xref: /openbmc/qemu/tcg/ppc/tcg-target.c.inc (revision 92e87950)
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#ifdef _CALL_SYSV
48# define TCG_TARGET_CALL_ALIGN_ARGS   1
49#endif
50
51/* For some memory operations, we need a scratch that isn't R0.  For the AIX
52   calling convention, we can re-use the TOC register since we'll be reloading
53   it at every call.  Otherwise R12 will do nicely as neither a call-saved
54   register nor a parameter register.  */
55#ifdef _CALL_AIX
56# define TCG_REG_TMP1   TCG_REG_R2
57#else
58# define TCG_REG_TMP1   TCG_REG_R12
59#endif
60
61#define TCG_VEC_TMP1    TCG_REG_V0
62#define TCG_VEC_TMP2    TCG_REG_V1
63
64#define TCG_REG_TB     TCG_REG_R31
65#define USE_REG_TB     (TCG_TARGET_REG_BITS == 64)
66
67/* Shorthand for size of a pointer.  Avoid promotion to unsigned.  */
68#define SZP  ((int)sizeof(void *))
69
70/* Shorthand for size of a register.  */
71#define SZR  (TCG_TARGET_REG_BITS / 8)
72
73#define TCG_CT_CONST_S16  0x100
74#define TCG_CT_CONST_U16  0x200
75#define TCG_CT_CONST_S32  0x400
76#define TCG_CT_CONST_U32  0x800
77#define TCG_CT_CONST_ZERO 0x1000
78#define TCG_CT_CONST_MONE 0x2000
79#define TCG_CT_CONST_WSZ  0x4000
80
81#define ALL_GENERAL_REGS  0xffffffffu
82#define ALL_VECTOR_REGS   0xffffffff00000000ull
83
84#ifdef CONFIG_SOFTMMU
85#define ALL_QLOAD_REGS \
86    (ALL_GENERAL_REGS & \
87     ~((1 << TCG_REG_R3) | (1 << TCG_REG_R4) | (1 << TCG_REG_R5)))
88#define ALL_QSTORE_REGS \
89    (ALL_GENERAL_REGS & ~((1 << TCG_REG_R3) | (1 << TCG_REG_R4) | \
90                          (1 << TCG_REG_R5) | (1 << TCG_REG_R6)))
91#else
92#define ALL_QLOAD_REGS  (ALL_GENERAL_REGS & ~(1 << TCG_REG_R3))
93#define ALL_QSTORE_REGS ALL_QLOAD_REGS
94#endif
95
96TCGPowerISA have_isa;
97static bool have_isel;
98bool have_altivec;
99bool have_vsx;
100
101#ifndef CONFIG_SOFTMMU
102#define TCG_GUEST_BASE_REG 30
103#endif
104
105#ifdef CONFIG_DEBUG_TCG
106static const char tcg_target_reg_names[TCG_TARGET_NB_REGS][4] = {
107    "r0",  "r1",  "r2",  "r3",  "r4",  "r5",  "r6",  "r7",
108    "r8",  "r9",  "r10", "r11", "r12", "r13", "r14", "r15",
109    "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
110    "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31",
111    "v0",  "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7",
112    "v8",  "v9",  "v10", "v11", "v12", "v13", "v14", "v15",
113    "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
114    "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",
115};
116#endif
117
118static const int tcg_target_reg_alloc_order[] = {
119    TCG_REG_R14,  /* call saved registers */
120    TCG_REG_R15,
121    TCG_REG_R16,
122    TCG_REG_R17,
123    TCG_REG_R18,
124    TCG_REG_R19,
125    TCG_REG_R20,
126    TCG_REG_R21,
127    TCG_REG_R22,
128    TCG_REG_R23,
129    TCG_REG_R24,
130    TCG_REG_R25,
131    TCG_REG_R26,
132    TCG_REG_R27,
133    TCG_REG_R28,
134    TCG_REG_R29,
135    TCG_REG_R30,
136    TCG_REG_R31,
137    TCG_REG_R12,  /* call clobbered, non-arguments */
138    TCG_REG_R11,
139    TCG_REG_R2,
140    TCG_REG_R13,
141    TCG_REG_R10,  /* call clobbered, arguments */
142    TCG_REG_R9,
143    TCG_REG_R8,
144    TCG_REG_R7,
145    TCG_REG_R6,
146    TCG_REG_R5,
147    TCG_REG_R4,
148    TCG_REG_R3,
149
150    /* V0 and V1 reserved as temporaries; V20 - V31 are call-saved */
151    TCG_REG_V2,   /* call clobbered, vectors */
152    TCG_REG_V3,
153    TCG_REG_V4,
154    TCG_REG_V5,
155    TCG_REG_V6,
156    TCG_REG_V7,
157    TCG_REG_V8,
158    TCG_REG_V9,
159    TCG_REG_V10,
160    TCG_REG_V11,
161    TCG_REG_V12,
162    TCG_REG_V13,
163    TCG_REG_V14,
164    TCG_REG_V15,
165    TCG_REG_V16,
166    TCG_REG_V17,
167    TCG_REG_V18,
168    TCG_REG_V19,
169};
170
171static const int tcg_target_call_iarg_regs[] = {
172    TCG_REG_R3,
173    TCG_REG_R4,
174    TCG_REG_R5,
175    TCG_REG_R6,
176    TCG_REG_R7,
177    TCG_REG_R8,
178    TCG_REG_R9,
179    TCG_REG_R10
180};
181
182static const int tcg_target_call_oarg_regs[] = {
183    TCG_REG_R3,
184    TCG_REG_R4
185};
186
187static const int tcg_target_callee_save_regs[] = {
188#ifdef _CALL_DARWIN
189    TCG_REG_R11,
190#endif
191    TCG_REG_R14,
192    TCG_REG_R15,
193    TCG_REG_R16,
194    TCG_REG_R17,
195    TCG_REG_R18,
196    TCG_REG_R19,
197    TCG_REG_R20,
198    TCG_REG_R21,
199    TCG_REG_R22,
200    TCG_REG_R23,
201    TCG_REG_R24,
202    TCG_REG_R25,
203    TCG_REG_R26,
204    TCG_REG_R27, /* currently used for the global env */
205    TCG_REG_R28,
206    TCG_REG_R29,
207    TCG_REG_R30,
208    TCG_REG_R31
209};
210
211static inline bool in_range_b(tcg_target_long target)
212{
213    return target == sextract64(target, 0, 26);
214}
215
216static uint32_t reloc_pc24_val(const tcg_insn_unit *pc,
217			       const tcg_insn_unit *target)
218{
219    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
220    tcg_debug_assert(in_range_b(disp));
221    return disp & 0x3fffffc;
222}
223
224static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
225{
226    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
227    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
228
229    if (in_range_b(disp)) {
230        *src_rw = (*src_rw & ~0x3fffffc) | (disp & 0x3fffffc);
231        return true;
232    }
233    return false;
234}
235
236static uint16_t reloc_pc14_val(const tcg_insn_unit *pc,
237			       const tcg_insn_unit *target)
238{
239    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
240    tcg_debug_assert(disp == (int16_t) disp);
241    return disp & 0xfffc;
242}
243
244static bool reloc_pc14(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
245{
246    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
247    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
248
249    if (disp == (int16_t) disp) {
250        *src_rw = (*src_rw & ~0xfffc) | (disp & 0xfffc);
251        return true;
252    }
253    return false;
254}
255
256/* test if a constant matches the constraint */
257static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
258{
259    if (ct & TCG_CT_CONST) {
260        return 1;
261    }
262
263    /* The only 32-bit constraint we use aside from
264       TCG_CT_CONST is TCG_CT_CONST_S16.  */
265    if (type == TCG_TYPE_I32) {
266        val = (int32_t)val;
267    }
268
269    if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
270        return 1;
271    } else if ((ct & TCG_CT_CONST_U16) && val == (uint16_t)val) {
272        return 1;
273    } else if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
274        return 1;
275    } else if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
276        return 1;
277    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
278        return 1;
279    } else if ((ct & TCG_CT_CONST_MONE) && val == -1) {
280        return 1;
281    } else if ((ct & TCG_CT_CONST_WSZ)
282               && val == (type == TCG_TYPE_I32 ? 32 : 64)) {
283        return 1;
284    }
285    return 0;
286}
287
288#define OPCD(opc) ((opc)<<26)
289#define XO19(opc) (OPCD(19)|((opc)<<1))
290#define MD30(opc) (OPCD(30)|((opc)<<2))
291#define MDS30(opc) (OPCD(30)|((opc)<<1))
292#define XO31(opc) (OPCD(31)|((opc)<<1))
293#define XO58(opc) (OPCD(58)|(opc))
294#define XO62(opc) (OPCD(62)|(opc))
295#define VX4(opc)  (OPCD(4)|(opc))
296
297#define B      OPCD( 18)
298#define BC     OPCD( 16)
299#define LBZ    OPCD( 34)
300#define LHZ    OPCD( 40)
301#define LHA    OPCD( 42)
302#define LWZ    OPCD( 32)
303#define LWZUX  XO31( 55)
304#define STB    OPCD( 38)
305#define STH    OPCD( 44)
306#define STW    OPCD( 36)
307
308#define STD    XO62(  0)
309#define STDU   XO62(  1)
310#define STDX   XO31(149)
311
312#define LD     XO58(  0)
313#define LDX    XO31( 21)
314#define LDU    XO58(  1)
315#define LDUX   XO31( 53)
316#define LWA    XO58(  2)
317#define LWAX   XO31(341)
318
319#define ADDIC  OPCD( 12)
320#define ADDI   OPCD( 14)
321#define ADDIS  OPCD( 15)
322#define ORI    OPCD( 24)
323#define ORIS   OPCD( 25)
324#define XORI   OPCD( 26)
325#define XORIS  OPCD( 27)
326#define ANDI   OPCD( 28)
327#define ANDIS  OPCD( 29)
328#define MULLI  OPCD(  7)
329#define CMPLI  OPCD( 10)
330#define CMPI   OPCD( 11)
331#define SUBFIC OPCD( 8)
332
333#define LWZU   OPCD( 33)
334#define STWU   OPCD( 37)
335
336#define RLWIMI OPCD( 20)
337#define RLWINM OPCD( 21)
338#define RLWNM  OPCD( 23)
339
340#define RLDICL MD30(  0)
341#define RLDICR MD30(  1)
342#define RLDIMI MD30(  3)
343#define RLDCL  MDS30( 8)
344
345#define BCLR   XO19( 16)
346#define BCCTR  XO19(528)
347#define CRAND  XO19(257)
348#define CRANDC XO19(129)
349#define CRNAND XO19(225)
350#define CROR   XO19(449)
351#define CRNOR  XO19( 33)
352
353#define EXTSB  XO31(954)
354#define EXTSH  XO31(922)
355#define EXTSW  XO31(986)
356#define ADD    XO31(266)
357#define ADDE   XO31(138)
358#define ADDME  XO31(234)
359#define ADDZE  XO31(202)
360#define ADDC   XO31( 10)
361#define AND    XO31( 28)
362#define SUBF   XO31( 40)
363#define SUBFC  XO31(  8)
364#define SUBFE  XO31(136)
365#define SUBFME XO31(232)
366#define SUBFZE XO31(200)
367#define OR     XO31(444)
368#define XOR    XO31(316)
369#define MULLW  XO31(235)
370#define MULHW  XO31( 75)
371#define MULHWU XO31( 11)
372#define DIVW   XO31(491)
373#define DIVWU  XO31(459)
374#define MODSW  XO31(779)
375#define MODUW  XO31(267)
376#define CMP    XO31(  0)
377#define CMPL   XO31( 32)
378#define LHBRX  XO31(790)
379#define LWBRX  XO31(534)
380#define LDBRX  XO31(532)
381#define STHBRX XO31(918)
382#define STWBRX XO31(662)
383#define STDBRX XO31(660)
384#define MFSPR  XO31(339)
385#define MTSPR  XO31(467)
386#define SRAWI  XO31(824)
387#define NEG    XO31(104)
388#define MFCR   XO31( 19)
389#define MFOCRF (MFCR | (1u << 20))
390#define NOR    XO31(124)
391#define CNTLZW XO31( 26)
392#define CNTLZD XO31( 58)
393#define CNTTZW XO31(538)
394#define CNTTZD XO31(570)
395#define CNTPOPW XO31(378)
396#define CNTPOPD XO31(506)
397#define ANDC   XO31( 60)
398#define ORC    XO31(412)
399#define EQV    XO31(284)
400#define NAND   XO31(476)
401#define ISEL   XO31( 15)
402
403#define MULLD  XO31(233)
404#define MULHD  XO31( 73)
405#define MULHDU XO31(  9)
406#define DIVD   XO31(489)
407#define DIVDU  XO31(457)
408#define MODSD  XO31(777)
409#define MODUD  XO31(265)
410
411#define LBZX   XO31( 87)
412#define LHZX   XO31(279)
413#define LHAX   XO31(343)
414#define LWZX   XO31( 23)
415#define STBX   XO31(215)
416#define STHX   XO31(407)
417#define STWX   XO31(151)
418
419#define EIEIO  XO31(854)
420#define HWSYNC XO31(598)
421#define LWSYNC (HWSYNC | (1u << 21))
422
423#define SPR(a, b) ((((a)<<5)|(b))<<11)
424#define LR     SPR(8, 0)
425#define CTR    SPR(9, 0)
426
427#define SLW    XO31( 24)
428#define SRW    XO31(536)
429#define SRAW   XO31(792)
430
431#define SLD    XO31( 27)
432#define SRD    XO31(539)
433#define SRAD   XO31(794)
434#define SRADI  XO31(413<<1)
435
436#define BRH    XO31(219)
437#define BRW    XO31(155)
438#define BRD    XO31(187)
439
440#define TW     XO31( 4)
441#define TRAP   (TW | TO(31))
442
443#define NOP    ORI  /* ori 0,0,0 */
444
445#define LVX        XO31(103)
446#define LVEBX      XO31(7)
447#define LVEHX      XO31(39)
448#define LVEWX      XO31(71)
449#define LXSDX      (XO31(588) | 1)  /* v2.06, force tx=1 */
450#define LXVDSX     (XO31(332) | 1)  /* v2.06, force tx=1 */
451#define LXSIWZX    (XO31(12) | 1)   /* v2.07, force tx=1 */
452#define LXV        (OPCD(61) | 8 | 1)  /* v3.00, force tx=1 */
453#define LXSD       (OPCD(57) | 2)   /* v3.00 */
454#define LXVWSX     (XO31(364) | 1)  /* v3.00, force tx=1 */
455
456#define STVX       XO31(231)
457#define STVEWX     XO31(199)
458#define STXSDX     (XO31(716) | 1)  /* v2.06, force sx=1 */
459#define STXSIWX    (XO31(140) | 1)  /* v2.07, force sx=1 */
460#define STXV       (OPCD(61) | 8 | 5) /* v3.00, force sx=1 */
461#define STXSD      (OPCD(61) | 2)   /* v3.00 */
462
463#define VADDSBS    VX4(768)
464#define VADDUBS    VX4(512)
465#define VADDUBM    VX4(0)
466#define VADDSHS    VX4(832)
467#define VADDUHS    VX4(576)
468#define VADDUHM    VX4(64)
469#define VADDSWS    VX4(896)
470#define VADDUWS    VX4(640)
471#define VADDUWM    VX4(128)
472#define VADDUDM    VX4(192)       /* v2.07 */
473
474#define VSUBSBS    VX4(1792)
475#define VSUBUBS    VX4(1536)
476#define VSUBUBM    VX4(1024)
477#define VSUBSHS    VX4(1856)
478#define VSUBUHS    VX4(1600)
479#define VSUBUHM    VX4(1088)
480#define VSUBSWS    VX4(1920)
481#define VSUBUWS    VX4(1664)
482#define VSUBUWM    VX4(1152)
483#define VSUBUDM    VX4(1216)      /* v2.07 */
484
485#define VNEGW      (VX4(1538) | (6 << 16))  /* v3.00 */
486#define VNEGD      (VX4(1538) | (7 << 16))  /* v3.00 */
487
488#define VMAXSB     VX4(258)
489#define VMAXSH     VX4(322)
490#define VMAXSW     VX4(386)
491#define VMAXSD     VX4(450)       /* v2.07 */
492#define VMAXUB     VX4(2)
493#define VMAXUH     VX4(66)
494#define VMAXUW     VX4(130)
495#define VMAXUD     VX4(194)       /* v2.07 */
496#define VMINSB     VX4(770)
497#define VMINSH     VX4(834)
498#define VMINSW     VX4(898)
499#define VMINSD     VX4(962)       /* v2.07 */
500#define VMINUB     VX4(514)
501#define VMINUH     VX4(578)
502#define VMINUW     VX4(642)
503#define VMINUD     VX4(706)       /* v2.07 */
504
505#define VCMPEQUB   VX4(6)
506#define VCMPEQUH   VX4(70)
507#define VCMPEQUW   VX4(134)
508#define VCMPEQUD   VX4(199)       /* v2.07 */
509#define VCMPGTSB   VX4(774)
510#define VCMPGTSH   VX4(838)
511#define VCMPGTSW   VX4(902)
512#define VCMPGTSD   VX4(967)       /* v2.07 */
513#define VCMPGTUB   VX4(518)
514#define VCMPGTUH   VX4(582)
515#define VCMPGTUW   VX4(646)
516#define VCMPGTUD   VX4(711)       /* v2.07 */
517#define VCMPNEB    VX4(7)         /* v3.00 */
518#define VCMPNEH    VX4(71)        /* v3.00 */
519#define VCMPNEW    VX4(135)       /* v3.00 */
520
521#define VSLB       VX4(260)
522#define VSLH       VX4(324)
523#define VSLW       VX4(388)
524#define VSLD       VX4(1476)      /* v2.07 */
525#define VSRB       VX4(516)
526#define VSRH       VX4(580)
527#define VSRW       VX4(644)
528#define VSRD       VX4(1732)      /* v2.07 */
529#define VSRAB      VX4(772)
530#define VSRAH      VX4(836)
531#define VSRAW      VX4(900)
532#define VSRAD      VX4(964)       /* v2.07 */
533#define VRLB       VX4(4)
534#define VRLH       VX4(68)
535#define VRLW       VX4(132)
536#define VRLD       VX4(196)       /* v2.07 */
537
538#define VMULEUB    VX4(520)
539#define VMULEUH    VX4(584)
540#define VMULEUW    VX4(648)       /* v2.07 */
541#define VMULOUB    VX4(8)
542#define VMULOUH    VX4(72)
543#define VMULOUW    VX4(136)       /* v2.07 */
544#define VMULUWM    VX4(137)       /* v2.07 */
545#define VMULLD     VX4(457)       /* v3.10 */
546#define VMSUMUHM   VX4(38)
547
548#define VMRGHB     VX4(12)
549#define VMRGHH     VX4(76)
550#define VMRGHW     VX4(140)
551#define VMRGLB     VX4(268)
552#define VMRGLH     VX4(332)
553#define VMRGLW     VX4(396)
554
555#define VPKUHUM    VX4(14)
556#define VPKUWUM    VX4(78)
557
558#define VAND       VX4(1028)
559#define VANDC      VX4(1092)
560#define VNOR       VX4(1284)
561#define VOR        VX4(1156)
562#define VXOR       VX4(1220)
563#define VEQV       VX4(1668)      /* v2.07 */
564#define VNAND      VX4(1412)      /* v2.07 */
565#define VORC       VX4(1348)      /* v2.07 */
566
567#define VSPLTB     VX4(524)
568#define VSPLTH     VX4(588)
569#define VSPLTW     VX4(652)
570#define VSPLTISB   VX4(780)
571#define VSPLTISH   VX4(844)
572#define VSPLTISW   VX4(908)
573
574#define VSLDOI     VX4(44)
575
576#define XXPERMDI   (OPCD(60) | (10 << 3) | 7)  /* v2.06, force ax=bx=tx=1 */
577#define XXSEL      (OPCD(60) | (3 << 4) | 0xf) /* v2.06, force ax=bx=cx=tx=1 */
578#define XXSPLTIB   (OPCD(60) | (360 << 1) | 1) /* v3.00, force tx=1 */
579
580#define MFVSRD     (XO31(51) | 1)   /* v2.07, force sx=1 */
581#define MFVSRWZ    (XO31(115) | 1)  /* v2.07, force sx=1 */
582#define MTVSRD     (XO31(179) | 1)  /* v2.07, force tx=1 */
583#define MTVSRWZ    (XO31(243) | 1)  /* v2.07, force tx=1 */
584#define MTVSRDD    (XO31(435) | 1)  /* v3.00, force tx=1 */
585#define MTVSRWS    (XO31(403) | 1)  /* v3.00, force tx=1 */
586
587#define RT(r) ((r)<<21)
588#define RS(r) ((r)<<21)
589#define RA(r) ((r)<<16)
590#define RB(r) ((r)<<11)
591#define TO(t) ((t)<<21)
