xref: /openbmc/qemu/tcg/ppc/tcg-target.c.inc (revision 9b89cdb2)
1/*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2008 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25#include "elf.h"
26#include "../tcg-pool.c.inc"
27
28/*
29 * Standardize on the _CALL_FOO symbols used by GCC:
30 * Apple XCode does not define _CALL_DARWIN.
31 * Clang defines _CALL_ELF (64-bit) but not _CALL_SYSV (32-bit).
32 */
33#if !defined(_CALL_SYSV) && \
34    !defined(_CALL_DARWIN) && \
35    !defined(_CALL_AIX) && \
36    !defined(_CALL_ELF)
37# if defined(__APPLE__)
38#  define _CALL_DARWIN
39# elif defined(__ELF__) && TCG_TARGET_REG_BITS == 32
40#  define _CALL_SYSV
41# else
42#  error "Unknown ABI"
43# endif
44#endif
45
46#ifdef _CALL_SYSV
47# define TCG_TARGET_CALL_ALIGN_ARGS   1
48#endif
49
50/* For some memory operations, we need a scratch that isn't R0.  For the AIX
51   calling convention, we can re-use the TOC register since we'll be reloading
52   it at every call.  Otherwise R12 will do nicely as neither a call-saved
53   register nor a parameter register.  */
54#ifdef _CALL_AIX
55# define TCG_REG_TMP1   TCG_REG_R2
56#else
57# define TCG_REG_TMP1   TCG_REG_R12
58#endif
59
60#define TCG_VEC_TMP1    TCG_REG_V0
61#define TCG_VEC_TMP2    TCG_REG_V1
62
63#define TCG_REG_TB     TCG_REG_R31
64#define USE_REG_TB     (TCG_TARGET_REG_BITS == 64)
65
66/* Shorthand for size of a pointer.  Avoid promotion to unsigned.  */
67#define SZP  ((int)sizeof(void *))
68
69/* Shorthand for size of a register.  */
70#define SZR  (TCG_TARGET_REG_BITS / 8)
71
72#define TCG_CT_CONST_S16  0x100
73#define TCG_CT_CONST_U16  0x200
74#define TCG_CT_CONST_S32  0x400
75#define TCG_CT_CONST_U32  0x800
76#define TCG_CT_CONST_ZERO 0x1000
77#define TCG_CT_CONST_MONE 0x2000
78#define TCG_CT_CONST_WSZ  0x4000
79
80#define ALL_GENERAL_REGS  0xffffffffu
81#define ALL_VECTOR_REGS   0xffffffff00000000ull
82
83#ifdef CONFIG_SOFTMMU
84#define ALL_QLOAD_REGS \
85    (ALL_GENERAL_REGS & \
86     ~((1 << TCG_REG_R3) | (1 << TCG_REG_R4) | (1 << TCG_REG_R5)))
87#define ALL_QSTORE_REGS \
88    (ALL_GENERAL_REGS & ~((1 << TCG_REG_R3) | (1 << TCG_REG_R4) | \
89                          (1 << TCG_REG_R5) | (1 << TCG_REG_R6)))
90#else
91#define ALL_QLOAD_REGS  (ALL_GENERAL_REGS & ~(1 << TCG_REG_R3))
92#define ALL_QSTORE_REGS ALL_QLOAD_REGS
93#endif
94
95TCGPowerISA have_isa;
96static bool have_isel;
97bool have_altivec;
98bool have_vsx;
99
100#ifndef CONFIG_SOFTMMU
101#define TCG_GUEST_BASE_REG 30
102#endif
103
104#ifdef CONFIG_DEBUG_TCG
105static const char tcg_target_reg_names[TCG_TARGET_NB_REGS][4] = {
106    "r0",  "r1",  "r2",  "r3",  "r4",  "r5",  "r6",  "r7",
107    "r8",  "r9",  "r10", "r11", "r12", "r13", "r14", "r15",
108    "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
109    "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31",
110    "v0",  "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7",
111    "v8",  "v9",  "v10", "v11", "v12", "v13", "v14", "v15",
112    "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
113    "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",
114};
115#endif
116
117static const int tcg_target_reg_alloc_order[] = {
118    TCG_REG_R14,  /* call saved registers */
119    TCG_REG_R15,
120    TCG_REG_R16,
121    TCG_REG_R17,
122    TCG_REG_R18,
123    TCG_REG_R19,
124    TCG_REG_R20,
125    TCG_REG_R21,
126    TCG_REG_R22,
127    TCG_REG_R23,
128    TCG_REG_R24,
129    TCG_REG_R25,
130    TCG_REG_R26,
131    TCG_REG_R27,
132    TCG_REG_R28,
133    TCG_REG_R29,
134    TCG_REG_R30,
135    TCG_REG_R31,
136    TCG_REG_R12,  /* call clobbered, non-arguments */
137    TCG_REG_R11,
138    TCG_REG_R2,
139    TCG_REG_R13,
140    TCG_REG_R10,  /* call clobbered, arguments */
141    TCG_REG_R9,
142    TCG_REG_R8,
143    TCG_REG_R7,
144    TCG_REG_R6,
145    TCG_REG_R5,
146    TCG_REG_R4,
147    TCG_REG_R3,
148
149    /* V0 and V1 reserved as temporaries; V20 - V31 are call-saved */
150    TCG_REG_V2,   /* call clobbered, vectors */
151    TCG_REG_V3,
152    TCG_REG_V4,
153    TCG_REG_V5,
154    TCG_REG_V6,
155    TCG_REG_V7,
156    TCG_REG_V8,
157    TCG_REG_V9,
158    TCG_REG_V10,
159    TCG_REG_V11,
160    TCG_REG_V12,
161    TCG_REG_V13,
162    TCG_REG_V14,
163    TCG_REG_V15,
164    TCG_REG_V16,
165    TCG_REG_V17,
166    TCG_REG_V18,
167    TCG_REG_V19,
168};
169
