xref: /openbmc/qemu/tcg/ppc/tcg-target.c.inc (revision 39920a04)
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#include "../tcg-ldst.c.inc"
28
29/*
30 * Standardize on the _CALL_FOO symbols used by GCC:
31 * Apple XCode does not define _CALL_DARWIN.
32 * Clang defines _CALL_ELF (64-bit) but not _CALL_SYSV (32-bit).
33 */
34#if !defined(_CALL_SYSV) && \
35    !defined(_CALL_DARWIN) && \
36    !defined(_CALL_AIX) && \
37    !defined(_CALL_ELF)
38# if defined(__APPLE__)
39#  define _CALL_DARWIN
40# elif defined(__ELF__) && TCG_TARGET_REG_BITS == 32
41#  define _CALL_SYSV
42# else
43#  error "Unknown ABI"
44# endif
45#endif
46
47#if TCG_TARGET_REG_BITS == 64
48# define TCG_TARGET_CALL_ARG_I32   TCG_CALL_ARG_EXTEND
49# define TCG_TARGET_CALL_RET_I128  TCG_CALL_RET_NORMAL
50#else
51# define TCG_TARGET_CALL_ARG_I32   TCG_CALL_ARG_NORMAL
52# define TCG_TARGET_CALL_RET_I128  TCG_CALL_RET_BY_REF
53#endif
54#ifdef _CALL_SYSV
55# define TCG_TARGET_CALL_ARG_I64   TCG_CALL_ARG_EVEN
56# define TCG_TARGET_CALL_ARG_I128  TCG_CALL_ARG_BY_REF
57#else
58# define TCG_TARGET_CALL_ARG_I64   TCG_CALL_ARG_NORMAL
59# define TCG_TARGET_CALL_ARG_I128  TCG_CALL_ARG_NORMAL
60#endif
61
62/* For some memory operations, we need a scratch that isn't R0.  For the AIX
63   calling convention, we can re-use the TOC register since we'll be reloading
64   it at every call.  Otherwise R12 will do nicely as neither a call-saved
65   register nor a parameter register.  */
66#ifdef _CALL_AIX
67# define TCG_REG_TMP1   TCG_REG_R2
68#else
69# define TCG_REG_TMP1   TCG_REG_R12
70#endif
71
72#define TCG_VEC_TMP1    TCG_REG_V0
73#define TCG_VEC_TMP2    TCG_REG_V1
74
75#define TCG_REG_TB     TCG_REG_R31
76#define USE_REG_TB     (TCG_TARGET_REG_BITS == 64)
77
78/* Shorthand for size of a pointer.  Avoid promotion to unsigned.  */
79#define SZP  ((int)sizeof(void *))
80
81/* Shorthand for size of a register.  */
82#define SZR  (TCG_TARGET_REG_BITS / 8)
83
84#define TCG_CT_CONST_S16  0x100
85#define TCG_CT_CONST_U16  0x200
86#define TCG_CT_CONST_S32  0x400
87#define TCG_CT_CONST_U32  0x800
88#define TCG_CT_CONST_ZERO 0x1000
89#define TCG_CT_CONST_MONE 0x2000
90#define TCG_CT_CONST_WSZ  0x4000
91
92#define ALL_GENERAL_REGS  0xffffffffu
93#define ALL_VECTOR_REGS   0xffffffff00000000ull
94
95#ifdef CONFIG_SOFTMMU
96#define ALL_QLOAD_REGS \
97    (ALL_GENERAL_REGS & \
98     ~((1 << TCG_REG_R3) | (1 << TCG_REG_R4) | (1 << TCG_REG_R5)))
99#define ALL_QSTORE_REGS \
100    (ALL_GENERAL_REGS & ~((1 << TCG_REG_R3) | (1 << TCG_REG_R4) | \
101                          (1 << TCG_REG_R5) | (1 << TCG_REG_R6)))
102#else
103#define ALL_QLOAD_REGS  (ALL_GENERAL_REGS & ~(1 << TCG_REG_R3))
104#define ALL_QSTORE_REGS ALL_QLOAD_REGS
105#endif
106
107TCGPowerISA have_isa;
108static bool have_isel;
109bool have_altivec;
110bool have_vsx;
111
112#ifndef CONFIG_SOFTMMU
113#define TCG_GUEST_BASE_REG 30
114#endif
115
116#ifdef CONFIG_DEBUG_TCG
117static const char tcg_target_reg_names[TCG_TARGET_NB_REGS][4] = {
118    "r0",  "r1",  "r2",  "r3",  "r4",  "r5",  "r6",  "r7",
119    "r8",  "r9",  "r10", "r11", "r12", "r13", "r14", "r15",
120    "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
121    "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31",
122    "v0",  "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7",
123    "v8",  "v9",  "v10", "v11", "v12", "v13", "v14", "v15",
124    "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
125    "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",
126};
127#endif
128
129static const int tcg_target_reg_alloc_order[] = {
130    TCG_REG_R14,  /* call saved registers */
131    TCG_REG_R15,
132    TCG_REG_R16,
133    TCG_REG_R17,
134    TCG_REG_R18,
135    TCG_REG_R19,
136    TCG_REG_R20,
137    TCG_REG_R21,
138    TCG_REG_R22,
139    TCG_REG_R23,
140    TCG_REG_R24,
141    TCG_REG_R25,
142    TCG_REG_R26,
143    TCG_REG_R27,
144    TCG_REG_R28,
145    TCG_REG_R29,
146    TCG_REG_R30,
147    TCG_REG_R31,
148    TCG_REG_R12,  /* call clobbered, non-arguments */
149    TCG_REG_R11,
150    TCG_REG_R2,
151    TCG_REG_R13,
152    TCG_REG_R10,  /* call clobbered, arguments */
153    TCG_REG_R9,
154    TCG_REG_R8,
155    TCG_REG_R7,
156    TCG_REG_R6,
157    TCG_REG_R5,
158    TCG_REG_R4,
159    TCG_REG_R3,
160
161    /* V0 and V1 reserved as temporaries; V20 - V31 are call-saved */
162    TCG_REG_V2,   /* call clobbered, vectors */
163    TCG_REG_V3,
164    TCG_REG_V4,
165    TCG_REG_V5,
166    TCG_REG_V6,
167    TCG_REG_V7,
168    TCG_REG_V8,
169    TCG_REG_V9,
170    TCG_REG_V10,
171    TCG_REG_V11,
172    TCG_REG_V12,
173    TCG_REG_V13,
174    TCG_REG_V14,
175    TCG_REG_V15,
176    TCG_REG_V16,
177    TCG_REG_V17,
178    TCG_REG_V18,
179    TCG_REG_V19,
180};
181
182static const int tcg_target_call_iarg_regs[] = {
183    TCG_REG_R3,
184    TCG_REG_R4,
185    TCG_REG_R5,
186    TCG_REG_R6,
187    TCG_REG_R7,
188    TCG_REG_R8,
189    TCG_REG_R9,
190    TCG_REG_R10
191};
192
193static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
194{
195    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
196    tcg_debug_assert(slot >= 0 && slot <= 1);
197    return TCG_REG_R3 + slot;
198}
199
200static const int tcg_target_callee_save_regs[] = {
201#ifdef _CALL_DARWIN
202    TCG_REG_R11,
203#endif
204    TCG_REG_R14,
205    TCG_REG_R15,
206    TCG_REG_R16,
207    TCG_REG_R17,
208    TCG_REG_R18,
209    TCG_REG_R19,
210    TCG_REG_R20,
211    TCG_REG_R21,
212    TCG_REG_R22,
213    TCG_REG_R23,
214    TCG_REG_R24,
215    TCG_REG_R25,
216    TCG_REG_R26,
217    TCG_REG_R27, /* currently used for the global env */
218    TCG_REG_R28,
219    TCG_REG_R29,
220    TCG_REG_R30,
221    TCG_REG_R31
222};
223
224static inline bool in_range_b(tcg_target_long target)
225{
226    return target == sextract64(target, 0, 26);
227}
228
229static uint32_t reloc_pc24_val(const tcg_insn_unit *pc,
230			       const tcg_insn_unit *target)
231{
232    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
233    tcg_debug_assert(in_range_b(disp));
234    return disp & 0x3fffffc;
235}
236
237static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
238{
239    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
240    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
241
242    if (in_range_b(disp)) {
243        *src_rw = (*src_rw & ~0x3fffffc) | (disp & 0x3fffffc);
244        return true;
245    }
246    return false;
247}
248
249static uint16_t reloc_pc14_val(const tcg_insn_unit *pc,
250			       const tcg_insn_unit *target)
251{
252    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
253    tcg_debug_assert(disp == (int16_t) disp);
254    return disp & 0xfffc;
255}
256
257static bool reloc_pc14(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
258{
259    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
260    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
261
262    if (disp == (int16_t) disp) {
263        *src_rw = (*src_rw & ~0xfffc) | (disp & 0xfffc);
264        return true;
265    }
266    return false;
267}
268
269/* test if a constant matches the constraint */
270static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
271{
272    if (ct & TCG_CT_CONST) {
273        return 1;
274    }
275
276    /* The only 32-bit constraint we use aside from
277       TCG_CT_CONST is TCG_CT_CONST_S16.  */
278    if (type == TCG_TYPE_I32) {
279        val = (int32_t)val;
280    }
281
282    if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
283        return 1;
284    } else if ((ct & TCG_CT_CONST_U16) && val == (uint16_t)val) {
285        return 1;
286    } else if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
287        return 1;
288    } else if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
289        return 1;
290    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
291        return 1;
292    } else if ((ct & TCG_CT_CONST_MONE) && val == -1) {
293        return 1;
294    } else if ((ct & TCG_CT_CONST_WSZ)
295               && val == (type == TCG_TYPE_I32 ? 32 : 64)) {
296        return 1;
297    }
298    return 0;
299}
300
301#define OPCD(opc) ((opc)<<26)
302#define XO19(opc) (OPCD(19)|((opc)<<1))
303#define MD30(opc) (OPCD(30)|((opc)<<2))
304#define MDS30(opc) (OPCD(30)|((opc)<<1))
305#define XO31(opc) (OPCD(31)|((opc)<<1))
306#define XO58(opc) (OPCD(58)|(opc))
307#define XO62(opc) (OPCD(62)|(opc))
308#define VX4(opc)  (OPCD(4)|(opc))
309
310#define B      OPCD( 18)
311#define BC     OPCD( 16)
312#define LBZ    OPCD( 34)
313#define LHZ    OPCD( 40)
314#define LHA    OPCD( 42)
315#define LWZ    OPCD( 32)
316#define LWZUX  XO31( 55)
317#define STB    OPCD( 38)
318#define STH    OPCD( 44)
319#define STW    OPCD( 36)
320
321#define STD    XO62(  0)
322#define STDU   XO62(  1)
323#define STDX   XO31(149)
324
325#define LD     XO58(  0)
326#define LDX    XO31( 21)
327#define LDU    XO58(  1)
328#define LDUX   XO31( 53)
329#define LWA    XO58(  2)
330#define LWAX   XO31(341)
331
332#define ADDIC  OPCD( 12)
333#define ADDI   OPCD( 14)
334#define ADDIS  OPCD( 15)
335#define ORI    OPCD( 24)
336#define ORIS   OPCD( 25)
337#define XORI   OPCD( 26)
338#define XORIS  OPCD( 27)
339#define ANDI   OPCD( 28)
340#define ANDIS  OPCD( 29)
341#define MULLI  OPCD(  7)
342#define CMPLI  OPCD( 10)
343#define CMPI   OPCD( 11)
344#define SUBFIC OPCD( 8)
345
346#define LWZU   OPCD( 33)
347#define STWU   OPCD( 37)
348
349#define RLWIMI OPCD( 20)
350#define RLWINM OPCD( 21)
351#define RLWNM  OPCD( 23)
352
353#define RLDICL MD30(  0)
354#define RLDICR MD30(  1)
355#define RLDIMI MD30(  3)
356#define RLDCL  MDS30( 8)
357
358#define BCLR   XO19( 16)
359#define BCCTR  XO19(528)
360#define CRAND  XO19(257)
361#define CRANDC XO19(129)
362#define CRNAND XO19(225)
363#define CROR   XO19(449)
364#define CRNOR  XO19( 33)
365
366#define EXTSB  XO31(954)
367#define EXTSH  XO31(922)
368#define EXTSW  XO31(986)
369#define ADD    XO31(266)
370#define ADDE   XO31(138)
371#define ADDME  XO31(234)
372#define ADDZE  XO31(202)
373#define ADDC   XO31( 10)
374#define AND    XO31( 28)
375#define SUBF   XO31( 40)
376#define SUBFC  XO31(  8)
377#define SUBFE  XO31(136)
378#define SUBFME XO31(232)
379#define SUBFZE XO31(200)
380#define OR     XO31(444)
381#define XOR    XO31(316)
382#define MULLW  XO31(235)
383#define MULHW  XO31( 75)
384#define MULHWU XO31( 11)
385#define DIVW   XO31(491)
386#define DIVWU  XO31(459)
387#define MODSW  XO31(779)
388#define MODUW  XO31(267)
389#define CMP    XO31(  0)
390#define CMPL   XO31( 32)
391#define LHBRX  XO31(790)
392#define LWBRX  XO31(534)
393#define LDBRX  XO31(532)
394#define STHBRX XO31(918)
395#define STWBRX XO31(662)
396#define STDBRX XO31(660)
397#define MFSPR  XO31(339)
398#define MTSPR  XO31(467)
399#define SRAWI  XO31(824)
400#define NEG    XO31(104)
401#define MFCR   XO31( 19)
402#define MFOCRF (MFCR | (1u << 20))
403#define NOR    XO31(124)
404#define CNTLZW XO31( 26)
405#define CNTLZD XO31( 58)
406#define CNTTZW XO31(538)
407#define CNTTZD XO31(570)
408#define CNTPOPW XO31(378)
409#define CNTPOPD XO31(506)
410#define ANDC   XO31( 60)
411#define ORC    XO31(412)
412#define EQV    XO31(284)
413#define NAND   XO31(476)
414#define ISEL   XO31( 15)
415
416#define MULLD  XO31(233)
417#define MULHD  XO31( 73)
418#define MULHDU XO31(  9)
419#define DIVD   XO31(489)
420#define DIVDU  XO31(457)
421#define MODSD  XO31(777)
422#define MODUD  XO31(265)
423
424#define LBZX   XO31( 87)
425#define LHZX   XO31(279)
426#define LHAX   XO31(343)
427#define LWZX   XO31( 23)
428#define STBX   XO31(215)
429#define STHX   XO31(407)
430#define STWX   XO31(151)
431
432#define EIEIO  XO31(854)
433#define HWSYNC XO31(598)
434#define LWSYNC (HWSYNC | (1u << 21))
435
436#define SPR(a, b) ((((a)<<5)|(b))<<11)
437#define LR     SPR(8, 0)
438#define CTR    SPR(9, 0)
439
440#define SLW    XO31( 24)
441#define SRW    XO31(536)
442#define SRAW   XO31(792)
443
444#define SLD    XO31( 27)
445#define SRD    XO31(539)
446#define SRAD   XO31(794)
447#define SRADI  XO31(413<<1)
448
449#define BRH    XO31(219)
450#define BRW    XO31(155)
451#define BRD    XO31(187)
452
453#define TW     XO31( 4)
454#define TRAP   (TW | TO(31))
455
456#define NOP    ORI  /* ori 0,0,0 */
457
458#define LVX        XO31(103)
459#define LVEBX      XO31(7)
460#define LVEHX      XO31(39)
461#define LVEWX      XO31(71)
462#define LXSDX      (XO31(588) | 1)  /* v2.06, force tx=1 */
463#define LXVDSX     (XO31(332) | 1)  /* v2.06, force tx=1 */
464#define LXSIWZX    (XO31(12) | 1)   /* v2.07, force tx=1 */
465#define LXV        (OPCD(61) | 8 | 1)  /* v3.00, force tx=1 */
466#define LXSD       (OPCD(57) | 2)   /* v3.00 */
467#define LXVWSX     (XO31(364) | 1)  /* v3.00, force tx=1 */
468
469#define STVX       XO31(231)
470#define STVEWX     XO31(199)
471#define STXSDX     (XO31(716) | 1)  /* v2.06, force sx=1 */
472#define STXSIWX    (XO31(140) | 1)  /* v2.07, force sx=1 */
473#define STXV       (OPCD(61) | 8 | 5) /* v3.00, force sx=1 */
474#define STXSD      (OPCD(61) | 2)   /* v3.00 */
475
476#define VADDSBS    VX4(768)
477#define VADDUBS    VX4(512)
478#define VADDUBM    VX4(0)
479#define VADDSHS    VX4(832)
480#define VADDUHS    VX4(576)
481#define VADDUHM    VX4(64)
482#define VADDSWS    VX4(896)
483#define VADDUWS    VX4(640)
484#define VADDUWM    VX4(128)
485#define VADDUDM    VX4(192)       /* v2.07 */
486
487#define VSUBSBS    VX4(1792)
488#define VSUBUBS    VX4(1536)
489#define VSUBUBM    VX4(1024)
490#define VSUBSHS    VX4(1856)
491#define VSUBUHS    VX4(1600)
492#define VSUBUHM    VX4(1088)
493#define VSUBSWS    VX4(1920)
494#define VSUBUWS    VX4(1664)
495#define VSUBUWM    VX4(1152)
496#define VSUBUDM    VX4(1216)      /* v2.07 */
497
498#define VNEGW      (VX4(1538) | (6 << 16))  /* v3.00 */
499#define VNEGD      (VX4(1538) | (7 << 16))  /* v3.00 */
500
501#define VMAXSB     VX4(258)
502#define VMAXSH     VX4(322)
503#define VMAXSW     VX4(386)
504#define VMAXSD     VX4(450)       /* v2.07 */
505#define VMAXUB     VX4(2)
506#define VMAXUH     VX4(66)
507#define VMAXUW     VX4(130)
508#define VMAXUD     VX4(194)       /* v2.07 */
509#define VMINSB     VX4(770)
510#define VMINSH     VX4(834)
511#define VMINSW     VX4(898)
512#define VMINSD     VX4(962)       /* v2.07 */
513#define VMINUB     VX4(514)
514#define VMINUH     VX4(578)
515#define VMINUW     VX4(642)
516#define VMINUD     VX4(706)       /* v2.07 */
517
518#define VCMPEQUB   VX4(6)
519#define VCMPEQUH   VX4(70)
520#define VCMPEQUW   VX4(134)
521#define VCMPEQUD   VX4(199)       /* v2.07 */
522#define VCMPGTSB   VX4(774)
523#define VCMPGTSH   VX4(838)
524#define VCMPGTSW   VX4(902)
525#define VCMPGTSD   VX4(967)       /* v2.07 */
526#define VCMPGTUB   VX4(518)
527#define VCMPGTUH   VX4(582)
528#define VCMPGTUW   VX4(646)
529#define VCMPGTUD   VX4(711)       /* v2.07 */
530#define VCMPNEB    VX4(7)         /* v3.00 */
531#define VCMPNEH    VX4(71)        /* v3.00 */
532#define VCMPNEW    VX4(135)       /* v3.00 */
533
534#define VSLB       VX4(260)
535#define VSLH       VX4(324)
536#define VSLW       VX4(388)
537#define VSLD       VX4(1476)      /* v2.07 */
538#define VSRB       VX4(516)
539#define VSRH       VX4(580)
540#define VSRW       VX4(644)
541#define VSRD       VX4(1732)      /* v2.07 */
542#define VSRAB      VX4(772)
543#define VSRAH      VX4(836)
544#define VSRAW      VX4(900)
545#define VSRAD      VX4(964)       /* v2.07 */
546#define VRLB       VX4(4)
547#define VRLH       VX4(68)
548#define VRLW       VX4(132)
549#define VRLD       VX4(196)       /* v2.07 */
550
551#define VMULEUB    VX4(520)
552#define VMULEUH    VX4(584)
553#define VMULEUW    VX4(648)       /* v2.07 */
554#define VMULOUB    VX4(8)
555#define VMULOUH    VX4(72)
556#define VMULOUW    VX4(136)       /* v2.07 */
557#define VMULUWM    VX4(137)       /* v2.07 */
558#define VMULLD     VX4(457)       /* v3.10 */
559#define VMSUMUHM   VX4(38)
560
561#define VMRGHB     VX4(12)
562#define VMRGHH     VX4(76)
563#define VMRGHW     VX4(140)
564#define VMRGLB     VX4(268)
565#define VMRGLH     VX4(332)
566#define VMRGLW     VX4(396)
567
568#define VPKUHUM    VX4(14)
569#define VPKUWUM    VX4(78)
570
571#define VAND       VX4(1028)
572#define VANDC      VX4(1092)
573#define VNOR       VX4(1284)
574#define VOR        VX4(1156)
575#define VXOR       VX4(1220)
576#define VEQV       VX4(1668)      /* v2.07 */
577#define VNAND      VX4(1412)      /* v2.07 */
578#define VORC       VX4(1348)      /* v2.07 */
579
580#define VSPLTB     VX4(524)
581#define VSPLTH     VX4(588)
582#define VSPLTW     VX4(652)
583#define VSPLTISB   VX4(780)
584#define VSPLTISH   VX4(844)
585#define VSPLTISW   VX4(908)
586
587#define VSLDOI     VX4(44)
588
589#define XXPERMDI   (OPCD(60) | (10 << 3) | 7)  /* v2.06, force ax=bx=tx=1 */
590#define XXSEL      (OPCD(60) | (3 << 4) | 0xf) /* v2.06, force ax=bx=cx=tx=1 */
591#define XXSPLTIB   (OPCD(60) | (360 << 1) | 1) /* v3.00, force tx=1 */
592
593#define MFVSRD     (XO31(51) | 1)   /* v2.07, force sx=1 */
594#define MFVSRWZ    (XO31(115) | 1)  /* v2.07, force sx=1 */
595#define MTVSRD     (XO31(179) | 1)  /* v2.07, force tx=1 */
596#define MTVSRWZ    (XO31(243) | 1)  /* v2.07, force tx=1 */
597#define MTVSRDD    (XO31(435) | 1)  /* v3.00, force tx=1 */