592#define SH(s) ((s)<<11)
593#define MB(b) ((b)<<6)
594#define ME(e) ((e)<<1)
595#define BO(o) ((o)<<21)
596#define MB64(b) ((b)<<5)
597#define FXM(b) (1 << (19 - (b)))
598
599#define VRT(r)  (((r) & 31) << 21)
600#define VRA(r)  (((r) & 31) << 16)
601#define VRB(r)  (((r) & 31) << 11)
602#define VRC(r)  (((r) & 31) <<  6)
603
604#define LK    1
605
606#define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
607#define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
608#define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
609#define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
610
611#define BF(n)    ((n)<<23)
612#define BI(n, c) (((c)+((n)*4))<<16)
613#define BT(n, c) (((c)+((n)*4))<<21)
614#define BA(n, c) (((c)+((n)*4))<<16)
615#define BB(n, c) (((c)+((n)*4))<<11)
616#define BC_(n, c) (((c)+((n)*4))<<6)
617
618#define BO_COND_TRUE  BO(12)
619#define BO_COND_FALSE BO( 4)
620#define BO_ALWAYS     BO(20)
621
622enum {
623    CR_LT,
624    CR_GT,
625    CR_EQ,
626    CR_SO
627};
628
629static const uint32_t tcg_to_bc[] = {
630    [TCG_COND_EQ]  = BC | BI(7, CR_EQ) | BO_COND_TRUE,
631    [TCG_COND_NE]  = BC | BI(7, CR_EQ) | BO_COND_FALSE,
632    [TCG_COND_LT]  = BC | BI(7, CR_LT) | BO_COND_TRUE,
633    [TCG_COND_GE]  = BC | BI(7, CR_LT) | BO_COND_FALSE,
634    [TCG_COND_LE]  = BC | BI(7, CR_GT) | BO_COND_FALSE,
635    [TCG_COND_GT]  = BC | BI(7, CR_GT) | BO_COND_TRUE,
636    [TCG_COND_LTU] = BC | BI(7, CR_LT) | BO_COND_TRUE,
637    [TCG_COND_GEU] = BC | BI(7, CR_LT) | BO_COND_FALSE,
638    [TCG_COND_LEU] = BC | BI(7, CR_GT) | BO_COND_FALSE,
639    [TCG_COND_GTU] = BC | BI(7, CR_GT) | BO_COND_TRUE,
640};
641
642/* The low bit here is set if the RA and RB fields must be inverted.  */
643static const uint32_t tcg_to_isel[] = {
644    [TCG_COND_EQ]  = ISEL | BC_(7, CR_EQ),
645    [TCG_COND_NE]  = ISEL | BC_(7, CR_EQ) | 1,
646    [TCG_COND_LT]  = ISEL | BC_(7, CR_LT),
647    [TCG_COND_GE]  = ISEL | BC_(7, CR_LT) | 1,
648    [TCG_COND_LE]  = ISEL | BC_(7, CR_GT) | 1,
649    [TCG_COND_GT]  = ISEL | BC_(7, CR_GT),
650    [TCG_COND_LTU] = ISEL | BC_(7, CR_LT),
651    [TCG_COND_GEU] = ISEL | BC_(7, CR_LT) | 1,
652    [TCG_COND_LEU] = ISEL | BC_(7, CR_GT) | 1,
653    [TCG_COND_GTU] = ISEL | BC_(7, CR_GT),
654};
655
656static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
657                        intptr_t value, intptr_t addend)
658{
659    const tcg_insn_unit *target;
660    int16_t lo;
661    int32_t hi;
662
663    value += addend;
664    target = (const tcg_insn_unit *)value;
665
666    switch (type) {
667    case R_PPC_REL14:
668        return reloc_pc14(code_ptr, target);
669    case R_PPC_REL24:
670        return reloc_pc24(code_ptr, target);
671    case R_PPC_ADDR16:
672        /*
673         * We are (slightly) abusing this relocation type.  In particular,
674         * assert that the low 2 bits are zero, and do not modify them.
675         * That way we can use this with LD et al that have opcode bits
676         * in the low 2 bits of the insn.
677         */
678        if ((value & 3) || value != (int16_t)value) {
679            return false;
680        }
681        *code_ptr = (*code_ptr & ~0xfffc) | (value & 0xfffc);
682        break;
683    case R_PPC_ADDR32:
684        /*
685         * We are abusing this relocation type.  Again, this points to
686         * a pair of insns, lis + load.  This is an absolute address
687         * relocation for PPC32 so the lis cannot be removed.
688         */
689        lo = value;
690        hi = value - lo;
691        if (hi + lo != value) {
692            return false;
693        }
694        code_ptr[0] = deposit32(code_ptr[0], 0, 16, hi >> 16);
695        code_ptr[1] = deposit32(code_ptr[1], 0, 16, lo);
696        break;
697    default:
698        g_assert_not_reached();
699    }
700    return true;
701}
702
703static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
704                             TCGReg base, tcg_target_long offset);
705
706static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
707{
708    if (ret == arg) {
709        return true;
710    }
711    switch (type) {
712    case TCG_TYPE_I64:
713        tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
714        /* fallthru */
715    case TCG_TYPE_I32:
716        if (ret < TCG_REG_V0) {
717            if (arg < TCG_REG_V0) {
718                tcg_out32(s, OR | SAB(arg, ret, arg));
719                break;
720            } else if (have_isa_2_07) {
721                tcg_out32(s, (type == TCG_TYPE_I32 ? MFVSRWZ : MFVSRD)
722                          | VRT(arg) | RA(ret));
723                break;
724            } else {
725                /* Altivec does not support vector->integer moves.  */
726                return false;
727            }
728        } else if (arg < TCG_REG_V0) {
729            if (have_isa_2_07) {
730                tcg_out32(s, (type == TCG_TYPE_I32 ? MTVSRWZ : MTVSRD)
731                          | VRT(ret) | RA(arg));
732                break;
733            } else {
734                /* Altivec does not support integer->vector moves.  */
735                return false;
736            }
737        }
738        /* fallthru */
739    case TCG_TYPE_V64:
740    case TCG_TYPE_V128:
741        tcg_debug_assert(ret >= TCG_REG_V0 && arg >= TCG_REG_V0);
742        tcg_out32(s, VOR | VRT(ret) | VRA(arg) | VRB(arg));
743        break;
744    default:
745        g_assert_not_reached();
746    }
747    return true;
748}
749
750static inline void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs,
751                               int sh, int mb)
752{
753    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
754    sh = SH(sh & 0x1f) | (((sh >> 5) & 1) << 1);
755    mb = MB64((mb >> 5) | ((mb << 1) & 0x3f));
756    tcg_out32(s, op | RA(ra) | RS(rs) | sh | mb);
757}
758
759static inline void tcg_out_rlw(TCGContext *s, int op, TCGReg ra, TCGReg rs,
760                               int sh, int mb, int me)
761{
762    tcg_out32(s, op | RA(ra) | RS(rs) | SH(sh) | MB(mb) | ME(me));
763}
764
765static inline void tcg_out_ext8s(TCGContext *s, TCGReg dst, TCGReg src)
766{
767    tcg_out32(s, EXTSB | RA(dst) | RS(src));
768}
769
770static inline void tcg_out_ext16s(TCGContext *s, TCGReg dst, TCGReg src)
771{
772    tcg_out32(s, EXTSH | RA(dst) | RS(src));
773}
774
775static inline void tcg_out_ext16u(TCGContext *s, TCGReg dst, TCGReg src)
776{
777    tcg_out32(s, ANDI | SAI(src, dst, 0xffff));
778}
779
780static inline void tcg_out_ext32s(TCGContext *s, TCGReg dst, TCGReg src)
781{
782    tcg_out32(s, EXTSW | RA(dst) | RS(src));
783}
784
785static inline void tcg_out_ext32u(TCGContext *s, TCGReg dst, TCGReg src)
786{
787    tcg_out_rld(s, RLDICL, dst, src, 0, 32);
788}
789
790static inline void tcg_out_shli32(TCGContext *s, TCGReg dst, TCGReg src, int c)
791{
792    tcg_out_rlw(s, RLWINM, dst, src, c, 0, 31 - c);
793}
794
795static inline void tcg_out_shli64(TCGContext *s, TCGReg dst, TCGReg src, int c)
796{
797    tcg_out_rld(s, RLDICR, dst, src, c, 63 - c);
798}
799
800static inline void tcg_out_sari32(TCGContext *s, TCGReg dst, TCGReg src, int c)
801{
802    /* Limit immediate shift count lest we create an illegal insn.  */
803    tcg_out32(s, SRAWI | RA(dst) | RS(src) | SH(c & 31));
804}
805
806static inline void tcg_out_shri32(TCGContext *s, TCGReg dst, TCGReg src, int c)
807{
808    tcg_out_rlw(s, RLWINM, dst, src, 32 - c, c, 31);
809}
810
811static inline void tcg_out_shri64(TCGContext *s, TCGReg dst, TCGReg src, int c)
812{
813    tcg_out_rld(s, RLDICL, dst, src, 64 - c, c);
814}
815
816static inline void tcg_out_sari64(TCGContext *s, TCGReg dst, TCGReg src, int c)
817{
818    tcg_out32(s, SRADI | RA(dst) | RS(src) | SH(c & 0x1f) | ((c >> 4) & 2));
819}
820
821static void tcg_out_bswap16(TCGContext *s, TCGReg dst, TCGReg src, int flags)
822{
823    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
824
825    if (have_isa_3_10) {
826        tcg_out32(s, BRH | RA(dst) | RS(src));
827        if (flags & TCG_BSWAP_OS) {
828            tcg_out_ext16s(s, dst, dst);
829        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
830            tcg_out_ext16u(s, dst, dst);
831        }
832        return;
833    }
834
835    /*
836     * In the following,
837     *   dep(a, b, m) -> (a & ~m) | (b & m)
838     *
839     * Begin with:                              src = xxxxabcd
840     */
841    /* tmp = rol32(src, 24) & 0x000000ff            = 0000000c */
842    tcg_out_rlw(s, RLWINM, tmp, src, 24, 24, 31);
843    /* tmp = dep(tmp, rol32(src, 8), 0x0000ff00)    = 000000dc */
844    tcg_out_rlw(s, RLWIMI, tmp, src, 8, 16, 23);
845
846    if (flags & TCG_BSWAP_OS) {
847        tcg_out_ext16s(s, dst, tmp);
848    } else {
849        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
850    }
851}
852
853static void tcg_out_bswap32(TCGContext *s, TCGReg dst, TCGReg src, int flags)
854{
855    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
856
857    if (have_isa_3_10) {
858        tcg_out32(s, BRW | RA(dst) | RS(src));
859        if (flags & TCG_BSWAP_OS) {
860            tcg_out_ext32s(s, dst, dst);
861        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
862            tcg_out_ext32u(s, dst, dst);
863        }
864        return;
865    }
866
867    /*
868     * Stolen from gcc's builtin_bswap32.
869     * In the following,
870     *   dep(a, b, m) -> (a & ~m) | (b & m)
871     *
872     * Begin with:                              src = xxxxabcd
873     */
874    /* tmp = rol32(src, 8) & 0xffffffff             = 0000bcda */
875    tcg_out_rlw(s, RLWINM, tmp, src, 8, 0, 31);
876    /* tmp = dep(tmp, rol32(src, 24), 0xff000000)   = 0000dcda */
877    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 0, 7);
878    /* tmp = dep(tmp, rol32(src, 24), 0x0000ff00)   = 0000dcba */
879    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 16, 23);
880
881    if (flags & TCG_BSWAP_OS) {
882        tcg_out_ext32s(s, dst, tmp);
883    } else {
884        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
885    }
886}
887
888static void tcg_out_bswap64(TCGContext *s, TCGReg dst, TCGReg src)
889{
890    TCGReg t0 = dst == src ? TCG_REG_R0 : dst;
891    TCGReg t1 = dst == src ? dst : TCG_REG_R0;
892
893    if (have_isa_3_10) {
894        tcg_out32(s, BRD | RA(dst) | RS(src));
895        return;
896    }
897
898    /*
899     * In the following,
900     *   dep(a, b, m) -> (a & ~m) | (b & m)
901     *
902     * Begin with:                              src = abcdefgh
903     */
904    /* t0 = rol32(src, 8) & 0xffffffff              = 0000fghe */
905    tcg_out_rlw(s, RLWINM, t0, src, 8, 0, 31);
906    /* t0 = dep(t0, rol32(src, 24), 0xff000000)     = 0000hghe */
907    tcg_out_rlw(s, RLWIMI, t0, src, 24, 0, 7);
908    /* t0 = dep(t0, rol32(src, 24), 0x0000ff00)     = 0000hgfe */
909    tcg_out_rlw(s, RLWIMI, t0, src, 24, 16, 23);
910
911    /* t0 = rol64(t0, 32)                           = hgfe0000 */
912    tcg_out_rld(s, RLDICL, t0, t0, 32, 0);
913    /* t1 = rol64(src, 32)                          = efghabcd */
914    tcg_out_rld(s, RLDICL, t1, src, 32, 0);
915
916    /* t0 = dep(t0, rol32(t1, 24), 0xffffffff)      = hgfebcda */
917    tcg_out_rlw(s, RLWIMI, t0, t1, 8, 0, 31);
918    /* t0 = dep(t0, rol32(t1, 24), 0xff000000)      = hgfedcda */
919    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 0, 7);
920    /* t0 = dep(t0, rol32(t1, 24), 0x0000ff00)      = hgfedcba */
921    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 16, 23);
922
923    tcg_out_mov(s, TCG_TYPE_REG, dst, t0);
924}
925
926/* Emit a move into ret of arg, if it can be done in one insn.  */
927static bool tcg_out_movi_one(TCGContext *s, TCGReg ret, tcg_target_long arg)
928{
929    if (arg == (int16_t)arg) {
930        tcg_out32(s, ADDI | TAI(ret, 0, arg));
931        return true;
932    }
933    if (arg == (int32_t)arg && (arg & 0xffff) == 0) {
934        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
935        return true;
936    }
937    return false;
938}
939
940static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
941                             tcg_target_long arg, bool in_prologue)
942{
943    intptr_t tb_diff;
944    tcg_target_long tmp;
945    int shift;
946
947    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
948
949    if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
950        arg = (int32_t)arg;
951    }
952
953    /* Load 16-bit immediates with one insn.  */
954    if (tcg_out_movi_one(s, ret, arg)) {
955        return;
956    }
957
958    /* Load addresses within the TB with one insn.  */
959    tb_diff = tcg_tbrel_diff(s, (void *)arg);
960    if (!in_prologue && USE_REG_TB && tb_diff == (int16_t)tb_diff) {
961        tcg_out32(s, ADDI | TAI(ret, TCG_REG_TB, tb_diff));
962        return;
963    }
964
965    /* Load 32-bit immediates with two insns.  Note that we've already
966       eliminated bare ADDIS, so we know both insns are required.  */
967    if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) {
968        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
969        tcg_out32(s, ORI | SAI(ret, ret, arg));
970        return;
971    }
972    if (arg == (uint32_t)arg && !(arg & 0x8000)) {
973        tcg_out32(s, ADDI | TAI(ret, 0, arg));
974        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
975        return;
976    }
977
978    /* Load masked 16-bit value.  */
979    if (arg > 0 && (arg & 0x8000)) {
980        tmp = arg | 0x7fff;
981        if ((tmp & (tmp + 1)) == 0) {
982            int mb = clz64(tmp + 1) + 1;
983            tcg_out32(s, ADDI | TAI(ret, 0, arg));
984            tcg_out_rld(s, RLDICL, ret, ret, 0, mb);
985            return;
986        }
987    }
988
989    /* Load common masks with 2 insns.  */
990    shift = ctz64(arg);
991    tmp = arg >> shift;
992    if (tmp == (int16_t)tmp) {
993        tcg_out32(s, ADDI | TAI(ret, 0, tmp));
994        tcg_out_shli64(s, ret, ret, shift);
995        return;
996    }
997    shift = clz64(arg);
998    if (tcg_out_movi_one(s, ret, arg << shift)) {
999        tcg_out_shri64(s, ret, ret, shift);
1000        return;
1001    }
1002
1003    /* Load addresses within 2GB of TB with 2 (or rarely 3) insns.  */
1004    if (!in_prologue && USE_REG_TB && tb_diff == (int32_t)tb_diff) {
1005        tcg_out_mem_long(s, ADDI, ADD, ret, TCG_REG_TB, tb_diff);
1006        return;
1007    }
1008
1009    /* Use the constant pool, if possible.  */
1010    if (!in_prologue && USE_REG_TB) {
1011        new_pool_label(s, arg, R_PPC_ADDR16, s->code_ptr,
1012                       tcg_tbrel_diff(s, NULL));
1013        tcg_out32(s, LD | TAI(ret, TCG_REG_TB, 0));
1014        return;
1015    }
1016
1017    tmp = arg >> 31 >> 1;
1018    tcg_out_movi(s, TCG_TYPE_I32, ret, tmp);
1019    if (tmp) {
1020        tcg_out_shli64(s, ret, ret, 32);
1021    }
1022    if (arg & 0xffff0000) {
1023        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
1024    }
1025    if (arg & 0xffff) {
1026        tcg_out32(s, ORI | SAI(ret, ret, arg));
1027    }
1028}
1029
1030static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
1031                             TCGReg ret, int64_t val)
1032{
1033    uint32_t load_insn;
1034    int rel, low;
1035    intptr_t add;
1036
1037    switch (vece) {
1038    case MO_8:
1039        low = (int8_t)val;
1040        if (low >= -16 && low < 16) {
1041            tcg_out32(s, VSPLTISB | VRT(ret) | ((val & 31) << 16));
1042            return;
1043        }
1044        if (have_isa_3_00) {
1045            tcg_out32(s, XXSPLTIB | VRT(ret) | ((val & 0xff) << 11));
1046            return;
1047        }
1048        break;
1049
1050    case MO_16:
1051        low = (int16_t)val;
1052        if (low >= -16 && low < 16) {
1053            tcg_out32(s, VSPLTISH | VRT(ret) | ((val & 31) << 16));
1054            return;
1055        }
1056        break;
1057
1058    case MO_32:
1059        low = (int32_t)val;
1060        if (low >= -16 && low < 16) {
1061            tcg_out32(s, VSPLTISW | VRT(ret) | ((val & 31) << 16));
1062            return;
1063        }
1064        break;
1065    }
1066
1067    /*
1068     * Otherwise we must load the value from the constant pool.
