2942            tcg_out32(s, SUBFIC | TAI(a0, args[4], args[2]));
2943        } else {
2944            tcg_out32(s, SUBFC | TAB(a0, args[4], args[2]));
2945        }
2946        if (const_args[3]) {
2947            tcg_out32(s, (args[3] ? SUBFME : SUBFZE) | RT(a1) | RA(args[5]));
2948        } else {
2949            tcg_out32(s, SUBFE | TAB(a1, args[5], args[3]));
2950        }
2951        if (a0 != args[0]) {
2952            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2953        }
2954        break;
2955
2956    case INDEX_op_muluh_i32:
2957        tcg_out32(s, MULHWU | TAB(args[0], args[1], args[2]));
2958        break;
2959    case INDEX_op_mulsh_i32:
2960        tcg_out32(s, MULHW | TAB(args[0], args[1], args[2]));
2961        break;
2962    case INDEX_op_muluh_i64:
2963        tcg_out32(s, MULHDU | TAB(args[0], args[1], args[2]));
2964        break;
2965    case INDEX_op_mulsh_i64:
2966        tcg_out32(s, MULHD | TAB(args[0], args[1], args[2]));
2967        break;
2968
2969    case INDEX_op_mb:
2970        tcg_out_mb(s, args[0]);
2971        break;
2972
2973    case INDEX_op_mov_i32:   /* Always emitted via tcg_out_mov.  */
2974    case INDEX_op_mov_i64:
2975    case INDEX_op_movi_i32:  /* Always emitted via tcg_out_movi.  */
2976    case INDEX_op_movi_i64:
2977    case INDEX_op_call:      /* Always emitted via tcg_out_call.  */
2978    default:
2979        tcg_abort();
2980    }
2981}
2982
2983int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
2984{
2985    switch (opc) {
2986    case INDEX_op_and_vec:
2987    case INDEX_op_or_vec:
2988    case INDEX_op_xor_vec:
2989    case INDEX_op_andc_vec:
2990    case INDEX_op_not_vec:
2991        return 1;
2992    case INDEX_op_orc_vec:
2993        return have_isa_2_07;
2994    case INDEX_op_add_vec:
2995    case INDEX_op_sub_vec:
2996    case INDEX_op_smax_vec:
2997    case INDEX_op_smin_vec:
2998    case INDEX_op_umax_vec:
2999    case INDEX_op_umin_vec:
3000    case INDEX_op_shlv_vec:
3001    case INDEX_op_shrv_vec:
3002    case INDEX_op_sarv_vec:
3003    case INDEX_op_rotlv_vec:
3004        return vece <= MO_32 || have_isa_2_07;
3005    case INDEX_op_ssadd_vec:
3006    case INDEX_op_sssub_vec:
3007    case INDEX_op_usadd_vec:
3008    case INDEX_op_ussub_vec:
3009        return vece <= MO_32;
3010    case INDEX_op_cmp_vec:
3011    case INDEX_op_shli_vec:
3012    case INDEX_op_shri_vec:
3013    case INDEX_op_sari_vec:
3014    case INDEX_op_rotli_vec:
3015        return vece <= MO_32 || have_isa_2_07 ? -1 : 0;
3016    case INDEX_op_neg_vec:
3017        return vece >= MO_32 && have_isa_3_00;
3018    case INDEX_op_mul_vec:
3019        switch (vece) {
3020        case MO_8:
3021        case MO_16:
3022            return -1;
3023        case MO_32:
3024            return have_isa_2_07 ? 1 : -1;
3025        case MO_64:
3026            return have_isa_3_10;
3027        }
3028        return 0;
3029    case INDEX_op_bitsel_vec:
3030        return have_vsx;
3031    case INDEX_op_rotrv_vec:
3032        return -1;
3033    default:
3034        return 0;
3035    }
3036}
3037
3038static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
3039                            TCGReg dst, TCGReg src)
3040{
3041    tcg_debug_assert(dst >= TCG_REG_V0);
3042
3043    /* Splat from integer reg allowed via constraints for v3.00.  */
3044    if (src < TCG_REG_V0) {
3045        tcg_debug_assert(have_isa_3_00);
3046        switch (vece) {
3047        case MO_64:
3048            tcg_out32(s, MTVSRDD | VRT(dst) | RA(src) | RB(src));
3049            return true;
3050        case MO_32:
3051            tcg_out32(s, MTVSRWS | VRT(dst) | RA(src));
3052            return true;
3053        default:
3054            /* Fail, so that we fall back on either dupm or mov+dup.  */
3055            return false;
3056        }
3057    }
3058
3059    /*
3060     * Recall we use (or emulate) VSX integer loads, so the integer is
3061     * right justified within the left (zero-index) double-word.