170static const int tcg_target_call_iarg_regs[] = {
171    TCG_REG_R3,
172    TCG_REG_R4,
173    TCG_REG_R5,
174    TCG_REG_R6,
175    TCG_REG_R7,
176    TCG_REG_R8,
177    TCG_REG_R9,
178    TCG_REG_R10
179};
180
181static const int tcg_target_call_oarg_regs[] = {
182    TCG_REG_R3,
183    TCG_REG_R4
184};
185
186static const int tcg_target_callee_save_regs[] = {
187#ifdef _CALL_DARWIN
188    TCG_REG_R11,
189#endif
190    TCG_REG_R14,
191    TCG_REG_R15,
192    TCG_REG_R16,
193    TCG_REG_R17,
194    TCG_REG_R18,
195    TCG_REG_R19,
196    TCG_REG_R20,
197    TCG_REG_R21,
198    TCG_REG_R22,
199    TCG_REG_R23,
200    TCG_REG_R24,
201    TCG_REG_R25,
202    TCG_REG_R26,
203    TCG_REG_R27, /* currently used for the global env */
204    TCG_REG_R28,
205    TCG_REG_R29,
206    TCG_REG_R30,
207    TCG_REG_R31
208};
209
210static inline bool in_range_b(tcg_target_long target)
211{
212    return target == sextract64(target, 0, 26);
213}
214
215static uint32_t reloc_pc24_val(const tcg_insn_unit *pc,
216			       const tcg_insn_unit *target)
217{
218    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
219    tcg_debug_assert(in_range_b(disp));
220    return disp & 0x3fffffc;
221}
222
223static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
224{
225    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
226    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
227
228    if (in_range_b(disp)) {
229        *src_rw = (*src_rw & ~0x3fffffc) | (disp & 0x3fffffc);
230        return true;
231    }
232    return false;
233}
234
235static uint16_t reloc_pc14_val(const tcg_insn_unit *pc,
236			       const tcg_insn_unit *target)
237{
238    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
239    tcg_debug_assert(disp == (int16_t) disp);
240    return disp & 0xfffc;
241}
242
243static bool reloc_pc14(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
244{
245    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
246    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
247
248    if (disp == (int16_t) disp) {
249        *src_rw = (*src_rw & ~0xfffc) | (disp & 0xfffc);
250        return true;
251    }
252    return false;
253}
254
255/* test if a constant matches the constraint */
256static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
257{
258    if (ct & TCG_CT_CONST) {
259        return 1;
260    }
261
262    /* The only 32-bit constraint we use aside from
263       TCG_CT_CONST is TCG_CT_CONST_S16.  */
264    if (type == TCG_TYPE_I32) {
265        val = (int32_t)val;
266    }
267
268    if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
269        return 1;
270    } else if ((ct & TCG_CT_CONST_U16) && val == (uint16_t)val) {
271        return 1;
272    } else if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
273        return 1;
274    } else if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
275        return 1;
276    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
277        return 1;
278    } else if ((ct & TCG_CT_CONST_MONE) && val == -1) {
279        return 1;
280    } else if ((ct & TCG_CT_CONST_WSZ)
281               && val == (type == TCG_TYPE_I32 ? 32 : 64)) {
282        return 1;
283    }
284    return 0;
285}
286
287#define OPCD(opc) ((opc)<<26)
288#define XO19(opc) (OPCD(19)|((opc)<<1))
289#define MD30(opc) (OPCD(30)|((opc)<<2))
290#define MDS30(opc) (OPCD(30)|((opc)<<1))
291#define XO31(opc) (OPCD(31)|((opc)<<1))
292#define XO58(opc) (OPCD(58)|(opc))
293#define XO62(opc) (OPCD(62)|(opc))
294#define VX4(opc)  (OPCD(4)|(opc))
295
296#define B      OPCD( 18)
297#define BC     OPCD( 16)
298#define LBZ    OPCD( 34)
299#define LHZ    OPCD( 40)
300#define LHA    OPCD( 42)
301#define LWZ    OPCD( 32)
302#define LWZUX  XO31( 55)
303#define STB    OPCD( 38)
304#define STH    OPCD( 44)
305#define STW    OPCD( 36)
306
307#define STD    XO62(  0)
308#define STDU   XO62(  1)
309#define STDX   XO31(149)
310
311#define LD     XO58(  0)
312#define LDX    XO31( 21)
313#define LDU    XO58(  1)
314#define LDUX   XO31( 53)
315#define LWA    XO58(  2)
316#define LWAX   XO31(341)
317
318#define ADDIC  OPCD( 12)
319#define ADDI   OPCD( 14)
320#define ADDIS  OPCD( 15)
321#define ORI    OPCD( 24)
322#define ORIS   OPCD( 25)
323#define XORI   OPCD( 26)
324#define XORIS  OPCD( 27)
325#define ANDI   OPCD( 28)
326#define ANDIS  OPCD( 29)
