xref: /openbmc/qemu/tcg/ppc/tcg-target.c.inc (revision 57223a4c)
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 or _CALL_AIX.
33 */
34#if TCG_TARGET_REG_BITS == 64
35# ifdef _CALL_AIX
36    /* ok */
37# elif defined(_CALL_ELF) && _CALL_ELF == 1
38#  define _CALL_AIX
39# elif defined(_CALL_ELF) && _CALL_ELF == 2
40    /* ok */
41# else
42#  error "Unknown ABI"
43# endif
44#else
45# if defined(_CALL_SYSV) || defined(_CALL_DARWIN)
46    /* ok */
47# elif defined(__APPLE__)
48#  define _CALL_DARWIN
49# elif defined(__ELF__)
50#  define _CALL_SYSV
51# else
52#  error "Unknown ABI"
53# endif
54#endif
55
56#if TCG_TARGET_REG_BITS == 64
57# define TCG_TARGET_CALL_ARG_I32   TCG_CALL_ARG_EXTEND
58# define TCG_TARGET_CALL_RET_I128  TCG_CALL_RET_NORMAL
59#else
60# define TCG_TARGET_CALL_ARG_I32   TCG_CALL_ARG_NORMAL
61# define TCG_TARGET_CALL_RET_I128  TCG_CALL_RET_BY_REF
62#endif
63#ifdef _CALL_SYSV
64# define TCG_TARGET_CALL_ARG_I64   TCG_CALL_ARG_EVEN
65# define TCG_TARGET_CALL_ARG_I128  TCG_CALL_ARG_BY_REF
66#else
67# define TCG_TARGET_CALL_ARG_I64   TCG_CALL_ARG_NORMAL
68# define TCG_TARGET_CALL_ARG_I128  TCG_CALL_ARG_NORMAL
69#endif
70
71/* For some memory operations, we need a scratch that isn't R0.  For the AIX
72   calling convention, we can re-use the TOC register since we'll be reloading
73   it at every call.  Otherwise R12 will do nicely as neither a call-saved
74   register nor a parameter register.  */
75#ifdef _CALL_AIX
76# define TCG_REG_TMP1   TCG_REG_R2
77#else
78# define TCG_REG_TMP1   TCG_REG_R12
79#endif
80#define TCG_REG_TMP2    TCG_REG_R11
81
82#define TCG_VEC_TMP1    TCG_REG_V0
83#define TCG_VEC_TMP2    TCG_REG_V1
84
85#define TCG_REG_TB     TCG_REG_R31
86#define USE_REG_TB     (TCG_TARGET_REG_BITS == 64)
87
88/* Shorthand for size of a pointer.  Avoid promotion to unsigned.  */
89#define SZP  ((int)sizeof(void *))
90
91/* Shorthand for size of a register.  */
92#define SZR  (TCG_TARGET_REG_BITS / 8)
93
94#define TCG_CT_CONST_S16  0x100
95#define TCG_CT_CONST_S32  0x400
96#define TCG_CT_CONST_U32  0x800
97#define TCG_CT_CONST_ZERO 0x1000
98#define TCG_CT_CONST_MONE 0x2000
99#define TCG_CT_CONST_WSZ  0x4000
100
101#define ALL_GENERAL_REGS  0xffffffffu
102#define ALL_VECTOR_REGS   0xffffffff00000000ull
103
104TCGPowerISA have_isa;
105static bool have_isel;
106bool have_altivec;
107bool have_vsx;
108
109#ifndef CONFIG_SOFTMMU
110#define TCG_GUEST_BASE_REG 30
111#endif
112
113#ifdef CONFIG_DEBUG_TCG
114static const char tcg_target_reg_names[TCG_TARGET_NB_REGS][4] = {
115    "r0",  "r1",  "r2",  "r3",  "r4",  "r5",  "r6",  "r7",
116    "r8",  "r9",  "r10", "r11", "r12", "r13", "r14", "r15",
117    "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
118    "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31",
119    "v0",  "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7",
120    "v8",  "v9",  "v10", "v11", "v12", "v13", "v14", "v15",
121    "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
122    "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",
123};
124#endif
125
126static const int tcg_target_reg_alloc_order[] = {
127    TCG_REG_R14,  /* call saved registers */
128    TCG_REG_R15,
129    TCG_REG_R16,
130    TCG_REG_R17,
131    TCG_REG_R18,
132    TCG_REG_R19,
133    TCG_REG_R20,
134    TCG_REG_R21,
135    TCG_REG_R22,
136    TCG_REG_R23,
137    TCG_REG_R24,
138    TCG_REG_R25,
139    TCG_REG_R26,
140    TCG_REG_R27,
141    TCG_REG_R28,
142    TCG_REG_R29,
143    TCG_REG_R30,
144    TCG_REG_R31,
145    TCG_REG_R12,  /* call clobbered, non-arguments */
146    TCG_REG_R11,
147    TCG_REG_R2,
148    TCG_REG_R13,
149    TCG_REG_R10,  /* call clobbered, arguments */
150    TCG_REG_R9,
151    TCG_REG_R8,
152    TCG_REG_R7,
153    TCG_REG_R6,
154    TCG_REG_R5,
155    TCG_REG_R4,
156    TCG_REG_R3,
157
158    /* V0 and V1 reserved as temporaries; V20 - V31 are call-saved */
159    TCG_REG_V2,   /* call clobbered, vectors */
160    TCG_REG_V3,
161    TCG_REG_V4,
162    TCG_REG_V5,
163    TCG_REG_V6,
164    TCG_REG_V7,
165    TCG_REG_V8,
166    TCG_REG_V9,
167    TCG_REG_V10,
168    TCG_REG_V11,
169    TCG_REG_V12,
170    TCG_REG_V13,
171    TCG_REG_V14,
172    TCG_REG_V15,
173    TCG_REG_V16,
174    TCG_REG_V17,
175    TCG_REG_V18,
176    TCG_REG_V19,
177};
178
179static const int tcg_target_call_iarg_regs[] = {
180    TCG_REG_R3,
181    TCG_REG_R4,
182    TCG_REG_R5,
183    TCG_REG_R6,
184    TCG_REG_R7,
185    TCG_REG_R8,
186    TCG_REG_R9,
187    TCG_REG_R10
188};
189
190static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
191{
192    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
193    tcg_debug_assert(slot >= 0 && slot <= 1);
194    return TCG_REG_R3 + slot;
195}
196
197static const int tcg_target_callee_save_regs[] = {
198#ifdef _CALL_DARWIN
199    TCG_REG_R11,
200#endif
201    TCG_REG_R14,
202    TCG_REG_R15,
203    TCG_REG_R16,
204    TCG_REG_R17,
205    TCG_REG_R18,
206    TCG_REG_R19,
207    TCG_REG_R20,
208    TCG_REG_R21,
209    TCG_REG_R22,
210    TCG_REG_R23,
211    TCG_REG_R24,
212    TCG_REG_R25,
213    TCG_REG_R26,
214    TCG_REG_R27, /* currently used for the global env */
215    TCG_REG_R28,
216    TCG_REG_R29,
217    TCG_REG_R30,
218    TCG_REG_R31
219};
220
221static inline bool in_range_b(tcg_target_long target)
222{
223    return target == sextract64(target, 0, 26);
224}
225
226static uint32_t reloc_pc24_val(const tcg_insn_unit *pc,
227			       const tcg_insn_unit *target)
228{
229    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
230    tcg_debug_assert(in_range_b(disp));
231    return disp & 0x3fffffc;
232}
233
234static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
235{
236    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
237    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
238
239    if (in_range_b(disp)) {
240        *src_rw = (*src_rw & ~0x3fffffc) | (disp & 0x3fffffc);
241        return true;
242    }
243    return false;
244}
245
246static uint16_t reloc_pc14_val(const tcg_insn_unit *pc,
247			       const tcg_insn_unit *target)
248{
249    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
250    tcg_debug_assert(disp == (int16_t) disp);
251    return disp & 0xfffc;
252}
253
254static bool reloc_pc14(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
255{
256    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
257    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
258
259    if (disp == (int16_t) disp) {
260        *src_rw = (*src_rw & ~0xfffc) | (disp & 0xfffc);
261        return true;
262    }
263    return false;
264}
265
266/* test if a constant matches the constraint */
267static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
268{
269    if (ct & TCG_CT_CONST) {
270        return 1;
271    }
272
273    /* The only 32-bit constraint we use aside from
274       TCG_CT_CONST is TCG_CT_CONST_S16.  */
275    if (type == TCG_TYPE_I32) {
276        val = (int32_t)val;
277    }
278
279    if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
280        return 1;
281    } else if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
282        return 1;
283    } else if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
284        return 1;
285    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
286        return 1;
287    } else if ((ct & TCG_CT_CONST_MONE) && val == -1) {
288        return 1;
289    } else if ((ct & TCG_CT_CONST_WSZ)
290               && val == (type == TCG_TYPE_I32 ? 32 : 64)) {
291        return 1;
292    }
293    return 0;
294}
295
296#define OPCD(opc) ((opc)<<26)
297#define XO19(opc) (OPCD(19)|((opc)<<1))
298#define MD30(opc) (OPCD(30)|((opc)<<2))
299#define MDS30(opc) (OPCD(30)|((opc)<<1))
300#define XO31(opc) (OPCD(31)|((opc)<<1))
301#define XO58(opc) (OPCD(58)|(opc))
302#define XO62(opc) (OPCD(62)|(opc))
303#define VX4(opc)  (OPCD(4)|(opc))
304
305#define B      OPCD( 18)
306#define BC     OPCD( 16)
307
308#define LBZ    OPCD( 34)
309#define LHZ    OPCD( 40)
310#define LHA    OPCD( 42)
311#define LWZ    OPCD( 32)
312#define LWZUX  XO31( 55)
313#define LD     XO58(  0)
314#define LDX    XO31( 21)
315#define LDU    XO58(  1)
316#define LDUX   XO31( 53)
317#define LWA    XO58(  2)
318#define LWAX   XO31(341)
319#define LQ     OPCD( 56)
320
321#define STB    OPCD( 38)
322#define STH    OPCD( 44)
323#define STW    OPCD( 36)
324#define STD    XO62(  0)
325#define STDU   XO62(  1)
326#define STDX   XO31(149)
327#define STQ    XO62(  2)
328
329#define ADDIC  OPCD( 12)
330#define ADDI   OPCD( 14)
331#define ADDIS  OPCD( 15)
332#define ORI    OPCD( 24)
333#define ORIS   OPCD( 25)
334#define XORI   OPCD( 26)
335#define XORIS  OPCD( 27)
336#define ANDI   OPCD( 28)
337#define ANDIS  OPCD( 29)
338#define MULLI  OPCD(  7)
339#define CMPLI  OPCD( 10)
340#define CMPI   OPCD( 11)
341#define SUBFIC OPCD( 8)
342
343#define LWZU   OPCD( 33)
344#define STWU   OPCD( 37)
345
346#define RLWIMI OPCD( 20)
347#define RLWINM OPCD( 21)
348#define RLWNM  OPCD( 23)
349
350#define RLDICL MD30(  0)
351#define RLDICR MD30(  1)
352#define RLDIMI MD30(  3)
353#define RLDCL  MDS30( 8)
354
355#define BCLR   XO19( 16)
356#define BCCTR  XO19(528)
357#define CRAND  XO19(257)
358#define CRANDC XO19(129)
359#define CRNAND XO19(225)
360#define CROR   XO19(449)
361#define CRNOR  XO19( 33)
362
363#define EXTSB  XO31(954)
364#define EXTSH  XO31(922)
365#define EXTSW  XO31(986)
366#define ADD    XO31(266)
367#define ADDE   XO31(138)
368#define ADDME  XO31(234)
369#define ADDZE  XO31(202)
370#define ADDC   XO31( 10)
371#define AND    XO31( 28)
372#define SUBF   XO31( 40)
373#define SUBFC  XO31(  8)
374#define SUBFE  XO31(136)
375#define SUBFME XO31(232)
376#define SUBFZE XO31(200)
377#define OR     XO31(444)
378#define XOR    XO31(316)
379#define MULLW  XO31(235)
380#define MULHW  XO31( 75)
381#define MULHWU XO31( 11)
382#define DIVW   XO31(491)
383#define DIVWU  XO31(459)
384#define MODSW  XO31(779)
385#define MODUW  XO31(267)
386#define CMP    XO31(  0)
387#define CMPL   XO31( 32)
388#define LHBRX  XO31(790)
389#define LWBRX  XO31(534)
390#define LDBRX  XO31(532)
391#define STHBRX XO31(918)
392#define STWBRX XO31(662)
393#define STDBRX XO31(660)
394#define MFSPR  XO31(339)
395#define MTSPR  XO31(467)
396#define SRAWI  XO31(824)
397#define NEG    XO31(104)
398#define MFCR   XO31( 19)
399#define MFOCRF (MFCR | (1u << 20))
400#define NOR    XO31(124)
401#define CNTLZW XO31( 26)
402#define CNTLZD XO31( 58)
403#define CNTTZW XO31(538)
404#define CNTTZD XO31(570)
405#define CNTPOPW XO31(378)
406#define CNTPOPD XO31(506)
407#define ANDC   XO31( 60)
408#define ORC    XO31(412)
409#define EQV    XO31(284)
410#define NAND   XO31(476)
411#define ISEL   XO31( 15)
412
413#define MULLD  XO31(233)
414#define MULHD  XO31( 73)
415#define MULHDU XO31(  9)
416#define DIVD   XO31(489)
417#define DIVDU  XO31(457)
418#define MODSD  XO31(777)
419#define MODUD  XO31(265)
420
421#define LBZX   XO31( 87)
422#define LHZX   XO31(279)
423#define LHAX   XO31(343)
424#define LWZX   XO31( 23)
425#define STBX   XO31(215)
426#define STHX   XO31(407)
427#define STWX   XO31(151)
428
429#define EIEIO  XO31(854)
430#define HWSYNC XO31(598)
431#define LWSYNC (HWSYNC | (1u << 21))
432
433#define SPR(a, b) ((((a)<<5)|(b))<<11)
434#define LR     SPR(8, 0)
435#define CTR    SPR(9, 0)
436
437#define SLW    XO31( 24)
438#define SRW    XO31(536)
439#define SRAW   XO31(792)
440
441#define SLD    XO31( 27)
442#define SRD    XO31(539)
443#define SRAD   XO31(794)
444#define SRADI  XO31(413<<1)
445
446#define BRH    XO31(219)
447#define BRW    XO31(155)
448#define BRD    XO31(187)
449
450#define TW     XO31( 4)
451#define TRAP   (TW | TO(31))
452
453#define NOP    ORI  /* ori 0,0,0 */
454
455#define LVX        XO31(103)
456#define LVEBX      XO31(7)
457#define LVEHX      XO31(39)
458#define LVEWX      XO31(71)
459#define LXSDX      (XO31(588) | 1)  /* v2.06, force tx=1 */
460#define LXVDSX     (XO31(332) | 1)  /* v2.06, force tx=1 */
461#define LXSIWZX    (XO31(12) | 1)   /* v2.07, force tx=1 */
462#define LXV        (OPCD(61) | 8 | 1)  /* v3.00, force tx=1 */
463#define LXSD       (OPCD(57) | 2)   /* v3.00 */
464#define LXVWSX     (XO31(364) | 1)  /* v3.00, force tx=1 */
465
466#define STVX       XO31(231)
467#define STVEWX     XO31(199)
468#define STXSDX     (XO31(716) | 1)  /* v2.06, force sx=1 */
469#define STXSIWX    (XO31(140) | 1)  /* v2.07, force sx=1 */
470#define STXV       (OPCD(61) | 8 | 5) /* v3.00, force sx=1 */
471#define STXSD      (OPCD(61) | 2)   /* v3.00 */
472
473#define VADDSBS    VX4(768)
474#define VADDUBS    VX4(512)
475#define VADDUBM    VX4(0)
476#define VADDSHS    VX4(832)
477#define VADDUHS    VX4(576)
478#define VADDUHM    VX4(64)
479#define VADDSWS    VX4(896)
480#define VADDUWS    VX4(640)
481#define VADDUWM    VX4(128)
482#define VADDUDM    VX4(192)       /* v2.07 */
483
484#define VSUBSBS    VX4(1792)
485#define VSUBUBS    VX4(1536)
486#define VSUBUBM    VX4(1024)
487#define VSUBSHS    VX4(1856)
488#define VSUBUHS    VX4(1600)
489#define VSUBUHM    VX4(1088)
490#define VSUBSWS    VX4(1920)
491#define VSUBUWS    VX4(1664)
492#define VSUBUWM    VX4(1152)
493#define VSUBUDM    VX4(1216)      /* v2.07 */
494
495#define VNEGW      (VX4(1538) | (6 << 16))  /* v3.00 */
496#define VNEGD      (VX4(1538) | (7 << 16))  /* v3.00 */
497
498#define VMAXSB     VX4(258)
499#define VMAXSH     VX4(322)
500#define VMAXSW     VX4(386)
501#define VMAXSD     VX4(450)       /* v2.07 */
502#define VMAXUB     VX4(2)
503#define VMAXUH     VX4(66)
504#define VMAXUW     VX4(130)
505#define VMAXUD     VX4(194)       /* v2.07 */
506#define VMINSB     VX4(770)
507#define VMINSH     VX4(834)
508#define VMINSW     VX4(898)
509#define VMINSD     VX4(962)       /* v2.07 */
510#define VMINUB     VX4(514)
511#define VMINUH     VX4(578)
512#define VMINUW     VX4(642)
513#define VMINUD     VX4(706)       /* v2.07 */
514
515#define VCMPEQUB   VX4(6)
516#define VCMPEQUH   VX4(70)
517#define VCMPEQUW   VX4(134)
518#define VCMPEQUD   VX4(199)       /* v2.07 */
519#define VCMPGTSB   VX4(774)
520#define VCMPGTSH   VX4(838)
521#define VCMPGTSW   VX4(902)
522#define VCMPGTSD   VX4(967)       /* v2.07 */
523#define VCMPGTUB   VX4(518)
524#define VCMPGTUH   VX4(582)
525#define VCMPGTUW   VX4(646)
526#define VCMPGTUD   VX4(711)       /* v2.07 */
527#define VCMPNEB    VX4(7)         /* v3.00 */
528#define VCMPNEH    VX4(71)        /* v3.00 */
529#define VCMPNEW    VX4(135)       /* v3.00 */
530
531#define VSLB       VX4(260)
532#define VSLH       VX4(324)
533#define VSLW       VX4(388)
534#define VSLD       VX4(1476)      /* v2.07 */
535#define VSRB       VX4(516)
536#define VSRH       VX4(580)
537#define VSRW       VX4(644)
538#define VSRD       VX4(1732)      /* v2.07 */
539#define VSRAB      VX4(772)
540#define VSRAH      VX4(836)
541#define VSRAW      VX4(900)
542#define VSRAD      VX4(964)       /* v2.07 */
543#define VRLB       VX4(4)
544#define VRLH       VX4(68)
545#define VRLW       VX4(132)
546#define VRLD       VX4(196)       /* v2.07 */
547
548#define VMULEUB    VX4(520)
549#define VMULEUH    VX4(584)
550#define VMULEUW    VX4(648)       /* v2.07 */
551#define VMULOUB    VX4(8)
552#define VMULOUH    VX4(72)
553#define VMULOUW    VX4(136)       /* v2.07 */
554#define VMULUWM    VX4(137)       /* v2.07 */
555#define VMULLD     VX4(457)       /* v3.10 */
556#define VMSUMUHM   VX4(38)
557
558#define VMRGHB     VX4(12)
559#define VMRGHH     VX4(76)
560#define VMRGHW     VX4(140)
561#define VMRGLB     VX4(268)
562#define VMRGLH     VX4(332)
563#define VMRGLW     VX4(396)
564
565#define VPKUHUM    VX4(14)
566#define VPKUWUM    VX4(78)
567
568#define VAND       VX4(1028)
569#define VANDC      VX4(1092)
570#define VNOR       VX4(1284)
571#define VOR        VX4(1156)
572#define VXOR       VX4(1220)
573#define VEQV       VX4(1668)      /* v2.07 */
574#define VNAND      VX4(1412)      /* v2.07 */
575#define VORC       VX4(1348)      /* v2.07 */
576
577#define VSPLTB     VX4(524)
578#define VSPLTH     VX4(588)
579#define VSPLTW     VX4(652)
580#define VSPLTISB   VX4(780)
581#define VSPLTISH   VX4(844)
582#define VSPLTISW   VX4(908)
583
584#define VSLDOI     VX4(44)
585
586#define XXPERMDI   (OPCD(60) | (10 << 3) | 7)  /* v2.06, force ax=bx=tx=1 */
587#define XXSEL      (OPCD(60) | (3 << 4) | 0xf) /* v2.06, force ax=bx=cx=tx=1 */
588#define XXSPLTIB   (OPCD(60) | (360 << 1) | 1) /* v3.00, force tx=1 */
589
590#define MFVSRD     (XO31(51) | 1)   /* v2.07, force sx=1 */
591#define MFVSRWZ    (XO31(115) | 1)  /* v2.07, force sx=1 */
592#define MTVSRD     (XO31(179) | 1)  /* v2.07, force tx=1 */
593#define MTVSRWZ    (XO31(243) | 1)  /* v2.07, force tx=1 */
594#define MTVSRDD    (XO31(435) | 1)  /* v3.00, force tx=1 */
595#define MTVSRWS    (XO31(403) | 1)  /* v3.00, force tx=1 */
596
597#define RT(r) ((r)<<21)
598#define RS(r) ((r)<<21)
599#define RA(r) ((r)<<16)
600#define RB(r) ((r)<<11)
601#define TO(t) ((t)<<21)
602#define SH(s) ((s)<<11)