3062     */
3063    switch (vece) {
3064    case MO_8:
3065        tcg_out32(s, VSPLTB | VRT(dst) | VRB(src) | (7 << 16));
3066        break;
3067    case MO_16:
3068        tcg_out32(s, VSPLTH | VRT(dst) | VRB(src) | (3 << 16));
3069        break;
3070    case MO_32:
3071        tcg_out32(s, VSPLTW | VRT(dst) | VRB(src) | (1 << 16));
3072        break;
3073    case MO_64:
3074        if (have_vsx) {
3075            tcg_out32(s, XXPERMDI | VRT(dst) | VRA(src) | VRB(src));
3076            break;
3077        }
3078        tcg_out_vsldoi(s, TCG_VEC_TMP1, src, src, 8);
3079        tcg_out_vsldoi(s, dst, TCG_VEC_TMP1, src, 8);
3080        break;
3081    default:
3082        g_assert_not_reached();
3083    }
3084    return true;
3085}
3086
3087static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
3088                             TCGReg out, TCGReg base, intptr_t offset)
3089{
3090    int elt;
3091
3092    tcg_debug_assert(out >= TCG_REG_V0);
3093    switch (vece) {
3094    case MO_8:
3095        if (have_isa_3_00) {
3096            tcg_out_mem_long(s, LXV, LVX, out, base, offset & -16);
3097        } else {
3098            tcg_out_mem_long(s, 0, LVEBX, out, base, offset);
3099        }
3100        elt = extract32(offset, 0, 4);
3101#ifndef HOST_WORDS_BIGENDIAN
3102        elt ^= 15;
3103#endif
3104        tcg_out32(s, VSPLTB | VRT(out) | VRB(out) | (elt << 16));
3105        break;
3106    case MO_16:
3107        tcg_debug_assert((offset & 1) == 0);
3108        if (have_isa_3_00) {
3109            tcg_out_mem_long(s, LXV | 8, LVX, out, base, offset & -16);
3110        } else {
3111            tcg_out_mem_long(s, 0, LVEHX, out, base, offset);
3112        }
3113        elt = extract32(offset, 1, 3);
3114#ifndef HOST_WORDS_BIGENDIAN
3115        elt ^= 7;
3116#endif
3117        tcg_out32(s, VSPLTH | VRT(out) | VRB(out) | (elt << 16));
3118        break;
3119    case MO_32:
3120        if (have_isa_3_00) {
3121            tcg_out_mem_long(s, 0, LXVWSX, out, base, offset);
3122            break;
3123        }
3124        tcg_debug_assert((offset & 3) == 0);
3125        tcg_out_mem_long(s, 0, LVEWX, out, base, offset);
3126        elt = extract32(offset, 2, 2);
3127#ifndef HOST_WORDS_BIGENDIAN
3128        elt ^= 3;
3129#endif
3130        tcg_out32(s, VSPLTW | VRT(out) | VRB(out) | (elt << 16));
3131        break;
3132    case MO_64:
3133        if (have_vsx) {
3134            tcg_out_mem_long(s, 0, LXVDSX, out, base, offset);
3135            break;
3136        }
3137        tcg_debug_assert((offset & 7) == 0);
3138        tcg_out_mem_long(s, 0, LVX, out, base, offset & -16);
3139        tcg_out_vsldoi(s, TCG_VEC_TMP1, out, out, 8);
3140        elt = extract32(offset, 3, 1);
3141#ifndef HOST_WORDS_BIGENDIAN
3142        elt = !elt;
3143#endif
3144        if (elt) {
3145            tcg_out_vsldoi(s, out, out, TCG_VEC_TMP1, 8);
3146        } else {
3147            tcg_out_vsldoi(s, out, TCG_VEC_TMP1, out, 8);
3148        }
3149        break;
3150    default:
3151        g_assert_not_reached();
3152    }
3153    return true;
3154}
3155
3156static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
3157                           unsigned vecl, unsigned vece,
3158                           const TCGArg *args, const int *const_args)
3159{
3160    static const uint32_t
3161        add_op[4] = { VADDUBM, VADDUHM, VADDUWM, VADDUDM },
3162        sub_op[4] = { VSUBUBM, VSUBUHM, VSUBUWM, VSUBUDM },
3163        mul_op[4] = { 0, 0, VMULUWM, VMULLD },
3164        neg_op[4] = { 0, 0, VNEGW, VNEGD },
3165        eq_op[4]  = { VCMPEQUB, VCMPEQUH, VCMPEQUW, VCMPEQUD },
3166        ne_op[4]  = { VCMPNEB, VCMPNEH, VCMPNEW, 0 },
3167        gts_op[4] = { VCMPGTSB, VCMPGTSH, VCMPGTSW, VCMPGTSD },
3168        gtu_op[4] = { VCMPGTUB, VCMPGTUH, VCMPGTUW, VCMPGTUD },
3169        ssadd_op[4] = { VADDSBS, VADDSHS, VADDSWS, 0 },
3170        usadd_op[4] = { VADDUBS, VADDUHS, VADDUWS, 0 },
3171        sssub_op[4] = { VSUBSBS, VSUBSHS, VSUBSWS, 0 },
3172        ussub_op[4] = { VSUBUBS, VSUBUHS, VSUBUWS, 0 },
3173        umin_op[4] = { VMINUB, VMINUH, VMINUW, VMINUD },
3174        smin_op[4] = { VMINSB, VMINSH, VMINSW, VMINSD },
