xref: /openbmc/qemu/tcg/ppc/tcg-target.c.inc (revision edf5ca5dbe8031e7814ea34eb109b8f7d4024ae5)
1/*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2008 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25#include "elf.h"
26#include "../tcg-pool.c.inc"
27#include "../tcg-ldst.c.inc"
28
29/*
30 * Standardize on the _CALL_FOO symbols used by GCC:
31 * Apple XCode does not define _CALL_DARWIN.
32 * Clang defines _CALL_ELF (64-bit) but not _CALL_SYSV (32-bit).
33 */
34#if !defined(_CALL_SYSV) && \
35    !defined(_CALL_DARWIN) && \
36    !defined(_CALL_AIX) && \
37    !defined(_CALL_ELF)
38# if defined(__APPLE__)
39#  define _CALL_DARWIN
40# elif defined(__ELF__) && TCG_TARGET_REG_BITS == 32
41#  define _CALL_SYSV
42# else
43#  error "Unknown ABI"
44# endif
45#endif
46
47#if TCG_TARGET_REG_BITS == 64
48# define TCG_TARGET_CALL_ARG_I32   TCG_CALL_ARG_EXTEND
49#else
50# define TCG_TARGET_CALL_ARG_I32   TCG_CALL_ARG_NORMAL
51#endif
52#ifdef _CALL_SYSV
53# define TCG_TARGET_CALL_ARG_I64   TCG_CALL_ARG_EVEN
54#else
55# define TCG_TARGET_CALL_ARG_I64   TCG_CALL_ARG_NORMAL
56#endif
57
58/* For some memory operations, we need a scratch that isn't R0.  For the AIX
59   calling convention, we can re-use the TOC register since we'll be reloading
60   it at every call.  Otherwise R12 will do nicely as neither a call-saved
61   register nor a parameter register.  */
62#ifdef _CALL_AIX
63# define TCG_REG_TMP1   TCG_REG_R2
64#else
65# define TCG_REG_TMP1   TCG_REG_R12
66#endif
67
68#define TCG_VEC_TMP1    TCG_REG_V0
69#define TCG_VEC_TMP2    TCG_REG_V1
70
71#define TCG_REG_TB     TCG_REG_R31
72#define USE_REG_TB     (TCG_TARGET_REG_BITS == 64)
73
74/* Shorthand for size of a pointer.  Avoid promotion to unsigned.  */
75#define SZP  ((int)sizeof(void *))
76
77/* Shorthand for size of a register.  */
78#define SZR  (TCG_TARGET_REG_BITS / 8)
79
80#define TCG_CT_CONST_S16  0x100
81#define TCG_CT_CONST_U16  0x200
82#define TCG_CT_CONST_S32  0x400
83#define TCG_CT_CONST_U32  0x800
84#define TCG_CT_CONST_ZERO 0x1000
85#define TCG_CT_CONST_MONE 0x2000
86#define TCG_CT_CONST_WSZ  0x4000
87
88#define ALL_GENERAL_REGS  0xffffffffu
89#define ALL_VECTOR_REGS   0xffffffff00000000ull
90
91#ifdef CONFIG_SOFTMMU
92#define ALL_QLOAD_REGS \
93    (ALL_GENERAL_REGS & \
94     ~((1 << TCG_REG_R3) | (1 << TCG_REG_R4) | (1 << TCG_REG_R5)))
95#define ALL_QSTORE_REGS \
96    (ALL_GENERAL_REGS & ~((1 << TCG_REG_R3) | (1 << TCG_REG_R4) | \
97                          (1 << TCG_REG_R5) | (1 << TCG_REG_R6)))
98#else
99#define ALL_QLOAD_REGS  (ALL_GENERAL_REGS & ~(1 << TCG_REG_R3))
100#define ALL_QSTORE_REGS ALL_QLOAD_REGS
101#endif
102
103TCGPowerISA have_isa;
104static bool have_isel;
105bool have_altivec;
106bool have_vsx;
107
108#ifndef CONFIG_SOFTMMU
109#define TCG_GUEST_BASE_REG 30
110#endif
111
112#ifdef CONFIG_DEBUG_TCG
113static const char tcg_target_reg_names[TCG_TARGET_NB_REGS][4] = {
114    "r0",  "r1",  "r2",  "r3",  "r4",  "r5",  "r6",  "r7",
115    "r8",  "r9",  "r10", "r11", "r12", "r13", "r14", "r15",
116    "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
117    "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31",
118    "v0",  "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7",
119    "v8",  "v9",  "v10", "v11", "v12", "v13", "v14", "v15",
120    "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
121    "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",
122};
123#endif
124
125static const int tcg_target_reg_alloc_order[] = {
126    TCG_REG_R14,  /* call saved registers */
127    TCG_REG_R15,
128    TCG_REG_R16,
129    TCG_REG_R17,
130    TCG_REG_R18,
131    TCG_REG_R19,
132    TCG_REG_R20,
133    TCG_REG_R21,
134    TCG_REG_R22,
135    TCG_REG_R23,
136    TCG_REG_R24,
137    TCG_REG_R25,
138    TCG_REG_R26,
139    TCG_REG_R27,
140    TCG_REG_R28,
141    TCG_REG_R29,
142    TCG_REG_R30,
143    TCG_REG_R31,
144    TCG_REG_R12,  /* call clobbered, non-arguments */
145    TCG_REG_R11,
146    TCG_REG_R2,
147    TCG_REG_R13,
148    TCG_REG_R10,  /* call clobbered, arguments */
149    TCG_REG_R9,
150    TCG_REG_R8,
151    TCG_REG_R7,
152    TCG_REG_R6,
153    TCG_REG_R5,
154    TCG_REG_R4,
155    TCG_REG_R3,
156
157    /* V0 and V1 reserved as temporaries; V20 - V31 are call-saved */
158    TCG_REG_V2,   /* call clobbered, vectors */
159    TCG_REG_V3,
160    TCG_REG_V4,
161    TCG_REG_V5,
162    TCG_REG_V6,
163    TCG_REG_V7,
164    TCG_REG_V8,
165    TCG_REG_V9,
166    TCG_REG_V10,
167    TCG_REG_V11,
168    TCG_REG_V12,
169    TCG_REG_V13,
170    TCG_REG_V14,
171    TCG_REG_V15,
172    TCG_REG_V16,
173    TCG_REG_V17,
174    TCG_REG_V18,
175    TCG_REG_V19,
176};
177
178static const int tcg_target_call_iarg_regs[] = {
179    TCG_REG_R3,
180    TCG_REG_R4,
181    TCG_REG_R5,
182    TCG_REG_R6,
183    TCG_REG_R7,
184    TCG_REG_R8,
185    TCG_REG_R9,
186    TCG_REG_R10
187};
188
189static const int tcg_target_call_oarg_regs[] = {
190    TCG_REG_R3,
191    TCG_REG_R4
192};
193
194static const int tcg_target_callee_save_regs[] = {
195#ifdef _CALL_DARWIN
196    TCG_REG_R11,
197#endif
198    TCG_REG_R14,
199    TCG_REG_R15,
200    TCG_REG_R16,
201    TCG_REG_R17,
202    TCG_REG_R18,
203    TCG_REG_R19,
204    TCG_REG_R20,
205    TCG_REG_R21,
206    TCG_REG_R22,
207    TCG_REG_R23,
208    TCG_REG_R24,
209    TCG_REG_R25,
210    TCG_REG_R26,
211    TCG_REG_R27, /* currently used for the global env */
212    TCG_REG_R28,
213    TCG_REG_R29,
214    TCG_REG_R30,
215    TCG_REG_R31
216};
217
218static inline bool in_range_b(tcg_target_long target)
219{
220    return target == sextract64(target, 0, 26);
221}
222
223static uint32_t reloc_pc24_val(const tcg_insn_unit *pc,
224			       const tcg_insn_unit *target)
225{
226    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
227    tcg_debug_assert(in_range_b(disp));
228    return disp & 0x3fffffc;
229}
230
231static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
232{
233    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
234    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
235
236    if (in_range_b(disp)) {
237        *src_rw = (*src_rw & ~0x3fffffc) | (disp & 0x3fffffc);
238        return true;
239    }
240    return false;
241}
242
243static uint16_t reloc_pc14_val(const tcg_insn_unit *pc,
244			       const tcg_insn_unit *target)
245{
246    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
247    tcg_debug_assert(disp == (int16_t) disp);
248    return disp & 0xfffc;
249}
250
251static bool reloc_pc14(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
252{
253    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
254    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
255
256    if (disp == (int16_t) disp) {
257        *src_rw = (*src_rw & ~0xfffc) | (disp & 0xfffc);
258        return true;
259    }
260    return false;
261}
262
263/* test if a constant matches the constraint */
264static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
265{
266    if (ct & TCG_CT_CONST) {
267        return 1;
268    }
269
270    /* The only 32-bit constraint we use aside from
271       TCG_CT_CONST is TCG_CT_CONST_S16.  */
272    if (type == TCG_TYPE_I32) {
273        val = (int32_t)val;
274    }
275
276    if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
277        return 1;
278    } else if ((ct & TCG_CT_CONST_U16) && val == (uint16_t)val) {
279        return 1;
280    } else if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
281        return 1;
282    } else if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
283        return 1;
284    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
285        return 1;
286    } else if ((ct & TCG_CT_CONST_MONE) && val == -1) {
287        return 1;
288    } else if ((ct & TCG_CT_CONST_WSZ)
289               && val == (type == TCG_TYPE_I32 ? 32 : 64)) {
290        return 1;
291    }
292    return 0;
293}
294
295#define OPCD(opc) ((opc)<<26)
296#define XO19(opc) (OPCD(19)|((opc)<<1))
297#define MD30(opc) (OPCD(30)|((opc)<<2))
298#define MDS30(opc) (OPCD(30)|((opc)<<1))
299#define XO31(opc) (OPCD(31)|((opc)<<1))
300#define XO58(opc) (OPCD(58)|(opc))
301#define XO62(opc) (OPCD(62)|(opc))
302#define VX4(opc)  (OPCD(4)|(opc))
303
304#define B      OPCD( 18)
305#define BC     OPCD( 16)
306#define LBZ    OPCD( 34)
307#define LHZ    OPCD( 40)
308#define LHA    OPCD( 42)
309#define LWZ    OPCD( 32)
310#define LWZUX  XO31( 55)
311#define STB    OPCD( 38)
312#define STH    OPCD( 44)
313#define STW    OPCD( 36)
314
315#define STD    XO62(  0)
316#define STDU   XO62(  1)
317#define STDX   XO31(149)
318
319#define LD     XO58(  0)
320#define LDX    XO31( 21)
321#define LDU    XO58(  1)
322#define LDUX   XO31( 53)
323#define LWA    XO58(  2)
324#define LWAX   XO31(341)
325
326#define ADDIC  OPCD( 12)
327#define ADDI   OPCD( 14)
328#define ADDIS  OPCD( 15)
329#define ORI    OPCD( 24)
330#define ORIS   OPCD( 25)
331#define XORI   OPCD( 26)
332#define XORIS  OPCD( 27)
333#define ANDI   OPCD( 28)
334#define ANDIS  OPCD( 29)
335#define MULLI  OPCD(  7)
336#define CMPLI  OPCD( 10)
337#define CMPI   OPCD( 11)
338#define SUBFIC OPCD( 8)
339
340#define LWZU   OPCD( 33)
341#define STWU   OPCD( 37)
342
343#define RLWIMI OPCD( 20)
344#define RLWINM OPCD( 21)
345#define RLWNM  OPCD( 23)
346
347#define RLDICL MD30(  0)
348#define RLDICR MD30(  1)
349#define RLDIMI MD30(  3)
350#define RLDCL  MDS30( 8)
351
352#define BCLR   XO19( 16)
353#define BCCTR  XO19(528)
354#define CRAND  XO19(257)
355#define CRANDC XO19(129)
356#define CRNAND XO19(225)
357#define CROR   XO19(449)
358#define CRNOR  XO19( 33)
359
360#define EXTSB  XO31(954)
361#define EXTSH  XO31(922)
362#define EXTSW  XO31(986)
363#define ADD    XO31(266)
364#define ADDE   XO31(138)
365#define ADDME  XO31(234)
366#define ADDZE  XO31(202)
367#define ADDC   XO31( 10)
368#define AND    XO31( 28)
369#define SUBF   XO31( 40)
370#define SUBFC  XO31(  8)
371#define SUBFE  XO31(136)
372#define SUBFME XO31(232)
373#define SUBFZE XO31(200)
374#define OR     XO31(444)
375#define XOR    XO31(316)
376#define MULLW  XO31(235)
377#define MULHW  XO31( 75)
378#define MULHWU XO31( 11)
379#define DIVW   XO31(491)
380#define DIVWU  XO31(459)
381#define MODSW  XO31(779)
382#define MODUW  XO31(267)
383#define CMP    XO31(  0)
384#define CMPL   XO31( 32)
385#define LHBRX  XO31(790)
386#define LWBRX  XO31(534)
387#define LDBRX  XO31(532)
388#define STHBRX XO31(918)
389#define STWBRX XO31(662)
390#define STDBRX XO31(660)
391#define MFSPR  XO31(339)
392#define MTSPR  XO31(467)
393#define SRAWI  XO31(824)
394#define NEG    XO31(104)
395#define MFCR   XO31( 19)
396#define MFOCRF (MFCR | (1u << 20))
397#define NOR    XO31(124)
398#define CNTLZW XO31( 26)
399#define CNTLZD XO31( 58)
400#define CNTTZW XO31(538)
401#define CNTTZD XO31(570)
402#define CNTPOPW XO31(378)
403#define CNTPOPD XO31(506)
404#define ANDC   XO31( 60)
405#define ORC    XO31(412)
406#define EQV    XO31(284)
407#define NAND   XO31(476)
408#define ISEL   XO31( 15)
409
410#define MULLD  XO31(233)
411#define MULHD  XO31( 73)
412#define MULHDU XO31(  9)
413#define DIVD   XO31(489)
414#define DIVDU  XO31(457)
415#define MODSD  XO31(777)
416#define MODUD  XO31(265)
417
418#define LBZX   XO31( 87)
419#define LHZX   XO31(279)
420#define LHAX   XO31(343)
421#define LWZX   XO31( 23)
422#define STBX   XO31(215)
423#define STHX   XO31(407)
424#define STWX   XO31(151)
425
426#define EIEIO  XO31(854)
427#define HWSYNC XO31(598)
428#define LWSYNC (HWSYNC | (1u << 21))
429
430#define SPR(a, b) ((((a)<<5)|(b))<<11)
431#define LR     SPR(8, 0)
432#define CTR    SPR(9, 0)
433
434#define SLW    XO31( 24)
435#define SRW    XO31(536)
436#define SRAW   XO31(792)
437
438#define SLD    XO31( 27)
439#define SRD    XO31(539)
440#define SRAD   XO31(794)
441#define SRADI  XO31(413<<1)
442
443#define BRH    XO31(219)
444#define BRW    XO31(155)
445#define BRD    XO31(187)
446
447#define TW     XO31( 4)
448#define TRAP   (TW | TO(31))
449
450#define NOP    ORI  /* ori 0,0,0 */
451
452#define LVX        XO31(103)
453#define LVEBX      XO31(7)
454#define LVEHX      XO31(39)
455#define LVEWX      XO31(71)
456#define LXSDX      (XO31(588) | 1)  /* v2.06, force tx=1 */
457#define LXVDSX     (XO31(332) | 1)  /* v2.06, force tx=1 */
458#define LXSIWZX    (XO31(12) | 1)   /* v2.07, force tx=1 */
459#define LXV        (OPCD(61) | 8 | 1)  /* v3.00, force tx=1 */
460#define LXSD       (OPCD(57) | 2)   /* v3.00 */
461#define LXVWSX     (XO31(364) | 1)  /* v3.00, force tx=1 */
462
463#define STVX       XO31(231)
464#define STVEWX     XO31(199)
465#define STXSDX     (XO31(716) | 1)  /* v2.06, force sx=1 */
466#define STXSIWX    (XO31(140) | 1)  /* v2.07, force sx=1 */
467#define STXV       (OPCD(61) | 8 | 5) /* v3.00, force sx=1 */
468#define STXSD      (OPCD(61) | 2)   /* v3.00 */
469
470#define VADDSBS    VX4(768)
471#define VADDUBS    VX4(512)
472#define VADDUBM    VX4(0)
473#define VADDSHS    VX4(832)
474#define VADDUHS    VX4(576)
475#define VADDUHM    VX4(64)
476#define VADDSWS    VX4(896)
477#define VADDUWS    VX4(640)
478#define VADDUWM    VX4(128)
479#define VADDUDM    VX4(192)       /* v2.07 */
480
481#define VSUBSBS    VX4(1792)
482#define VSUBUBS    VX4(1536)
483#define VSUBUBM    VX4(1024)
484#define VSUBSHS    VX4(1856)
485#define VSUBUHS    VX4(1600)
486#define VSUBUHM    VX4(1088)
487#define VSUBSWS    VX4(1920)
488#define VSUBUWS    VX4(1664)
489#define VSUBUWM    VX4(1152)
490#define VSUBUDM    VX4(1216)      /* v2.07 */
491
492#define VNEGW      (VX4(1538) | (6 << 16))  /* v3.00 */
493#define VNEGD      (VX4(1538) | (7 << 16))  /* v3.00 */
494
495#define VMAXSB     VX4(258)
496#define VMAXSH     VX4(322)
497#define VMAXSW     VX4(386)
498#define VMAXSD     VX4(450)       /* v2.07 */
499#define VMAXUB     VX4(2)
500#define VMAXUH     VX4(66)
501#define VMAXUW     VX4(130)
502#define VMAXUD     VX4(194)       /* v2.07 */
503#define VMINSB     VX4(770)
504#define VMINSH     VX4(834)
505#define VMINSW     VX4(898)
506#define VMINSD     VX4(962)       /* v2.07 */
507#define VMINUB     VX4(514)
508#define VMINUH     VX4(578)
509#define VMINUW     VX4(642)
510#define VMINUD     VX4(706)       /* v2.07 */
511
512#define VCMPEQUB   VX4(6)
513#define VCMPEQUH   VX4(70)
514#define VCMPEQUW   VX4(134)
515#define VCMPEQUD   VX4(199)       /* v2.07 */
516#define VCMPGTSB   VX4(774)
517#define VCMPGTSH   VX4(838)
518#define VCMPGTSW   VX4(902)
519#define VCMPGTSD   VX4(967)       /* v2.07 */
520#define VCMPGTUB   VX4(518)
521#define VCMPGTUH   VX4(582)
522#define VCMPGTUW   VX4(646)
523#define VCMPGTUD   VX4(711)       /* v2.07 */
524#define VCMPNEB    VX4(7)         /* v3.00 */
525#define VCMPNEH    VX4(71)        /* v3.00 */
526#define VCMPNEW    VX4(135)       /* v3.00 */
527
528#define VSLB       VX4(260)
529#define VSLH       VX4(324)
530#define VSLW       VX4(388)
531#define VSLD       VX4(1476)      /* v2.07 */
532#define VSRB       VX4(516)
533#define VSRH       VX4(580)
534#define VSRW       VX4(644)
535#define VSRD       VX4(1732)      /* v2.07 */
536#define VSRAB      VX4(772)
537#define VSRAH      VX4(836)
538#define VSRAW      VX4(900)
539#define VSRAD      VX4(964)       /* v2.07 */
540#define VRLB       VX4(4)
541#define VRLH       VX4(68)
542#define VRLW       VX4(132)
543#define VRLD       VX4(196)       /* v2.07 */
544
545#define VMULEUB    VX4(520)
546#define VMULEUH    VX4(584)
547#define VMULEUW    VX4(648)       /* v2.07 */
548#define VMULOUB    VX4(8)
549#define VMULOUH    VX4(72)
550#define VMULOUW    VX4(136)       /* v2.07 */
551#define VMULUWM    VX4(137)       /* v2.07 */
552#define VMULLD     VX4(457)       /* v3.10 */
553#define VMSUMUHM   VX4(38)
554
555#define VMRGHB     VX4(12)
556#define VMRGHH     VX4(76)
557#define VMRGHW     VX4(140)
558#define VMRGLB     VX4(268)
559#define VMRGLH     VX4(332)
560#define VMRGLW     VX4(396)
561
562#define VPKUHUM    VX4(14)
563#define VPKUWUM    VX4(78)
564
565#define VAND       VX4(1028)
566#define VANDC      VX4(1092)
567#define VNOR       VX4(1284)
568#define VOR        VX4(1156)
569#define VXOR       VX4(1220)
570#define VEQV       VX4(1668)      /* v2.07 */
571#define VNAND      VX4(1412)      /* v2.07 */
572#define VORC       VX4(1348)      /* v2.07 */
573
574#define VSPLTB     VX4(524)
575#define VSPLTH     VX4(588)
576#define VSPLTW     VX4(652)
577#define VSPLTISB   VX4(780)
578#define VSPLTISH   VX4(844)
579#define VSPLTISW   VX4(908)
580
581#define VSLDOI     VX4(44)
582
583#define XXPERMDI   (OPCD(60) | (10 << 3) | 7)  /* v2.06, force ax=bx=tx=1 */
584#define XXSEL      (OPCD(60) | (3 << 4) | 0xf) /* v2.06, force ax=bx=cx=tx=1 */
585#define XXSPLTIB   (OPCD(60) | (360 << 1) | 1) /* v3.00, force tx=1 */
586
587#define MFVSRD     (XO31(51) | 1)   /* v2.07, force sx=1 */