327#define MULLI  OPCD(  7)
328#define CMPLI  OPCD( 10)
329#define CMPI   OPCD( 11)
330#define SUBFIC OPCD( 8)
331
332#define LWZU   OPCD( 33)
333#define STWU   OPCD( 37)
334
335#define RLWIMI OPCD( 20)
336#define RLWINM OPCD( 21)
337#define RLWNM  OPCD( 23)
338
339#define RLDICL MD30(  0)
340#define RLDICR MD30(  1)
341#define RLDIMI MD30(  3)
342#define RLDCL  MDS30( 8)
343
344#define BCLR   XO19( 16)
345#define BCCTR  XO19(528)
346#define CRAND  XO19(257)
347#define CRANDC XO19(129)
348#define CRNAND XO19(225)
349#define CROR   XO19(449)
350#define CRNOR  XO19( 33)
351
352#define EXTSB  XO31(954)
353#define EXTSH  XO31(922)
354#define EXTSW  XO31(986)
355#define ADD    XO31(266)
356#define ADDE   XO31(138)
357#define ADDME  XO31(234)
358#define ADDZE  XO31(202)
359#define ADDC   XO31( 10)
360#define AND    XO31( 28)
361#define SUBF   XO31( 40)
362#define SUBFC  XO31(  8)
363#define SUBFE  XO31(136)
364#define SUBFME XO31(232)
365#define SUBFZE XO31(200)
366#define OR     XO31(444)
367#define XOR    XO31(316)
368#define MULLW  XO31(235)
369#define MULHW  XO31( 75)
370#define MULHWU XO31( 11)
371#define DIVW   XO31(491)
372#define DIVWU  XO31(459)
373#define CMP    XO31(  0)
374#define CMPL   XO31( 32)
375#define LHBRX  XO31(790)
376#define LWBRX  XO31(534)
377#define LDBRX  XO31(532)
378#define STHBRX XO31(918)
379#define STWBRX XO31(662)
380#define STDBRX XO31(660)
381#define MFSPR  XO31(339)
382#define MTSPR  XO31(467)
383#define SRAWI  XO31(824)
384#define NEG    XO31(104)
385#define MFCR   XO31( 19)
386#define MFOCRF (MFCR | (1u << 20))
387#define NOR    XO31(124)
388#define CNTLZW XO31( 26)
389#define CNTLZD XO31( 58)
390#define CNTTZW XO31(538)
391#define CNTTZD XO31(570)
392#define CNTPOPW XO31(378)
393#define CNTPOPD XO31(506)
394#define ANDC   XO31( 60)
395#define ORC    XO31(412)
396#define EQV    XO31(284)
397#define NAND   XO31(476)
398#define ISEL   XO31( 15)
399
400#define MULLD  XO31(233)
401#define MULHD  XO31( 73)
402#define MULHDU XO31(  9)
403#define DIVD   XO31(489)
404#define DIVDU  XO31(457)
405
406#define LBZX   XO31( 87)
407#define LHZX   XO31(279)
408#define LHAX   XO31(343)
409#define LWZX   XO31( 23)
410#define STBX   XO31(215)
411#define STHX   XO31(407)
412#define STWX   XO31(151)
413
414#define EIEIO  XO31(854)
415#define HWSYNC XO31(598)
416#define LWSYNC (HWSYNC | (1u << 21))
417
418#define SPR(a, b) ((((a)<<5)|(b))<<11)
419#define LR     SPR(8, 0)
420#define CTR    SPR(9, 0)
421
422#define SLW    XO31( 24)
423#define SRW    XO31(536)
424#define SRAW   XO31(792)
425
426#define SLD    XO31( 27)
427#define SRD    XO31(539)
428#define SRAD   XO31(794)
429#define SRADI  XO31(413<<1)
430
431#define BRH    XO31(219)
432#define BRW    XO31(155)
433#define BRD    XO31(187)
434
435#define TW     XO31( 4)
436#define TRAP   (TW | TO(31))
437
438#define NOP    ORI  /* ori 0,0,0 */
439
440#define LVX        XO31(103)
441#define LVEBX      XO31(7)
442#define LVEHX      XO31(39)
443#define LVEWX      XO31(71)
444#define LXSDX      (XO31(588) | 1)  /* v2.06, force tx=1 */
445#define LXVDSX     (XO31(332) | 1)  /* v2.06, force tx=1 */
446#define LXSIWZX    (XO31(12) | 1)   /* v2.07, force tx=1 */
447#define LXV        (OPCD(61) | 8 | 1)  /* v3.00, force tx=1 */
448#define LXSD       (OPCD(57) | 2)   /* v3.00 */
449#define LXVWSX     (XO31(364) | 1)  /* v3.00, force tx=1 */
450
451#define STVX       XO31(231)
452#define STVEWX     XO31(199)
453#define STXSDX     (XO31(716) | 1)  /* v2.06, force sx=1 */
454#define STXSIWX    (XO31(140) | 1)  /* v2.07, force sx=1 */
455#define STXV       (OPCD(61) | 8 | 5) /* v3.00, force sx=1 */
456#define STXSD      (OPCD(61) | 2)   /* v3.00 */
457
458#define VADDSBS    VX4(768)
459#define VADDUBS    VX4(512)
460#define VADDUBM    VX4(0)
461#define VADDSHS    VX4(832)
462#define VADDUHS    VX4(576)
463#define VADDUHM    VX4(64)
464#define VADDSWS    VX4(896)
465#define VADDUWS    VX4(640)
466#define VADDUWM    VX4(128)
467#define VADDUDM    VX4(192)       /* v2.07 */
468
469#define VSUBSBS    VX4(1792)
470#define VSUBUBS    VX4(1536)
471#define VSUBUBM    VX4(1024)
472#define VSUBSHS    VX4(1856)
473#define VSUBUHS    VX4(1600)
474#define VSUBUHM    VX4(1088)
475#define VSUBSWS    VX4(1920)
476#define VSUBUWS    VX4(1664)
477#define VSUBUWM    VX4(1152)
478#define VSUBUDM    VX4(1216)      /* v2.07 */
479
480#define VNEGW      (VX4(1538) | (6 << 16))  /* v3.00 */