1069     */
1070    if (USE_REG_TB) {
1071        rel = R_PPC_ADDR16;
1072        add = tcg_tbrel_diff(s, NULL);
1073    } else {
1074        rel = R_PPC_ADDR32;
1075        add = 0;
1076    }
1077
1078    if (have_vsx) {
1079        load_insn = type == TCG_TYPE_V64 ? LXSDX : LXVDSX;
1080        load_insn |= VRT(ret) | RB(TCG_REG_TMP1);
1081        if (TCG_TARGET_REG_BITS == 64) {
1082            new_pool_label(s, val, rel, s->code_ptr, add);
1083        } else {
1084            new_pool_l2(s, rel, s->code_ptr, add, val >> 32, val);
1085        }
1086    } else {
1087        load_insn = LVX | VRT(ret) | RB(TCG_REG_TMP1);
1088        if (TCG_TARGET_REG_BITS == 64) {
1089            new_pool_l2(s, rel, s->code_ptr, add, val, val);
1090        } else {
1091            new_pool_l4(s, rel, s->code_ptr, add,
1092                        val >> 32, val, val >> 32, val);
1093        }
1094    }
1095
1096    if (USE_REG_TB) {
1097        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, 0, 0));
1098        load_insn |= RA(TCG_REG_TB);
1099    } else {
1100        tcg_out32(s, ADDIS | TAI(TCG_REG_TMP1, 0, 0));
1101        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, TCG_REG_TMP1, 0));
1102    }
1103    tcg_out32(s, load_insn);
1104}
1105
1106static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,
1107                         tcg_target_long arg)
1108{
1109    switch (type) {
1110    case TCG_TYPE_I32:
1111    case TCG_TYPE_I64:
1112        tcg_debug_assert(ret < TCG_REG_V0);
1113        tcg_out_movi_int(s, type, ret, arg, false);
1114        break;
1115
1116    default:
1117        g_assert_not_reached();
1118    }
1119}
1120
1121static bool mask_operand(uint32_t c, int *mb, int *me)
1122{
1123    uint32_t lsb, test;
1124
1125    /* Accept a bit pattern like:
1126           0....01....1
1127           1....10....0
1128           0..01..10..0
1129       Keep track of the transitions.  */
1130    if (c == 0 || c == -1) {
1131        return false;
1132    }
1133    test = c;
1134    lsb = test & -test;
1135    test += lsb;
1136    if (test & (test - 1)) {
1137        return false;
1138    }
1139
1140    *me = clz32(lsb);
1141    *mb = test ? clz32(test & -test) + 1 : 0;
1142    return true;
1143}
1144
1145static bool mask64_operand(uint64_t c, int *mb, int *me)
1146{
1147    uint64_t lsb;
1148
1149    if (c == 0) {
1150        return false;
1151    }
1152
1153    lsb = c & -c;
1154    /* Accept 1..10..0.  */
1155    if (c == -lsb) {
1156        *mb = 0;
1157        *me = clz64(lsb);
1158        return true;
1159    }
1160    /* Accept 0..01..1.  */
1161    if (lsb == 1 && (c & (c + 1)) == 0) {
1162        *mb = clz64(c + 1) + 1;
1163        *me = 63;
1164        return true;
1165    }
1166    return false;
1167}
1168
1169static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1170{
1171    int mb, me;
1172
1173    if (mask_operand(c, &mb, &me)) {
1174        tcg_out_rlw(s, RLWINM, dst, src, 0, mb, me);
1175    } else if ((c & 0xffff) == c) {
1176        tcg_out32(s, ANDI | SAI(src, dst, c));
1177        return;
1178    } else if ((c & 0xffff0000) == c) {
1179        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1180        return;
1181    } else {
1182        tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R0, c);
1183        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1184    }
1185}
1186
1187static void tcg_out_andi64(TCGContext *s, TCGReg dst, TCGReg src, uint64_t c)
1188{
1189    int mb, me;
1190
1191    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1192    if (mask64_operand(c, &mb, &me)) {
1193        if (mb == 0) {
1194            tcg_out_rld(s, RLDICR, dst, src, 0, me);
1195        } else {
1196            tcg_out_rld(s, RLDICL, dst, src, 0, mb);
1197        }
1198    } else if ((c & 0xffff) == c) {
1199        tcg_out32(s, ANDI | SAI(src, dst, c));
1200        return;
1201    } else if ((c & 0xffff0000) == c) {
1202        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1203        return;
1204    } else {
1205        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, c);
1206        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1207    }
1208}
1209
1210static void tcg_out_zori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c,
1211                           int op_lo, int op_hi)
1212{
1213    if (c >> 16) {
1214        tcg_out32(s, op_hi | SAI(src, dst, c >> 16));
1215        src = dst;
1216    }
1217    if (c & 0xffff) {
1218        tcg_out32(s, op_lo | SAI(src, dst, c));
1219        src = dst;
1220    }
1221}
1222
1223static void tcg_out_ori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1224{
1225    tcg_out_zori32(s, dst, src, c, ORI, ORIS);
1226}
1227
1228static void tcg_out_xori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1229{
1230    tcg_out_zori32(s, dst, src, c, XORI, XORIS);
1231}
1232
1233static void tcg_out_b(TCGContext *s, int mask, const tcg_insn_unit *target)
1234{
1235    ptrdiff_t disp = tcg_pcrel_diff(s, target);
1236    if (in_range_b(disp)) {
1237        tcg_out32(s, B | (disp & 0x3fffffc) | mask);
1238    } else {
1239        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, (uintptr_t)target);
1240        tcg_out32(s, MTSPR | RS(TCG_REG_R0) | CTR);
1241        tcg_out32(s, BCCTR | BO_ALWAYS | mask);
1242    }
1243}
1244
1245static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
1246                             TCGReg base, tcg_target_long offset)
1247{
1248    tcg_target_long orig = offset, l0, l1, extra = 0, align = 0;
1249    bool is_int_store = false;
1250    TCGReg rs = TCG_REG_TMP1;
1251
1252    switch (opi) {
1253    case LD: case LWA:
1254        align = 3;
1255        /* FALLTHRU */
1256    default:
1257        if (rt > TCG_REG_R0 && rt < TCG_REG_V0) {
1258            rs = rt;
1259            break;
1260        }
1261        break;
1262    case LXSD:
1263    case STXSD:
1264        align = 3;
1265        break;
1266    case LXV:
1267    case STXV:
1268        align = 15;
1269        break;
1270    case STD:
1271        align = 3;
1272        /* FALLTHRU */
1273    case STB: case STH: case STW:
1274        is_int_store = true;
1275        break;
1276    }
1277
1278    /* For unaligned, or very large offsets, use the indexed form.  */
1279    if (offset & align || offset != (int32_t)offset || opi == 0) {
1280        if (rs == base) {
1281            rs = TCG_REG_R0;
1282        }
1283        tcg_debug_assert(!is_int_store || rs != rt);
1284        tcg_out_movi(s, TCG_TYPE_PTR, rs, orig);
1285        tcg_out32(s, opx | TAB(rt & 31, base, rs));
1286        return;
1287    }
1288
1289    l0 = (int16_t)offset;
1290    offset = (offset - l0) >> 16;
1291    l1 = (int16_t)offset;
1292
1293    if (l1 < 0 && orig >= 0) {
1294        extra = 0x4000;
1295        l1 = (int16_t)(offset - 0x4000);
1296    }
1297    if (l1) {
1298        tcg_out32(s, ADDIS | TAI(rs, base, l1));
1299        base = rs;
1300    }
1301    if (extra) {
1302        tcg_out32(s, ADDIS | TAI(rs, base, extra));
1303        base = rs;
1304    }
1305    if (opi != ADDI || base != rt || l0 != 0) {
1306        tcg_out32(s, opi | TAI(rt & 31, base, l0));
1307    }
1308}
1309
1310static void tcg_out_vsldoi(TCGContext *s, TCGReg ret,
1311                           TCGReg va, TCGReg vb, int shb)
1312{
1313    tcg_out32(s, VSLDOI | VRT(ret) | VRA(va) | VRB(vb) | (shb << 6));
1314}
1315
1316static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
1317                       TCGReg base, intptr_t offset)
1318{
1319    int shift;
1320
1321    switch (type) {
1322    case TCG_TYPE_I32:
1323        if (ret < TCG_REG_V0) {
1324            tcg_out_mem_long(s, LWZ, LWZX, ret, base, offset);
1325            break;
1326        }
1327        if (have_isa_2_07 && have_vsx) {
1328            tcg_out_mem_long(s, 0, LXSIWZX, ret, base, offset);
1329            break;
1330        }
1331        tcg_debug_assert((offset & 3) == 0);
1332        tcg_out_mem_long(s, 0, LVEWX, ret, base, offset);
1333        shift = (offset - 4) & 0xc;
1334        if (shift) {
1335            tcg_out_vsldoi(s, ret, ret, ret, shift);
1336        }
1337        break;
1338    case TCG_TYPE_I64:
1339        if (ret < TCG_REG_V0) {
1340            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1341            tcg_out_mem_long(s, LD, LDX, ret, base, offset);
1342            break;
1343        }
1344        /* fallthru */
1345    case TCG_TYPE_V64:
1346        tcg_debug_assert(ret >= TCG_REG_V0);
1347        if (have_vsx) {
1348            tcg_out_mem_long(s, have_isa_3_00 ? LXSD : 0, LXSDX,
1349                             ret, base, offset);
1350            break;
1351        }
1352        tcg_debug_assert((offset & 7) == 0);
1353        tcg_out_mem_long(s, 0, LVX, ret, base, offset & -16);
1354        if (offset & 8) {
1355            tcg_out_vsldoi(s, ret, ret, ret, 8);
1356        }
1357        break;
1358    case TCG_TYPE_V128:
1359        tcg_debug_assert(ret >= TCG_REG_V0);
1360        tcg_debug_assert((offset & 15) == 0);
1361        tcg_out_mem_long(s, have_isa_3_00 ? LXV : 0,
1362                         LVX, ret, base, offset);
1363        break;
1364    default:
1365        g_assert_not_reached();
1366    }
1367}
1368
1369static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
1370                              TCGReg base, intptr_t offset)
1371{
1372    int shift;
1373
1374    switch (type) {
1375    case TCG_TYPE_I32:
1376        if (arg < TCG_REG_V0) {
1377            tcg_out_mem_long(s, STW, STWX, arg, base, offset);
1378            break;
1379        }
1380        if (have_isa_2_07 && have_vsx) {
1381            tcg_out_mem_long(s, 0, STXSIWX, arg, base, offset);
1382            break;
1383        }
1384        assert((offset & 3) == 0);
1385        tcg_debug_assert((offset & 3) == 0);
1386        shift = (offset - 4) & 0xc;
1387        if (shift) {
1388            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, shift);
1389            arg = TCG_VEC_TMP1;
1390        }
1391        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1392        break;
1393    case TCG_TYPE_I64:
1394        if (arg < TCG_REG_V0) {
1395            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1396            tcg_out_mem_long(s, STD, STDX, arg, base, offset);
1397            break;
1398        }
1399        /* fallthru */
1400    case TCG_TYPE_V64:
1401        tcg_debug_assert(arg >= TCG_REG_V0);
1402        if (have_vsx) {
1403            tcg_out_mem_long(s, have_isa_3_00 ? STXSD : 0,
1404                             STXSDX, arg, base, offset);
1405            break;
1406        }
1407        tcg_debug_assert((offset & 7) == 0);
1408        if (offset & 8) {
1409            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, 8);
1410            arg = TCG_VEC_TMP1;
1411        }
1412        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1413        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset + 4);
1414        break;
1415    case TCG_TYPE_V128:
1416        tcg_debug_assert(arg >= TCG_REG_V0);
1417        tcg_out_mem_long(s, have_isa_3_00 ? STXV : 0,
1418                         STVX, arg, base, offset);
1419        break;
1420    default:
1421        g_assert_not_reached();
1422    }
1423}
1424
1425static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1426                               TCGReg base, intptr_t ofs)
1427{
1428    return false;
1429}
1430
1431static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2,
1432                        int const_arg2, int cr, TCGType type)
1433{
1434    int imm;
1435    uint32_t op;
1436
1437    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1438
1439    /* Simplify the comparisons below wrt CMPI.  */
1440    if (type == TCG_TYPE_I32) {
1441        arg2 = (int32_t)arg2;
1442    }
1443
1444    switch (cond) {
1445    case TCG_COND_EQ:
1446    case TCG_COND_NE:
1447        if (const_arg2) {
1448            if ((int16_t) arg2 == arg2) {
1449                op = CMPI;
1450                imm = 1;
1451                break;
1452            } else if ((uint16_t) arg2 == arg2) {
1453                op = CMPLI;
1454                imm = 1;
1455                break;
1456            }
1457        }
1458        op = CMPL;
1459        imm = 0;
1460        break;
1461
1462    case TCG_COND_LT:
1463    case TCG_COND_GE:
1464    case TCG_COND_LE:
1465    case TCG_COND_GT:
1466        if (const_arg2) {
1467            if ((int16_t) arg2 == arg2) {
1468                op = CMPI;
1469                imm = 1;
1470                break;
1471            }
1472        }
1473        op = CMP;
1474        imm = 0;
1475        break;
1476
1477    case TCG_COND_LTU:
1478    case TCG_COND_GEU:
1479    case TCG_COND_LEU:
1480    case TCG_COND_GTU:
1481        if (const_arg2) {
1482            if ((uint16_t) arg2 == arg2) {
1483                op = CMPLI;
1484                imm = 1;
1485                break;
1486            }
1487        }
1488        op = CMPL;
1489        imm = 0;
1490        break;
1491
1492    default:
1493        tcg_abort();
1494    }
1495    op |= BF(cr) | ((type == TCG_TYPE_I64) << 21);
1496
1497    if (imm) {
1498        tcg_out32(s, op | RA(arg1) | (arg2 & 0xffff));
1499    } else {
1500        if (const_arg2) {
1501            tcg_out_movi(s, type, TCG_REG_R0, arg2);
1502            arg2 = TCG_REG_R0;
1503        }
1504        tcg_out32(s, op | RA(arg1) | RB(arg2));
1505    }
1506}
1507
1508static void tcg_out_setcond_eq0(TCGContext *s, TCGType type,
1509                                TCGReg dst, TCGReg src)
1510{
1511    if (type == TCG_TYPE_I32) {
1512        tcg_out32(s, CNTLZW | RS(src) | RA(dst));
1513        tcg_out_shri32(s, dst, dst, 5);
1514    } else {
1515        tcg_out32(s, CNTLZD | RS(src) | RA(dst));
1516        tcg_out_shri64(s, dst, dst, 6);
1517    }
1518}
1519
1520static void tcg_out_setcond_ne0(TCGContext *s, TCGReg dst, TCGReg src)
1521{
1522    /* X != 0 implies X + -1 generates a carry.  Extra addition
1523       trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C.  */
1524    if (dst != src) {
1525        tcg_out32(s, ADDIC | TAI(dst, src, -1));
1526        tcg_out32(s, SUBFE | TAB(dst, dst, src));
1527    } else {
1528        tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1));
1529        tcg_out32(s, SUBFE | TAB(dst, TCG_REG_R0, src));
1530    }
1531}
1532
1533static TCGReg tcg_gen_setcond_xor(TCGContext *s, TCGReg arg1, TCGArg arg2,
1534                                  bool const_arg2)
1535{
1536    if (const_arg2) {
1537        if ((uint32_t)arg2 == arg2) {
1538            tcg_out_xori32(s, TCG_REG_R0, arg1, arg2);
1539        } else {
1540            tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, arg2);
1541            tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, TCG_REG_R0));
1542        }
1543    } else {
1544        tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, arg2));
1545    }
1546    return TCG_REG_R0;
1547}
1548
1549static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
1550                            TCGArg arg0, TCGArg arg1, TCGArg arg2,
1551                            int const_arg2)
1552{
1553    int crop, sh;
1554
1555    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1556
1557    /* Ignore high bits of a potential constant arg2.  */
1558    if (type == TCG_TYPE_I32) {
1559        arg2 = (uint32_t)arg2;
1560    }
1561
1562    /* Handle common and trivial cases before handling anything else.  */
1563    if (arg2 == 0) {
1564        switch (cond) {
1565        case TCG_COND_EQ:
1566            tcg_out_setcond_eq0(s, type, arg0, arg1);
1567            return;
1568        case TCG_COND_NE:
1569            if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1570                tcg_out_ext32u(s, TCG_REG_R0, arg1);
1571                arg1 = TCG_REG_R0;
1572            }
1573            tcg_out_setcond_ne0(s, arg0, arg1);
1574            return;
1575        case TCG_COND_GE:
1576            tcg_out32(s, NOR | SAB(arg1, arg0, arg1));
1577            arg1 = arg0;
1578            /* FALLTHRU */
1579        case TCG_COND_LT:
1580            /* Extract the sign bit.  */
1581            if (type == TCG_TYPE_I32) {
1582                tcg_out_shri32(s, arg0, arg1, 31);
1583            } else {
1584                tcg_out_shri64(s, arg0, arg1, 63);
1585            }
1586            return;
1587        default:
1588            break;
1589        }
1590    }
1591
1592    /* If we have ISEL, we can implement everything with 3 or 4 insns.