588#define MFVSRWZ    (XO31(115) | 1)  /* v2.07, force sx=1 */
589#define MTVSRD     (XO31(179) | 1)  /* v2.07, force tx=1 */
590#define MTVSRWZ    (XO31(243) | 1)  /* v2.07, force tx=1 */
591#define MTVSRDD    (XO31(435) | 1)  /* v3.00, force tx=1 */
592#define MTVSRWS    (XO31(403) | 1)  /* v3.00, force tx=1 */
593
594#define RT(r) ((r)<<21)
595#define RS(r) ((r)<<21)
596#define RA(r) ((r)<<16)
597#define RB(r) ((r)<<11)
598#define TO(t) ((t)<<21)
599#define SH(s) ((s)<<11)
600#define MB(b) ((b)<<6)
601#define ME(e) ((e)<<1)
602#define BO(o) ((o)<<21)
603#define MB64(b) ((b)<<5)
604#define FXM(b) (1 << (19 - (b)))
605
606#define VRT(r)  (((r) & 31) << 21)
607#define VRA(r)  (((r) & 31) << 16)
608#define VRB(r)  (((r) & 31) << 11)
609#define VRC(r)  (((r) & 31) <<  6)
610
611#define LK    1
612
613#define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
614#define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
615#define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
616#define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
617
618#define BF(n)    ((n)<<23)
619#define BI(n, c) (((c)+((n)*4))<<16)
620#define BT(n, c) (((c)+((n)*4))<<21)
621#define BA(n, c) (((c)+((n)*4))<<16)
622#define BB(n, c) (((c)+((n)*4))<<11)
623#define BC_(n, c) (((c)+((n)*4))<<6)
624
625#define BO_COND_TRUE  BO(12)
626#define BO_COND_FALSE BO( 4)
627#define BO_ALWAYS     BO(20)
628
629enum {
630    CR_LT,
631    CR_GT,
632    CR_EQ,
633    CR_SO
634};
635
636static const uint32_t tcg_to_bc[] = {
637    [TCG_COND_EQ]  = BC | BI(7, CR_EQ) | BO_COND_TRUE,
638    [TCG_COND_NE]  = BC | BI(7, CR_EQ) | BO_COND_FALSE,
639    [TCG_COND_LT]  = BC | BI(7, CR_LT) | BO_COND_TRUE,
640    [TCG_COND_GE]  = BC | BI(7, CR_LT) | BO_COND_FALSE,
641    [TCG_COND_LE]  = BC | BI(7, CR_GT) | BO_COND_FALSE,
642    [TCG_COND_GT]  = BC | BI(7, CR_GT) | BO_COND_TRUE,
643    [TCG_COND_LTU] = BC | BI(7, CR_LT) | BO_COND_TRUE,
644    [TCG_COND_GEU] = BC | BI(7, CR_LT) | BO_COND_FALSE,
645    [TCG_COND_LEU] = BC | BI(7, CR_GT) | BO_COND_FALSE,
646    [TCG_COND_GTU] = BC | BI(7, CR_GT) | BO_COND_TRUE,
647};
648
649/* The low bit here is set if the RA and RB fields must be inverted.  */
650static const uint32_t tcg_to_isel[] = {
651    [TCG_COND_EQ]  = ISEL | BC_(7, CR_EQ),
652    [TCG_COND_NE]  = ISEL | BC_(7, CR_EQ) | 1,
653    [TCG_COND_LT]  = ISEL | BC_(7, CR_LT),
654    [TCG_COND_GE]  = ISEL | BC_(7, CR_LT) | 1,
655    [TCG_COND_LE]  = ISEL | BC_(7, CR_GT) | 1,
656    [TCG_COND_GT]  = ISEL | BC_(7, CR_GT),
657    [TCG_COND_LTU] = ISEL | BC_(7, CR_LT),
658    [TCG_COND_GEU] = ISEL | BC_(7, CR_LT) | 1,
659    [TCG_COND_LEU] = ISEL | BC_(7, CR_GT) | 1,
660    [TCG_COND_GTU] = ISEL | BC_(7, CR_GT),
661};
662
663static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
664                        intptr_t value, intptr_t addend)
665{
666    const tcg_insn_unit *target;
667    int16_t lo;
668    int32_t hi;
669
670    value += addend;
671    target = (const tcg_insn_unit *)value;
672
673    switch (type) {
674    case R_PPC_REL14:
675        return reloc_pc14(code_ptr, target);
676    case R_PPC_REL24:
677        return reloc_pc24(code_ptr, target);
678    case R_PPC_ADDR16:
679        /*
680         * We are (slightly) abusing this relocation type.  In particular,
681         * assert that the low 2 bits are zero, and do not modify them.
682         * That way we can use this with LD et al that have opcode bits
683         * in the low 2 bits of the insn.
684         */
685        if ((value & 3) || value != (int16_t)value) {
686            return false;
687        }
688        *code_ptr = (*code_ptr & ~0xfffc) | (value & 0xfffc);
689        break;
690    case R_PPC_ADDR32:
691        /*
692         * We are abusing this relocation type.  Again, this points to
693         * a pair of insns, lis + load.  This is an absolute address
694         * relocation for PPC32 so the lis cannot be removed.
695         */
696        lo = value;
697        hi = value - lo;
698        if (hi + lo != value) {
699            return false;
700        }
701        code_ptr[0] = deposit32(code_ptr[0], 0, 16, hi >> 16);
702        code_ptr[1] = deposit32(code_ptr[1], 0, 16, lo);
703        break;
704    default:
705        g_assert_not_reached();
706    }
707    return true;
708}
709
710static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
711                             TCGReg base, tcg_target_long offset);
712
713static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
714{
715    if (ret == arg) {
716        return true;
717    }
718    switch (type) {
719    case TCG_TYPE_I64:
720        tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
721        /* fallthru */
722    case TCG_TYPE_I32:
723        if (ret < TCG_REG_V0) {
724            if (arg < TCG_REG_V0) {
725                tcg_out32(s, OR | SAB(arg, ret, arg));
726                break;
727            } else if (have_isa_2_07) {
728                tcg_out32(s, (type == TCG_TYPE_I32 ? MFVSRWZ : MFVSRD)
729                          | VRT(arg) | RA(ret));
730                break;
731            } else {
732                /* Altivec does not support vector->integer moves.  */
733                return false;
734            }
735        } else if (arg < TCG_REG_V0) {
736            if (have_isa_2_07) {
737                tcg_out32(s, (type == TCG_TYPE_I32 ? MTVSRWZ : MTVSRD)
738                          | VRT(ret) | RA(arg));
739                break;
740            } else {
741                /* Altivec does not support integer->vector moves.  */
742                return false;
743            }
744        }
745        /* fallthru */
746    case TCG_TYPE_V64:
747    case TCG_TYPE_V128:
748        tcg_debug_assert(ret >= TCG_REG_V0 && arg >= TCG_REG_V0);
749        tcg_out32(s, VOR | VRT(ret) | VRA(arg) | VRB(arg));
750        break;
751    default:
752        g_assert_not_reached();
753    }
754    return true;
755}
756
757static inline void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs,
758                               int sh, int mb)
759{
760    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
761    sh = SH(sh & 0x1f) | (((sh >> 5) & 1) << 1);
762    mb = MB64((mb >> 5) | ((mb << 1) & 0x3f));
763    tcg_out32(s, op | RA(ra) | RS(rs) | sh | mb);
764}
765
766static inline void tcg_out_rlw(TCGContext *s, int op, TCGReg ra, TCGReg rs,
767                               int sh, int mb, int me)
768{
769    tcg_out32(s, op | RA(ra) | RS(rs) | SH(sh) | MB(mb) | ME(me));
770}
771
772static inline void tcg_out_ext8s(TCGContext *s, TCGReg dst, TCGReg src)
773{
774    tcg_out32(s, EXTSB | RA(dst) | RS(src));
775}
776
777static inline void tcg_out_ext16s(TCGContext *s, TCGReg dst, TCGReg src)
778{
779    tcg_out32(s, EXTSH | RA(dst) | RS(src));
780}
781
782static inline void tcg_out_ext16u(TCGContext *s, TCGReg dst, TCGReg src)
783{
784    tcg_out32(s, ANDI | SAI(src, dst, 0xffff));
785}
786
787static inline void tcg_out_ext32s(TCGContext *s, TCGReg dst, TCGReg src)
788{
789    tcg_out32(s, EXTSW | RA(dst) | RS(src));
790}
791
792static inline void tcg_out_ext32u(TCGContext *s, TCGReg dst, TCGReg src)
793{
794    tcg_out_rld(s, RLDICL, dst, src, 0, 32);
795}
796
797static inline void tcg_out_shli32(TCGContext *s, TCGReg dst, TCGReg src, int c)
798{
799    tcg_out_rlw(s, RLWINM, dst, src, c, 0, 31 - c);
800}
801
802static inline void tcg_out_shli64(TCGContext *s, TCGReg dst, TCGReg src, int c)
803{
804    tcg_out_rld(s, RLDICR, dst, src, c, 63 - c);
805}
806
807static inline void tcg_out_sari32(TCGContext *s, TCGReg dst, TCGReg src, int c)
808{
809    /* Limit immediate shift count lest we create an illegal insn.  */
810    tcg_out32(s, SRAWI | RA(dst) | RS(src) | SH(c & 31));
811}
812
813static inline void tcg_out_shri32(TCGContext *s, TCGReg dst, TCGReg src, int c)
814{
815    tcg_out_rlw(s, RLWINM, dst, src, 32 - c, c, 31);
816}
817
818static inline void tcg_out_shri64(TCGContext *s, TCGReg dst, TCGReg src, int c)
819{
820    tcg_out_rld(s, RLDICL, dst, src, 64 - c, c);
821}
822
823static inline void tcg_out_sari64(TCGContext *s, TCGReg dst, TCGReg src, int c)
824{
825    tcg_out32(s, SRADI | RA(dst) | RS(src) | SH(c & 0x1f) | ((c >> 4) & 2));
826}
827
828static void tcg_out_bswap16(TCGContext *s, TCGReg dst, TCGReg src, int flags)
829{
830    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
831
832    if (have_isa_3_10) {
833        tcg_out32(s, BRH | RA(dst) | RS(src));
834        if (flags & TCG_BSWAP_OS) {
835            tcg_out_ext16s(s, dst, dst);
836        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
837            tcg_out_ext16u(s, dst, dst);
838        }
839        return;
840    }
841
842    /*
843     * In the following,
844     *   dep(a, b, m) -> (a & ~m) | (b & m)
845     *
846     * Begin with:                              src = xxxxabcd
847     */
848    /* tmp = rol32(src, 24) & 0x000000ff            = 0000000c */
849    tcg_out_rlw(s, RLWINM, tmp, src, 24, 24, 31);
850    /* tmp = dep(tmp, rol32(src, 8), 0x0000ff00)    = 000000dc */
851    tcg_out_rlw(s, RLWIMI, tmp, src, 8, 16, 23);
852
853    if (flags & TCG_BSWAP_OS) {
854        tcg_out_ext16s(s, dst, tmp);
855    } else {
856        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
857    }
858}
859
860static void tcg_out_bswap32(TCGContext *s, TCGReg dst, TCGReg src, int flags)
861{
862    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
863
864    if (have_isa_3_10) {
865        tcg_out32(s, BRW | RA(dst) | RS(src));
866        if (flags & TCG_BSWAP_OS) {
867            tcg_out_ext32s(s, dst, dst);
868        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
869            tcg_out_ext32u(s, dst, dst);
870        }
871        return;
872    }
873
874    /*
875     * Stolen from gcc's builtin_bswap32.
876     * In the following,
877     *   dep(a, b, m) -> (a & ~m) | (b & m)
878     *
879     * Begin with:                              src = xxxxabcd
880     */
881    /* tmp = rol32(src, 8) & 0xffffffff             = 0000bcda */
882    tcg_out_rlw(s, RLWINM, tmp, src, 8, 0, 31);
883    /* tmp = dep(tmp, rol32(src, 24), 0xff000000)   = 0000dcda */
884    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 0, 7);
885    /* tmp = dep(tmp, rol32(src, 24), 0x0000ff00)   = 0000dcba */
886    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 16, 23);
887
888    if (flags & TCG_BSWAP_OS) {
889        tcg_out_ext32s(s, dst, tmp);
890    } else {
891        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
892    }
893}
894
895static void tcg_out_bswap64(TCGContext *s, TCGReg dst, TCGReg src)
896{
897    TCGReg t0 = dst == src ? TCG_REG_R0 : dst;
898    TCGReg t1 = dst == src ? dst : TCG_REG_R0;
899
900    if (have_isa_3_10) {
901        tcg_out32(s, BRD | RA(dst) | RS(src));
902        return;
903    }
904
905    /*
906     * In the following,
907     *   dep(a, b, m) -> (a & ~m) | (b & m)
908     *
909     * Begin with:                              src = abcdefgh
910     */
911    /* t0 = rol32(src, 8) & 0xffffffff              = 0000fghe */
912    tcg_out_rlw(s, RLWINM, t0, src, 8, 0, 31);
913    /* t0 = dep(t0, rol32(src, 24), 0xff000000)     = 0000hghe */
914    tcg_out_rlw(s, RLWIMI, t0, src, 24, 0, 7);
915    /* t0 = dep(t0, rol32(src, 24), 0x0000ff00)     = 0000hgfe */
916    tcg_out_rlw(s, RLWIMI, t0, src, 24, 16, 23);
917
918    /* t0 = rol64(t0, 32)                           = hgfe0000 */
919    tcg_out_rld(s, RLDICL, t0, t0, 32, 0);
920    /* t1 = rol64(src, 32)                          = efghabcd */
921    tcg_out_rld(s, RLDICL, t1, src, 32, 0);
922
923    /* t0 = dep(t0, rol32(t1, 24), 0xffffffff)      = hgfebcda */
924    tcg_out_rlw(s, RLWIMI, t0, t1, 8, 0, 31);
925    /* t0 = dep(t0, rol32(t1, 24), 0xff000000)      = hgfedcda */
926    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 0, 7);
927    /* t0 = dep(t0, rol32(t1, 24), 0x0000ff00)      = hgfedcba */
928    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 16, 23);
929
930    tcg_out_mov(s, TCG_TYPE_REG, dst, t0);
931}
932
933/* Emit a move into ret of arg, if it can be done in one insn.  */
934static bool tcg_out_movi_one(TCGContext *s, TCGReg ret, tcg_target_long arg)
935{
936    if (arg == (int16_t)arg) {
937        tcg_out32(s, ADDI | TAI(ret, 0, arg));
938        return true;
939    }
940    if (arg == (int32_t)arg && (arg & 0xffff) == 0) {
941        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
942        return true;
943    }
944    return false;
945}
946
947static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
948                             tcg_target_long arg, bool in_prologue)
949{
950    intptr_t tb_diff;
951    tcg_target_long tmp;
952    int shift;
953
954    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
955
956    if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
957        arg = (int32_t)arg;
958    }
959
960    /* Load 16-bit immediates with one insn.  */
961    if (tcg_out_movi_one(s, ret, arg)) {
962        return;
963    }
964
965    /* Load addresses within the TB with one insn.  */
966    tb_diff = tcg_tbrel_diff(s, (void *)arg);
967    if (!in_prologue && USE_REG_TB && tb_diff == (int16_t)tb_diff) {
968        tcg_out32(s, ADDI | TAI(ret, TCG_REG_TB, tb_diff));
969        return;
970    }
971
972    /* Load 32-bit immediates with two insns.  Note that we've already
973       eliminated bare ADDIS, so we know both insns are required.  */
974    if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) {
975        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
976        tcg_out32(s, ORI | SAI(ret, ret, arg));
977        return;
978    }
979    if (arg == (uint32_t)arg && !(arg & 0x8000)) {
980        tcg_out32(s, ADDI | TAI(ret, 0, arg));
981        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
982        return;
983    }
984
985    /* Load masked 16-bit value.  */
986    if (arg > 0 && (arg & 0x8000)) {
987        tmp = arg | 0x7fff;
988        if ((tmp & (tmp + 1)) == 0) {
989            int mb = clz64(tmp + 1) + 1;
990            tcg_out32(s, ADDI | TAI(ret, 0, arg));
991            tcg_out_rld(s, RLDICL, ret, ret, 0, mb);
992            return;
993        }
994    }
995
996    /* Load common masks with 2 insns.  */
997    shift = ctz64(arg);
998    tmp = arg >> shift;
999    if (tmp == (int16_t)tmp) {
1000        tcg_out32(s, ADDI | TAI(ret, 0, tmp));
1001        tcg_out_shli64(s, ret, ret, shift);
1002        return;
1003    }
1004    shift = clz64(arg);
1005    if (tcg_out_movi_one(s, ret, arg << shift)) {
1006        tcg_out_shri64(s, ret, ret, shift);
1007        return;
1008    }
1009
1010    /* Load addresses within 2GB of TB with 2 (or rarely 3) insns.  */
1011    if (!in_prologue && USE_REG_TB && tb_diff == (int32_t)tb_diff) {
1012        tcg_out_mem_long(s, ADDI, ADD, ret, TCG_REG_TB, tb_diff);
1013        return;
1014    }
1015
1016    /* Use the constant pool, if possible.  */
1017    if (!in_prologue && USE_REG_TB) {
1018        new_pool_label(s, arg, R_PPC_ADDR16, s->code_ptr,
1019                       tcg_tbrel_diff(s, NULL));
1020        tcg_out32(s, LD | TAI(ret, TCG_REG_TB, 0));
1021        return;
1022    }
1023
1024    tmp = arg >> 31 >> 1;
1025    tcg_out_movi(s, TCG_TYPE_I32, ret, tmp);
1026    if (tmp) {
1027        tcg_out_shli64(s, ret, ret, 32);
1028    }
1029    if (arg & 0xffff0000) {
1030        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
1031    }
1032    if (arg & 0xffff) {
1033        tcg_out32(s, ORI | SAI(ret, ret, arg));
1034    }
1035}
1036
1037static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
1038                             TCGReg ret, int64_t val)
1039{
1040    uint32_t load_insn;
1041    int rel, low;
1042    intptr_t add;
1043
1044    switch (vece) {
1045    case MO_8:
1046        low = (int8_t)val;
1047        if (low >= -16 && low < 16) {
1048            tcg_out32(s, VSPLTISB | VRT(ret) | ((val & 31) << 16));
1049            return;
1050        }
1051        if (have_isa_3_00) {
1052            tcg_out32(s, XXSPLTIB | VRT(ret) | ((val & 0xff) << 11));
1053            return;
1054        }
1055        break;
1056
1057    case MO_16:
1058        low = (int16_t)val;
1059        if (low >= -16 && low < 16) {
1060            tcg_out32(s, VSPLTISH | VRT(ret) | ((val & 31) << 16));
1061            return;
1062        }
1063        break;
1064
1065    case MO_32:
1066        low = (int32_t)val;
1067        if (low >= -16 && low < 16) {
1068            tcg_out32(s, VSPLTISW | VRT(ret) | ((val & 31) << 16));
1069            return;
1070        }
1071        break;
1072    }
1073
1074    /*
1075     * Otherwise we must load the value from the constant pool.