603#define MB(b) ((b)<<6)
604#define ME(e) ((e)<<1)
605#define BO(o) ((o)<<21)
606#define MB64(b) ((b)<<5)
607#define FXM(b) (1 << (19 - (b)))
608
609#define VRT(r)  (((r) & 31) << 21)
610#define VRA(r)  (((r) & 31) << 16)
611#define VRB(r)  (((r) & 31) << 11)
612#define VRC(r)  (((r) & 31) <<  6)
613
614#define LK    1
615
616#define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
617#define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
618#define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
619#define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
620
621#define BF(n)    ((n)<<23)
622#define BI(n, c) (((c)+((n)*4))<<16)
623#define BT(n, c) (((c)+((n)*4))<<21)
624#define BA(n, c) (((c)+((n)*4))<<16)
625#define BB(n, c) (((c)+((n)*4))<<11)
626#define BC_(n, c) (((c)+((n)*4))<<6)
627
628#define BO_COND_TRUE  BO(12)
629#define BO_COND_FALSE BO( 4)
630#define BO_ALWAYS     BO(20)
631
632enum {
633    CR_LT,
634    CR_GT,
635    CR_EQ,
636    CR_SO
637};
638
639static const uint32_t tcg_to_bc[] = {
640    [TCG_COND_EQ]  = BC | BI(7, CR_EQ) | BO_COND_TRUE,
641    [TCG_COND_NE]  = BC | BI(7, CR_EQ) | BO_COND_FALSE,
642    [TCG_COND_LT]  = BC | BI(7, CR_LT) | BO_COND_TRUE,
643    [TCG_COND_GE]  = BC | BI(7, CR_LT) | BO_COND_FALSE,
644    [TCG_COND_LE]  = BC | BI(7, CR_GT) | BO_COND_FALSE,
645    [TCG_COND_GT]  = BC | BI(7, CR_GT) | BO_COND_TRUE,
646    [TCG_COND_LTU] = BC | BI(7, CR_LT) | BO_COND_TRUE,
647    [TCG_COND_GEU] = BC | BI(7, CR_LT) | BO_COND_FALSE,
648    [TCG_COND_LEU] = BC | BI(7, CR_GT) | BO_COND_FALSE,
649    [TCG_COND_GTU] = BC | BI(7, CR_GT) | BO_COND_TRUE,
650};
651
652/* The low bit here is set if the RA and RB fields must be inverted.  */
653static const uint32_t tcg_to_isel[] = {
654    [TCG_COND_EQ]  = ISEL | BC_(7, CR_EQ),
655    [TCG_COND_NE]  = ISEL | BC_(7, CR_EQ) | 1,
656    [TCG_COND_LT]  = ISEL | BC_(7, CR_LT),
657    [TCG_COND_GE]  = ISEL | BC_(7, CR_LT) | 1,
658    [TCG_COND_LE]  = ISEL | BC_(7, CR_GT) | 1,
659    [TCG_COND_GT]  = ISEL | BC_(7, CR_GT),
660    [TCG_COND_LTU] = ISEL | BC_(7, CR_LT),
661    [TCG_COND_GEU] = ISEL | BC_(7, CR_LT) | 1,
662    [TCG_COND_LEU] = ISEL | BC_(7, CR_GT) | 1,
663    [TCG_COND_GTU] = ISEL | BC_(7, CR_GT),
664};
665
666static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
667                        intptr_t value, intptr_t addend)
668{
669    const tcg_insn_unit *target;
670    int16_t lo;
671    int32_t hi;
672
673    value += addend;
674    target = (const tcg_insn_unit *)value;
675
676    switch (type) {
677    case R_PPC_REL14:
678        return reloc_pc14(code_ptr, target);
679    case R_PPC_REL24:
680        return reloc_pc24(code_ptr, target);
681    case R_PPC_ADDR16:
682        /*
683         * We are (slightly) abusing this relocation type.  In particular,
684         * assert that the low 2 bits are zero, and do not modify them.
685         * That way we can use this with LD et al that have opcode bits
686         * in the low 2 bits of the insn.
687         */
688        if ((value & 3) || value != (int16_t)value) {
689            return false;
690        }
691        *code_ptr = (*code_ptr & ~0xfffc) | (value & 0xfffc);
692        break;
693    case R_PPC_ADDR32:
694        /*
695         * We are abusing this relocation type.  Again, this points to
696         * a pair of insns, lis + load.  This is an absolute address
697         * relocation for PPC32 so the lis cannot be removed.
698         */
699        lo = value;
700        hi = value - lo;
701        if (hi + lo != value) {
702            return false;
703        }
704        code_ptr[0] = deposit32(code_ptr[0], 0, 16, hi >> 16);
705        code_ptr[1] = deposit32(code_ptr[1], 0, 16, lo);
706        break;
707    default:
708        g_assert_not_reached();
709    }
710    return true;
711}
712
713static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
714                             TCGReg base, tcg_target_long offset);
715
716static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
717{
718    if (ret == arg) {
719        return true;
720    }
721    switch (type) {
722    case TCG_TYPE_I64:
723        tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
724        /* fallthru */
725    case TCG_TYPE_I32:
726        if (ret < TCG_REG_V0) {
727            if (arg < TCG_REG_V0) {
728                tcg_out32(s, OR | SAB(arg, ret, arg));
729                break;
730            } else if (have_isa_2_07) {
731                tcg_out32(s, (type == TCG_TYPE_I32 ? MFVSRWZ : MFVSRD)
732                          | VRT(arg) | RA(ret));
733                break;
734            } else {
735                /* Altivec does not support vector->integer moves.  */
736                return false;
737            }
738        } else if (arg < TCG_REG_V0) {
739            if (have_isa_2_07) {
740                tcg_out32(s, (type == TCG_TYPE_I32 ? MTVSRWZ : MTVSRD)
741                          | VRT(ret) | RA(arg));
742                break;
743            } else {
744                /* Altivec does not support integer->vector moves.  */
745                return false;
746            }
747        }
748        /* fallthru */
749    case TCG_TYPE_V64:
750    case TCG_TYPE_V128:
751        tcg_debug_assert(ret >= TCG_REG_V0 && arg >= TCG_REG_V0);
752        tcg_out32(s, VOR | VRT(ret) | VRA(arg) | VRB(arg));
753        break;
754    default:
755        g_assert_not_reached();
756    }
757    return true;
758}
759
760static inline void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs,
761                               int sh, int mb)
762{
763    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
764    sh = SH(sh & 0x1f) | (((sh >> 5) & 1) << 1);
765    mb = MB64((mb >> 5) | ((mb << 1) & 0x3f));
766    tcg_out32(s, op | RA(ra) | RS(rs) | sh | mb);
767}
768
769static inline void tcg_out_rlw(TCGContext *s, int op, TCGReg ra, TCGReg rs,
770                               int sh, int mb, int me)
771{
772    tcg_out32(s, op | RA(ra) | RS(rs) | SH(sh) | MB(mb) | ME(me));
773}
774
775static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
776{
777    tcg_out32(s, EXTSB | RA(dst) | RS(src));
778}
779
780static void tcg_out_ext8u(TCGContext *s, TCGReg dst, TCGReg src)
781{
782    tcg_out32(s, ANDI | SAI(src, dst, 0xff));
783}
784
785static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
786{
787    tcg_out32(s, EXTSH | RA(dst) | RS(src));
788}
789
790static void tcg_out_ext16u(TCGContext *s, TCGReg dst, TCGReg src)
791{
792    tcg_out32(s, ANDI | SAI(src, dst, 0xffff));
793}
794
795static void tcg_out_ext32s(TCGContext *s, TCGReg dst, TCGReg src)
796{
797    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
798    tcg_out32(s, EXTSW | RA(dst) | RS(src));
799}
800
801static void tcg_out_ext32u(TCGContext *s, TCGReg dst, TCGReg src)
802{
803    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
804    tcg_out_rld(s, RLDICL, dst, src, 0, 32);
805}
806
807static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dst, TCGReg src)
808{
809    tcg_out_ext32s(s, dst, src);
810}
811
812static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dst, TCGReg src)
813{
814    tcg_out_ext32u(s, dst, src);
815}
816
817static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rn)
818{
819    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
820    tcg_out_mov(s, TCG_TYPE_I32, rd, rn);
821}
822
823static inline void tcg_out_shli32(TCGContext *s, TCGReg dst, TCGReg src, int c)
824{
825    tcg_out_rlw(s, RLWINM, dst, src, c, 0, 31 - c);
826}
827
828static inline void tcg_out_shli64(TCGContext *s, TCGReg dst, TCGReg src, int c)
829{
830    tcg_out_rld(s, RLDICR, dst, src, c, 63 - c);
831}
832
833static inline void tcg_out_sari32(TCGContext *s, TCGReg dst, TCGReg src, int c)
834{
835    /* Limit immediate shift count lest we create an illegal insn.  */
836    tcg_out32(s, SRAWI | RA(dst) | RS(src) | SH(c & 31));
837}
838
839static inline void tcg_out_shri32(TCGContext *s, TCGReg dst, TCGReg src, int c)
840{
841    tcg_out_rlw(s, RLWINM, dst, src, 32 - c, c, 31);
842}
843
844static inline void tcg_out_shri64(TCGContext *s, TCGReg dst, TCGReg src, int c)
845{
846    tcg_out_rld(s, RLDICL, dst, src, 64 - c, c);
847}
848
849static inline void tcg_out_sari64(TCGContext *s, TCGReg dst, TCGReg src, int c)
850{
851    tcg_out32(s, SRADI | RA(dst) | RS(src) | SH(c & 0x1f) | ((c >> 4) & 2));
852}
853
854static void tcg_out_bswap16(TCGContext *s, TCGReg dst, TCGReg src, int flags)
855{
856    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
857
858    if (have_isa_3_10) {
859        tcg_out32(s, BRH | RA(dst) | RS(src));
860        if (flags & TCG_BSWAP_OS) {
861            tcg_out_ext16s(s, TCG_TYPE_REG, dst, dst);
862        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
863            tcg_out_ext16u(s, dst, dst);
864        }
865        return;
866    }
867
868    /*
869     * In the following,
870     *   dep(a, b, m) -> (a & ~m) | (b & m)
871     *
872     * Begin with:                              src = xxxxabcd
873     */
874    /* tmp = rol32(src, 24) & 0x000000ff            = 0000000c */
875    tcg_out_rlw(s, RLWINM, tmp, src, 24, 24, 31);
876    /* tmp = dep(tmp, rol32(src, 8), 0x0000ff00)    = 000000dc */
877    tcg_out_rlw(s, RLWIMI, tmp, src, 8, 16, 23);
878
879    if (flags & TCG_BSWAP_OS) {
880        tcg_out_ext16s(s, TCG_TYPE_REG, dst, tmp);
881    } else {
882        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
883    }
884}
885
886static void tcg_out_bswap32(TCGContext *s, TCGReg dst, TCGReg src, int flags)
887{
888    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
889
890    if (have_isa_3_10) {
891        tcg_out32(s, BRW | RA(dst) | RS(src));
892        if (flags & TCG_BSWAP_OS) {
893            tcg_out_ext32s(s, dst, dst);
894        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
895            tcg_out_ext32u(s, dst, dst);
896        }
897        return;
898    }
899
900    /*
901     * Stolen from gcc's builtin_bswap32.
902     * In the following,
903     *   dep(a, b, m) -> (a & ~m) | (b & m)
904     *
905     * Begin with:                              src = xxxxabcd
906     */
907    /* tmp = rol32(src, 8) & 0xffffffff             = 0000bcda */
908    tcg_out_rlw(s, RLWINM, tmp, src, 8, 0, 31);
909    /* tmp = dep(tmp, rol32(src, 24), 0xff000000)   = 0000dcda */
910    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 0, 7);
911    /* tmp = dep(tmp, rol32(src, 24), 0x0000ff00)   = 0000dcba */
912    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 16, 23);
913
914    if (flags & TCG_BSWAP_OS) {
915        tcg_out_ext32s(s, dst, tmp);
916    } else {
917        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
918    }
919}
920
921static void tcg_out_bswap64(TCGContext *s, TCGReg dst, TCGReg src)
922{
923    TCGReg t0 = dst == src ? TCG_REG_R0 : dst;
924    TCGReg t1 = dst == src ? dst : TCG_REG_R0;
925
926    if (have_isa_3_10) {
927        tcg_out32(s, BRD | RA(dst) | RS(src));
928        return;
929    }
930
931    /*
932     * In the following,
933     *   dep(a, b, m) -> (a & ~m) | (b & m)
934     *
935     * Begin with:                              src = abcdefgh
936     */
937    /* t0 = rol32(src, 8) & 0xffffffff              = 0000fghe */
938    tcg_out_rlw(s, RLWINM, t0, src, 8, 0, 31);
939    /* t0 = dep(t0, rol32(src, 24), 0xff000000)     = 0000hghe */
940    tcg_out_rlw(s, RLWIMI, t0, src, 24, 0, 7);
941    /* t0 = dep(t0, rol32(src, 24), 0x0000ff00)     = 0000hgfe */
942    tcg_out_rlw(s, RLWIMI, t0, src, 24, 16, 23);
943
944    /* t0 = rol64(t0, 32)                           = hgfe0000 */
945    tcg_out_rld(s, RLDICL, t0, t0, 32, 0);
946    /* t1 = rol64(src, 32)                          = efghabcd */
947    tcg_out_rld(s, RLDICL, t1, src, 32, 0);
948
949    /* t0 = dep(t0, rol32(t1, 24), 0xffffffff)      = hgfebcda */
950    tcg_out_rlw(s, RLWIMI, t0, t1, 8, 0, 31);
951    /* t0 = dep(t0, rol32(t1, 24), 0xff000000)      = hgfedcda */
952    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 0, 7);
953    /* t0 = dep(t0, rol32(t1, 24), 0x0000ff00)      = hgfedcba */
954    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 16, 23);
955
956    tcg_out_mov(s, TCG_TYPE_REG, dst, t0);
957}
958
959/* Emit a move into ret of arg, if it can be done in one insn.  */
960static bool tcg_out_movi_one(TCGContext *s, TCGReg ret, tcg_target_long arg)
961{
962    if (arg == (int16_t)arg) {
963        tcg_out32(s, ADDI | TAI(ret, 0, arg));
964        return true;
965    }
966    if (arg == (int32_t)arg && (arg & 0xffff) == 0) {
967        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
968        return true;
969    }
970    return false;
971}
972
973static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
974                             tcg_target_long arg, bool in_prologue)
975{
976    intptr_t tb_diff;
977    tcg_target_long tmp;
978    int shift;
979
980    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
981
982    if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
983        arg = (int32_t)arg;
984    }
985
986    /* Load 16-bit immediates with one insn.  */
987    if (tcg_out_movi_one(s, ret, arg)) {
988        return;
989    }
990
991    /* Load addresses within the TB with one insn.  */
992    tb_diff = tcg_tbrel_diff(s, (void *)arg);
993    if (!in_prologue && USE_REG_TB && tb_diff == (int16_t)tb_diff) {
994        tcg_out32(s, ADDI | TAI(ret, TCG_REG_TB, tb_diff));
995        return;
996    }
997
998    /* Load 32-bit immediates with two insns.  Note that we've already
999       eliminated bare ADDIS, so we know both insns are required.  */
1000    if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) {
1001        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
1002        tcg_out32(s, ORI | SAI(ret, ret, arg));
1003        return;
1004    }
1005    if (arg == (uint32_t)arg && !(arg & 0x8000)) {
1006        tcg_out32(s, ADDI | TAI(ret, 0, arg));
1007        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
1008        return;
1009    }
1010
1011    /* Load masked 16-bit value.  */
1012    if (arg > 0 && (arg & 0x8000)) {
1013        tmp = arg | 0x7fff;
1014        if ((tmp & (tmp + 1)) == 0) {
1015            int mb = clz64(tmp + 1) + 1;
1016            tcg_out32(s, ADDI | TAI(ret, 0, arg));
1017            tcg_out_rld(s, RLDICL, ret, ret, 0, mb);
1018            return;
1019        }
1020    }
1021
1022    /* Load common masks with 2 insns.  */
1023    shift = ctz64(arg);
1024    tmp = arg >> shift;
1025    if (tmp == (int16_t)tmp) {
1026        tcg_out32(s, ADDI | TAI(ret, 0, tmp));
1027        tcg_out_shli64(s, ret, ret, shift);
1028        return;
1029    }
1030    shift = clz64(arg);
1031    if (tcg_out_movi_one(s, ret, arg << shift)) {
1032        tcg_out_shri64(s, ret, ret, shift);
1033        return;
1034    }
1035
1036    /* Load addresses within 2GB of TB with 2 (or rarely 3) insns.  */
1037    if (!in_prologue && USE_REG_TB && tb_diff == (int32_t)tb_diff) {
1038        tcg_out_mem_long(s, ADDI, ADD, ret, TCG_REG_TB, tb_diff);
1039        return;
1040    }
1041
1042    /* Use the constant pool, if possible.  */
1043    if (!in_prologue && USE_REG_TB) {
1044        new_pool_label(s, arg, R_PPC_ADDR16, s->code_ptr,
1045                       tcg_tbrel_diff(s, NULL));
1046        tcg_out32(s, LD | TAI(ret, TCG_REG_TB, 0));
1047        return;
1048    }
1049
1050    tmp = arg >> 31 >> 1;
1051    tcg_out_movi(s, TCG_TYPE_I32, ret, tmp);
1052    if (tmp) {
1053        tcg_out_shli64(s, ret, ret, 32);
1054    }
1055    if (arg & 0xffff0000) {
1056        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
1057    }
1058    if (arg & 0xffff) {
1059        tcg_out32(s, ORI | SAI(ret, ret, arg));
1060    }
1061}
1062
1063static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
1064                             TCGReg ret, int64_t val)
1065{
1066    uint32_t load_insn;
1067    int rel, low;
1068    intptr_t add;
1069
1070    switch (vece) {
1071    case MO_8:
1072        low = (int8_t)val;
1073        if (low >= -16 && low < 16) {
1074            tcg_out32(s, VSPLTISB | VRT(ret) | ((val & 31) << 16));
1075            return;
1076        }
1077        if (have_isa_3_00) {
1078            tcg_out32(s, XXSPLTIB | VRT(ret) | ((val & 0xff) << 11));
1079            return;
1080        }
1081        break;
1082
1083    case MO_16:
1084        low = (int16_t)val;
1085        if (low >= -16 && low < 16) {
1086            tcg_out32(s, VSPLTISH | VRT(ret) | ((val & 31) << 16));
1087            return;
1088        }
1089        break;
1090
1091    case MO_32:
1092        low = (int32_t)val;
1093        if (low >= -16 && low < 16) {
1094            tcg_out32(s, VSPLTISW | VRT(ret) | ((val & 31) << 16));
1095            return;
1096        }
1097        break;
1098    }
1099
1100    /*
1101     * Otherwise we must load the value from the constant pool.