598#define MTVSRWS    (XO31(403) | 1)  /* v3.00, force tx=1 */
599
600#define RT(r) ((r)<<21)
601#define RS(r) ((r)<<21)
602#define RA(r) ((r)<<16)
603#define RB(r) ((r)<<11)
604#define TO(t) ((t)<<21)
605#define SH(s) ((s)<<11)
606#define MB(b) ((b)<<6)
607#define ME(e) ((e)<<1)
608#define BO(o) ((o)<<21)
609#define MB64(b) ((b)<<5)
610#define FXM(b) (1 << (19 - (b)))
611
612#define VRT(r)  (((r) & 31) << 21)
613#define VRA(r)  (((r) & 31) << 16)
614#define VRB(r)  (((r) & 31) << 11)
615#define VRC(r)  (((r) & 31) <<  6)
616
617#define LK    1
618
619#define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
620#define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
621#define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
622#define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
623
624#define BF(n)    ((n)<<23)
625#define BI(n, c) (((c)+((n)*4))<<16)
626#define BT(n, c) (((c)+((n)*4))<<21)
627#define BA(n, c) (((c)+((n)*4))<<16)
628#define BB(n, c) (((c)+((n)*4))<<11)
629#define BC_(n, c) (((c)+((n)*4))<<6)
630
631#define BO_COND_TRUE  BO(12)
632#define BO_COND_FALSE BO( 4)
633#define BO_ALWAYS     BO(20)
634
635enum {
636    CR_LT,
637    CR_GT,
638    CR_EQ,
639    CR_SO
640};
641
642static const uint32_t tcg_to_bc[] = {
643    [TCG_COND_EQ]  = BC | BI(7, CR_EQ) | BO_COND_TRUE,
644    [TCG_COND_NE]  = BC | BI(7, CR_EQ) | BO_COND_FALSE,
645    [TCG_COND_LT]  = BC | BI(7, CR_LT) | BO_COND_TRUE,
646    [TCG_COND_GE]  = BC | BI(7, CR_LT) | BO_COND_FALSE,
647    [TCG_COND_LE]  = BC | BI(7, CR_GT) | BO_COND_FALSE,
648    [TCG_COND_GT]  = BC | BI(7, CR_GT) | BO_COND_TRUE,
649    [TCG_COND_LTU] = BC | BI(7, CR_LT) | BO_COND_TRUE,
650    [TCG_COND_GEU] = BC | BI(7, CR_LT) | BO_COND_FALSE,
651    [TCG_COND_LEU] = BC | BI(7, CR_GT) | BO_COND_FALSE,
652    [TCG_COND_GTU] = BC | BI(7, CR_GT) | BO_COND_TRUE,
653};
654
655/* The low bit here is set if the RA and RB fields must be inverted.  */
656static const uint32_t tcg_to_isel[] = {
657    [TCG_COND_EQ]  = ISEL | BC_(7, CR_EQ),
658    [TCG_COND_NE]  = ISEL | BC_(7, CR_EQ) | 1,
659    [TCG_COND_LT]  = ISEL | BC_(7, CR_LT),
660    [TCG_COND_GE]  = ISEL | BC_(7, CR_LT) | 1,
661    [TCG_COND_LE]  = ISEL | BC_(7, CR_GT) | 1,
662    [TCG_COND_GT]  = ISEL | BC_(7, CR_GT),
663    [TCG_COND_LTU] = ISEL | BC_(7, CR_LT),
664    [TCG_COND_GEU] = ISEL | BC_(7, CR_LT) | 1,
665    [TCG_COND_LEU] = ISEL | BC_(7, CR_GT) | 1,
666    [TCG_COND_GTU] = ISEL | BC_(7, CR_GT),
667};
668
669static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
670                        intptr_t value, intptr_t addend)
671{
672    const tcg_insn_unit *target;
673    int16_t lo;
674    int32_t hi;
675
676    value += addend;
677    target = (const tcg_insn_unit *)value;
678
679    switch (type) {
680    case R_PPC_REL14:
681        return reloc_pc14(code_ptr, target);
682    case R_PPC_REL24:
683        return reloc_pc24(code_ptr, target);
684    case R_PPC_ADDR16:
685        /*
686         * We are (slightly) abusing this relocation type.  In particular,
687         * assert that the low 2 bits are zero, and do not modify them.
688         * That way we can use this with LD et al that have opcode bits
689         * in the low 2 bits of the insn.
690         */
691        if ((value & 3) || value != (int16_t)value) {
692            return false;
693        }
694        *code_ptr = (*code_ptr & ~0xfffc) | (value & 0xfffc);
695        break;
696    case R_PPC_ADDR32:
697        /*
698         * We are abusing this relocation type.  Again, this points to
699         * a pair of insns, lis + load.  This is an absolute address
700         * relocation for PPC32 so the lis cannot be removed.
701         */
702        lo = value;
703        hi = value - lo;
704        if (hi + lo != value) {
705            return false;
706        }
707        code_ptr[0] = deposit32(code_ptr[0], 0, 16, hi >> 16);
708        code_ptr[1] = deposit32(code_ptr[1], 0, 16, lo);
709        break;
710    default:
711        g_assert_not_reached();
712    }
713    return true;
714}
715
716static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
717                             TCGReg base, tcg_target_long offset);
718
719static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
720{
721    if (ret == arg) {
722        return true;
723    }
724    switch (type) {
725    case TCG_TYPE_I64:
726        tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
727        /* fallthru */
728    case TCG_TYPE_I32:
729        if (ret < TCG_REG_V0) {
730            if (arg < TCG_REG_V0) {
731                tcg_out32(s, OR | SAB(arg, ret, arg));
732                break;
733            } else if (have_isa_2_07) {
734                tcg_out32(s, (type == TCG_TYPE_I32 ? MFVSRWZ : MFVSRD)
735                          | VRT(arg) | RA(ret));
736                break;
737            } else {
738                /* Altivec does not support vector->integer moves.  */
739                return false;
740            }
741        } else if (arg < TCG_REG_V0) {
742            if (have_isa_2_07) {
743                tcg_out32(s, (type == TCG_TYPE_I32 ? MTVSRWZ : MTVSRD)
744                          | VRT(ret) | RA(arg));
745                break;
746            } else {
747                /* Altivec does not support integer->vector moves.  */
748                return false;
749            }
750        }
751        /* fallthru */
752    case TCG_TYPE_V64:
753    case TCG_TYPE_V128:
754        tcg_debug_assert(ret >= TCG_REG_V0 && arg >= TCG_REG_V0);
755        tcg_out32(s, VOR | VRT(ret) | VRA(arg) | VRB(arg));
756        break;
757    default:
758        g_assert_not_reached();
759    }
760    return true;
761}
762
763static inline void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs,
764                               int sh, int mb)
765{
766    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
767    sh = SH(sh & 0x1f) | (((sh >> 5) & 1) << 1);
768    mb = MB64((mb >> 5) | ((mb << 1) & 0x3f));
769    tcg_out32(s, op | RA(ra) | RS(rs) | sh | mb);
770}
771
772static inline void tcg_out_rlw(TCGContext *s, int op, TCGReg ra, TCGReg rs,
773                               int sh, int mb, int me)
774{
775    tcg_out32(s, op | RA(ra) | RS(rs) | SH(sh) | MB(mb) | ME(me));
776}
777
778static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
779{
780    tcg_out32(s, EXTSB | RA(dst) | RS(src));
781}
782
783static void tcg_out_ext8u(TCGContext *s, TCGReg dst, TCGReg src)
784{
785    tcg_out32(s, ANDI | SAI(src, dst, 0xff));
786}
787
788static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
789{
790    tcg_out32(s, EXTSH | RA(dst) | RS(src));
791}
792
793static void tcg_out_ext16u(TCGContext *s, TCGReg dst, TCGReg src)
794{
795    tcg_out32(s, ANDI | SAI(src, dst, 0xffff));
796}
797
798static void tcg_out_ext32s(TCGContext *s, TCGReg dst, TCGReg src)
799{
800    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
801    tcg_out32(s, EXTSW | RA(dst) | RS(src));
802}
803
804static void tcg_out_ext32u(TCGContext *s, TCGReg dst, TCGReg src)
805{
806    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
807    tcg_out_rld(s, RLDICL, dst, src, 0, 32);
808}
809
810static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dst, TCGReg src)
811{
812    tcg_out_ext32s(s, dst, src);
813}
814
815static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dst, TCGReg src)
816{
817    tcg_out_ext32u(s, dst, src);
818}
819
820static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rn)
821{
822    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
823    tcg_out_mov(s, TCG_TYPE_I32, rd, rn);
824}
825
826static inline void tcg_out_shli32(TCGContext *s, TCGReg dst, TCGReg src, int c)
827{
828    tcg_out_rlw(s, RLWINM, dst, src, c, 0, 31 - c);
829}
830
831static inline void tcg_out_shli64(TCGContext *s, TCGReg dst, TCGReg src, int c)
832{
833    tcg_out_rld(s, RLDICR, dst, src, c, 63 - c);
834}
835
836static inline void tcg_out_sari32(TCGContext *s, TCGReg dst, TCGReg src, int c)
837{
838    /* Limit immediate shift count lest we create an illegal insn.  */
839    tcg_out32(s, SRAWI | RA(dst) | RS(src) | SH(c & 31));
840}
841
842static inline void tcg_out_shri32(TCGContext *s, TCGReg dst, TCGReg src, int c)
843{
844    tcg_out_rlw(s, RLWINM, dst, src, 32 - c, c, 31);
845}
846
847static inline void tcg_out_shri64(TCGContext *s, TCGReg dst, TCGReg src, int c)
848{
849    tcg_out_rld(s, RLDICL, dst, src, 64 - c, c);
850}
851
852static inline void tcg_out_sari64(TCGContext *s, TCGReg dst, TCGReg src, int c)
853{
854    tcg_out32(s, SRADI | RA(dst) | RS(src) | SH(c & 0x1f) | ((c >> 4) & 2));
855}
856
857static void tcg_out_bswap16(TCGContext *s, TCGReg dst, TCGReg src, int flags)
858{
859    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
860
861    if (have_isa_3_10) {
862        tcg_out32(s, BRH | RA(dst) | RS(src));
863        if (flags & TCG_BSWAP_OS) {
864            tcg_out_ext16s(s, TCG_TYPE_REG, dst, dst);
865        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
866            tcg_out_ext16u(s, dst, dst);
867        }
868        return;
869    }
870
871    /*
872     * In the following,
873     *   dep(a, b, m) -> (a & ~m) | (b & m)
874     *
875     * Begin with:                              src = xxxxabcd
876     */
877    /* tmp = rol32(src, 24) & 0x000000ff            = 0000000c */
878    tcg_out_rlw(s, RLWINM, tmp, src, 24, 24, 31);
879    /* tmp = dep(tmp, rol32(src, 8), 0x0000ff00)    = 000000dc */
880    tcg_out_rlw(s, RLWIMI, tmp, src, 8, 16, 23);
881
882    if (flags & TCG_BSWAP_OS) {
883        tcg_out_ext16s(s, TCG_TYPE_REG, dst, tmp);
884    } else {
885        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
886    }
887}
888
889static void tcg_out_bswap32(TCGContext *s, TCGReg dst, TCGReg src, int flags)
890{
891    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
892
893    if (have_isa_3_10) {
894        tcg_out32(s, BRW | RA(dst) | RS(src));
895        if (flags & TCG_BSWAP_OS) {
896            tcg_out_ext32s(s, dst, dst);
897        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
898            tcg_out_ext32u(s, dst, dst);
899        }
900        return;
901    }
902
903    /*
904     * Stolen from gcc's builtin_bswap32.
905     * In the following,
906     *   dep(a, b, m) -> (a & ~m) | (b & m)
907     *
908     * Begin with:                              src = xxxxabcd
909     */
910    /* tmp = rol32(src, 8) & 0xffffffff             = 0000bcda */
911    tcg_out_rlw(s, RLWINM, tmp, src, 8, 0, 31);
912    /* tmp = dep(tmp, rol32(src, 24), 0xff000000)   = 0000dcda */
913    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 0, 7);
914    /* tmp = dep(tmp, rol32(src, 24), 0x0000ff00)   = 0000dcba */
915    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 16, 23);
916
917    if (flags & TCG_BSWAP_OS) {
918        tcg_out_ext32s(s, dst, tmp);
919    } else {
920        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
921    }
922}
923
924static void tcg_out_bswap64(TCGContext *s, TCGReg dst, TCGReg src)
925{
926    TCGReg t0 = dst == src ? TCG_REG_R0 : dst;
927    TCGReg t1 = dst == src ? dst : TCG_REG_R0;
928
929    if (have_isa_3_10) {
930        tcg_out32(s, BRD | RA(dst) | RS(src));
931        return;
932    }
933
934    /*
935     * In the following,
936     *   dep(a, b, m) -> (a & ~m) | (b & m)
937     *
938     * Begin with:                              src = abcdefgh
939     */
940    /* t0 = rol32(src, 8) & 0xffffffff              = 0000fghe */
941    tcg_out_rlw(s, RLWINM, t0, src, 8, 0, 31);
942    /* t0 = dep(t0, rol32(src, 24), 0xff000000)     = 0000hghe */
943    tcg_out_rlw(s, RLWIMI, t0, src, 24, 0, 7);
944    /* t0 = dep(t0, rol32(src, 24), 0x0000ff00)     = 0000hgfe */
945    tcg_out_rlw(s, RLWIMI, t0, src, 24, 16, 23);
946
947    /* t0 = rol64(t0, 32)                           = hgfe0000 */
948    tcg_out_rld(s, RLDICL, t0, t0, 32, 0);
949    /* t1 = rol64(src, 32)                          = efghabcd */
950    tcg_out_rld(s, RLDICL, t1, src, 32, 0);
951
952    /* t0 = dep(t0, rol32(t1, 24), 0xffffffff)      = hgfebcda */
953    tcg_out_rlw(s, RLWIMI, t0, t1, 8, 0, 31);
954    /* t0 = dep(t0, rol32(t1, 24), 0xff000000)      = hgfedcda */
955    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 0, 7);
956    /* t0 = dep(t0, rol32(t1, 24), 0x0000ff00)      = hgfedcba */
957    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 16, 23);
958
959    tcg_out_mov(s, TCG_TYPE_REG, dst, t0);
960}
961
962/* Emit a move into ret of arg, if it can be done in one insn.  */
963static bool tcg_out_movi_one(TCGContext *s, TCGReg ret, tcg_target_long arg)
964{
965    if (arg == (int16_t)arg) {
966        tcg_out32(s, ADDI | TAI(ret, 0, arg));
967        return true;
968    }
969    if (arg == (int32_t)arg && (arg & 0xffff) == 0) {
970        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
971        return true;
972    }
973    return false;
974}
975
976static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
977                             tcg_target_long arg, bool in_prologue)
978{
979    intptr_t tb_diff;
980    tcg_target_long tmp;
981    int shift;
982
983    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
984
985    if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
986        arg = (int32_t)arg;
987    }
988
989    /* Load 16-bit immediates with one insn.  */
990    if (tcg_out_movi_one(s, ret, arg)) {
991        return;
992    }
993
994    /* Load addresses within the TB with one insn.  */
995    tb_diff = tcg_tbrel_diff(s, (void *)arg);
996    if (!in_prologue && USE_REG_TB && tb_diff == (int16_t)tb_diff) {
997        tcg_out32(s, ADDI | TAI(ret, TCG_REG_TB, tb_diff));
998        return;
999    }
1000
1001    /* Load 32-bit immediates with two insns.  Note that we've already
1002       eliminated bare ADDIS, so we know both insns are required.  */
1003    if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) {
1004        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
1005        tcg_out32(s, ORI | SAI(ret, ret, arg));
1006        return;
1007    }
1008    if (arg == (uint32_t)arg && !(arg & 0x8000)) {
1009        tcg_out32(s, ADDI | TAI(ret, 0, arg));
1010        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
1011        return;
1012    }
1013
1014    /* Load masked 16-bit value.  */
1015    if (arg > 0 && (arg & 0x8000)) {
1016        tmp = arg | 0x7fff;
1017        if ((tmp & (tmp + 1)) == 0) {
1018            int mb = clz64(tmp + 1) + 1;
1019            tcg_out32(s, ADDI | TAI(ret, 0, arg));
1020            tcg_out_rld(s, RLDICL, ret, ret, 0, mb);
1021            return;
1022        }
1023    }
1024
1025    /* Load common masks with 2 insns.  */
1026    shift = ctz64(arg);
1027    tmp = arg >> shift;
1028    if (tmp == (int16_t)tmp) {
1029        tcg_out32(s, ADDI | TAI(ret, 0, tmp));
1030        tcg_out_shli64(s, ret, ret, shift);
1031        return;
1032    }
1033    shift = clz64(arg);
1034    if (tcg_out_movi_one(s, ret, arg << shift)) {
1035        tcg_out_shri64(s, ret, ret, shift);
1036        return;
1037    }
1038
1039    /* Load addresses within 2GB of TB with 2 (or rarely 3) insns.  */
1040    if (!in_prologue && USE_REG_TB && tb_diff == (int32_t)tb_diff) {
1041        tcg_out_mem_long(s, ADDI, ADD, ret, TCG_REG_TB, tb_diff);
1042        return;
1043    }
1044
1045    /* Use the constant pool, if possible.  */
1046    if (!in_prologue && USE_REG_TB) {
1047        new_pool_label(s, arg, R_PPC_ADDR16, s->code_ptr,
1048                       tcg_tbrel_diff(s, NULL));
1049        tcg_out32(s, LD | TAI(ret, TCG_REG_TB, 0));
1050        return;
1051    }
1052
1053    tmp = arg >> 31 >> 1;
1054    tcg_out_movi(s, TCG_TYPE_I32, ret, tmp);
1055    if (tmp) {
1056        tcg_out_shli64(s, ret, ret, 32);
1057    }
1058    if (arg & 0xffff0000) {
1059        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
1060    }
1061    if (arg & 0xffff) {
1062        tcg_out32(s, ORI | SAI(ret, ret, arg));
1063    }
1064}
1065
1066static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
1067                             TCGReg ret, int64_t val)
1068{
1069    uint32_t load_insn;
1070    int rel, low;
1071    intptr_t add;
1072
1073    switch (vece) {
1074    case MO_8:
1075        low = (int8_t)val;
1076        if (low >= -16 && low < 16) {
1077            tcg_out32(s, VSPLTISB | VRT(ret) | ((val & 31) << 16));
1078            return;
1079        }
1080        if (have_isa_3_00) {
1081            tcg_out32(s, XXSPLTIB | VRT(ret) | ((val & 0xff) << 11));
1082            return;
1083        }
1084        break;
1085
1086    case MO_16:
1087        low = (int16_t)val;
1088        if (low >= -16 && low < 16) {
1089            tcg_out32(s, VSPLTISH | VRT(ret) | ((val & 31) << 16));
1090            return;
1091        }
1092        break;
1093
1094    case MO_32:
1095        low = (int32_t)val;
1096        if (low >= -16 && low < 16) {
1097            tcg_out32(s, VSPLTISW | VRT(ret) | ((val & 31) << 16));
1098            return;
1099        }
1100        break;
1101    }
1102
1103    /*
1104     * Otherwise we must load the value from the constant pool.