1593       All other cases below are also at least 3 insns, so speed up the
1594       code generator by not considering them and always using ISEL.  */
1595    if (have_isel) {
1596        int isel, tab;
1597
1598        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1599
1600        isel = tcg_to_isel[cond];
1601
1602        tcg_out_movi(s, type, arg0, 1);
1603        if (isel & 1) {
1604            /* arg0 = (bc ? 0 : 1) */
1605            tab = TAB(arg0, 0, arg0);
1606            isel &= ~1;
1607        } else {
1608            /* arg0 = (bc ? 1 : 0) */
1609            tcg_out_movi(s, type, TCG_REG_R0, 0);
1610            tab = TAB(arg0, arg0, TCG_REG_R0);
1611        }
1612        tcg_out32(s, isel | tab);
1613        return;
1614    }
1615
1616    switch (cond) {
1617    case TCG_COND_EQ:
1618        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1619        tcg_out_setcond_eq0(s, type, arg0, arg1);
1620        return;
1621
1622    case TCG_COND_NE:
1623        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1624        /* Discard the high bits only once, rather than both inputs.  */
1625        if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1626            tcg_out_ext32u(s, TCG_REG_R0, arg1);
1627            arg1 = TCG_REG_R0;
1628        }
1629        tcg_out_setcond_ne0(s, arg0, arg1);
1630        return;
1631
1632    case TCG_COND_GT:
1633    case TCG_COND_GTU:
1634        sh = 30;
1635        crop = 0;
1636        goto crtest;
1637
1638    case TCG_COND_LT:
1639    case TCG_COND_LTU:
1640        sh = 29;
1641        crop = 0;
1642        goto crtest;
1643
1644    case TCG_COND_GE:
1645    case TCG_COND_GEU:
1646        sh = 31;
1647        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_LT) | BB(7, CR_LT);
1648        goto crtest;
1649
1650    case TCG_COND_LE:
1651    case TCG_COND_LEU:
1652        sh = 31;
1653        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_GT) | BB(7, CR_GT);
1654    crtest:
1655        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1656        if (crop) {
1657            tcg_out32(s, crop);
1658        }
1659        tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1660        tcg_out_rlw(s, RLWINM, arg0, TCG_REG_R0, sh, 31, 31);
1661        break;
1662
1663    default:
1664        tcg_abort();
1665    }
1666}
1667
1668static void tcg_out_bc(TCGContext *s, int bc, TCGLabel *l)
1669{
1670    if (l->has_value) {
1671        bc |= reloc_pc14_val(tcg_splitwx_to_rx(s->code_ptr), l->u.value_ptr);
1672    } else {
1673        tcg_out_reloc(s, s->code_ptr, R_PPC_REL14, l, 0);
1674    }
1675    tcg_out32(s, bc);
1676}
1677
1678static void tcg_out_brcond(TCGContext *s, TCGCond cond,
1679                           TCGArg arg1, TCGArg arg2, int const_arg2,
1680                           TCGLabel *l, TCGType type)
1681{
1682    tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1683    tcg_out_bc(s, tcg_to_bc[cond], l);
1684}
1685
1686static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
1687                            TCGArg dest, TCGArg c1, TCGArg c2, TCGArg v1,
1688                            TCGArg v2, bool const_c2)
1689{
1690    /* If for some reason both inputs are zero, don't produce bad code.  */
1691    if (v1 == 0 && v2 == 0) {
1692        tcg_out_movi(s, type, dest, 0);
1693        return;
1694    }
1695
1696    tcg_out_cmp(s, cond, c1, c2, const_c2, 7, type);
1697
1698    if (have_isel) {
1699        int isel = tcg_to_isel[cond];
1700
1701        /* Swap the V operands if the operation indicates inversion.  */
1702        if (isel & 1) {
1703            int t = v1;
1704            v1 = v2;
1705            v2 = t;
1706            isel &= ~1;
1707        }
1708        /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand.  */
1709        if (v2 == 0) {
1710            tcg_out_movi(s, type, TCG_REG_R0, 0);
1711        }
1712        tcg_out32(s, isel | TAB(dest, v1, v2));
1713    } else {
1714        if (dest == v2) {
1715            cond = tcg_invert_cond(cond);
1716            v2 = v1;
1717        } else if (dest != v1) {
1718            if (v1 == 0) {
1719                tcg_out_movi(s, type, dest, 0);
1720            } else {
1721                tcg_out_mov(s, type, dest, v1);
1722            }
1723        }
1724        /* Branch forward over one insn */
1725        tcg_out32(s, tcg_to_bc[cond] | 8);
1726        if (v2 == 0) {
1727            tcg_out_movi(s, type, dest, 0);
1728        } else {
1729            tcg_out_mov(s, type, dest, v2);
1730        }
1731    }
1732}
1733
1734static void tcg_out_cntxz(TCGContext *s, TCGType type, uint32_t opc,
1735                          TCGArg a0, TCGArg a1, TCGArg a2, bool const_a2)
1736{
1737    if (const_a2 && a2 == (type == TCG_TYPE_I32 ? 32 : 64)) {
1738        tcg_out32(s, opc | RA(a0) | RS(a1));
1739    } else {
1740        tcg_out_cmp(s, TCG_COND_EQ, a1, 0, 1, 7, type);
1741        /* Note that the only other valid constant for a2 is 0.  */
1742        if (have_isel) {
1743            tcg_out32(s, opc | RA(TCG_REG_R0) | RS(a1));
1744            tcg_out32(s, tcg_to_isel[TCG_COND_EQ] | TAB(a0, a2, TCG_REG_R0));
1745        } else if (!const_a2 && a0 == a2) {
1746            tcg_out32(s, tcg_to_bc[TCG_COND_EQ] | 8);
1747            tcg_out32(s, opc | RA(a0) | RS(a1));
1748        } else {
1749            tcg_out32(s, opc | RA(a0) | RS(a1));
1750            tcg_out32(s, tcg_to_bc[TCG_COND_NE] | 8);
1751            if (const_a2) {
1752                tcg_out_movi(s, type, a0, 0);
1753            } else {
1754                tcg_out_mov(s, type, a0, a2);
1755            }
1756        }
1757    }
1758}
1759
1760static void tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1761                         const int *const_args)
1762{
1763    static const struct { uint8_t bit1, bit2; } bits[] = {
1764        [TCG_COND_LT ] = { CR_LT, CR_LT },
1765        [TCG_COND_LE ] = { CR_LT, CR_GT },
1766        [TCG_COND_GT ] = { CR_GT, CR_GT },
1767        [TCG_COND_GE ] = { CR_GT, CR_LT },
1768        [TCG_COND_LTU] = { CR_LT, CR_LT },
1769        [TCG_COND_LEU] = { CR_LT, CR_GT },
1770        [TCG_COND_GTU] = { CR_GT, CR_GT },
1771        [TCG_COND_GEU] = { CR_GT, CR_LT },
1772    };
1773
1774    TCGCond cond = args[4], cond2;
1775    TCGArg al, ah, bl, bh;
1776    int blconst, bhconst;
1777    int op, bit1, bit2;
1778
1779    al = args[0];
1780    ah = args[1];
1781    bl = args[2];
1782    bh = args[3];
1783    blconst = const_args[2];
1784    bhconst = const_args[3];
1785
1786    switch (cond) {
1787    case TCG_COND_EQ:
1788        op = CRAND;
1789        goto do_equality;
1790    case TCG_COND_NE:
1791        op = CRNAND;
1792    do_equality:
1793        tcg_out_cmp(s, cond, al, bl, blconst, 6, TCG_TYPE_I32);
1794        tcg_out_cmp(s, cond, ah, bh, bhconst, 7, TCG_TYPE_I32);
1795        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1796        break;
1797
1798    case TCG_COND_LT:
1799    case TCG_COND_LE:
1800    case TCG_COND_GT:
1801    case TCG_COND_GE:
1802    case TCG_COND_LTU:
1803    case TCG_COND_LEU:
1804    case TCG_COND_GTU:
1805    case TCG_COND_GEU:
1806        bit1 = bits[cond].bit1;
1807        bit2 = bits[cond].bit2;
1808        op = (bit1 != bit2 ? CRANDC : CRAND);
1809        cond2 = tcg_unsigned_cond(cond);
1810
1811        tcg_out_cmp(s, cond, ah, bh, bhconst, 6, TCG_TYPE_I32);
1812        tcg_out_cmp(s, cond2, al, bl, blconst, 7, TCG_TYPE_I32);
1813        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, bit2));
1814        tcg_out32(s, CROR | BT(7, CR_EQ) | BA(6, bit1) | BB(7, CR_EQ));
1815        break;
1816
1817    default:
1818        tcg_abort();
1819    }
1820}
1821
1822static void tcg_out_setcond2(TCGContext *s, const TCGArg *args,
1823                             const int *const_args)
1824{
1825    tcg_out_cmp2(s, args + 1, const_args + 1);
1826    tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1827    tcg_out_rlw(s, RLWINM, args[0], TCG_REG_R0, 31, 31, 31);
1828}
1829
1830static void tcg_out_brcond2 (TCGContext *s, const TCGArg *args,
1831                             const int *const_args)
1832{
1833    tcg_out_cmp2(s, args, const_args);
1834    tcg_out_bc(s, BC | BI(7, CR_EQ) | BO_COND_TRUE, arg_label(args[5]));
1835}
1836
1837static void tcg_out_mb(TCGContext *s, TCGArg a0)
1838{
1839    uint32_t insn;
1840
1841    if (a0 & TCG_MO_ST_LD) {
1842        insn = HWSYNC;
1843    } else {
1844        insn = LWSYNC;
1845    }
1846
1847    tcg_out32(s, insn);
1848}
1849
1850void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_rx,
1851                              uintptr_t jmp_rw, uintptr_t addr)
1852{
1853    if (TCG_TARGET_REG_BITS == 64) {
1854        tcg_insn_unit i1, i2;
1855        intptr_t tb_diff = addr - tc_ptr;
1856        intptr_t br_diff = addr - (jmp_rx + 4);
1857        uint64_t pair;
1858
1859        /* This does not exercise the range of the branch, but we do
1860           still need to be able to load the new value of TCG_REG_TB.
1861           But this does still happen quite often.  */
1862        if (tb_diff == (int16_t)tb_diff) {
1863            i1 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, tb_diff);
1864            i2 = B | (br_diff & 0x3fffffc);
1865        } else {
1866            intptr_t lo = (int16_t)tb_diff;
1867            intptr_t hi = (int32_t)(tb_diff - lo);
1868            assert(tb_diff == hi + lo);
1869            i1 = ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, hi >> 16);
1870            i2 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, lo);
1871        }
1872#if HOST_BIG_ENDIAN
1873        pair = (uint64_t)i1 << 32 | i2;
1874#else
1875        pair = (uint64_t)i2 << 32 | i1;
1876#endif
1877
1878        /* As per the enclosing if, this is ppc64.  Avoid the _Static_assert
1879           within qatomic_set that would fail to build a ppc32 host.  */
1880        qatomic_set__nocheck((uint64_t *)jmp_rw, pair);
1881        flush_idcache_range(jmp_rx, jmp_rw, 8);
1882    } else {
1883        intptr_t diff = addr - jmp_rx;
1884        tcg_debug_assert(in_range_b(diff));
1885        qatomic_set((uint32_t *)jmp_rw, B | (diff & 0x3fffffc));
1886        flush_idcache_range(jmp_rx, jmp_rw, 4);
1887    }
1888}
1889
1890static void tcg_out_call_int(TCGContext *s, int lk,
1891                             const tcg_insn_unit *target)
1892{
1893#ifdef _CALL_AIX
1894    /* Look through the descriptor.  If the branch is in range, and we
1895       don't have to spend too much effort on building the toc.  */
1896    const void *tgt = ((const void * const *)target)[0];
1897    uintptr_t toc = ((const uintptr_t *)target)[1];
1898    intptr_t diff = tcg_pcrel_diff(s, tgt);
1899
1900    if (in_range_b(diff) && toc == (uint32_t)toc) {
1901        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, toc);
1902        tcg_out_b(s, lk, tgt);
1903    } else {
1904        /* Fold the low bits of the constant into the addresses below.  */
1905        intptr_t arg = (intptr_t)target;
1906        int ofs = (int16_t)arg;
1907
1908        if (ofs + 8 < 0x8000) {
1909            arg -= ofs;
1910        } else {
1911            ofs = 0;
1912        }
1913        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, arg);
1914        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_TMP1, ofs);
1915        tcg_out32(s, MTSPR | RA(TCG_REG_R0) | CTR);
1916        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_REG_TMP1, ofs + SZP);
1917        tcg_out32(s, BCCTR | BO_ALWAYS | lk);
1918    }
1919#elif defined(_CALL_ELF) && _CALL_ELF == 2
1920    intptr_t diff;
1921
1922    /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1923       address, which the callee uses to compute its TOC address.  */
1924    /* FIXME: when the branch is in range, we could avoid r12 load if we
1925       knew that the destination uses the same TOC, and what its local
1926       entry point offset is.  */
1927    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R12, (intptr_t)target);
1928
1929    diff = tcg_pcrel_diff(s, target);
1930    if (in_range_b(diff)) {
1931        tcg_out_b(s, lk, target);
1932    } else {
1933        tcg_out32(s, MTSPR | RS(TCG_REG_R12) | CTR);
1934        tcg_out32(s, BCCTR | BO_ALWAYS | lk);
1935    }
1936#else
1937    tcg_out_b(s, lk, target);
1938#endif
1939}
1940
1941static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target)
1942{
1943    tcg_out_call_int(s, LK, target);
1944}
1945
1946static const uint32_t qemu_ldx_opc[(MO_SSIZE + MO_BSWAP) + 1] = {
1947    [MO_UB] = LBZX,
1948    [MO_UW] = LHZX,
1949    [MO_UL] = LWZX,
1950    [MO_UQ] = LDX,
1951    [MO_SW] = LHAX,
1952    [MO_SL] = LWAX,
1953    [MO_BSWAP | MO_UB] = LBZX,
1954    [MO_BSWAP | MO_UW] = LHBRX,
1955    [MO_BSWAP | MO_UL] = LWBRX,
1956    [MO_BSWAP | MO_UQ] = LDBRX,
1957};
1958
1959static const uint32_t qemu_stx_opc[(MO_SIZE + MO_BSWAP) + 1] = {
1960    [MO_UB] = STBX,
1961    [MO_UW] = STHX,
1962    [MO_UL] = STWX,
1963    [MO_UQ] = STDX,
1964    [MO_BSWAP | MO_UB] = STBX,
1965    [MO_BSWAP | MO_UW] = STHBRX,
1966    [MO_BSWAP | MO_UL] = STWBRX,
1967    [MO_BSWAP | MO_UQ] = STDBRX,
1968};
1969
1970static const uint32_t qemu_exts_opc[4] = {
1971    EXTSB, EXTSH, EXTSW, 0
1972};
1973
1974#if defined (CONFIG_SOFTMMU)
1975/* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
1976 *                                 int mmu_idx, uintptr_t ra)
1977 */
1978static void * const qemu_ld_helpers[(MO_SIZE | MO_BSWAP) + 1] = {
1979    [MO_UB]   = helper_ret_ldub_mmu,
1980    [MO_LEUW] = helper_le_lduw_mmu,
1981    [MO_LEUL] = helper_le_ldul_mmu,
1982    [MO_LEUQ] = helper_le_ldq_mmu,
1983    [MO_BEUW] = helper_be_lduw_mmu,
1984    [MO_BEUL] = helper_be_ldul_mmu,
1985    [MO_BEUQ] = helper_be_ldq_mmu,
1986};
1987
1988/* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
1989 *                                 uintxx_t val, int mmu_idx, uintptr_t ra)
1990 */
1991static void * const qemu_st_helpers[(MO_SIZE | MO_BSWAP) + 1] = {
1992    [MO_UB]   = helper_ret_stb_mmu,
1993    [MO_LEUW] = helper_le_stw_mmu,
1994    [MO_LEUL] = helper_le_stl_mmu,
1995    [MO_LEUQ] = helper_le_stq_mmu,
1996    [MO_BEUW] = helper_be_stw_mmu,
1997    [MO_BEUL] = helper_be_stl_mmu,
1998    [MO_BEUQ] = helper_be_stq_mmu,
1999};
2000
2001/* We expect to use a 16-bit negative offset from ENV.  */
2002QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
2003QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -32768);
2004
2005/* Perform the TLB load and compare.  Places the result of the comparison
2006   in CR7, loads the addend of the TLB into R3, and returns the register
2007   containing the guest address (zero-extended into R4).  Clobbers R0 and R2. */
2008
2009static TCGReg tcg_out_tlb_read(TCGContext *s, MemOp opc,
2010                               TCGReg addrlo, TCGReg addrhi,
2011                               int mem_index, bool is_read)
2012{
2013    int cmp_off
2014        = (is_read
2015           ? offsetof(CPUTLBEntry, addr_read)
2016           : offsetof(CPUTLBEntry, addr_write));
2017    int fast_off = TLB_MASK_TABLE_OFS(mem_index);
2018    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
2019    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
2020    unsigned s_bits = opc & MO_SIZE;
2021    unsigned a_bits = get_alignment_bits(opc);
2022
2023    /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx].  */
2024    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_AREG0, mask_off);
2025    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R4, TCG_AREG0, table_off);
2026
2027    /* Extract the page index, shifted into place for tlb index.  */
2028    if (TCG_TARGET_REG_BITS == 32) {
2029        tcg_out_shri32(s, TCG_REG_TMP1, addrlo,
2030                       TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
2031    } else {
2032        tcg_out_shri64(s, TCG_REG_TMP1, addrlo,
2033                       TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
2034    }
2035    tcg_out32(s, AND | SAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_TMP1));
2036
2037    /* Load the TLB comparator.  */
2038    if (cmp_off == 0 && TCG_TARGET_REG_BITS >= TARGET_LONG_BITS) {
2039        uint32_t lxu = (TCG_TARGET_REG_BITS == 32 || TARGET_LONG_BITS == 32
2040                        ? LWZUX : LDUX);
2041        tcg_out32(s, lxu | TAB(TCG_REG_TMP1, TCG_REG_R3, TCG_REG_R4));
2042    } else {
2043        tcg_out32(s, ADD | TAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_R4));
2044        if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2045            tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP1, TCG_REG_R3, cmp_off + 4);
2046            tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R4, TCG_REG_R3, cmp_off);
2047        } else {
2048            tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP1, TCG_REG_R3, cmp_off);
2049        }
2050    }
2051
2052    /* Load the TLB addend for use on the fast path.  Do this asap
2053       to minimize any load use delay.  */
2054    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_REG_R3,
2055               offsetof(CPUTLBEntry, addend));
2056
2057    /* Clear the non-page, non-alignment bits from the address */
2058    if (TCG_TARGET_REG_BITS == 32) {
2059        /* We don't support unaligned accesses on 32-bits.
2060         * Preserve the bottom bits and thus trigger a comparison
2061         * failure on unaligned accesses.
2062         */
2063        if (a_bits < s_bits) {
2064            a_bits = s_bits;
2065        }
2066        tcg_out_rlw(s, RLWINM, TCG_REG_R0, addrlo, 0,
2067                    (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
2068    } else {
2069        TCGReg t = addrlo;
2070
2071        /* If the access is unaligned, we need to make sure we fail if we
2072         * cross a page boundary.  The trick is to add the access size-1
2073         * to the address before masking the low bits.  That will make the
2074         * address overflow to the next page if we cross a page boundary,
2075         * which will then force a mismatch of the TLB compare.