1102     */
1103    if (USE_REG_TB) {
1104        rel = R_PPC_ADDR16;
1105        add = tcg_tbrel_diff(s, NULL);
1106    } else {
1107        rel = R_PPC_ADDR32;
1108        add = 0;
1109    }
1110
1111    if (have_vsx) {
1112        load_insn = type == TCG_TYPE_V64 ? LXSDX : LXVDSX;
1113        load_insn |= VRT(ret) | RB(TCG_REG_TMP1);
1114        if (TCG_TARGET_REG_BITS == 64) {
1115            new_pool_label(s, val, rel, s->code_ptr, add);
1116        } else {
1117            new_pool_l2(s, rel, s->code_ptr, add, val >> 32, val);
1118        }
1119    } else {
1120        load_insn = LVX | VRT(ret) | RB(TCG_REG_TMP1);
1121        if (TCG_TARGET_REG_BITS == 64) {
1122            new_pool_l2(s, rel, s->code_ptr, add, val, val);
1123        } else {
1124            new_pool_l4(s, rel, s->code_ptr, add,
1125                        val >> 32, val, val >> 32, val);
1126        }
1127    }
1128
1129    if (USE_REG_TB) {
1130        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, 0, 0));
1131        load_insn |= RA(TCG_REG_TB);
1132    } else {
1133        tcg_out32(s, ADDIS | TAI(TCG_REG_TMP1, 0, 0));
1134        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, TCG_REG_TMP1, 0));
1135    }
1136    tcg_out32(s, load_insn);
1137}
1138
1139static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,
1140                         tcg_target_long arg)
1141{
1142    switch (type) {
1143    case TCG_TYPE_I32:
1144    case TCG_TYPE_I64:
1145        tcg_debug_assert(ret < TCG_REG_V0);
1146        tcg_out_movi_int(s, type, ret, arg, false);
1147        break;
1148
1149    default:
1150        g_assert_not_reached();
1151    }
1152}
1153
1154static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
1155{
1156    return false;
1157}
1158
1159static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
1160                             tcg_target_long imm)
1161{
1162    /* This function is only used for passing structs by reference. */
1163    g_assert_not_reached();
1164}
1165
1166static bool mask_operand(uint32_t c, int *mb, int *me)
1167{
1168    uint32_t lsb, test;
1169
1170    /* Accept a bit pattern like:
1171           0....01....1
1172           1....10....0
1173           0..01..10..0
1174       Keep track of the transitions.  */
1175    if (c == 0 || c == -1) {
1176        return false;
1177    }
1178    test = c;
1179    lsb = test & -test;
1180    test += lsb;
1181    if (test & (test - 1)) {
1182        return false;
1183    }
1184
1185    *me = clz32(lsb);
1186    *mb = test ? clz32(test & -test) + 1 : 0;
1187    return true;
1188}
1189
1190static bool mask64_operand(uint64_t c, int *mb, int *me)
1191{
1192    uint64_t lsb;
1193
1194    if (c == 0) {
1195        return false;
1196    }
1197
1198    lsb = c & -c;
1199    /* Accept 1..10..0.  */
1200    if (c == -lsb) {
1201        *mb = 0;
1202        *me = clz64(lsb);
1203        return true;
1204    }
1205    /* Accept 0..01..1.  */
1206    if (lsb == 1 && (c & (c + 1)) == 0) {
1207        *mb = clz64(c + 1) + 1;
1208        *me = 63;
1209        return true;
1210    }
1211    return false;
1212}
1213
1214static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1215{
1216    int mb, me;
1217
1218    if (mask_operand(c, &mb, &me)) {
1219        tcg_out_rlw(s, RLWINM, dst, src, 0, mb, me);
1220    } else if ((c & 0xffff) == c) {
1221        tcg_out32(s, ANDI | SAI(src, dst, c));
1222        return;
1223    } else if ((c & 0xffff0000) == c) {
1224        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1225        return;
1226    } else {
1227        tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R0, c);
1228        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1229    }
1230}
1231
1232static void tcg_out_andi64(TCGContext *s, TCGReg dst, TCGReg src, uint64_t c)
1233{
1234    int mb, me;
1235
1236    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1237    if (mask64_operand(c, &mb, &me)) {
1238        if (mb == 0) {
1239            tcg_out_rld(s, RLDICR, dst, src, 0, me);
1240        } else {
1241            tcg_out_rld(s, RLDICL, dst, src, 0, mb);
1242        }
1243    } else if ((c & 0xffff) == c) {
1244        tcg_out32(s, ANDI | SAI(src, dst, c));
1245        return;
1246    } else if ((c & 0xffff0000) == c) {
1247        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1248        return;
1249    } else {
1250        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, c);
1251        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1252    }
1253}
1254
1255static void tcg_out_zori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c,
1256                           int op_lo, int op_hi)
1257{
1258    if (c >> 16) {
1259        tcg_out32(s, op_hi | SAI(src, dst, c >> 16));
1260        src = dst;
1261    }
1262    if (c & 0xffff) {
1263        tcg_out32(s, op_lo | SAI(src, dst, c));
1264        src = dst;
1265    }
1266}
1267
1268static void tcg_out_ori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1269{
1270    tcg_out_zori32(s, dst, src, c, ORI, ORIS);
1271}
1272
1273static void tcg_out_xori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1274{
1275    tcg_out_zori32(s, dst, src, c, XORI, XORIS);
1276}
1277
1278static void tcg_out_b(TCGContext *s, int mask, const tcg_insn_unit *target)
1279{
1280    ptrdiff_t disp = tcg_pcrel_diff(s, target);
1281    if (in_range_b(disp)) {
1282        tcg_out32(s, B | (disp & 0x3fffffc) | mask);
1283    } else {
1284        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, (uintptr_t)target);
1285        tcg_out32(s, MTSPR | RS(TCG_REG_R0) | CTR);
1286        tcg_out32(s, BCCTR | BO_ALWAYS | mask);
1287    }
1288}
1289
1290static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
1291                             TCGReg base, tcg_target_long offset)
1292{
1293    tcg_target_long orig = offset, l0, l1, extra = 0, align = 0;
1294    bool is_int_store = false;
1295    TCGReg rs = TCG_REG_TMP1;
1296
1297    switch (opi) {
1298    case LD: case LWA:
1299        align = 3;
1300        /* FALLTHRU */
1301    default:
1302        if (rt > TCG_REG_R0 && rt < TCG_REG_V0) {
1303            rs = rt;
1304            break;
1305        }
1306        break;
1307    case LXSD:
1308    case STXSD:
1309        align = 3;
1310        break;
1311    case LXV:
1312    case STXV:
1313        align = 15;
1314        break;
1315    case STD:
1316        align = 3;
1317        /* FALLTHRU */
1318    case STB: case STH: case STW:
1319        is_int_store = true;
1320        break;
1321    }
1322
1323    /* For unaligned, or very large offsets, use the indexed form.  */
1324    if (offset & align || offset != (int32_t)offset || opi == 0) {
1325        if (rs == base) {
1326            rs = TCG_REG_R0;
1327        }
1328        tcg_debug_assert(!is_int_store || rs != rt);
1329        tcg_out_movi(s, TCG_TYPE_PTR, rs, orig);
1330        tcg_out32(s, opx | TAB(rt & 31, base, rs));
1331        return;
1332    }
1333
1334    l0 = (int16_t)offset;
1335    offset = (offset - l0) >> 16;
1336    l1 = (int16_t)offset;
1337
1338    if (l1 < 0 && orig >= 0) {
1339        extra = 0x4000;
1340        l1 = (int16_t)(offset - 0x4000);
1341    }
1342    if (l1) {
1343        tcg_out32(s, ADDIS | TAI(rs, base, l1));
1344        base = rs;
1345    }
1346    if (extra) {
1347        tcg_out32(s, ADDIS | TAI(rs, base, extra));
1348        base = rs;
1349    }
1350    if (opi != ADDI || base != rt || l0 != 0) {
1351        tcg_out32(s, opi | TAI(rt & 31, base, l0));
1352    }
1353}
1354
1355static void tcg_out_vsldoi(TCGContext *s, TCGReg ret,
1356                           TCGReg va, TCGReg vb, int shb)
1357{
1358    tcg_out32(s, VSLDOI | VRT(ret) | VRA(va) | VRB(vb) | (shb << 6));
1359}
1360
1361static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
1362                       TCGReg base, intptr_t offset)
1363{
1364    int shift;
1365
1366    switch (type) {
1367    case TCG_TYPE_I32:
1368        if (ret < TCG_REG_V0) {
1369            tcg_out_mem_long(s, LWZ, LWZX, ret, base, offset);
1370            break;
1371        }
1372        if (have_isa_2_07 && have_vsx) {
1373            tcg_out_mem_long(s, 0, LXSIWZX, ret, base, offset);
1374            break;
1375        }
1376        tcg_debug_assert((offset & 3) == 0);
1377        tcg_out_mem_long(s, 0, LVEWX, ret, base, offset);
1378        shift = (offset - 4) & 0xc;
1379        if (shift) {
1380            tcg_out_vsldoi(s, ret, ret, ret, shift);
1381        }
1382        break;
1383    case TCG_TYPE_I64:
1384        if (ret < TCG_REG_V0) {
1385            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1386            tcg_out_mem_long(s, LD, LDX, ret, base, offset);
1387            break;
1388        }
1389        /* fallthru */
1390    case TCG_TYPE_V64:
1391        tcg_debug_assert(ret >= TCG_REG_V0);
1392        if (have_vsx) {
1393            tcg_out_mem_long(s, have_isa_3_00 ? LXSD : 0, LXSDX,
1394                             ret, base, offset);
1395            break;
1396        }
1397        tcg_debug_assert((offset & 7) == 0);
1398        tcg_out_mem_long(s, 0, LVX, ret, base, offset & -16);
1399        if (offset & 8) {
1400            tcg_out_vsldoi(s, ret, ret, ret, 8);
1401        }
1402        break;
1403    case TCG_TYPE_V128:
1404        tcg_debug_assert(ret >= TCG_REG_V0);
1405        tcg_debug_assert((offset & 15) == 0);
1406        tcg_out_mem_long(s, have_isa_3_00 ? LXV : 0,
1407                         LVX, ret, base, offset);
1408        break;
1409    default:
1410        g_assert_not_reached();
1411    }
1412}
1413
1414static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
1415                              TCGReg base, intptr_t offset)
1416{
1417    int shift;
1418
1419    switch (type) {
1420    case TCG_TYPE_I32:
1421        if (arg < TCG_REG_V0) {
1422            tcg_out_mem_long(s, STW, STWX, arg, base, offset);
1423            break;
1424        }
1425        if (have_isa_2_07 && have_vsx) {
1426            tcg_out_mem_long(s, 0, STXSIWX, arg, base, offset);
1427            break;
1428        }
1429        assert((offset & 3) == 0);
1430        tcg_debug_assert((offset & 3) == 0);
1431        shift = (offset - 4) & 0xc;
1432        if (shift) {
1433            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, shift);
1434            arg = TCG_VEC_TMP1;
1435        }
1436        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1437        break;
1438    case TCG_TYPE_I64:
1439        if (arg < TCG_REG_V0) {
1440            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1441            tcg_out_mem_long(s, STD, STDX, arg, base, offset);
1442            break;
1443        }
1444        /* fallthru */
1445    case TCG_TYPE_V64:
1446        tcg_debug_assert(arg >= TCG_REG_V0);
1447        if (have_vsx) {
1448            tcg_out_mem_long(s, have_isa_3_00 ? STXSD : 0,
1449                             STXSDX, arg, base, offset);
1450            break;
1451        }
1452        tcg_debug_assert((offset & 7) == 0);
1453        if (offset & 8) {
1454            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, 8);
1455            arg = TCG_VEC_TMP1;
1456        }
1457        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1458        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset + 4);
1459        break;
1460    case TCG_TYPE_V128:
1461        tcg_debug_assert(arg >= TCG_REG_V0);
1462        tcg_out_mem_long(s, have_isa_3_00 ? STXV : 0,
1463                         STVX, arg, base, offset);
1464        break;
1465    default:
1466        g_assert_not_reached();
1467    }
1468}
1469
1470static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1471                               TCGReg base, intptr_t ofs)
1472{
1473    return false;
1474}
1475
1476static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2,
1477                        int const_arg2, int cr, TCGType type)
1478{
1479    int imm;
1480    uint32_t op;
1481
1482    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1483
1484    /* Simplify the comparisons below wrt CMPI.  */
1485    if (type == TCG_TYPE_I32) {
1486        arg2 = (int32_t)arg2;
1487    }
1488
1489    switch (cond) {
1490    case TCG_COND_EQ:
1491    case TCG_COND_NE:
1492        if (const_arg2) {
1493            if ((int16_t) arg2 == arg2) {
1494                op = CMPI;
1495                imm = 1;
1496                break;
1497            } else if ((uint16_t) arg2 == arg2) {
1498                op = CMPLI;
1499                imm = 1;
1500                break;
1501            }
1502        }
1503        op = CMPL;
1504        imm = 0;
1505        break;
1506
1507    case TCG_COND_LT:
1508    case TCG_COND_GE:
1509    case TCG_COND_LE:
1510    case TCG_COND_GT:
1511        if (const_arg2) {
1512            if ((int16_t) arg2 == arg2) {
1513                op = CMPI;
1514                imm = 1;
1515                break;
1516            }
1517        }
1518        op = CMP;
1519        imm = 0;
1520        break;
1521
1522    case TCG_COND_LTU:
1523    case TCG_COND_GEU:
1524    case TCG_COND_LEU:
1525    case TCG_COND_GTU:
1526        if (const_arg2) {
1527            if ((uint16_t) arg2 == arg2) {
1528                op = CMPLI;
1529                imm = 1;
1530                break;
1531            }
1532        }
1533        op = CMPL;
1534        imm = 0;
1535        break;
1536
1537    default:
1538        g_assert_not_reached();
1539    }
1540    op |= BF(cr) | ((type == TCG_TYPE_I64) << 21);
1541
1542    if (imm) {
1543        tcg_out32(s, op | RA(arg1) | (arg2 & 0xffff));
1544    } else {
1545        if (const_arg2) {
1546            tcg_out_movi(s, type, TCG_REG_R0, arg2);
1547            arg2 = TCG_REG_R0;
1548        }
1549        tcg_out32(s, op | RA(arg1) | RB(arg2));
1550    }
1551}
1552
1553static void tcg_out_setcond_eq0(TCGContext *s, TCGType type,
1554                                TCGReg dst, TCGReg src)
1555{
1556    if (type == TCG_TYPE_I32) {
1557        tcg_out32(s, CNTLZW | RS(src) | RA(dst));
1558        tcg_out_shri32(s, dst, dst, 5);
1559    } else {
1560        tcg_out32(s, CNTLZD | RS(src) | RA(dst));
1561        tcg_out_shri64(s, dst, dst, 6);
1562    }
1563}
1564
1565static void tcg_out_setcond_ne0(TCGContext *s, TCGReg dst, TCGReg src)
1566{
1567    /* X != 0 implies X + -1 generates a carry.  Extra addition
1568       trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C.  */
1569    if (dst != src) {
1570        tcg_out32(s, ADDIC | TAI(dst, src, -1));
1571        tcg_out32(s, SUBFE | TAB(dst, dst, src));
1572    } else {
1573        tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1));
1574        tcg_out32(s, SUBFE | TAB(dst, TCG_REG_R0, src));
1575    }
1576}
1577
1578static TCGReg tcg_gen_setcond_xor(TCGContext *s, TCGReg arg1, TCGArg arg2,
1579                                  bool const_arg2)
1580{
1581    if (const_arg2) {
1582        if ((uint32_t)arg2 == arg2) {
1583            tcg_out_xori32(s, TCG_REG_R0, arg1, arg2);
1584        } else {
1585            tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, arg2);
1586            tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, TCG_REG_R0));
1587        }
1588    } else {
1589        tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, arg2));
1590    }
1591    return TCG_REG_R0;
1592}
1593
1594static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
1595                            TCGArg arg0, TCGArg arg1, TCGArg arg2,
1596                            int const_arg2)
1597{
1598    int crop, sh;
1599
1600    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1601
1602    /* Ignore high bits of a potential constant arg2.  */
1603    if (type == TCG_TYPE_I32) {
1604        arg2 = (uint32_t)arg2;
1605    }
1606
1607    /* Handle common and trivial cases before handling anything else.  */
1608    if (arg2 == 0) {
1609        switch (cond) {
1610        case TCG_COND_EQ:
1611            tcg_out_setcond_eq0(s, type, arg0, arg1);
1612            return;
1613        case TCG_COND_NE:
1614            if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1615                tcg_out_ext32u(s, TCG_REG_R0, arg1);
1616                arg1 = TCG_REG_R0;
1617            }
1618            tcg_out_setcond_ne0(s, arg0, arg1);
1619            return;
1620        case TCG_COND_GE:
1621            tcg_out32(s, NOR | SAB(arg1, arg0, arg1));
1622            arg1 = arg0;
1623            /* FALLTHRU */
1624        case TCG_COND_LT:
1625            /* Extract the sign bit.  */
1626            if (type == TCG_TYPE_I32) {
1627                tcg_out_shri32(s, arg0, arg1, 31);
1628            } else {
1629                tcg_out_shri64(s, arg0, arg1, 63);
1630            }
1631            return;
1632        default:
1633            break;
1634        }
1635    }
1636
1637    /* If we have ISEL, we can implement everything with 3 or 4 insns.