1105     */
1106    if (USE_REG_TB) {
1107        rel = R_PPC_ADDR16;
1108        add = tcg_tbrel_diff(s, NULL);
1109    } else {
1110        rel = R_PPC_ADDR32;
1111        add = 0;
1112    }
1113
1114    if (have_vsx) {
1115        load_insn = type == TCG_TYPE_V64 ? LXSDX : LXVDSX;
1116        load_insn |= VRT(ret) | RB(TCG_REG_TMP1);
1117        if (TCG_TARGET_REG_BITS == 64) {
1118            new_pool_label(s, val, rel, s->code_ptr, add);
1119        } else {
1120            new_pool_l2(s, rel, s->code_ptr, add, val >> 32, val);
1121        }
1122    } else {
1123        load_insn = LVX | VRT(ret) | RB(TCG_REG_TMP1);
1124        if (TCG_TARGET_REG_BITS == 64) {
1125            new_pool_l2(s, rel, s->code_ptr, add, val, val);
1126        } else {
1127            new_pool_l4(s, rel, s->code_ptr, add,
1128                        val >> 32, val, val >> 32, val);
1129        }
1130    }
1131
1132    if (USE_REG_TB) {
1133        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, 0, 0));
1134        load_insn |= RA(TCG_REG_TB);
1135    } else {
1136        tcg_out32(s, ADDIS | TAI(TCG_REG_TMP1, 0, 0));
1137        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, TCG_REG_TMP1, 0));
1138    }
1139    tcg_out32(s, load_insn);
1140}
1141
1142static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,
1143                         tcg_target_long arg)
1144{
1145    switch (type) {
1146    case TCG_TYPE_I32:
1147    case TCG_TYPE_I64:
1148        tcg_debug_assert(ret < TCG_REG_V0);
1149        tcg_out_movi_int(s, type, ret, arg, false);
1150        break;
1151
1152    default:
1153        g_assert_not_reached();
1154    }
1155}
1156
1157static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
1158{
1159    return false;
1160}
1161
1162static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
1163                             tcg_target_long imm)
1164{
1165    /* This function is only used for passing structs by reference. */
1166    g_assert_not_reached();
1167}
1168
1169static bool mask_operand(uint32_t c, int *mb, int *me)
1170{
1171    uint32_t lsb, test;
1172
1173    /* Accept a bit pattern like:
1174           0....01....1
1175           1....10....0
1176           0..01..10..0
1177       Keep track of the transitions.  */
1178    if (c == 0 || c == -1) {
1179        return false;
1180    }
1181    test = c;
1182    lsb = test & -test;
1183    test += lsb;
1184    if (test & (test - 1)) {
1185        return false;
1186    }
1187
1188    *me = clz32(lsb);
1189    *mb = test ? clz32(test & -test) + 1 : 0;
1190    return true;
1191}
1192
1193static bool mask64_operand(uint64_t c, int *mb, int *me)
1194{
1195    uint64_t lsb;
1196
1197    if (c == 0) {
1198        return false;
1199    }
1200
1201    lsb = c & -c;
1202    /* Accept 1..10..0.  */
1203    if (c == -lsb) {
1204        *mb = 0;
1205        *me = clz64(lsb);
1206        return true;
1207    }
1208    /* Accept 0..01..1.  */
1209    if (lsb == 1 && (c & (c + 1)) == 0) {
1210        *mb = clz64(c + 1) + 1;
1211        *me = 63;
1212        return true;
1213    }
1214    return false;
1215}
1216
1217static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1218{
1219    int mb, me;
1220
1221    if (mask_operand(c, &mb, &me)) {
1222        tcg_out_rlw(s, RLWINM, dst, src, 0, mb, me);
1223    } else if ((c & 0xffff) == c) {
1224        tcg_out32(s, ANDI | SAI(src, dst, c));
1225        return;
1226    } else if ((c & 0xffff0000) == c) {
1227        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1228        return;
1229    } else {
1230        tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R0, c);
1231        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1232    }
1233}
1234
1235static void tcg_out_andi64(TCGContext *s, TCGReg dst, TCGReg src, uint64_t c)
1236{
1237    int mb, me;
1238
1239    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1240    if (mask64_operand(c, &mb, &me)) {
1241        if (mb == 0) {
1242            tcg_out_rld(s, RLDICR, dst, src, 0, me);
1243        } else {
1244            tcg_out_rld(s, RLDICL, dst, src, 0, mb);
1245        }
1246    } else if ((c & 0xffff) == c) {
1247        tcg_out32(s, ANDI | SAI(src, dst, c));
1248        return;
1249    } else if ((c & 0xffff0000) == c) {
1250        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1251        return;
1252    } else {
1253        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, c);
1254        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1255    }
1256}
1257
1258static void tcg_out_zori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c,
1259                           int op_lo, int op_hi)
1260{
1261    if (c >> 16) {
1262        tcg_out32(s, op_hi | SAI(src, dst, c >> 16));
1263        src = dst;
1264    }
1265    if (c & 0xffff) {
1266        tcg_out32(s, op_lo | SAI(src, dst, c));
1267        src = dst;
1268    }
1269}
1270
1271static void tcg_out_ori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1272{
1273    tcg_out_zori32(s, dst, src, c, ORI, ORIS);
1274}
1275
1276static void tcg_out_xori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1277{
1278    tcg_out_zori32(s, dst, src, c, XORI, XORIS);
1279}
1280
1281static void tcg_out_b(TCGContext *s, int mask, const tcg_insn_unit *target)
1282{
1283    ptrdiff_t disp = tcg_pcrel_diff(s, target);
1284    if (in_range_b(disp)) {
1285        tcg_out32(s, B | (disp & 0x3fffffc) | mask);
1286    } else {
1287        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, (uintptr_t)target);
1288        tcg_out32(s, MTSPR | RS(TCG_REG_R0) | CTR);
1289        tcg_out32(s, BCCTR | BO_ALWAYS | mask);
1290    }
1291}
1292
1293static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
1294                             TCGReg base, tcg_target_long offset)
1295{
1296    tcg_target_long orig = offset, l0, l1, extra = 0, align = 0;
1297    bool is_int_store = false;
1298    TCGReg rs = TCG_REG_TMP1;
1299
1300    switch (opi) {
1301    case LD: case LWA:
1302        align = 3;
1303        /* FALLTHRU */
1304    default:
1305        if (rt > TCG_REG_R0 && rt < TCG_REG_V0) {
1306            rs = rt;
1307            break;
1308        }
1309        break;
1310    case LXSD:
1311    case STXSD:
1312        align = 3;
1313        break;
1314    case LXV:
1315    case STXV:
1316        align = 15;
1317        break;
1318    case STD:
1319        align = 3;
1320        /* FALLTHRU */
1321    case STB: case STH: case STW:
1322        is_int_store = true;
1323        break;
1324    }
1325
1326    /* For unaligned, or very large offsets, use the indexed form.  */
1327    if (offset & align || offset != (int32_t)offset || opi == 0) {
1328        if (rs == base) {
1329            rs = TCG_REG_R0;
1330        }
1331        tcg_debug_assert(!is_int_store || rs != rt);
1332        tcg_out_movi(s, TCG_TYPE_PTR, rs, orig);
1333        tcg_out32(s, opx | TAB(rt & 31, base, rs));
1334        return;
1335    }
1336
1337    l0 = (int16_t)offset;
1338    offset = (offset - l0) >> 16;
1339    l1 = (int16_t)offset;
1340
1341    if (l1 < 0 && orig >= 0) {
1342        extra = 0x4000;
1343        l1 = (int16_t)(offset - 0x4000);
1344    }
1345    if (l1) {
1346        tcg_out32(s, ADDIS | TAI(rs, base, l1));
1347        base = rs;
1348    }
1349    if (extra) {
1350        tcg_out32(s, ADDIS | TAI(rs, base, extra));
1351        base = rs;
1352    }
1353    if (opi != ADDI || base != rt || l0 != 0) {
1354        tcg_out32(s, opi | TAI(rt & 31, base, l0));
1355    }
1356}
1357
1358static void tcg_out_vsldoi(TCGContext *s, TCGReg ret,
1359                           TCGReg va, TCGReg vb, int shb)
1360{
1361    tcg_out32(s, VSLDOI | VRT(ret) | VRA(va) | VRB(vb) | (shb << 6));
1362}
1363
1364static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
1365                       TCGReg base, intptr_t offset)
1366{
1367    int shift;
1368
1369    switch (type) {
1370    case TCG_TYPE_I32:
1371        if (ret < TCG_REG_V0) {
1372            tcg_out_mem_long(s, LWZ, LWZX, ret, base, offset);
1373            break;
1374        }
1375        if (have_isa_2_07 && have_vsx) {
1376            tcg_out_mem_long(s, 0, LXSIWZX, ret, base, offset);
1377            break;
1378        }
1379        tcg_debug_assert((offset & 3) == 0);
1380        tcg_out_mem_long(s, 0, LVEWX, ret, base, offset);
1381        shift = (offset - 4) & 0xc;
1382        if (shift) {
1383            tcg_out_vsldoi(s, ret, ret, ret, shift);
1384        }
1385        break;
1386    case TCG_TYPE_I64:
1387        if (ret < TCG_REG_V0) {
1388            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1389            tcg_out_mem_long(s, LD, LDX, ret, base, offset);
1390            break;
1391        }
1392        /* fallthru */
1393    case TCG_TYPE_V64:
1394        tcg_debug_assert(ret >= TCG_REG_V0);
1395        if (have_vsx) {
1396            tcg_out_mem_long(s, have_isa_3_00 ? LXSD : 0, LXSDX,
1397                             ret, base, offset);
1398            break;
1399        }
1400        tcg_debug_assert((offset & 7) == 0);
1401        tcg_out_mem_long(s, 0, LVX, ret, base, offset & -16);
1402        if (offset & 8) {
1403            tcg_out_vsldoi(s, ret, ret, ret, 8);
1404        }
1405        break;
1406    case TCG_TYPE_V128:
1407        tcg_debug_assert(ret >= TCG_REG_V0);
1408        tcg_debug_assert((offset & 15) == 0);
1409        tcg_out_mem_long(s, have_isa_3_00 ? LXV : 0,
1410                         LVX, ret, base, offset);
1411        break;
1412    default:
1413        g_assert_not_reached();
1414    }
1415}
1416
1417static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
1418                              TCGReg base, intptr_t offset)
1419{
1420    int shift;
1421
1422    switch (type) {
1423    case TCG_TYPE_I32:
1424        if (arg < TCG_REG_V0) {
1425            tcg_out_mem_long(s, STW, STWX, arg, base, offset);
1426            break;
1427        }
1428        if (have_isa_2_07 && have_vsx) {
1429            tcg_out_mem_long(s, 0, STXSIWX, arg, base, offset);
1430            break;
1431        }
1432        assert((offset & 3) == 0);
1433        tcg_debug_assert((offset & 3) == 0);
1434        shift = (offset - 4) & 0xc;
1435        if (shift) {
1436            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, shift);
1437            arg = TCG_VEC_TMP1;
1438        }
1439        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1440        break;
1441    case TCG_TYPE_I64:
1442        if (arg < TCG_REG_V0) {
1443            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1444            tcg_out_mem_long(s, STD, STDX, arg, base, offset);
1445            break;
1446        }
1447        /* fallthru */
1448    case TCG_TYPE_V64:
1449        tcg_debug_assert(arg >= TCG_REG_V0);
1450        if (have_vsx) {
1451            tcg_out_mem_long(s, have_isa_3_00 ? STXSD : 0,
1452                             STXSDX, arg, base, offset);
1453            break;
1454        }
1455        tcg_debug_assert((offset & 7) == 0);
1456        if (offset & 8) {
1457            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, 8);
1458            arg = TCG_VEC_TMP1;
1459        }
1460        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1461        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset + 4);
1462        break;
1463    case TCG_TYPE_V128:
1464        tcg_debug_assert(arg >= TCG_REG_V0);
1465        tcg_out_mem_long(s, have_isa_3_00 ? STXV : 0,
1466                         STVX, arg, base, offset);
1467        break;
1468    default:
1469        g_assert_not_reached();
1470    }
1471}
1472
1473static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1474                               TCGReg base, intptr_t ofs)
1475{
1476    return false;
1477}
1478
1479static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2,
1480                        int const_arg2, int cr, TCGType type)
1481{
1482    int imm;
1483    uint32_t op;
1484
1485    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1486
1487    /* Simplify the comparisons below wrt CMPI.  */
1488    if (type == TCG_TYPE_I32) {
1489        arg2 = (int32_t)arg2;
1490    }
1491
1492    switch (cond) {
1493    case TCG_COND_EQ:
1494    case TCG_COND_NE:
1495        if (const_arg2) {
1496            if ((int16_t) arg2 == arg2) {
1497                op = CMPI;
1498                imm = 1;
1499                break;
1500            } else if ((uint16_t) arg2 == arg2) {
1501                op = CMPLI;
1502                imm = 1;
1503                break;
1504            }
1505        }
1506        op = CMPL;
1507        imm = 0;
1508        break;
1509
1510    case TCG_COND_LT:
1511    case TCG_COND_GE:
1512    case TCG_COND_LE:
1513    case TCG_COND_GT:
1514        if (const_arg2) {
1515            if ((int16_t) arg2 == arg2) {
1516                op = CMPI;
1517                imm = 1;
1518                break;
1519            }
1520        }
1521        op = CMP;
1522        imm = 0;
1523        break;
1524
1525    case TCG_COND_LTU:
1526    case TCG_COND_GEU:
1527    case TCG_COND_LEU:
1528    case TCG_COND_GTU:
1529        if (const_arg2) {
1530            if ((uint16_t) arg2 == arg2) {
1531                op = CMPLI;
1532                imm = 1;
1533                break;
1534            }
1535        }
1536        op = CMPL;
1537        imm = 0;
1538        break;
1539
1540    default:
1541        g_assert_not_reached();
1542    }
1543    op |= BF(cr) | ((type == TCG_TYPE_I64) << 21);
1544
1545    if (imm) {
1546        tcg_out32(s, op | RA(arg1) | (arg2 & 0xffff));
1547    } else {
1548        if (const_arg2) {
1549            tcg_out_movi(s, type, TCG_REG_R0, arg2);
1550            arg2 = TCG_REG_R0;
1551        }
1552        tcg_out32(s, op | RA(arg1) | RB(arg2));
1553    }
1554}
1555
1556static void tcg_out_setcond_eq0(TCGContext *s, TCGType type,
1557                                TCGReg dst, TCGReg src)
1558{
1559    if (type == TCG_TYPE_I32) {
1560        tcg_out32(s, CNTLZW | RS(src) | RA(dst));
1561        tcg_out_shri32(s, dst, dst, 5);
1562    } else {
1563        tcg_out32(s, CNTLZD | RS(src) | RA(dst));
1564        tcg_out_shri64(s, dst, dst, 6);
1565    }
1566}
1567
1568static void tcg_out_setcond_ne0(TCGContext *s, TCGReg dst, TCGReg src)
1569{
1570    /* X != 0 implies X + -1 generates a carry.  Extra addition
1571       trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C.  */
1572    if (dst != src) {
1573        tcg_out32(s, ADDIC | TAI(dst, src, -1));
1574        tcg_out32(s, SUBFE | TAB(dst, dst, src));
1575    } else {
1576        tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1));
1577        tcg_out32(s, SUBFE | TAB(dst, TCG_REG_R0, src));
1578    }
1579}
1580
1581static TCGReg tcg_gen_setcond_xor(TCGContext *s, TCGReg arg1, TCGArg arg2,
1582                                  bool const_arg2)
1583{
1584    if (const_arg2) {
1585        if ((uint32_t)arg2 == arg2) {
1586            tcg_out_xori32(s, TCG_REG_R0, arg1, arg2);
1587        } else {
1588            tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, arg2);
1589            tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, TCG_REG_R0));
1590        }
1591    } else {
1592        tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, arg2));
1593    }
1594    return TCG_REG_R0;
1595}
1596
1597static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
1598                            TCGArg arg0, TCGArg arg1, TCGArg arg2,
1599                            int const_arg2)
1600{
1601    int crop, sh;
1602
1603    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1604
1605    /* Ignore high bits of a potential constant arg2.  */
1606    if (type == TCG_TYPE_I32) {
1607        arg2 = (uint32_t)arg2;
1608    }
1609
1610    /* Handle common and trivial cases before handling anything else.  */
1611    if (arg2 == 0) {
1612        switch (cond) {
1613        case TCG_COND_EQ:
1614            tcg_out_setcond_eq0(s, type, arg0, arg1);
1615            return;
1616        case TCG_COND_NE:
1617            if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1618                tcg_out_ext32u(s, TCG_REG_R0, arg1);
1619                arg1 = TCG_REG_R0;
1620            }
1621            tcg_out_setcond_ne0(s, arg0, arg1);
1622            return;
1623        case TCG_COND_GE:
1624            tcg_out32(s, NOR | SAB(arg1, arg0, arg1));
1625            arg1 = arg0;
1626            /* FALLTHRU */
1627        case TCG_COND_LT:
1628            /* Extract the sign bit.  */
1629            if (type == TCG_TYPE_I32) {
1630                tcg_out_shri32(s, arg0, arg1, 31);
1631            } else {
1632                tcg_out_shri64(s, arg0, arg1, 63);
1633            }
1634            return;
1635        default:
1636            break;
1637        }
1638    }
1639
1640    /* If we have ISEL, we can implement everything with 3 or 4 insns.