2076         */
2077        if (a_bits < s_bits) {
2078            unsigned a_mask = (1 << a_bits) - 1;
2079            unsigned s_mask = (1 << s_bits) - 1;
2080            tcg_out32(s, ADDI | TAI(TCG_REG_R0, t, s_mask - a_mask));
2081            t = TCG_REG_R0;
2082        }
2083
2084        /* Mask the address for the requested alignment.  */
2085        if (TARGET_LONG_BITS == 32) {
2086            tcg_out_rlw(s, RLWINM, TCG_REG_R0, t, 0,
2087                        (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
2088            /* Zero-extend the address for use in the final address.  */
2089            tcg_out_ext32u(s, TCG_REG_R4, addrlo);
2090            addrlo = TCG_REG_R4;
2091        } else if (a_bits == 0) {
2092            tcg_out_rld(s, RLDICR, TCG_REG_R0, t, 0, 63 - TARGET_PAGE_BITS);
2093        } else {
2094            tcg_out_rld(s, RLDICL, TCG_REG_R0, t,
2095                        64 - TARGET_PAGE_BITS, TARGET_PAGE_BITS - a_bits);
2096            tcg_out_rld(s, RLDICL, TCG_REG_R0, TCG_REG_R0, TARGET_PAGE_BITS, 0);
2097        }
2098    }
2099
2100    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2101        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
2102                    0, 7, TCG_TYPE_I32);
2103        tcg_out_cmp(s, TCG_COND_EQ, addrhi, TCG_REG_R4, 0, 6, TCG_TYPE_I32);
2104        tcg_out32(s, CRAND | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
2105    } else {
2106        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
2107                    0, 7, TCG_TYPE_TL);
2108    }
2109
2110    return addrlo;
2111}
2112
2113/* Record the context of a call to the out of line helper code for the slow
2114   path for a load or store, so that we can later generate the correct
2115   helper code.  */
2116static void add_qemu_ldst_label(TCGContext *s, bool is_ld, MemOpIdx oi,
2117                                TCGReg datalo_reg, TCGReg datahi_reg,
2118                                TCGReg addrlo_reg, TCGReg addrhi_reg,
2119                                tcg_insn_unit *raddr, tcg_insn_unit *lptr)
2120{
2121    TCGLabelQemuLdst *label = new_ldst_label(s);
2122
2123    label->is_ld = is_ld;
2124    label->oi = oi;
2125    label->datalo_reg = datalo_reg;
2126    label->datahi_reg = datahi_reg;
2127    label->addrlo_reg = addrlo_reg;
2128    label->addrhi_reg = addrhi_reg;
2129    label->raddr = tcg_splitwx_to_rx(raddr);
2130    label->label_ptr[0] = lptr;
2131}
2132
2133static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2134{
2135    MemOpIdx oi = lb->oi;
2136    MemOp opc = get_memop(oi);
2137    TCGReg hi, lo, arg = TCG_REG_R3;
2138
2139    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2140        return false;
2141    }
2142
2143    tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
2144
2145    lo = lb->addrlo_reg;
2146    hi = lb->addrhi_reg;
2147    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2148#ifdef TCG_TARGET_CALL_ALIGN_ARGS
2149        arg |= 1;
2150#endif
2151        tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2152        tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2153    } else {
2154        /* If the address needed to be zero-extended, we'll have already
2155           placed it in R4.  The only remaining case is 64-bit guest.  */
2156        tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
2157    }
2158
2159    tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
2160    tcg_out32(s, MFSPR | RT(arg) | LR);
2161
2162    tcg_out_call(s, qemu_ld_helpers[opc & (MO_BSWAP | MO_SIZE)]);
2163
2164    lo = lb->datalo_reg;
2165    hi = lb->datahi_reg;
2166    if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
2167        tcg_out_mov(s, TCG_TYPE_I32, lo, TCG_REG_R4);
2168        tcg_out_mov(s, TCG_TYPE_I32, hi, TCG_REG_R3);
2169    } else if (opc & MO_SIGN) {
2170        uint32_t insn = qemu_exts_opc[opc & MO_SIZE];
2171        tcg_out32(s, insn | RA(lo) | RS(TCG_REG_R3));
2172    } else {
2173        tcg_out_mov(s, TCG_TYPE_REG, lo, TCG_REG_R3);
2174    }
2175
2176    tcg_out_b(s, 0, lb->raddr);
2177    return true;
2178}
2179
2180static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2181{
2182    MemOpIdx oi = lb->oi;
2183    MemOp opc = get_memop(oi);
2184    MemOp s_bits = opc & MO_SIZE;
2185    TCGReg hi, lo, arg = TCG_REG_R3;
2186
2187    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2188        return false;
2189    }
2190
2191    tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
2192
2193    lo = lb->addrlo_reg;
2194    hi = lb->addrhi_reg;
2195    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2196#ifdef TCG_TARGET_CALL_ALIGN_ARGS
2197        arg |= 1;
2198#endif
2199        tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2200        tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2201    } else {
2202        /* If the address needed to be zero-extended, we'll have already
2203           placed it in R4.  The only remaining case is 64-bit guest.  */
2204        tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
2205    }
2206
2207    lo = lb->datalo_reg;
2208    hi = lb->datahi_reg;
2209    if (TCG_TARGET_REG_BITS == 32) {
2210        switch (s_bits) {
2211        case MO_64:
2212#ifdef TCG_TARGET_CALL_ALIGN_ARGS
2213            arg |= 1;
2214#endif
2215            tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2216            /* FALLTHRU */
2217        case MO_32:
2218            tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2219            break;
2220        default:
2221            tcg_out_rlw(s, RLWINM, arg++, lo, 0, 32 - (8 << s_bits), 31);
2222            break;
2223        }
2224    } else {
2225        if (s_bits == MO_64) {
2226            tcg_out_mov(s, TCG_TYPE_I64, arg++, lo);
2227        } else {
2228            tcg_out_rld(s, RLDICL, arg++, lo, 0, 64 - (8 << s_bits));
2229        }
2230    }
2231
2232    tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
2233    tcg_out32(s, MFSPR | RT(arg) | LR);
2234
2235    tcg_out_call(s, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
2236
2237    tcg_out_b(s, 0, lb->raddr);
2238    return true;
2239}
2240#else
2241
2242static void tcg_out_test_alignment(TCGContext *s, bool is_ld, TCGReg addrlo,
2243                                   TCGReg addrhi, unsigned a_bits)
2244{
2245    unsigned a_mask = (1 << a_bits) - 1;
2246    TCGLabelQemuLdst *label = new_ldst_label(s);
2247
2248    label->is_ld = is_ld;
2249    label->addrlo_reg = addrlo;
2250    label->addrhi_reg = addrhi;
2251
2252    /* We are expecting a_bits to max out at 7, much lower than ANDI. */
2253    tcg_debug_assert(a_bits < 16);
2254    tcg_out32(s, ANDI | SAI(addrlo, TCG_REG_R0, a_mask));
2255
2256    label->label_ptr[0] = s->code_ptr;
2257    tcg_out32(s, BC | BI(0, CR_EQ) | BO_COND_FALSE | LK);
2258
2259    label->raddr = tcg_splitwx_to_rx(s->code_ptr);
2260}
2261
2262static bool tcg_out_fail_alignment(TCGContext *s, TCGLabelQemuLdst *l)
2263{
2264    if (!reloc_pc14(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2265        return false;
2266    }
2267
2268    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2269        TCGReg arg = TCG_REG_R4;
2270#ifdef TCG_TARGET_CALL_ALIGN_ARGS
2271        arg |= 1;
2272#endif
2273        if (l->addrlo_reg != arg) {
2274            tcg_out_mov(s, TCG_TYPE_I32, arg, l->addrhi_reg);
2275            tcg_out_mov(s, TCG_TYPE_I32, arg + 1, l->addrlo_reg);
2276        } else if (l->addrhi_reg != arg + 1) {
2277            tcg_out_mov(s, TCG_TYPE_I32, arg + 1, l->addrlo_reg);
2278            tcg_out_mov(s, TCG_TYPE_I32, arg, l->addrhi_reg);
2279        } else {
2280            tcg_out_mov(s, TCG_TYPE_I32, TCG_REG_R0, arg);
2281            tcg_out_mov(s, TCG_TYPE_I32, arg, arg + 1);
2282            tcg_out_mov(s, TCG_TYPE_I32, arg + 1, TCG_REG_R0);
2283        }
2284    } else {
2285        tcg_out_mov(s, TCG_TYPE_TL, TCG_REG_R4, l->addrlo_reg);
2286    }
2287    tcg_out_mov(s, TCG_TYPE_TL, TCG_REG_R3, TCG_AREG0);
2288
2289    /* "Tail call" to the helper, with the return address back inline. */
2290    tcg_out_call_int(s, 0, (const void *)(l->is_ld ? helper_unaligned_ld
2291                                          : helper_unaligned_st));
2292    return true;
2293}
2294
2295static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
2296{
2297    return tcg_out_fail_alignment(s, l);
2298}
2299
2300static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
2301{
2302    return tcg_out_fail_alignment(s, l);
2303}
2304
2305#endif /* SOFTMMU */
2306
2307static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64)
2308{
2309    TCGReg datalo, datahi, addrlo, rbase;
2310    TCGReg addrhi __attribute__((unused));
2311    MemOpIdx oi;
2312    MemOp opc, s_bits;
2313#ifdef CONFIG_SOFTMMU
2314    int mem_index;
2315    tcg_insn_unit *label_ptr;
2316#else
2317    unsigned a_bits;
2318#endif
2319
2320    datalo = *args++;
2321    datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
2322    addrlo = *args++;
2323    addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
2324    oi = *args++;
2325    opc = get_memop(oi);
2326    s_bits = opc & MO_SIZE;
2327
2328#ifdef CONFIG_SOFTMMU
2329    mem_index = get_mmuidx(oi);
2330    addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, true);
2331
2332    /* Load a pointer into the current opcode w/conditional branch-link. */
2333    label_ptr = s->code_ptr;
2334    tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
2335
2336    rbase = TCG_REG_R3;
2337#else  /* !CONFIG_SOFTMMU */
2338    a_bits = get_alignment_bits(opc);
2339    if (a_bits) {
2340        tcg_out_test_alignment(s, true, addrlo, addrhi, a_bits);
2341    }
2342    rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
2343    if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
2344        tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
2345        addrlo = TCG_REG_TMP1;
2346    }
2347#endif
2348
2349    if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
2350        if (opc & MO_BSWAP) {
2351            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2352            tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
2353            tcg_out32(s, LWBRX | TAB(datahi, rbase, TCG_REG_R0));
2354        } else if (rbase != 0) {
2355            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2356            tcg_out32(s, LWZX | TAB(datahi, rbase, addrlo));
2357            tcg_out32(s, LWZX | TAB(datalo, rbase, TCG_REG_R0));
2358        } else if (addrlo == datahi) {
2359            tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
2360            tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
2361        } else {
2362            tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
2363            tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
2364        }
2365    } else {
2366        uint32_t insn = qemu_ldx_opc[opc & (MO_BSWAP | MO_SSIZE)];
2367        if (!have_isa_2_06 && insn == LDBRX) {
2368            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2369            tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
2370            tcg_out32(s, LWBRX | TAB(TCG_REG_R0, rbase, TCG_REG_R0));
2371            tcg_out_rld(s, RLDIMI, datalo, TCG_REG_R0, 32, 0);
2372        } else if (insn) {
2373            tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
2374        } else {
2375            insn = qemu_ldx_opc[opc & (MO_SIZE | MO_BSWAP)];
2376            tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
2377            insn = qemu_exts_opc[s_bits];
2378            tcg_out32(s, insn | RA(datalo) | RS(datalo));
2379        }
2380    }
2381
2382#ifdef CONFIG_SOFTMMU
2383    add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
2384                        s->code_ptr, label_ptr);
2385#endif
2386}
2387
2388static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
2389{
2390    TCGReg datalo, datahi, addrlo, rbase;
2391    TCGReg addrhi __attribute__((unused));
2392    MemOpIdx oi;
2393    MemOp opc, s_bits;
2394#ifdef CONFIG_SOFTMMU
2395    int mem_index;
2396    tcg_insn_unit *label_ptr;
2397#else
2398    unsigned a_bits;
2399#endif
2400
2401    datalo = *args++;
2402    datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
2403    addrlo = *args++;
2404    addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
2405    oi = *args++;
2406    opc = get_memop(oi);
2407    s_bits = opc & MO_SIZE;
2408
2409#ifdef CONFIG_SOFTMMU
2410    mem_index = get_mmuidx(oi);
2411    addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, false);
2412
2413    /* Load a pointer into the current opcode w/conditional branch-link. */
2414    label_ptr = s->code_ptr;
2415    tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
2416
2417    rbase = TCG_REG_R3;
2418#else  /* !CONFIG_SOFTMMU */
2419    a_bits = get_alignment_bits(opc);
2420    if (a_bits) {
2421        tcg_out_test_alignment(s, false, addrlo, addrhi, a_bits);
2422    }
2423    rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
2424    if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
2425        tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
2426        addrlo = TCG_REG_TMP1;
2427    }
2428#endif
2429
2430    if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
2431        if (opc & MO_BSWAP) {
2432            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2433            tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
2434            tcg_out32(s, STWBRX | SAB(datahi, rbase, TCG_REG_R0));
2435        } else if (rbase != 0) {
2436            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2437            tcg_out32(s, STWX | SAB(datahi, rbase, addrlo));
2438            tcg_out32(s, STWX | SAB(datalo, rbase, TCG_REG_R0));
2439        } else {
2440            tcg_out32(s, STW | TAI(datahi, addrlo, 0));
2441            tcg_out32(s, STW | TAI(datalo, addrlo, 4));
2442        }
2443    } else {
2444        uint32_t insn = qemu_stx_opc[opc & (MO_BSWAP | MO_SIZE)];
2445        if (!have_isa_2_06 && insn == STDBRX) {
2446            tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
2447            tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, addrlo, 4));
2448            tcg_out_shri64(s, TCG_REG_R0, datalo, 32);
2449            tcg_out32(s, STWBRX | SAB(TCG_REG_R0, rbase, TCG_REG_TMP1));
2450        } else {
2451            tcg_out32(s, insn | SAB(datalo, rbase, addrlo));
2452        }
2453    }
2454
2455#ifdef CONFIG_SOFTMMU
2456    add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
2457                        s->code_ptr, label_ptr);
2458#endif
2459}
2460
2461static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
2462{
2463    int i;
2464    for (i = 0; i < count; ++i) {
2465        p[i] = NOP;
2466    }
2467}
2468
2469/* Parameters for function call generation, used in tcg.c.  */
2470#define TCG_TARGET_STACK_ALIGN       16
2471#define TCG_TARGET_EXTEND_ARGS       1
2472
2473#ifdef _CALL_AIX
2474# define LINK_AREA_SIZE                (6 * SZR)
2475# define LR_OFFSET                     (1 * SZR)
2476# define TCG_TARGET_CALL_STACK_OFFSET  (LINK_AREA_SIZE + 8 * SZR)
2477#elif defined(_CALL_DARWIN)
2478# define LINK_AREA_SIZE                (6 * SZR)
2479# define LR_OFFSET                     (2 * SZR)
2480#elif TCG_TARGET_REG_BITS == 64
2481# if defined(_CALL_ELF) && _CALL_ELF == 2
2482#  define LINK_AREA_SIZE               (4 * SZR)
2483#  define LR_OFFSET                    (1 * SZR)
2484# endif
2485#else /* TCG_TARGET_REG_BITS == 32 */
2486# if defined(_CALL_SYSV)
2487#  define LINK_AREA_SIZE               (2 * SZR)
2488#  define LR_OFFSET                    (1 * SZR)
2489# endif
2490#endif
2491#ifndef LR_OFFSET
2492# error "Unhandled abi"
2493#endif
2494#ifndef TCG_TARGET_CALL_STACK_OFFSET
2495# define TCG_TARGET_CALL_STACK_OFFSET  LINK_AREA_SIZE
2496#endif
2497
2498#define CPU_TEMP_BUF_SIZE  (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
2499#define REG_SAVE_SIZE      ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
2500
2501#define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET   \
2502                     + TCG_STATIC_CALL_ARGS_SIZE    \
2503                     + CPU_TEMP_BUF_SIZE            \
2504                     + REG_SAVE_SIZE                \
2505                     + TCG_TARGET_STACK_ALIGN - 1)  \
2506                    & -TCG_TARGET_STACK_ALIGN)
2507
2508#define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
2509
2510static void tcg_target_qemu_prologue(TCGContext *s)
2511{
2512    int i;
2513
2514#ifdef _CALL_AIX
2515    const void **desc = (const void **)s->code_ptr;
2516    desc[0] = tcg_splitwx_to_rx(desc + 2);  /* entry point */
2517    desc[1] = 0;                            /* environment pointer */
2518    s->code_ptr = (void *)(desc + 2);       /* skip over descriptor */
2519#endif
2520
2521    tcg_set_frame(s, TCG_REG_CALL_STACK, REG_SAVE_BOT - CPU_TEMP_BUF_SIZE,
2522                  CPU_TEMP_BUF_SIZE);
2523
2524    /* Prologue */
2525    tcg_out32(s, MFSPR | RT(TCG_REG_R0) | LR);
2526    tcg_out32(s, (SZR == 8 ? STDU : STWU)
2527              | SAI(TCG_REG_R1, TCG_REG_R1, -FRAME_SIZE));
2528
2529    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2530        tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2531                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2532    }
2533    tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2534
2535#ifndef CONFIG_SOFTMMU
2536    if (guest_base) {
2537        tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true);
2538        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2539    }
2540#endif
2541
2542    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2543    tcg_out32(s, MTSPR | RS(tcg_target_call_iarg_regs[1]) | CTR);
2544    if (USE_REG_TB) {
2545        tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, tcg_target_call_iarg_regs[1]);
2546    }
2547    tcg_out32(s, BCCTR | BO_ALWAYS);
2548
2549    /* Epilogue */
2550    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
2551
2552    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2553    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2554        tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2555                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2556    }
2557    tcg_out32(s, MTSPR | RS(TCG_REG_R0) | LR);
2558    tcg_out32(s, ADDI | TAI(TCG_REG_R1, TCG_REG_R1, FRAME_SIZE));
2559    tcg_out32(s, BCLR | BO_ALWAYS);
2560}
2561
2562static void tcg_out_op(TCGContext *s, TCGOpcode opc,
2563                       const TCGArg args[TCG_MAX_OP_ARGS],