1638       All other cases below are also at least 3 insns, so speed up the
1639       code generator by not considering them and always using ISEL.  */
1640    if (have_isel) {
1641        int isel, tab;
1642
1643        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1644
1645        isel = tcg_to_isel[cond];
1646
1647        tcg_out_movi(s, type, arg0, 1);
1648        if (isel & 1) {
1649            /* arg0 = (bc ? 0 : 1) */
1650            tab = TAB(arg0, 0, arg0);
1651            isel &= ~1;
1652        } else {
1653            /* arg0 = (bc ? 1 : 0) */
1654            tcg_out_movi(s, type, TCG_REG_R0, 0);
1655            tab = TAB(arg0, arg0, TCG_REG_R0);
1656        }
1657        tcg_out32(s, isel | tab);
1658        return;
1659    }
1660
1661    switch (cond) {
1662    case TCG_COND_EQ:
1663        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1664        tcg_out_setcond_eq0(s, type, arg0, arg1);
1665        return;
1666
1667    case TCG_COND_NE:
1668        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1669        /* Discard the high bits only once, rather than both inputs.  */
1670        if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1671            tcg_out_ext32u(s, TCG_REG_R0, arg1);
1672            arg1 = TCG_REG_R0;
1673        }
1674        tcg_out_setcond_ne0(s, arg0, arg1);
1675        return;
1676
1677    case TCG_COND_GT:
1678    case TCG_COND_GTU:
1679        sh = 30;
1680        crop = 0;
1681        goto crtest;
1682
1683    case TCG_COND_LT:
1684    case TCG_COND_LTU:
1685        sh = 29;
1686        crop = 0;
1687        goto crtest;
1688
1689    case TCG_COND_GE:
1690    case TCG_COND_GEU:
1691        sh = 31;
1692        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_LT) | BB(7, CR_LT);
1693        goto crtest;
1694
1695    case TCG_COND_LE:
1696    case TCG_COND_LEU:
1697        sh = 31;
1698        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_GT) | BB(7, CR_GT);
1699    crtest:
1700        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1701        if (crop) {
1702            tcg_out32(s, crop);
1703        }
1704        tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1705        tcg_out_rlw(s, RLWINM, arg0, TCG_REG_R0, sh, 31, 31);
1706        break;
1707
1708    default:
1709        g_assert_not_reached();
1710    }
1711}
1712
1713static void tcg_out_bc(TCGContext *s, int bc, TCGLabel *l)
1714{
1715    if (l->has_value) {
1716        bc |= reloc_pc14_val(tcg_splitwx_to_rx(s->code_ptr), l->u.value_ptr);
1717    } else {
1718        tcg_out_reloc(s, s->code_ptr, R_PPC_REL14, l, 0);
1719    }
1720    tcg_out32(s, bc);
1721}
1722
1723static void tcg_out_brcond(TCGContext *s, TCGCond cond,
1724                           TCGArg arg1, TCGArg arg2, int const_arg2,
1725                           TCGLabel *l, TCGType type)
1726{
1727    tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1728    tcg_out_bc(s, tcg_to_bc[cond], l);
1729}
1730
1731static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
1732                            TCGArg dest, TCGArg c1, TCGArg c2, TCGArg v1,
1733                            TCGArg v2, bool const_c2)
1734{
1735    /* If for some reason both inputs are zero, don't produce bad code.  */
1736    if (v1 == 0 && v2 == 0) {
1737        tcg_out_movi(s, type, dest, 0);
1738        return;
1739    }
1740
1741    tcg_out_cmp(s, cond, c1, c2, const_c2, 7, type);
1742
1743    if (have_isel) {
1744        int isel = tcg_to_isel[cond];
1745
1746        /* Swap the V operands if the operation indicates inversion.  */
1747        if (isel & 1) {
1748            int t = v1;
1749            v1 = v2;
1750            v2 = t;
1751            isel &= ~1;
1752        }
1753        /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand.  */
1754        if (v2 == 0) {
1755            tcg_out_movi(s, type, TCG_REG_R0, 0);
1756        }
1757        tcg_out32(s, isel | TAB(dest, v1, v2));
1758    } else {
1759        if (dest == v2) {
1760            cond = tcg_invert_cond(cond);
1761            v2 = v1;
1762        } else if (dest != v1) {
1763            if (v1 == 0) {
1764                tcg_out_movi(s, type, dest, 0);
1765            } else {
1766                tcg_out_mov(s, type, dest, v1);
1767            }
1768        }
1769        /* Branch forward over one insn */
1770        tcg_out32(s, tcg_to_bc[cond] | 8);
1771        if (v2 == 0) {
1772            tcg_out_movi(s, type, dest, 0);
1773        } else {
1774            tcg_out_mov(s, type, dest, v2);
1775        }
1776    }
1777}
1778
1779static void tcg_out_cntxz(TCGContext *s, TCGType type, uint32_t opc,
1780                          TCGArg a0, TCGArg a1, TCGArg a2, bool const_a2)
1781{
1782    if (const_a2 && a2 == (type == TCG_TYPE_I32 ? 32 : 64)) {
1783        tcg_out32(s, opc | RA(a0) | RS(a1));
1784    } else {
1785        tcg_out_cmp(s, TCG_COND_EQ, a1, 0, 1, 7, type);
1786        /* Note that the only other valid constant for a2 is 0.  */
1787        if (have_isel) {
1788            tcg_out32(s, opc | RA(TCG_REG_R0) | RS(a1));
1789            tcg_out32(s, tcg_to_isel[TCG_COND_EQ] | TAB(a0, a2, TCG_REG_R0));
1790        } else if (!const_a2 && a0 == a2) {
1791            tcg_out32(s, tcg_to_bc[TCG_COND_EQ] | 8);
1792            tcg_out32(s, opc | RA(a0) | RS(a1));
1793        } else {
1794            tcg_out32(s, opc | RA(a0) | RS(a1));
1795            tcg_out32(s, tcg_to_bc[TCG_COND_NE] | 8);
1796            if (const_a2) {
1797                tcg_out_movi(s, type, a0, 0);
1798            } else {
1799                tcg_out_mov(s, type, a0, a2);
1800            }
1801        }
1802    }
1803}
1804
1805static void tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1806                         const int *const_args)
1807{
1808    static const struct { uint8_t bit1, bit2; } bits[] = {
1809        [TCG_COND_LT ] = { CR_LT, CR_LT },
1810        [TCG_COND_LE ] = { CR_LT, CR_GT },
1811        [TCG_COND_GT ] = { CR_GT, CR_GT },
1812        [TCG_COND_GE ] = { CR_GT, CR_LT },
1813        [TCG_COND_LTU] = { CR_LT, CR_LT },
1814        [TCG_COND_LEU] = { CR_LT, CR_GT },
1815        [TCG_COND_GTU] = { CR_GT, CR_GT },
1816        [TCG_COND_GEU] = { CR_GT, CR_LT },
1817    };
1818
1819    TCGCond cond = args[4], cond2;
1820    TCGArg al, ah, bl, bh;
1821    int blconst, bhconst;
1822    int op, bit1, bit2;
1823
1824    al = args[0];
1825    ah = args[1];
1826    bl = args[2];
1827    bh = args[3];
1828    blconst = const_args[2];
1829    bhconst = const_args[3];
1830
1831    switch (cond) {
1832    case TCG_COND_EQ:
1833        op = CRAND;
1834        goto do_equality;
1835    case TCG_COND_NE:
1836        op = CRNAND;
1837    do_equality:
1838        tcg_out_cmp(s, cond, al, bl, blconst, 6, TCG_TYPE_I32);
1839        tcg_out_cmp(s, cond, ah, bh, bhconst, 7, TCG_TYPE_I32);
1840        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1841        break;
1842
1843    case TCG_COND_LT:
1844    case TCG_COND_LE:
1845    case TCG_COND_GT:
1846    case TCG_COND_GE:
1847    case TCG_COND_LTU:
1848    case TCG_COND_LEU:
1849    case TCG_COND_GTU:
1850    case TCG_COND_GEU:
1851        bit1 = bits[cond].bit1;
1852        bit2 = bits[cond].bit2;
1853        op = (bit1 != bit2 ? CRANDC : CRAND);
1854        cond2 = tcg_unsigned_cond(cond);
1855
1856        tcg_out_cmp(s, cond, ah, bh, bhconst, 6, TCG_TYPE_I32);
1857        tcg_out_cmp(s, cond2, al, bl, blconst, 7, TCG_TYPE_I32);
1858        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, bit2));
1859        tcg_out32(s, CROR | BT(7, CR_EQ) | BA(6, bit1) | BB(7, CR_EQ));
1860        break;
1861
1862    default:
1863        g_assert_not_reached();
1864    }
1865}
1866
1867static void tcg_out_setcond2(TCGContext *s, const TCGArg *args,
1868                             const int *const_args)
1869{
1870    tcg_out_cmp2(s, args + 1, const_args + 1);
1871    tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1872    tcg_out_rlw(s, RLWINM, args[0], TCG_REG_R0, 31, 31, 31);
1873}
1874
1875static void tcg_out_brcond2 (TCGContext *s, const TCGArg *args,
1876                             const int *const_args)
1877{
1878    tcg_out_cmp2(s, args, const_args);
1879    tcg_out_bc(s, BC | BI(7, CR_EQ) | BO_COND_TRUE, arg_label(args[5]));
1880}
1881
1882static void tcg_out_mb(TCGContext *s, TCGArg a0)
1883{
1884    uint32_t insn;
1885
1886    if (a0 & TCG_MO_ST_LD) {
1887        insn = HWSYNC;
1888    } else {
1889        insn = LWSYNC;
1890    }
1891
1892    tcg_out32(s, insn);
1893}
1894
1895static void tcg_out_call_int(TCGContext *s, int lk,
1896                             const tcg_insn_unit *target)
1897{
1898#ifdef _CALL_AIX
1899    /* Look through the descriptor.  If the branch is in range, and we
1900       don't have to spend too much effort on building the toc.  */
1901    const void *tgt = ((const void * const *)target)[0];
1902    uintptr_t toc = ((const uintptr_t *)target)[1];
1903    intptr_t diff = tcg_pcrel_diff(s, tgt);
1904
1905    if (in_range_b(diff) && toc == (uint32_t)toc) {
1906        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, toc);
1907        tcg_out_b(s, lk, tgt);
1908    } else {
1909        /* Fold the low bits of the constant into the addresses below.  */
1910        intptr_t arg = (intptr_t)target;
1911        int ofs = (int16_t)arg;
1912
1913        if (ofs + 8 < 0x8000) {
1914            arg -= ofs;
1915        } else {
1916            ofs = 0;
1917        }
1918        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, arg);
1919        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_TMP1, ofs);
1920        tcg_out32(s, MTSPR | RA(TCG_REG_R0) | CTR);
1921        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_REG_TMP1, ofs + SZP);
1922        tcg_out32(s, BCCTR | BO_ALWAYS | lk);
1923    }
1924#elif defined(_CALL_ELF) && _CALL_ELF == 2
1925    intptr_t diff;
1926
1927    /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1928       address, which the callee uses to compute its TOC address.  */
1929    /* FIXME: when the branch is in range, we could avoid r12 load if we
1930       knew that the destination uses the same TOC, and what its local
1931       entry point offset is.  */
1932    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R12, (intptr_t)target);
1933
1934    diff = tcg_pcrel_diff(s, target);
1935    if (in_range_b(diff)) {
1936        tcg_out_b(s, lk, target);
1937    } else {
1938        tcg_out32(s, MTSPR | RS(TCG_REG_R12) | CTR);
1939        tcg_out32(s, BCCTR | BO_ALWAYS | lk);
1940    }
1941#else
1942    tcg_out_b(s, lk, target);
1943#endif
1944}
1945
1946static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target,
1947                         const TCGHelperInfo *info)
1948{
1949    tcg_out_call_int(s, LK, target);
1950}
1951
1952static const uint32_t qemu_ldx_opc[(MO_SSIZE + MO_BSWAP) + 1] = {
1953    [MO_UB] = LBZX,
1954    [MO_UW] = LHZX,
1955    [MO_UL] = LWZX,
1956    [MO_UQ] = LDX,
1957    [MO_SW] = LHAX,
1958    [MO_SL] = LWAX,
1959    [MO_BSWAP | MO_UB] = LBZX,
1960    [MO_BSWAP | MO_UW] = LHBRX,
1961    [MO_BSWAP | MO_UL] = LWBRX,
1962    [MO_BSWAP | MO_UQ] = LDBRX,
1963};
1964
1965static const uint32_t qemu_stx_opc[(MO_SIZE + MO_BSWAP) + 1] = {
1966    [MO_UB] = STBX,
1967    [MO_UW] = STHX,
1968    [MO_UL] = STWX,
1969    [MO_UQ] = STDX,
1970    [MO_BSWAP | MO_UB] = STBX,
1971    [MO_BSWAP | MO_UW] = STHBRX,
1972    [MO_BSWAP | MO_UL] = STWBRX,
1973    [MO_BSWAP | MO_UQ] = STDBRX,
1974};
1975
1976static TCGReg ldst_ra_gen(TCGContext *s, const TCGLabelQemuLdst *l, int arg)
1977{
1978    if (arg < 0) {
1979        arg = TCG_REG_TMP1;
1980    }
1981    tcg_out32(s, MFSPR | RT(arg) | LR);
1982    return arg;
1983}
1984
1985/*
1986 * For the purposes of ppc32 sorting 4 input registers into 4 argument
1987 * registers, there is an outside chance we would require 3 temps.
1988 */
1989static const TCGLdstHelperParam ldst_helper_param = {
1990    .ra_gen = ldst_ra_gen,
1991    .ntmp = 3,
1992    .tmp = { TCG_REG_TMP1, TCG_REG_TMP2, TCG_REG_R0 }
1993};
1994
1995static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1996{
1997    MemOp opc = get_memop(lb->oi);
1998
1999    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2000        return false;
2001    }
2002
2003    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
2004    tcg_out_call_int(s, LK, qemu_ld_helpers[opc & MO_SIZE]);
2005    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
2006
2007    tcg_out_b(s, 0, lb->raddr);
2008    return true;
2009}
2010
2011static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2012{
2013    MemOp opc = get_memop(lb->oi);
2014
2015    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2016        return false;
2017    }
2018
2019    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
2020    tcg_out_call_int(s, LK, qemu_st_helpers[opc & MO_SIZE]);
2021
2022    tcg_out_b(s, 0, lb->raddr);
2023    return true;
2024}
2025
2026typedef struct {
2027    TCGReg base;
2028    TCGReg index;
2029    TCGAtomAlign aa;
2030} HostAddress;
2031
2032bool tcg_target_has_memory_bswap(MemOp memop)
2033{
2034    TCGAtomAlign aa;
2035
2036    if ((memop & MO_SIZE) <= MO_64) {
2037        return true;
2038    }
2039
2040    /*
2041     * Reject 16-byte memop with 16-byte atomicity,
2042     * but do allow a pair of 64-bit operations.
2043     */
2044    aa = atom_and_align_for_opc(tcg_ctx, memop, MO_ATOM_IFALIGN, true);
2045    return aa.atom <= MO_64;
2046}
2047
2048/* We expect to use a 16-bit negative offset from ENV.  */
2049#define MIN_TLB_MASK_TABLE_OFS  -32768
2050
2051/*
2052 * For softmmu, perform the TLB load and compare.
2053 * For useronly, perform any required alignment tests.
2054 * In both cases, return a TCGLabelQemuLdst structure if the slow path
2055 * is required and fill in @h with the host address for the fast path.
2056 */
2057static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
2058                                           TCGReg addrlo, TCGReg addrhi,
2059                                           MemOpIdx oi, bool is_ld)
2060{
2061    TCGType addr_type = s->addr_type;
2062    TCGLabelQemuLdst *ldst = NULL;
2063    MemOp opc = get_memop(oi);
2064    MemOp a_bits, s_bits;
2065
2066    /*
2067     * Book II, Section 1.4, Single-Copy Atomicity, specifies:
2068     *
2069     * Before 3.0, "An access that is not atomic is performed as a set of
2070     * smaller disjoint atomic accesses. In general, the number and alignment
2071     * of these accesses are implementation-dependent."  Thus MO_ATOM_IFALIGN.
2072     *
2073     * As of 3.0, "the non-atomic access is performed as described in
2074     * the corresponding list", which matches MO_ATOM_SUBALIGN.
2075     */
2076    s_bits = opc & MO_SIZE;
2077    h->aa = atom_and_align_for_opc(s, opc,
2078                                   have_isa_3_00 ? MO_ATOM_SUBALIGN
2079                                                 : MO_ATOM_IFALIGN,
2080                                   s_bits == MO_128);
2081    a_bits = h->aa.align;
2082
2083#ifdef CONFIG_SOFTMMU
2084    int mem_index = get_mmuidx(oi);
2085    int cmp_off = is_ld ? offsetof(CPUTLBEntry, addr_read)
2086                        : offsetof(CPUTLBEntry, addr_write);
2087    int fast_off = tlb_mask_table_ofs(s, mem_index);
2088    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
2089    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
2090
2091    ldst = new_ldst_label(s);
2092    ldst->is_ld = is_ld;
2093    ldst->oi = oi;
2094    ldst->addrlo_reg = addrlo;
2095    ldst->addrhi_reg = addrhi;
2096
2097    /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx].  */
2098    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_AREG0, mask_off);
2099    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP2, TCG_AREG0, table_off);
2100
2101    /* Extract the page index, shifted into place for tlb index.  */
2102    if (TCG_TARGET_REG_BITS == 32) {
2103        tcg_out_shri32(s, TCG_REG_R0, addrlo,
2104                       s->page_bits - CPU_TLB_ENTRY_BITS);
2105    } else {
2106        tcg_out_shri64(s, TCG_REG_R0, addrlo,
2107                       s->page_bits - CPU_TLB_ENTRY_BITS);
2108    }
2109    tcg_out32(s, AND | SAB(TCG_REG_TMP1, TCG_REG_TMP1, TCG_REG_R0));
2110
2111    /*
2112     * Load the (low part) TLB comparator into TMP2.
2113     * For 64-bit host, always load the entire 64-bit slot for simplicity.
2114     * We will ignore the high bits with tcg_out_cmp(..., addr_type).
2115     */
2116    if (TCG_TARGET_REG_BITS == 64) {
2117        if (cmp_off == 0) {
2118            tcg_out32(s, LDUX | TAB(TCG_REG_TMP2, TCG_REG_TMP1, TCG_REG_TMP2));
2119        } else {
2120            tcg_out32(s, ADD | TAB(TCG_REG_TMP1, TCG_REG_TMP1, TCG_REG_TMP2));
2121            tcg_out_ld(s, TCG_TYPE_I64, TCG_REG_TMP2, TCG_REG_TMP1, cmp_off);
2122        }
2123    } else if (cmp_off == 0 && !HOST_BIG_ENDIAN) {
2124        tcg_out32(s, LWZUX | TAB(TCG_REG_TMP2, TCG_REG_TMP1, TCG_REG_TMP2));
2125    } else {
2126        tcg_out32(s, ADD | TAB(TCG_REG_TMP1, TCG_REG_TMP1, TCG_REG_TMP2));
2127        tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP2, TCG_REG_TMP1,
2128                   cmp_off + 4 * HOST_BIG_ENDIAN);
2129    }
2130
2131    /*
2132     * Load the TLB addend for use on the fast path.
2133     * Do this asap to minimize any load use delay.