1076     */
1077    if (USE_REG_TB) {
1078        rel = R_PPC_ADDR16;
1079        add = tcg_tbrel_diff(s, NULL);
1080    } else {
1081        rel = R_PPC_ADDR32;
1082        add = 0;
1083    }
1084
1085    if (have_vsx) {
1086        load_insn = type == TCG_TYPE_V64 ? LXSDX : LXVDSX;
1087        load_insn |= VRT(ret) | RB(TCG_REG_TMP1);
1088        if (TCG_TARGET_REG_BITS == 64) {
1089            new_pool_label(s, val, rel, s->code_ptr, add);
1090        } else {
1091            new_pool_l2(s, rel, s->code_ptr, add, val >> 32, val);
1092        }
1093    } else {
1094        load_insn = LVX | VRT(ret) | RB(TCG_REG_TMP1);
1095        if (TCG_TARGET_REG_BITS == 64) {
1096            new_pool_l2(s, rel, s->code_ptr, add, val, val);
1097        } else {
1098            new_pool_l4(s, rel, s->code_ptr, add,
1099                        val >> 32, val, val >> 32, val);
1100        }
1101    }
1102
1103    if (USE_REG_TB) {
1104        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, 0, 0));
1105        load_insn |= RA(TCG_REG_TB);
1106    } else {
1107        tcg_out32(s, ADDIS | TAI(TCG_REG_TMP1, 0, 0));
1108        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, TCG_REG_TMP1, 0));
1109    }
1110    tcg_out32(s, load_insn);
1111}
1112
1113static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,
1114                         tcg_target_long arg)
1115{
1116    switch (type) {
1117    case TCG_TYPE_I32:
1118    case TCG_TYPE_I64:
1119        tcg_debug_assert(ret < TCG_REG_V0);
1120        tcg_out_movi_int(s, type, ret, arg, false);
1121        break;
1122
1123    default:
1124        g_assert_not_reached();
1125    }
1126}
1127
1128static bool mask_operand(uint32_t c, int *mb, int *me)
1129{
1130    uint32_t lsb, test;
1131
1132    /* Accept a bit pattern like:
1133           0....01....1
1134           1....10....0
1135           0..01..10..0
1136       Keep track of the transitions.  */
1137    if (c == 0 || c == -1) {
1138        return false;
1139    }
1140    test = c;
1141    lsb = test & -test;
1142    test += lsb;
1143    if (test & (test - 1)) {
1144        return false;
1145    }
1146
1147    *me = clz32(lsb);
1148    *mb = test ? clz32(test & -test) + 1 : 0;
1149    return true;
1150}
1151
1152static bool mask64_operand(uint64_t c, int *mb, int *me)
1153{
1154    uint64_t lsb;
1155
1156    if (c == 0) {
1157        return false;
1158    }
1159
1160    lsb = c & -c;
1161    /* Accept 1..10..0.  */
1162    if (c == -lsb) {
1163        *mb = 0;
1164        *me = clz64(lsb);
1165        return true;
1166    }
1167    /* Accept 0..01..1.  */
1168    if (lsb == 1 && (c & (c + 1)) == 0) {
1169        *mb = clz64(c + 1) + 1;
1170        *me = 63;
1171        return true;
1172    }
1173    return false;
1174}
1175
1176static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1177{
1178    int mb, me;
1179
1180    if (mask_operand(c, &mb, &me)) {
1181        tcg_out_rlw(s, RLWINM, dst, src, 0, mb, me);
1182    } else if ((c & 0xffff) == c) {
1183        tcg_out32(s, ANDI | SAI(src, dst, c));
1184        return;
1185    } else if ((c & 0xffff0000) == c) {
1186        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1187        return;
1188    } else {
1189        tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R0, c);
1190        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1191    }
1192}
1193
1194static void tcg_out_andi64(TCGContext *s, TCGReg dst, TCGReg src, uint64_t c)
1195{
1196    int mb, me;
1197
1198    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1199    if (mask64_operand(c, &mb, &me)) {
1200        if (mb == 0) {
1201            tcg_out_rld(s, RLDICR, dst, src, 0, me);
1202        } else {
1203            tcg_out_rld(s, RLDICL, dst, src, 0, mb);
1204        }
1205    } else if ((c & 0xffff) == c) {
1206        tcg_out32(s, ANDI | SAI(src, dst, c));
1207        return;
1208    } else if ((c & 0xffff0000) == c) {
1209        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1210        return;
1211    } else {
1212        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, c);
1213        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1214    }
1215}
1216
1217static void tcg_out_zori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c,
1218                           int op_lo, int op_hi)
1219{
1220    if (c >> 16) {
1221        tcg_out32(s, op_hi | SAI(src, dst, c >> 16));
1222        src = dst;
1223    }
1224    if (c & 0xffff) {
1225        tcg_out32(s, op_lo | SAI(src, dst, c));
1226        src = dst;
1227    }
1228}
1229
1230static void tcg_out_ori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1231{
1232    tcg_out_zori32(s, dst, src, c, ORI, ORIS);
1233}
1234
1235static void tcg_out_xori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1236{
1237    tcg_out_zori32(s, dst, src, c, XORI, XORIS);
1238}
1239
1240static void tcg_out_b(TCGContext *s, int mask, const tcg_insn_unit *target)
1241{
1242    ptrdiff_t disp = tcg_pcrel_diff(s, target);
1243    if (in_range_b(disp)) {
1244        tcg_out32(s, B | (disp & 0x3fffffc) | mask);
1245    } else {
1246        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, (uintptr_t)target);
1247        tcg_out32(s, MTSPR | RS(TCG_REG_R0) | CTR);
1248        tcg_out32(s, BCCTR | BO_ALWAYS | mask);
1249    }
1250}
1251
1252static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
1253                             TCGReg base, tcg_target_long offset)
1254{
1255    tcg_target_long orig = offset, l0, l1, extra = 0, align = 0;
1256    bool is_int_store = false;
1257    TCGReg rs = TCG_REG_TMP1;
1258
1259    switch (opi) {
1260    case LD: case LWA:
1261        align = 3;
1262        /* FALLTHRU */
1263    default:
1264        if (rt > TCG_REG_R0 && rt < TCG_REG_V0) {
1265            rs = rt;
1266            break;
1267        }
1268        break;
1269    case LXSD:
1270    case STXSD:
1271        align = 3;
1272        break;
1273    case LXV:
1274    case STXV:
1275        align = 15;
1276        break;
1277    case STD:
1278        align = 3;
1279        /* FALLTHRU */
1280    case STB: case STH: case STW:
1281        is_int_store = true;
1282        break;
1283    }
1284
1285    /* For unaligned, or very large offsets, use the indexed form.  */
1286    if (offset & align || offset != (int32_t)offset || opi == 0) {
1287        if (rs == base) {
1288            rs = TCG_REG_R0;
1289        }
1290        tcg_debug_assert(!is_int_store || rs != rt);
1291        tcg_out_movi(s, TCG_TYPE_PTR, rs, orig);
1292        tcg_out32(s, opx | TAB(rt & 31, base, rs));
1293        return;
1294    }
1295
1296    l0 = (int16_t)offset;
1297    offset = (offset - l0) >> 16;
1298    l1 = (int16_t)offset;
1299
1300    if (l1 < 0 && orig >= 0) {
1301        extra = 0x4000;
1302        l1 = (int16_t)(offset - 0x4000);
1303    }
1304    if (l1) {
1305        tcg_out32(s, ADDIS | TAI(rs, base, l1));
1306        base = rs;
1307    }
1308    if (extra) {
1309        tcg_out32(s, ADDIS | TAI(rs, base, extra));
1310        base = rs;
1311    }
1312    if (opi != ADDI || base != rt || l0 != 0) {
1313        tcg_out32(s, opi | TAI(rt & 31, base, l0));
1314    }
1315}
1316
1317static void tcg_out_vsldoi(TCGContext *s, TCGReg ret,
1318                           TCGReg va, TCGReg vb, int shb)
1319{
1320    tcg_out32(s, VSLDOI | VRT(ret) | VRA(va) | VRB(vb) | (shb << 6));
1321}
1322
1323static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
1324                       TCGReg base, intptr_t offset)
1325{
1326    int shift;
1327
1328    switch (type) {
1329    case TCG_TYPE_I32:
1330        if (ret < TCG_REG_V0) {
1331            tcg_out_mem_long(s, LWZ, LWZX, ret, base, offset);
1332            break;
1333        }
1334        if (have_isa_2_07 && have_vsx) {
1335            tcg_out_mem_long(s, 0, LXSIWZX, ret, base, offset);
1336            break;
1337        }
1338        tcg_debug_assert((offset & 3) == 0);
1339        tcg_out_mem_long(s, 0, LVEWX, ret, base, offset);
1340        shift = (offset - 4) & 0xc;
1341        if (shift) {
1342            tcg_out_vsldoi(s, ret, ret, ret, shift);
1343        }
1344        break;
1345    case TCG_TYPE_I64:
1346        if (ret < TCG_REG_V0) {
1347            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1348            tcg_out_mem_long(s, LD, LDX, ret, base, offset);
1349            break;
1350        }
1351        /* fallthru */
1352    case TCG_TYPE_V64:
1353        tcg_debug_assert(ret >= TCG_REG_V0);
1354        if (have_vsx) {
1355            tcg_out_mem_long(s, have_isa_3_00 ? LXSD : 0, LXSDX,
1356                             ret, base, offset);
1357            break;
1358        }
1359        tcg_debug_assert((offset & 7) == 0);
1360        tcg_out_mem_long(s, 0, LVX, ret, base, offset & -16);
1361        if (offset & 8) {
1362            tcg_out_vsldoi(s, ret, ret, ret, 8);
1363        }
1364        break;
1365    case TCG_TYPE_V128:
1366        tcg_debug_assert(ret >= TCG_REG_V0);
1367        tcg_debug_assert((offset & 15) == 0);
1368        tcg_out_mem_long(s, have_isa_3_00 ? LXV : 0,
1369                         LVX, ret, base, offset);
1370        break;
1371    default:
1372        g_assert_not_reached();
1373    }
1374}
1375
1376static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
1377                              TCGReg base, intptr_t offset)
1378{
1379    int shift;
1380
1381    switch (type) {
1382    case TCG_TYPE_I32:
1383        if (arg < TCG_REG_V0) {
1384            tcg_out_mem_long(s, STW, STWX, arg, base, offset);
1385            break;
1386        }
1387        if (have_isa_2_07 && have_vsx) {
1388            tcg_out_mem_long(s, 0, STXSIWX, arg, base, offset);
1389            break;
1390        }
1391        assert((offset & 3) == 0);
1392        tcg_debug_assert((offset & 3) == 0);
1393        shift = (offset - 4) & 0xc;
1394        if (shift) {
1395            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, shift);
1396            arg = TCG_VEC_TMP1;
1397        }
1398        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1399        break;
1400    case TCG_TYPE_I64:
1401        if (arg < TCG_REG_V0) {
1402            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1403            tcg_out_mem_long(s, STD, STDX, arg, base, offset);
1404            break;
1405        }
1406        /* fallthru */
1407    case TCG_TYPE_V64:
1408        tcg_debug_assert(arg >= TCG_REG_V0);
1409        if (have_vsx) {
1410            tcg_out_mem_long(s, have_isa_3_00 ? STXSD : 0,
1411                             STXSDX, arg, base, offset);
1412            break;
1413        }
1414        tcg_debug_assert((offset & 7) == 0);
1415        if (offset & 8) {
1416            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, 8);
1417            arg = TCG_VEC_TMP1;
1418        }
1419        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1420        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset + 4);
1421        break;
1422    case TCG_TYPE_V128:
1423        tcg_debug_assert(arg >= TCG_REG_V0);
1424        tcg_out_mem_long(s, have_isa_3_00 ? STXV : 0,
1425                         STVX, arg, base, offset);
1426        break;
1427    default:
1428        g_assert_not_reached();
1429    }
1430}
1431
1432static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1433                               TCGReg base, intptr_t ofs)
1434{
1435    return false;
1436}
1437
1438static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2,
1439                        int const_arg2, int cr, TCGType type)
1440{
1441    int imm;
1442    uint32_t op;
1443
1444    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1445
1446    /* Simplify the comparisons below wrt CMPI.  */
1447    if (type == TCG_TYPE_I32) {
1448        arg2 = (int32_t)arg2;
1449    }
1450
1451    switch (cond) {
1452    case TCG_COND_EQ:
1453    case TCG_COND_NE:
1454        if (const_arg2) {
1455            if ((int16_t) arg2 == arg2) {
1456                op = CMPI;
1457                imm = 1;
1458                break;
1459            } else if ((uint16_t) arg2 == arg2) {
1460                op = CMPLI;
1461                imm = 1;
1462                break;
1463            }
1464        }
1465        op = CMPL;
1466        imm = 0;
1467        break;
1468
1469    case TCG_COND_LT:
1470    case TCG_COND_GE:
1471    case TCG_COND_LE:
1472    case TCG_COND_GT:
1473        if (const_arg2) {
1474            if ((int16_t) arg2 == arg2) {
1475                op = CMPI;
1476                imm = 1;
1477                break;
1478            }
1479        }
1480        op = CMP;
1481        imm = 0;
1482        break;
1483
1484    case TCG_COND_LTU:
1485    case TCG_COND_GEU:
1486    case TCG_COND_LEU:
1487    case TCG_COND_GTU:
1488        if (const_arg2) {
1489            if ((uint16_t) arg2 == arg2) {
1490                op = CMPLI;
1491                imm = 1;
1492                break;
1493            }
1494        }
1495        op = CMPL;
1496        imm = 0;
1497        break;
1498
1499    default:
1500        tcg_abort();
1501    }
1502    op |= BF(cr) | ((type == TCG_TYPE_I64) << 21);
1503
1504    if (imm) {
1505        tcg_out32(s, op | RA(arg1) | (arg2 & 0xffff));
1506    } else {
1507        if (const_arg2) {
1508            tcg_out_movi(s, type, TCG_REG_R0, arg2);
1509            arg2 = TCG_REG_R0;
1510        }
1511        tcg_out32(s, op | RA(arg1) | RB(arg2));
1512    }
1513}
1514
1515static void tcg_out_setcond_eq0(TCGContext *s, TCGType type,
1516                                TCGReg dst, TCGReg src)
1517{
1518    if (type == TCG_TYPE_I32) {
1519        tcg_out32(s, CNTLZW | RS(src) | RA(dst));
1520        tcg_out_shri32(s, dst, dst, 5);
1521    } else {
1522        tcg_out32(s, CNTLZD | RS(src) | RA(dst));
1523        tcg_out_shri64(s, dst, dst, 6);
1524    }
1525}
1526
1527static void tcg_out_setcond_ne0(TCGContext *s, TCGReg dst, TCGReg src)
1528{
1529    /* X != 0 implies X + -1 generates a carry.  Extra addition
1530       trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C.  */
1531    if (dst != src) {
1532        tcg_out32(s, ADDIC | TAI(dst, src, -1));
1533        tcg_out32(s, SUBFE | TAB(dst, dst, src));
1534    } else {
1535        tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1));
1536        tcg_out32(s, SUBFE | TAB(dst, TCG_REG_R0, src));
1537    }
1538}
1539
1540static TCGReg tcg_gen_setcond_xor(TCGContext *s, TCGReg arg1, TCGArg arg2,
1541                                  bool const_arg2)
1542{
1543    if (const_arg2) {
1544        if ((uint32_t)arg2 == arg2) {
1545            tcg_out_xori32(s, TCG_REG_R0, arg1, arg2);
1546        } else {
1547            tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, arg2);
1548            tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, TCG_REG_R0));
1549        }
1550    } else {
1551        tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, arg2));
1552    }
1553    return TCG_REG_R0;
1554}
1555
1556static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
1557                            TCGArg arg0, TCGArg arg1, TCGArg arg2,
1558                            int const_arg2)
1559{
1560    int crop, sh;
1561
1562    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1563
1564    /* Ignore high bits of a potential constant arg2.  */
1565    if (type == TCG_TYPE_I32) {
1566        arg2 = (uint32_t)arg2;
1567    }
1568
1569    /* Handle common and trivial cases before handling anything else.  */
1570    if (arg2 == 0) {
1571        switch (cond) {
1572        case TCG_COND_EQ:
1573            tcg_out_setcond_eq0(s, type, arg0, arg1);
1574            return;
1575        case TCG_COND_NE:
1576            if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1577                tcg_out_ext32u(s, TCG_REG_R0, arg1);
1578                arg1 = TCG_REG_R0;
1579            }
1580            tcg_out_setcond_ne0(s, arg0, arg1);
1581            return;
1582        case TCG_COND_GE:
1583            tcg_out32(s, NOR | SAB(arg1, arg0, arg1));
1584            arg1 = arg0;
1585            /* FALLTHRU */
1586        case TCG_COND_LT:
1587            /* Extract the sign bit.  */
1588            if (type == TCG_TYPE_I32) {
1589                tcg_out_shri32(s, arg0, arg1, 31);
1590            } else {
1591                tcg_out_shri64(s, arg0, arg1, 63);
1592            }
1593            return;
1594        default:
1595            break;
1596        }
1597    }
1598
1599    /* If we have ISEL, we can implement everything with 3 or 4 insns.