1641       All other cases below are also at least 3 insns, so speed up the
1642       code generator by not considering them and always using ISEL.  */
1643    if (have_isel) {
1644        int isel, tab;
1645
1646        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1647
1648        isel = tcg_to_isel[cond];
1649
1650        tcg_out_movi(s, type, arg0, 1);
1651        if (isel & 1) {
1652            /* arg0 = (bc ? 0 : 1) */
1653            tab = TAB(arg0, 0, arg0);
1654            isel &= ~1;
1655        } else {
1656            /* arg0 = (bc ? 1 : 0) */
1657            tcg_out_movi(s, type, TCG_REG_R0, 0);
1658            tab = TAB(arg0, arg0, TCG_REG_R0);
1659        }
1660        tcg_out32(s, isel | tab);
1661        return;
1662    }
1663
1664    switch (cond) {
1665    case TCG_COND_EQ:
1666        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1667        tcg_out_setcond_eq0(s, type, arg0, arg1);
1668        return;
1669
1670    case TCG_COND_NE:
1671        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1672        /* Discard the high bits only once, rather than both inputs.  */
1673        if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1674            tcg_out_ext32u(s, TCG_REG_R0, arg1);
1675            arg1 = TCG_REG_R0;
1676        }
1677        tcg_out_setcond_ne0(s, arg0, arg1);
1678        return;
1679
1680    case TCG_COND_GT:
1681    case TCG_COND_GTU:
1682        sh = 30;
1683        crop = 0;
1684        goto crtest;
1685
1686    case TCG_COND_LT:
1687    case TCG_COND_LTU:
1688        sh = 29;
1689        crop = 0;
1690        goto crtest;
1691
1692    case TCG_COND_GE:
1693    case TCG_COND_GEU:
1694        sh = 31;
1695        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_LT) | BB(7, CR_LT);
1696        goto crtest;
1697
1698    case TCG_COND_LE:
1699    case TCG_COND_LEU:
1700        sh = 31;
1701        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_GT) | BB(7, CR_GT);
1702    crtest:
1703        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1704        if (crop) {
1705            tcg_out32(s, crop);
1706        }
1707        tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1708        tcg_out_rlw(s, RLWINM, arg0, TCG_REG_R0, sh, 31, 31);
1709        break;
1710
1711    default:
1712        g_assert_not_reached();
1713    }
1714}
1715
1716static void tcg_out_bc(TCGContext *s, int bc, TCGLabel *l)
1717{
1718    if (l->has_value) {
1719        bc |= reloc_pc14_val(tcg_splitwx_to_rx(s->code_ptr), l->u.value_ptr);
1720    } else {
1721        tcg_out_reloc(s, s->code_ptr, R_PPC_REL14, l, 0);
1722    }
1723    tcg_out32(s, bc);
1724}
1725
1726static void tcg_out_brcond(TCGContext *s, TCGCond cond,
1727                           TCGArg arg1, TCGArg arg2, int const_arg2,
1728                           TCGLabel *l, TCGType type)
1729{
1730    tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1731    tcg_out_bc(s, tcg_to_bc[cond], l);
1732}
1733
1734static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
1735                            TCGArg dest, TCGArg c1, TCGArg c2, TCGArg v1,
1736                            TCGArg v2, bool const_c2)
1737{
1738    /* If for some reason both inputs are zero, don't produce bad code.  */
1739    if (v1 == 0 && v2 == 0) {
1740        tcg_out_movi(s, type, dest, 0);
1741        return;
1742    }
1743
1744    tcg_out_cmp(s, cond, c1, c2, const_c2, 7, type);
1745
1746    if (have_isel) {
1747        int isel = tcg_to_isel[cond];
1748
1749        /* Swap the V operands if the operation indicates inversion.  */
1750        if (isel & 1) {
1751            int t = v1;
1752            v1 = v2;
1753            v2 = t;
1754            isel &= ~1;
1755        }
1756        /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand.  */
1757        if (v2 == 0) {
1758            tcg_out_movi(s, type, TCG_REG_R0, 0);
1759        }
1760        tcg_out32(s, isel | TAB(dest, v1, v2));
1761    } else {
1762        if (dest == v2) {
1763            cond = tcg_invert_cond(cond);
1764            v2 = v1;
1765        } else if (dest != v1) {
1766            if (v1 == 0) {
1767                tcg_out_movi(s, type, dest, 0);
1768            } else {
1769                tcg_out_mov(s, type, dest, v1);
1770            }
1771        }
1772        /* Branch forward over one insn */
1773        tcg_out32(s, tcg_to_bc[cond] | 8);
1774        if (v2 == 0) {
1775            tcg_out_movi(s, type, dest, 0);
1776        } else {
1777            tcg_out_mov(s, type, dest, v2);
1778        }
1779    }
1780}
1781
1782static void tcg_out_cntxz(TCGContext *s, TCGType type, uint32_t opc,
1783                          TCGArg a0, TCGArg a1, TCGArg a2, bool const_a2)
1784{
1785    if (const_a2 && a2 == (type == TCG_TYPE_I32 ? 32 : 64)) {
1786        tcg_out32(s, opc | RA(a0) | RS(a1));
1787    } else {
1788        tcg_out_cmp(s, TCG_COND_EQ, a1, 0, 1, 7, type);
1789        /* Note that the only other valid constant for a2 is 0.  */
1790        if (have_isel) {
1791            tcg_out32(s, opc | RA(TCG_REG_R0) | RS(a1));
1792            tcg_out32(s, tcg_to_isel[TCG_COND_EQ] | TAB(a0, a2, TCG_REG_R0));
1793        } else if (!const_a2 && a0 == a2) {
1794            tcg_out32(s, tcg_to_bc[TCG_COND_EQ] | 8);
1795            tcg_out32(s, opc | RA(a0) | RS(a1));
1796        } else {
1797            tcg_out32(s, opc | RA(a0) | RS(a1));
1798            tcg_out32(s, tcg_to_bc[TCG_COND_NE] | 8);
1799            if (const_a2) {
1800                tcg_out_movi(s, type, a0, 0);
1801            } else {
1802                tcg_out_mov(s, type, a0, a2);
1803            }
1804        }
1805    }
1806}
1807
1808static void tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1809                         const int *const_args)
1810{
1811    static const struct { uint8_t bit1, bit2; } bits[] = {
1812        [TCG_COND_LT ] = { CR_LT, CR_LT },
1813        [TCG_COND_LE ] = { CR_LT, CR_GT },
1814        [TCG_COND_GT ] = { CR_GT, CR_GT },
1815        [TCG_COND_GE ] = { CR_GT, CR_LT },
1816        [TCG_COND_LTU] = { CR_LT, CR_LT },
1817        [TCG_COND_LEU] = { CR_LT, CR_GT },
1818        [TCG_COND_GTU] = { CR_GT, CR_GT },
1819        [TCG_COND_GEU] = { CR_GT, CR_LT },
1820    };
1821
1822    TCGCond cond = args[4], cond2;
1823    TCGArg al, ah, bl, bh;
1824    int blconst, bhconst;
1825    int op, bit1, bit2;
1826
1827    al = args[0];
1828    ah = args[1];
1829    bl = args[2];
1830    bh = args[3];
1831    blconst = const_args[2];
1832    bhconst = const_args[3];
1833
1834    switch (cond) {
1835    case TCG_COND_EQ:
1836        op = CRAND;
1837        goto do_equality;
1838    case TCG_COND_NE:
1839        op = CRNAND;
1840    do_equality:
1841        tcg_out_cmp(s, cond, al, bl, blconst, 6, TCG_TYPE_I32);
1842        tcg_out_cmp(s, cond, ah, bh, bhconst, 7, TCG_TYPE_I32);
1843        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1844        break;
1845
1846    case TCG_COND_LT:
1847    case TCG_COND_LE:
1848    case TCG_COND_GT:
1849    case TCG_COND_GE:
1850    case TCG_COND_LTU:
1851    case TCG_COND_LEU:
1852    case TCG_COND_GTU:
1853    case TCG_COND_GEU:
1854        bit1 = bits[cond].bit1;
1855        bit2 = bits[cond].bit2;
1856        op = (bit1 != bit2 ? CRANDC : CRAND);
1857        cond2 = tcg_unsigned_cond(cond);
1858
1859        tcg_out_cmp(s, cond, ah, bh, bhconst, 6, TCG_TYPE_I32);
1860        tcg_out_cmp(s, cond2, al, bl, blconst, 7, TCG_TYPE_I32);
1861        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, bit2));
1862        tcg_out32(s, CROR | BT(7, CR_EQ) | BA(6, bit1) | BB(7, CR_EQ));
1863        break;
1864
1865    default:
1866        g_assert_not_reached();
1867    }
1868}
1869
1870static void tcg_out_setcond2(TCGContext *s, const TCGArg *args,
1871                             const int *const_args)
1872{
1873    tcg_out_cmp2(s, args + 1, const_args + 1);
1874    tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1875    tcg_out_rlw(s, RLWINM, args[0], TCG_REG_R0, 31, 31, 31);
1876}
1877
1878static void tcg_out_brcond2 (TCGContext *s, const TCGArg *args,
1879                             const int *const_args)
1880{
1881    tcg_out_cmp2(s, args, const_args);
1882    tcg_out_bc(s, BC | BI(7, CR_EQ) | BO_COND_TRUE, arg_label(args[5]));
1883}
1884
1885static void tcg_out_mb(TCGContext *s, TCGArg a0)
1886{
1887    uint32_t insn;
1888
1889    if (a0 & TCG_MO_ST_LD) {
1890        insn = HWSYNC;
1891    } else {
1892        insn = LWSYNC;
1893    }
1894
1895    tcg_out32(s, insn);
1896}
1897
1898static void tcg_out_call_int(TCGContext *s, int lk,
1899                             const tcg_insn_unit *target)
1900{
1901#ifdef _CALL_AIX
1902    /* Look through the descriptor.  If the branch is in range, and we
1903       don't have to spend too much effort on building the toc.  */
1904    const void *tgt = ((const void * const *)target)[0];
1905    uintptr_t toc = ((const uintptr_t *)target)[1];
1906    intptr_t diff = tcg_pcrel_diff(s, tgt);
1907
1908    if (in_range_b(diff) && toc == (uint32_t)toc) {
1909        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, toc);
1910        tcg_out_b(s, lk, tgt);
1911    } else {
1912        /* Fold the low bits of the constant into the addresses below.  */
1913        intptr_t arg = (intptr_t)target;
1914        int ofs = (int16_t)arg;
1915
1916        if (ofs + 8 < 0x8000) {
1917            arg -= ofs;
1918        } else {
1919            ofs = 0;
1920        }
1921        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, arg);
1922        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_TMP1, ofs);
1923        tcg_out32(s, MTSPR | RA(TCG_REG_R0) | CTR);
1924        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_REG_TMP1, ofs + SZP);
1925        tcg_out32(s, BCCTR | BO_ALWAYS | lk);
1926    }
1927#elif defined(_CALL_ELF) && _CALL_ELF == 2
1928    intptr_t diff;
1929
1930    /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1931       address, which the callee uses to compute its TOC address.  */
1932    /* FIXME: when the branch is in range, we could avoid r12 load if we
1933       knew that the destination uses the same TOC, and what its local
1934       entry point offset is.  */
1935    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R12, (intptr_t)target);
1936
1937    diff = tcg_pcrel_diff(s, target);
1938    if (in_range_b(diff)) {
1939        tcg_out_b(s, lk, target);
1940    } else {
1941        tcg_out32(s, MTSPR | RS(TCG_REG_R12) | CTR);
1942        tcg_out32(s, BCCTR | BO_ALWAYS | lk);
1943    }
1944#else
1945    tcg_out_b(s, lk, target);
1946#endif
1947}
1948
1949static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target,
1950                         const TCGHelperInfo *info)
1951{
1952    tcg_out_call_int(s, LK, target);
1953}
1954
1955static const uint32_t qemu_ldx_opc[(MO_SSIZE + MO_BSWAP) + 1] = {
1956    [MO_UB] = LBZX,
1957    [MO_UW] = LHZX,
1958    [MO_UL] = LWZX,
1959    [MO_UQ] = LDX,
1960    [MO_SW] = LHAX,
1961    [MO_SL] = LWAX,
1962    [MO_BSWAP | MO_UB] = LBZX,
1963    [MO_BSWAP | MO_UW] = LHBRX,
1964    [MO_BSWAP | MO_UL] = LWBRX,
1965    [MO_BSWAP | MO_UQ] = LDBRX,
1966};
1967
1968static const uint32_t qemu_stx_opc[(MO_SIZE + MO_BSWAP) + 1] = {
1969    [MO_UB] = STBX,
1970    [MO_UW] = STHX,
1971    [MO_UL] = STWX,
1972    [MO_UQ] = STDX,
1973    [MO_BSWAP | MO_UB] = STBX,
1974    [MO_BSWAP | MO_UW] = STHBRX,
1975    [MO_BSWAP | MO_UL] = STWBRX,
1976    [MO_BSWAP | MO_UQ] = STDBRX,
1977};
1978
1979#if defined (CONFIG_SOFTMMU)
1980/* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
1981 *                                 int mmu_idx, uintptr_t ra)
1982 */
1983static void * const qemu_ld_helpers[(MO_SIZE | MO_BSWAP) + 1] = {
1984    [MO_UB]   = helper_ret_ldub_mmu,
1985    [MO_LEUW] = helper_le_lduw_mmu,
1986    [MO_LEUL] = helper_le_ldul_mmu,
1987    [MO_LEUQ] = helper_le_ldq_mmu,
1988    [MO_BEUW] = helper_be_lduw_mmu,
1989    [MO_BEUL] = helper_be_ldul_mmu,
1990    [MO_BEUQ] = helper_be_ldq_mmu,
1991};
1992
1993/* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
1994 *                                 uintxx_t val, int mmu_idx, uintptr_t ra)
1995 */
1996static void * const qemu_st_helpers[(MO_SIZE | MO_BSWAP) + 1] = {
1997    [MO_UB]   = helper_ret_stb_mmu,
1998    [MO_LEUW] = helper_le_stw_mmu,
1999    [MO_LEUL] = helper_le_stl_mmu,
2000    [MO_LEUQ] = helper_le_stq_mmu,
2001    [MO_BEUW] = helper_be_stw_mmu,
2002    [MO_BEUL] = helper_be_stl_mmu,
2003    [MO_BEUQ] = helper_be_stq_mmu,
2004};
2005
2006/* We expect to use a 16-bit negative offset from ENV.  */
2007QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
2008QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -32768);
2009
2010/* Perform the TLB load and compare.  Places the result of the comparison
2011   in CR7, loads the addend of the TLB into R3, and returns the register
2012   containing the guest address (zero-extended into R4).  Clobbers R0 and R2. */
2013
2014static TCGReg tcg_out_tlb_read(TCGContext *s, MemOp opc,
2015                               TCGReg addrlo, TCGReg addrhi,
2016                               int mem_index, bool is_read)
2017{
2018    int cmp_off
2019        = (is_read
2020           ? offsetof(CPUTLBEntry, addr_read)
2021           : offsetof(CPUTLBEntry, addr_write));
2022    int fast_off = TLB_MASK_TABLE_OFS(mem_index);
2023    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
2024    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
2025    unsigned s_bits = opc & MO_SIZE;
2026    unsigned a_bits = get_alignment_bits(opc);
2027
2028    /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx].  */
2029    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_AREG0, mask_off);
2030    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R4, TCG_AREG0, table_off);
2031
2032    /* Extract the page index, shifted into place for tlb index.  */
2033    if (TCG_TARGET_REG_BITS == 32) {
2034        tcg_out_shri32(s, TCG_REG_TMP1, addrlo,
2035                       TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
2036    } else {
2037        tcg_out_shri64(s, TCG_REG_TMP1, addrlo,
2038                       TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
2039    }
2040    tcg_out32(s, AND | SAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_TMP1));
2041
2042    /* Load the TLB comparator.  */
2043    if (cmp_off == 0 && TCG_TARGET_REG_BITS >= TARGET_LONG_BITS) {
2044        uint32_t lxu = (TCG_TARGET_REG_BITS == 32 || TARGET_LONG_BITS == 32
2045                        ? LWZUX : LDUX);
2046        tcg_out32(s, lxu | TAB(TCG_REG_TMP1, TCG_REG_R3, TCG_REG_R4));
2047    } else {
2048        tcg_out32(s, ADD | TAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_R4));
2049        if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2050            tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP1, TCG_REG_R3, cmp_off + 4);
2051            tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R4, TCG_REG_R3, cmp_off);
2052        } else {
2053            tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP1, TCG_REG_R3, cmp_off);
2054        }
2055    }
2056
2057    /* Load the TLB addend for use on the fast path.  Do this asap
2058       to minimize any load use delay.  */
2059    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_REG_R3,
2060               offsetof(CPUTLBEntry, addend));
2061
2062    /* Clear the non-page, non-alignment bits from the address */
2063    if (TCG_TARGET_REG_BITS == 32) {
2064        /* We don't support unaligned accesses on 32-bits.
2065         * Preserve the bottom bits and thus trigger a comparison
2066         * failure on unaligned accesses.
2067         */
2068        if (a_bits < s_bits) {
2069            a_bits = s_bits;
2070        }
2071        tcg_out_rlw(s, RLWINM, TCG_REG_R0, addrlo, 0,
2072                    (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
2073    } else {
2074        TCGReg t = addrlo;
2075
2076        /* If the access is unaligned, we need to make sure we fail if we
2077         * cross a page boundary.  The trick is to add the access size-1
2078         * to the address before masking the low bits.  That will make the
2079         * address overflow to the next page if we cross a page boundary,
2080         * which will then force a mismatch of the TLB compare.