2134     */
2135    if (TCG_TARGET_REG_BITS == 64 || addr_type == TCG_TYPE_I32) {
2136        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1,
2137                   offsetof(CPUTLBEntry, addend));
2138    }
2139
2140    /* Clear the non-page, non-alignment bits from the address in R0. */
2141    if (TCG_TARGET_REG_BITS == 32) {
2142        /*
2143         * We don't support unaligned accesses on 32-bits.
2144         * Preserve the bottom bits and thus trigger a comparison
2145         * failure on unaligned accesses.
2146         */
2147        if (a_bits < s_bits) {
2148            a_bits = s_bits;
2149        }
2150        tcg_out_rlw(s, RLWINM, TCG_REG_R0, addrlo, 0,
2151                    (32 - a_bits) & 31, 31 - s->page_bits);
2152    } else {
2153        TCGReg t = addrlo;
2154
2155        /*
2156         * If the access is unaligned, we need to make sure we fail if we
2157         * cross a page boundary.  The trick is to add the access size-1
2158         * to the address before masking the low bits.  That will make the
2159         * address overflow to the next page if we cross a page boundary,
2160         * which will then force a mismatch of the TLB compare.
2161         */
2162        if (a_bits < s_bits) {
2163            unsigned a_mask = (1 << a_bits) - 1;
2164            unsigned s_mask = (1 << s_bits) - 1;
2165            tcg_out32(s, ADDI | TAI(TCG_REG_R0, t, s_mask - a_mask));
2166            t = TCG_REG_R0;
2167        }
2168
2169        /* Mask the address for the requested alignment.  */
2170        if (addr_type == TCG_TYPE_I32) {
2171            tcg_out_rlw(s, RLWINM, TCG_REG_R0, t, 0,
2172                        (32 - a_bits) & 31, 31 - s->page_bits);
2173        } else if (a_bits == 0) {
2174            tcg_out_rld(s, RLDICR, TCG_REG_R0, t, 0, 63 - s->page_bits);
2175        } else {
2176            tcg_out_rld(s, RLDICL, TCG_REG_R0, t,
2177                        64 - s->page_bits, s->page_bits - a_bits);
2178            tcg_out_rld(s, RLDICL, TCG_REG_R0, TCG_REG_R0, s->page_bits, 0);
2179        }
2180    }
2181
2182    if (TCG_TARGET_REG_BITS == 32 && addr_type != TCG_TYPE_I32) {
2183        /* Low part comparison into cr7. */
2184        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP2,
2185                    0, 7, TCG_TYPE_I32);
2186
2187        /* Load the high part TLB comparator into TMP2.  */
2188        tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP2, TCG_REG_TMP1,
2189                   cmp_off + 4 * !HOST_BIG_ENDIAN);
2190
2191        /* Load addend, deferred for this case. */
2192        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1,
2193                   offsetof(CPUTLBEntry, addend));
2194
2195        /* High part comparison into cr6. */
2196        tcg_out_cmp(s, TCG_COND_EQ, addrhi, TCG_REG_TMP2, 0, 6, TCG_TYPE_I32);
2197
2198        /* Combine comparisons into cr7. */
2199        tcg_out32(s, CRAND | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
2200    } else {
2201        /* Full comparison into cr7. */
2202        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP2, 0, 7, addr_type);
2203    }
2204
2205    /* Load a pointer into the current opcode w/conditional branch-link. */
2206    ldst->label_ptr[0] = s->code_ptr;
2207    tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
2208
2209    h->base = TCG_REG_TMP1;
2210#else
2211    if (a_bits) {
2212        ldst = new_ldst_label(s);
2213        ldst->is_ld = is_ld;
2214        ldst->oi = oi;
2215        ldst->addrlo_reg = addrlo;
2216        ldst->addrhi_reg = addrhi;
2217
2218        /* We are expecting a_bits to max out at 7, much lower than ANDI. */
2219        tcg_debug_assert(a_bits < 16);
2220        tcg_out32(s, ANDI | SAI(addrlo, TCG_REG_R0, (1 << a_bits) - 1));
2221
2222        ldst->label_ptr[0] = s->code_ptr;
2223        tcg_out32(s, BC | BI(0, CR_EQ) | BO_COND_FALSE | LK);
2224    }
2225
2226    h->base = guest_base ? TCG_GUEST_BASE_REG : 0;
2227#endif
2228
2229    if (TCG_TARGET_REG_BITS == 64 && addr_type == TCG_TYPE_I32) {
2230        /* Zero-extend the guest address for use in the host address. */
2231        tcg_out_ext32u(s, TCG_REG_R0, addrlo);
2232        h->index = TCG_REG_R0;
2233    } else {
2234        h->index = addrlo;
2235    }
2236
2237    return ldst;
2238}
2239
2240static void tcg_out_qemu_ld(TCGContext *s, TCGReg datalo, TCGReg datahi,
2241                            TCGReg addrlo, TCGReg addrhi,
2242                            MemOpIdx oi, TCGType data_type)
2243{
2244    MemOp opc = get_memop(oi);
2245    TCGLabelQemuLdst *ldst;
2246    HostAddress h;
2247
2248    ldst = prepare_host_addr(s, &h, addrlo, addrhi, oi, true);
2249
2250    if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
2251        if (opc & MO_BSWAP) {
2252            tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4));
2253            tcg_out32(s, LWBRX | TAB(datalo, h.base, h.index));
2254            tcg_out32(s, LWBRX | TAB(datahi, h.base, TCG_REG_R0));
2255        } else if (h.base != 0) {
2256            tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4));
2257            tcg_out32(s, LWZX | TAB(datahi, h.base, h.index));
2258            tcg_out32(s, LWZX | TAB(datalo, h.base, TCG_REG_R0));
2259        } else if (h.index == datahi) {
2260            tcg_out32(s, LWZ | TAI(datalo, h.index, 4));
2261            tcg_out32(s, LWZ | TAI(datahi, h.index, 0));
2262        } else {
2263            tcg_out32(s, LWZ | TAI(datahi, h.index, 0));
2264            tcg_out32(s, LWZ | TAI(datalo, h.index, 4));
2265        }
2266    } else {
2267        uint32_t insn = qemu_ldx_opc[opc & (MO_BSWAP | MO_SSIZE)];
2268        if (!have_isa_2_06 && insn == LDBRX) {
2269            tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4));
2270            tcg_out32(s, LWBRX | TAB(datalo, h.base, h.index));
2271            tcg_out32(s, LWBRX | TAB(TCG_REG_R0, h.base, TCG_REG_R0));
2272            tcg_out_rld(s, RLDIMI, datalo, TCG_REG_R0, 32, 0);
2273        } else if (insn) {
2274            tcg_out32(s, insn | TAB(datalo, h.base, h.index));
2275        } else {
2276            insn = qemu_ldx_opc[opc & (MO_SIZE | MO_BSWAP)];
2277            tcg_out32(s, insn | TAB(datalo, h.base, h.index));
2278            tcg_out_movext(s, TCG_TYPE_REG, datalo,
2279                           TCG_TYPE_REG, opc & MO_SSIZE, datalo);
2280        }
2281    }
2282
2283    if (ldst) {
2284        ldst->type = data_type;
2285        ldst->datalo_reg = datalo;
2286        ldst->datahi_reg = datahi;
2287        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2288    }
2289}
2290
2291static void tcg_out_qemu_st(TCGContext *s, TCGReg datalo, TCGReg datahi,
2292                            TCGReg addrlo, TCGReg addrhi,
2293                            MemOpIdx oi, TCGType data_type)
2294{
2295    MemOp opc = get_memop(oi);
2296    TCGLabelQemuLdst *ldst;
2297    HostAddress h;
2298
2299    ldst = prepare_host_addr(s, &h, addrlo, addrhi, oi, false);
2300
2301    if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
2302        if (opc & MO_BSWAP) {
2303            tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4));
2304            tcg_out32(s, STWBRX | SAB(datalo, h.base, h.index));
2305            tcg_out32(s, STWBRX | SAB(datahi, h.base, TCG_REG_R0));
2306        } else if (h.base != 0) {
2307            tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4));
2308            tcg_out32(s, STWX | SAB(datahi, h.base, h.index));
2309            tcg_out32(s, STWX | SAB(datalo, h.base, TCG_REG_R0));
2310        } else {
2311            tcg_out32(s, STW | TAI(datahi, h.index, 0));
2312            tcg_out32(s, STW | TAI(datalo, h.index, 4));
2313        }
2314    } else {
2315        uint32_t insn = qemu_stx_opc[opc & (MO_BSWAP | MO_SIZE)];
2316        if (!have_isa_2_06 && insn == STDBRX) {
2317            tcg_out32(s, STWBRX | SAB(datalo, h.base, h.index));
2318            tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, h.index, 4));
2319            tcg_out_shri64(s, TCG_REG_R0, datalo, 32);
2320            tcg_out32(s, STWBRX | SAB(TCG_REG_R0, h.base, TCG_REG_TMP1));
2321        } else {
2322            tcg_out32(s, insn | SAB(datalo, h.base, h.index));
2323        }
2324    }
2325
2326    if (ldst) {
2327        ldst->type = data_type;
2328        ldst->datalo_reg = datalo;
2329        ldst->datahi_reg = datahi;
2330        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2331    }
2332}
2333
2334static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi,
2335                                   TCGReg addr_reg, MemOpIdx oi, bool is_ld)
2336{
2337    TCGLabelQemuLdst *ldst;
2338    HostAddress h;
2339    bool need_bswap;
2340    uint32_t insn;
2341    TCGReg index;
2342
2343    ldst = prepare_host_addr(s, &h, addr_reg, -1, oi, is_ld);
2344
2345    /* Compose the final address, as LQ/STQ have no indexing. */
2346    index = h.index;
2347    if (h.base != 0) {
2348        index = TCG_REG_TMP1;
2349        tcg_out32(s, ADD | TAB(index, h.base, h.index));
2350    }
2351    need_bswap = get_memop(oi) & MO_BSWAP;
2352
2353    if (h.aa.atom == MO_128) {
2354        tcg_debug_assert(!need_bswap);
2355        tcg_debug_assert(datalo & 1);
2356        tcg_debug_assert(datahi == datalo - 1);
2357        insn = is_ld ? LQ : STQ;
2358        tcg_out32(s, insn | TAI(datahi, index, 0));
2359    } else {
2360        TCGReg d1, d2;
2361
2362        if (HOST_BIG_ENDIAN ^ need_bswap) {
2363            d1 = datahi, d2 = datalo;
2364        } else {
2365            d1 = datalo, d2 = datahi;
2366        }
2367
2368        if (need_bswap) {
2369            tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, 8);
2370            insn = is_ld ? LDBRX : STDBRX;
2371            tcg_out32(s, insn | TAB(d1, 0, index));
2372            tcg_out32(s, insn | TAB(d2, index, TCG_REG_R0));
2373        } else {
2374            insn = is_ld ? LD : STD;
2375            tcg_out32(s, insn | TAI(d1, index, 0));
2376            tcg_out32(s, insn | TAI(d2, index, 8));
2377        }
2378    }
2379
2380    if (ldst) {
2381        ldst->type = TCG_TYPE_I128;
2382        ldst->datalo_reg = datalo;
2383        ldst->datahi_reg = datahi;
2384        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2385    }
2386}
2387
2388static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
2389{
2390    int i;
2391    for (i = 0; i < count; ++i) {
2392        p[i] = NOP;
2393    }
2394}
2395
2396/* Parameters for function call generation, used in tcg.c.  */
2397#define TCG_TARGET_STACK_ALIGN       16
2398
2399#ifdef _CALL_AIX
2400# define LINK_AREA_SIZE                (6 * SZR)
2401# define LR_OFFSET                     (1 * SZR)
2402# define TCG_TARGET_CALL_STACK_OFFSET  (LINK_AREA_SIZE + 8 * SZR)
2403#elif defined(_CALL_DARWIN)
2404# define LINK_AREA_SIZE                (6 * SZR)
2405# define LR_OFFSET                     (2 * SZR)
2406#elif TCG_TARGET_REG_BITS == 64
2407# if defined(_CALL_ELF) && _CALL_ELF == 2
2408#  define LINK_AREA_SIZE               (4 * SZR)
2409#  define LR_OFFSET                    (1 * SZR)
2410# endif
2411#else /* TCG_TARGET_REG_BITS == 32 */
2412# if defined(_CALL_SYSV)
2413#  define LINK_AREA_SIZE               (2 * SZR)
2414#  define LR_OFFSET                    (1 * SZR)
2415# endif
2416#endif
2417#ifndef LR_OFFSET
2418# error "Unhandled abi"
2419#endif
2420#ifndef TCG_TARGET_CALL_STACK_OFFSET
2421# define TCG_TARGET_CALL_STACK_OFFSET  LINK_AREA_SIZE
2422#endif
2423
2424#define CPU_TEMP_BUF_SIZE  (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
2425#define REG_SAVE_SIZE      ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
2426
2427#define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET   \
2428                     + TCG_STATIC_CALL_ARGS_SIZE    \
2429                     + CPU_TEMP_BUF_SIZE            \
2430                     + REG_SAVE_SIZE                \
2431                     + TCG_TARGET_STACK_ALIGN - 1)  \
2432                    & -TCG_TARGET_STACK_ALIGN)
2433
2434#define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
2435
2436static void tcg_target_qemu_prologue(TCGContext *s)
2437{
2438    int i;
2439
2440#ifdef _CALL_AIX
2441    const void **desc = (const void **)s->code_ptr;
2442    desc[0] = tcg_splitwx_to_rx(desc + 2);  /* entry point */
2443    desc[1] = 0;                            /* environment pointer */
2444    s->code_ptr = (void *)(desc + 2);       /* skip over descriptor */
2445#endif
2446
2447    tcg_set_frame(s, TCG_REG_CALL_STACK, REG_SAVE_BOT - CPU_TEMP_BUF_SIZE,
2448                  CPU_TEMP_BUF_SIZE);
2449
2450    /* Prologue */
2451    tcg_out32(s, MFSPR | RT(TCG_REG_R0) | LR);
2452    tcg_out32(s, (SZR == 8 ? STDU : STWU)
2453              | SAI(TCG_REG_R1, TCG_REG_R1, -FRAME_SIZE));
2454
2455    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2456        tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2457                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2458    }
2459    tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2460
2461#ifndef CONFIG_SOFTMMU
2462    if (guest_base) {
2463        tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true);
2464        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2465    }
2466#endif
2467
2468    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2469    tcg_out32(s, MTSPR | RS(tcg_target_call_iarg_regs[1]) | CTR);
2470    if (USE_REG_TB) {
2471        tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, tcg_target_call_iarg_regs[1]);
2472    }
2473    tcg_out32(s, BCCTR | BO_ALWAYS);
2474
2475    /* Epilogue */
2476    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
2477
2478    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2479    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2480        tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2481                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2482    }
2483    tcg_out32(s, MTSPR | RS(TCG_REG_R0) | LR);
2484    tcg_out32(s, ADDI | TAI(TCG_REG_R1, TCG_REG_R1, FRAME_SIZE));
2485    tcg_out32(s, BCLR | BO_ALWAYS);
2486}
2487
2488static void tcg_out_exit_tb(TCGContext *s, uintptr_t arg)
2489{
2490    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, arg);
2491    tcg_out_b(s, 0, tcg_code_gen_epilogue);
2492}
2493
2494static void tcg_out_goto_tb(TCGContext *s, int which)
2495{
2496    uintptr_t ptr = get_jmp_target_addr(s, which);
2497
2498    if (USE_REG_TB) {
2499        ptrdiff_t offset = tcg_tbrel_diff(s, (void *)ptr);
2500        tcg_out_mem_long(s, LD, LDX, TCG_REG_TB, TCG_REG_TB, offset);
2501
2502        /* Direct branch will be patched by tb_target_set_jmp_target. */
2503        set_jmp_insn_offset(s, which);
2504        tcg_out32(s, MTSPR | RS(TCG_REG_TB) | CTR);
2505
2506        /* When branch is out of range, fall through to indirect. */
2507        tcg_out32(s, BCCTR | BO_ALWAYS);
2508
2509        /* For the unlinked case, need to reset TCG_REG_TB.  */
2510        set_jmp_reset_offset(s, which);
2511        tcg_out_mem_long(s, ADDI, ADD, TCG_REG_TB, TCG_REG_TB,
2512                         -tcg_current_code_size(s));
2513    } else {
2514        /* Direct branch will be patched by tb_target_set_jmp_target. */
2515        set_jmp_insn_offset(s, which);
2516        tcg_out32(s, NOP);
2517
2518        /* When branch is out of range, fall through to indirect. */
2519        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, ptr - (int16_t)ptr);
2520        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1, (int16_t)ptr);
2521        tcg_out32(s, MTSPR | RS(TCG_REG_TMP1) | CTR);
2522        tcg_out32(s, BCCTR | BO_ALWAYS);
2523        set_jmp_reset_offset(s, which);
2524    }
2525}
2526
2527void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
2528                              uintptr_t jmp_rx, uintptr_t jmp_rw)
2529{
2530    uintptr_t addr = tb->jmp_target_addr[n];
2531    intptr_t diff = addr - jmp_rx;
2532    tcg_insn_unit insn;
2533
2534    if (in_range_b(diff)) {
2535        insn = B | (diff & 0x3fffffc);
2536    } else if (USE_REG_TB) {
2537        insn = MTSPR | RS(TCG_REG_TB) | CTR;
2538    } else {
2539        insn = NOP;
2540    }
2541
2542    qatomic_set((uint32_t *)jmp_rw, insn);
2543    flush_idcache_range(jmp_rx, jmp_rw, 4);
2544}
2545
2546static void tcg_out_op(TCGContext *s, TCGOpcode opc,
2547                       const TCGArg args[TCG_MAX_OP_ARGS],
2548                       const int const_args[TCG_MAX_OP_ARGS])
2549{
2550    TCGArg a0, a1, a2;
2551
2552    switch (opc) {
2553    case INDEX_op_goto_ptr:
2554        tcg_out32(s, MTSPR | RS(args[0]) | CTR);
2555        if (USE_REG_TB) {
2556            tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, args[0]);
2557        }
2558        tcg_out32(s, ADDI | TAI(TCG_REG_R3, 0, 0));
2559        tcg_out32(s, BCCTR | BO_ALWAYS);
2560        break;
2561    case INDEX_op_br:
2562        {
2563            TCGLabel *l = arg_label(args[0]);
2564            uint32_t insn = B;
2565
2566            if (l->has_value) {
2567                insn |= reloc_pc24_val(tcg_splitwx_to_rx(s->code_ptr),
2568                                       l->u.value_ptr);
2569            } else {
2570                tcg_out_reloc(s, s->code_ptr, R_PPC_REL24, l, 0);
2571            }
2572            tcg_out32(s, insn);
2573        }
2574        break;
2575    case INDEX_op_ld8u_i32:
2576    case INDEX_op_ld8u_i64:
2577        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2578        break;
2579    case INDEX_op_ld8s_i32:
2580    case INDEX_op_ld8s_i64:
2581        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2582        tcg_out_ext8s(s, TCG_TYPE_REG, args[0], args[0]);
2583        break;
2584    case INDEX_op_ld16u_i32:
2585    case INDEX_op_ld16u_i64:
2586        tcg_out_mem_long(s, LHZ, LHZX, args[0], args[1], args[2]);
2587        break;
2588    case INDEX_op_ld16s_i32:
2589    case INDEX_op_ld16s_i64:
2590        tcg_out_mem_long(s, LHA, LHAX, args[0], args[1], args[2]);
2591        break;
2592    case INDEX_op_ld_i32:
2593    case INDEX_op_ld32u_i64:
2594        tcg_out_mem_long(s, LWZ, LWZX, args[0], args[1], args[2]);
2595        break;
2596    case INDEX_op_ld32s_i64:
2597        tcg_out_mem_long(s, LWA, LWAX, args[0], args[1], args[2]);
2598        break;
2599    case INDEX_op_ld_i64:
2600        tcg_out_mem_long(s, LD, LDX, args[0], args[1], args[2]);
2601        break;
2602    case INDEX_op_st8_i32:
2603    case INDEX_op_st8_i64:
2604        tcg_out_mem_long(s, STB, STBX, args[0], args[1], args[2]);
2605        break;
2606    case INDEX_op_st16_i32:
2607    case INDEX_op_st16_i64:
2608        tcg_out_mem_long(s, STH, STHX, args[0], args[1], args[2]);
2609        break;
2610    case INDEX_op_st_i32:
2611    case INDEX_op_st32_i64:
2612        tcg_out_mem_long(s, STW, STWX, args[0], args[1], args[2]);
2613        break;
2614    case INDEX_op_st_i64:
2615        tcg_out_mem_long(s, STD, STDX, args[0], args[1], args[2]);
2616        break;
2617
2618    case INDEX_op_add_i32:
2619        a0 = args[0], a1 = args[1], a2 = args[2];
2620        if (const_args[2]) {
2621        do_addi_32:
2622            tcg_out_mem_long(s, ADDI, ADD, a0, a1, (int32_t)a2);
2623        } else {
2624            tcg_out32(s, ADD | TAB(a0, a1, a2));
2625        }
2626        break;
2627    case INDEX_op_sub_i32:
2628        a0 = args[0], a1 = args[1], a2 = args[2];
2629        if (const_args[1]) {
2630            if (const_args[2]) {
2631                tcg_out_movi(s, TCG_TYPE_I32, a0, a1 - a2);
2632            } else {
2633                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2634            }
2635        } else if (const_args[2]) {
2636            a2 = -a2;
2637            goto do_addi_32;
2638        } else {
2639            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2640        }
2641        break;
2642
2643    case INDEX_op_and_i32:
2644        a0 = args[0], a1 = args[1], a2 = args[2];
2645        if (const_args[2]) {
2646            tcg_out_andi32(s, a0, a1, a2);
2647        } else {
2648            tcg_out32(s, AND | SAB(a1, a0, a2));
2649        }
2650        break;
2651    case INDEX_op_and_i64:
2652        a0 = args[0], a1 = args[1], a2 = args[2];
2653        if (const_args[2]) {
2654            tcg_out_andi64(s, a0, a1, a2);
2655        } else {
2656            tcg_out32(s, AND | SAB(a1, a0, a2));
2657        }
2658        break;
2659    case INDEX_op_or_i64:
2660    case INDEX_op_or_i32:
2661        a0 = args[0], a1 = args[1], a2 = args[2];
2662        if (const_args[2]) {
2663            tcg_out_ori32(s, a0, a1, a2);
2664        } else {
2665            tcg_out32(s, OR | SAB(a1, a0, a2));
2666        }
2667        break;
2668    case INDEX_op_xor_i64:
2669    case INDEX_op_xor_i32:
2670        a0 = args[0], a1 = args[1], a2 = args[2];
2671        if (const_args[2]) {
2672            tcg_out_xori32(s, a0, a1, a2);
2673        } else {
2674            tcg_out32(s, XOR | SAB(a1, a0, a2));
2675        }
2676        break;
2677    case INDEX_op_andc_i32:
2678        a0 = args[0], a1 = args[1], a2 = args[2];
2679        if (const_args[2]) {
2680            tcg_out_andi32(s, a0, a1, ~a2);
2681        } else {
2682            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2683        }
2684        break;
2685    case INDEX_op_andc_i64:
2686        a0 = args[0], a1 = args[1], a2 = args[2];
2687        if (const_args[2]) {
2688            tcg_out_andi64(s, a0, a1, ~a2);
2689        } else {
2690            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2691        }
2692        break;
2693    case INDEX_op_orc_i32:
2694        if (const_args[2]) {
2695            tcg_out_ori32(s, args[0], args[1], ~args[2]);
2696            break;
2697        }
2698        /* FALLTHRU */
2699    case INDEX_op_orc_i64:
2700        tcg_out32(s, ORC | SAB(args[1], args[0], args[2]));
2701        break;
2702    case INDEX_op_eqv_i32:
2703        if (const_args[2]) {
2704            tcg_out_xori32(s, args[0], args[1], ~args[2]);
2705            break;
2706        }
2707        /* FALLTHRU */
2708    case INDEX_op_eqv_i64:
2709        tcg_out32(s, EQV | SAB(args[1], args[0], args[2]));
2710        break;
2711    case INDEX_op_nand_i32:
2712    case INDEX_op_nand_i64:
2713        tcg_out32(s, NAND | SAB(args[1], args[0], args[2]));
2714        break;
2715    case INDEX_op_nor_i32:
2716    case INDEX_op_nor_i64:
2717        tcg_out32(s, NOR | SAB(args[1], args[0], args[2]));
2718        break;
2719
2720    case INDEX_op_clz_i32:
2721        tcg_out_cntxz(s, TCG_TYPE_I32, CNTLZW, args[0], args[1],
2722                      args[2], const_args[2]);
2723        break;
2724    case INDEX_op_ctz_i32:
2725        tcg_out_cntxz(s, TCG_TYPE_I32, CNTTZW, args[0], args[1],
2726                      args[2], const_args[2]);
2727        break;
2728    case INDEX_op_ctpop_i32:
2729        tcg_out32(s, CNTPOPW | SAB(args[1], args[0], 0));
2730        break;
2731
2732    case INDEX_op_clz_i64:
2733        tcg_out_cntxz(s, TCG_TYPE_I64, CNTLZD, args[0], args[1],
2734                      args[2], const_args[2]);
2735        break;
2736    case INDEX_op_ctz_i64:
2737        tcg_out_cntxz(s, TCG_TYPE_I64, CNTTZD, args[0], args[1],
2738                      args[2], const_args[2]);
2739        break;
2740    case INDEX_op_ctpop_i64:
2741        tcg_out32(s, CNTPOPD | SAB(args[1], args[0], 0));
2742        break;
2743
2744    case INDEX_op_mul_i32:
2745        a0 = args[0], a1 = args[1], a2 = args[2];
2746        if (const_args[2]) {
2747            tcg_out32(s, MULLI | TAI(a0, a1, a2));
2748        } else {
2749            tcg_out32(s, MULLW | TAB(a0, a1, a2));
2750        }
2751        break;
2752
2753    case INDEX_op_div_i32:
2754        tcg_out32(s, DIVW | TAB(args[0], args[1], args[2]));
2755        break;
2756
2757    case INDEX_op_divu_i32:
2758        tcg_out32(s, DIVWU | TAB(args[0], args[1], args[2]));
2759        break;
2760
2761    case INDEX_op_rem_i32:
2762        tcg_out32(s, MODSW | TAB(args[0], args[1], args[2]));
2763        break;
2764
2765    case INDEX_op_remu_i32:
2766        tcg_out32(s, MODUW | TAB(args[0], args[1], args[2]));
2767        break;
2768
2769    case INDEX_op_shl_i32:
2770        if (const_args[2]) {
2771            /* Limit immediate shift count lest we create an illegal insn.  */
2772            tcg_out_shli32(s, args[0], args[1], args[2] & 31);
2773        } else {
2774            tcg_out32(s, SLW | SAB(args[1], args[0], args[2]));
2775        }
2776        break;
2777    case INDEX_op_shr_i32:
2778        if (const_args[2]) {
2779            /* Limit immediate shift count lest we create an illegal insn.  */
2780            tcg_out_shri32(s, args[0], args[1], args[2] & 31);
2781        } else {
2782            tcg_out32(s, SRW | SAB(args[1], args[0], args[2]));
2783        }
2784        break;
2785    case INDEX_op_sar_i32:
2786        if (const_args[2]) {
2787            tcg_out_sari32(s, args[0], args[1], args[2]);
2788        } else {
2789            tcg_out32(s, SRAW | SAB(args[1], args[0], args[2]));
2790        }
2791        break;
2792    case INDEX_op_rotl_i32:
2793        if (const_args[2]) {
2794            tcg_out_rlw(s, RLWINM, args[0], args[1], args[2], 0, 31);
2795        } else {
2796            tcg_out32(s, RLWNM | SAB(args[1], args[0], args[2])
2797                         | MB(0) | ME(31));
2798        }
2799        break;
2800    case INDEX_op_rotr_i32:
2801        if (const_args[2]) {
2802            tcg_out_rlw(s, RLWINM, args[0], args[1], 32 - args[2], 0, 31);
2803        } else {
2804            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 32));
2805            tcg_out32(s, RLWNM | SAB(args[1], args[0], TCG_REG_R0)
2806                         | MB(0) | ME(31));
2807        }
2808        break;
2809
2810    case INDEX_op_brcond_i32:
2811        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2812                       arg_label(args[3]), TCG_TYPE_I32);
2813        break;
2814    case INDEX_op_brcond_i64:
2815        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2816                       arg_label(args[3]), TCG_TYPE_I64);
2817        break;
2818    case INDEX_op_brcond2_i32:
2819        tcg_out_brcond2(s, args, const_args);
2820        break;
2821
2822    case INDEX_op_neg_i32:
2823    case INDEX_op_neg_i64:
2824        tcg_out32(s, NEG | RT(args[0]) | RA(args[1]));
2825        break;
2826
2827    case INDEX_op_not_i32:
2828    case INDEX_op_not_i64:
2829        tcg_out32(s, NOR | SAB(args[1], args[0], args[1]));
2830        break;
2831
2832    case INDEX_op_add_i64:
2833        a0 = args[0], a1 = args[1], a2 = args[2];
2834        if (const_args[2]) {
2835        do_addi_64:
2836            tcg_out_mem_long(s, ADDI, ADD, a0, a1, a2);
2837        } else {
2838            tcg_out32(s, ADD | TAB(a0, a1, a2));
2839        }
2840        break;
2841    case INDEX_op_sub_i64:
2842        a0 = args[0], a1 = args[1], a2 = args[2];
2843        if (const_args[1]) {
2844            if (const_args[2]) {
2845                tcg_out_movi(s, TCG_TYPE_I64, a0, a1 - a2);
2846            } else {
2847                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2848            }
2849        } else if (const_args[2]) {
2850            a2 = -a2;
2851            goto do_addi_64;
2852        } else {
2853            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2854        }
2855        break;
2856
2857    case INDEX_op_shl_i64:
2858        if (const_args[2]) {
2859            /* Limit immediate shift count lest we create an illegal insn.  */
2860            tcg_out_shli64(s, args[0], args[1], args[2] & 63);
2861        } else {
2862            tcg_out32(s, SLD | SAB(args[1], args[0], args[2]));
2863        }
2864        break;
2865    case INDEX_op_shr_i64:
2866        if (const_args[2]) {
2867            /* Limit immediate shift count lest we create an illegal insn.  */
2868            tcg_out_shri64(s, args[0], args[1], args[2] & 63);
2869        } else {
2870            tcg_out32(s, SRD | SAB(args[1], args[0], args[2]));
2871        }
2872        break;
2873    case INDEX_op_sar_i64:
2874        if (const_args[2]) {
2875            tcg_out_sari64(s, args[0], args[1], args[2]);
2876        } else {
2877            tcg_out32(s, SRAD | SAB(args[1], args[0], args[2]));
2878        }
2879        break;
2880    case INDEX_op_rotl_i64:
2881        if (const_args[2]) {
2882            tcg_out_rld(s, RLDICL, args[0], args[1], args[2], 0);
2883        } else {
2884            tcg_out32(s, RLDCL | SAB(args[1], args[0], args[2]) | MB64(0));
2885        }
2886        break;
2887    case INDEX_op_rotr_i64:
2888        if (const_args[2]) {
2889            tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 0);
2890        } else {
2891            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 64));
2892            tcg_out32(s, RLDCL | SAB(args[1], args[0], TCG_REG_R0) | MB64(0));
2893        }
2894        break;
2895
2896    case INDEX_op_mul_i64:
2897        a0 = args[0], a1 = args[1], a2 = args[2];
2898        if (const_args[2]) {
2899            tcg_out32(s, MULLI | TAI(a0, a1, a2));
2900        } else {
2901            tcg_out32(s, MULLD | TAB(a0, a1, a2));
2902        }
2903        break;
2904    case INDEX_op_div_i64:
2905        tcg_out32(s, DIVD | TAB(args[0], args[1], args[2]));
2906        break;
2907    case INDEX_op_divu_i64:
2908        tcg_out32(s, DIVDU | TAB(args[0], args[1], args[2]));
2909        break;
2910    case INDEX_op_rem_i64:
2911        tcg_out32(s, MODSD | TAB(args[0], args[1], args[2]));
2912        break;
2913    case INDEX_op_remu_i64:
2914        tcg_out32(s, MODUD | TAB(args[0], args[1], args[2]));
2915        break;
2916
2917    case INDEX_op_qemu_ld_a64_i32:
2918        if (TCG_TARGET_REG_BITS == 32) {
2919            tcg_out_qemu_ld(s, args[0], -1, args[1], args[2],
2920                            args[3], TCG_TYPE_I32);
2921            break;
2922        }
2923        /* fall through */
2924    case INDEX_op_qemu_ld_a32_i32:
2925        tcg_out_qemu_ld(s, args[0], -1, args[1], -1, args[2], TCG_TYPE_I32);
2926        break;
2927    case INDEX_op_qemu_ld_a32_i64:
2928        if (TCG_TARGET_REG_BITS == 64) {
2929            tcg_out_qemu_ld(s, args[0], -1, args[1], -1,
2930                            args[2], TCG_TYPE_I64);
2931        } else {
2932            tcg_out_qemu_ld(s, args[0], args[1], args[2], -1,
2933                            args[3], TCG_TYPE_I64);
2934        }
2935        break;
2936    case INDEX_op_qemu_ld_a64_i64:
2937        if (TCG_TARGET_REG_BITS == 64) {
2938            tcg_out_qemu_ld(s, args[0], -1, args[1], -1,
2939                            args[2], TCG_TYPE_I64);
2940        } else {
2941            tcg_out_qemu_ld(s, args[0], args[1], args[2], args[3],
2942                            args[4], TCG_TYPE_I64);
2943        }
2944        break;
2945    case INDEX_op_qemu_ld_a32_i128:
2946    case INDEX_op_qemu_ld_a64_i128:
2947        tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
2948        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], true);
2949        break;
2950
2951    case INDEX_op_qemu_st_a64_i32:
2952        if (TCG_TARGET_REG_BITS == 32) {
2953            tcg_out_qemu_st(s, args[0], -1, args[1], args[2],
2954                            args[3], TCG_TYPE_I32);
2955            break;
2956        }
2957        /* fall through */
2958    case INDEX_op_qemu_st_a32_i32:
2959        tcg_out_qemu_st(s, args[0], -1, args[1], -1, args[2], TCG_TYPE_I32);
2960        break;
2961    case INDEX_op_qemu_st_a32_i64:
2962        if (TCG_TARGET_REG_BITS == 64) {
2963            tcg_out_qemu_st(s, args[0], -1, args[1], -1,
2964                            args[2], TCG_TYPE_I64);
2965        } else {
2966            tcg_out_qemu_st(s, args[0], args[1], args[2], -1,
2967                            args[3], TCG_TYPE_I64);
2968        }
2969        break;
2970    case INDEX_op_qemu_st_a64_i64:
2971        if (TCG_TARGET_REG_BITS == 64) {
2972            tcg_out_qemu_st(s, args[0], -1, args[1], -1,
2973                            args[2], TCG_TYPE_I64);
2974        } else {
2975            tcg_out_qemu_st(s, args[0], args[1], args[2], args[3],
2976                            args[4], TCG_TYPE_I64);
2977        }
2978        break;
2979    case INDEX_op_qemu_st_a32_i128:
2980    case INDEX_op_qemu_st_a64_i128:
2981        tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
2982        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], false);
2983        break;
2984
2985    case INDEX_op_setcond_i32:
2986        tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2],
2987                        const_args[2]);
2988        break;
2989    case INDEX_op_setcond_i64:
2990        tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2],
2991                        const_args[2]);
2992        break;
2993    case INDEX_op_setcond2_i32:
2994        tcg_out_setcond2(s, args, const_args);
2995        break;
2996
2997    case INDEX_op_bswap16_i32:
2998    case INDEX_op_bswap16_i64:
2999        tcg_out_bswap16(s, args[0], args[1], args[2]);
3000        break;
3001    case INDEX_op_bswap32_i32:
3002        tcg_out_bswap32(s, args[0], args[1], 0);
3003        break;
3004    case INDEX_op_bswap32_i64:
3005        tcg_out_bswap32(s, args[0], args[1], args[2]);
3006        break;
3007    case INDEX_op_bswap64_i64:
3008        tcg_out_bswap64(s, args[0], args[1]);
3009        break;
3010
3011    case INDEX_op_deposit_i32:
3012        if (const_args[2]) {
3013            uint32_t mask = ((2u << (args[4] - 1)) - 1) << args[3];
3014            tcg_out_andi32(s, args[0], args[0], ~mask);
3015        } else {
3016            tcg_out_rlw(s, RLWIMI, args[0], args[2], args[3],
3017                        32 - args[3] - args[4], 31 - args[3]);
3018        }
3019        break;
3020    case INDEX_op_deposit_i64:
3021        if (const_args[2]) {
3022            uint64_t mask = ((2ull << (args[4] - 1)) - 1) << args[3];
3023            tcg_out_andi64(s, args[0], args[0], ~mask);
3024        } else {
3025            tcg_out_rld(s, RLDIMI, args[0], args[2], args[3],
3026                        64 - args[3] - args[4]);
3027        }
3028        break;
3029
3030    case INDEX_op_extract_i32:
3031        tcg_out_rlw(s, RLWINM, args[0], args[1],
3032                    32 - args[2], 32 - args[3], 31);
3033        break;
3034    case INDEX_op_extract_i64:
3035        tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 64 - args[3]);
3036        break;
3037
3038    case INDEX_op_movcond_i32:
3039        tcg_out_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], args[2],
3040                        args[3], args[4], const_args[2]);
3041        break;
3042    case INDEX_op_movcond_i64:
3043        tcg_out_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], args[2],
3044                        args[3], args[4], const_args[2]);
3045        break;
3046
3047#if TCG_TARGET_REG_BITS == 64
3048    case INDEX_op_add2_i64:
3049#else
3050    case INDEX_op_add2_i32:
3051#endif
3052        /* Note that the CA bit is defined based on the word size of the
3053           environment.  So in 64-bit mode it's always carry-out of bit 63.