3175        umax_op[4] = { VMAXUB, VMAXUH, VMAXUW, VMAXUD },
3176        smax_op[4] = { VMAXSB, VMAXSH, VMAXSW, VMAXSD },
3177        shlv_op[4] = { VSLB, VSLH, VSLW, VSLD },
3178        shrv_op[4] = { VSRB, VSRH, VSRW, VSRD },
3179        sarv_op[4] = { VSRAB, VSRAH, VSRAW, VSRAD },
3180        mrgh_op[4] = { VMRGHB, VMRGHH, VMRGHW, 0 },
3181        mrgl_op[4] = { VMRGLB, VMRGLH, VMRGLW, 0 },
3182        muleu_op[4] = { VMULEUB, VMULEUH, VMULEUW, 0 },
3183        mulou_op[4] = { VMULOUB, VMULOUH, VMULOUW, 0 },
3184        pkum_op[4] = { VPKUHUM, VPKUWUM, 0, 0 },
3185        rotl_op[4] = { VRLB, VRLH, VRLW, VRLD };
3186
3187    TCGType type = vecl + TCG_TYPE_V64;
3188    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
3189    uint32_t insn;
3190
3191    switch (opc) {
3192    case INDEX_op_ld_vec:
3193        tcg_out_ld(s, type, a0, a1, a2);
3194        return;
3195    case INDEX_op_st_vec:
3196        tcg_out_st(s, type, a0, a1, a2);
3197        return;
3198    case INDEX_op_dupm_vec:
3199        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
3200        return;
3201
3202    case INDEX_op_add_vec:
3203        insn = add_op[vece];
3204        break;
3205    case INDEX_op_sub_vec:
3206        insn = sub_op[vece];
3207        break;
3208    case INDEX_op_neg_vec:
3209        insn = neg_op[vece];
3210        a2 = a1;
3211        a1 = 0;
3212        break;
3213    case INDEX_op_mul_vec:
3214        insn = mul_op[vece];
3215        break;
3216    case INDEX_op_ssadd_vec:
3217        insn = ssadd_op[vece];
3218        break;
3219    case INDEX_op_sssub_vec:
3220        insn = sssub_op[vece];
3221        break;
3222    case INDEX_op_usadd_vec:
3223        insn = usadd_op[vece];
3224        break;
3225    case INDEX_op_ussub_vec:
3226        insn = ussub_op[vece];
3227        break;
3228    case INDEX_op_smin_vec:
3229        insn = smin_op[vece];
3230        break;
3231    case INDEX_op_umin_vec:
3232        insn = umin_op[vece];
3233        break;
3234    case INDEX_op_smax_vec:
3235        insn = smax_op[vece];
3236        break;
3237    case INDEX_op_umax_vec:
3238        insn = umax_op[vece];
3239        break;
3240    case INDEX_op_shlv_vec:
3241        insn = shlv_op[vece];
3242        break;
3243    case INDEX_op_shrv_vec:
3244        insn = shrv_op[vece];
3245        break;
3246    case INDEX_op_sarv_vec:
3247        insn = sarv_op[vece];
3248        break;
3249    case INDEX_op_and_vec:
3250        insn = VAND;
3251        break;
3252    case INDEX_op_or_vec:
3253        insn = VOR;
3254        break;
3255    case INDEX_op_xor_vec:
3256        insn = VXOR;
3257        break;
3258    case INDEX_op_andc_vec:
3259        insn = VANDC;
3260        break;
3261    case INDEX_op_not_vec:
3262        insn = VNOR;
3263        a2 = a1;
3264        break;
3265    case INDEX_op_orc_vec:
3266        insn = VORC;
3267        break;
3268
3269    case INDEX_op_cmp_vec:
3270        switch (args[3]) {
3271        case TCG_COND_EQ:
3272            insn = eq_op[vece];
3273            break;
3274        case TCG_COND_NE:
3275            insn = ne_op[vece];
3276            break;
3277        case TCG_COND_GT:
3278            insn = gts_op[vece];
3279            break;
3280        case TCG_COND_GTU:
3281            insn = gtu_op[vece];
3282            break;
3283        default:
3284            g_assert_not_reached();
3285        }
3286        break;
3287
3288    case INDEX_op_bitsel_vec:
3289        tcg_out32(s, XXSEL | VRT(a0) | VRC(a1) | VRB(a2) | VRA(args[3]));
3290        return;
3291
3292    case INDEX_op_dup2_vec:
3293        assert(TCG_TARGET_REG_BITS == 32);
3294        /* With inputs a1 = xLxx, a2 = xHxx  */
3295        tcg_out32(s, VMRGHW | VRT(a0) | VRA(a2) | VRB(a1));  /* a0  = xxHL */
3296        tcg_out_vsldoi(s, TCG_VEC_TMP1, a0, a0, 8);          /* tmp = HLxx */
3297        tcg_out_vsldoi(s, a0, a0, TCG_VEC_TMP1, 8);          /* a0  = HLHL */
3298        return;
3299
3300    case INDEX_op_ppc_mrgh_vec:
3301        insn = mrgh_op[vece];
3302        break;
3303    case INDEX_op_ppc_mrgl_vec:
3304        insn = mrgl_op[vece];
3305        break;
3306    case INDEX_op_ppc_muleu_vec:
3307        insn = muleu_op[vece];
3308        break;