2081         */
2082        if (a_bits < s_bits) {
2083            unsigned a_mask = (1 << a_bits) - 1;
2084            unsigned s_mask = (1 << s_bits) - 1;
2085            tcg_out32(s, ADDI | TAI(TCG_REG_R0, t, s_mask - a_mask));
2086            t = TCG_REG_R0;
2087        }
2088
2089        /* Mask the address for the requested alignment.  */
2090        if (TARGET_LONG_BITS == 32) {
2091            tcg_out_rlw(s, RLWINM, TCG_REG_R0, t, 0,
2092                        (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
2093            /* Zero-extend the address for use in the final address.  */
2094            tcg_out_ext32u(s, TCG_REG_R4, addrlo);
2095            addrlo = TCG_REG_R4;
2096        } else if (a_bits == 0) {
2097            tcg_out_rld(s, RLDICR, TCG_REG_R0, t, 0, 63 - TARGET_PAGE_BITS);
2098        } else {
2099            tcg_out_rld(s, RLDICL, TCG_REG_R0, t,
2100                        64 - TARGET_PAGE_BITS, TARGET_PAGE_BITS - a_bits);
2101            tcg_out_rld(s, RLDICL, TCG_REG_R0, TCG_REG_R0, TARGET_PAGE_BITS, 0);
2102        }
2103    }
2104
2105    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2106        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
2107                    0, 7, TCG_TYPE_I32);
2108        tcg_out_cmp(s, TCG_COND_EQ, addrhi, TCG_REG_R4, 0, 6, TCG_TYPE_I32);
2109        tcg_out32(s, CRAND | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
2110    } else {
2111        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
2112                    0, 7, TCG_TYPE_TL);
2113    }
2114
2115    return addrlo;
2116}
2117
2118/* Record the context of a call to the out of line helper code for the slow
2119   path for a load or store, so that we can later generate the correct
2120   helper code.  */
2121static void add_qemu_ldst_label(TCGContext *s, bool is_ld, MemOpIdx oi,
2122                                TCGReg datalo_reg, TCGReg datahi_reg,
2123                                TCGReg addrlo_reg, TCGReg addrhi_reg,
2124                                tcg_insn_unit *raddr, tcg_insn_unit *lptr)
2125{
2126    TCGLabelQemuLdst *label = new_ldst_label(s);
2127
2128    label->is_ld = is_ld;
2129    label->oi = oi;
2130    label->datalo_reg = datalo_reg;
2131    label->datahi_reg = datahi_reg;
2132    label->addrlo_reg = addrlo_reg;
2133    label->addrhi_reg = addrhi_reg;
2134    label->raddr = tcg_splitwx_to_rx(raddr);
2135    label->label_ptr[0] = lptr;
2136}
2137
2138static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2139{
2140    MemOpIdx oi = lb->oi;
2141    MemOp opc = get_memop(oi);
2142    TCGReg hi, lo, arg = TCG_REG_R3;
2143
2144    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2145        return false;
2146    }
2147
2148    tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
2149
2150    lo = lb->addrlo_reg;
2151    hi = lb->addrhi_reg;
2152    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2153        arg |= (TCG_TARGET_CALL_ARG_I64 == TCG_CALL_ARG_EVEN);
2154        tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2155        tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2156    } else {
2157        /* If the address needed to be zero-extended, we'll have already
2158           placed it in R4.  The only remaining case is 64-bit guest.  */
2159        tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
2160    }
2161
2162    tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
2163    tcg_out32(s, MFSPR | RT(arg) | LR);
2164
2165    tcg_out_call_int(s, LK, qemu_ld_helpers[opc & (MO_BSWAP | MO_SIZE)]);
2166
2167    lo = lb->datalo_reg;
2168    hi = lb->datahi_reg;
2169    if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
2170        tcg_out_mov(s, TCG_TYPE_I32, lo, TCG_REG_R4);
2171        tcg_out_mov(s, TCG_TYPE_I32, hi, TCG_REG_R3);
2172    } else {
2173        tcg_out_movext(s, lb->type, lo,
2174                       TCG_TYPE_REG, opc & MO_SSIZE, TCG_REG_R3);
2175    }
2176
2177    tcg_out_b(s, 0, lb->raddr);
2178    return true;
2179}
2180
2181static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2182{
2183    MemOpIdx oi = lb->oi;
2184    MemOp opc = get_memop(oi);
2185    MemOp s_bits = opc & MO_SIZE;
2186    TCGReg hi, lo, arg = TCG_REG_R3;
2187
2188    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2189        return false;
2190    }
2191
2192    tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
2193
2194    lo = lb->addrlo_reg;
2195    hi = lb->addrhi_reg;
2196    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2197        arg |= (TCG_TARGET_CALL_ARG_I64 == TCG_CALL_ARG_EVEN);
2198        tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2199        tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2200    } else {
2201        /* If the address needed to be zero-extended, we'll have already
2202           placed it in R4.  The only remaining case is 64-bit guest.  */
2203        tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
2204    }
2205
2206    lo = lb->datalo_reg;
2207    hi = lb->datahi_reg;
2208    if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
2209        arg |= (TCG_TARGET_CALL_ARG_I64 == TCG_CALL_ARG_EVEN);
2210        tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2211        tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2212    } else {
2213        tcg_out_movext(s, s_bits == MO_64 ? TCG_TYPE_I64 : TCG_TYPE_I32,
2214                       arg++, lb->type, s_bits, lo);
2215    }
2216
2217    tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
2218    tcg_out32(s, MFSPR | RT(arg) | LR);
2219
2220    tcg_out_call_int(s, LK, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
2221
2222    tcg_out_b(s, 0, lb->raddr);
2223    return true;
2224}
2225#else
2226
2227static void tcg_out_test_alignment(TCGContext *s, bool is_ld, TCGReg addrlo,
2228                                   TCGReg addrhi, unsigned a_bits)
2229{
2230    unsigned a_mask = (1 << a_bits) - 1;
2231    TCGLabelQemuLdst *label = new_ldst_label(s);
2232
2233    label->is_ld = is_ld;
2234    label->addrlo_reg = addrlo;
2235    label->addrhi_reg = addrhi;
2236
2237    /* We are expecting a_bits to max out at 7, much lower than ANDI. */
2238    tcg_debug_assert(a_bits < 16);
2239    tcg_out32(s, ANDI | SAI(addrlo, TCG_REG_R0, a_mask));
2240
2241    label->label_ptr[0] = s->code_ptr;
2242    tcg_out32(s, BC | BI(0, CR_EQ) | BO_COND_FALSE | LK);
2243
2244    label->raddr = tcg_splitwx_to_rx(s->code_ptr);
2245}
2246
2247static bool tcg_out_fail_alignment(TCGContext *s, TCGLabelQemuLdst *l)
2248{
2249    if (!reloc_pc14(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2250        return false;
2251    }
2252
2253    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2254        TCGReg arg = TCG_REG_R4;
2255
2256        arg |= (TCG_TARGET_CALL_ARG_I64 == TCG_CALL_ARG_EVEN);
2257        if (l->addrlo_reg != arg) {
2258            tcg_out_mov(s, TCG_TYPE_I32, arg, l->addrhi_reg);
2259            tcg_out_mov(s, TCG_TYPE_I32, arg + 1, l->addrlo_reg);
2260        } else if (l->addrhi_reg != arg + 1) {
2261            tcg_out_mov(s, TCG_TYPE_I32, arg + 1, l->addrlo_reg);
2262            tcg_out_mov(s, TCG_TYPE_I32, arg, l->addrhi_reg);
2263        } else {
2264            tcg_out_mov(s, TCG_TYPE_I32, TCG_REG_R0, arg);
2265            tcg_out_mov(s, TCG_TYPE_I32, arg, arg + 1);
2266            tcg_out_mov(s, TCG_TYPE_I32, arg + 1, TCG_REG_R0);
2267        }
2268    } else {
2269        tcg_out_mov(s, TCG_TYPE_TL, TCG_REG_R4, l->addrlo_reg);
2270    }
2271    tcg_out_mov(s, TCG_TYPE_TL, TCG_REG_R3, TCG_AREG0);
2272
2273    /* "Tail call" to the helper, with the return address back inline. */
2274    tcg_out_call_int(s, 0, (const void *)(l->is_ld ? helper_unaligned_ld
2275                                          : helper_unaligned_st));
2276    return true;
2277}
2278
2279static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
2280{
2281    return tcg_out_fail_alignment(s, l);
2282}
2283
2284static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
2285{
2286    return tcg_out_fail_alignment(s, l);
2287}
2288
2289#endif /* SOFTMMU */
2290
2291static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64)
2292{
2293    TCGReg datalo, datahi, addrlo, rbase;
2294    TCGReg addrhi __attribute__((unused));
2295    MemOpIdx oi;
2296    MemOp opc, s_bits;
2297#ifdef CONFIG_SOFTMMU
2298    int mem_index;
2299    tcg_insn_unit *label_ptr;
2300#else
2301    unsigned a_bits;
2302#endif
2303
2304    datalo = *args++;
2305    datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
2306    addrlo = *args++;
2307    addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
2308    oi = *args++;
2309    opc = get_memop(oi);
2310    s_bits = opc & MO_SIZE;
2311
2312#ifdef CONFIG_SOFTMMU
2313    mem_index = get_mmuidx(oi);
2314    addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, true);
2315
2316    /* Load a pointer into the current opcode w/conditional branch-link. */
2317    label_ptr = s->code_ptr;
2318    tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
2319
2320    rbase = TCG_REG_R3;
2321#else  /* !CONFIG_SOFTMMU */
2322    a_bits = get_alignment_bits(opc);
2323    if (a_bits) {
2324        tcg_out_test_alignment(s, true, addrlo, addrhi, a_bits);
2325    }
2326    rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
2327    if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
2328        tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
2329        addrlo = TCG_REG_TMP1;
2330    }
2331#endif
2332
2333    if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
2334        if (opc & MO_BSWAP) {
2335            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2336            tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
2337            tcg_out32(s, LWBRX | TAB(datahi, rbase, TCG_REG_R0));
2338        } else if (rbase != 0) {
2339            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2340            tcg_out32(s, LWZX | TAB(datahi, rbase, addrlo));
2341            tcg_out32(s, LWZX | TAB(datalo, rbase, TCG_REG_R0));
2342        } else if (addrlo == datahi) {
2343            tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
2344            tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
2345        } else {
2346            tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
2347            tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
2348        }
2349    } else {
2350        uint32_t insn = qemu_ldx_opc[opc & (MO_BSWAP | MO_SSIZE)];
2351        if (!have_isa_2_06 && insn == LDBRX) {
2352            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2353            tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
2354            tcg_out32(s, LWBRX | TAB(TCG_REG_R0, rbase, TCG_REG_R0));
2355            tcg_out_rld(s, RLDIMI, datalo, TCG_REG_R0, 32, 0);
2356        } else if (insn) {
2357            tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
2358        } else {
2359            insn = qemu_ldx_opc[opc & (MO_SIZE | MO_BSWAP)];
2360            tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
2361            tcg_out_movext(s, TCG_TYPE_REG, datalo,
2362                           TCG_TYPE_REG, opc & MO_SSIZE, datalo);
2363        }
2364    }
2365
2366#ifdef CONFIG_SOFTMMU
2367    add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
2368                        s->code_ptr, label_ptr);
2369#endif
2370}
2371
2372static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
2373{
2374    TCGReg datalo, datahi, addrlo, rbase;
2375    TCGReg addrhi __attribute__((unused));
2376    MemOpIdx oi;
2377    MemOp opc, s_bits;
2378#ifdef CONFIG_SOFTMMU
2379    int mem_index;
2380    tcg_insn_unit *label_ptr;
2381#else
2382    unsigned a_bits;
2383#endif
2384
2385    datalo = *args++;
2386    datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
2387    addrlo = *args++;
2388    addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
2389    oi = *args++;
2390    opc = get_memop(oi);
2391    s_bits = opc & MO_SIZE;
2392
2393#ifdef CONFIG_SOFTMMU
2394    mem_index = get_mmuidx(oi);
2395    addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, false);
2396
2397    /* Load a pointer into the current opcode w/conditional branch-link. */
2398    label_ptr = s->code_ptr;
2399    tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
2400
2401    rbase = TCG_REG_R3;
2402#else  /* !CONFIG_SOFTMMU */
2403    a_bits = get_alignment_bits(opc);
2404    if (a_bits) {
2405        tcg_out_test_alignment(s, false, addrlo, addrhi, a_bits);
2406    }
2407    rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
2408    if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
2409        tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
2410        addrlo = TCG_REG_TMP1;
2411    }
2412#endif
2413
2414    if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
2415        if (opc & MO_BSWAP) {
2416            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2417            tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
2418            tcg_out32(s, STWBRX | SAB(datahi, rbase, TCG_REG_R0));
2419        } else if (rbase != 0) {
2420            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2421            tcg_out32(s, STWX | SAB(datahi, rbase, addrlo));
2422            tcg_out32(s, STWX | SAB(datalo, rbase, TCG_REG_R0));
2423        } else {
2424            tcg_out32(s, STW | TAI(datahi, addrlo, 0));
2425            tcg_out32(s, STW | TAI(datalo, addrlo, 4));
2426        }
2427    } else {
2428        uint32_t insn = qemu_stx_opc[opc & (MO_BSWAP | MO_SIZE)];
2429        if (!have_isa_2_06 && insn == STDBRX) {
2430            tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
2431            tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, addrlo, 4));
2432            tcg_out_shri64(s, TCG_REG_R0, datalo, 32);
2433            tcg_out32(s, STWBRX | SAB(TCG_REG_R0, rbase, TCG_REG_TMP1));
2434        } else {
2435            tcg_out32(s, insn | SAB(datalo, rbase, addrlo));
2436        }
2437    }
2438
2439#ifdef CONFIG_SOFTMMU
2440    add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
2441                        s->code_ptr, label_ptr);
2442#endif
2443}
2444
2445static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
2446{
2447    int i;
2448    for (i = 0; i < count; ++i) {
2449        p[i] = NOP;
2450    }
2451}
2452
2453/* Parameters for function call generation, used in tcg.c.  */
2454#define TCG_TARGET_STACK_ALIGN       16
2455
2456#ifdef _CALL_AIX
2457# define LINK_AREA_SIZE                (6 * SZR)
2458# define LR_OFFSET                     (1 * SZR)
2459# define TCG_TARGET_CALL_STACK_OFFSET  (LINK_AREA_SIZE + 8 * SZR)
2460#elif defined(_CALL_DARWIN)
2461# define LINK_AREA_SIZE                (6 * SZR)
2462# define LR_OFFSET                     (2 * SZR)
2463#elif TCG_TARGET_REG_BITS == 64
2464# if defined(_CALL_ELF) && _CALL_ELF == 2
2465#  define LINK_AREA_SIZE               (4 * SZR)
2466#  define LR_OFFSET                    (1 * SZR)
2467# endif
2468#else /* TCG_TARGET_REG_BITS == 32 */
2469# if defined(_CALL_SYSV)
2470#  define LINK_AREA_SIZE               (2 * SZR)
2471#  define LR_OFFSET                    (1 * SZR)
2472# endif
2473#endif
2474#ifndef LR_OFFSET
2475# error "Unhandled abi"
2476#endif
2477#ifndef TCG_TARGET_CALL_STACK_OFFSET
2478# define TCG_TARGET_CALL_STACK_OFFSET  LINK_AREA_SIZE
2479#endif
2480
2481#define CPU_TEMP_BUF_SIZE  (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
2482#define REG_SAVE_SIZE      ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
2483
2484#define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET   \
2485                     + TCG_STATIC_CALL_ARGS_SIZE    \
2486                     + CPU_TEMP_BUF_SIZE            \
2487                     + REG_SAVE_SIZE                \
2488                     + TCG_TARGET_STACK_ALIGN - 1)  \
2489                    & -TCG_TARGET_STACK_ALIGN)
2490
2491#define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
2492
2493static void tcg_target_qemu_prologue(TCGContext *s)
2494{
2495    int i;
2496
2497#ifdef _CALL_AIX
2498    const void **desc = (const void **)s->code_ptr;
2499    desc[0] = tcg_splitwx_to_rx(desc + 2);  /* entry point */
2500    desc[1] = 0;                            /* environment pointer */
2501    s->code_ptr = (void *)(desc + 2);       /* skip over descriptor */
2502#endif
2503
2504    tcg_set_frame(s, TCG_REG_CALL_STACK, REG_SAVE_BOT - CPU_TEMP_BUF_SIZE,
2505                  CPU_TEMP_BUF_SIZE);
2506
2507    /* Prologue */
2508    tcg_out32(s, MFSPR | RT(TCG_REG_R0) | LR);
2509    tcg_out32(s, (SZR == 8 ? STDU : STWU)
2510              | SAI(TCG_REG_R1, TCG_REG_R1, -FRAME_SIZE));
2511
2512    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2513        tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2514                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2515    }
2516    tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2517
2518#ifndef CONFIG_SOFTMMU
2519    if (guest_base) {
2520        tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true);
2521        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2522    }
2523#endif
2524
2525    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2526    tcg_out32(s, MTSPR | RS(tcg_target_call_iarg_regs[1]) | CTR);
2527    if (USE_REG_TB) {
2528        tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, tcg_target_call_iarg_regs[1]);
2529    }
2530    tcg_out32(s, BCCTR | BO_ALWAYS);
2531
2532    /* Epilogue */
2533    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
2534
2535    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2536    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2537        tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2538                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2539    }
2540    tcg_out32(s, MTSPR | RS(TCG_REG_R0) | LR);
2541    tcg_out32(s, ADDI | TAI(TCG_REG_R1, TCG_REG_R1, FRAME_SIZE));
2542    tcg_out32(s, BCLR | BO_ALWAYS);
2543}
2544
2545static void tcg_out_exit_tb(TCGContext *s, uintptr_t arg)
2546{
2547    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, arg);
2548    tcg_out_b(s, 0, tcg_code_gen_epilogue);