1600       All other cases below are also at least 3 insns, so speed up the
1601       code generator by not considering them and always using ISEL.  */
1602    if (have_isel) {
1603        int isel, tab;
1604
1605        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1606
1607        isel = tcg_to_isel[cond];
1608
1609        tcg_out_movi(s, type, arg0, 1);
1610        if (isel & 1) {
1611            /* arg0 = (bc ? 0 : 1) */
1612            tab = TAB(arg0, 0, arg0);
1613            isel &= ~1;
1614        } else {
1615            /* arg0 = (bc ? 1 : 0) */
1616            tcg_out_movi(s, type, TCG_REG_R0, 0);
1617            tab = TAB(arg0, arg0, TCG_REG_R0);
1618        }
1619        tcg_out32(s, isel | tab);
1620        return;
1621    }
1622
1623    switch (cond) {
1624    case TCG_COND_EQ:
1625        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1626        tcg_out_setcond_eq0(s, type, arg0, arg1);
1627        return;
1628
1629    case TCG_COND_NE:
1630        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1631        /* Discard the high bits only once, rather than both inputs.  */
1632        if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1633            tcg_out_ext32u(s, TCG_REG_R0, arg1);
1634            arg1 = TCG_REG_R0;
1635        }
1636        tcg_out_setcond_ne0(s, arg0, arg1);
1637        return;
1638
1639    case TCG_COND_GT:
1640    case TCG_COND_GTU:
1641        sh = 30;
1642        crop = 0;
1643        goto crtest;
1644
1645    case TCG_COND_LT:
1646    case TCG_COND_LTU:
1647        sh = 29;
1648        crop = 0;
1649        goto crtest;
1650
1651    case TCG_COND_GE:
1652    case TCG_COND_GEU:
1653        sh = 31;
1654        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_LT) | BB(7, CR_LT);
1655        goto crtest;
1656
1657    case TCG_COND_LE:
1658    case TCG_COND_LEU:
1659        sh = 31;
1660        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_GT) | BB(7, CR_GT);
1661    crtest:
1662        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1663        if (crop) {
1664            tcg_out32(s, crop);
1665        }
1666        tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1667        tcg_out_rlw(s, RLWINM, arg0, TCG_REG_R0, sh, 31, 31);
1668        break;
1669
1670    default:
1671        tcg_abort();
1672    }
1673}
1674
1675static void tcg_out_bc(TCGContext *s, int bc, TCGLabel *l)
1676{
1677    if (l->has_value) {
1678        bc |= reloc_pc14_val(tcg_splitwx_to_rx(s->code_ptr), l->u.value_ptr);
1679    } else {
1680        tcg_out_reloc(s, s->code_ptr, R_PPC_REL14, l, 0);
1681    }
1682    tcg_out32(s, bc);
1683}
1684
1685static void tcg_out_brcond(TCGContext *s, TCGCond cond,
1686                           TCGArg arg1, TCGArg arg2, int const_arg2,
1687                           TCGLabel *l, TCGType type)
1688{
1689    tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1690    tcg_out_bc(s, tcg_to_bc[cond], l);
1691}
1692
1693static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
1694                            TCGArg dest, TCGArg c1, TCGArg c2, TCGArg v1,
1695                            TCGArg v2, bool const_c2)
1696{
1697    /* If for some reason both inputs are zero, don't produce bad code.  */
1698    if (v1 == 0 && v2 == 0) {
1699        tcg_out_movi(s, type, dest, 0);
1700        return;
1701    }
1702
1703    tcg_out_cmp(s, cond, c1, c2, const_c2, 7, type);
1704
1705    if (have_isel) {
1706        int isel = tcg_to_isel[cond];
1707
1708        /* Swap the V operands if the operation indicates inversion.  */
1709        if (isel & 1) {
1710            int t = v1;
1711            v1 = v2;
1712            v2 = t;
1713            isel &= ~1;
1714        }
1715        /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand.  */
1716        if (v2 == 0) {
1717            tcg_out_movi(s, type, TCG_REG_R0, 0);
1718        }
1719        tcg_out32(s, isel | TAB(dest, v1, v2));
1720    } else {
1721        if (dest == v2) {
1722            cond = tcg_invert_cond(cond);
1723            v2 = v1;
1724        } else if (dest != v1) {
1725            if (v1 == 0) {
1726                tcg_out_movi(s, type, dest, 0);
1727            } else {
1728                tcg_out_mov(s, type, dest, v1);
1729            }
1730        }
1731        /* Branch forward over one insn */
1732        tcg_out32(s, tcg_to_bc[cond] | 8);
1733        if (v2 == 0) {
1734            tcg_out_movi(s, type, dest, 0);
1735        } else {
1736            tcg_out_mov(s, type, dest, v2);
1737        }
1738    }
1739}
1740
1741static void tcg_out_cntxz(TCGContext *s, TCGType type, uint32_t opc,
1742                          TCGArg a0, TCGArg a1, TCGArg a2, bool const_a2)
1743{
1744    if (const_a2 && a2 == (type == TCG_TYPE_I32 ? 32 : 64)) {
1745        tcg_out32(s, opc | RA(a0) | RS(a1));
1746    } else {
1747        tcg_out_cmp(s, TCG_COND_EQ, a1, 0, 1, 7, type);
1748        /* Note that the only other valid constant for a2 is 0.  */
1749        if (have_isel) {
1750            tcg_out32(s, opc | RA(TCG_REG_R0) | RS(a1));
1751            tcg_out32(s, tcg_to_isel[TCG_COND_EQ] | TAB(a0, a2, TCG_REG_R0));
1752        } else if (!const_a2 && a0 == a2) {
1753            tcg_out32(s, tcg_to_bc[TCG_COND_EQ] | 8);
1754            tcg_out32(s, opc | RA(a0) | RS(a1));
1755        } else {
1756            tcg_out32(s, opc | RA(a0) | RS(a1));
1757            tcg_out32(s, tcg_to_bc[TCG_COND_NE] | 8);
1758            if (const_a2) {
1759                tcg_out_movi(s, type, a0, 0);
1760            } else {
1761                tcg_out_mov(s, type, a0, a2);
1762            }
1763        }
1764    }
1765}
1766
1767static void tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1768                         const int *const_args)
1769{
1770    static const struct { uint8_t bit1, bit2; } bits[] = {
1771        [TCG_COND_LT ] = { CR_LT, CR_LT },
1772        [TCG_COND_LE ] = { CR_LT, CR_GT },
1773        [TCG_COND_GT ] = { CR_GT, CR_GT },
1774        [TCG_COND_GE ] = { CR_GT, CR_LT },
1775        [TCG_COND_LTU] = { CR_LT, CR_LT },
1776        [TCG_COND_LEU] = { CR_LT, CR_GT },
1777        [TCG_COND_GTU] = { CR_GT, CR_GT },
1778        [TCG_COND_GEU] = { CR_GT, CR_LT },
1779    };
1780
1781    TCGCond cond = args[4], cond2;
1782    TCGArg al, ah, bl, bh;
1783    int blconst, bhconst;
1784    int op, bit1, bit2;
1785
1786    al = args[0];
1787    ah = args[1];
1788    bl = args[2];
1789    bh = args[3];
1790    blconst = const_args[2];
1791    bhconst = const_args[3];
1792
1793    switch (cond) {
1794    case TCG_COND_EQ:
1795        op = CRAND;
1796        goto do_equality;
1797    case TCG_COND_NE:
1798        op = CRNAND;
1799    do_equality:
1800        tcg_out_cmp(s, cond, al, bl, blconst, 6, TCG_TYPE_I32);
1801        tcg_out_cmp(s, cond, ah, bh, bhconst, 7, TCG_TYPE_I32);
1802        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1803        break;
1804
1805    case TCG_COND_LT:
1806    case TCG_COND_LE:
1807    case TCG_COND_GT:
1808    case TCG_COND_GE:
1809    case TCG_COND_LTU:
1810    case TCG_COND_LEU:
1811    case TCG_COND_GTU:
1812    case TCG_COND_GEU:
1813        bit1 = bits[cond].bit1;
1814        bit2 = bits[cond].bit2;
1815        op = (bit1 != bit2 ? CRANDC : CRAND);
1816        cond2 = tcg_unsigned_cond(cond);
1817
1818        tcg_out_cmp(s, cond, ah, bh, bhconst, 6, TCG_TYPE_I32);
1819        tcg_out_cmp(s, cond2, al, bl, blconst, 7, TCG_TYPE_I32);
1820        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, bit2));
1821        tcg_out32(s, CROR | BT(7, CR_EQ) | BA(6, bit1) | BB(7, CR_EQ));
1822        break;
1823
1824    default:
1825        tcg_abort();
1826    }
1827}
1828
1829static void tcg_out_setcond2(TCGContext *s, const TCGArg *args,
1830                             const int *const_args)
1831{
1832    tcg_out_cmp2(s, args + 1, const_args + 1);
1833    tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1834    tcg_out_rlw(s, RLWINM, args[0], TCG_REG_R0, 31, 31, 31);
1835}
1836
1837static void tcg_out_brcond2 (TCGContext *s, const TCGArg *args,
1838                             const int *const_args)
1839{
1840    tcg_out_cmp2(s, args, const_args);
1841    tcg_out_bc(s, BC | BI(7, CR_EQ) | BO_COND_TRUE, arg_label(args[5]));
1842}
1843
1844static void tcg_out_mb(TCGContext *s, TCGArg a0)
1845{
1846    uint32_t insn;
1847
1848    if (a0 & TCG_MO_ST_LD) {
1849        insn = HWSYNC;
1850    } else {
1851        insn = LWSYNC;
1852    }
1853
1854    tcg_out32(s, insn);
1855}
1856
1857static inline uint64_t make_pair(tcg_insn_unit i1, tcg_insn_unit i2)
1858{
1859    if (HOST_BIG_ENDIAN) {
1860        return (uint64_t)i1 << 32 | i2;
1861    }
1862    return (uint64_t)i2 << 32 | i1;
1863}
1864
1865static inline void ppc64_replace2(uintptr_t rx, uintptr_t rw,
1866                                  tcg_insn_unit i0, tcg_insn_unit i1)
1867{
1868#if TCG_TARGET_REG_BITS == 64
1869    qatomic_set((uint64_t *)rw, make_pair(i0, i1));
1870    flush_idcache_range(rx, rw, 8);
1871#else
1872    qemu_build_not_reached();
1873#endif
1874}
1875
1876static inline void ppc64_replace4(uintptr_t rx, uintptr_t rw,
1877                                  tcg_insn_unit i0, tcg_insn_unit i1,
1878                                  tcg_insn_unit i2, tcg_insn_unit i3)
1879{
1880    uint64_t p[2];
1881
1882    p[!HOST_BIG_ENDIAN] = make_pair(i0, i1);
1883    p[HOST_BIG_ENDIAN] = make_pair(i2, i3);
1884
1885    /*
1886     * There's no convenient way to get the compiler to allocate a pair
1887     * of registers at an even index, so copy into r6/r7 and clobber.
1888     */
1889    asm("mr  %%r6, %1\n\t"
1890        "mr  %%r7, %2\n\t"
1891        "stq %%r6, %0"
1892        : "=Q"(*(__int128 *)rw) : "r"(p[0]), "r"(p[1]) : "r6", "r7");
1893    flush_idcache_range(rx, rw, 16);
1894}
1895
1896void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_rx,
1897                              uintptr_t jmp_rw, uintptr_t addr)
1898{
1899    tcg_insn_unit i0, i1, i2, i3;
1900    intptr_t tb_diff = addr - tc_ptr;
1901    intptr_t br_diff = addr - (jmp_rx + 4);
1902    intptr_t lo, hi;
1903
1904    if (TCG_TARGET_REG_BITS == 32) {
1905        intptr_t diff = addr - jmp_rx;
1906        tcg_debug_assert(in_range_b(diff));
1907        qatomic_set((uint32_t *)jmp_rw, B | (diff & 0x3fffffc));
1908        flush_idcache_range(jmp_rx, jmp_rw, 4);
1909        return;
1910    }
1911
1912    /*
1913     * For 16-bit displacements, we can use a single add + branch.
1914     * This happens quite often.
1915     */
1916    if (tb_diff == (int16_t)tb_diff) {
1917        i0 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, tb_diff);
1918        i1 = B | (br_diff & 0x3fffffc);
1919        ppc64_replace2(jmp_rx, jmp_rw, i0, i1);
1920        return;
1921    }
1922
1923    lo = (int16_t)tb_diff;
1924    hi = (int32_t)(tb_diff - lo);
1925    assert(tb_diff == hi + lo);
1926    i0 = ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, hi >> 16);
1927    i1 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, lo);
1928
1929    /*
1930     * Without stq from 2.07, we can only update two insns,
1931     * and those must be the ones that load the target address.
1932     */
1933    if (!have_isa_2_07) {
1934        ppc64_replace2(jmp_rx, jmp_rw, i0, i1);
1935        return;
1936    }
1937
1938    /*
1939     * For 26-bit displacements, we can use a direct branch.
1940     * Otherwise we still need the indirect branch, which we
1941     * must restore after a potential direct branch write.
1942     */
1943    br_diff -= 4;
1944    if (in_range_b(br_diff)) {
1945        i2 = B | (br_diff & 0x3fffffc);
1946        i3 = NOP;
1947    } else {
1948        i2 = MTSPR | RS(TCG_REG_TB) | CTR;
1949        i3 = BCCTR | BO_ALWAYS;
1950    }
1951    ppc64_replace4(jmp_rx, jmp_rw, i0, i1, i2, i3);
1952}
1953
1954static void tcg_out_call_int(TCGContext *s, int lk,
1955                             const tcg_insn_unit *target)
1956{
1957#ifdef _CALL_AIX
1958    /* Look through the descriptor.  If the branch is in range, and we
1959       don't have to spend too much effort on building the toc.  */
1960    const void *tgt = ((const void * const *)target)[0];
1961    uintptr_t toc = ((const uintptr_t *)target)[1];
1962    intptr_t diff = tcg_pcrel_diff(s, tgt);
1963
1964    if (in_range_b(diff) && toc == (uint32_t)toc) {
1965        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, toc);
1966        tcg_out_b(s, lk, tgt);
1967    } else {
1968        /* Fold the low bits of the constant into the addresses below.  */
1969        intptr_t arg = (intptr_t)target;
1970        int ofs = (int16_t)arg;
1971
1972        if (ofs + 8 < 0x8000) {
1973            arg -= ofs;
1974        } else {
1975            ofs = 0;
1976        }
1977        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, arg);
1978        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_TMP1, ofs);
1979        tcg_out32(s, MTSPR | RA(TCG_REG_R0) | CTR);
1980        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_REG_TMP1, ofs + SZP);
1981        tcg_out32(s, BCCTR | BO_ALWAYS | lk);
1982    }
1983#elif defined(_CALL_ELF) && _CALL_ELF == 2
1984    intptr_t diff;
1985
1986    /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1987       address, which the callee uses to compute its TOC address.  */
1988    /* FIXME: when the branch is in range, we could avoid r12 load if we
1989       knew that the destination uses the same TOC, and what its local
1990       entry point offset is.  */
1991    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R12, (intptr_t)target);
1992
1993    diff = tcg_pcrel_diff(s, target);
1994    if (in_range_b(diff)) {
1995        tcg_out_b(s, lk, target);
1996    } else {
1997        tcg_out32(s, MTSPR | RS(TCG_REG_R12) | CTR);
1998        tcg_out32(s, BCCTR | BO_ALWAYS | lk);
1999    }
2000#else
2001    tcg_out_b(s, lk, target);
2002#endif
2003}
2004
2005static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target,
2006                         const TCGHelperInfo *info)
2007{
2008    tcg_out_call_int(s, LK, target);
2009}
2010
2011static const uint32_t qemu_ldx_opc[(MO_SSIZE + MO_BSWAP) + 1] = {
2012    [MO_UB] = LBZX,
2013    [MO_UW] = LHZX,
2014    [MO_UL] = LWZX,
2015    [MO_UQ] = LDX,
2016    [MO_SW] = LHAX,
2017    [MO_SL] = LWAX,
2018    [MO_BSWAP | MO_UB] = LBZX,
2019    [MO_BSWAP | MO_UW] = LHBRX,
2020    [MO_BSWAP | MO_UL] = LWBRX,
2021    [MO_BSWAP | MO_UQ] = LDBRX,
2022};
2023
2024static const uint32_t qemu_stx_opc[(MO_SIZE + MO_BSWAP) + 1] = {
2025    [MO_UB] = STBX,
2026    [MO_UW] = STHX,
2027    [MO_UL] = STWX,
2028    [MO_UQ] = STDX,
2029    [MO_BSWAP | MO_UB] = STBX,
2030    [MO_BSWAP | MO_UW] = STHBRX,
2031    [MO_BSWAP | MO_UL] = STWBRX,
2032    [MO_BSWAP | MO_UQ] = STDBRX,
2033};
2034
2035static const uint32_t qemu_exts_opc[4] = {
2036    EXTSB, EXTSH, EXTSW, 0
2037};
2038
2039#if defined (CONFIG_SOFTMMU)
2040/* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
2041 *                                 int mmu_idx, uintptr_t ra)
2042 */
2043static void * const qemu_ld_helpers[(MO_SIZE | MO_BSWAP) + 1] = {
2044    [MO_UB]   = helper_ret_ldub_mmu,
2045    [MO_LEUW] = helper_le_lduw_mmu,
2046    [MO_LEUL] = helper_le_ldul_mmu,
2047    [MO_LEUQ] = helper_le_ldq_mmu,
2048    [MO_BEUW] = helper_be_lduw_mmu,
2049    [MO_BEUL] = helper_be_ldul_mmu,
2050    [MO_BEUQ] = helper_be_ldq_mmu,
2051};
2052
2053/* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
2054 *                                 uintxx_t val, int mmu_idx, uintptr_t ra)
2055 */
2056static void * const qemu_st_helpers[(MO_SIZE | MO_BSWAP) + 1] = {
2057    [MO_UB]   = helper_ret_stb_mmu,
2058    [MO_LEUW] = helper_le_stw_mmu,
2059    [MO_LEUL] = helper_le_stl_mmu,
2060    [MO_LEUQ] = helper_le_stq_mmu,
2061    [MO_BEUW] = helper_be_stw_mmu,
2062    [MO_BEUL] = helper_be_stl_mmu,
2063    [MO_BEUQ] = helper_be_stq_mmu,
2064};
2065
2066/* We expect to use a 16-bit negative offset from ENV.  */
2067QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
2068QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -32768);
2069
2070/* Perform the TLB load and compare.  Places the result of the comparison
2071   in CR7, loads the addend of the TLB into R3, and returns the register
2072   containing the guest address (zero-extended into R4).  Clobbers R0 and R2. */
2073
2074static TCGReg tcg_out_tlb_read(TCGContext *s, MemOp opc,
2075                               TCGReg addrlo, TCGReg addrhi,
2076                               int mem_index, bool is_read)
2077{
2078    int cmp_off
2079        = (is_read
2080           ? offsetof(CPUTLBEntry, addr_read)
2081           : offsetof(CPUTLBEntry, addr_write));
2082    int fast_off = TLB_MASK_TABLE_OFS(mem_index);
2083    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
2084    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
2085    unsigned s_bits = opc & MO_SIZE;
2086    unsigned a_bits = get_alignment_bits(opc);
2087
2088    /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx].  */
2089    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_AREG0, mask_off);
2090    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R4, TCG_AREG0, table_off);
2091
2092    /* Extract the page index, shifted into place for tlb index.  */
2093    if (TCG_TARGET_REG_BITS == 32) {
2094        tcg_out_shri32(s, TCG_REG_TMP1, addrlo,
2095                       TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
2096    } else {
2097        tcg_out_shri64(s, TCG_REG_TMP1, addrlo,
2098                       TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
2099    }
2100    tcg_out32(s, AND | SAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_TMP1));
2101
2102    /* Load the TLB comparator.  */
2103    if (cmp_off == 0 && TCG_TARGET_REG_BITS >= TARGET_LONG_BITS) {
2104        uint32_t lxu = (TCG_TARGET_REG_BITS == 32 || TARGET_LONG_BITS == 32
2105                        ? LWZUX : LDUX);
2106        tcg_out32(s, lxu | TAB(TCG_REG_TMP1, TCG_REG_R3, TCG_REG_R4));
2107    } else {
2108        tcg_out32(s, ADD | TAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_R4));
2109        if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2110            tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP1, TCG_REG_R3, cmp_off + 4);
2111            tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R4, TCG_REG_R3, cmp_off);
2112        } else {
2113            tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP1, TCG_REG_R3, cmp_off);
2114        }
2115    }
2116
2117    /* Load the TLB addend for use on the fast path.  Do this asap
2118       to minimize any load use delay.  */
2119    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_REG_R3,
2120               offsetof(CPUTLBEntry, addend));
2121
2122    /* Clear the non-page, non-alignment bits from the address */
2123    if (TCG_TARGET_REG_BITS == 32) {
2124        /* We don't support unaligned accesses on 32-bits.