3309    case INDEX_op_ppc_mulou_vec:
3310        insn = mulou_op[vece];
3311        break;
3312    case INDEX_op_ppc_pkum_vec:
3313        insn = pkum_op[vece];
3314        break;
3315    case INDEX_op_rotlv_vec:
3316        insn = rotl_op[vece];
3317        break;
3318    case INDEX_op_ppc_msum_vec:
3319        tcg_debug_assert(vece == MO_16);
3320        tcg_out32(s, VMSUMUHM | VRT(a0) | VRA(a1) | VRB(a2) | VRC(args[3]));
3321        return;
3322
3323    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
3324    case INDEX_op_dupi_vec: /* Always emitted via tcg_out_movi.  */
3325    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
3326    default:
3327        g_assert_not_reached();
3328    }
3329
3330    tcg_debug_assert(insn != 0);
3331    tcg_out32(s, insn | VRT(a0) | VRA(a1) | VRB(a2));
3332}
3333
3334static void expand_vec_shi(TCGType type, unsigned vece, TCGv_vec v0,
3335                           TCGv_vec v1, TCGArg imm, TCGOpcode opci)
3336{
3337    TCGv_vec t1 = tcg_temp_new_vec(type);
3338
3339    /* Splat w/bytes for xxspltib.  */
3340    tcg_gen_dupi_vec(MO_8, t1, imm & ((8 << vece) - 1));
3341    vec_gen_3(opci, type, vece, tcgv_vec_arg(v0),
3342              tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3343    tcg_temp_free_vec(t1);
3344}
3345
3346static void expand_vec_cmp(TCGType type, unsigned vece, TCGv_vec v0,
3347                           TCGv_vec v1, TCGv_vec v2, TCGCond cond)
3348{
3349    bool need_swap = false, need_inv = false;
3350
3351    tcg_debug_assert(vece <= MO_32 || have_isa_2_07);
3352
3353    switch (cond) {
3354    case TCG_COND_EQ:
3355    case TCG_COND_GT:
3356    case TCG_COND_GTU:
3357        break;
3358    case TCG_COND_NE:
3359        if (have_isa_3_00 && vece <= MO_32) {
3360            break;
3361        }
3362        /* fall through */
3363    case TCG_COND_LE:
3364    case TCG_COND_LEU:
3365        need_inv = true;
3366        break;
3367    case TCG_COND_LT:
3368    case TCG_COND_LTU:
3369        need_swap = true;
3370        break;
3371    case TCG_COND_GE:
3372    case TCG_COND_GEU:
3373        need_swap = need_inv = true;
3374        break;
3375    default:
3376        g_assert_not_reached();
3377    }
3378
3379    if (need_inv) {
3380        cond = tcg_invert_cond(cond);
3381    }
3382    if (need_swap) {
3383        TCGv_vec t1;
3384        t1 = v1, v1 = v2, v2 = t1;
3385        cond = tcg_swap_cond(cond);
3386    }
3387
3388    vec_gen_4(INDEX_op_cmp_vec, type, vece, tcgv_vec_arg(v0),
3389              tcgv_vec_arg(v1), tcgv_vec_arg(v2), cond);
3390
3391    if (need_inv) {
3392        tcg_gen_not_vec(vece, v0, v0);
3393    }
3394}
3395
3396static void expand_vec_mul(TCGType type, unsigned vece, TCGv_vec v0,
3397                           TCGv_vec v1, TCGv_vec v2)
3398{
3399    TCGv_vec t1 = tcg_temp_new_vec(type);
3400    TCGv_vec t2 = tcg_temp_new_vec(type);
3401    TCGv_vec t3, t4;
3402
3403    switch (vece) {
3404    case MO_8:
3405    case MO_16:
3406        vec_gen_3(INDEX_op_ppc_muleu_vec, type, vece, tcgv_vec_arg(t1),
3407                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3408        vec_gen_3(INDEX_op_ppc_mulou_vec, type, vece, tcgv_vec_arg(t2),
3409                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3410        vec_gen_3(INDEX_op_ppc_mrgh_vec, type, vece + 1, tcgv_vec_arg(v0),
3411                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3412        vec_gen_3(INDEX_op_ppc_mrgl_vec, type, vece + 1, tcgv_vec_arg(t1),
3413                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3414        vec_gen_3(INDEX_op_ppc_pkum_vec, type, vece, tcgv_vec_arg(v0),
3415                  tcgv_vec_arg(v0), tcgv_vec_arg(t1));
3416	break;
3417
3418    case MO_32:
3419        tcg_debug_assert(!have_isa_2_07);
3420        t3 = tcg_temp_new_vec(type);
3421        t4 = tcg_temp_new_vec(type);
3422        tcg_gen_dupi_vec(MO_8, t4, -16);
3423        vec_gen_3(INDEX_op_rotlv_vec, type, MO_32, tcgv_vec_arg(t1),
3424                  tcgv_vec_arg(v2), tcgv_vec_arg(t4));
3425        vec_gen_3(INDEX_op_ppc_mulou_vec, type, MO_16, tcgv_vec_arg(t2),
3426                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3427        tcg_gen_dupi_vec(MO_8, t3, 0);