2549}
2550
2551static void tcg_out_goto_tb(TCGContext *s, int which)
2552{
2553    uintptr_t ptr = get_jmp_target_addr(s, which);
2554
2555    if (USE_REG_TB) {
2556        ptrdiff_t offset = tcg_tbrel_diff(s, (void *)ptr);
2557        tcg_out_mem_long(s, LD, LDX, TCG_REG_TB, TCG_REG_TB, offset);
2558
2559        /* Direct branch will be patched by tb_target_set_jmp_target. */
2560        set_jmp_insn_offset(s, which);
2561        tcg_out32(s, MTSPR | RS(TCG_REG_TB) | CTR);
2562
2563        /* When branch is out of range, fall through to indirect. */
2564        tcg_out32(s, BCCTR | BO_ALWAYS);
2565
2566        /* For the unlinked case, need to reset TCG_REG_TB.  */
2567        set_jmp_reset_offset(s, which);
2568        tcg_out_mem_long(s, ADDI, ADD, TCG_REG_TB, TCG_REG_TB,
2569                         -tcg_current_code_size(s));
2570    } else {
2571        /* Direct branch will be patched by tb_target_set_jmp_target. */
2572        set_jmp_insn_offset(s, which);
2573        tcg_out32(s, NOP);
2574
2575        /* When branch is out of range, fall through to indirect. */
2576        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, ptr - (int16_t)ptr);
2577        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1, (int16_t)ptr);
2578        tcg_out32(s, MTSPR | RS(TCG_REG_TMP1) | CTR);
2579        tcg_out32(s, BCCTR | BO_ALWAYS);
2580        set_jmp_reset_offset(s, which);
2581    }
2582}
2583
2584void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
2585                              uintptr_t jmp_rx, uintptr_t jmp_rw)
2586{
2587    uintptr_t addr = tb->jmp_target_addr[n];
2588    intptr_t diff = addr - jmp_rx;
2589    tcg_insn_unit insn;
2590
2591    if (in_range_b(diff)) {
2592        insn = B | (diff & 0x3fffffc);
2593    } else if (USE_REG_TB) {
2594        insn = MTSPR | RS(TCG_REG_TB) | CTR;
2595    } else {
2596        insn = NOP;
2597    }
2598
2599    qatomic_set((uint32_t *)jmp_rw, insn);
2600    flush_idcache_range(jmp_rx, jmp_rw, 4);
2601}
2602
2603static void tcg_out_op(TCGContext *s, TCGOpcode opc,
2604                       const TCGArg args[TCG_MAX_OP_ARGS],
2605                       const int const_args[TCG_MAX_OP_ARGS])
2606{
2607    TCGArg a0, a1, a2;
2608
2609    switch (opc) {
2610    case INDEX_op_goto_ptr:
2611        tcg_out32(s, MTSPR | RS(args[0]) | CTR);
2612        if (USE_REG_TB) {
2613            tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, args[0]);
2614        }
2615        tcg_out32(s, ADDI | TAI(TCG_REG_R3, 0, 0));
2616        tcg_out32(s, BCCTR | BO_ALWAYS);
2617        break;
2618    case INDEX_op_br:
2619        {
2620            TCGLabel *l = arg_label(args[0]);
2621            uint32_t insn = B;
2622
2623            if (l->has_value) {
2624                insn |= reloc_pc24_val(tcg_splitwx_to_rx(s->code_ptr),
2625                                       l->u.value_ptr);
2626            } else {
2627                tcg_out_reloc(s, s->code_ptr, R_PPC_REL24, l, 0);
2628            }
2629            tcg_out32(s, insn);
2630        }
2631        break;
2632    case INDEX_op_ld8u_i32:
2633    case INDEX_op_ld8u_i64:
2634        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2635        break;
2636    case INDEX_op_ld8s_i32:
2637    case INDEX_op_ld8s_i64:
2638        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2639        tcg_out_ext8s(s, TCG_TYPE_REG, args[0], args[0]);
2640        break;
2641    case INDEX_op_ld16u_i32:
2642    case INDEX_op_ld16u_i64:
2643        tcg_out_mem_long(s, LHZ, LHZX, args[0], args[1], args[2]);
2644        break;
2645    case INDEX_op_ld16s_i32:
2646    case INDEX_op_ld16s_i64:
2647        tcg_out_mem_long(s, LHA, LHAX, args[0], args[1], args[2]);
2648        break;
2649    case INDEX_op_ld_i32:
2650    case INDEX_op_ld32u_i64:
2651        tcg_out_mem_long(s, LWZ, LWZX, args[0], args[1], args[2]);
2652        break;
2653    case INDEX_op_ld32s_i64:
2654        tcg_out_mem_long(s, LWA, LWAX, args[0], args[1], args[2]);
2655        break;
2656    case INDEX_op_ld_i64:
2657        tcg_out_mem_long(s, LD, LDX, args[0], args[1], args[2]);
2658        break;
2659    case INDEX_op_st8_i32:
2660    case INDEX_op_st8_i64:
2661        tcg_out_mem_long(s, STB, STBX, args[0], args[1], args[2]);
2662        break;
2663    case INDEX_op_st16_i32:
2664    case INDEX_op_st16_i64:
2665        tcg_out_mem_long(s, STH, STHX, args[0], args[1], args[2]);
2666        break;
2667    case INDEX_op_st_i32:
2668    case INDEX_op_st32_i64:
2669        tcg_out_mem_long(s, STW, STWX, args[0], args[1], args[2]);
2670        break;
2671    case INDEX_op_st_i64:
2672        tcg_out_mem_long(s, STD, STDX, args[0], args[1], args[2]);
2673        break;
2674
2675    case INDEX_op_add_i32:
2676        a0 = args[0], a1 = args[1], a2 = args[2];
2677        if (const_args[2]) {
2678        do_addi_32:
2679            tcg_out_mem_long(s, ADDI, ADD, a0, a1, (int32_t)a2);
2680        } else {
2681            tcg_out32(s, ADD | TAB(a0, a1, a2));
2682        }
2683        break;
2684    case INDEX_op_sub_i32:
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_I32, 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_32;
2695        } else {
2696            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2697        }
2698        break;
2699
2700    case INDEX_op_and_i32:
2701        a0 = args[0], a1 = args[1], a2 = args[2];
2702        if (const_args[2]) {
2703            tcg_out_andi32(s, a0, a1, a2);
2704        } else {
2705            tcg_out32(s, AND | SAB(a1, a0, a2));
2706        }
2707        break;
2708    case INDEX_op_and_i64:
2709        a0 = args[0], a1 = args[1], a2 = args[2];
2710        if (const_args[2]) {
2711            tcg_out_andi64(s, a0, a1, a2);
2712        } else {
2713            tcg_out32(s, AND | SAB(a1, a0, a2));
2714        }
2715        break;
2716    case INDEX_op_or_i64:
2717    case INDEX_op_or_i32:
2718        a0 = args[0], a1 = args[1], a2 = args[2];
2719        if (const_args[2]) {
2720            tcg_out_ori32(s, a0, a1, a2);
2721        } else {
2722            tcg_out32(s, OR | SAB(a1, a0, a2));
2723        }
2724        break;
2725    case INDEX_op_xor_i64:
2726    case INDEX_op_xor_i32:
2727        a0 = args[0], a1 = args[1], a2 = args[2];
2728        if (const_args[2]) {
2729            tcg_out_xori32(s, a0, a1, a2);
2730        } else {
2731            tcg_out32(s, XOR | SAB(a1, a0, a2));
2732        }
2733        break;
2734    case INDEX_op_andc_i32:
2735        a0 = args[0], a1 = args[1], a2 = args[2];
2736        if (const_args[2]) {
2737            tcg_out_andi32(s, a0, a1, ~a2);
2738        } else {
2739            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2740        }
2741        break;
2742    case INDEX_op_andc_i64:
2743        a0 = args[0], a1 = args[1], a2 = args[2];
2744        if (const_args[2]) {
2745            tcg_out_andi64(s, a0, a1, ~a2);
2746        } else {
2747            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2748        }
2749        break;
2750    case INDEX_op_orc_i32:
2751        if (const_args[2]) {
2752            tcg_out_ori32(s, args[0], args[1], ~args[2]);
2753            break;
2754        }
2755        /* FALLTHRU */
2756    case INDEX_op_orc_i64:
2757        tcg_out32(s, ORC | SAB(args[1], args[0], args[2]));
2758        break;
2759    case INDEX_op_eqv_i32:
2760        if (const_args[2]) {
2761            tcg_out_xori32(s, args[0], args[1], ~args[2]);
2762            break;
2763        }
2764        /* FALLTHRU */
2765    case INDEX_op_eqv_i64:
2766        tcg_out32(s, EQV | SAB(args[1], args[0], args[2]));
2767        break;
2768    case INDEX_op_nand_i32:
2769    case INDEX_op_nand_i64:
2770        tcg_out32(s, NAND | SAB(args[1], args[0], args[2]));
2771        break;
2772    case INDEX_op_nor_i32:
2773    case INDEX_op_nor_i64:
2774        tcg_out32(s, NOR | SAB(args[1], args[0], args[2]));
2775        break;
2776
2777    case INDEX_op_clz_i32:
2778        tcg_out_cntxz(s, TCG_TYPE_I32, CNTLZW, args[0], args[1],
2779                      args[2], const_args[2]);
2780        break;
2781    case INDEX_op_ctz_i32:
2782        tcg_out_cntxz(s, TCG_TYPE_I32, CNTTZW, args[0], args[1],
2783                      args[2], const_args[2]);
2784        break;
2785    case INDEX_op_ctpop_i32:
2786        tcg_out32(s, CNTPOPW | SAB(args[1], args[0], 0));
2787        break;
2788
2789    case INDEX_op_clz_i64:
2790        tcg_out_cntxz(s, TCG_TYPE_I64, CNTLZD, args[0], args[1],
2791                      args[2], const_args[2]);
2792        break;
2793    case INDEX_op_ctz_i64:
2794        tcg_out_cntxz(s, TCG_TYPE_I64, CNTTZD, args[0], args[1],
2795                      args[2], const_args[2]);
2796        break;
2797    case INDEX_op_ctpop_i64:
2798        tcg_out32(s, CNTPOPD | SAB(args[1], args[0], 0));
2799        break;
2800
2801    case INDEX_op_mul_i32:
2802        a0 = args[0], a1 = args[1], a2 = args[2];
2803        if (const_args[2]) {
2804            tcg_out32(s, MULLI | TAI(a0, a1, a2));
2805        } else {
2806            tcg_out32(s, MULLW | TAB(a0, a1, a2));
2807        }
2808        break;
2809
2810    case INDEX_op_div_i32:
2811        tcg_out32(s, DIVW | TAB(args[0], args[1], args[2]));
2812        break;
2813
2814    case INDEX_op_divu_i32:
2815        tcg_out32(s, DIVWU | TAB(args[0], args[1], args[2]));
2816        break;
2817
2818    case INDEX_op_rem_i32:
2819        tcg_out32(s, MODSW | TAB(args[0], args[1], args[2]));
2820        break;
2821
2822    case INDEX_op_remu_i32:
2823        tcg_out32(s, MODUW | TAB(args[0], args[1], args[2]));
2824        break;
2825
2826    case INDEX_op_shl_i32:
2827        if (const_args[2]) {
2828            /* Limit immediate shift count lest we create an illegal insn.  */
2829            tcg_out_shli32(s, args[0], args[1], args[2] & 31);
2830        } else {
2831            tcg_out32(s, SLW | SAB(args[1], args[0], args[2]));
2832        }
2833        break;
2834    case INDEX_op_shr_i32:
2835        if (const_args[2]) {
2836            /* Limit immediate shift count lest we create an illegal insn.  */
2837            tcg_out_shri32(s, args[0], args[1], args[2] & 31);
2838        } else {
2839            tcg_out32(s, SRW | SAB(args[1], args[0], args[2]));
2840        }
2841        break;
2842    case INDEX_op_sar_i32:
2843        if (const_args[2]) {
2844            tcg_out_sari32(s, args[0], args[1], args[2]);
2845        } else {
2846            tcg_out32(s, SRAW | SAB(args[1], args[0], args[2]));
2847        }
2848        break;
2849    case INDEX_op_rotl_i32:
2850        if (const_args[2]) {
2851            tcg_out_rlw(s, RLWINM, args[0], args[1], args[2], 0, 31);
2852        } else {
2853            tcg_out32(s, RLWNM | SAB(args[1], args[0], args[2])
2854                         | MB(0) | ME(31));
2855        }
2856        break;
2857    case INDEX_op_rotr_i32:
2858        if (const_args[2]) {
2859            tcg_out_rlw(s, RLWINM, args[0], args[1], 32 - args[2], 0, 31);
2860        } else {
2861            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 32));
2862            tcg_out32(s, RLWNM | SAB(args[1], args[0], TCG_REG_R0)
2863                         | MB(0) | ME(31));
2864        }
2865        break;
2866
2867    case INDEX_op_brcond_i32:
2868        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2869                       arg_label(args[3]), TCG_TYPE_I32);
2870        break;
2871    case INDEX_op_brcond_i64:
2872        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2873                       arg_label(args[3]), TCG_TYPE_I64);
2874        break;
2875    case INDEX_op_brcond2_i32:
2876        tcg_out_brcond2(s, args, const_args);
2877        break;
2878
2879    case INDEX_op_neg_i32:
2880    case INDEX_op_neg_i64:
2881        tcg_out32(s, NEG | RT(args[0]) | RA(args[1]));
2882        break;
2883
2884    case INDEX_op_not_i32:
2885    case INDEX_op_not_i64:
2886        tcg_out32(s, NOR | SAB(args[1], args[0], args[1]));
2887        break;
2888
2889    case INDEX_op_add_i64:
2890        a0 = args[0], a1 = args[1], a2 = args[2];
2891        if (const_args[2]) {
2892        do_addi_64:
2893            tcg_out_mem_long(s, ADDI, ADD, a0, a1, a2);
2894        } else {
2895            tcg_out32(s, ADD | TAB(a0, a1, a2));
2896        }
2897        break;
2898    case INDEX_op_sub_i64:
2899        a0 = args[0], a1 = args[1], a2 = args[2];
2900        if (const_args[1]) {
2901            if (const_args[2]) {
2902                tcg_out_movi(s, TCG_TYPE_I64, a0, a1 - a2);
2903            } else {
2904                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2905            }
2906        } else if (const_args[2]) {
2907            a2 = -a2;
2908            goto do_addi_64;
2909        } else {
2910            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2911        }
2912        break;
2913
2914    case INDEX_op_shl_i64:
2915        if (const_args[2]) {
2916            /* Limit immediate shift count lest we create an illegal insn.  */
2917            tcg_out_shli64(s, args[0], args[1], args[2] & 63);
2918        } else {
2919            tcg_out32(s, SLD | SAB(args[1], args[0], args[2]));
2920        }
2921        break;
2922    case INDEX_op_shr_i64:
2923        if (const_args[2]) {
2924            /* Limit immediate shift count lest we create an illegal insn.  */
2925            tcg_out_shri64(s, args[0], args[1], args[2] & 63);
2926        } else {
2927            tcg_out32(s, SRD | SAB(args[1], args[0], args[2]));
2928        }
2929        break;
2930    case INDEX_op_sar_i64:
2931        if (const_args[2]) {
2932            tcg_out_sari64(s, args[0], args[1], args[2]);
2933        } else {
2934            tcg_out32(s, SRAD | SAB(args[1], args[0], args[2]));
2935        }
2936        break;
2937    case INDEX_op_rotl_i64:
2938        if (const_args[2]) {
2939            tcg_out_rld(s, RLDICL, args[0], args[1], args[2], 0);
2940        } else {
2941            tcg_out32(s, RLDCL | SAB(args[1], args[0], args[2]) | MB64(0));
2942        }
2943        break;
2944    case INDEX_op_rotr_i64:
2945        if (const_args[2]) {
2946            tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 0);
2947        } else {
2948            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 64));
2949            tcg_out32(s, RLDCL | SAB(args[1], args[0], TCG_REG_R0) | MB64(0));
2950        }
2951        break;
2952
2953    case INDEX_op_mul_i64:
2954        a0 = args[0], a1 = args[1], a2 = args[2];
2955        if (const_args[2]) {
2956            tcg_out32(s, MULLI | TAI(a0, a1, a2));
2957        } else {
2958            tcg_out32(s, MULLD | TAB(a0, a1, a2));
2959        }
2960        break;
2961    case INDEX_op_div_i64:
2962        tcg_out32(s, DIVD | TAB(args[0], args[1], args[2]));
2963        break;
2964    case INDEX_op_divu_i64:
2965        tcg_out32(s, DIVDU | TAB(args[0], args[1], args[2]));
2966        break;
2967    case INDEX_op_rem_i64:
2968        tcg_out32(s, MODSD | TAB(args[0], args[1], args[2]));
2969        break;
2970    case INDEX_op_remu_i64:
2971        tcg_out32(s, MODUD | TAB(args[0], args[1], args[2]));
2972        break;
2973
2974    case INDEX_op_qemu_ld_i32:
2975        tcg_out_qemu_ld(s, args, false);
2976        break;
2977    case INDEX_op_qemu_ld_i64:
2978        tcg_out_qemu_ld(s, args, true);
2979        break;
2980    case INDEX_op_qemu_st_i32:
2981        tcg_out_qemu_st(s, args, false);
2982        break;
2983    case INDEX_op_qemu_st_i64:
2984        tcg_out_qemu_st(s, args, true);
2985        break;
2986
2987    case INDEX_op_setcond_i32:
2988        tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2],
2989                        const_args[2]);
2990        break;
2991    case INDEX_op_setcond_i64:
2992        tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2],
2993                        const_args[2]);
2994        break;
2995    case INDEX_op_setcond2_i32:
2996        tcg_out_setcond2(s, args, const_args);
2997        break;
2998
2999    case INDEX_op_bswap16_i32:
3000    case INDEX_op_bswap16_i64:
3001        tcg_out_bswap16(s, args[0], args[1], args[2]);
3002        break;
3003    case INDEX_op_bswap32_i32:
3004        tcg_out_bswap32(s, args[0], args[1], 0);
3005        break;
3006    case INDEX_op_bswap32_i64:
3007        tcg_out_bswap32(s, args[0], args[1], args[2]);
3008        break;
3009    case INDEX_op_bswap64_i64:
3010        tcg_out_bswap64(s, args[0], args[1]);
3011        break;
3012
3013    case INDEX_op_deposit_i32:
3014        if (const_args[2]) {
3015            uint32_t mask = ((2u << (args[4] - 1)) - 1) << args[3];
3016            tcg_out_andi32(s, args[0], args[0], ~mask);
3017        } else {
3018            tcg_out_rlw(s, RLWIMI, args[0], args[2], args[3],
3019                        32 - args[3] - args[4], 31 - args[3]);
3020        }
3021        break;
3022    case INDEX_op_deposit_i64:
3023        if (const_args[2]) {
3024            uint64_t mask = ((2ull << (args[4] - 1)) - 1) << args[3];
3025            tcg_out_andi64(s, args[0], args[0], ~mask);
3026        } else {
3027            tcg_out_rld(s, RLDIMI, args[0], args[2], args[3],
3028                        64 - args[3] - args[4]);
3029        }
3030        break;
3031
3032    case INDEX_op_extract_i32:
3033        tcg_out_rlw(s, RLWINM, args[0], args[1],
3034                    32 - args[2], 32 - args[3], 31);
3035        break;
3036    case INDEX_op_extract_i64:
3037        tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 64 - args[3]);
3038        break;
3039
3040    case INDEX_op_movcond_i32:
3041        tcg_out_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], args[2],
3042                        args[3], args[4], const_args[2]);
3043        break;
3044    case INDEX_op_movcond_i64:
3045        tcg_out_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], args[2],
3046                        args[3], args[4], const_args[2]);
3047        break;
3048
3049#if TCG_TARGET_REG_BITS == 64
3050    case INDEX_op_add2_i64:
3051#else
3052    case INDEX_op_add2_i32:
3053#endif
3054        /* Note that the CA bit is defined based on the word size of the
3055           environment.  So in 64-bit mode it's always carry-out of bit 63.