481#define VNEGD      (VX4(1538) | (7 << 16))  /* v3.00 */
482
483#define VMAXSB     VX4(258)
484#define VMAXSH     VX4(322)
485#define VMAXSW     VX4(386)
486#define VMAXSD     VX4(450)       /* v2.07 */
487#define VMAXUB     VX4(2)
488#define VMAXUH     VX4(66)
489#define VMAXUW     VX4(130)
490#define VMAXUD     VX4(194)       /* v2.07 */
491#define VMINSB     VX4(770)
492#define VMINSH     VX4(834)
493#define VMINSW     VX4(898)
494#define VMINSD     VX4(962)       /* v2.07 */
495#define VMINUB     VX4(514)
496#define VMINUH     VX4(578)
497#define VMINUW     VX4(642)
498#define VMINUD     VX4(706)       /* v2.07 */
499
500#define VCMPEQUB   VX4(6)
501#define VCMPEQUH   VX4(70)
502#define VCMPEQUW   VX4(134)
503#define VCMPEQUD   VX4(199)       /* v2.07 */
504#define VCMPGTSB   VX4(774)
505#define VCMPGTSH   VX4(838)
506#define VCMPGTSW   VX4(902)
507#define VCMPGTSD   VX4(967)       /* v2.07 */
508#define VCMPGTUB   VX4(518)
509#define VCMPGTUH   VX4(582)
510#define VCMPGTUW   VX4(646)
511#define VCMPGTUD   VX4(711)       /* v2.07 */
512#define VCMPNEB    VX4(7)         /* v3.00 */
513#define VCMPNEH    VX4(71)        /* v3.00 */
514#define VCMPNEW    VX4(135)       /* v3.00 */
515
516#define VSLB       VX4(260)
517#define VSLH       VX4(324)
518#define VSLW       VX4(388)
519#define VSLD       VX4(1476)      /* v2.07 */
520#define VSRB       VX4(516)
521#define VSRH       VX4(580)
522#define VSRW       VX4(644)
523#define VSRD       VX4(1732)      /* v2.07 */
524#define VSRAB      VX4(772)
525#define VSRAH      VX4(836)
526#define VSRAW      VX4(900)
527#define VSRAD      VX4(964)       /* v2.07 */
528#define VRLB       VX4(4)
529#define VRLH       VX4(68)
530#define VRLW       VX4(132)
531#define VRLD       VX4(196)       /* v2.07 */
532
533#define VMULEUB    VX4(520)
534#define VMULEUH    VX4(584)
535#define VMULEUW    VX4(648)       /* v2.07 */
536#define VMULOUB    VX4(8)
537#define VMULOUH    VX4(72)
538#define VMULOUW    VX4(136)       /* v2.07 */
539#define VMULUWM    VX4(137)       /* v2.07 */
540#define VMULLD     VX4(457)       /* v3.10 */
541#define VMSUMUHM   VX4(38)
542
543#define VMRGHB     VX4(12)
544#define VMRGHH     VX4(76)
545#define VMRGHW     VX4(140)
546#define VMRGLB     VX4(268)
547#define VMRGLH     VX4(332)
548#define VMRGLW     VX4(396)
549
550#define VPKUHUM    VX4(14)
551#define VPKUWUM    VX4(78)
552
553#define VAND       VX4(1028)
554#define VANDC      VX4(1092)
555#define VNOR       VX4(1284)
556#define VOR        VX4(1156)
557#define VXOR       VX4(1220)
558#define VEQV       VX4(1668)      /* v2.07 */
559#define VNAND      VX4(1412)      /* v2.07 */
560#define VORC       VX4(1348)      /* v2.07 */
561
562#define VSPLTB     VX4(524)
563#define VSPLTH     VX4(588)
564#define VSPLTW     VX4(652)
565#define VSPLTISB   VX4(780)
566#define VSPLTISH   VX4(844)
567#define VSPLTISW   VX4(908)
568
569#define VSLDOI     VX4(44)
570
571#define XXPERMDI   (OPCD(60) | (10 << 3) | 7)  /* v2.06, force ax=bx=tx=1 */
572#define XXSEL      (OPCD(60) | (3 << 4) | 0xf) /* v2.06, force ax=bx=cx=tx=1 */
573#define XXSPLTIB   (OPCD(60) | (360 << 1) | 1) /* v3.00, force tx=1 */
574
575#define MFVSRD     (XO31(51) | 1)   /* v2.07, force sx=1 */
576#define MFVSRWZ    (XO31(115) | 1)  /* v2.07, force sx=1 */
577#define MTVSRD     (XO31(179) | 1)  /* v2.07, force tx=1 */
578#define MTVSRWZ    (XO31(243) | 1)  /* v2.07, force tx=1 */
579#define MTVSRDD    (XO31(435) | 1)  /* v3.00, force tx=1 */
580#define MTVSRWS    (XO31(403) | 1)  /* v3.00, force tx=1 */
581
582#define RT(r) ((r)<<21)
583#define RS(r) ((r)<<21)
584#define RA(r) ((r)<<16)
585#define RB(r) ((r)<<11)
586#define TO(t) ((t)<<21)
587#define SH(s) ((s)<<11)
588#define MB(b) ((b)<<6)
589#define ME(e) ((e)<<1)
590#define BO(o) ((o)<<21)
591#define MB64(b) ((b)<<5)
592#define FXM(b) (1 << (19 - (b)))
593
594#define VRT(r)  (((r) & 31) << 21)
595#define VRA(r)  (((r) & 31) << 16)
596#define VRB(r)  (((r) & 31) << 11)
597#define VRC(r)  (((r) & 31) <<  6)
598
599#define LK    1
600
601#define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
602#define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
603#define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
604#define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
605
606#define BF(n)    ((n)<<23)