3428        vec_gen_4(INDEX_op_ppc_msum_vec, type, MO_16, tcgv_vec_arg(t3),
3429                  tcgv_vec_arg(v1), tcgv_vec_arg(t1), tcgv_vec_arg(t3));
3430        vec_gen_3(INDEX_op_shlv_vec, type, MO_32, tcgv_vec_arg(t3),
3431                  tcgv_vec_arg(t3), tcgv_vec_arg(t4));
3432        tcg_gen_add_vec(MO_32, v0, t2, t3);
3433        tcg_temp_free_vec(t3);
3434        tcg_temp_free_vec(t4);
3435        break;
3436
3437    default:
3438        g_assert_not_reached();
3439    }
3440    tcg_temp_free_vec(t1);
3441    tcg_temp_free_vec(t2);
3442}
3443
3444void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3445                       TCGArg a0, ...)
3446{
3447    va_list va;
3448    TCGv_vec v0, v1, v2, t0;
3449    TCGArg a2;
3450
3451    va_start(va, a0);
3452    v0 = temp_tcgv_vec(arg_temp(a0));
3453    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3454    a2 = va_arg(va, TCGArg);
3455
3456    switch (opc) {
3457    case INDEX_op_shli_vec:
3458        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shlv_vec);
3459        break;
3460    case INDEX_op_shri_vec:
3461        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shrv_vec);
3462        break;
3463    case INDEX_op_sari_vec:
3464        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_sarv_vec);
3465        break;
3466    case INDEX_op_rotli_vec:
3467        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_rotlv_vec);
3468        break;
3469    case INDEX_op_cmp_vec:
3470        v2 = temp_tcgv_vec(arg_temp(a2));
3471        expand_vec_cmp(type, vece, v0, v1, v2, va_arg(va, TCGArg));
3472        break;
3473    case INDEX_op_mul_vec:
3474        v2 = temp_tcgv_vec(arg_temp(a2));
3475        expand_vec_mul(type, vece, v0, v1, v2);
3476        break;
3477    case INDEX_op_rotlv_vec:
3478        v2 = temp_tcgv_vec(arg_temp(a2));
3479        t0 = tcg_temp_new_vec(type);
3480        tcg_gen_neg_vec(vece, t0, v2);
3481        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3482        tcg_temp_free_vec(t0);
3483        break;
3484    default:
3485        g_assert_not_reached();
3486    }
3487    va_end(va);
3488}
3489
3490static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op)
3491{
3492    static const TCGTargetOpDef r = { .args_ct_str = { "r" } };
3493    static const TCGTargetOpDef r_r = { .args_ct_str = { "r", "r" } };
3494    static const TCGTargetOpDef r_L = { .args_ct_str = { "r", "L" } };
3495    static const TCGTargetOpDef S_S = { .args_ct_str = { "S", "S" } };
3496    static const TCGTargetOpDef r_ri = { .args_ct_str = { "r", "ri" } };
3497    static const TCGTargetOpDef r_r_r = { .args_ct_str = { "r", "r", "r" } };
3498    static const TCGTargetOpDef r_L_L = { .args_ct_str = { "r", "L", "L" } };
3499    static const TCGTargetOpDef L_L_L = { .args_ct_str = { "L", "L", "L" } };
3500    static const TCGTargetOpDef S_S_S = { .args_ct_str = { "S", "S", "S" } };
3501    static const TCGTargetOpDef r_r_ri = { .args_ct_str = { "r", "r", "ri" } };
3502    static const TCGTargetOpDef r_r_rI = { .args_ct_str = { "r", "r", "rI" } };
3503    static const TCGTargetOpDef r_r_rT = { .args_ct_str = { "r", "r", "rT" } };
3504    static const TCGTargetOpDef r_r_rU = { .args_ct_str = { "r", "r", "rU" } };
3505    static const TCGTargetOpDef r_rI_ri
3506        = { .args_ct_str = { "r", "rI", "ri" } };
3507    static const TCGTargetOpDef r_rI_rT
3508        = { .args_ct_str = { "r", "rI", "rT" } };
3509    static const TCGTargetOpDef r_r_rZW
3510        = { .args_ct_str = { "r", "r", "rZW" } };
3511    static const TCGTargetOpDef L_L_L_L
3512        = { .args_ct_str = { "L", "L", "L", "L" } };
3513    static const TCGTargetOpDef S_S_S_S
3514        = { .args_ct_str = { "S", "S", "S", "S" } };
3515    static const TCGTargetOpDef movc
3516        = { .args_ct_str = { "r", "r", "ri", "rZ", "rZ" } };
3517    static const TCGTargetOpDef dep
3518        = { .args_ct_str = { "r", "0", "rZ" } };
3519    static const TCGTargetOpDef br2
3520        = { .args_ct_str = { "r", "r", "ri", "ri" } };
3521    static const TCGTargetOpDef setc2
3522        = { .args_ct_str = { "r", "r", "r", "ri", "ri" } };