2125         * Preserve the bottom bits and thus trigger a comparison
2126         * failure on unaligned accesses.
2127         */
2128        if (a_bits < s_bits) {
2129            a_bits = s_bits;
2130        }
2131        tcg_out_rlw(s, RLWINM, TCG_REG_R0, addrlo, 0,
2132                    (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
2133    } else {
2134        TCGReg t = addrlo;
2135
2136        /* If the access is unaligned, we need to make sure we fail if we
2137         * cross a page boundary.  The trick is to add the access size-1
2138         * to the address before masking the low bits.  That will make the
2139         * address overflow to the next page if we cross a page boundary,
2140         * which will then force a mismatch of the TLB compare.
2141         */
2142        if (a_bits < s_bits) {
2143            unsigned a_mask = (1 << a_bits) - 1;
2144            unsigned s_mask = (1 << s_bits) - 1;
2145            tcg_out32(s, ADDI | TAI(TCG_REG_R0, t, s_mask - a_mask));
2146            t = TCG_REG_R0;
2147        }
2148
2149        /* Mask the address for the requested alignment.  */
2150        if (TARGET_LONG_BITS == 32) {
2151            tcg_out_rlw(s, RLWINM, TCG_REG_R0, t, 0,
2152                        (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
2153            /* Zero-extend the address for use in the final address.  */
2154            tcg_out_ext32u(s, TCG_REG_R4, addrlo);
2155            addrlo = TCG_REG_R4;
2156        } else if (a_bits == 0) {
2157            tcg_out_rld(s, RLDICR, TCG_REG_R0, t, 0, 63 - TARGET_PAGE_BITS);
2158        } else {
2159            tcg_out_rld(s, RLDICL, TCG_REG_R0, t,
2160                        64 - TARGET_PAGE_BITS, TARGET_PAGE_BITS - a_bits);
2161            tcg_out_rld(s, RLDICL, TCG_REG_R0, TCG_REG_R0, TARGET_PAGE_BITS, 0);
2162        }
2163    }
2164
2165    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2166        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
2167                    0, 7, TCG_TYPE_I32);
2168        tcg_out_cmp(s, TCG_COND_EQ, addrhi, TCG_REG_R4, 0, 6, TCG_TYPE_I32);
2169        tcg_out32(s, CRAND | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
2170    } else {
2171        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
2172                    0, 7, TCG_TYPE_TL);
2173    }
2174
2175    return addrlo;
2176}
2177
2178/* Record the context of a call to the out of line helper code for the slow
2179   path for a load or store, so that we can later generate the correct
2180   helper code.  */
2181static void add_qemu_ldst_label(TCGContext *s, bool is_ld, MemOpIdx oi,
2182                                TCGReg datalo_reg, TCGReg datahi_reg,
2183                                TCGReg addrlo_reg, TCGReg addrhi_reg,
2184                                tcg_insn_unit *raddr, tcg_insn_unit *lptr)
2185{
2186    TCGLabelQemuLdst *label = new_ldst_label(s);
2187
2188    label->is_ld = is_ld;
2189    label->oi = oi;
2190    label->datalo_reg = datalo_reg;
2191    label->datahi_reg = datahi_reg;
2192    label->addrlo_reg = addrlo_reg;
2193    label->addrhi_reg = addrhi_reg;
2194    label->raddr = tcg_splitwx_to_rx(raddr);
2195    label->label_ptr[0] = lptr;
2196}
2197
2198static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2199{
2200    MemOpIdx oi = lb->oi;
2201    MemOp opc = get_memop(oi);
2202    TCGReg hi, lo, arg = TCG_REG_R3;
2203
2204    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2205        return false;
2206    }
2207
2208    tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
2209
2210    lo = lb->addrlo_reg;
2211    hi = lb->addrhi_reg;
2212    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2213        arg |= (TCG_TARGET_CALL_ARG_I64 == TCG_CALL_ARG_EVEN);
2214        tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2215        tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2216    } else {
2217        /* If the address needed to be zero-extended, we'll have already
2218           placed it in R4.  The only remaining case is 64-bit guest.  */
2219        tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
2220    }
2221
2222    tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
2223    tcg_out32(s, MFSPR | RT(arg) | LR);
2224
2225    tcg_out_call_int(s, LK, qemu_ld_helpers[opc & (MO_BSWAP | MO_SIZE)]);
2226
2227    lo = lb->datalo_reg;
2228    hi = lb->datahi_reg;
2229    if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
2230        tcg_out_mov(s, TCG_TYPE_I32, lo, TCG_REG_R4);
2231        tcg_out_mov(s, TCG_TYPE_I32, hi, TCG_REG_R3);
2232    } else if (opc & MO_SIGN) {
2233        uint32_t insn = qemu_exts_opc[opc & MO_SIZE];
2234        tcg_out32(s, insn | RA(lo) | RS(TCG_REG_R3));
2235    } else {
2236        tcg_out_mov(s, TCG_TYPE_REG, lo, TCG_REG_R3);
2237    }
2238
2239    tcg_out_b(s, 0, lb->raddr);
2240    return true;
2241}
2242
2243static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2244{
2245    MemOpIdx oi = lb->oi;
2246    MemOp opc = get_memop(oi);
2247    MemOp s_bits = opc & MO_SIZE;
2248    TCGReg hi, lo, arg = TCG_REG_R3;
2249
2250    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2251        return false;
2252    }
2253
2254    tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
2255
2256    lo = lb->addrlo_reg;
2257    hi = lb->addrhi_reg;
2258    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2259        arg |= (TCG_TARGET_CALL_ARG_I64 == TCG_CALL_ARG_EVEN);
2260        tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2261        tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2262    } else {
2263        /* If the address needed to be zero-extended, we'll have already
2264           placed it in R4.  The only remaining case is 64-bit guest.  */
2265        tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
2266    }
2267
2268    lo = lb->datalo_reg;
2269    hi = lb->datahi_reg;
2270    if (TCG_TARGET_REG_BITS == 32) {
2271        switch (s_bits) {
2272        case MO_64:
2273            arg |= (TCG_TARGET_CALL_ARG_I64 == TCG_CALL_ARG_EVEN);
2274            tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2275            /* FALLTHRU */
2276        case MO_32:
2277            tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2278            break;
2279        default:
2280            tcg_out_rlw(s, RLWINM, arg++, lo, 0, 32 - (8 << s_bits), 31);
2281            break;
2282        }
2283    } else {
2284        if (s_bits == MO_64) {
2285            tcg_out_mov(s, TCG_TYPE_I64, arg++, lo);
2286        } else {
2287            tcg_out_rld(s, RLDICL, arg++, lo, 0, 64 - (8 << s_bits));
2288        }
2289    }
2290
2291    tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
2292    tcg_out32(s, MFSPR | RT(arg) | LR);
2293
2294    tcg_out_call_int(s, LK, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
2295
2296    tcg_out_b(s, 0, lb->raddr);
2297    return true;
2298}
2299#else
2300
2301static void tcg_out_test_alignment(TCGContext *s, bool is_ld, TCGReg addrlo,
2302                                   TCGReg addrhi, unsigned a_bits)
2303{
2304    unsigned a_mask = (1 << a_bits) - 1;
2305    TCGLabelQemuLdst *label = new_ldst_label(s);
2306
2307    label->is_ld = is_ld;
2308    label->addrlo_reg = addrlo;
2309    label->addrhi_reg = addrhi;
2310
2311    /* We are expecting a_bits to max out at 7, much lower than ANDI. */
2312    tcg_debug_assert(a_bits < 16);
2313    tcg_out32(s, ANDI | SAI(addrlo, TCG_REG_R0, a_mask));
2314
2315    label->label_ptr[0] = s->code_ptr;
2316    tcg_out32(s, BC | BI(0, CR_EQ) | BO_COND_FALSE | LK);
2317
2318    label->raddr = tcg_splitwx_to_rx(s->code_ptr);
2319}
2320
2321static bool tcg_out_fail_alignment(TCGContext *s, TCGLabelQemuLdst *l)
2322{
2323    if (!reloc_pc14(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2324        return false;
2325    }
2326
2327    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2328        TCGReg arg = TCG_REG_R4;
2329
2330        arg |= (TCG_TARGET_CALL_ARG_I64 == TCG_CALL_ARG_EVEN);
2331        if (l->addrlo_reg != arg) {
2332            tcg_out_mov(s, TCG_TYPE_I32, arg, l->addrhi_reg);
2333            tcg_out_mov(s, TCG_TYPE_I32, arg + 1, l->addrlo_reg);
2334        } else if (l->addrhi_reg != arg + 1) {
2335            tcg_out_mov(s, TCG_TYPE_I32, arg + 1, l->addrlo_reg);
2336            tcg_out_mov(s, TCG_TYPE_I32, arg, l->addrhi_reg);
2337        } else {
2338            tcg_out_mov(s, TCG_TYPE_I32, TCG_REG_R0, arg);
2339            tcg_out_mov(s, TCG_TYPE_I32, arg, arg + 1);
2340            tcg_out_mov(s, TCG_TYPE_I32, arg + 1, TCG_REG_R0);
2341        }
2342    } else {
2343        tcg_out_mov(s, TCG_TYPE_TL, TCG_REG_R4, l->addrlo_reg);
2344    }
2345    tcg_out_mov(s, TCG_TYPE_TL, TCG_REG_R3, TCG_AREG0);
2346
2347    /* "Tail call" to the helper, with the return address back inline. */
2348    tcg_out_call_int(s, 0, (const void *)(l->is_ld ? helper_unaligned_ld
2349                                          : helper_unaligned_st));
2350    return true;
2351}
2352
2353static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
2354{
2355    return tcg_out_fail_alignment(s, l);
2356}
2357
2358static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
2359{
2360    return tcg_out_fail_alignment(s, l);
2361}
2362
2363#endif /* SOFTMMU */
2364
2365static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64)
2366{
2367    TCGReg datalo, datahi, addrlo, rbase;
2368    TCGReg addrhi __attribute__((unused));
2369    MemOpIdx oi;
2370    MemOp opc, s_bits;
2371#ifdef CONFIG_SOFTMMU
2372    int mem_index;
2373    tcg_insn_unit *label_ptr;
2374#else
2375    unsigned a_bits;
2376#endif
2377
2378    datalo = *args++;
2379    datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
2380    addrlo = *args++;
2381    addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
2382    oi = *args++;
2383    opc = get_memop(oi);
2384    s_bits = opc & MO_SIZE;
2385
2386#ifdef CONFIG_SOFTMMU
2387    mem_index = get_mmuidx(oi);
2388    addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, true);
2389
2390    /* Load a pointer into the current opcode w/conditional branch-link. */
2391    label_ptr = s->code_ptr;
2392    tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
2393
2394    rbase = TCG_REG_R3;
2395#else  /* !CONFIG_SOFTMMU */
2396    a_bits = get_alignment_bits(opc);
2397    if (a_bits) {
2398        tcg_out_test_alignment(s, true, addrlo, addrhi, a_bits);
2399    }
2400    rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
2401    if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
2402        tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
2403        addrlo = TCG_REG_TMP1;
2404    }
2405#endif
2406
2407    if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
2408        if (opc & MO_BSWAP) {
2409            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2410            tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
2411            tcg_out32(s, LWBRX | TAB(datahi, rbase, TCG_REG_R0));
2412        } else if (rbase != 0) {
2413            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2414            tcg_out32(s, LWZX | TAB(datahi, rbase, addrlo));
2415            tcg_out32(s, LWZX | TAB(datalo, rbase, TCG_REG_R0));
2416        } else if (addrlo == datahi) {
2417            tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
2418            tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
2419        } else {
2420            tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
2421            tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
2422        }
2423    } else {
2424        uint32_t insn = qemu_ldx_opc[opc & (MO_BSWAP | MO_SSIZE)];
2425        if (!have_isa_2_06 && insn == LDBRX) {
2426            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2427            tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
2428            tcg_out32(s, LWBRX | TAB(TCG_REG_R0, rbase, TCG_REG_R0));
2429            tcg_out_rld(s, RLDIMI, datalo, TCG_REG_R0, 32, 0);
2430        } else if (insn) {
2431            tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
2432        } else {
2433            insn = qemu_ldx_opc[opc & (MO_SIZE | MO_BSWAP)];
2434            tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
2435            insn = qemu_exts_opc[s_bits];
2436            tcg_out32(s, insn | RA(datalo) | RS(datalo));
2437        }
2438    }
2439
2440#ifdef CONFIG_SOFTMMU
2441    add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
2442                        s->code_ptr, label_ptr);
2443#endif
2444}
2445
2446static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
2447{
2448    TCGReg datalo, datahi, addrlo, rbase;
2449    TCGReg addrhi __attribute__((unused));
2450    MemOpIdx oi;
2451    MemOp opc, s_bits;
2452#ifdef CONFIG_SOFTMMU
2453    int mem_index;
2454    tcg_insn_unit *label_ptr;
2455#else
2456    unsigned a_bits;
2457#endif
2458
2459    datalo = *args++;
2460    datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
2461    addrlo = *args++;
2462    addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
2463    oi = *args++;
2464    opc = get_memop(oi);
2465    s_bits = opc & MO_SIZE;
2466
2467#ifdef CONFIG_SOFTMMU
2468    mem_index = get_mmuidx(oi);
2469    addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, false);
2470
2471    /* Load a pointer into the current opcode w/conditional branch-link. */
2472    label_ptr = s->code_ptr;
2473    tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
2474
2475    rbase = TCG_REG_R3;
2476#else  /* !CONFIG_SOFTMMU */
2477    a_bits = get_alignment_bits(opc);
2478    if (a_bits) {
2479        tcg_out_test_alignment(s, false, addrlo, addrhi, a_bits);
2480    }
2481    rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
2482    if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
2483        tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
2484        addrlo = TCG_REG_TMP1;
2485    }
2486#endif
2487
2488    if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
2489        if (opc & MO_BSWAP) {
2490            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2491            tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
2492            tcg_out32(s, STWBRX | SAB(datahi, rbase, TCG_REG_R0));
2493        } else if (rbase != 0) {
2494            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2495            tcg_out32(s, STWX | SAB(datahi, rbase, addrlo));
2496            tcg_out32(s, STWX | SAB(datalo, rbase, TCG_REG_R0));
2497        } else {
2498            tcg_out32(s, STW | TAI(datahi, addrlo, 0));
2499            tcg_out32(s, STW | TAI(datalo, addrlo, 4));
2500        }
2501    } else {
2502        uint32_t insn = qemu_stx_opc[opc & (MO_BSWAP | MO_SIZE)];
2503        if (!have_isa_2_06 && insn == STDBRX) {
2504            tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
2505            tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, addrlo, 4));
2506            tcg_out_shri64(s, TCG_REG_R0, datalo, 32);
2507            tcg_out32(s, STWBRX | SAB(TCG_REG_R0, rbase, TCG_REG_TMP1));
2508        } else {
2509            tcg_out32(s, insn | SAB(datalo, rbase, addrlo));
2510        }
2511    }
2512
2513#ifdef CONFIG_SOFTMMU
2514    add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
2515                        s->code_ptr, label_ptr);
2516#endif
2517}
2518
2519static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
2520{
2521    int i;
2522    for (i = 0; i < count; ++i) {
2523        p[i] = NOP;
2524    }
2525}
2526
2527/* Parameters for function call generation, used in tcg.c.  */
2528#define TCG_TARGET_STACK_ALIGN       16
2529
2530#ifdef _CALL_AIX
2531# define LINK_AREA_SIZE                (6 * SZR)
2532# define LR_OFFSET                     (1 * SZR)
2533# define TCG_TARGET_CALL_STACK_OFFSET  (LINK_AREA_SIZE + 8 * SZR)
2534#elif defined(_CALL_DARWIN)
2535# define LINK_AREA_SIZE                (6 * SZR)
2536# define LR_OFFSET                     (2 * SZR)
2537#elif TCG_TARGET_REG_BITS == 64
2538# if defined(_CALL_ELF) && _CALL_ELF == 2
2539#  define LINK_AREA_SIZE               (4 * SZR)
2540#  define LR_OFFSET                    (1 * SZR)
2541# endif
2542#else /* TCG_TARGET_REG_BITS == 32 */
2543# if defined(_CALL_SYSV)
2544#  define LINK_AREA_SIZE               (2 * SZR)
2545#  define LR_OFFSET                    (1 * SZR)
2546# endif
2547#endif
2548#ifndef LR_OFFSET
2549# error "Unhandled abi"
2550#endif
2551#ifndef TCG_TARGET_CALL_STACK_OFFSET
2552# define TCG_TARGET_CALL_STACK_OFFSET  LINK_AREA_SIZE
2553#endif
2554
2555#define CPU_TEMP_BUF_SIZE  (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
2556#define REG_SAVE_SIZE      ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
2557
2558#define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET   \
2559                     + TCG_STATIC_CALL_ARGS_SIZE    \
2560                     + CPU_TEMP_BUF_SIZE            \
2561                     + REG_SAVE_SIZE                \
2562                     + TCG_TARGET_STACK_ALIGN - 1)  \
2563                    & -TCG_TARGET_STACK_ALIGN)
2564
2565#define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
2566
2567static void tcg_target_qemu_prologue(TCGContext *s)
2568{
2569    int i;
2570
2571#ifdef _CALL_AIX
2572    const void **desc = (const void **)s->code_ptr;
2573    desc[0] = tcg_splitwx_to_rx(desc + 2);  /* entry point */
2574    desc[1] = 0;                            /* environment pointer */
2575    s->code_ptr = (void *)(desc + 2);       /* skip over descriptor */
2576#endif
2577
2578    tcg_set_frame(s, TCG_REG_CALL_STACK, REG_SAVE_BOT - CPU_TEMP_BUF_SIZE,
2579                  CPU_TEMP_BUF_SIZE);
2580
2581    /* Prologue */
2582    tcg_out32(s, MFSPR | RT(TCG_REG_R0) | LR);
2583    tcg_out32(s, (SZR == 8 ? STDU : STWU)
2584              | SAI(TCG_REG_R1, TCG_REG_R1, -FRAME_SIZE));
2585
2586    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2587        tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2588                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2589    }
2590    tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2591
2592#ifndef CONFIG_SOFTMMU
2593    if (guest_base) {
2594        tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true);
2595        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2596    }
2597#endif
2598
2599    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2600    tcg_out32(s, MTSPR | RS(tcg_target_call_iarg_regs[1]) | CTR);
2601    if (USE_REG_TB) {
2602        tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, tcg_target_call_iarg_regs[1]);
2603    }
2604    tcg_out32(s, BCCTR | BO_ALWAYS);
2605
2606    /* Epilogue */
2607    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
2608
2609    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2610    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2611        tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2612                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2613    }