2564                       const int const_args[TCG_MAX_OP_ARGS])
2565{
2566    TCGArg a0, a1, a2;
2567
2568    switch (opc) {
2569    case INDEX_op_exit_tb:
2570        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, args[0]);
2571        tcg_out_b(s, 0, tcg_code_gen_epilogue);
2572        break;
2573    case INDEX_op_goto_tb:
2574        if (s->tb_jmp_insn_offset) {
2575            /* Direct jump. */
2576            if (TCG_TARGET_REG_BITS == 64) {
2577                /* Ensure the next insns are 8-byte aligned. */
2578                if ((uintptr_t)s->code_ptr & 7) {
2579                    tcg_out32(s, NOP);
2580                }
2581                s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2582                tcg_out32(s, ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2583                tcg_out32(s, ADDI | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2584            } else {
2585                s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2586                tcg_out32(s, B);
2587                s->tb_jmp_reset_offset[args[0]] = tcg_current_code_size(s);
2588                break;
2589            }
2590        } else {
2591            /* Indirect jump. */
2592            tcg_debug_assert(s->tb_jmp_insn_offset == NULL);
2593            tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, 0,
2594                       (intptr_t)(s->tb_jmp_insn_offset + args[0]));
2595        }
2596        tcg_out32(s, MTSPR | RS(TCG_REG_TB) | CTR);
2597        tcg_out32(s, BCCTR | BO_ALWAYS);
2598        set_jmp_reset_offset(s, args[0]);
2599        if (USE_REG_TB) {
2600            /* For the unlinked case, need to reset TCG_REG_TB.  */
2601            tcg_out_mem_long(s, ADDI, ADD, TCG_REG_TB, TCG_REG_TB,
2602                             -tcg_current_code_size(s));
2603        }
2604        break;
2605    case INDEX_op_goto_ptr:
2606        tcg_out32(s, MTSPR | RS(args[0]) | CTR);
2607        if (USE_REG_TB) {
2608            tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, args[0]);
2609        }
2610        tcg_out32(s, ADDI | TAI(TCG_REG_R3, 0, 0));
2611        tcg_out32(s, BCCTR | BO_ALWAYS);
2612        break;
2613    case INDEX_op_br:
2614        {
2615            TCGLabel *l = arg_label(args[0]);
2616            uint32_t insn = B;
2617
2618            if (l->has_value) {
2619                insn |= reloc_pc24_val(tcg_splitwx_to_rx(s->code_ptr),
2620                                       l->u.value_ptr);
2621            } else {
2622                tcg_out_reloc(s, s->code_ptr, R_PPC_REL24, l, 0);
2623            }
2624            tcg_out32(s, insn);
2625        }
2626        break;
2627    case INDEX_op_ld8u_i32:
2628    case INDEX_op_ld8u_i64:
2629        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2630        break;
2631    case INDEX_op_ld8s_i32:
2632    case INDEX_op_ld8s_i64:
2633        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2634        tcg_out_ext8s(s, args[0], args[0]);
2635        break;
2636    case INDEX_op_ld16u_i32:
2637    case INDEX_op_ld16u_i64:
2638        tcg_out_mem_long(s, LHZ, LHZX, args[0], args[1], args[2]);
2639        break;
2640    case INDEX_op_ld16s_i32:
2641    case INDEX_op_ld16s_i64:
2642        tcg_out_mem_long(s, LHA, LHAX, args[0], args[1], args[2]);
2643        break;
2644    case INDEX_op_ld_i32:
2645    case INDEX_op_ld32u_i64:
2646        tcg_out_mem_long(s, LWZ, LWZX, args[0], args[1], args[2]);
2647        break;
2648    case INDEX_op_ld32s_i64:
2649        tcg_out_mem_long(s, LWA, LWAX, args[0], args[1], args[2]);
2650        break;
2651    case INDEX_op_ld_i64:
2652        tcg_out_mem_long(s, LD, LDX, args[0], args[1], args[2]);
2653        break;
2654    case INDEX_op_st8_i32:
2655    case INDEX_op_st8_i64:
2656        tcg_out_mem_long(s, STB, STBX, args[0], args[1], args[2]);
2657        break;
2658    case INDEX_op_st16_i32:
2659    case INDEX_op_st16_i64:
2660        tcg_out_mem_long(s, STH, STHX, args[0], args[1], args[2]);
2661        break;
2662    case INDEX_op_st_i32:
2663    case INDEX_op_st32_i64:
2664        tcg_out_mem_long(s, STW, STWX, args[0], args[1], args[2]);
2665        break;
2666    case INDEX_op_st_i64:
2667        tcg_out_mem_long(s, STD, STDX, args[0], args[1], args[2]);
2668        break;
2669
2670    case INDEX_op_add_i32:
2671        a0 = args[0], a1 = args[1], a2 = args[2];
2672        if (const_args[2]) {
2673        do_addi_32:
2674            tcg_out_mem_long(s, ADDI, ADD, a0, a1, (int32_t)a2);
2675        } else {
2676            tcg_out32(s, ADD | TAB(a0, a1, a2));
2677        }
2678        break;
2679    case INDEX_op_sub_i32:
2680        a0 = args[0], a1 = args[1], a2 = args[2];
2681        if (const_args[1]) {
2682            if (const_args[2]) {
2683                tcg_out_movi(s, TCG_TYPE_I32, a0, a1 - a2);
2684            } else {
2685                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2686            }
2687        } else if (const_args[2]) {
2688            a2 = -a2;
2689            goto do_addi_32;
2690        } else {
2691            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2692        }
2693        break;
2694
2695    case INDEX_op_and_i32:
2696        a0 = args[0], a1 = args[1], a2 = args[2];
2697        if (const_args[2]) {
2698            tcg_out_andi32(s, a0, a1, a2);
2699        } else {
2700            tcg_out32(s, AND | SAB(a1, a0, a2));
2701        }
2702        break;
2703    case INDEX_op_and_i64:
2704        a0 = args[0], a1 = args[1], a2 = args[2];
2705        if (const_args[2]) {
2706            tcg_out_andi64(s, a0, a1, a2);
2707        } else {
2708            tcg_out32(s, AND | SAB(a1, a0, a2));
2709        }
2710        break;
2711    case INDEX_op_or_i64:
2712    case INDEX_op_or_i32:
2713        a0 = args[0], a1 = args[1], a2 = args[2];
2714        if (const_args[2]) {
2715            tcg_out_ori32(s, a0, a1, a2);
2716        } else {
2717            tcg_out32(s, OR | SAB(a1, a0, a2));
2718        }
2719        break;
2720    case INDEX_op_xor_i64:
2721    case INDEX_op_xor_i32:
2722        a0 = args[0], a1 = args[1], a2 = args[2];
2723        if (const_args[2]) {
2724            tcg_out_xori32(s, a0, a1, a2);
2725        } else {
2726            tcg_out32(s, XOR | SAB(a1, a0, a2));
2727        }
2728        break;
2729    case INDEX_op_andc_i32:
2730        a0 = args[0], a1 = args[1], a2 = args[2];
2731        if (const_args[2]) {
2732            tcg_out_andi32(s, a0, a1, ~a2);
2733        } else {
2734            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2735        }
2736        break;
2737    case INDEX_op_andc_i64:
2738        a0 = args[0], a1 = args[1], a2 = args[2];
2739        if (const_args[2]) {
2740            tcg_out_andi64(s, a0, a1, ~a2);
2741        } else {
2742            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2743        }
2744        break;
2745    case INDEX_op_orc_i32:
2746        if (const_args[2]) {
2747            tcg_out_ori32(s, args[0], args[1], ~args[2]);
2748            break;
2749        }
2750        /* FALLTHRU */
2751    case INDEX_op_orc_i64:
2752        tcg_out32(s, ORC | SAB(args[1], args[0], args[2]));
2753        break;
2754    case INDEX_op_eqv_i32:
2755        if (const_args[2]) {
2756            tcg_out_xori32(s, args[0], args[1], ~args[2]);
2757            break;
2758        }
2759        /* FALLTHRU */
2760    case INDEX_op_eqv_i64:
2761        tcg_out32(s, EQV | SAB(args[1], args[0], args[2]));
2762        break;
2763    case INDEX_op_nand_i32:
2764    case INDEX_op_nand_i64:
2765        tcg_out32(s, NAND | SAB(args[1], args[0], args[2]));
2766        break;
2767    case INDEX_op_nor_i32:
2768    case INDEX_op_nor_i64:
2769        tcg_out32(s, NOR | SAB(args[1], args[0], args[2]));
2770        break;
2771
2772    case INDEX_op_clz_i32:
2773        tcg_out_cntxz(s, TCG_TYPE_I32, CNTLZW, args[0], args[1],
2774                      args[2], const_args[2]);
2775        break;
2776    case INDEX_op_ctz_i32:
2777        tcg_out_cntxz(s, TCG_TYPE_I32, CNTTZW, args[0], args[1],
2778                      args[2], const_args[2]);
2779        break;
2780    case INDEX_op_ctpop_i32:
2781        tcg_out32(s, CNTPOPW | SAB(args[1], args[0], 0));
2782        break;
2783
2784    case INDEX_op_clz_i64:
2785        tcg_out_cntxz(s, TCG_TYPE_I64, CNTLZD, args[0], args[1],
2786                      args[2], const_args[2]);
2787        break;
2788    case INDEX_op_ctz_i64:
2789        tcg_out_cntxz(s, TCG_TYPE_I64, CNTTZD, args[0], args[1],
2790                      args[2], const_args[2]);
2791        break;
2792    case INDEX_op_ctpop_i64:
2793        tcg_out32(s, CNTPOPD | SAB(args[1], args[0], 0));
2794        break;
2795
2796    case INDEX_op_mul_i32:
2797        a0 = args[0], a1 = args[1], a2 = args[2];
2798        if (const_args[2]) {
2799            tcg_out32(s, MULLI | TAI(a0, a1, a2));
2800        } else {
2801            tcg_out32(s, MULLW | TAB(a0, a1, a2));
2802        }
2803        break;
2804
2805    case INDEX_op_div_i32:
2806        tcg_out32(s, DIVW | TAB(args[0], args[1], args[2]));
2807        break;
2808
2809    case INDEX_op_divu_i32:
2810        tcg_out32(s, DIVWU | TAB(args[0], args[1], args[2]));
2811        break;
2812
2813    case INDEX_op_rem_i32:
2814        tcg_out32(s, MODSW | TAB(args[0], args[1], args[2]));
2815        break;
2816
2817    case INDEX_op_remu_i32:
2818        tcg_out32(s, MODUW | TAB(args[0], args[1], args[2]));
2819        break;
2820
2821    case INDEX_op_shl_i32:
2822        if (const_args[2]) {
2823            /* Limit immediate shift count lest we create an illegal insn.  */
2824            tcg_out_shli32(s, args[0], args[1], args[2] & 31);
2825        } else {
2826            tcg_out32(s, SLW | SAB(args[1], args[0], args[2]));
2827        }
2828        break;
2829    case INDEX_op_shr_i32:
2830        if (const_args[2]) {
2831            /* Limit immediate shift count lest we create an illegal insn.  */
2832            tcg_out_shri32(s, args[0], args[1], args[2] & 31);
2833        } else {
2834            tcg_out32(s, SRW | SAB(args[1], args[0], args[2]));
2835        }
2836        break;
2837    case INDEX_op_sar_i32:
2838        if (const_args[2]) {
2839            tcg_out_sari32(s, args[0], args[1], args[2]);
2840        } else {
2841            tcg_out32(s, SRAW | SAB(args[1], args[0], args[2]));
2842        }
2843        break;
2844    case INDEX_op_rotl_i32:
2845        if (const_args[2]) {
2846            tcg_out_rlw(s, RLWINM, args[0], args[1], args[2], 0, 31);
2847        } else {
2848            tcg_out32(s, RLWNM | SAB(args[1], args[0], args[2])
2849                         | MB(0) | ME(31));
2850        }
2851        break;
2852    case INDEX_op_rotr_i32:
2853        if (const_args[2]) {
2854            tcg_out_rlw(s, RLWINM, args[0], args[1], 32 - args[2], 0, 31);
2855        } else {
2856            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 32));
2857            tcg_out32(s, RLWNM | SAB(args[1], args[0], TCG_REG_R0)
2858                         | MB(0) | ME(31));
2859        }
2860        break;
2861
2862    case INDEX_op_brcond_i32:
2863        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2864                       arg_label(args[3]), TCG_TYPE_I32);
2865        break;
2866    case INDEX_op_brcond_i64:
2867        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2868                       arg_label(args[3]), TCG_TYPE_I64);
2869        break;
2870    case INDEX_op_brcond2_i32:
2871        tcg_out_brcond2(s, args, const_args);
2872        break;
2873
2874    case INDEX_op_neg_i32:
2875    case INDEX_op_neg_i64:
2876        tcg_out32(s, NEG | RT(args[0]) | RA(args[1]));
2877        break;
2878
2879    case INDEX_op_not_i32:
2880    case INDEX_op_not_i64:
2881        tcg_out32(s, NOR | SAB(args[1], args[0], args[1]));
2882        break;
2883
2884    case INDEX_op_add_i64:
2885        a0 = args[0], a1 = args[1], a2 = args[2];
2886        if (const_args[2]) {
2887        do_addi_64:
2888            tcg_out_mem_long(s, ADDI, ADD, a0, a1, a2);
2889        } else {
2890            tcg_out32(s, ADD | TAB(a0, a1, a2));
2891        }
2892        break;
2893    case INDEX_op_sub_i64:
2894        a0 = args[0], a1 = args[1], a2 = args[2];
2895        if (const_args[1]) {
2896            if (const_args[2]) {
2897                tcg_out_movi(s, TCG_TYPE_I64, a0, a1 - a2);
2898            } else {
2899                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2900            }
2901        } else if (const_args[2]) {
2902            a2 = -a2;
2903            goto do_addi_64;
2904        } else {
2905            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2906        }
2907        break;
2908
2909    case INDEX_op_shl_i64:
2910        if (const_args[2]) {
2911            /* Limit immediate shift count lest we create an illegal insn.  */
2912            tcg_out_shli64(s, args[0], args[1], args[2] & 63);
2913        } else {
2914            tcg_out32(s, SLD | SAB(args[1], args[0], args[2]));
2915        }
2916        break;
2917    case INDEX_op_shr_i64:
2918        if (const_args[2]) {
2919            /* Limit immediate shift count lest we create an illegal insn.  */
2920            tcg_out_shri64(s, args[0], args[1], args[2] & 63);
2921        } else {
2922            tcg_out32(s, SRD | SAB(args[1], args[0], args[2]));
2923        }
2924        break;
2925    case INDEX_op_sar_i64:
2926        if (const_args[2]) {
2927            tcg_out_sari64(s, args[0], args[1], args[2]);
2928        } else {
2929            tcg_out32(s, SRAD | SAB(args[1], args[0], args[2]));
2930        }
2931        break;
2932    case INDEX_op_rotl_i64:
2933        if (const_args[2]) {
2934            tcg_out_rld(s, RLDICL, args[0], args[1], args[2], 0);
2935        } else {
2936            tcg_out32(s, RLDCL | SAB(args[1], args[0], args[2]) | MB64(0));
2937        }
2938        break;
2939    case INDEX_op_rotr_i64:
2940        if (const_args[2]) {
2941            tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 0);
2942        } else {
2943            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 64));
2944            tcg_out32(s, RLDCL | SAB(args[1], args[0], TCG_REG_R0) | MB64(0));
2945        }
2946        break;
2947
2948    case INDEX_op_mul_i64:
2949        a0 = args[0], a1 = args[1], a2 = args[2];
2950        if (const_args[2]) {
2951            tcg_out32(s, MULLI | TAI(a0, a1, a2));
2952        } else {
2953            tcg_out32(s, MULLD | TAB(a0, a1, a2));
2954        }
2955        break;
2956    case INDEX_op_div_i64:
2957        tcg_out32(s, DIVD | TAB(args[0], args[1], args[2]));
2958        break;
2959    case INDEX_op_divu_i64:
2960        tcg_out32(s, DIVDU | TAB(args[0], args[1], args[2]));
2961        break;
2962    case INDEX_op_rem_i64:
2963        tcg_out32(s, MODSD | TAB(args[0], args[1], args[2]));
2964        break;
2965    case INDEX_op_remu_i64:
2966        tcg_out32(s, MODUD | TAB(args[0], args[1], args[2]));
2967        break;
2968
2969    case INDEX_op_qemu_ld_i32:
2970        tcg_out_qemu_ld(s, args, false);
2971        break;
2972    case INDEX_op_qemu_ld_i64:
2973        tcg_out_qemu_ld(s, args, true);
2974        break;
2975    case INDEX_op_qemu_st_i32:
2976        tcg_out_qemu_st(s, args, false);
2977        break;
2978    case INDEX_op_qemu_st_i64:
2979        tcg_out_qemu_st(s, args, true);
2980        break;
2981
2982    case INDEX_op_ext8s_i32:
2983    case INDEX_op_ext8s_i64:
2984        tcg_out_ext8s(s, args[0], args[1]);
2985        break;
2986    case INDEX_op_ext16s_i32:
2987    case INDEX_op_ext16s_i64:
2988        tcg_out_ext16s(s, args[0], args[1]);
2989        break;
2990    case INDEX_op_ext_i32_i64:
2991    case INDEX_op_ext32s_i64:
2992        tcg_out_ext32s(s, args[0], args[1]);
2993        break;
2994    case INDEX_op_extu_i32_i64:
2995        tcg_out_ext32u(s, args[0], args[1]);
2996        break;
2997
2998    case INDEX_op_setcond_i32:
2999        tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2],
3000                        const_args[2]);
3001        break;
3002    case INDEX_op_setcond_i64:
3003        tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2],
3004                        const_args[2]);
3005        break;
3006    case INDEX_op_setcond2_i32:
3007        tcg_out_setcond2(s, args, const_args);
3008        break;
3009
3010    case INDEX_op_bswap16_i32:
3011    case INDEX_op_bswap16_i64:
3012        tcg_out_bswap16(s, args[0], args[1], args[2]);
3013        break;
3014    case INDEX_op_bswap32_i32:
3015        tcg_out_bswap32(s, args[0], args[1], 0);
3016        break;
3017    case INDEX_op_bswap32_i64:
3018        tcg_out_bswap32(s, args[0], args[1], args[2]);
3019        break;
3020    case INDEX_op_bswap64_i64:
3021        tcg_out_bswap64(s, args[0], args[1]);
3022        break;
3023
3024    case INDEX_op_deposit_i32:
3025        if (const_args[2]) {
3026            uint32_t mask = ((2u << (args[4] - 1)) - 1) << args[3];
3027            tcg_out_andi32(s, args[0], args[0], ~mask);
3028        } else {
3029            tcg_out_rlw(s, RLWIMI, args[0], args[2], args[3],
3030                        32 - args[3] - args[4], 31 - args[3]);
3031        }
3032        break;
3033    case INDEX_op_deposit_i64:
3034        if (const_args[2]) {
3035            uint64_t mask = ((2ull << (args[4] - 1)) - 1) << args[3];
3036            tcg_out_andi64(s, args[0], args[0], ~mask);
3037        } else {
3038            tcg_out_rld(s, RLDIMI, args[0], args[2], args[3],
3039                        64 - args[3] - args[4]);
3040        }
3041        break;
3042
3043    case INDEX_op_extract_i32:
3044        tcg_out_rlw(s, RLWINM, args[0], args[1],
3045                    32 - args[2], 32 - args[3], 31);
3046        break;
3047    case INDEX_op_extract_i64:
3048        tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 64 - args[3]);
3049        break;
3050
3051    case INDEX_op_movcond_i32:
3052        tcg_out_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], args[2],
3053                        args[3], args[4], const_args[2]);
3054        break;
3055    case INDEX_op_movcond_i64:
3056        tcg_out_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], args[2],
3057                        args[3], args[4], const_args[2]);
3058        break;
3059
3060#if TCG_TARGET_REG_BITS == 64
3061    case INDEX_op_add2_i64:
3062#else
3063    case INDEX_op_add2_i32:
3064#endif
3065        /* Note that the CA bit is defined based on the word size of the
3066           environment.  So in 64-bit mode it's always carry-out of bit 63.