3523    static const TCGTargetOpDef add2
3524        = { .args_ct_str = { "r", "r", "r", "r", "rI", "rZM" } };
3525    static const TCGTargetOpDef sub2
3526        = { .args_ct_str = { "r", "r", "rI", "rZM", "r", "r" } };
3527    static const TCGTargetOpDef v_r = { .args_ct_str = { "v", "r" } };
3528    static const TCGTargetOpDef v_vr = { .args_ct_str = { "v", "vr" } };
3529    static const TCGTargetOpDef v_v = { .args_ct_str = { "v", "v" } };
3530    static const TCGTargetOpDef v_v_v = { .args_ct_str = { "v", "v", "v" } };
3531    static const TCGTargetOpDef v_v_v_v
3532        = { .args_ct_str = { "v", "v", "v", "v" } };
3533
3534    switch (op) {
3535    case INDEX_op_goto_ptr:
3536        return &r;
3537
3538    case INDEX_op_ld8u_i32:
3539    case INDEX_op_ld8s_i32:
3540    case INDEX_op_ld16u_i32:
3541    case INDEX_op_ld16s_i32:
3542    case INDEX_op_ld_i32:
3543    case INDEX_op_st8_i32:
3544    case INDEX_op_st16_i32:
3545    case INDEX_op_st_i32:
3546    case INDEX_op_ctpop_i32:
3547    case INDEX_op_neg_i32:
3548    case INDEX_op_not_i32:
3549    case INDEX_op_ext8s_i32:
3550    case INDEX_op_ext16s_i32:
3551    case INDEX_op_bswap16_i32:
3552    case INDEX_op_bswap32_i32:
3553    case INDEX_op_extract_i32:
3554    case INDEX_op_ld8u_i64:
3555    case INDEX_op_ld8s_i64:
3556    case INDEX_op_ld16u_i64:
3557    case INDEX_op_ld16s_i64:
3558    case INDEX_op_ld32u_i64:
3559    case INDEX_op_ld32s_i64:
3560    case INDEX_op_ld_i64:
3561    case INDEX_op_st8_i64:
3562    case INDEX_op_st16_i64:
3563    case INDEX_op_st32_i64:
3564    case INDEX_op_st_i64:
3565    case INDEX_op_ctpop_i64:
3566    case INDEX_op_neg_i64:
3567    case INDEX_op_not_i64:
3568    case INDEX_op_ext8s_i64:
3569    case INDEX_op_ext16s_i64:
3570    case INDEX_op_ext32s_i64:
3571    case INDEX_op_ext_i32_i64:
3572    case INDEX_op_extu_i32_i64:
3573    case INDEX_op_bswap16_i64:
3574    case INDEX_op_bswap32_i64:
3575    case INDEX_op_bswap64_i64:
3576    case INDEX_op_extract_i64:
3577        return &r_r;
3578
3579    case INDEX_op_add_i32:
3580    case INDEX_op_and_i32:
3581    case INDEX_op_or_i32:
3582    case INDEX_op_xor_i32:
3583    case INDEX_op_andc_i32:
3584    case INDEX_op_orc_i32:
3585    case INDEX_op_eqv_i32:
3586    case INDEX_op_shl_i32:
3587    case INDEX_op_shr_i32:
3588    case INDEX_op_sar_i32:
3589    case INDEX_op_rotl_i32:
3590    case INDEX_op_rotr_i32:
3591    case INDEX_op_setcond_i32:
3592    case INDEX_op_and_i64:
3593    case INDEX_op_andc_i64:
3594    case INDEX_op_shl_i64:
3595    case INDEX_op_shr_i64:
3596    case INDEX_op_sar_i64:
3597    case INDEX_op_rotl_i64:
3598    case INDEX_op_rotr_i64:
3599    case INDEX_op_setcond_i64:
3600        return &r_r_ri;
3601    case INDEX_op_mul_i32:
3602    case INDEX_op_mul_i64:
3603        return &r_r_rI;
3604    case INDEX_op_div_i32:
3605    case INDEX_op_divu_i32:
3606    case INDEX_op_nand_i32:
3607    case INDEX_op_nor_i32:
3608    case INDEX_op_muluh_i32:
3609    case INDEX_op_mulsh_i32:
3610    case INDEX_op_orc_i64:
3611    case INDEX_op_eqv_i64:
3612    case INDEX_op_nand_i64:
3613    case INDEX_op_nor_i64:
3614    case INDEX_op_div_i64:
3615    case INDEX_op_divu_i64:
3616    case INDEX_op_mulsh_i64:
3617    case INDEX_op_muluh_i64:
3618        return &r_r_r;
3619    case INDEX_op_sub_i32:
3620        return &r_rI_ri;
3621    case INDEX_op_add_i64:
3622        return &r_r_rT;
3623    case INDEX_op_or_i64:
3624    case INDEX_op_xor_i64:
3625        return &r_r_rU;
3626    case INDEX_op_sub_i64:
3627        return &r_rI_rT;
3628    case INDEX_op_clz_i32:
3629    case INDEX_op_ctz_i32:
3630    case INDEX_op_clz_i64:
3631    case INDEX_op_ctz_i64:
3632        return &r_r_rZW;
3633
3634    case INDEX_op_brcond_i32:
3635    case INDEX_op_brcond_i64:
3636        return &r_ri;
3637
3638    case INDEX_op_movcond_i32:
3639    case INDEX_op_movcond_i64:
3640        return &movc;
3641    case INDEX_op_deposit_i32:
3642    case INDEX_op_deposit_i64:
3643        return &dep;
3644    case INDEX_op_brcond2_i32:
3645        return &br2;
3646    case INDEX_op_setcond2_i32:
3647        return &setc2;
3648    case INDEX_op_add2_i64:
3649    case INDEX_op_add2_i32:
3650        return &add2;
3651    case INDEX_op_sub2_i64:
3652    case INDEX_op_sub2_i32:
3653        return &sub2;
3654
3655    case INDEX_op_qemu_ld_i32:
3656        return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
3657                ? &r_L : &r_L_L);
3658    case INDEX_op_qemu_st_i32:
3659        return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
3660                ? &S_S : &S_S_S);
3661    case INDEX_op_qemu_ld_i64:
3662        return (TCG_TARGET_REG_BITS == 64 ? &r_L
3663                : TARGET_LONG_BITS == 32 ? &L_L_L : &L_L_L_L);
3664    case INDEX_op_qemu_st_i64:
3665        return (TCG_TARGET_REG_BITS == 64 ? &S_S
3666                : TARGET_LONG_BITS == 32 ? &S_S_S : &S_S_S_S);
3667
3668    case INDEX_op_add_vec:
3669    case INDEX_op_sub_vec:
3670    case INDEX_op_mul_vec:
3671    case INDEX_op_and_vec:
3672    case INDEX_op_or_vec:
3673    case INDEX_op_xor_vec:
3674    case INDEX_op_andc_vec:
3675    case INDEX_op_orc_vec:
3676    case INDEX_op_cmp_vec:
3677    case INDEX_op_ssadd_vec:
3678    case INDEX_op_sssub_vec:
3679    case INDEX_op_usadd_vec:
3680    case INDEX_op_ussub_vec:
3681    case INDEX_op_smax_vec:
3682    case INDEX_op_smin_vec:
3683    case INDEX_op_umax_vec:
3684    case INDEX_op_umin_vec:
3685    case INDEX_op_shlv_vec:
3686    case INDEX_op_shrv_vec:
3687    case INDEX_op_sarv_vec:
3688    case INDEX_op_rotlv_vec:
3689    case INDEX_op_rotrv_vec:
3690    case INDEX_op_ppc_mrgh_vec:
3691    case INDEX_op_ppc_mrgl_vec:
3692    case INDEX_op_ppc_muleu_vec:
3693    case INDEX_op_ppc_mulou_vec:
3694    case INDEX_op_ppc_pkum_vec:
3695    case INDEX_op_dup2_vec:
3696        return &v_v_v;
3697    case INDEX_op_not_vec:
3698    case INDEX_op_neg_vec:
3699        return &v_v;
3700    case INDEX_op_dup_vec:
3701        return have_isa_3_00 ? &v_vr : &v_v;
3702    case INDEX_op_ld_vec:
3703    case INDEX_op_st_vec:
3704    case INDEX_op_dupm_vec:
3705        return &v_r;
3706    case INDEX_op_bitsel_vec:
3707    case INDEX_op_ppc_msum_vec:
3708        return &v_v_v_v;
3709
3710    default:
3711        return NULL;
3712    }
3713}
3714
3715static void tcg_target_init(TCGContext *s)
3716{
3717    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3718    unsigned long hwcap2 = qemu_getauxval(AT_HWCAP2);
3719
3720    have_isa = tcg_isa_base;
3721    if (hwcap & PPC_FEATURE_ARCH_2_06) {
3722        have_isa = tcg_isa_2_06;
3723    }
3724#ifdef PPC_FEATURE2_ARCH_2_07
3725    if (hwcap2 & PPC_FEATURE2_ARCH_2_07) {
3726        have_isa = tcg_isa_2_07;
3727    }
3728#endif
3729#ifdef PPC_FEATURE2_ARCH_3_00
3730    if (hwcap2 & PPC_FEATURE2_ARCH_3_00) {
3731        have_isa = tcg_isa_3_00;
3732    }
3733#endif
3734#ifdef PPC_FEATURE2_ARCH_3_10
3735    if (hwcap2 & PPC_FEATURE2_ARCH_3_10) {
3736        have_isa = tcg_isa_3_10;
3737    }
3738#endif
3739
3740#ifdef PPC_FEATURE2_HAS_ISEL
3741    /* Prefer explicit instruction from the kernel. */
3742    have_isel = (hwcap2 & PPC_FEATURE2_HAS_ISEL) != 0;
3743#else
3744    /* Fall back to knowing Power7 (2.06) has ISEL. */
3745    have_isel = have_isa_2_06;
3746#endif
3747
3748    if (hwcap & PPC_FEATURE_HAS_ALTIVEC) {
3749        have_altivec = true;
3750        /* We only care about the portion of VSX that overlaps Altivec. */
3751        if (hwcap & PPC_FEATURE_HAS_VSX) {
3752            have_vsx = true;
3753        }
3754    }
3755
3756    tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
3757    tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
3758    if (have_altivec) {
3759        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3760        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3761    }
3762
3763    tcg_target_call_clobber_regs = 0;
3764    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3765    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3766    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3767    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3768    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3769    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3770    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R7);
3771    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R8);