2614    tcg_out32(s, MTSPR | RS(TCG_REG_R0) | LR);
2615    tcg_out32(s, ADDI | TAI(TCG_REG_R1, TCG_REG_R1, FRAME_SIZE));
2616    tcg_out32(s, BCLR | BO_ALWAYS);
2617}
2618
2619static void tcg_out_op(TCGContext *s, TCGOpcode opc,
2620                       const TCGArg args[TCG_MAX_OP_ARGS],
2621                       const int const_args[TCG_MAX_OP_ARGS])
2622{
2623    TCGArg a0, a1, a2;
2624
2625    switch (opc) {
2626    case INDEX_op_exit_tb:
2627        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, args[0]);
2628        tcg_out_b(s, 0, tcg_code_gen_epilogue);
2629        break;
2630    case INDEX_op_goto_tb:
2631        if (s->tb_jmp_insn_offset) {
2632            /* Direct jump. */
2633            if (TCG_TARGET_REG_BITS == 64) {
2634                /* Ensure the next insns are 8 or 16-byte aligned. */
2635                while ((uintptr_t)s->code_ptr & (have_isa_2_07 ? 15 : 7)) {
2636                    tcg_out32(s, NOP);
2637                }
2638                s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2639                tcg_out32(s, ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2640                tcg_out32(s, ADDI | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2641            } else {
2642                s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2643                tcg_out32(s, B);
2644                s->tb_jmp_reset_offset[args[0]] = tcg_current_code_size(s);
2645                break;
2646            }
2647        } else {
2648            /* Indirect jump. */
2649            tcg_debug_assert(s->tb_jmp_insn_offset == NULL);
2650            tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, 0,
2651                       (intptr_t)(s->tb_jmp_insn_offset + args[0]));
2652        }
2653        tcg_out32(s, MTSPR | RS(TCG_REG_TB) | CTR);
2654        tcg_out32(s, BCCTR | BO_ALWAYS);
2655        set_jmp_reset_offset(s, args[0]);
2656        if (USE_REG_TB) {
2657            /* For the unlinked case, need to reset TCG_REG_TB.  */
2658            tcg_out_mem_long(s, ADDI, ADD, TCG_REG_TB, TCG_REG_TB,
2659                             -tcg_current_code_size(s));
2660        }
2661        break;
2662    case INDEX_op_goto_ptr:
2663        tcg_out32(s, MTSPR | RS(args[0]) | CTR);
2664        if (USE_REG_TB) {
2665            tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, args[0]);
2666        }
2667        tcg_out32(s, ADDI | TAI(TCG_REG_R3, 0, 0));
2668        tcg_out32(s, BCCTR | BO_ALWAYS);
2669        break;
2670    case INDEX_op_br:
2671        {
2672            TCGLabel *l = arg_label(args[0]);
2673            uint32_t insn = B;
2674
2675            if (l->has_value) {
2676                insn |= reloc_pc24_val(tcg_splitwx_to_rx(s->code_ptr),
2677                                       l->u.value_ptr);
2678            } else {
2679                tcg_out_reloc(s, s->code_ptr, R_PPC_REL24, l, 0);
2680            }
2681            tcg_out32(s, insn);
2682        }
2683        break;
2684    case INDEX_op_ld8u_i32:
2685    case INDEX_op_ld8u_i64:
2686        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2687        break;
2688    case INDEX_op_ld8s_i32:
2689    case INDEX_op_ld8s_i64:
2690        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2691        tcg_out_ext8s(s, args[0], args[0]);
2692        break;
2693    case INDEX_op_ld16u_i32:
2694    case INDEX_op_ld16u_i64:
2695        tcg_out_mem_long(s, LHZ, LHZX, args[0], args[1], args[2]);
2696        break;
2697    case INDEX_op_ld16s_i32:
2698    case INDEX_op_ld16s_i64:
2699        tcg_out_mem_long(s, LHA, LHAX, args[0], args[1], args[2]);
2700        break;
2701    case INDEX_op_ld_i32:
2702    case INDEX_op_ld32u_i64:
2703        tcg_out_mem_long(s, LWZ, LWZX, args[0], args[1], args[2]);
2704        break;
2705    case INDEX_op_ld32s_i64:
2706        tcg_out_mem_long(s, LWA, LWAX, args[0], args[1], args[2]);
2707        break;
2708    case INDEX_op_ld_i64:
2709        tcg_out_mem_long(s, LD, LDX, args[0], args[1], args[2]);
2710        break;
2711    case INDEX_op_st8_i32:
2712    case INDEX_op_st8_i64:
2713        tcg_out_mem_long(s, STB, STBX, args[0], args[1], args[2]);
2714        break;
2715    case INDEX_op_st16_i32:
2716    case INDEX_op_st16_i64:
2717        tcg_out_mem_long(s, STH, STHX, args[0], args[1], args[2]);
2718        break;
2719    case INDEX_op_st_i32:
2720    case INDEX_op_st32_i64:
2721        tcg_out_mem_long(s, STW, STWX, args[0], args[1], args[2]);
2722        break;
2723    case INDEX_op_st_i64:
2724        tcg_out_mem_long(s, STD, STDX, args[0], args[1], args[2]);
2725        break;
2726
2727    case INDEX_op_add_i32:
2728        a0 = args[0], a1 = args[1], a2 = args[2];
2729        if (const_args[2]) {
2730        do_addi_32:
2731            tcg_out_mem_long(s, ADDI, ADD, a0, a1, (int32_t)a2);
2732        } else {
2733            tcg_out32(s, ADD | TAB(a0, a1, a2));
2734        }
2735        break;
2736    case INDEX_op_sub_i32:
2737        a0 = args[0], a1 = args[1], a2 = args[2];
2738        if (const_args[1]) {
2739            if (const_args[2]) {
2740                tcg_out_movi(s, TCG_TYPE_I32, a0, a1 - a2);
2741            } else {
2742                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2743            }
2744        } else if (const_args[2]) {
2745            a2 = -a2;
2746            goto do_addi_32;
2747        } else {
2748            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2749        }
2750        break;
2751
2752    case INDEX_op_and_i32:
2753        a0 = args[0], a1 = args[1], a2 = args[2];
2754        if (const_args[2]) {
2755            tcg_out_andi32(s, a0, a1, a2);
2756        } else {
2757            tcg_out32(s, AND | SAB(a1, a0, a2));
2758        }
2759        break;
2760    case INDEX_op_and_i64:
2761        a0 = args[0], a1 = args[1], a2 = args[2];
2762        if (const_args[2]) {
2763            tcg_out_andi64(s, a0, a1, a2);
2764        } else {
2765            tcg_out32(s, AND | SAB(a1, a0, a2));
2766        }
2767        break;
2768    case INDEX_op_or_i64:
2769    case INDEX_op_or_i32:
2770        a0 = args[0], a1 = args[1], a2 = args[2];
2771        if (const_args[2]) {
2772            tcg_out_ori32(s, a0, a1, a2);
2773        } else {
2774            tcg_out32(s, OR | SAB(a1, a0, a2));
2775        }
2776        break;
2777    case INDEX_op_xor_i64:
2778    case INDEX_op_xor_i32:
2779        a0 = args[0], a1 = args[1], a2 = args[2];
2780        if (const_args[2]) {
2781            tcg_out_xori32(s, a0, a1, a2);
2782        } else {
2783            tcg_out32(s, XOR | SAB(a1, a0, a2));
2784        }
2785        break;
2786    case INDEX_op_andc_i32:
2787        a0 = args[0], a1 = args[1], a2 = args[2];
2788        if (const_args[2]) {
2789            tcg_out_andi32(s, a0, a1, ~a2);
2790        } else {
2791            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2792        }
2793        break;
2794    case INDEX_op_andc_i64:
2795        a0 = args[0], a1 = args[1], a2 = args[2];
2796        if (const_args[2]) {
2797            tcg_out_andi64(s, a0, a1, ~a2);
2798        } else {
2799            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2800        }
2801        break;
2802    case INDEX_op_orc_i32:
2803        if (const_args[2]) {
2804            tcg_out_ori32(s, args[0], args[1], ~args[2]);
2805            break;
2806        }
2807        /* FALLTHRU */
2808    case INDEX_op_orc_i64:
2809        tcg_out32(s, ORC | SAB(args[1], args[0], args[2]));
2810        break;
2811    case INDEX_op_eqv_i32:
2812        if (const_args[2]) {
2813            tcg_out_xori32(s, args[0], args[1], ~args[2]);
2814            break;
2815        }
2816        /* FALLTHRU */
2817    case INDEX_op_eqv_i64:
2818        tcg_out32(s, EQV | SAB(args[1], args[0], args[2]));
2819        break;
2820    case INDEX_op_nand_i32:
2821    case INDEX_op_nand_i64:
2822        tcg_out32(s, NAND | SAB(args[1], args[0], args[2]));
2823        break;
2824    case INDEX_op_nor_i32:
2825    case INDEX_op_nor_i64:
2826        tcg_out32(s, NOR | SAB(args[1], args[0], args[2]));
2827        break;
2828
2829    case INDEX_op_clz_i32:
2830        tcg_out_cntxz(s, TCG_TYPE_I32, CNTLZW, args[0], args[1],
2831                      args[2], const_args[2]);
2832        break;
2833    case INDEX_op_ctz_i32:
2834        tcg_out_cntxz(s, TCG_TYPE_I32, CNTTZW, args[0], args[1],
2835                      args[2], const_args[2]);
2836        break;
2837    case INDEX_op_ctpop_i32:
2838        tcg_out32(s, CNTPOPW | SAB(args[1], args[0], 0));
2839        break;
2840
2841    case INDEX_op_clz_i64:
2842        tcg_out_cntxz(s, TCG_TYPE_I64, CNTLZD, args[0], args[1],
2843                      args[2], const_args[2]);
2844        break;
2845    case INDEX_op_ctz_i64:
2846        tcg_out_cntxz(s, TCG_TYPE_I64, CNTTZD, args[0], args[1],
2847                      args[2], const_args[2]);
2848        break;
2849    case INDEX_op_ctpop_i64:
2850        tcg_out32(s, CNTPOPD | SAB(args[1], args[0], 0));
2851        break;
2852
2853    case INDEX_op_mul_i32:
2854        a0 = args[0], a1 = args[1], a2 = args[2];
2855        if (const_args[2]) {
2856            tcg_out32(s, MULLI | TAI(a0, a1, a2));
2857        } else {
2858            tcg_out32(s, MULLW | TAB(a0, a1, a2));
2859        }
2860        break;
2861
2862    case INDEX_op_div_i32:
2863        tcg_out32(s, DIVW | TAB(args[0], args[1], args[2]));
2864        break;
2865
2866    case INDEX_op_divu_i32:
2867        tcg_out32(s, DIVWU | TAB(args[0], args[1], args[2]));
2868        break;
2869
2870    case INDEX_op_rem_i32:
2871        tcg_out32(s, MODSW | TAB(args[0], args[1], args[2]));
2872        break;
2873
2874    case INDEX_op_remu_i32:
2875        tcg_out32(s, MODUW | TAB(args[0], args[1], args[2]));
2876        break;
2877
2878    case INDEX_op_shl_i32:
2879        if (const_args[2]) {
2880            /* Limit immediate shift count lest we create an illegal insn.  */
2881            tcg_out_shli32(s, args[0], args[1], args[2] & 31);
2882        } else {
2883            tcg_out32(s, SLW | SAB(args[1], args[0], args[2]));
2884        }
2885        break;
2886    case INDEX_op_shr_i32:
2887        if (const_args[2]) {
2888            /* Limit immediate shift count lest we create an illegal insn.  */
2889            tcg_out_shri32(s, args[0], args[1], args[2] & 31);
2890        } else {
2891            tcg_out32(s, SRW | SAB(args[1], args[0], args[2]));
2892        }
2893        break;
2894    case INDEX_op_sar_i32:
2895        if (const_args[2]) {
2896            tcg_out_sari32(s, args[0], args[1], args[2]);
2897        } else {
2898            tcg_out32(s, SRAW | SAB(args[1], args[0], args[2]));
2899        }
2900        break;
2901    case INDEX_op_rotl_i32:
2902        if (const_args[2]) {
2903            tcg_out_rlw(s, RLWINM, args[0], args[1], args[2], 0, 31);
2904        } else {
2905            tcg_out32(s, RLWNM | SAB(args[1], args[0], args[2])
2906                         | MB(0) | ME(31));
2907        }
2908        break;
2909    case INDEX_op_rotr_i32:
2910        if (const_args[2]) {
2911            tcg_out_rlw(s, RLWINM, args[0], args[1], 32 - args[2], 0, 31);
2912        } else {
2913            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 32));
2914            tcg_out32(s, RLWNM | SAB(args[1], args[0], TCG_REG_R0)
2915                         | MB(0) | ME(31));
2916        }
2917        break;
2918
2919    case INDEX_op_brcond_i32:
2920        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2921                       arg_label(args[3]), TCG_TYPE_I32);
2922        break;
2923    case INDEX_op_brcond_i64:
2924        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2925                       arg_label(args[3]), TCG_TYPE_I64);
2926        break;
2927    case INDEX_op_brcond2_i32:
2928        tcg_out_brcond2(s, args, const_args);
2929        break;
2930
2931    case INDEX_op_neg_i32:
2932    case INDEX_op_neg_i64:
2933        tcg_out32(s, NEG | RT(args[0]) | RA(args[1]));
2934        break;
2935
2936    case INDEX_op_not_i32:
2937    case INDEX_op_not_i64:
2938        tcg_out32(s, NOR | SAB(args[1], args[0], args[1]));
2939        break;
2940
2941    case INDEX_op_add_i64:
2942        a0 = args[0], a1 = args[1], a2 = args[2];
2943        if (const_args[2]) {
2944        do_addi_64:
2945            tcg_out_mem_long(s, ADDI, ADD, a0, a1, a2);
2946        } else {
2947            tcg_out32(s, ADD | TAB(a0, a1, a2));
2948        }
2949        break;
2950    case INDEX_op_sub_i64:
2951        a0 = args[0], a1 = args[1], a2 = args[2];
2952        if (const_args[1]) {
2953            if (const_args[2]) {
2954                tcg_out_movi(s, TCG_TYPE_I64, a0, a1 - a2);
2955            } else {
2956                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2957            }
2958        } else if (const_args[2]) {
2959            a2 = -a2;
2960            goto do_addi_64;
2961        } else {
2962            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2963        }
2964        break;
2965
2966    case INDEX_op_shl_i64:
2967        if (const_args[2]) {
2968            /* Limit immediate shift count lest we create an illegal insn.  */
2969            tcg_out_shli64(s, args[0], args[1], args[2] & 63);
2970        } else {
2971            tcg_out32(s, SLD | SAB(args[1], args[0], args[2]));
2972        }
2973        break;
2974    case INDEX_op_shr_i64:
2975        if (const_args[2]) {
2976            /* Limit immediate shift count lest we create an illegal insn.  */
2977            tcg_out_shri64(s, args[0], args[1], args[2] & 63);
2978        } else {
2979            tcg_out32(s, SRD | SAB(args[1], args[0], args[2]));
2980        }
2981        break;
2982    case INDEX_op_sar_i64:
2983        if (const_args[2]) {
2984            tcg_out_sari64(s, args[0], args[1], args[2]);
2985        } else {
2986            tcg_out32(s, SRAD | SAB(args[1], args[0], args[2]));
2987        }
2988        break;
2989    case INDEX_op_rotl_i64:
2990        if (const_args[2]) {
2991            tcg_out_rld(s, RLDICL, args[0], args[1], args[2], 0);
2992        } else {
2993            tcg_out32(s, RLDCL | SAB(args[1], args[0], args[2]) | MB64(0));
2994        }
2995        break;
2996    case INDEX_op_rotr_i64:
2997        if (const_args[2]) {
2998            tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 0);
2999        } else {
3000            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 64));
3001            tcg_out32(s, RLDCL | SAB(args[1], args[0], TCG_REG_R0) | MB64(0));
3002        }
3003        break;
3004
3005    case INDEX_op_mul_i64:
3006        a0 = args[0], a1 = args[1], a2 = args[2];
3007        if (const_args[2]) {
3008            tcg_out32(s, MULLI | TAI(a0, a1, a2));
3009        } else {
3010            tcg_out32(s, MULLD | TAB(a0, a1, a2));
3011        }
3012        break;
3013    case INDEX_op_div_i64:
3014        tcg_out32(s, DIVD | TAB(args[0], args[1], args[2]));
3015        break;
3016    case INDEX_op_divu_i64:
3017        tcg_out32(s, DIVDU | TAB(args[0], args[1], args[2]));
3018        break;
3019    case INDEX_op_rem_i64:
3020        tcg_out32(s, MODSD | TAB(args[0], args[1], args[2]));
3021        break;
3022    case INDEX_op_remu_i64:
3023        tcg_out32(s, MODUD | TAB(args[0], args[1], args[2]));
3024        break;
3025
3026    case INDEX_op_qemu_ld_i32:
3027        tcg_out_qemu_ld(s, args, false);
3028        break;
3029    case INDEX_op_qemu_ld_i64:
3030        tcg_out_qemu_ld(s, args, true);
3031        break;
3032    case INDEX_op_qemu_st_i32:
3033        tcg_out_qemu_st(s, args, false);
3034        break;
3035    case INDEX_op_qemu_st_i64:
3036        tcg_out_qemu_st(s, args, true);
3037        break;
3038
3039    case INDEX_op_ext8s_i32:
3040    case INDEX_op_ext8s_i64:
3041        tcg_out_ext8s(s, args[0], args[1]);
3042        break;
3043    case INDEX_op_ext16s_i32:
3044    case INDEX_op_ext16s_i64:
3045        tcg_out_ext16s(s, args[0], args[1]);
3046        break;
3047    case INDEX_op_ext_i32_i64:
3048    case INDEX_op_ext32s_i64:
3049        tcg_out_ext32s(s, args[0], args[1]);
3050        break;
3051    case INDEX_op_extu_i32_i64:
3052        tcg_out_ext32u(s, args[0], args[1]);
3053        break;
3054
3055    case INDEX_op_setcond_i32:
3056        tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2],
3057                        const_args[2]);
3058        break;
3059    case INDEX_op_setcond_i64:
3060        tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2],
3061                        const_args[2]);
3062        break;
3063    case INDEX_op_setcond2_i32:
3064        tcg_out_setcond2(s, args, const_args);
3065        break;
3066
3067    case INDEX_op_bswap16_i32:
3068    case INDEX_op_bswap16_i64:
3069        tcg_out_bswap16(s, args[0], args[1], args[2]);
3070        break;
3071    case INDEX_op_bswap32_i32:
3072        tcg_out_bswap32(s, args[0], args[1], 0);
3073        break;
3074    case INDEX_op_bswap32_i64:
3075        tcg_out_bswap32(s, args[0], args[1], args[2]);
3076        break;
3077    case INDEX_op_bswap64_i64:
3078        tcg_out_bswap64(s, args[0], args[1]);
3079        break;
3080
3081    case INDEX_op_deposit_i32:
3082        if (const_args[2]) {
3083            uint32_t mask = ((2u << (args[4] - 1)) - 1) << args[3];
3084            tcg_out_andi32(s, args[0], args[0], ~mask);
3085        } else {
3086            tcg_out_rlw(s, RLWIMI, args[0], args[2], args[3],
3087                        32 - args[3] - args[4], 31 - args[3]);
3088        }
3089        break;
3090    case INDEX_op_deposit_i64:
3091        if (const_args[2]) {
3092            uint64_t mask = ((2ull << (args[4] - 1)) - 1) << args[3];
3093            tcg_out_andi64(s, args[0], args[0], ~mask);
3094        } else {
3095            tcg_out_rld(s, RLDIMI, args[0], args[2], args[3],
3096                        64 - args[3] - args[4]);
3097        }
3098        break;
3099
3100    case INDEX_op_extract_i32:
3101        tcg_out_rlw(s, RLWINM, args[0], args[1],
3102                    32 - args[2], 32 - args[3], 31);
3103        break;
3104    case INDEX_op_extract_i64:
3105        tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 64 - args[3]);
3106        break;
3107
3108    case INDEX_op_movcond_i32:
3109        tcg_out_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], args[2],
3110                        args[3], args[4], const_args[2]);
3111        break;
3112    case INDEX_op_movcond_i64:
3113        tcg_out_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], args[2],
3114                        args[3], args[4], const_args[2]);
3115        break;
3116
3117#if TCG_TARGET_REG_BITS == 64
3118    case INDEX_op_add2_i64:
3119#else
3120    case INDEX_op_add2_i32:
3121#endif
3122        /* Note that the CA bit is defined based on the word size of the
3123           environment.  So in 64-bit mode it's always carry-out of bit 63.