3067           The fallback code using deposit works just as well for 32-bit.  */
3068        a0 = args[0], a1 = args[1];
3069        if (a0 == args[3] || (!const_args[5] && a0 == args[5])) {
3070            a0 = TCG_REG_R0;
3071        }
3072        if (const_args[4]) {
3073            tcg_out32(s, ADDIC | TAI(a0, args[2], args[4]));
3074        } else {
3075            tcg_out32(s, ADDC | TAB(a0, args[2], args[4]));
3076        }
3077        if (const_args[5]) {
3078            tcg_out32(s, (args[5] ? ADDME : ADDZE) | RT(a1) | RA(args[3]));
3079        } else {
3080            tcg_out32(s, ADDE | TAB(a1, args[3], args[5]));
3081        }
3082        if (a0 != args[0]) {
3083            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
3084        }
3085        break;
3086
3087#if TCG_TARGET_REG_BITS == 64
3088    case INDEX_op_sub2_i64:
3089#else
3090    case INDEX_op_sub2_i32:
3091#endif
3092        a0 = args[0], a1 = args[1];
3093        if (a0 == args[5] || (!const_args[3] && a0 == args[3])) {
3094            a0 = TCG_REG_R0;
3095        }
3096        if (const_args[2]) {
3097            tcg_out32(s, SUBFIC | TAI(a0, args[4], args[2]));
3098        } else {
3099            tcg_out32(s, SUBFC | TAB(a0, args[4], args[2]));
3100        }
3101        if (const_args[3]) {
3102            tcg_out32(s, (args[3] ? SUBFME : SUBFZE) | RT(a1) | RA(args[5]));
3103        } else {
3104            tcg_out32(s, SUBFE | TAB(a1, args[5], args[3]));
3105        }
3106        if (a0 != args[0]) {
3107            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
3108        }
3109        break;
3110
3111    case INDEX_op_muluh_i32:
3112        tcg_out32(s, MULHWU | TAB(args[0], args[1], args[2]));
3113        break;
3114    case INDEX_op_mulsh_i32:
3115        tcg_out32(s, MULHW | TAB(args[0], args[1], args[2]));
3116        break;
3117    case INDEX_op_muluh_i64:
3118        tcg_out32(s, MULHDU | TAB(args[0], args[1], args[2]));
3119        break;
3120    case INDEX_op_mulsh_i64:
3121        tcg_out32(s, MULHD | TAB(args[0], args[1], args[2]));
3122        break;
3123
3124    case INDEX_op_mb:
3125        tcg_out_mb(s, args[0]);
3126        break;
3127
3128    case INDEX_op_mov_i32:   /* Always emitted via tcg_out_mov.  */
3129    case INDEX_op_mov_i64:
3130    case INDEX_op_call:      /* Always emitted via tcg_out_call.  */
3131    default:
3132        tcg_abort();
3133    }
3134}
3135
3136int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3137{
3138    switch (opc) {
3139    case INDEX_op_and_vec:
3140    case INDEX_op_or_vec:
3141    case INDEX_op_xor_vec:
3142    case INDEX_op_andc_vec:
3143    case INDEX_op_not_vec:
3144    case INDEX_op_nor_vec:
3145    case INDEX_op_eqv_vec:
3146    case INDEX_op_nand_vec:
3147        return 1;
3148    case INDEX_op_orc_vec:
3149        return have_isa_2_07;
3150    case INDEX_op_add_vec:
3151    case INDEX_op_sub_vec:
3152    case INDEX_op_smax_vec:
3153    case INDEX_op_smin_vec:
3154    case INDEX_op_umax_vec:
3155    case INDEX_op_umin_vec:
3156    case INDEX_op_shlv_vec:
3157    case INDEX_op_shrv_vec:
3158    case INDEX_op_sarv_vec:
3159    case INDEX_op_rotlv_vec:
3160        return vece <= MO_32 || have_isa_2_07;
3161    case INDEX_op_ssadd_vec:
3162    case INDEX_op_sssub_vec:
3163    case INDEX_op_usadd_vec:
3164    case INDEX_op_ussub_vec:
3165        return vece <= MO_32;
3166    case INDEX_op_cmp_vec:
3167    case INDEX_op_shli_vec:
3168    case INDEX_op_shri_vec:
3169    case INDEX_op_sari_vec:
3170    case INDEX_op_rotli_vec:
3171        return vece <= MO_32 || have_isa_2_07 ? -1 : 0;
3172    case INDEX_op_neg_vec:
3173        return vece >= MO_32 && have_isa_3_00;
3174    case INDEX_op_mul_vec:
3175        switch (vece) {
3176        case MO_8:
3177        case MO_16:
3178            return -1;
3179        case MO_32:
3180            return have_isa_2_07 ? 1 : -1;
3181        case MO_64:
3182            return have_isa_3_10;
3183        }
3184        return 0;
3185    case INDEX_op_bitsel_vec:
3186        return have_vsx;
3187    case INDEX_op_rotrv_vec:
3188        return -1;
3189    default:
3190        return 0;
3191    }
3192}
3193
3194static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
3195                            TCGReg dst, TCGReg src)
3196{
3197    tcg_debug_assert(dst >= TCG_REG_V0);
3198
3199    /* Splat from integer reg allowed via constraints for v3.00.  */
3200    if (src < TCG_REG_V0) {
3201        tcg_debug_assert(have_isa_3_00);
3202        switch (vece) {
3203        case MO_64:
3204            tcg_out32(s, MTVSRDD | VRT(dst) | RA(src) | RB(src));
3205            return true;
3206        case MO_32:
3207            tcg_out32(s, MTVSRWS | VRT(dst) | RA(src));
3208            return true;
3209        default:
3210            /* Fail, so that we fall back on either dupm or mov+dup.  */
3211            return false;
3212        }
3213    }
3214
3215    /*
3216     * Recall we use (or emulate) VSX integer loads, so the integer is
3217     * right justified within the left (zero-index) double-word.
3218     */
3219    switch (vece) {
3220    case MO_8:
3221        tcg_out32(s, VSPLTB | VRT(dst) | VRB(src) | (7 << 16));
3222        break;
3223    case MO_16:
3224        tcg_out32(s, VSPLTH | VRT(dst) | VRB(src) | (3 << 16));
3225        break;
3226    case MO_32:
3227        tcg_out32(s, VSPLTW | VRT(dst) | VRB(src) | (1 << 16));
3228        break;
3229    case MO_64:
3230        if (have_vsx) {
3231            tcg_out32(s, XXPERMDI | VRT(dst) | VRA(src) | VRB(src));
3232            break;
3233        }
3234        tcg_out_vsldoi(s, TCG_VEC_TMP1, src, src, 8);
3235        tcg_out_vsldoi(s, dst, TCG_VEC_TMP1, src, 8);
3236        break;
3237    default:
3238        g_assert_not_reached();
3239    }
3240    return true;
3241}
3242
3243static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
3244                             TCGReg out, TCGReg base, intptr_t offset)
3245{
3246    int elt;
3247
3248    tcg_debug_assert(out >= TCG_REG_V0);
3249    switch (vece) {
3250    case MO_8:
3251        if (have_isa_3_00) {
3252            tcg_out_mem_long(s, LXV, LVX, out, base, offset & -16);
3253        } else {
3254            tcg_out_mem_long(s, 0, LVEBX, out, base, offset);
3255        }
3256        elt = extract32(offset, 0, 4);
3257#if !HOST_BIG_ENDIAN
3258        elt ^= 15;
3259#endif
3260        tcg_out32(s, VSPLTB | VRT(out) | VRB(out) | (elt << 16));
3261        break;
3262    case MO_16:
3263        tcg_debug_assert((offset & 1) == 0);
3264        if (have_isa_3_00) {
3265            tcg_out_mem_long(s, LXV | 8, LVX, out, base, offset & -16);
3266        } else {
3267            tcg_out_mem_long(s, 0, LVEHX, out, base, offset);
3268        }
3269        elt = extract32(offset, 1, 3);
3270#if !HOST_BIG_ENDIAN
3271        elt ^= 7;
3272#endif
3273        tcg_out32(s, VSPLTH | VRT(out) | VRB(out) | (elt << 16));
3274        break;
3275    case MO_32:
3276        if (have_isa_3_00) {
3277            tcg_out_mem_long(s, 0, LXVWSX, out, base, offset);
3278            break;
3279        }
3280        tcg_debug_assert((offset & 3) == 0);
3281        tcg_out_mem_long(s, 0, LVEWX, out, base, offset);
3282        elt = extract32(offset, 2, 2);
3283#if !HOST_BIG_ENDIAN
3284        elt ^= 3;
3285#endif
3286        tcg_out32(s, VSPLTW | VRT(out) | VRB(out) | (elt << 16));
3287        break;
3288    case MO_64:
3289        if (have_vsx) {
3290            tcg_out_mem_long(s, 0, LXVDSX, out, base, offset);
3291            break;
3292        }
3293        tcg_debug_assert((offset & 7) == 0);
3294        tcg_out_mem_long(s, 0, LVX, out, base, offset & -16);
3295        tcg_out_vsldoi(s, TCG_VEC_TMP1, out, out, 8);
3296        elt = extract32(offset, 3, 1);
3297#if !HOST_BIG_ENDIAN
3298        elt = !elt;
3299#endif
3300        if (elt) {
3301            tcg_out_vsldoi(s, out, out, TCG_VEC_TMP1, 8);
3302        } else {
3303            tcg_out_vsldoi(s, out, TCG_VEC_TMP1, out, 8);
3304        }
3305        break;
3306    default:
3307        g_assert_not_reached();
3308    }
3309    return true;
3310}
3311
3312static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
3313                           unsigned vecl, unsigned vece,
3314                           const TCGArg args[TCG_MAX_OP_ARGS],
3315                           const int const_args[TCG_MAX_OP_ARGS])
3316{
3317    static const uint32_t
3318        add_op[4] = { VADDUBM, VADDUHM, VADDUWM, VADDUDM },
3319        sub_op[4] = { VSUBUBM, VSUBUHM, VSUBUWM, VSUBUDM },
3320        mul_op[4] = { 0, 0, VMULUWM, VMULLD },
3321        neg_op[4] = { 0, 0, VNEGW, VNEGD },
3322        eq_op[4]  = { VCMPEQUB, VCMPEQUH, VCMPEQUW, VCMPEQUD },
3323        ne_op[4]  = { VCMPNEB, VCMPNEH, VCMPNEW, 0 },
3324        gts_op[4] = { VCMPGTSB, VCMPGTSH, VCMPGTSW, VCMPGTSD },
3325        gtu_op[4] = { VCMPGTUB, VCMPGTUH, VCMPGTUW, VCMPGTUD },
3326        ssadd_op[4] = { VADDSBS, VADDSHS, VADDSWS, 0 },
3327        usadd_op[4] = { VADDUBS, VADDUHS, VADDUWS, 0 },
3328        sssub_op[4] = { VSUBSBS, VSUBSHS, VSUBSWS, 0 },
3329        ussub_op[4] = { VSUBUBS, VSUBUHS, VSUBUWS, 0 },
3330        umin_op[4] = { VMINUB, VMINUH, VMINUW, VMINUD },
3331        smin_op[4] = { VMINSB, VMINSH, VMINSW, VMINSD },
3332        umax_op[4] = { VMAXUB, VMAXUH, VMAXUW, VMAXUD },
3333        smax_op[4] = { VMAXSB, VMAXSH, VMAXSW, VMAXSD },
3334        shlv_op[4] = { VSLB, VSLH, VSLW, VSLD },
3335        shrv_op[4] = { VSRB, VSRH, VSRW, VSRD },
3336        sarv_op[4] = { VSRAB, VSRAH, VSRAW, VSRAD },
3337        mrgh_op[4] = { VMRGHB, VMRGHH, VMRGHW, 0 },
3338        mrgl_op[4] = { VMRGLB, VMRGLH, VMRGLW, 0 },
3339        muleu_op[4] = { VMULEUB, VMULEUH, VMULEUW, 0 },
3340        mulou_op[4] = { VMULOUB, VMULOUH, VMULOUW, 0 },
3341        pkum_op[4] = { VPKUHUM, VPKUWUM, 0, 0 },
3342        rotl_op[4] = { VRLB, VRLH, VRLW, VRLD };
3343
3344    TCGType type = vecl + TCG_TYPE_V64;
3345    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
3346    uint32_t insn;
3347
3348    switch (opc) {
3349    case INDEX_op_ld_vec:
3350        tcg_out_ld(s, type, a0, a1, a2);
3351        return;
3352    case INDEX_op_st_vec:
3353        tcg_out_st(s, type, a0, a1, a2);
3354        return;
3355    case INDEX_op_dupm_vec:
3356        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
3357        return;
3358
3359    case INDEX_op_add_vec:
3360        insn = add_op[vece];
3361        break;
3362    case INDEX_op_sub_vec:
3363        insn = sub_op[vece];
3364        break;
3365    case INDEX_op_neg_vec:
3366        insn = neg_op[vece];
3367        a2 = a1;
3368        a1 = 0;
3369        break;
3370    case INDEX_op_mul_vec:
3371        insn = mul_op[vece];
3372        break;
3373    case INDEX_op_ssadd_vec:
3374        insn = ssadd_op[vece];
3375        break;
3376    case INDEX_op_sssub_vec:
3377        insn = sssub_op[vece];
3378        break;
3379    case INDEX_op_usadd_vec:
3380        insn = usadd_op[vece];
3381        break;
3382    case INDEX_op_ussub_vec:
3383        insn = ussub_op[vece];
3384        break;
3385    case INDEX_op_smin_vec:
3386        insn = smin_op[vece];
3387        break;
3388    case INDEX_op_umin_vec:
3389        insn = umin_op[vece];
3390        break;
3391    case INDEX_op_smax_vec:
3392        insn = smax_op[vece];
3393        break;
3394    case INDEX_op_umax_vec:
3395        insn = umax_op[vece];
3396        break;
3397    case INDEX_op_shlv_vec:
3398        insn = shlv_op[vece];
3399        break;
3400    case INDEX_op_shrv_vec:
3401        insn = shrv_op[vece];
3402        break;
3403    case INDEX_op_sarv_vec:
3404        insn = sarv_op[vece];
3405        break;
3406    case INDEX_op_and_vec:
3407        insn = VAND;
3408        break;
3409    case INDEX_op_or_vec:
3410        insn = VOR;
3411        break;
3412    case INDEX_op_xor_vec:
3413        insn = VXOR;
3414        break;
3415    case INDEX_op_andc_vec:
3416        insn = VANDC;
3417        break;
3418    case INDEX_op_not_vec:
3419        insn = VNOR;
3420        a2 = a1;
3421        break;
3422    case INDEX_op_orc_vec:
3423        insn = VORC;
3424        break;
3425    case INDEX_op_nand_vec:
3426        insn = VNAND;
3427        break;
3428    case INDEX_op_nor_vec:
3429        insn = VNOR;
3430        break;
3431    case INDEX_op_eqv_vec:
3432        insn = VEQV;
3433        break;
3434
3435    case INDEX_op_cmp_vec:
3436        switch (args[3]) {
3437        case TCG_COND_EQ:
3438            insn = eq_op[vece];
3439            break;
3440        case TCG_COND_NE:
3441            insn = ne_op[vece];
3442            break;
3443        case TCG_COND_GT:
3444            insn = gts_op[vece];
3445            break;
3446        case TCG_COND_GTU:
3447            insn = gtu_op[vece];
3448            break;
3449        default:
3450            g_assert_not_reached();
3451        }
3452        break;
3453
3454    case INDEX_op_bitsel_vec:
3455        tcg_out32(s, XXSEL | VRT(a0) | VRC(a1) | VRB(a2) | VRA(args[3]));
3456        return;
3457
3458    case INDEX_op_dup2_vec:
3459        assert(TCG_TARGET_REG_BITS == 32);
3460        /* With inputs a1 = xLxx, a2 = xHxx  */
3461        tcg_out32(s, VMRGHW | VRT(a0) | VRA(a2) | VRB(a1));  /* a0  = xxHL */
3462        tcg_out_vsldoi(s, TCG_VEC_TMP1, a0, a0, 8);          /* tmp = HLxx */
3463        tcg_out_vsldoi(s, a0, a0, TCG_VEC_TMP1, 8);          /* a0  = HLHL */
3464        return;
3465
3466    case INDEX_op_ppc_mrgh_vec:
3467        insn = mrgh_op[vece];
3468        break;
3469    case INDEX_op_ppc_mrgl_vec:
3470        insn = mrgl_op[vece];
3471        break;
3472    case INDEX_op_ppc_muleu_vec:
3473        insn = muleu_op[vece];
3474        break;
3475    case INDEX_op_ppc_mulou_vec:
3476        insn = mulou_op[vece];
3477        break;
3478    case INDEX_op_ppc_pkum_vec:
3479        insn = pkum_op[vece];
3480        break;
3481    case INDEX_op_rotlv_vec:
3482        insn = rotl_op[vece];
3483        break;
3484    case INDEX_op_ppc_msum_vec:
3485        tcg_debug_assert(vece == MO_16);
3486        tcg_out32(s, VMSUMUHM | VRT(a0) | VRA(a1) | VRB(a2) | VRC(args[3]));
3487        return;
3488
3489    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
3490    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
3491    default:
3492        g_assert_not_reached();
3493    }
3494
3495    tcg_debug_assert(insn != 0);
3496    tcg_out32(s, insn | VRT(a0) | VRA(a1) | VRB(a2));
3497}
3498
3499static void expand_vec_shi(TCGType type, unsigned vece, TCGv_vec v0,
3500                           TCGv_vec v1, TCGArg imm, TCGOpcode opci)
3501{
3502    TCGv_vec t1;
3503
3504    if (vece == MO_32) {
3505        /*
3506         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
3507         * So using negative numbers gets us the 4th bit easily.