3772    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R9);
3773    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R10);
3774    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R11);
3775    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
3776
3777    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3778    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3779    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3780    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3781    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3782    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3783    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3784    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3785    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V8);
3786    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V9);
3787    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V10);
3788    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V11);
3789    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V12);
3790    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V13);
3791    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V14);
3792    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V15);
3793    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3794    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3795    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3796    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3797
3798    s->reserved_regs = 0;
3799    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); /* tcg temp */
3800    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R1); /* stack pointer */
3801#if defined(_CALL_SYSV)
3802    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R2); /* toc pointer */
3803#endif
3804#if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
3805    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R13); /* thread pointer */
3806#endif
3807    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1); /* mem temp */
3808    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP1);
3809    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP2);
3810    if (USE_REG_TB) {
3811        tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);  /* tb->tc_ptr */
3812    }
3813}
3814
3815#ifdef __ELF__
3816typedef struct {
3817    DebugFrameCIE cie;
3818    DebugFrameFDEHeader fde;
3819    uint8_t fde_def_cfa[4];
3820    uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2 + 3];
3821} DebugFrame;
3822
3823/* We're expecting a 2 byte uleb128 encoded value.  */
3824QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3825
3826#if TCG_TARGET_REG_BITS == 64
3827# define ELF_HOST_MACHINE EM_PPC64
3828#else
3829# define ELF_HOST_MACHINE EM_PPC
3830#endif
3831
3832static DebugFrame debug_frame = {
3833    .cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3834    .cie.id = -1,
3835    .cie.version = 1,
3836    .cie.code_align = 1,
3837    .cie.data_align = (-SZR & 0x7f),         /* sleb128 -SZR */
3838    .cie.return_column = 65,
3839
3840    /* Total FDE size does not include the "len" member.  */
3841    .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset),
3842
3843    .fde_def_cfa = {
3844        12, TCG_REG_R1,                 /* DW_CFA_def_cfa r1, ... */
3845        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3846        (FRAME_SIZE >> 7)
3847    },
3848    .fde_reg_ofs = {
3849        /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
3850        0x11, 65, (LR_OFFSET / -SZR) & 0x7f,
3851    }
3852};
3853
3854void tcg_register_jit(const void *buf, size_t buf_size)
3855{
3856    uint8_t *p = &debug_frame.fde_reg_ofs[3];
3857    int i;
3858
3859    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i, p += 2) {
3860        p[0] = 0x80 + tcg_target_callee_save_regs[i];
3861        p[1] = (FRAME_SIZE - (REG_SAVE_BOT + i * SZR)) / SZR;
3862    }
3863
3864    debug_frame.fde.func_start = (uintptr_t)buf;
3865    debug_frame.fde.func_len = buf_size;
3866
3867    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3868}
3869#endif /* __ELF__ */
3870