607#define BI(n, c) (((c)+((n)*4))<<16)
608#define BT(n, c) (((c)+((n)*4))<<21)
609#define BA(n, c) (((c)+((n)*4))<<16)
610#define BB(n, c) (((c)+((n)*4))<<11)
611#define BC_(n, c) (((c)+((n)*4))<<6)
612
613#define BO_COND_TRUE  BO(12)
614#define BO_COND_FALSE BO( 4)
615#define BO_ALWAYS     BO(20)
616
617enum {
618    CR_LT,
619    CR_GT,
620    CR_EQ,
621    CR_SO
622};
623
624static const uint32_t tcg_to_bc[] = {
625    [TCG_COND_EQ]  = BC | BI(7, CR_EQ) | BO_COND_TRUE,
626    [TCG_COND_NE]  = BC | BI(7, CR_EQ) | BO_COND_FALSE,
627    [TCG_COND_LT]  = BC | BI(7, CR_LT) | BO_COND_TRUE,
628    [TCG_COND_GE]  = BC | BI(7, CR_LT) | BO_COND_FALSE,
629    [TCG_COND_LE]  = BC | BI(7, CR_GT) | BO_COND_FALSE,
630    [TCG_COND_GT]  = BC | BI(7, CR_GT) | BO_COND_TRUE,
631    [TCG_COND_LTU] = BC | BI(7, CR_LT) | BO_COND_TRUE,
632    [TCG_COND_GEU] = BC | BI(7, CR_LT) | BO_COND_FALSE,
633    [TCG_COND_LEU] = BC | BI(7, CR_GT) | BO_COND_FALSE,
634    [TCG_COND_GTU] = BC | BI(7, CR_GT) | BO_COND_TRUE,
635};
636
637/* The low bit here is set if the RA and RB fields must be inverted.  */
638static const uint32_t tcg_to_isel[] = {
639    [TCG_COND_EQ]  = ISEL | BC_(7, CR_EQ),
640    [TCG_COND_NE]  = ISEL | BC_(7, CR_EQ) | 1,
641    [TCG_COND_LT]  = ISEL | BC_(7, CR_LT),
642    [TCG_COND_GE]  = ISEL | BC_(7, CR_LT) | 1,
643    [TCG_COND_LE]  = ISEL | BC_(7, CR_GT) | 1,
644    [TCG_COND_GT]  = ISEL | BC_(7, CR_GT),
645    [TCG_COND_LTU] = ISEL | BC_(7, CR_LT),
646    [TCG_COND_GEU] = ISEL | BC_(7, CR_LT) | 1,
647    [TCG_COND_LEU] = ISEL | BC_(7, CR_GT) | 1,
648    [TCG_COND_GTU] = ISEL | BC_(7, CR_GT),
649};
650
651static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
652                        intptr_t value, intptr_t addend)
653{
654    const tcg_insn_unit *target;
655    int16_t lo;
656    int32_t hi;
657
658    value += addend;
659    target = (const tcg_insn_unit *)value;
660
661    switch (type) {
662    case R_PPC_REL14:
663        return reloc_pc14(code_ptr, target);
664    case R_PPC_REL24:
665        return reloc_pc24(code_ptr, target);
666    case R_PPC_ADDR16:
667        /*
668         * We are (slightly) abusing this relocation type.  In particular,
669         * assert that the low 2 bits are zero, and do not modify them.
670         * That way we can use this with LD et al that have opcode bits
671         * in the low 2 bits of the insn.
672         */
673        if ((value & 3) || value != (int16_t)value) {
674            return false;
675        }
676        *code_ptr = (*code_ptr & ~0xfffc) | (value & 0xfffc);
677        break;
678    case R_PPC_ADDR32:
679        /*
680         * We are abusing this relocation type.  Again, this points to
681         * a pair of insns, lis + load.  This is an absolute address
682         * relocation for PPC32 so the lis cannot be removed.
683         */
684        lo = value;
685        hi = value - lo;
686        if (hi + lo != value) {
687            return false;
688        }
689        code_ptr[0] = deposit32(code_ptr[0], 0, 16, hi >> 16);
690        code_ptr[1] = deposit32(code_ptr[1], 0, 16, lo);
691        break;
692    default:
693        g_assert_not_reached();
694    }
695    return true;
696}
697
698static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
699                             TCGReg base, tcg_target_long offset);
700
701static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
702{
703    if (ret == arg) {
704        return true;
705    }
706    switch (type) {
707    case TCG_TYPE_I64:
708        tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
709        /* fallthru */
710    case TCG_TYPE_I32:
711        if (ret < TCG_REG_V0) {
712            if (arg < TCG_REG_V0) {
713                tcg_out32(s, OR | SAB(arg, ret, arg));
714                break;
715            } else if (have_isa_2_07) {
716                tcg_out32(s, (type == TCG_TYPE_I32 ? MFVSRWZ : MFVSRD)
717                          | VRT(arg) | RA(ret));
718                break;
719            } else {
720                /* Altivec does not support vector->integer moves.  */
721                return false;
722            }
723        } else if (arg < TCG_REG_V0) {
724            if (have_isa_2_07) {
725                tcg_out32(s, (type == TCG_TYPE_I32 ? MTVSRWZ : MTVSRD)
726                          | VRT(ret) | RA(arg));
727                break;