3508         */
3509        imm = sextract32(imm, 0, 5);
3510    } else {
3511        imm &= (8 << vece) - 1;
3512    }
3513
3514    /* Splat w/bytes for xxspltib when 2.07 allows MO_64. */
3515    t1 = tcg_constant_vec(type, MO_8, imm);
3516    vec_gen_3(opci, type, vece, tcgv_vec_arg(v0),
3517              tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3518}
3519
3520static void expand_vec_cmp(TCGType type, unsigned vece, TCGv_vec v0,
3521                           TCGv_vec v1, TCGv_vec v2, TCGCond cond)
3522{
3523    bool need_swap = false, need_inv = false;
3524
3525    tcg_debug_assert(vece <= MO_32 || have_isa_2_07);
3526
3527    switch (cond) {
3528    case TCG_COND_EQ:
3529    case TCG_COND_GT:
3530    case TCG_COND_GTU:
3531        break;
3532    case TCG_COND_NE:
3533        if (have_isa_3_00 && vece <= MO_32) {
3534            break;
3535        }
3536        /* fall through */
3537    case TCG_COND_LE:
3538    case TCG_COND_LEU:
3539        need_inv = true;
3540        break;
3541    case TCG_COND_LT:
3542    case TCG_COND_LTU:
3543        need_swap = true;
3544        break;
3545    case TCG_COND_GE:
3546    case TCG_COND_GEU:
3547        need_swap = need_inv = true;
3548        break;
3549    default:
3550        g_assert_not_reached();
3551    }
3552
3553    if (need_inv) {
3554        cond = tcg_invert_cond(cond);
3555    }
3556    if (need_swap) {
3557        TCGv_vec t1;
3558        t1 = v1, v1 = v2, v2 = t1;
3559        cond = tcg_swap_cond(cond);
3560    }
3561
3562    vec_gen_4(INDEX_op_cmp_vec, type, vece, tcgv_vec_arg(v0),
3563              tcgv_vec_arg(v1), tcgv_vec_arg(v2), cond);
3564
3565    if (need_inv) {
3566        tcg_gen_not_vec(vece, v0, v0);
3567    }
3568}
3569
3570static void expand_vec_mul(TCGType type, unsigned vece, TCGv_vec v0,
3571                           TCGv_vec v1, TCGv_vec v2)
3572{
3573    TCGv_vec t1 = tcg_temp_new_vec(type);
3574    TCGv_vec t2 = tcg_temp_new_vec(type);
3575    TCGv_vec c0, c16;
3576
3577    switch (vece) {
3578    case MO_8:
3579    case MO_16:
3580        vec_gen_3(INDEX_op_ppc_muleu_vec, type, vece, tcgv_vec_arg(t1),
3581                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3582        vec_gen_3(INDEX_op_ppc_mulou_vec, type, vece, tcgv_vec_arg(t2),
3583                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3584        vec_gen_3(INDEX_op_ppc_mrgh_vec, type, vece + 1, tcgv_vec_arg(v0),
3585                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3586        vec_gen_3(INDEX_op_ppc_mrgl_vec, type, vece + 1, tcgv_vec_arg(t1),
3587                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3588        vec_gen_3(INDEX_op_ppc_pkum_vec, type, vece, tcgv_vec_arg(v0),
3589                  tcgv_vec_arg(v0), tcgv_vec_arg(t1));
3590	break;
3591
3592    case MO_32:
3593        tcg_debug_assert(!have_isa_2_07);
3594        /*
3595         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
3596         * So using -16 is a quick way to represent 16.
3597         */
3598        c16 = tcg_constant_vec(type, MO_8, -16);
3599        c0 = tcg_constant_vec(type, MO_8, 0);
3600
3601        vec_gen_3(INDEX_op_rotlv_vec, type, MO_32, tcgv_vec_arg(t1),
3602                  tcgv_vec_arg(v2), tcgv_vec_arg(c16));
3603        vec_gen_3(INDEX_op_ppc_mulou_vec, type, MO_16, tcgv_vec_arg(t2),
3604                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3605        vec_gen_4(INDEX_op_ppc_msum_vec, type, MO_16, tcgv_vec_arg(t1),
3606                  tcgv_vec_arg(v1), tcgv_vec_arg(t1), tcgv_vec_arg(c0));
3607        vec_gen_3(INDEX_op_shlv_vec, type, MO_32, tcgv_vec_arg(t1),
3608                  tcgv_vec_arg(t1), tcgv_vec_arg(c16));
3609        tcg_gen_add_vec(MO_32, v0, t1, t2);
3610        break;
3611
3612    default:
3613        g_assert_not_reached();
3614    }
3615    tcg_temp_free_vec(t1);
3616    tcg_temp_free_vec(t2);
3617}
3618
3619void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3620                       TCGArg a0, ...)
3621{
3622    va_list va;
3623    TCGv_vec v0, v1, v2, t0;
3624    TCGArg a2;
3625
3626    va_start(va, a0);
3627    v0 = temp_tcgv_vec(arg_temp(a0));
3628    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3629    a2 = va_arg(va, TCGArg);
3630
3631    switch (opc) {
3632    case INDEX_op_shli_vec:
3633        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shlv_vec);
3634        break;
3635    case INDEX_op_shri_vec:
3636        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shrv_vec);
3637        break;
3638    case INDEX_op_sari_vec:
3639        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_sarv_vec);
3640        break;
3641    case INDEX_op_rotli_vec:
3642        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_rotlv_vec);
3643        break;
3644    case INDEX_op_cmp_vec:
3645        v2 = temp_tcgv_vec(arg_temp(a2));
3646        expand_vec_cmp(type, vece, v0, v1, v2, va_arg(va, TCGArg));
3647        break;
3648    case INDEX_op_mul_vec:
3649        v2 = temp_tcgv_vec(arg_temp(a2));
3650        expand_vec_mul(type, vece, v0, v1, v2);
3651        break;
3652    case INDEX_op_rotlv_vec:
3653        v2 = temp_tcgv_vec(arg_temp(a2));
3654        t0 = tcg_temp_new_vec(type);
3655        tcg_gen_neg_vec(vece, t0, v2);
3656        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3657        tcg_temp_free_vec(t0);
3658        break;
3659    default:
3660        g_assert_not_reached();
3661    }
3662    va_end(va);
3663}
3664
3665static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
3666{
3667    switch (op) {
3668    case INDEX_op_goto_ptr:
3669        return C_O0_I1(r);
3670
3671    case INDEX_op_ld8u_i32:
3672    case INDEX_op_ld8s_i32:
3673    case INDEX_op_ld16u_i32:
3674    case INDEX_op_ld16s_i32:
3675    case INDEX_op_ld_i32:
3676    case INDEX_op_ctpop_i32:
3677    case INDEX_op_neg_i32:
3678    case INDEX_op_not_i32:
3679    case INDEX_op_ext8s_i32:
3680    case INDEX_op_ext16s_i32:
3681    case INDEX_op_bswap16_i32:
3682    case INDEX_op_bswap32_i32:
3683    case INDEX_op_extract_i32:
3684    case INDEX_op_ld8u_i64:
3685    case INDEX_op_ld8s_i64:
3686    case INDEX_op_ld16u_i64:
3687    case INDEX_op_ld16s_i64:
3688    case INDEX_op_ld32u_i64:
3689    case INDEX_op_ld32s_i64:
3690    case INDEX_op_ld_i64:
3691    case INDEX_op_ctpop_i64:
3692    case INDEX_op_neg_i64:
3693    case INDEX_op_not_i64:
3694    case INDEX_op_ext8s_i64:
3695    case INDEX_op_ext16s_i64:
3696    case INDEX_op_ext32s_i64:
3697    case INDEX_op_ext_i32_i64:
3698    case INDEX_op_extu_i32_i64:
3699    case INDEX_op_bswap16_i64:
3700    case INDEX_op_bswap32_i64:
3701    case INDEX_op_bswap64_i64:
3702    case INDEX_op_extract_i64:
3703        return C_O1_I1(r, r);
3704
3705    case INDEX_op_st8_i32:
3706    case INDEX_op_st16_i32:
3707    case INDEX_op_st_i32:
3708    case INDEX_op_st8_i64:
3709    case INDEX_op_st16_i64:
3710    case INDEX_op_st32_i64:
3711    case INDEX_op_st_i64:
3712        return C_O0_I2(r, r);
3713
3714    case INDEX_op_add_i32:
3715    case INDEX_op_and_i32:
3716    case INDEX_op_or_i32:
3717    case INDEX_op_xor_i32:
3718    case INDEX_op_andc_i32:
3719    case INDEX_op_orc_i32:
3720    case INDEX_op_eqv_i32:
3721    case INDEX_op_shl_i32:
3722    case INDEX_op_shr_i32:
3723    case INDEX_op_sar_i32:
3724    case INDEX_op_rotl_i32:
3725    case INDEX_op_rotr_i32:
3726    case INDEX_op_setcond_i32:
3727    case INDEX_op_and_i64:
3728    case INDEX_op_andc_i64:
3729    case INDEX_op_shl_i64:
3730    case INDEX_op_shr_i64:
3731    case INDEX_op_sar_i64:
3732    case INDEX_op_rotl_i64:
3733    case INDEX_op_rotr_i64:
3734    case INDEX_op_setcond_i64:
3735        return C_O1_I2(r, r, ri);
3736
3737    case INDEX_op_mul_i32:
3738    case INDEX_op_mul_i64:
3739        return C_O1_I2(r, r, rI);
3740
3741    case INDEX_op_div_i32:
3742    case INDEX_op_divu_i32:
3743    case INDEX_op_rem_i32:
3744    case INDEX_op_remu_i32:
3745    case INDEX_op_nand_i32:
3746    case INDEX_op_nor_i32:
3747    case INDEX_op_muluh_i32:
3748    case INDEX_op_mulsh_i32:
3749    case INDEX_op_orc_i64:
3750    case INDEX_op_eqv_i64:
3751    case INDEX_op_nand_i64:
3752    case INDEX_op_nor_i64:
3753    case INDEX_op_div_i64:
3754    case INDEX_op_divu_i64:
3755    case INDEX_op_rem_i64:
3756    case INDEX_op_remu_i64:
3757    case INDEX_op_mulsh_i64:
3758    case INDEX_op_muluh_i64:
3759        return C_O1_I2(r, r, r);
3760
3761    case INDEX_op_sub_i32:
3762        return C_O1_I2(r, rI, ri);
3763    case INDEX_op_add_i64:
3764        return C_O1_I2(r, r, rT);
3765    case INDEX_op_or_i64:
3766    case INDEX_op_xor_i64:
3767        return C_O1_I2(r, r, rU);
3768    case INDEX_op_sub_i64:
3769        return C_O1_I2(r, rI, rT);
3770    case INDEX_op_clz_i32:
3771    case INDEX_op_ctz_i32:
3772    case INDEX_op_clz_i64:
3773    case INDEX_op_ctz_i64:
3774        return C_O1_I2(r, r, rZW);
3775
3776    case INDEX_op_brcond_i32:
3777    case INDEX_op_brcond_i64:
3778        return C_O0_I2(r, ri);
3779
3780    case INDEX_op_movcond_i32:
3781    case INDEX_op_movcond_i64:
3782        return C_O1_I4(r, r, ri, rZ, rZ);
3783    case INDEX_op_deposit_i32:
3784    case INDEX_op_deposit_i64:
3785        return C_O1_I2(r, 0, rZ);
3786    case INDEX_op_brcond2_i32:
3787        return C_O0_I4(r, r, ri, ri);
3788    case INDEX_op_setcond2_i32:
3789        return C_O1_I4(r, r, r, ri, ri);
3790    case INDEX_op_add2_i64:
3791    case INDEX_op_add2_i32:
3792        return C_O2_I4(r, r, r, r, rI, rZM);
3793    case INDEX_op_sub2_i64:
3794    case INDEX_op_sub2_i32:
3795        return C_O2_I4(r, r, rI, rZM, r, r);
3796
3797    case INDEX_op_qemu_ld_i32:
3798        return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
3799                ? C_O1_I1(r, L)
3800                : C_O1_I2(r, L, L));
3801
3802    case INDEX_op_qemu_st_i32:
3803        return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
3804                ? C_O0_I2(S, S)
3805                : C_O0_I3(S, S, S));
3806
3807    case INDEX_op_qemu_ld_i64:
3808        return (TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, L)
3809                : TARGET_LONG_BITS == 32 ? C_O2_I1(L, L, L)
3810                : C_O2_I2(L, L, L, L));
3811
3812    case INDEX_op_qemu_st_i64:
3813        return (TCG_TARGET_REG_BITS == 64 ? C_O0_I2(S, S)
3814                : TARGET_LONG_BITS == 32 ? C_O0_I3(S, S, S)
3815                : C_O0_I4(S, S, S, S));
3816
3817    case INDEX_op_add_vec:
3818    case INDEX_op_sub_vec:
3819    case INDEX_op_mul_vec:
3820    case INDEX_op_and_vec:
3821    case INDEX_op_or_vec:
3822    case INDEX_op_xor_vec:
3823    case INDEX_op_andc_vec:
3824    case INDEX_op_orc_vec:
3825    case INDEX_op_nor_vec:
3826    case INDEX_op_eqv_vec:
3827    case INDEX_op_nand_vec:
3828    case INDEX_op_cmp_vec:
3829    case INDEX_op_ssadd_vec:
3830    case INDEX_op_sssub_vec:
3831    case INDEX_op_usadd_vec:
3832    case INDEX_op_ussub_vec:
3833    case INDEX_op_smax_vec:
3834    case INDEX_op_smin_vec:
3835    case INDEX_op_umax_vec:
3836    case INDEX_op_umin_vec:
3837    case INDEX_op_shlv_vec:
3838    case INDEX_op_shrv_vec:
3839    case INDEX_op_sarv_vec:
3840    case INDEX_op_rotlv_vec:
3841    case INDEX_op_rotrv_vec:
3842    case INDEX_op_ppc_mrgh_vec:
3843    case INDEX_op_ppc_mrgl_vec:
3844    case INDEX_op_ppc_muleu_vec:
3845    case INDEX_op_ppc_mulou_vec:
3846    case INDEX_op_ppc_pkum_vec:
3847    case INDEX_op_dup2_vec:
3848        return C_O1_I2(v, v, v);
3849
3850    case INDEX_op_not_vec:
3851    case INDEX_op_neg_vec:
3852        return C_O1_I1(v, v);
3853
3854    case INDEX_op_dup_vec:
3855        return have_isa_3_00 ? C_O1_I1(v, vr) : C_O1_I1(v, v);
3856
3857    case INDEX_op_ld_vec:
3858    case INDEX_op_dupm_vec:
3859        return C_O1_I1(v, r);
3860
3861    case INDEX_op_st_vec:
3862        return C_O0_I2(v, r);
3863
3864    case INDEX_op_bitsel_vec:
3865    case INDEX_op_ppc_msum_vec:
3866        return C_O1_I3(v, v, v, v);
3867
3868    default:
3869        g_assert_not_reached();
3870    }
3871}
3872
3873static void tcg_target_init(TCGContext *s)
3874{
3875    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3876    unsigned long hwcap2 = qemu_getauxval(AT_HWCAP2);
3877
3878    have_isa = tcg_isa_base;
3879    if (hwcap & PPC_FEATURE_ARCH_2_06) {
3880        have_isa = tcg_isa_2_06;
3881    }
3882#ifdef PPC_FEATURE2_ARCH_2_07
3883    if (hwcap2 & PPC_FEATURE2_ARCH_2_07) {
3884        have_isa = tcg_isa_2_07;
3885    }
3886#endif
3887#ifdef PPC_FEATURE2_ARCH_3_00
3888    if (hwcap2 & PPC_FEATURE2_ARCH_3_00) {
3889        have_isa = tcg_isa_3_00;
3890    }
3891#endif
3892#ifdef PPC_FEATURE2_ARCH_3_10
3893    if (hwcap2 & PPC_FEATURE2_ARCH_3_10) {
3894        have_isa = tcg_isa_3_10;
3895    }
3896#endif
3897
3898#ifdef PPC_FEATURE2_HAS_ISEL
3899    /* Prefer explicit instruction from the kernel. */
3900    have_isel = (hwcap2 & PPC_FEATURE2_HAS_ISEL) != 0;
3901#else
3902    /* Fall back to knowing Power7 (2.06) has ISEL. */
3903    have_isel = have_isa_2_06;
3904#endif
3905
3906    if (hwcap & PPC_FEATURE_HAS_ALTIVEC) {
3907        have_altivec = true;
3908        /* We only care about the portion of VSX that overlaps Altivec. */
3909        if (hwcap & PPC_FEATURE_HAS_VSX) {
3910            have_vsx = true;
3911        }
3912    }
3913
3914    tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
3915    tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
3916    if (have_altivec) {
3917        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3918        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3919    }
3920
3921    tcg_target_call_clobber_regs = 0;
3922    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3923    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3924    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3925    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3926    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3927    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3928    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R7);
3929    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R8);
3930    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R9);
3931    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R10);
3932    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R11);
3933    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
3934
3935    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3936    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3937    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3938    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3939    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3940    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3941    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3942    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3943    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V8);
3944    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V9);
3945    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V10);
3946    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V11);
3947    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V12);
3948    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V13);
3949    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V14);
3950    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V15);
3951    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3952    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3953    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3954    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3955
3956    s->reserved_regs = 0;
3957    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); /* tcg temp */
3958    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R1); /* stack pointer */
3959#if defined(_CALL_SYSV)
3960    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R2); /* toc pointer */
3961#endif
3962#if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
3963    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R13); /* thread pointer */
3964#endif
3965    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1); /* mem temp */
3966    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP1);
3967    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP2);
3968    if (USE_REG_TB) {
3969        tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);  /* tb->tc_ptr */
3970    }
3971}
3972
3973#ifdef __ELF__
3974typedef struct {
3975    DebugFrameCIE cie;
3976    DebugFrameFDEHeader fde;
3977    uint8_t fde_def_cfa[4];
3978    uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2 + 3];
3979} DebugFrame;
3980
3981/* We're expecting a 2 byte uleb128 encoded value.  */
3982QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3983
3984#if TCG_TARGET_REG_BITS == 64
3985# define ELF_HOST_MACHINE EM_PPC64
3986#else
3987# define ELF_HOST_MACHINE EM_PPC
3988#endif
3989
3990static DebugFrame debug_frame = {
3991    .cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3992    .cie.id = -1,
3993    .cie.version = 1,
3994    .cie.code_align = 1,
3995    .cie.data_align = (-SZR & 0x7f),         /* sleb128 -SZR */
3996    .cie.return_column = 65,
3997
3998    /* Total FDE size does not include the "len" member.  */
3999    .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset),
4000
4001    .fde_def_cfa = {
4002        12, TCG_REG_R1,                 /* DW_CFA_def_cfa r1, ... */
4003        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
4004        (FRAME_SIZE >> 7)
4005    },
4006    .fde_reg_ofs = {
4007        /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
4008        0x11, 65, (LR_OFFSET / -SZR) & 0x7f,
4009    }
4010};
4011
4012void tcg_register_jit(const void *buf, size_t buf_size)
4013{
4014    uint8_t *p = &debug_frame.fde_reg_ofs[3];
4015    int i;
4016
4017    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i, p += 2) {
4018        p[0] = 0x80 + tcg_target_callee_save_regs[i];
4019        p[1] = (FRAME_SIZE - (REG_SAVE_BOT + i * SZR)) / SZR;
4020    }
4021
4022    debug_frame.fde.func_start = (uintptr_t)buf;
4023    debug_frame.fde.func_len = buf_size;
4024
4025    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
4026}
4027#endif /* __ELF__ */
4028#undef VMULEUB
4029#undef VMULEUH
4030#undef VMULEUW
4031#undef VMULOUB
4032#undef VMULOUH
4033#undef VMULOUW
4034#undef VMSUMUHM
4035