3054           The fallback code using deposit works just as well for 32-bit.  */
3055        a0 = args[0], a1 = args[1];
3056        if (a0 == args[3] || (!const_args[5] && a0 == args[5])) {
3057            a0 = TCG_REG_R0;
3058        }
3059        if (const_args[4]) {
3060            tcg_out32(s, ADDIC | TAI(a0, args[2], args[4]));
3061        } else {
3062            tcg_out32(s, ADDC | TAB(a0, args[2], args[4]));
3063        }
3064        if (const_args[5]) {
3065            tcg_out32(s, (args[5] ? ADDME : ADDZE) | RT(a1) | RA(args[3]));
3066        } else {
3067            tcg_out32(s, ADDE | TAB(a1, args[3], args[5]));
3068        }
3069        if (a0 != args[0]) {
3070            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
3071        }
3072        break;
3073
3074#if TCG_TARGET_REG_BITS == 64
3075    case INDEX_op_sub2_i64:
3076#else
3077    case INDEX_op_sub2_i32:
3078#endif
3079        a0 = args[0], a1 = args[1];
3080        if (a0 == args[5] || (!const_args[3] && a0 == args[3])) {
3081            a0 = TCG_REG_R0;
3082        }
3083        if (const_args[2]) {
3084            tcg_out32(s, SUBFIC | TAI(a0, args[4], args[2]));
3085        } else {
3086            tcg_out32(s, SUBFC | TAB(a0, args[4], args[2]));
3087        }
3088        if (const_args[3]) {
3089            tcg_out32(s, (args[3] ? SUBFME : SUBFZE) | RT(a1) | RA(args[5]));
3090        } else {
3091            tcg_out32(s, SUBFE | TAB(a1, args[5], args[3]));
3092        }
3093        if (a0 != args[0]) {
3094            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
3095        }
3096        break;
3097
3098    case INDEX_op_muluh_i32:
3099        tcg_out32(s, MULHWU | TAB(args[0], args[1], args[2]));
3100        break;
3101    case INDEX_op_mulsh_i32:
3102        tcg_out32(s, MULHW | TAB(args[0], args[1], args[2]));
3103        break;
3104    case INDEX_op_muluh_i64:
3105        tcg_out32(s, MULHDU | TAB(args[0], args[1], args[2]));
3106        break;
3107    case INDEX_op_mulsh_i64:
3108        tcg_out32(s, MULHD | TAB(args[0], args[1], args[2]));
3109        break;
3110
3111    case INDEX_op_mb:
3112        tcg_out_mb(s, args[0]);
3113        break;
3114
3115    case INDEX_op_mov_i32:   /* Always emitted via tcg_out_mov.  */
3116    case INDEX_op_mov_i64:
3117    case INDEX_op_call:      /* Always emitted via tcg_out_call.  */
3118    case INDEX_op_exit_tb:   /* Always emitted via tcg_out_exit_tb.  */
3119    case INDEX_op_goto_tb:   /* Always emitted via tcg_out_goto_tb.  */
3120    case INDEX_op_ext8s_i32:  /* Always emitted via tcg_reg_alloc_op.  */
3121    case INDEX_op_ext8s_i64:
3122    case INDEX_op_ext8u_i32:
3123    case INDEX_op_ext8u_i64:
3124    case INDEX_op_ext16s_i32:
3125    case INDEX_op_ext16s_i64:
3126    case INDEX_op_ext16u_i32:
3127    case INDEX_op_ext16u_i64:
3128    case INDEX_op_ext32s_i64:
3129    case INDEX_op_ext32u_i64:
3130    case INDEX_op_ext_i32_i64:
3131    case INDEX_op_extu_i32_i64:
3132    case INDEX_op_extrl_i64_i32:
3133    default:
3134        g_assert_not_reached();
3135    }
3136}
3137
3138int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3139{
3140    switch (opc) {
3141    case INDEX_op_and_vec:
3142    case INDEX_op_or_vec:
3143    case INDEX_op_xor_vec:
3144    case INDEX_op_andc_vec:
3145    case INDEX_op_not_vec:
3146    case INDEX_op_nor_vec:
3147    case INDEX_op_eqv_vec:
3148    case INDEX_op_nand_vec:
3149        return 1;
3150    case INDEX_op_orc_vec:
3151        return have_isa_2_07;
3152    case INDEX_op_add_vec:
3153    case INDEX_op_sub_vec:
3154    case INDEX_op_smax_vec:
3155    case INDEX_op_smin_vec:
3156    case INDEX_op_umax_vec:
3157    case INDEX_op_umin_vec:
3158    case INDEX_op_shlv_vec:
3159    case INDEX_op_shrv_vec:
3160    case INDEX_op_sarv_vec:
3161    case INDEX_op_rotlv_vec:
3162        return vece <= MO_32 || have_isa_2_07;
3163    case INDEX_op_ssadd_vec:
3164    case INDEX_op_sssub_vec:
3165    case INDEX_op_usadd_vec:
3166    case INDEX_op_ussub_vec:
3167        return vece <= MO_32;
3168    case INDEX_op_cmp_vec:
3169    case INDEX_op_shli_vec:
3170    case INDEX_op_shri_vec:
3171    case INDEX_op_sari_vec:
3172    case INDEX_op_rotli_vec:
3173        return vece <= MO_32 || have_isa_2_07 ? -1 : 0;
3174    case INDEX_op_neg_vec:
3175        return vece >= MO_32 && have_isa_3_00;
3176    case INDEX_op_mul_vec:
3177        switch (vece) {
3178        case MO_8:
3179        case MO_16:
3180            return -1;
3181        case MO_32:
3182            return have_isa_2_07 ? 1 : -1;
3183        case MO_64:
3184            return have_isa_3_10;
3185        }
3186        return 0;
3187    case INDEX_op_bitsel_vec:
3188        return have_vsx;
3189    case INDEX_op_rotrv_vec:
3190        return -1;
3191    default:
3192        return 0;
3193    }
3194}
3195
3196static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
3197                            TCGReg dst, TCGReg src)
3198{
3199    tcg_debug_assert(dst >= TCG_REG_V0);
3200
3201    /* Splat from integer reg allowed via constraints for v3.00.  */
3202    if (src < TCG_REG_V0) {
3203        tcg_debug_assert(have_isa_3_00);
3204        switch (vece) {
3205        case MO_64:
3206            tcg_out32(s, MTVSRDD | VRT(dst) | RA(src) | RB(src));
3207            return true;
3208        case MO_32:
3209            tcg_out32(s, MTVSRWS | VRT(dst) | RA(src));
3210            return true;
3211        default:
3212            /* Fail, so that we fall back on either dupm or mov+dup.  */
3213            return false;
3214        }
3215    }
3216
3217    /*
3218     * Recall we use (or emulate) VSX integer loads, so the integer is
3219     * right justified within the left (zero-index) double-word.
3220     */
3221    switch (vece) {
3222    case MO_8:
3223        tcg_out32(s, VSPLTB | VRT(dst) | VRB(src) | (7 << 16));
3224        break;
3225    case MO_16:
3226        tcg_out32(s, VSPLTH | VRT(dst) | VRB(src) | (3 << 16));
3227        break;
3228    case MO_32:
3229        tcg_out32(s, VSPLTW | VRT(dst) | VRB(src) | (1 << 16));
3230        break;
3231    case MO_64:
3232        if (have_vsx) {
3233            tcg_out32(s, XXPERMDI | VRT(dst) | VRA(src) | VRB(src));
3234            break;
3235        }
3236        tcg_out_vsldoi(s, TCG_VEC_TMP1, src, src, 8);
3237        tcg_out_vsldoi(s, dst, TCG_VEC_TMP1, src, 8);
3238        break;
3239    default:
3240        g_assert_not_reached();
3241    }
3242    return true;
3243}
3244
3245static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
3246                             TCGReg out, TCGReg base, intptr_t offset)
3247{
3248    int elt;
3249
3250    tcg_debug_assert(out >= TCG_REG_V0);
3251    switch (vece) {
3252    case MO_8:
3253        if (have_isa_3_00) {
3254            tcg_out_mem_long(s, LXV, LVX, out, base, offset & -16);
3255        } else {
3256            tcg_out_mem_long(s, 0, LVEBX, out, base, offset);
3257        }
3258        elt = extract32(offset, 0, 4);
3259#if !HOST_BIG_ENDIAN
3260        elt ^= 15;
3261#endif
3262        tcg_out32(s, VSPLTB | VRT(out) | VRB(out) | (elt << 16));
3263        break;
3264    case MO_16:
3265        tcg_debug_assert((offset & 1) == 0);
3266        if (have_isa_3_00) {
3267            tcg_out_mem_long(s, LXV | 8, LVX, out, base, offset & -16);
3268        } else {
3269            tcg_out_mem_long(s, 0, LVEHX, out, base, offset);
3270        }
3271        elt = extract32(offset, 1, 3);
3272#if !HOST_BIG_ENDIAN
3273        elt ^= 7;
3274#endif
3275        tcg_out32(s, VSPLTH | VRT(out) | VRB(out) | (elt << 16));
3276        break;
3277    case MO_32:
3278        if (have_isa_3_00) {
3279            tcg_out_mem_long(s, 0, LXVWSX, out, base, offset);
3280            break;
3281        }
3282        tcg_debug_assert((offset & 3) == 0);
3283        tcg_out_mem_long(s, 0, LVEWX, out, base, offset);
3284        elt = extract32(offset, 2, 2);
3285#if !HOST_BIG_ENDIAN
3286        elt ^= 3;
3287#endif
3288        tcg_out32(s, VSPLTW | VRT(out) | VRB(out) | (elt << 16));
3289        break;
3290    case MO_64:
3291        if (have_vsx) {
3292            tcg_out_mem_long(s, 0, LXVDSX, out, base, offset);
3293            break;
3294        }
3295        tcg_debug_assert((offset & 7) == 0);
3296        tcg_out_mem_long(s, 0, LVX, out, base, offset & -16);
3297        tcg_out_vsldoi(s, TCG_VEC_TMP1, out, out, 8);
3298        elt = extract32(offset, 3, 1);
3299#if !HOST_BIG_ENDIAN
3300        elt = !elt;
3301#endif
3302        if (elt) {
3303            tcg_out_vsldoi(s, out, out, TCG_VEC_TMP1, 8);
3304        } else {
3305            tcg_out_vsldoi(s, out, TCG_VEC_TMP1, out, 8);
3306        }
3307        break;
3308    default:
3309        g_assert_not_reached();
3310    }
3311    return true;
3312}
3313
3314static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
3315                           unsigned vecl, unsigned vece,
3316                           const TCGArg args[TCG_MAX_OP_ARGS],
3317                           const int const_args[TCG_MAX_OP_ARGS])
3318{
3319    static const uint32_t
3320        add_op[4] = { VADDUBM, VADDUHM, VADDUWM, VADDUDM },
3321        sub_op[4] = { VSUBUBM, VSUBUHM, VSUBUWM, VSUBUDM },
3322        mul_op[4] = { 0, 0, VMULUWM, VMULLD },
3323        neg_op[4] = { 0, 0, VNEGW, VNEGD },
3324        eq_op[4]  = { VCMPEQUB, VCMPEQUH, VCMPEQUW, VCMPEQUD },
3325        ne_op[4]  = { VCMPNEB, VCMPNEH, VCMPNEW, 0 },
3326        gts_op[4] = { VCMPGTSB, VCMPGTSH, VCMPGTSW, VCMPGTSD },
3327        gtu_op[4] = { VCMPGTUB, VCMPGTUH, VCMPGTUW, VCMPGTUD },
3328        ssadd_op[4] = { VADDSBS, VADDSHS, VADDSWS, 0 },
3329        usadd_op[4] = { VADDUBS, VADDUHS, VADDUWS, 0 },
3330        sssub_op[4] = { VSUBSBS, VSUBSHS, VSUBSWS, 0 },
3331        ussub_op[4] = { VSUBUBS, VSUBUHS, VSUBUWS, 0 },
3332        umin_op[4] = { VMINUB, VMINUH, VMINUW, VMINUD },
3333        smin_op[4] = { VMINSB, VMINSH, VMINSW, VMINSD },
3334        umax_op[4] = { VMAXUB, VMAXUH, VMAXUW, VMAXUD },
3335        smax_op[4] = { VMAXSB, VMAXSH, VMAXSW, VMAXSD },
3336        shlv_op[4] = { VSLB, VSLH, VSLW, VSLD },
3337        shrv_op[4] = { VSRB, VSRH, VSRW, VSRD },
3338        sarv_op[4] = { VSRAB, VSRAH, VSRAW, VSRAD },
3339        mrgh_op[4] = { VMRGHB, VMRGHH, VMRGHW, 0 },
3340        mrgl_op[4] = { VMRGLB, VMRGLH, VMRGLW, 0 },
3341        muleu_op[4] = { VMULEUB, VMULEUH, VMULEUW, 0 },
3342        mulou_op[4] = { VMULOUB, VMULOUH, VMULOUW, 0 },
3343        pkum_op[4] = { VPKUHUM, VPKUWUM, 0, 0 },
3344        rotl_op[4] = { VRLB, VRLH, VRLW, VRLD };
3345
3346    TCGType type = vecl + TCG_TYPE_V64;
3347    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
3348    uint32_t insn;
3349
3350    switch (opc) {
3351    case INDEX_op_ld_vec:
3352        tcg_out_ld(s, type, a0, a1, a2);
3353        return;
3354    case INDEX_op_st_vec:
3355        tcg_out_st(s, type, a0, a1, a2);
3356        return;
3357    case INDEX_op_dupm_vec:
3358        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
3359        return;
3360
3361    case INDEX_op_add_vec:
3362        insn = add_op[vece];
3363        break;
3364    case INDEX_op_sub_vec:
3365        insn = sub_op[vece];
3366        break;
3367    case INDEX_op_neg_vec:
3368        insn = neg_op[vece];
3369        a2 = a1;
3370        a1 = 0;
3371        break;
3372    case INDEX_op_mul_vec:
3373        insn = mul_op[vece];
3374        break;
3375    case INDEX_op_ssadd_vec:
3376        insn = ssadd_op[vece];
3377        break;
3378    case INDEX_op_sssub_vec:
3379        insn = sssub_op[vece];
3380        break;
3381    case INDEX_op_usadd_vec:
3382        insn = usadd_op[vece];
3383        break;
3384    case INDEX_op_ussub_vec:
3385        insn = ussub_op[vece];
3386        break;
3387    case INDEX_op_smin_vec:
3388        insn = smin_op[vece];
3389        break;
3390    case INDEX_op_umin_vec:
3391        insn = umin_op[vece];
3392        break;
3393    case INDEX_op_smax_vec:
3394        insn = smax_op[vece];
3395        break;
3396    case INDEX_op_umax_vec:
3397        insn = umax_op[vece];
3398        break;
3399    case INDEX_op_shlv_vec:
3400        insn = shlv_op[vece];
3401        break;
3402    case INDEX_op_shrv_vec:
3403        insn = shrv_op[vece];
3404        break;
3405    case INDEX_op_sarv_vec:
3406        insn = sarv_op[vece];
3407        break;
3408    case INDEX_op_and_vec:
3409        insn = VAND;
3410        break;
3411    case INDEX_op_or_vec:
3412        insn = VOR;
3413        break;
3414    case INDEX_op_xor_vec:
3415        insn = VXOR;
3416        break;
3417    case INDEX_op_andc_vec:
3418        insn = VANDC;
3419        break;
3420    case INDEX_op_not_vec:
3421        insn = VNOR;
3422        a2 = a1;
3423        break;
3424    case INDEX_op_orc_vec:
3425        insn = VORC;
3426        break;
3427    case INDEX_op_nand_vec:
3428        insn = VNAND;
3429        break;
3430    case INDEX_op_nor_vec:
3431        insn = VNOR;
3432        break;
3433    case INDEX_op_eqv_vec:
3434        insn = VEQV;
3435        break;
3436
3437    case INDEX_op_cmp_vec:
3438        switch (args[3]) {
3439        case TCG_COND_EQ:
3440            insn = eq_op[vece];
3441            break;
3442        case TCG_COND_NE:
3443            insn = ne_op[vece];
3444            break;
3445        case TCG_COND_GT:
3446            insn = gts_op[vece];
3447            break;
3448        case TCG_COND_GTU:
3449            insn = gtu_op[vece];
3450            break;
3451        default:
3452            g_assert_not_reached();
3453        }
3454        break;
3455
3456    case INDEX_op_bitsel_vec:
3457        tcg_out32(s, XXSEL | VRT(a0) | VRC(a1) | VRB(a2) | VRA(args[3]));
3458        return;
3459
3460    case INDEX_op_dup2_vec:
3461        assert(TCG_TARGET_REG_BITS == 32);
3462        /* With inputs a1 = xLxx, a2 = xHxx  */
3463        tcg_out32(s, VMRGHW | VRT(a0) | VRA(a2) | VRB(a1));  /* a0  = xxHL */
3464        tcg_out_vsldoi(s, TCG_VEC_TMP1, a0, a0, 8);          /* tmp = HLxx */
3465        tcg_out_vsldoi(s, a0, a0, TCG_VEC_TMP1, 8);          /* a0  = HLHL */
3466        return;
3467
3468    case INDEX_op_ppc_mrgh_vec:
3469        insn = mrgh_op[vece];
3470        break;
3471    case INDEX_op_ppc_mrgl_vec:
3472        insn = mrgl_op[vece];
3473        break;
3474    case INDEX_op_ppc_muleu_vec:
3475        insn = muleu_op[vece];
3476        break;
3477    case INDEX_op_ppc_mulou_vec:
3478        insn = mulou_op[vece];
3479        break;
3480    case INDEX_op_ppc_pkum_vec:
3481        insn = pkum_op[vece];
3482        break;
3483    case INDEX_op_rotlv_vec:
3484        insn = rotl_op[vece];
3485        break;
3486    case INDEX_op_ppc_msum_vec:
3487        tcg_debug_assert(vece == MO_16);
3488        tcg_out32(s, VMSUMUHM | VRT(a0) | VRA(a1) | VRB(a2) | VRC(args[3]));
3489        return;
3490
3491    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
3492    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
3493    default:
3494        g_assert_not_reached();
3495    }
3496
3497    tcg_debug_assert(insn != 0);
3498    tcg_out32(s, insn | VRT(a0) | VRA(a1) | VRB(a2));
3499}
3500
3501static void expand_vec_shi(TCGType type, unsigned vece, TCGv_vec v0,
3502                           TCGv_vec v1, TCGArg imm, TCGOpcode opci)
3503{
3504    TCGv_vec t1;
3505
3506    if (vece == MO_32) {
3507        /*
3508         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
3509         * So using negative numbers gets us the 4th bit easily.