728            } else {
729                /* Altivec does not support integer->vector moves.  */
730                return false;
731            }
732        }
733        /* fallthru */
734    case TCG_TYPE_V64:
735    case TCG_TYPE_V128:
736        tcg_debug_assert(ret >= TCG_REG_V0 && arg >= TCG_REG_V0);
737        tcg_out32(s, VOR | VRT(ret) | VRA(arg) | VRB(arg));
738        break;
739    default:
740        g_assert_not_reached();
741    }
742    return true;
743}
744
745static inline void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs,
746                               int sh, int mb)
747{
748    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
749    sh = SH(sh & 0x1f) | (((sh >> 5) & 1) << 1);
750    mb = MB64((mb >> 5) | ((mb << 1) & 0x3f));
751    tcg_out32(s, op | RA(ra) | RS(rs) | sh | mb);
752}
753
754static inline void tcg_out_rlw(TCGContext *s, int op, TCGReg ra, TCGReg rs,
755                               int sh, int mb, int me)
756{
757    tcg_out32(s, op | RA(ra) | RS(rs) | SH(sh) | MB(mb) | ME(me));
758}
759
760static inline void tcg_out_ext8s(TCGContext *s, TCGReg dst, TCGReg src)
761{
762    tcg_out32(s, EXTSB | RA(dst) | RS(src));
763}
764
765static inline void tcg_out_ext16s(TCGContext *s, TCGReg dst, TCGReg src)
766{
767    tcg_out32(s, EXTSH | RA(dst) | RS(src));
768}
769
770static inline void tcg_out_ext16u(TCGContext *s, TCGReg dst, TCGReg src)
771{
772    tcg_out32(s, ANDI | SAI(src, dst, 0xffff));
773}
774
775static inline void tcg_out_ext32s(TCGContext *s, TCGReg dst, TCGReg src)
776{
777    tcg_out32(s, EXTSW | RA(dst) | RS(src));
778}
779
780static inline void tcg_out_ext32u(TCGContext *s, TCGReg dst, TCGReg src)
781{
782    tcg_out_rld(s, RLDICL, dst, src, 0, 32);
783}
784
785static inline void tcg_out_shli32(TCGContext *s, TCGReg dst, TCGReg src, int c)
786{
787    tcg_out_rlw(s, RLWINM, dst, src, c, 0, 31 - c);
788}
789
790static inline void tcg_out_shli64(TCGContext *s, TCGReg dst, TCGReg src, int c)
791{
792    tcg_out_rld(s, RLDICR, dst, src, c, 63 - c);
793}
794
795static inline void tcg_out_sari32(TCGContext *s, TCGReg dst, TCGReg src, int c)
796{
797    /* Limit immediate shift count lest we create an illegal insn.  */
798    tcg_out32(s, SRAWI | RA(dst) | RS(src) | SH(c & 31));
799}
800
801static inline void tcg_out_shri32(TCGContext *s, TCGReg dst, TCGReg src, int c)
802{
803    tcg_out_rlw(s, RLWINM, dst, src, 32 - c, c, 31);
804}
805
806static inline void tcg_out_shri64(TCGContext *s, TCGReg dst, TCGReg src, int c)
807{
808    tcg_out_rld(s, RLDICL, dst, src, 64 - c, c);
809}
810
811static inline void tcg_out_sari64(TCGContext *s, TCGReg dst, TCGReg src, int c)
812{
813    tcg_out32(s, SRADI | RA(dst) | RS(src) | SH(c & 0x1f) | ((c >> 4) & 2));
814}
815
816static void tcg_out_bswap16(TCGContext *s, TCGReg dst, TCGReg src, int flags)
817{
818    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
819
820    if (have_isa_3_10) {
821        tcg_out32(s, BRH | RA(dst) | RS(src));
822        if (flags & TCG_BSWAP_OS) {
823            tcg_out_ext16s(s, dst, dst);
824        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
825            tcg_out_ext16u(s, dst, dst);
826        }
827        return;
828    }
829
830    /*
831     * In the following,
832     *   dep(a, b, m) -> (a & ~m) | (b & m)
833     *
834     * Begin with:                              src = xxxxabcd
835     */
836    /* tmp = rol32(src, 24) & 0x000000ff            = 0000000c */
837    tcg_out_rlw(s, RLWINM, tmp, src, 24, 24, 31);
838    /* tmp = dep(tmp, rol32(src, 8), 0x0000ff00)    = 000000dc */
839    tcg_out_rlw(s, RLWIMI, tmp, src, 8, 16, 23);
840
841    if (flags & TCG_BSWAP_OS) {
842        tcg_out_ext16s(s, dst, tmp);
843    } else {
844        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
845    }
846}
847
848static void tcg_out_bswap32(TCGContext *s, TCGReg dst, TCGReg src, int flags)
849{
850    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
851
852    if (have_isa_3_10) {
853        tcg_out32(s, BRW | RA(dst) | RS(src));
854        if (flags & TCG_BSWAP_OS) {
855            tcg_out_ext32s(s, dst, dst);
856        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
857            tcg_out_ext32u(s, dst, dst);
858        }
859        return;
860    }
861
862    /*
863     * Stolen from gcc's builtin_bswap32.