3124           The fallback code using deposit works just as well for 32-bit.  */
3125        a0 = args[0], a1 = args[1];
3126        if (a0 == args[3] || (!const_args[5] && a0 == args[5])) {
3127            a0 = TCG_REG_R0;
3128        }
3129        if (const_args[4]) {
3130            tcg_out32(s, ADDIC | TAI(a0, args[2], args[4]));
3131        } else {
3132            tcg_out32(s, ADDC | TAB(a0, args[2], args[4]));
3133        }
3134        if (const_args[5]) {
3135            tcg_out32(s, (args[5] ? ADDME : ADDZE) | RT(a1) | RA(args[3]));
3136        } else {
3137            tcg_out32(s, ADDE | TAB(a1, args[3], args[5]));
3138        }
3139        if (a0 != args[0]) {
3140            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
3141        }
3142        break;
3143
3144#if TCG_TARGET_REG_BITS == 64
3145    case INDEX_op_sub2_i64:
3146#else
3147    case INDEX_op_sub2_i32:
3148#endif
3149        a0 = args[0], a1 = args[1];
3150        if (a0 == args[5] || (!const_args[3] && a0 == args[3])) {
3151            a0 = TCG_REG_R0;
3152        }
3153        if (const_args[2]) {
3154            tcg_out32(s, SUBFIC | TAI(a0, args[4], args[2]));
3155        } else {
3156            tcg_out32(s, SUBFC | TAB(a0, args[4], args[2]));
3157        }
3158        if (const_args[3]) {
3159            tcg_out32(s, (args[3] ? SUBFME : SUBFZE) | RT(a1) | RA(args[5]));
3160        } else {
3161            tcg_out32(s, SUBFE | TAB(a1, args[5], args[3]));
3162        }
3163        if (a0 != args[0]) {
3164            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
3165        }
3166        break;
3167
3168    case INDEX_op_muluh_i32:
3169        tcg_out32(s, MULHWU | TAB(args[0], args[1], args[2]));
3170        break;
3171    case INDEX_op_mulsh_i32:
3172        tcg_out32(s, MULHW | TAB(args[0], args[1], args[2]));
3173        break;
3174    case INDEX_op_muluh_i64:
3175        tcg_out32(s, MULHDU | TAB(args[0], args[1], args[2]));
3176        break;
3177    case INDEX_op_mulsh_i64:
3178        tcg_out32(s, MULHD | TAB(args[0], args[1], args[2]));
3179        break;
3180
3181    case INDEX_op_mb:
3182        tcg_out_mb(s, args[0]);
3183        break;
3184
3185    case INDEX_op_mov_i32:   /* Always emitted via tcg_out_mov.  */
3186    case INDEX_op_mov_i64:
3187    case INDEX_op_call:      /* Always emitted via tcg_out_call.  */
3188    default:
3189        tcg_abort();
3190    }
3191}
3192
3193int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3194{
3195    switch (opc) {
3196    case INDEX_op_and_vec:
3197    case INDEX_op_or_vec:
3198    case INDEX_op_xor_vec:
3199    case INDEX_op_andc_vec:
3200    case INDEX_op_not_vec:
3201    case INDEX_op_nor_vec:
3202    case INDEX_op_eqv_vec:
3203    case INDEX_op_nand_vec:
3204        return 1;
3205    case INDEX_op_orc_vec:
3206        return have_isa_2_07;
3207    case INDEX_op_add_vec:
3208    case INDEX_op_sub_vec:
3209    case INDEX_op_smax_vec:
3210    case INDEX_op_smin_vec:
3211    case INDEX_op_umax_vec:
3212    case INDEX_op_umin_vec:
3213    case INDEX_op_shlv_vec:
3214    case INDEX_op_shrv_vec:
3215    case INDEX_op_sarv_vec:
3216    case INDEX_op_rotlv_vec:
3217        return vece <= MO_32 || have_isa_2_07;
3218    case INDEX_op_ssadd_vec:
3219    case INDEX_op_sssub_vec:
3220    case INDEX_op_usadd_vec:
3221    case INDEX_op_ussub_vec:
3222        return vece <= MO_32;
3223    case INDEX_op_cmp_vec:
3224    case INDEX_op_shli_vec:
3225    case INDEX_op_shri_vec:
3226    case INDEX_op_sari_vec:
3227    case INDEX_op_rotli_vec:
3228        return vece <= MO_32 || have_isa_2_07 ? -1 : 0;
3229    case INDEX_op_neg_vec:
3230        return vece >= MO_32 && have_isa_3_00;
3231    case INDEX_op_mul_vec:
3232        switch (vece) {
3233        case MO_8:
3234        case MO_16:
3235            return -1;
3236        case MO_32:
3237            return have_isa_2_07 ? 1 : -1;
3238        case MO_64:
3239            return have_isa_3_10;
3240        }
3241        return 0;
3242    case INDEX_op_bitsel_vec:
3243        return have_vsx;
3244    case INDEX_op_rotrv_vec:
3245        return -1;
3246    default:
3247        return 0;
3248    }
3249}
3250
3251static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
3252                            TCGReg dst, TCGReg src)
3253{
3254    tcg_debug_assert(dst >= TCG_REG_V0);
3255
3256    /* Splat from integer reg allowed via constraints for v3.00.  */
3257    if (src < TCG_REG_V0) {
3258        tcg_debug_assert(have_isa_3_00);
3259        switch (vece) {
3260        case MO_64:
3261            tcg_out32(s, MTVSRDD | VRT(dst) | RA(src) | RB(src));
3262            return true;
3263        case MO_32:
3264            tcg_out32(s, MTVSRWS | VRT(dst) | RA(src));
3265            return true;
3266        default:
3267            /* Fail, so that we fall back on either dupm or mov+dup.  */
3268            return false;
3269        }
3270    }
3271
3272    /*
3273     * Recall we use (or emulate) VSX integer loads, so the integer is
3274     * right justified within the left (zero-index) double-word.
3275     */
3276    switch (vece) {
3277    case MO_8:
3278        tcg_out32(s, VSPLTB | VRT(dst) | VRB(src) | (7 << 16));
3279        break;
3280    case MO_16:
3281        tcg_out32(s, VSPLTH | VRT(dst) | VRB(src) | (3 << 16));
3282        break;
3283    case MO_32:
3284        tcg_out32(s, VSPLTW | VRT(dst) | VRB(src) | (1 << 16));
3285        break;
3286    case MO_64:
3287        if (have_vsx) {
3288            tcg_out32(s, XXPERMDI | VRT(dst) | VRA(src) | VRB(src));
3289            break;
3290        }
3291        tcg_out_vsldoi(s, TCG_VEC_TMP1, src, src, 8);
3292        tcg_out_vsldoi(s, dst, TCG_VEC_TMP1, src, 8);
3293        break;
3294    default:
3295        g_assert_not_reached();
3296    }
3297    return true;
3298}
3299
3300static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
3301                             TCGReg out, TCGReg base, intptr_t offset)
3302{
3303    int elt;
3304
3305    tcg_debug_assert(out >= TCG_REG_V0);
3306    switch (vece) {
3307    case MO_8:
3308        if (have_isa_3_00) {
3309            tcg_out_mem_long(s, LXV, LVX, out, base, offset & -16);
3310        } else {
3311            tcg_out_mem_long(s, 0, LVEBX, out, base, offset);
3312        }
3313        elt = extract32(offset, 0, 4);
3314#if !HOST_BIG_ENDIAN
3315        elt ^= 15;
3316#endif
3317        tcg_out32(s, VSPLTB | VRT(out) | VRB(out) | (elt << 16));
3318        break;
3319    case MO_16:
3320        tcg_debug_assert((offset & 1) == 0);
3321        if (have_isa_3_00) {
3322            tcg_out_mem_long(s, LXV | 8, LVX, out, base, offset & -16);
3323        } else {
3324            tcg_out_mem_long(s, 0, LVEHX, out, base, offset);
3325        }
3326        elt = extract32(offset, 1, 3);
3327#if !HOST_BIG_ENDIAN
3328        elt ^= 7;
3329#endif
3330        tcg_out32(s, VSPLTH | VRT(out) | VRB(out) | (elt << 16));
3331        break;
3332    case MO_32:
3333        if (have_isa_3_00) {
3334            tcg_out_mem_long(s, 0, LXVWSX, out, base, offset);
3335            break;
3336        }
3337        tcg_debug_assert((offset & 3) == 0);
3338        tcg_out_mem_long(s, 0, LVEWX, out, base, offset);
3339        elt = extract32(offset, 2, 2);
3340#if !HOST_BIG_ENDIAN
3341        elt ^= 3;
3342#endif
3343        tcg_out32(s, VSPLTW | VRT(out) | VRB(out) | (elt << 16));
3344        break;
3345    case MO_64:
3346        if (have_vsx) {
3347            tcg_out_mem_long(s, 0, LXVDSX, out, base, offset);
3348            break;
3349        }
3350        tcg_debug_assert((offset & 7) == 0);
3351        tcg_out_mem_long(s, 0, LVX, out, base, offset & -16);
3352        tcg_out_vsldoi(s, TCG_VEC_TMP1, out, out, 8);
3353        elt = extract32(offset, 3, 1);
3354#if !HOST_BIG_ENDIAN
3355        elt = !elt;
3356#endif
3357        if (elt) {
3358            tcg_out_vsldoi(s, out, out, TCG_VEC_TMP1, 8);
3359        } else {
3360            tcg_out_vsldoi(s, out, TCG_VEC_TMP1, out, 8);
3361        }
3362        break;
3363    default:
3364        g_assert_not_reached();
3365    }
3366    return true;
3367}
3368
3369static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
3370                           unsigned vecl, unsigned vece,
3371                           const TCGArg args[TCG_MAX_OP_ARGS],
3372                           const int const_args[TCG_MAX_OP_ARGS])
3373{
3374    static const uint32_t
3375        add_op[4] = { VADDUBM, VADDUHM, VADDUWM, VADDUDM },
3376        sub_op[4] = { VSUBUBM, VSUBUHM, VSUBUWM, VSUBUDM },
3377        mul_op[4] = { 0, 0, VMULUWM, VMULLD },
3378        neg_op[4] = { 0, 0, VNEGW, VNEGD },
3379        eq_op[4]  = { VCMPEQUB, VCMPEQUH, VCMPEQUW, VCMPEQUD },
3380        ne_op[4]  = { VCMPNEB, VCMPNEH, VCMPNEW, 0 },
3381        gts_op[4] = { VCMPGTSB, VCMPGTSH, VCMPGTSW, VCMPGTSD },
3382        gtu_op[4] = { VCMPGTUB, VCMPGTUH, VCMPGTUW, VCMPGTUD },
3383        ssadd_op[4] = { VADDSBS, VADDSHS, VADDSWS, 0 },
3384        usadd_op[4] = { VADDUBS, VADDUHS, VADDUWS, 0 },
3385        sssub_op[4] = { VSUBSBS, VSUBSHS, VSUBSWS, 0 },
3386        ussub_op[4] = { VSUBUBS, VSUBUHS, VSUBUWS, 0 },
3387        umin_op[4] = { VMINUB, VMINUH, VMINUW, VMINUD },
3388        smin_op[4] = { VMINSB, VMINSH, VMINSW, VMINSD },
3389        umax_op[4] = { VMAXUB, VMAXUH, VMAXUW, VMAXUD },
3390        smax_op[4] = { VMAXSB, VMAXSH, VMAXSW, VMAXSD },
3391        shlv_op[4] = { VSLB, VSLH, VSLW, VSLD },
3392        shrv_op[4] = { VSRB, VSRH, VSRW, VSRD },
3393        sarv_op[4] = { VSRAB, VSRAH, VSRAW, VSRAD },
3394        mrgh_op[4] = { VMRGHB, VMRGHH, VMRGHW, 0 },
3395        mrgl_op[4] = { VMRGLB, VMRGLH, VMRGLW, 0 },
3396        muleu_op[4] = { VMULEUB, VMULEUH, VMULEUW, 0 },
3397        mulou_op[4] = { VMULOUB, VMULOUH, VMULOUW, 0 },
3398        pkum_op[4] = { VPKUHUM, VPKUWUM, 0, 0 },
3399        rotl_op[4] = { VRLB, VRLH, VRLW, VRLD };
3400
3401    TCGType type = vecl + TCG_TYPE_V64;
3402    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
3403    uint32_t insn;
3404
3405    switch (opc) {
3406    case INDEX_op_ld_vec:
3407        tcg_out_ld(s, type, a0, a1, a2);
3408        return;
3409    case INDEX_op_st_vec:
3410        tcg_out_st(s, type, a0, a1, a2);
3411        return;
3412    case INDEX_op_dupm_vec:
3413        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
3414        return;
3415
3416    case INDEX_op_add_vec:
3417        insn = add_op[vece];
3418        break;
3419    case INDEX_op_sub_vec:
3420        insn = sub_op[vece];
3421        break;
3422    case INDEX_op_neg_vec:
3423        insn = neg_op[vece];
3424        a2 = a1;
3425        a1 = 0;
3426        break;
3427    case INDEX_op_mul_vec:
3428        insn = mul_op[vece];
3429        break;
3430    case INDEX_op_ssadd_vec:
3431        insn = ssadd_op[vece];
3432        break;
3433    case INDEX_op_sssub_vec:
3434        insn = sssub_op[vece];
3435        break;
3436    case INDEX_op_usadd_vec:
3437        insn = usadd_op[vece];
3438        break;
3439    case INDEX_op_ussub_vec:
3440        insn = ussub_op[vece];
3441        break;
3442    case INDEX_op_smin_vec:
3443        insn = smin_op[vece];
3444        break;
3445    case INDEX_op_umin_vec:
3446        insn = umin_op[vece];
3447        break;
3448    case INDEX_op_smax_vec:
3449        insn = smax_op[vece];
3450        break;
3451    case INDEX_op_umax_vec:
3452        insn = umax_op[vece];
3453        break;
3454    case INDEX_op_shlv_vec:
3455        insn = shlv_op[vece];
3456        break;
3457    case INDEX_op_shrv_vec:
3458        insn = shrv_op[vece];
3459        break;
3460    case INDEX_op_sarv_vec:
3461        insn = sarv_op[vece];
3462        break;
3463    case INDEX_op_and_vec:
3464        insn = VAND;
3465        break;
3466    case INDEX_op_or_vec:
3467        insn = VOR;
3468        break;
3469    case INDEX_op_xor_vec:
3470        insn = VXOR;
3471        break;
3472    case INDEX_op_andc_vec:
3473        insn = VANDC;
3474        break;
3475    case INDEX_op_not_vec:
3476        insn = VNOR;
3477        a2 = a1;
3478        break;
3479    case INDEX_op_orc_vec:
3480        insn = VORC;
3481        break;
3482    case INDEX_op_nand_vec:
3483        insn = VNAND;
3484        break;
3485    case INDEX_op_nor_vec:
3486        insn = VNOR;
3487        break;
3488    case INDEX_op_eqv_vec:
3489        insn = VEQV;
3490        break;
3491
3492    case INDEX_op_cmp_vec:
3493        switch (args[3]) {
3494        case TCG_COND_EQ:
3495            insn = eq_op[vece];
3496            break;
3497        case TCG_COND_NE:
3498            insn = ne_op[vece];
3499            break;
3500        case TCG_COND_GT:
3501            insn = gts_op[vece];
3502            break;
3503        case TCG_COND_GTU:
3504            insn = gtu_op[vece];
3505            break;
3506        default:
3507            g_assert_not_reached();
3508        }
3509        break;
3510
3511    case INDEX_op_bitsel_vec:
3512        tcg_out32(s, XXSEL | VRT(a0) | VRC(a1) | VRB(a2) | VRA(args[3]));
3513        return;
3514
3515    case INDEX_op_dup2_vec:
3516        assert(TCG_TARGET_REG_BITS == 32);
3517        /* With inputs a1 = xLxx, a2 = xHxx  */
3518        tcg_out32(s, VMRGHW | VRT(a0) | VRA(a2) | VRB(a1));  /* a0  = xxHL */
3519        tcg_out_vsldoi(s, TCG_VEC_TMP1, a0, a0, 8);          /* tmp = HLxx */
3520        tcg_out_vsldoi(s, a0, a0, TCG_VEC_TMP1, 8);          /* a0  = HLHL */
3521        return;
3522
3523    case INDEX_op_ppc_mrgh_vec:
3524        insn = mrgh_op[vece];
3525        break;
3526    case INDEX_op_ppc_mrgl_vec:
3527        insn = mrgl_op[vece];
3528        break;
3529    case INDEX_op_ppc_muleu_vec:
3530        insn = muleu_op[vece];
3531        break;
3532    case INDEX_op_ppc_mulou_vec:
3533        insn = mulou_op[vece];
3534        break;
3535    case INDEX_op_ppc_pkum_vec:
3536        insn = pkum_op[vece];
3537        break;
3538    case INDEX_op_rotlv_vec:
3539        insn = rotl_op[vece];
3540        break;
3541    case INDEX_op_ppc_msum_vec:
3542        tcg_debug_assert(vece == MO_16);
3543        tcg_out32(s, VMSUMUHM | VRT(a0) | VRA(a1) | VRB(a2) | VRC(args[3]));
3544        return;
3545
3546    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
3547    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
3548    default:
3549        g_assert_not_reached();
3550    }
3551
3552    tcg_debug_assert(insn != 0);
3553    tcg_out32(s, insn | VRT(a0) | VRA(a1) | VRB(a2));
3554}
3555
3556static void expand_vec_shi(TCGType type, unsigned vece, TCGv_vec v0,
3557                           TCGv_vec v1, TCGArg imm, TCGOpcode opci)
3558{
3559    TCGv_vec t1;
3560
3561    if (vece == MO_32) {
3562        /*
3563         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
3564         * So using negative numbers gets us the 4th bit easily.