3510         */
3511        imm = sextract32(imm, 0, 5);
3512    } else {
3513        imm &= (8 << vece) - 1;
3514    }
3515
3516    /* Splat w/bytes for xxspltib when 2.07 allows MO_64. */
3517    t1 = tcg_constant_vec(type, MO_8, imm);
3518    vec_gen_3(opci, type, vece, tcgv_vec_arg(v0),
3519              tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3520}
3521
3522static void expand_vec_cmp(TCGType type, unsigned vece, TCGv_vec v0,
3523                           TCGv_vec v1, TCGv_vec v2, TCGCond cond)
3524{
3525    bool need_swap = false, need_inv = false;
3526
3527    tcg_debug_assert(vece <= MO_32 || have_isa_2_07);
3528
3529    switch (cond) {
3530    case TCG_COND_EQ:
3531    case TCG_COND_GT:
3532    case TCG_COND_GTU:
3533        break;
3534    case TCG_COND_NE:
3535        if (have_isa_3_00 && vece <= MO_32) {
3536            break;
3537        }
3538        /* fall through */
3539    case TCG_COND_LE:
3540    case TCG_COND_LEU:
3541        need_inv = true;
3542        break;
3543    case TCG_COND_LT:
3544    case TCG_COND_LTU:
3545        need_swap = true;
3546        break;
3547    case TCG_COND_GE:
3548    case TCG_COND_GEU:
3549        need_swap = need_inv = true;
3550        break;
3551    default:
3552        g_assert_not_reached();
3553    }
3554
3555    if (need_inv) {
3556        cond = tcg_invert_cond(cond);
3557    }
3558    if (need_swap) {
3559        TCGv_vec t1;
3560        t1 = v1, v1 = v2, v2 = t1;
3561        cond = tcg_swap_cond(cond);
3562    }
3563
3564    vec_gen_4(INDEX_op_cmp_vec, type, vece, tcgv_vec_arg(v0),
3565              tcgv_vec_arg(v1), tcgv_vec_arg(v2), cond);
3566
3567    if (need_inv) {
3568        tcg_gen_not_vec(vece, v0, v0);
3569    }
3570}
3571
3572static void expand_vec_mul(TCGType type, unsigned vece, TCGv_vec v0,
3573                           TCGv_vec v1, TCGv_vec v2)
3574{
3575    TCGv_vec t1 = tcg_temp_new_vec(type);
3576    TCGv_vec t2 = tcg_temp_new_vec(type);
3577    TCGv_vec c0, c16;
3578
3579    switch (vece) {
3580    case MO_8:
3581    case MO_16:
3582        vec_gen_3(INDEX_op_ppc_muleu_vec, type, vece, tcgv_vec_arg(t1),
3583                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3584        vec_gen_3(INDEX_op_ppc_mulou_vec, type, vece, tcgv_vec_arg(t2),
3585                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3586        vec_gen_3(INDEX_op_ppc_mrgh_vec, type, vece + 1, tcgv_vec_arg(v0),
3587                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3588        vec_gen_3(INDEX_op_ppc_mrgl_vec, type, vece + 1, tcgv_vec_arg(t1),
3589                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3590        vec_gen_3(INDEX_op_ppc_pkum_vec, type, vece, tcgv_vec_arg(v0),
3591                  tcgv_vec_arg(v0), tcgv_vec_arg(t1));
3592	break;
3593
3594    case MO_32:
3595        tcg_debug_assert(!have_isa_2_07);
3596        /*
3597         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
3598         * So using -16 is a quick way to represent 16.
3599         */
3600        c16 = tcg_constant_vec(type, MO_8, -16);
3601        c0 = tcg_constant_vec(type, MO_8, 0);
3602
3603        vec_gen_3(INDEX_op_rotlv_vec, type, MO_32, tcgv_vec_arg(t1),
3604                  tcgv_vec_arg(v2), tcgv_vec_arg(c16));
3605        vec_gen_3(INDEX_op_ppc_mulou_vec, type, MO_16, tcgv_vec_arg(t2),
3606                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3607        vec_gen_4(INDEX_op_ppc_msum_vec, type, MO_16, tcgv_vec_arg(t1),
3608                  tcgv_vec_arg(v1), tcgv_vec_arg(t1), tcgv_vec_arg(c0));
3609        vec_gen_3(INDEX_op_shlv_vec, type, MO_32, tcgv_vec_arg(t1),
3610                  tcgv_vec_arg(t1), tcgv_vec_arg(c16));
3611        tcg_gen_add_vec(MO_32, v0, t1, t2);
3612        break;
3613
3614    default:
3615        g_assert_not_reached();
3616    }
3617    tcg_temp_free_vec(t1);
3618    tcg_temp_free_vec(t2);
3619}
3620
3621void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3622                       TCGArg a0, ...)
3623{
3624    va_list va;
3625    TCGv_vec v0, v1, v2, t0;
3626    TCGArg a2;
3627
3628    va_start(va, a0);
3629    v0 = temp_tcgv_vec(arg_temp(a0));
3630    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3631    a2 = va_arg(va, TCGArg);
3632
3633    switch (opc) {
3634    case INDEX_op_shli_vec:
3635        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shlv_vec);
3636        break;
3637    case INDEX_op_shri_vec:
3638        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shrv_vec);
3639        break;
3640    case INDEX_op_sari_vec:
3641        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_sarv_vec);
3642        break;
3643    case INDEX_op_rotli_vec:
3644        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_rotlv_vec);
3645        break;
3646    case INDEX_op_cmp_vec:
3647        v2 = temp_tcgv_vec(arg_temp(a2));
3648        expand_vec_cmp(type, vece, v0, v1, v2, va_arg(va, TCGArg));
3649        break;
3650    case INDEX_op_mul_vec:
3651        v2 = temp_tcgv_vec(arg_temp(a2));
3652        expand_vec_mul(type, vece, v0, v1, v2);
3653        break;
3654    case INDEX_op_rotlv_vec:
3655        v2 = temp_tcgv_vec(arg_temp(a2));
3656        t0 = tcg_temp_new_vec(type);
3657        tcg_gen_neg_vec(vece, t0, v2);
3658        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3659        tcg_temp_free_vec(t0);
3660        break;
3661    default:
3662        g_assert_not_reached();
3663    }
3664    va_end(va);
3665}
3666
3667static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
3668{
3669    switch (op) {
3670    case INDEX_op_goto_ptr:
3671        return C_O0_I1(r);
3672
3673    case INDEX_op_ld8u_i32:
3674    case INDEX_op_ld8s_i32:
3675    case INDEX_op_ld16u_i32:
3676    case INDEX_op_ld16s_i32:
3677    case INDEX_op_ld_i32:
3678    case INDEX_op_ctpop_i32:
3679    case INDEX_op_neg_i32:
3680    case INDEX_op_not_i32:
3681    case INDEX_op_ext8s_i32:
3682    case INDEX_op_ext16s_i32:
3683    case INDEX_op_bswap16_i32:
3684    case INDEX_op_bswap32_i32:
3685    case INDEX_op_extract_i32:
3686    case INDEX_op_ld8u_i64:
3687    case INDEX_op_ld8s_i64:
3688    case INDEX_op_ld16u_i64:
3689    case INDEX_op_ld16s_i64:
3690    case INDEX_op_ld32u_i64:
3691    case INDEX_op_ld32s_i64:
3692    case INDEX_op_ld_i64:
3693    case INDEX_op_ctpop_i64:
3694    case INDEX_op_neg_i64:
3695    case INDEX_op_not_i64:
3696    case INDEX_op_ext8s_i64:
3697    case INDEX_op_ext16s_i64:
3698    case INDEX_op_ext32s_i64:
3699    case INDEX_op_ext_i32_i64:
3700    case INDEX_op_extu_i32_i64:
3701    case INDEX_op_bswap16_i64:
3702    case INDEX_op_bswap32_i64:
3703    case INDEX_op_bswap64_i64:
3704    case INDEX_op_extract_i64:
3705        return C_O1_I1(r, r);
3706
3707    case INDEX_op_st8_i32:
3708    case INDEX_op_st16_i32:
3709    case INDEX_op_st_i32:
3710    case INDEX_op_st8_i64:
3711    case INDEX_op_st16_i64:
3712    case INDEX_op_st32_i64:
3713    case INDEX_op_st_i64:
3714        return C_O0_I2(r, r);
3715
3716    case INDEX_op_add_i32:
3717    case INDEX_op_and_i32:
3718    case INDEX_op_or_i32:
3719    case INDEX_op_xor_i32:
3720    case INDEX_op_andc_i32:
3721    case INDEX_op_orc_i32:
3722    case INDEX_op_eqv_i32:
3723    case INDEX_op_shl_i32:
3724    case INDEX_op_shr_i32:
3725    case INDEX_op_sar_i32:
3726    case INDEX_op_rotl_i32:
3727    case INDEX_op_rotr_i32:
3728    case INDEX_op_setcond_i32:
3729    case INDEX_op_and_i64:
3730    case INDEX_op_andc_i64:
3731    case INDEX_op_shl_i64:
3732    case INDEX_op_shr_i64:
3733    case INDEX_op_sar_i64:
3734    case INDEX_op_rotl_i64:
3735    case INDEX_op_rotr_i64:
3736    case INDEX_op_setcond_i64:
3737        return C_O1_I2(r, r, ri);
3738
3739    case INDEX_op_mul_i32:
3740    case INDEX_op_mul_i64:
3741        return C_O1_I2(r, r, rI);
3742
3743    case INDEX_op_div_i32:
3744    case INDEX_op_divu_i32:
3745    case INDEX_op_rem_i32:
3746    case INDEX_op_remu_i32:
3747    case INDEX_op_nand_i32:
3748    case INDEX_op_nor_i32:
3749    case INDEX_op_muluh_i32:
3750    case INDEX_op_mulsh_i32:
3751    case INDEX_op_orc_i64:
3752    case INDEX_op_eqv_i64:
3753    case INDEX_op_nand_i64:
3754    case INDEX_op_nor_i64:
3755    case INDEX_op_div_i64:
3756    case INDEX_op_divu_i64:
3757    case INDEX_op_rem_i64:
3758    case INDEX_op_remu_i64:
3759    case INDEX_op_mulsh_i64:
3760    case INDEX_op_muluh_i64:
3761        return C_O1_I2(r, r, r);
3762
3763    case INDEX_op_sub_i32:
3764        return C_O1_I2(r, rI, ri);
3765    case INDEX_op_add_i64:
3766        return C_O1_I2(r, r, rT);
3767    case INDEX_op_or_i64:
3768    case INDEX_op_xor_i64:
3769        return C_O1_I2(r, r, rU);
3770    case INDEX_op_sub_i64:
3771        return C_O1_I2(r, rI, rT);
3772    case INDEX_op_clz_i32:
3773    case INDEX_op_ctz_i32:
3774    case INDEX_op_clz_i64:
3775    case INDEX_op_ctz_i64:
3776        return C_O1_I2(r, r, rZW);
3777
3778    case INDEX_op_brcond_i32:
3779    case INDEX_op_brcond_i64:
3780        return C_O0_I2(r, ri);
3781
3782    case INDEX_op_movcond_i32:
3783    case INDEX_op_movcond_i64:
3784        return C_O1_I4(r, r, ri, rZ, rZ);
3785    case INDEX_op_deposit_i32:
3786    case INDEX_op_deposit_i64:
3787        return C_O1_I2(r, 0, rZ);
3788    case INDEX_op_brcond2_i32:
3789        return C_O0_I4(r, r, ri, ri);
3790    case INDEX_op_setcond2_i32:
3791        return C_O1_I4(r, r, r, ri, ri);
3792    case INDEX_op_add2_i64:
3793    case INDEX_op_add2_i32:
3794        return C_O2_I4(r, r, r, r, rI, rZM);
3795    case INDEX_op_sub2_i64:
3796    case INDEX_op_sub2_i32:
3797        return C_O2_I4(r, r, rI, rZM, r, r);
3798
3799    case INDEX_op_qemu_ld_a32_i32:
3800        return C_O1_I1(r, r);
3801    case INDEX_op_qemu_ld_a64_i32:
3802        return TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, r) : C_O1_I2(r, r, r);
3803    case INDEX_op_qemu_ld_a32_i64:
3804        return TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, r) : C_O2_I1(r, r, r);
3805    case INDEX_op_qemu_ld_a64_i64:
3806        return TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, r) : C_O2_I2(r, r, r, r);
3807
3808    case INDEX_op_qemu_st_a32_i32:
3809        return C_O0_I2(r, r);
3810    case INDEX_op_qemu_st_a64_i32:
3811        return TCG_TARGET_REG_BITS == 64 ? C_O0_I2(r, r) : C_O0_I3(r, r, r);
3812    case INDEX_op_qemu_st_a32_i64:
3813        return TCG_TARGET_REG_BITS == 64 ? C_O0_I2(r, r) : C_O0_I3(r, r, r);
3814    case INDEX_op_qemu_st_a64_i64:
3815        return TCG_TARGET_REG_BITS == 64 ? C_O0_I2(r, r) : C_O0_I4(r, r, r, r);
3816
3817    case INDEX_op_qemu_ld_a32_i128:
3818    case INDEX_op_qemu_ld_a64_i128:
3819        return C_O2_I1(o, m, r);
3820    case INDEX_op_qemu_st_a32_i128:
3821    case INDEX_op_qemu_st_a64_i128:
3822        return C_O0_I3(o, m, r);
3823
3824    case INDEX_op_add_vec:
3825    case INDEX_op_sub_vec:
3826    case INDEX_op_mul_vec:
3827    case INDEX_op_and_vec:
3828    case INDEX_op_or_vec:
3829    case INDEX_op_xor_vec:
3830    case INDEX_op_andc_vec:
3831    case INDEX_op_orc_vec:
3832    case INDEX_op_nor_vec:
3833    case INDEX_op_eqv_vec:
3834    case INDEX_op_nand_vec:
3835    case INDEX_op_cmp_vec:
3836    case INDEX_op_ssadd_vec:
3837    case INDEX_op_sssub_vec:
3838    case INDEX_op_usadd_vec:
3839    case INDEX_op_ussub_vec:
3840    case INDEX_op_smax_vec:
3841    case INDEX_op_smin_vec:
3842    case INDEX_op_umax_vec:
3843    case INDEX_op_umin_vec:
3844    case INDEX_op_shlv_vec:
3845    case INDEX_op_shrv_vec:
3846    case INDEX_op_sarv_vec:
3847    case INDEX_op_rotlv_vec:
3848    case INDEX_op_rotrv_vec:
3849    case INDEX_op_ppc_mrgh_vec:
3850    case INDEX_op_ppc_mrgl_vec:
3851    case INDEX_op_ppc_muleu_vec:
3852    case INDEX_op_ppc_mulou_vec:
3853    case INDEX_op_ppc_pkum_vec:
3854    case INDEX_op_dup2_vec:
3855        return C_O1_I2(v, v, v);
3856
3857    case INDEX_op_not_vec:
3858    case INDEX_op_neg_vec:
3859        return C_O1_I1(v, v);
3860
3861    case INDEX_op_dup_vec:
3862        return have_isa_3_00 ? C_O1_I1(v, vr) : C_O1_I1(v, v);
3863
3864    case INDEX_op_ld_vec:
3865    case INDEX_op_dupm_vec:
3866        return C_O1_I1(v, r);
3867
3868    case INDEX_op_st_vec:
3869        return C_O0_I2(v, r);
3870
3871    case INDEX_op_bitsel_vec:
3872    case INDEX_op_ppc_msum_vec:
3873        return C_O1_I3(v, v, v, v);
3874
3875    default:
3876        g_assert_not_reached();
3877    }
3878}
3879
3880static void tcg_target_init(TCGContext *s)
3881{
3882    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3883    unsigned long hwcap2 = qemu_getauxval(AT_HWCAP2);
3884
3885    have_isa = tcg_isa_base;
3886    if (hwcap & PPC_FEATURE_ARCH_2_06) {
3887        have_isa = tcg_isa_2_06;
3888    }
3889#ifdef PPC_FEATURE2_ARCH_2_07
3890    if (hwcap2 & PPC_FEATURE2_ARCH_2_07) {
3891        have_isa = tcg_isa_2_07;
3892    }
3893#endif
3894#ifdef PPC_FEATURE2_ARCH_3_00
3895    if (hwcap2 & PPC_FEATURE2_ARCH_3_00) {
3896        have_isa = tcg_isa_3_00;
3897    }
3898#endif
3899#ifdef PPC_FEATURE2_ARCH_3_10
3900    if (hwcap2 & PPC_FEATURE2_ARCH_3_10) {
3901        have_isa = tcg_isa_3_10;
3902    }
3903#endif
3904
3905#ifdef PPC_FEATURE2_HAS_ISEL
3906    /* Prefer explicit instruction from the kernel. */
3907    have_isel = (hwcap2 & PPC_FEATURE2_HAS_ISEL) != 0;
3908#else
3909    /* Fall back to knowing Power7 (2.06) has ISEL. */
3910    have_isel = have_isa_2_06;
3911#endif
3912
3913    if (hwcap & PPC_FEATURE_HAS_ALTIVEC) {
3914        have_altivec = true;
3915        /* We only care about the portion of VSX that overlaps Altivec. */
3916        if (hwcap & PPC_FEATURE_HAS_VSX) {
3917            have_vsx = true;
3918        }
3919    }
3920
3921    tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
3922    tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
3923    if (have_altivec) {
3924        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3925        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3926    }
3927
3928    tcg_target_call_clobber_regs = 0;
3929    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3930    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3931    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3932    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3933    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3934    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3935    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R7);
3936    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R8);
3937    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R9);
3938    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R10);
3939    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R11);
3940    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
3941
3942    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3943    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3944    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3945    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3946    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3947    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3948    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3949    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3950    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V8);
3951    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V9);
3952    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V10);
3953    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V11);
3954    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V12);
3955    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V13);
3956    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V14);
3957    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V15);
3958    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3959    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3960    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3961    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3962
3963    s->reserved_regs = 0;
3964    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); /* tcg temp */
3965    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R1); /* stack pointer */
3966#if defined(_CALL_SYSV)
3967    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R2); /* toc pointer */
3968#endif
3969#if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
3970    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R13); /* thread pointer */
3971#endif
3972    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1);
3973    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP2);
3974    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP1);
3975    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP2);
3976    if (USE_REG_TB) {
3977        tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);  /* tb->tc_ptr */
3978    }
3979}
3980
3981#ifdef __ELF__
3982typedef struct {
3983    DebugFrameCIE cie;
3984    DebugFrameFDEHeader fde;
3985    uint8_t fde_def_cfa[4];
3986    uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2 + 3];
3987} DebugFrame;
3988
3989/* We're expecting a 2 byte uleb128 encoded value.  */
3990QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3991
3992#if TCG_TARGET_REG_BITS == 64
3993# define ELF_HOST_MACHINE EM_PPC64
3994#else
3995# define ELF_HOST_MACHINE EM_PPC
3996#endif
3997
3998static DebugFrame debug_frame = {
3999    .cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
4000    .cie.id = -1,
4001    .cie.version = 1,
4002    .cie.code_align = 1,
4003    .cie.data_align = (-SZR & 0x7f),         /* sleb128 -SZR */
4004    .cie.return_column = 65,
4005
4006    /* Total FDE size does not include the "len" member.  */
4007    .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset),
4008
4009    .fde_def_cfa = {
4010        12, TCG_REG_R1,                 /* DW_CFA_def_cfa r1, ... */
4011        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
4012        (FRAME_SIZE >> 7)
4013    },
4014    .fde_reg_ofs = {
4015        /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
4016        0x11, 65, (LR_OFFSET / -SZR) & 0x7f,
4017    }
4018};
4019
4020void tcg_register_jit(const void *buf, size_t buf_size)
4021{
4022    uint8_t *p = &debug_frame.fde_reg_ofs[3];
4023    int i;
4024
4025    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i, p += 2) {
4026        p[0] = 0x80 + tcg_target_callee_save_regs[i];
4027        p[1] = (FRAME_SIZE - (REG_SAVE_BOT + i * SZR)) / SZR;
4028    }
4029
4030    debug_frame.fde.func_start = (uintptr_t)buf;
4031    debug_frame.fde.func_len = buf_size;
4032
4033    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
4034}
4035#endif /* __ELF__ */
4036#undef VMULEUB
4037#undef VMULEUH
4038#undef VMULEUW
4039#undef VMULOUB
4040#undef VMULOUH
4041#undef VMULOUW
4042#undef VMSUMUHM
4043