3056           The fallback code using deposit works just as well for 32-bit.  */
3057        a0 = args[0], a1 = args[1];
3058        if (a0 == args[3] || (!const_args[5] && a0 == args[5])) {
3059            a0 = TCG_REG_R0;
3060        }
3061        if (const_args[4]) {
3062            tcg_out32(s, ADDIC | TAI(a0, args[2], args[4]));
3063        } else {
3064            tcg_out32(s, ADDC | TAB(a0, args[2], args[4]));
3065        }
3066        if (const_args[5]) {
3067            tcg_out32(s, (args[5] ? ADDME : ADDZE) | RT(a1) | RA(args[3]));
3068        } else {
3069            tcg_out32(s, ADDE | TAB(a1, args[3], args[5]));
3070        }
3071        if (a0 != args[0]) {
3072            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
3073        }
3074        break;
3075
3076#if TCG_TARGET_REG_BITS == 64
3077    case INDEX_op_sub2_i64:
3078#else
3079    case INDEX_op_sub2_i32:
3080#endif
3081        a0 = args[0], a1 = args[1];
3082        if (a0 == args[5] || (!const_args[3] && a0 == args[3])) {
3083            a0 = TCG_REG_R0;
3084        }
3085        if (const_args[2]) {
3086            tcg_out32(s, SUBFIC | TAI(a0, args[4], args[2]));
3087        } else {
3088            tcg_out32(s, SUBFC | TAB(a0, args[4], args[2]));
3089        }
3090        if (const_args[3]) {
3091            tcg_out32(s, (args[3] ? SUBFME : SUBFZE) | RT(a1) | RA(args[5]));
3092        } else {
3093            tcg_out32(s, SUBFE | TAB(a1, args[5], args[3]));
3094        }
3095        if (a0 != args[0]) {
3096            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
3097        }
3098        break;
3099
3100    case INDEX_op_muluh_i32:
3101        tcg_out32(s, MULHWU | TAB(args[0], args[1], args[2]));
3102        break;
3103    case INDEX_op_mulsh_i32:
3104        tcg_out32(s, MULHW | TAB(args[0], args[1], args[2]));
3105        break;
3106    case INDEX_op_muluh_i64:
3107        tcg_out32(s, MULHDU | TAB(args[0], args[1], args[2]));
3108        break;
3109    case INDEX_op_mulsh_i64:
3110        tcg_out32(s, MULHD | TAB(args[0], args[1], args[2]));
3111        break;
3112
3113    case INDEX_op_mb:
3114        tcg_out_mb(s, args[0]);
3115        break;
3116
3117    case INDEX_op_mov_i32:   /* Always emitted via tcg_out_mov.  */
3118    case INDEX_op_mov_i64:
3119    case INDEX_op_call:      /* Always emitted via tcg_out_call.  */
3120    case INDEX_op_exit_tb:   /* Always emitted via tcg_out_exit_tb.  */
3121    case INDEX_op_goto_tb:   /* Always emitted via tcg_out_goto_tb.  */
3122    case INDEX_op_ext8s_i32:  /* Always emitted via tcg_reg_alloc_op.  */
3123    case INDEX_op_ext8s_i64:
3124    case INDEX_op_ext8u_i32:
3125    case INDEX_op_ext8u_i64:
3126    case INDEX_op_ext16s_i32:
3127    case INDEX_op_ext16s_i64:
3128    case INDEX_op_ext16u_i32:
3129    case INDEX_op_ext16u_i64:
3130    case INDEX_op_ext32s_i64:
3131    case INDEX_op_ext32u_i64:
3132    case INDEX_op_ext_i32_i64:
3133    case INDEX_op_extu_i32_i64:
3134    case INDEX_op_extrl_i64_i32:
3135    default:
3136        g_assert_not_reached();
3137    }
3138}
3139
3140int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3141{
3142    switch (opc) {
3143    case INDEX_op_and_vec:
3144    case INDEX_op_or_vec:
3145    case INDEX_op_xor_vec:
3146    case INDEX_op_andc_vec:
3147    case INDEX_op_not_vec:
3148    case INDEX_op_nor_vec:
3149    case INDEX_op_eqv_vec:
3150    case INDEX_op_nand_vec:
3151        return 1;
3152    case INDEX_op_orc_vec:
3153        return have_isa_2_07;
3154    case INDEX_op_add_vec:
3155    case INDEX_op_sub_vec:
3156    case INDEX_op_smax_vec:
3157    case INDEX_op_smin_vec:
3158    case INDEX_op_umax_vec:
3159    case INDEX_op_umin_vec:
3160    case INDEX_op_shlv_vec:
3161    case INDEX_op_shrv_vec:
3162    case INDEX_op_sarv_vec:
3163    case INDEX_op_rotlv_vec:
3164        return vece <= MO_32 || have_isa_2_07;
3165    case INDEX_op_ssadd_vec:
3166    case INDEX_op_sssub_vec:
3167    case INDEX_op_usadd_vec:
3168    case INDEX_op_ussub_vec:
3169        return vece <= MO_32;
3170    case INDEX_op_cmp_vec:
3171    case INDEX_op_shli_vec:
3172    case INDEX_op_shri_vec:
3173    case INDEX_op_sari_vec:
3174    case INDEX_op_rotli_vec:
3175        return vece <= MO_32 || have_isa_2_07 ? -1 : 0;
3176    case INDEX_op_neg_vec:
3177        return vece >= MO_32 && have_isa_3_00;
3178    case INDEX_op_mul_vec:
3179        switch (vece) {
3180        case MO_8:
3181        case MO_16:
3182            return -1;
3183        case MO_32:
3184            return have_isa_2_07 ? 1 : -1;
3185        case MO_64:
3186            return have_isa_3_10;
3187        }
3188        return 0;
3189    case INDEX_op_bitsel_vec:
3190        return have_vsx;
3191    case INDEX_op_rotrv_vec:
3192        return -1;
3193    default:
3194        return 0;
3195    }
3196}
3197
3198static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
3199                            TCGReg dst, TCGReg src)
3200{
3201    tcg_debug_assert(dst >= TCG_REG_V0);
3202
3203    /* Splat from integer reg allowed via constraints for v3.00.  */
3204    if (src < TCG_REG_V0) {
3205        tcg_debug_assert(have_isa_3_00);
3206        switch (vece) {
3207        case MO_64:
3208            tcg_out32(s, MTVSRDD | VRT(dst) | RA(src) | RB(src));
3209            return true;
3210        case MO_32:
3211            tcg_out32(s, MTVSRWS | VRT(dst) | RA(src));
3212            return true;
3213        default:
3214            /* Fail, so that we fall back on either dupm or mov+dup.  */
3215            return false;
3216        }
3217    }
3218
3219    /*
3220     * Recall we use (or emulate) VSX integer loads, so the integer is
3221     * right justified within the left (zero-index) double-word.
3222     */
3223    switch (vece) {
3224    case MO_8:
3225        tcg_out32(s, VSPLTB | VRT(dst) | VRB(src) | (7 << 16));
3226        break;
3227    case MO_16:
3228        tcg_out32(s, VSPLTH | VRT(dst) | VRB(src) | (3 << 16));
3229        break;
3230    case MO_32:
3231        tcg_out32(s, VSPLTW | VRT(dst) | VRB(src) | (1 << 16));
3232        break;
3233    case MO_64:
3234        if (have_vsx) {
3235            tcg_out32(s, XXPERMDI | VRT(dst) | VRA(src) | VRB(src));
3236            break;
3237        }
3238        tcg_out_vsldoi(s, TCG_VEC_TMP1, src, src, 8);
3239        tcg_out_vsldoi(s, dst, TCG_VEC_TMP1, src, 8);
3240        break;
3241    default:
3242        g_assert_not_reached();
3243    }
3244    return true;
3245}
3246
3247static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
3248                             TCGReg out, TCGReg base, intptr_t offset)
3249{
3250    int elt;
3251
3252    tcg_debug_assert(out >= TCG_REG_V0);
3253    switch (vece) {
3254    case MO_8:
3255        if (have_isa_3_00) {
3256            tcg_out_mem_long(s, LXV, LVX, out, base, offset & -16);
3257        } else {
3258            tcg_out_mem_long(s, 0, LVEBX, out, base, offset);
3259        }
3260        elt = extract32(offset, 0, 4);
3261#if !HOST_BIG_ENDIAN
3262        elt ^= 15;
3263#endif
3264        tcg_out32(s, VSPLTB | VRT(out) | VRB(out) | (elt << 16));
3265        break;
3266    case MO_16:
3267        tcg_debug_assert((offset & 1) == 0);
3268        if (have_isa_3_00) {
3269            tcg_out_mem_long(s, LXV | 8, LVX, out, base, offset & -16);
3270        } else {
3271            tcg_out_mem_long(s, 0, LVEHX, out, base, offset);
3272        }
3273        elt = extract32(offset, 1, 3);
3274#if !HOST_BIG_ENDIAN
3275        elt ^= 7;
3276#endif
3277        tcg_out32(s, VSPLTH | VRT(out) | VRB(out) | (elt << 16));
3278        break;
3279    case MO_32:
3280        if (have_isa_3_00) {
3281            tcg_out_mem_long(s, 0, LXVWSX, out, base, offset);
3282            break;
3283        }
3284        tcg_debug_assert((offset & 3) == 0);
3285        tcg_out_mem_long(s, 0, LVEWX, out, base, offset);
3286        elt = extract32(offset, 2, 2);
3287#if !HOST_BIG_ENDIAN
3288        elt ^= 3;
3289#endif
3290        tcg_out32(s, VSPLTW | VRT(out) | VRB(out) | (elt << 16));
3291        break;
3292    case MO_64:
3293        if (have_vsx) {
3294            tcg_out_mem_long(s, 0, LXVDSX, out, base, offset);
3295            break;
3296        }
3297        tcg_debug_assert((offset & 7) == 0);
3298        tcg_out_mem_long(s, 0, LVX, out, base, offset & -16);
3299        tcg_out_vsldoi(s, TCG_VEC_TMP1, out, out, 8);
3300        elt = extract32(offset, 3, 1);
3301#if !HOST_BIG_ENDIAN
3302        elt = !elt;
3303#endif
3304        if (elt) {
3305            tcg_out_vsldoi(s, out, out, TCG_VEC_TMP1, 8);
3306        } else {
3307            tcg_out_vsldoi(s, out, TCG_VEC_TMP1, out, 8);
3308        }
3309        break;
3310    default:
3311        g_assert_not_reached();
3312    }
3313    return true;
3314}
3315
3316static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
3317                           unsigned vecl, unsigned vece,
3318                           const TCGArg args[TCG_MAX_OP_ARGS],
3319                           const int const_args[TCG_MAX_OP_ARGS])
3320{
3321    static const uint32_t
3322        add_op[4] = { VADDUBM, VADDUHM, VADDUWM, VADDUDM },
3323        sub_op[4] = { VSUBUBM, VSUBUHM, VSUBUWM, VSUBUDM },
3324        mul_op[4] = { 0, 0, VMULUWM, VMULLD },
3325        neg_op[4] = { 0, 0, VNEGW, VNEGD },
3326        eq_op[4]  = { VCMPEQUB, VCMPEQUH, VCMPEQUW, VCMPEQUD },
3327        ne_op[4]  = { VCMPNEB, VCMPNEH, VCMPNEW, 0 },
3328        gts_op[4] = { VCMPGTSB, VCMPGTSH, VCMPGTSW, VCMPGTSD },
3329        gtu_op[4] = { VCMPGTUB, VCMPGTUH, VCMPGTUW, VCMPGTUD },
3330        ssadd_op[4] = { VADDSBS, VADDSHS, VADDSWS, 0 },
3331        usadd_op[4] = { VADDUBS, VADDUHS, VADDUWS, 0 },
3332        sssub_op[4] = { VSUBSBS, VSUBSHS, VSUBSWS, 0 },
3333        ussub_op[4] = { VSUBUBS, VSUBUHS, VSUBUWS, 0 },
3334        umin_op[4] = { VMINUB, VMINUH, VMINUW, VMINUD },
3335        smin_op[4] = { VMINSB, VMINSH, VMINSW, VMINSD },
3336        umax_op[4] = { VMAXUB, VMAXUH, VMAXUW, VMAXUD },
3337        smax_op[4] = { VMAXSB, VMAXSH, VMAXSW, VMAXSD },
3338        shlv_op[4] = { VSLB, VSLH, VSLW, VSLD },
3339        shrv_op[4] = { VSRB, VSRH, VSRW, VSRD },
3340        sarv_op[4] = { VSRAB, VSRAH, VSRAW, VSRAD },
3341        mrgh_op[4] = { VMRGHB, VMRGHH, VMRGHW, 0 },
3342        mrgl_op[4] = { VMRGLB, VMRGLH, VMRGLW, 0 },
3343        muleu_op[4] = { VMULEUB, VMULEUH, VMULEUW, 0 },
3344        mulou_op[4] = { VMULOUB, VMULOUH, VMULOUW, 0 },
3345        pkum_op[4] = { VPKUHUM, VPKUWUM, 0, 0 },
3346        rotl_op[4] = { VRLB, VRLH, VRLW, VRLD };
3347
3348    TCGType type = vecl + TCG_TYPE_V64;
3349    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
3350    uint32_t insn;
3351
3352    switch (opc) {
3353    case INDEX_op_ld_vec:
3354        tcg_out_ld(s, type, a0, a1, a2);
3355        return;
3356    case INDEX_op_st_vec:
3357        tcg_out_st(s, type, a0, a1, a2);
3358        return;
3359    case INDEX_op_dupm_vec:
3360        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
3361        return;
3362
3363    case INDEX_op_add_vec:
3364        insn = add_op[vece];
3365        break;
3366    case INDEX_op_sub_vec:
3367        insn = sub_op[vece];
3368        break;
3369    case INDEX_op_neg_vec:
3370        insn = neg_op[vece];
3371        a2 = a1;
3372        a1 = 0;
3373        break;
3374    case INDEX_op_mul_vec:
3375        insn = mul_op[vece];
3376        break;
3377    case INDEX_op_ssadd_vec:
3378        insn = ssadd_op[vece];
3379        break;
3380    case INDEX_op_sssub_vec:
3381        insn = sssub_op[vece];
3382        break;
3383    case INDEX_op_usadd_vec:
3384        insn = usadd_op[vece];
3385        break;
3386    case INDEX_op_ussub_vec:
3387        insn = ussub_op[vece];
3388        break;
3389    case INDEX_op_smin_vec:
3390        insn = smin_op[vece];
3391        break;
3392    case INDEX_op_umin_vec:
3393        insn = umin_op[vece];
3394        break;
3395    case INDEX_op_smax_vec:
3396        insn = smax_op[vece];
3397        break;
3398    case INDEX_op_umax_vec:
3399        insn = umax_op[vece];
3400        break;
3401    case INDEX_op_shlv_vec:
3402        insn = shlv_op[vece];
3403        break;
3404    case INDEX_op_shrv_vec:
3405        insn = shrv_op[vece];
3406        break;
3407    case INDEX_op_sarv_vec:
3408        insn = sarv_op[vece];
3409        break;
3410    case INDEX_op_and_vec:
3411        insn = VAND;
3412        break;
3413    case INDEX_op_or_vec:
3414        insn = VOR;
3415        break;
3416    case INDEX_op_xor_vec:
3417        insn = VXOR;
3418        break;
3419    case INDEX_op_andc_vec:
3420        insn = VANDC;
3421        break;
3422    case INDEX_op_not_vec:
3423        insn = VNOR;
3424        a2 = a1;
3425        break;
3426    case INDEX_op_orc_vec:
3427        insn = VORC;
3428        break;
3429    case INDEX_op_nand_vec:
3430        insn = VNAND;
3431        break;
3432    case INDEX_op_nor_vec:
3433        insn = VNOR;
3434        break;
3435    case INDEX_op_eqv_vec:
3436        insn = VEQV;
3437        break;
3438
3439    case INDEX_op_cmp_vec:
3440        switch (args[3]) {
3441        case TCG_COND_EQ:
3442            insn = eq_op[vece];
3443            break;
3444        case TCG_COND_NE:
3445            insn = ne_op[vece];
3446            break;
3447        case TCG_COND_GT:
3448            insn = gts_op[vece];
3449            break;
3450        case TCG_COND_GTU:
3451            insn = gtu_op[vece];
3452            break;
3453        default:
3454            g_assert_not_reached();
3455        }
3456        break;
3457
3458    case INDEX_op_bitsel_vec:
3459        tcg_out32(s, XXSEL | VRT(a0) | VRC(a1) | VRB(a2) | VRA(args[3]));
3460        return;
3461
3462    case INDEX_op_dup2_vec:
3463        assert(TCG_TARGET_REG_BITS == 32);
3464        /* With inputs a1 = xLxx, a2 = xHxx  */
3465        tcg_out32(s, VMRGHW | VRT(a0) | VRA(a2) | VRB(a1));  /* a0  = xxHL */
3466        tcg_out_vsldoi(s, TCG_VEC_TMP1, a0, a0, 8);          /* tmp = HLxx */
3467        tcg_out_vsldoi(s, a0, a0, TCG_VEC_TMP1, 8);          /* a0  = HLHL */
3468        return;
3469
3470    case INDEX_op_ppc_mrgh_vec:
3471        insn = mrgh_op[vece];
3472        break;
3473    case INDEX_op_ppc_mrgl_vec:
3474        insn = mrgl_op[vece];
3475        break;
3476    case INDEX_op_ppc_muleu_vec:
3477        insn = muleu_op[vece];
3478        break;
3479    case INDEX_op_ppc_mulou_vec:
3480        insn = mulou_op[vece];
3481        break;
3482    case INDEX_op_ppc_pkum_vec:
3483        insn = pkum_op[vece];
3484        break;
3485    case INDEX_op_rotlv_vec:
3486        insn = rotl_op[vece];
3487        break;
3488    case INDEX_op_ppc_msum_vec:
3489        tcg_debug_assert(vece == MO_16);
3490        tcg_out32(s, VMSUMUHM | VRT(a0) | VRA(a1) | VRB(a2) | VRC(args[3]));
3491        return;
3492
3493    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
3494    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
3495    default:
3496        g_assert_not_reached();
3497    }
3498
3499    tcg_debug_assert(insn != 0);
3500    tcg_out32(s, insn | VRT(a0) | VRA(a1) | VRB(a2));
3501}
3502
3503static void expand_vec_shi(TCGType type, unsigned vece, TCGv_vec v0,
3504                           TCGv_vec v1, TCGArg imm, TCGOpcode opci)
3505{
3506    TCGv_vec t1;
3507
3508    if (vece == MO_32) {
3509        /*
3510         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
3511         * So using negative numbers gets us the 4th bit easily.