3565         */
3566        imm = sextract32(imm, 0, 5);
3567    } else {
3568        imm &= (8 << vece) - 1;
3569    }
3570
3571    /* Splat w/bytes for xxspltib when 2.07 allows MO_64. */
3572    t1 = tcg_constant_vec(type, MO_8, imm);
3573    vec_gen_3(opci, type, vece, tcgv_vec_arg(v0),
3574              tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3575}
3576
3577static void expand_vec_cmp(TCGType type, unsigned vece, TCGv_vec v0,
3578                           TCGv_vec v1, TCGv_vec v2, TCGCond cond)
3579{
3580    bool need_swap = false, need_inv = false;
3581
3582    tcg_debug_assert(vece <= MO_32 || have_isa_2_07);
3583
3584    switch (cond) {
3585    case TCG_COND_EQ:
3586    case TCG_COND_GT:
3587    case TCG_COND_GTU:
3588        break;
3589    case TCG_COND_NE:
3590        if (have_isa_3_00 && vece <= MO_32) {
3591            break;
3592        }
3593        /* fall through */
3594    case TCG_COND_LE:
3595    case TCG_COND_LEU:
3596        need_inv = true;
3597        break;
3598    case TCG_COND_LT:
3599    case TCG_COND_LTU:
3600        need_swap = true;
3601        break;
3602    case TCG_COND_GE:
3603    case TCG_COND_GEU:
3604        need_swap = need_inv = true;
3605        break;
3606    default:
3607        g_assert_not_reached();
3608    }
3609
3610    if (need_inv) {
3611        cond = tcg_invert_cond(cond);
3612    }
3613    if (need_swap) {
3614        TCGv_vec t1;
3615        t1 = v1, v1 = v2, v2 = t1;
3616        cond = tcg_swap_cond(cond);
3617    }
3618
3619    vec_gen_4(INDEX_op_cmp_vec, type, vece, tcgv_vec_arg(v0),
3620              tcgv_vec_arg(v1), tcgv_vec_arg(v2), cond);
3621
3622    if (need_inv) {
3623        tcg_gen_not_vec(vece, v0, v0);
3624    }
3625}
3626
3627static void expand_vec_mul(TCGType type, unsigned vece, TCGv_vec v0,
3628                           TCGv_vec v1, TCGv_vec v2)
3629{
3630    TCGv_vec t1 = tcg_temp_new_vec(type);
3631    TCGv_vec t2 = tcg_temp_new_vec(type);
3632    TCGv_vec c0, c16;
3633
3634    switch (vece) {
3635    case MO_8:
3636    case MO_16:
3637        vec_gen_3(INDEX_op_ppc_muleu_vec, type, vece, tcgv_vec_arg(t1),
3638                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3639        vec_gen_3(INDEX_op_ppc_mulou_vec, type, vece, tcgv_vec_arg(t2),
3640                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3641        vec_gen_3(INDEX_op_ppc_mrgh_vec, type, vece + 1, tcgv_vec_arg(v0),
3642                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3643        vec_gen_3(INDEX_op_ppc_mrgl_vec, type, vece + 1, tcgv_vec_arg(t1),
3644                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3645        vec_gen_3(INDEX_op_ppc_pkum_vec, type, vece, tcgv_vec_arg(v0),
3646                  tcgv_vec_arg(v0), tcgv_vec_arg(t1));
3647	break;
3648
3649    case MO_32:
3650        tcg_debug_assert(!have_isa_2_07);
3651        /*
3652         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
3653         * So using -16 is a quick way to represent 16.
3654         */
3655        c16 = tcg_constant_vec(type, MO_8, -16);
3656        c0 = tcg_constant_vec(type, MO_8, 0);
3657
3658        vec_gen_3(INDEX_op_rotlv_vec, type, MO_32, tcgv_vec_arg(t1),
3659                  tcgv_vec_arg(v2), tcgv_vec_arg(c16));
3660        vec_gen_3(INDEX_op_ppc_mulou_vec, type, MO_16, tcgv_vec_arg(t2),
3661                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3662        vec_gen_4(INDEX_op_ppc_msum_vec, type, MO_16, tcgv_vec_arg(t1),
3663                  tcgv_vec_arg(v1), tcgv_vec_arg(t1), tcgv_vec_arg(c0));
3664        vec_gen_3(INDEX_op_shlv_vec, type, MO_32, tcgv_vec_arg(t1),
3665                  tcgv_vec_arg(t1), tcgv_vec_arg(c16));
3666        tcg_gen_add_vec(MO_32, v0, t1, t2);
3667        break;
3668
3669    default:
3670        g_assert_not_reached();
3671    }
3672    tcg_temp_free_vec(t1);
3673    tcg_temp_free_vec(t2);
3674}
3675
3676void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3677                       TCGArg a0, ...)
3678{
3679    va_list va;
3680    TCGv_vec v0, v1, v2, t0;
3681    TCGArg a2;
3682
3683    va_start(va, a0);
3684    v0 = temp_tcgv_vec(arg_temp(a0));
3685    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3686    a2 = va_arg(va, TCGArg);
3687
3688    switch (opc) {
3689    case INDEX_op_shli_vec:
3690        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shlv_vec);
3691        break;
3692    case INDEX_op_shri_vec:
3693        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shrv_vec);
3694        break;
3695    case INDEX_op_sari_vec:
3696        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_sarv_vec);
3697        break;
3698    case INDEX_op_rotli_vec:
3699        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_rotlv_vec);
3700        break;
3701    case INDEX_op_cmp_vec:
3702        v2 = temp_tcgv_vec(arg_temp(a2));
3703        expand_vec_cmp(type, vece, v0, v1, v2, va_arg(va, TCGArg));
3704        break;
3705    case INDEX_op_mul_vec:
3706        v2 = temp_tcgv_vec(arg_temp(a2));
3707        expand_vec_mul(type, vece, v0, v1, v2);
3708        break;
3709    case INDEX_op_rotlv_vec:
3710        v2 = temp_tcgv_vec(arg_temp(a2));
3711        t0 = tcg_temp_new_vec(type);
3712        tcg_gen_neg_vec(vece, t0, v2);
3713        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3714        tcg_temp_free_vec(t0);
3715        break;
3716    default:
3717        g_assert_not_reached();
3718    }
3719    va_end(va);
3720}
3721
3722static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
3723{
3724    switch (op) {
3725    case INDEX_op_goto_ptr:
3726        return C_O0_I1(r);
3727
3728    case INDEX_op_ld8u_i32:
3729    case INDEX_op_ld8s_i32:
3730    case INDEX_op_ld16u_i32:
3731    case INDEX_op_ld16s_i32:
3732    case INDEX_op_ld_i32:
3733    case INDEX_op_ctpop_i32:
3734    case INDEX_op_neg_i32:
3735    case INDEX_op_not_i32:
3736    case INDEX_op_ext8s_i32:
3737    case INDEX_op_ext16s_i32:
3738    case INDEX_op_bswap16_i32:
3739    case INDEX_op_bswap32_i32:
3740    case INDEX_op_extract_i32:
3741    case INDEX_op_ld8u_i64:
3742    case INDEX_op_ld8s_i64:
3743    case INDEX_op_ld16u_i64:
3744    case INDEX_op_ld16s_i64:
3745    case INDEX_op_ld32u_i64:
3746    case INDEX_op_ld32s_i64:
3747    case INDEX_op_ld_i64:
3748    case INDEX_op_ctpop_i64:
3749    case INDEX_op_neg_i64:
3750    case INDEX_op_not_i64:
3751    case INDEX_op_ext8s_i64:
3752    case INDEX_op_ext16s_i64:
3753    case INDEX_op_ext32s_i64:
3754    case INDEX_op_ext_i32_i64:
3755    case INDEX_op_extu_i32_i64:
3756    case INDEX_op_bswap16_i64:
3757    case INDEX_op_bswap32_i64:
3758    case INDEX_op_bswap64_i64:
3759    case INDEX_op_extract_i64:
3760        return C_O1_I1(r, r);
3761
3762    case INDEX_op_st8_i32:
3763    case INDEX_op_st16_i32:
3764    case INDEX_op_st_i32:
3765    case INDEX_op_st8_i64:
3766    case INDEX_op_st16_i64:
3767    case INDEX_op_st32_i64:
3768    case INDEX_op_st_i64:
3769        return C_O0_I2(r, r);
3770
3771    case INDEX_op_add_i32:
3772    case INDEX_op_and_i32:
3773    case INDEX_op_or_i32:
3774    case INDEX_op_xor_i32:
3775    case INDEX_op_andc_i32:
3776    case INDEX_op_orc_i32:
3777    case INDEX_op_eqv_i32:
3778    case INDEX_op_shl_i32:
3779    case INDEX_op_shr_i32:
3780    case INDEX_op_sar_i32:
3781    case INDEX_op_rotl_i32:
3782    case INDEX_op_rotr_i32:
3783    case INDEX_op_setcond_i32:
3784    case INDEX_op_and_i64:
3785    case INDEX_op_andc_i64:
3786    case INDEX_op_shl_i64:
3787    case INDEX_op_shr_i64:
3788    case INDEX_op_sar_i64:
3789    case INDEX_op_rotl_i64:
3790    case INDEX_op_rotr_i64:
3791    case INDEX_op_setcond_i64:
3792        return C_O1_I2(r, r, ri);
3793
3794    case INDEX_op_mul_i32:
3795    case INDEX_op_mul_i64:
3796        return C_O1_I2(r, r, rI);
3797
3798    case INDEX_op_div_i32:
3799    case INDEX_op_divu_i32:
3800    case INDEX_op_rem_i32:
3801    case INDEX_op_remu_i32:
3802    case INDEX_op_nand_i32:
3803    case INDEX_op_nor_i32:
3804    case INDEX_op_muluh_i32:
3805    case INDEX_op_mulsh_i32:
3806    case INDEX_op_orc_i64:
3807    case INDEX_op_eqv_i64:
3808    case INDEX_op_nand_i64:
3809    case INDEX_op_nor_i64:
3810    case INDEX_op_div_i64:
3811    case INDEX_op_divu_i64:
3812    case INDEX_op_rem_i64:
3813    case INDEX_op_remu_i64:
3814    case INDEX_op_mulsh_i64:
3815    case INDEX_op_muluh_i64:
3816        return C_O1_I2(r, r, r);
3817
3818    case INDEX_op_sub_i32:
3819        return C_O1_I2(r, rI, ri);
3820    case INDEX_op_add_i64:
3821        return C_O1_I2(r, r, rT);
3822    case INDEX_op_or_i64:
3823    case INDEX_op_xor_i64:
3824        return C_O1_I2(r, r, rU);
3825    case INDEX_op_sub_i64:
3826        return C_O1_I2(r, rI, rT);
3827    case INDEX_op_clz_i32:
3828    case INDEX_op_ctz_i32:
3829    case INDEX_op_clz_i64:
3830    case INDEX_op_ctz_i64:
3831        return C_O1_I2(r, r, rZW);
3832
3833    case INDEX_op_brcond_i32:
3834    case INDEX_op_brcond_i64:
3835        return C_O0_I2(r, ri);
3836
3837    case INDEX_op_movcond_i32:
3838    case INDEX_op_movcond_i64:
3839        return C_O1_I4(r, r, ri, rZ, rZ);
3840    case INDEX_op_deposit_i32:
3841    case INDEX_op_deposit_i64:
3842        return C_O1_I2(r, 0, rZ);
3843    case INDEX_op_brcond2_i32:
3844        return C_O0_I4(r, r, ri, ri);
3845    case INDEX_op_setcond2_i32:
3846        return C_O1_I4(r, r, r, ri, ri);
3847    case INDEX_op_add2_i64:
3848    case INDEX_op_add2_i32:
3849        return C_O2_I4(r, r, r, r, rI, rZM);
3850    case INDEX_op_sub2_i64:
3851    case INDEX_op_sub2_i32:
3852        return C_O2_I4(r, r, rI, rZM, r, r);
3853
3854    case INDEX_op_qemu_ld_i32:
3855        return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
3856                ? C_O1_I1(r, L)
3857                : C_O1_I2(r, L, L));
3858
3859    case INDEX_op_qemu_st_i32:
3860        return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
3861                ? C_O0_I2(S, S)
3862                : C_O0_I3(S, S, S));
3863
3864    case INDEX_op_qemu_ld_i64:
3865        return (TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, L)
3866                : TARGET_LONG_BITS == 32 ? C_O2_I1(L, L, L)
3867                : C_O2_I2(L, L, L, L));
3868
3869    case INDEX_op_qemu_st_i64:
3870        return (TCG_TARGET_REG_BITS == 64 ? C_O0_I2(S, S)
3871                : TARGET_LONG_BITS == 32 ? C_O0_I3(S, S, S)
3872                : C_O0_I4(S, S, S, S));
3873
3874    case INDEX_op_add_vec:
3875    case INDEX_op_sub_vec:
3876    case INDEX_op_mul_vec:
3877    case INDEX_op_and_vec:
3878    case INDEX_op_or_vec:
3879    case INDEX_op_xor_vec:
3880    case INDEX_op_andc_vec:
3881    case INDEX_op_orc_vec:
3882    case INDEX_op_nor_vec:
3883    case INDEX_op_eqv_vec:
3884    case INDEX_op_nand_vec:
3885    case INDEX_op_cmp_vec:
3886    case INDEX_op_ssadd_vec:
3887    case INDEX_op_sssub_vec:
3888    case INDEX_op_usadd_vec:
3889    case INDEX_op_ussub_vec:
3890    case INDEX_op_smax_vec:
3891    case INDEX_op_smin_vec:
3892    case INDEX_op_umax_vec:
3893    case INDEX_op_umin_vec:
3894    case INDEX_op_shlv_vec:
3895    case INDEX_op_shrv_vec:
3896    case INDEX_op_sarv_vec:
3897    case INDEX_op_rotlv_vec:
3898    case INDEX_op_rotrv_vec:
3899    case INDEX_op_ppc_mrgh_vec:
3900    case INDEX_op_ppc_mrgl_vec:
3901    case INDEX_op_ppc_muleu_vec:
3902    case INDEX_op_ppc_mulou_vec:
3903    case INDEX_op_ppc_pkum_vec:
3904    case INDEX_op_dup2_vec:
3905        return C_O1_I2(v, v, v);
3906
3907    case INDEX_op_not_vec:
3908    case INDEX_op_neg_vec:
3909        return C_O1_I1(v, v);
3910
3911    case INDEX_op_dup_vec:
3912        return have_isa_3_00 ? C_O1_I1(v, vr) : C_O1_I1(v, v);
3913
3914    case INDEX_op_ld_vec:
3915    case INDEX_op_dupm_vec:
3916        return C_O1_I1(v, r);
3917
3918    case INDEX_op_st_vec:
3919        return C_O0_I2(v, r);
3920
3921    case INDEX_op_bitsel_vec:
3922    case INDEX_op_ppc_msum_vec:
3923        return C_O1_I3(v, v, v, v);
3924
3925    default:
3926        g_assert_not_reached();
3927    }
3928}
3929
3930static void tcg_target_init(TCGContext *s)
3931{
3932    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3933    unsigned long hwcap2 = qemu_getauxval(AT_HWCAP2);
3934
3935    have_isa = tcg_isa_base;
3936    if (hwcap & PPC_FEATURE_ARCH_2_06) {
3937        have_isa = tcg_isa_2_06;
3938    }
3939#ifdef PPC_FEATURE2_ARCH_2_07
3940    if (hwcap2 & PPC_FEATURE2_ARCH_2_07) {
3941        have_isa = tcg_isa_2_07;
3942    }
3943#endif
3944#ifdef PPC_FEATURE2_ARCH_3_00
3945    if (hwcap2 & PPC_FEATURE2_ARCH_3_00) {
3946        have_isa = tcg_isa_3_00;
3947    }
3948#endif
3949#ifdef PPC_FEATURE2_ARCH_3_10
3950    if (hwcap2 & PPC_FEATURE2_ARCH_3_10) {
3951        have_isa = tcg_isa_3_10;
3952    }
3953#endif
3954
3955#ifdef PPC_FEATURE2_HAS_ISEL
3956    /* Prefer explicit instruction from the kernel. */
3957    have_isel = (hwcap2 & PPC_FEATURE2_HAS_ISEL) != 0;
3958#else
3959    /* Fall back to knowing Power7 (2.06) has ISEL. */
3960    have_isel = have_isa_2_06;
3961#endif
3962
3963    if (hwcap & PPC_FEATURE_HAS_ALTIVEC) {
3964        have_altivec = true;
3965        /* We only care about the portion of VSX that overlaps Altivec. */
3966        if (hwcap & PPC_FEATURE_HAS_VSX) {
3967            have_vsx = true;
3968        }
3969    }
3970
3971    tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
3972    tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
3973    if (have_altivec) {
3974        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3975        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3976    }
3977
3978    tcg_target_call_clobber_regs = 0;
3979    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3980    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3981    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3982    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3983    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3984    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3985    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R7);
3986    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R8);
3987    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R9);
3988    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R10);
3989    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R11);
3990    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
3991
3992    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3993    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3994    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3995    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3996    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3997    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3998    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3999    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
4000    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V8);
4001    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V9);
4002    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V10);
4003    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V11);
4004    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V12);
4005    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V13);
4006    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V14);
4007    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V15);
4008    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
4009    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
4010    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
4011    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
4012
4013    s->reserved_regs = 0;
4014    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); /* tcg temp */
4015    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R1); /* stack pointer */
4016#if defined(_CALL_SYSV)
4017    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R2); /* toc pointer */
4018#endif
4019#if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
4020    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R13); /* thread pointer */
4021#endif
4022    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1); /* mem temp */
4023    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP1);
4024    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP2);
4025    if (USE_REG_TB) {
4026        tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);  /* tb->tc_ptr */
4027    }
4028}
4029
4030#ifdef __ELF__
4031typedef struct {
4032    DebugFrameCIE cie;
4033    DebugFrameFDEHeader fde;
4034    uint8_t fde_def_cfa[4];
4035    uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2 + 3];
4036} DebugFrame;
4037
4038/* We're expecting a 2 byte uleb128 encoded value.  */
4039QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
4040
4041#if TCG_TARGET_REG_BITS == 64
4042# define ELF_HOST_MACHINE EM_PPC64
4043#else
4044# define ELF_HOST_MACHINE EM_PPC
4045#endif
4046
4047static DebugFrame debug_frame = {
4048    .cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
4049    .cie.id = -1,
4050    .cie.version = 1,
4051    .cie.code_align = 1,
4052    .cie.data_align = (-SZR & 0x7f),         /* sleb128 -SZR */
4053    .cie.return_column = 65,
4054
4055    /* Total FDE size does not include the "len" member.  */
4056    .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset),
4057
4058    .fde_def_cfa = {
4059        12, TCG_REG_R1,                 /* DW_CFA_def_cfa r1, ... */
4060        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
4061        (FRAME_SIZE >> 7)
4062    },
4063    .fde_reg_ofs = {
4064        /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
4065        0x11, 65, (LR_OFFSET / -SZR) & 0x7f,
4066    }
4067};
4068
4069void tcg_register_jit(const void *buf, size_t buf_size)
4070{
4071    uint8_t *p = &debug_frame.fde_reg_ofs[3];
4072    int i;
4073
4074    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i, p += 2) {
4075        p[0] = 0x80 + tcg_target_callee_save_regs[i];
4076        p[1] = (FRAME_SIZE - (REG_SAVE_BOT + i * SZR)) / SZR;
4077    }
4078
4079    debug_frame.fde.func_start = (uintptr_t)buf;
4080    debug_frame.fde.func_len = buf_size;
4081
4082    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
4083}
4084#endif /* __ELF__ */
4085#undef VMULEUB
4086#undef VMULEUH
4087#undef VMULEUW
4088#undef VMULOUB
4089#undef VMULOUH
4090#undef VMULOUW
4091#undef VMSUMUHM
4092