864     * In the following,
865     *   dep(a, b, m) -> (a & ~m) | (b & m)
866     *
867     * Begin with:                              src = xxxxabcd
868     */
869    /* tmp = rol32(src, 8) & 0xffffffff             = 0000bcda */
870    tcg_out_rlw(s, RLWINM, tmp, src, 8, 0, 31);
871    /* tmp = dep(tmp, rol32(src, 24), 0xff000000)   = 0000dcda */
872    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 0, 7);
873    /* tmp = dep(tmp, rol32(src, 24), 0x0000ff00)   = 0000dcba */
874    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 16, 23);
875
876    if (flags & TCG_BSWAP_OS) {
877        tcg_out_ext32s(s, dst, tmp);
878    } else {
879        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
880    }
881}
882
883static void tcg_out_bswap64(TCGContext *s, TCGReg dst, TCGReg src)
884{
885    TCGReg t0 = dst == src ? TCG_REG_R0 : dst;
886    TCGReg t1 = dst == src ? dst : TCG_REG_R0;
887
888    if (have_isa_3_10) {
889        tcg_out32(s, BRD | RA(dst) | RS(src));
890        return;
891    }
892
893    /*
894     * In the following,
895     *   dep(a, b, m) -> (a & ~m) | (b & m)
896     *
897     * Begin with:                              src = abcdefgh
898     */
899    /* t0 = rol32(src, 8) & 0xffffffff              = 0000fghe */
900    tcg_out_rlw(s, RLWINM, t0, src, 8, 0, 31);
901    /* t0 = dep(t0, rol32(src, 24), 0xff000000)     = 0000hghe */
902    tcg_out_rlw(s, RLWIMI, t0, src, 24, 0, 7);
903    /* t0 = dep(t0, rol32(src, 24), 0x0000ff00)     = 0000hgfe */
904    tcg_out_rlw(s, RLWIMI, t0, src, 24, 16, 23);
905
906    /* t0 = rol64(t0, 32)                           = hgfe0000 */
907    tcg_out_rld(s, RLDICL, t0, t0, 32, 0);
908    /* t1 = rol64(src, 32)                          = efghabcd */
909    tcg_out_rld(s, RLDICL, t1, src, 32, 0);
910
911    /* t0 = dep(t0, rol32(t1, 24), 0xffffffff)      = hgfebcda */
912    tcg_out_rlw(s, RLWIMI, t0, t1, 8, 0, 31);
913    /* t0 = dep(t0, rol32(t1, 24), 0xff000000)      = hgfedcda */
914    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 0, 7);
915    /* t0 = dep(t0, rol32(t1, 24), 0x0000ff00)      = hgfedcba */
916    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 16, 23);
917
918    tcg_out_mov(s, TCG_TYPE_REG, dst, t0);
919}
920
921/* Emit a move into ret of arg, if it can be done in one insn.  */
922static bool tcg_out_movi_one(TCGContext *s, TCGReg ret, tcg_target_long arg)
923{
924    if (arg == (int16_t)arg) {
925        tcg_out32(s, ADDI | TAI(ret, 0, arg));
926        return true;
927    }
928    if (arg == (int32_t)arg && (arg & 0xffff) == 0) {
929        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
930        return true;
931    }
932    return false;
933}
934
935static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
936                             tcg_target_long arg, bool in_prologue)
937{
938    intptr_t tb_diff;
939    tcg_target_long tmp;
940    int shift;
941
942    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
943
944    if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
945        arg = (int32_t)arg;
946    }
947
948    /* Load 16-bit immediates with one insn.  */
949    if (tcg_out_movi_one(s, ret, arg)) {
950        return;
951    }
952
953    /* Load addresses within the TB with one insn.  */
954    tb_diff = tcg_tbrel_diff(s, (void *)arg);
955    if (!in_prologue && USE_REG_TB && tb_diff == (int16_t)tb_diff) {
956        tcg_out32(s, ADDI | TAI(ret, TCG_REG_TB, tb_diff));
957        return;
958    }
959
960    /* Load 32-bit immediates with two insns.  Note that we've already
961       eliminated bare ADDIS, so we know both insns are required.  */
962    if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) {
963        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
964        tcg_out32(s, ORI | SAI(ret, ret, arg));
965        return;
966    }
967    if (arg == (uint32_t)arg && !(arg & 0x8000)) {
968        tcg_out32(s, ADDI | TAI(ret, 0, arg));
969        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
970        return;
971    }
972
973    /* Load masked 16-bit value.  */
974    if (arg > 0 && (arg & 0x8000)) {
975        tmp = arg | 0x7fff;
976        if ((tmp & (tmp + 1)) == 0) {
977            int mb = clz64(tmp + 1) + 1;
978            tcg_out32(s, ADDI | TAI(ret, 0, arg));
979            tcg_out_rld(s, RLDICL, ret, ret, 0, mb);
980            return;
981        }
982    }
983
984    /* Load common masks with 2 insns.  */
985    shift = ctz64(arg);
986    tmp = arg >> shift;
987    if (tmp == (int16_t)tmp) {
988        tcg_out32(s, ADDI | TAI(ret, 0, tmp));
989        tcg_out_shli64(s, ret, ret, shift);
990        return;
991    }
992    shift = clz64(arg);
993    if (tcg_out_movi_one(s, ret, arg << shift)) {
994        tcg_out_shri64(s, ret, ret, shift);
995        return;
996    }
997
998    /* Load addresses within 2GB of TB with 2 (or rarely 3) insns.  */
999    if (!in_prologue && USE_REG_TB && tb_diff == (int32_t)tb_diff) {
1000        tcg_out_mem_long(s, ADDI, ADD, ret, TCG_REG_TB, tb_diff);
1001        return;
1002    }
1003
1004    /* Use the constant pool, if possible.  */
1005    if (!in_prologue && USE_REG_TB) {
1006        new_pool_label(s, arg, R_PPC_ADDR16, s->code_ptr,
1007                       tcg_tbrel_diff(s, NULL));
1008        tcg_out32(s, LD | TAI(ret, TCG_REG_TB, 0));
1009        return;
1010    }
1011
1012    tmp = arg >> 31 >> 1;
1013    tcg_out_movi(s, TCG_TYPE_I32, ret, tmp);
1014    if (tmp) {
1015        tcg_out_shli64(s, ret, ret, 32);
1016    }
1017    if (arg & 0xffff0000) {
1018        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
1019    }
1020    if (arg & 0xffff) {
1021        tcg_out32(s, ORI | SAI(ret, ret, arg));
1022    }
1023}
1024
1025static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
1026                             TCGReg ret, int64_t val)
1027{
1028    uint32_t load_insn;
1029    int rel, low;
1030    intptr_t add;
1031
1032    switch (vece) {
1033    case MO_8:
1034        low = (int8_t)val;
1035        if (low >= -16 && low < 16) {
1036            tcg_out32(s, VSPLTISB | VRT(ret) | ((val & 31) << 16));
1037            return;
1038        }
1039        if (have_isa_3_00) {
1040            tcg_out32(s, XXSPLTIB | VRT(ret) | ((val & 0xff) << 11));
1041            return;
1042        }
1043        break;
1044
1045    case MO_16:
1046        low = (int16_t)val;
1047        if (low >= -16 && low < 16) {
1048            tcg_out32(s, VSPLTISH | VRT(ret) | ((val & 31) << 16));
1049            return;
1050        }
1051        break;
1052
1053    case MO_32:
1054        low = (int32_t)val;
1055        if (low >= -16 && low < 16) {
1056            tcg_out32(s, VSPLTISW | VRT(ret) | ((val & 31) << 16));
1057            return;
1058        }
1059        break;
1060    }
1061
1062    /*
1063     * Otherwise we must load the value from the constant pool.