3512         */
3513        imm = sextract32(imm, 0, 5);
3514    } else {
3515        imm &= (8 << vece) - 1;
3516    }
3517
3518    /* Splat w/bytes for xxspltib when 2.07 allows MO_64. */
3519    t1 = tcg_constant_vec(type, MO_8, imm);
3520    vec_gen_3(opci, type, vece, tcgv_vec_arg(v0),
3521              tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3522}
3523
3524static void expand_vec_cmp(TCGType type, unsigned vece, TCGv_vec v0,
3525                           TCGv_vec v1, TCGv_vec v2, TCGCond cond)
3526{
3527    bool need_swap = false, need_inv = false;
3528
3529    tcg_debug_assert(vece <= MO_32 || have_isa_2_07);
3530
3531    switch (cond) {
3532    case TCG_COND_EQ:
3533    case TCG_COND_GT:
3534    case TCG_COND_GTU:
3535        break;
3536    case TCG_COND_NE:
3537        if (have_isa_3_00 && vece <= MO_32) {
3538            break;
3539        }
3540        /* fall through */
3541    case TCG_COND_LE:
3542    case TCG_COND_LEU:
3543        need_inv = true;
3544        break;
3545    case TCG_COND_LT:
3546    case TCG_COND_LTU:
3547        need_swap = true;
3548        break;
3549    case TCG_COND_GE:
3550    case TCG_COND_GEU:
3551        need_swap = need_inv = true;
3552        break;
3553    default:
3554        g_assert_not_reached();
3555    }
3556
3557    if (need_inv) {
3558        cond = tcg_invert_cond(cond);
3559    }
3560    if (need_swap) {
3561        TCGv_vec t1;
3562        t1 = v1, v1 = v2, v2 = t1;
3563        cond = tcg_swap_cond(cond);
3564    }
3565
3566    vec_gen_4(INDEX_op_cmp_vec, type, vece, tcgv_vec_arg(v0),
3567              tcgv_vec_arg(v1), tcgv_vec_arg(v2), cond);
3568
3569    if (need_inv) {
3570        tcg_gen_not_vec(vece, v0, v0);
3571    }
3572}
3573
3574static void expand_vec_mul(TCGType type, unsigned vece, TCGv_vec v0,
3575                           TCGv_vec v1, TCGv_vec v2)
3576{
3577    TCGv_vec t1 = tcg_temp_new_vec(type);
3578    TCGv_vec t2 = tcg_temp_new_vec(type);
3579    TCGv_vec c0, c16;
3580
3581    switch (vece) {
3582    case MO_8:
3583    case MO_16:
3584        vec_gen_3(INDEX_op_ppc_muleu_vec, type, vece, tcgv_vec_arg(t1),
3585                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3586        vec_gen_3(INDEX_op_ppc_mulou_vec, type, vece, tcgv_vec_arg(t2),
3587                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3588        vec_gen_3(INDEX_op_ppc_mrgh_vec, type, vece + 1, tcgv_vec_arg(v0),
3589                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3590        vec_gen_3(INDEX_op_ppc_mrgl_vec, type, vece + 1, tcgv_vec_arg(t1),
3591                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3592        vec_gen_3(INDEX_op_ppc_pkum_vec, type, vece, tcgv_vec_arg(v0),
3593                  tcgv_vec_arg(v0), tcgv_vec_arg(t1));
3594	break;
3595
3596    case MO_32:
3597        tcg_debug_assert(!have_isa_2_07);
3598        /*
3599         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
3600         * So using -16 is a quick way to represent 16.
3601         */
3602        c16 = tcg_constant_vec(type, MO_8, -16);
3603        c0 = tcg_constant_vec(type, MO_8, 0);
3604
3605        vec_gen_3(INDEX_op_rotlv_vec, type, MO_32, tcgv_vec_arg(t1),
3606                  tcgv_vec_arg(v2), tcgv_vec_arg(c16));
3607        vec_gen_3(INDEX_op_ppc_mulou_vec, type, MO_16, tcgv_vec_arg(t2),
3608                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3609        vec_gen_4(INDEX_op_ppc_msum_vec, type, MO_16, tcgv_vec_arg(t1),
3610                  tcgv_vec_arg(v1), tcgv_vec_arg(t1), tcgv_vec_arg(c0));
3611        vec_gen_3(INDEX_op_shlv_vec, type, MO_32, tcgv_vec_arg(t1),
3612                  tcgv_vec_arg(t1), tcgv_vec_arg(c16));
3613        tcg_gen_add_vec(MO_32, v0, t1, t2);
3614        break;
3615
3616    default:
3617        g_assert_not_reached();
3618    }
3619    tcg_temp_free_vec(t1);
3620    tcg_temp_free_vec(t2);
3621}
3622
3623void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3624                       TCGArg a0, ...)
3625{
3626    va_list va;
3627    TCGv_vec v0, v1, v2, t0;
3628    TCGArg a2;
3629
3630    va_start(va, a0);
3631    v0 = temp_tcgv_vec(arg_temp(a0));
3632    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3633    a2 = va_arg(va, TCGArg);
3634
3635    switch (opc) {
3636    case INDEX_op_shli_vec:
3637        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shlv_vec);
3638        break;
3639    case INDEX_op_shri_vec:
3640        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shrv_vec);
3641        break;
3642    case INDEX_op_sari_vec:
3643        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_sarv_vec);
3644        break;
3645    case INDEX_op_rotli_vec:
3646        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_rotlv_vec);
3647        break;
3648    case INDEX_op_cmp_vec:
3649        v2 = temp_tcgv_vec(arg_temp(a2));
3650        expand_vec_cmp(type, vece, v0, v1, v2, va_arg(va, TCGArg));
3651        break;
3652    case INDEX_op_mul_vec:
3653        v2 = temp_tcgv_vec(arg_temp(a2));
3654        expand_vec_mul(type, vece, v0, v1, v2);
3655        break;
3656    case INDEX_op_rotlv_vec:
3657        v2 = temp_tcgv_vec(arg_temp(a2));
3658        t0 = tcg_temp_new_vec(type);
3659        tcg_gen_neg_vec(vece, t0, v2);
3660        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3661        tcg_temp_free_vec(t0);
3662        break;
3663    default:
3664        g_assert_not_reached();
3665    }
3666    va_end(va);
3667}
3668
3669static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
3670{
3671    switch (op) {
3672    case INDEX_op_goto_ptr:
3673        return C_O0_I1(r);
3674
3675    case INDEX_op_ld8u_i32:
3676    case INDEX_op_ld8s_i32:
3677    case INDEX_op_ld16u_i32:
3678    case INDEX_op_ld16s_i32:
3679    case INDEX_op_ld_i32:
3680    case INDEX_op_ctpop_i32:
3681    case INDEX_op_neg_i32:
3682    case INDEX_op_not_i32:
3683    case INDEX_op_ext8s_i32:
3684    case INDEX_op_ext16s_i32:
3685    case INDEX_op_bswap16_i32:
3686    case INDEX_op_bswap32_i32:
3687    case INDEX_op_extract_i32:
3688    case INDEX_op_ld8u_i64:
3689    case INDEX_op_ld8s_i64:
3690    case INDEX_op_ld16u_i64:
3691    case INDEX_op_ld16s_i64:
3692    case INDEX_op_ld32u_i64:
3693    case INDEX_op_ld32s_i64:
3694    case INDEX_op_ld_i64:
3695    case INDEX_op_ctpop_i64:
3696    case INDEX_op_neg_i64:
3697    case INDEX_op_not_i64:
3698    case INDEX_op_ext8s_i64:
3699    case INDEX_op_ext16s_i64:
3700    case INDEX_op_ext32s_i64:
3701    case INDEX_op_ext_i32_i64:
3702    case INDEX_op_extu_i32_i64:
3703    case INDEX_op_bswap16_i64:
3704    case INDEX_op_bswap32_i64:
3705    case INDEX_op_bswap64_i64:
3706    case INDEX_op_extract_i64:
3707        return C_O1_I1(r, r);
3708
3709    case INDEX_op_st8_i32:
3710    case INDEX_op_st16_i32:
3711    case INDEX_op_st_i32:
3712    case INDEX_op_st8_i64:
3713    case INDEX_op_st16_i64:
3714    case INDEX_op_st32_i64:
3715    case INDEX_op_st_i64:
3716        return C_O0_I2(r, r);
3717
3718    case INDEX_op_add_i32:
3719    case INDEX_op_and_i32:
3720    case INDEX_op_or_i32:
3721    case INDEX_op_xor_i32:
3722    case INDEX_op_andc_i32:
3723    case INDEX_op_orc_i32:
3724    case INDEX_op_eqv_i32:
3725    case INDEX_op_shl_i32:
3726    case INDEX_op_shr_i32:
3727    case INDEX_op_sar_i32:
3728    case INDEX_op_rotl_i32:
3729    case INDEX_op_rotr_i32:
3730    case INDEX_op_setcond_i32:
3731    case INDEX_op_and_i64:
3732    case INDEX_op_andc_i64:
3733    case INDEX_op_shl_i64:
3734    case INDEX_op_shr_i64:
3735    case INDEX_op_sar_i64:
3736    case INDEX_op_rotl_i64:
3737    case INDEX_op_rotr_i64:
3738    case INDEX_op_setcond_i64:
3739        return C_O1_I2(r, r, ri);
3740
3741    case INDEX_op_mul_i32:
3742    case INDEX_op_mul_i64:
3743        return C_O1_I2(r, r, rI);
3744
3745    case INDEX_op_div_i32:
3746    case INDEX_op_divu_i32:
3747    case INDEX_op_rem_i32:
3748    case INDEX_op_remu_i32:
3749    case INDEX_op_nand_i32:
3750    case INDEX_op_nor_i32:
3751    case INDEX_op_muluh_i32:
3752    case INDEX_op_mulsh_i32:
3753    case INDEX_op_orc_i64:
3754    case INDEX_op_eqv_i64:
3755    case INDEX_op_nand_i64:
3756    case INDEX_op_nor_i64:
3757    case INDEX_op_div_i64:
3758    case INDEX_op_divu_i64:
3759    case INDEX_op_rem_i64:
3760    case INDEX_op_remu_i64:
3761    case INDEX_op_mulsh_i64:
3762    case INDEX_op_muluh_i64:
3763        return C_O1_I2(r, r, r);
3764
3765    case INDEX_op_sub_i32:
3766        return C_O1_I2(r, rI, ri);
3767    case INDEX_op_add_i64:
3768        return C_O1_I2(r, r, rT);
3769    case INDEX_op_or_i64:
3770    case INDEX_op_xor_i64:
3771        return C_O1_I2(r, r, rU);
3772    case INDEX_op_sub_i64:
3773        return C_O1_I2(r, rI, rT);
3774    case INDEX_op_clz_i32:
3775    case INDEX_op_ctz_i32:
3776    case INDEX_op_clz_i64:
3777    case INDEX_op_ctz_i64:
3778        return C_O1_I2(r, r, rZW);
3779
3780    case INDEX_op_brcond_i32:
3781    case INDEX_op_brcond_i64:
3782        return C_O0_I2(r, ri);
3783
3784    case INDEX_op_movcond_i32:
3785    case INDEX_op_movcond_i64:
3786        return C_O1_I4(r, r, ri, rZ, rZ);
3787    case INDEX_op_deposit_i32:
3788    case INDEX_op_deposit_i64:
3789        return C_O1_I2(r, 0, rZ);
3790    case INDEX_op_brcond2_i32:
3791        return C_O0_I4(r, r, ri, ri);
3792    case INDEX_op_setcond2_i32:
3793        return C_O1_I4(r, r, r, ri, ri);
3794    case INDEX_op_add2_i64:
3795    case INDEX_op_add2_i32:
3796        return C_O2_I4(r, r, r, r, rI, rZM);
3797    case INDEX_op_sub2_i64:
3798    case INDEX_op_sub2_i32:
3799        return C_O2_I4(r, r, rI, rZM, r, r);
3800
3801    case INDEX_op_qemu_ld_i32:
3802        return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
3803                ? C_O1_I1(r, L)
3804                : C_O1_I2(r, L, L));
3805
3806    case INDEX_op_qemu_st_i32:
3807        return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
3808                ? C_O0_I2(S, S)
3809                : C_O0_I3(S, S, S));
3810
3811    case INDEX_op_qemu_ld_i64:
3812        return (TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, L)
3813                : TARGET_LONG_BITS == 32 ? C_O2_I1(L, L, L)
3814                : C_O2_I2(L, L, L, L));
3815
3816    case INDEX_op_qemu_st_i64:
3817        return (TCG_TARGET_REG_BITS == 64 ? C_O0_I2(S, S)
3818                : TARGET_LONG_BITS == 32 ? C_O0_I3(S, S, S)
3819                : C_O0_I4(S, S, S, S));
3820
3821    case INDEX_op_add_vec:
3822    case INDEX_op_sub_vec:
3823    case INDEX_op_mul_vec:
3824    case INDEX_op_and_vec:
3825    case INDEX_op_or_vec:
3826    case INDEX_op_xor_vec:
3827    case INDEX_op_andc_vec:
3828    case INDEX_op_orc_vec:
3829    case INDEX_op_nor_vec:
3830    case INDEX_op_eqv_vec:
3831    case INDEX_op_nand_vec:
3832    case INDEX_op_cmp_vec:
3833    case INDEX_op_ssadd_vec:
3834    case INDEX_op_sssub_vec:
3835    case INDEX_op_usadd_vec:
3836    case INDEX_op_ussub_vec:
3837    case INDEX_op_smax_vec:
3838    case INDEX_op_smin_vec:
3839    case INDEX_op_umax_vec:
3840    case INDEX_op_umin_vec:
3841    case INDEX_op_shlv_vec:
3842    case INDEX_op_shrv_vec:
3843    case INDEX_op_sarv_vec:
3844    case INDEX_op_rotlv_vec:
3845    case INDEX_op_rotrv_vec:
3846    case INDEX_op_ppc_mrgh_vec:
3847    case INDEX_op_ppc_mrgl_vec:
3848    case INDEX_op_ppc_muleu_vec:
3849    case INDEX_op_ppc_mulou_vec:
3850    case INDEX_op_ppc_pkum_vec:
3851    case INDEX_op_dup2_vec:
3852        return C_O1_I2(v, v, v);
3853
3854    case INDEX_op_not_vec:
3855    case INDEX_op_neg_vec:
3856        return C_O1_I1(v, v);
3857
3858    case INDEX_op_dup_vec:
3859        return have_isa_3_00 ? C_O1_I1(v, vr) : C_O1_I1(v, v);
3860
3861    case INDEX_op_ld_vec:
3862    case INDEX_op_dupm_vec:
3863        return C_O1_I1(v, r);
3864
3865    case INDEX_op_st_vec:
3866        return C_O0_I2(v, r);
3867
3868    case INDEX_op_bitsel_vec:
3869    case INDEX_op_ppc_msum_vec:
3870        return C_O1_I3(v, v, v, v);
3871
3872    default:
3873        g_assert_not_reached();
3874    }
3875}
3876
3877static void tcg_target_init(TCGContext *s)
3878{
3879    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3880    unsigned long hwcap2 = qemu_getauxval(AT_HWCAP2);
3881
3882    have_isa = tcg_isa_base;
3883    if (hwcap & PPC_FEATURE_ARCH_2_06) {
3884        have_isa = tcg_isa_2_06;
3885    }
3886#ifdef PPC_FEATURE2_ARCH_2_07
3887    if (hwcap2 & PPC_FEATURE2_ARCH_2_07) {
3888        have_isa = tcg_isa_2_07;
3889    }
3890#endif
3891#ifdef PPC_FEATURE2_ARCH_3_00
3892    if (hwcap2 & PPC_FEATURE2_ARCH_3_00) {
3893        have_isa = tcg_isa_3_00;
3894    }
3895#endif
3896#ifdef PPC_FEATURE2_ARCH_3_10
3897    if (hwcap2 & PPC_FEATURE2_ARCH_3_10) {
3898        have_isa = tcg_isa_3_10;
3899    }
3900#endif
3901
3902#ifdef PPC_FEATURE2_HAS_ISEL
3903    /* Prefer explicit instruction from the kernel. */
3904    have_isel = (hwcap2 & PPC_FEATURE2_HAS_ISEL) != 0;
3905#else
3906    /* Fall back to knowing Power7 (2.06) has ISEL. */
3907    have_isel = have_isa_2_06;
3908#endif
3909
3910    if (hwcap & PPC_FEATURE_HAS_ALTIVEC) {
3911        have_altivec = true;
3912        /* We only care about the portion of VSX that overlaps Altivec. */
3913        if (hwcap & PPC_FEATURE_HAS_VSX) {
3914            have_vsx = true;
3915        }
3916    }
3917
3918    tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
3919    tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
3920    if (have_altivec) {
3921        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3922        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3923    }
3924
3925    tcg_target_call_clobber_regs = 0;
3926    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3927    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3928    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3929    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3930    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3931    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3932    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R7);
3933    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R8);
3934    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R9);
3935    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R10);
3936    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R11);
3937    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
3938
3939    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3940    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3941    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3942    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3943    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3944    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3945    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3946    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3947    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V8);
3948    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V9);
3949    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V10);
3950    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V11);
3951    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V12);
3952    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V13);
3953    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V14);
3954    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V15);
3955    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3956    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3957    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3958    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3959
3960    s->reserved_regs = 0;
3961    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); /* tcg temp */
3962    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R1); /* stack pointer */
3963#if defined(_CALL_SYSV)
3964    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R2); /* toc pointer */
3965#endif
3966#if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
3967    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R13); /* thread pointer */
3968#endif
3969    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1); /* mem temp */
3970    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP1);
3971    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP2);
3972    if (USE_REG_TB) {
3973        tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);  /* tb->tc_ptr */
3974    }
3975}
3976
3977#ifdef __ELF__
3978typedef struct {
3979    DebugFrameCIE cie;
3980    DebugFrameFDEHeader fde;
3981    uint8_t fde_def_cfa[4];
3982    uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2 + 3];
3983} DebugFrame;
3984
3985/* We're expecting a 2 byte uleb128 encoded value.  */
3986QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3987
3988#if TCG_TARGET_REG_BITS == 64
3989# define ELF_HOST_MACHINE EM_PPC64
3990#else
3991# define ELF_HOST_MACHINE EM_PPC
3992#endif
3993
3994static DebugFrame debug_frame = {
3995    .cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3996    .cie.id = -1,
3997    .cie.version = 1,
3998    .cie.code_align = 1,
3999    .cie.data_align = (-SZR & 0x7f),         /* sleb128 -SZR */
4000    .cie.return_column = 65,
4001
4002    /* Total FDE size does not include the "len" member.  */
4003    .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset),
4004
4005    .fde_def_cfa = {
4006        12, TCG_REG_R1,                 /* DW_CFA_def_cfa r1, ... */
4007        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
4008        (FRAME_SIZE >> 7)
4009    },
4010    .fde_reg_ofs = {
4011        /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
4012        0x11, 65, (LR_OFFSET / -SZR) & 0x7f,
4013    }
4014};
4015
4016void tcg_register_jit(const void *buf, size_t buf_size)
4017{
4018    uint8_t *p = &debug_frame.fde_reg_ofs[3];
4019    int i;
4020
4021    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i, p += 2) {
4022        p[0] = 0x80 + tcg_target_callee_save_regs[i];
4023        p[1] = (FRAME_SIZE - (REG_SAVE_BOT + i * SZR)) / SZR;
4024    }
4025
4026    debug_frame.fde.func_start = (uintptr_t)buf;
4027    debug_frame.fde.func_len = buf_size;
4028
4029    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
4030}
4031#endif /* __ELF__ */
4032#undef VMULEUB
4033#undef VMULEUH
4034#undef VMULEUW
4035#undef VMULOUB
4036#undef VMULOUH
4037#undef VMULOUW
4038#undef VMSUMUHM
4039