1064     */
1065    if (USE_REG_TB) {
1066        rel = R_PPC_ADDR16;
1067        add = tcg_tbrel_diff(s, NULL);
1068    } else {
1069        rel = R_PPC_ADDR32;
1070        add = 0;
1071    }
1072
1073    if (have_vsx) {
1074        load_insn = type == TCG_TYPE_V64 ? LXSDX : LXVDSX;
1075        load_insn |= VRT(ret) | RB(TCG_REG_TMP1);
1076        if (TCG_TARGET_REG_BITS == 64) {
1077            new_pool_label(s, val, rel, s->code_ptr, add);
1078        } else {
1079            new_pool_l2(s, rel, s->code_ptr, add, val >> 32, val);
1080        }
1081    } else {
1082        load_insn = LVX | VRT(ret) | RB(TCG_REG_TMP1);
1083        if (TCG_TARGET_REG_BITS == 64) {
1084            new_pool_l2(s, rel, s->code_ptr, add, val, val);
1085        } else {
1086            new_pool_l4(s, rel, s->code_ptr, add,
1087                        val >> 32, val, val >> 32, val);
1088        }
1089    }
1090
1091    if (USE_REG_TB) {
1092        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, 0, 0));
1093        load_insn |= RA(TCG_REG_TB);
1094    } else {
1095        tcg_out32(s, ADDIS | TAI(TCG_REG_TMP1, 0, 0));
1096        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, TCG_REG_TMP1, 0));
1097    }
1098    tcg_out32(s, load_insn);
1099}
1100
1101static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,
1102                         tcg_target_long arg)
1103{
1104    switch (type) {
1105    case TCG_TYPE_I32:
1106    case TCG_TYPE_I64:
1107        tcg_debug_assert(ret < TCG_REG_V0);
1108        tcg_out_movi_int(s, type, ret, arg, false);
1109        break;
1110
1111    default:
1112        g_assert_not_reached();
1113    }
1114}
1115
1116static bool mask_operand(uint32_t c, int *mb, int *me)
1117{
1118    uint32_t lsb, test;
1119
1120    /* Accept a bit pattern like:
1121           0....01....1
1122           1....10....0
1123           0..01..10..0
1124       Keep track of the transitions.  */
1125    if (c == 0 || c == -1) {
1126        return false;
1127    }
1128    test = c;
1129    lsb = test & -test;
1130    test += lsb;
1131    if (test & (test - 1)) {
1132        return false;
1133    }
1134
1135    *me = clz32(lsb);
1136    *mb = test ? clz32(test & -test) + 1 : 0;
1137    return true;
1138}
1139
1140static bool mask64_operand(uint64_t c, int *mb, int *me)
1141{
1142    uint64_t lsb;
1143
1144    if (c == 0) {
1145        return false;
1146    }
1147
1148    lsb = c & -c;
1149    /* Accept 1..10..0.  */
1150    if (c == -lsb) {
1151        *mb = 0;
1152        *me = clz64(lsb);
1153        return true;
1154    }
1155    /* Accept 0..01..1.  */
1156    if (lsb == 1 && (c & (c + 1)) == 0) {
1157        *mb = clz64(c + 1) + 1;
1158        *me = 63;
1159        return true;
1160    }
1161    return false;
1162}
1163
1164static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1165{
1166    int mb, me;
1167
1168    if (mask_operand(c, &mb, &me)) {
1169        tcg_out_rlw(s, RLWINM, dst, src, 0, mb, me);
1170    } else if ((c & 0xffff) == c) {
1171        tcg_out32(s, ANDI | SAI(src, dst, c));
1172        return;
1173    } else if ((c & 0xffff0000) == c) {
1174        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1175        return;
1176    } else {
1177        tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R0, c);
1178        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1179    }
1180}
1181
1182static void tcg_out_andi64(TCGContext *s, TCGReg dst, TCGReg src, uint64_t c)
1183{
1184    int mb, me;
1185
1186    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1187    if (mask64_operand(c, &mb, &me)) {
1188        if (mb == 0) {
1189            tcg_out_rld(s, RLDICR, dst, src, 0, me);
1190        } else {
1191            tcg_out_rld(s, RLDICL, dst, src, 0, mb);
1192        }
1193    } else if ((c & 0xffff) == c) {
1194        tcg_out32(s, ANDI | SAI(src, dst, c));
1195        return;
1196    } else if ((c & 0xffff0000) == c) {
1197        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1198        return;
1199    } else {
1200        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, c);
1201        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1202    }
1203}
1204
1205static void tcg_out_zori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c,
1206                           int op_lo, int op_hi)
1207{
1208    if (c >> 16) {
1209        tcg_out32(s, op_hi | SAI(src, dst, c >> 16));
1210        src = dst;
1211    }
1212    if (c & 0xffff) {
1213        tcg_out32(s, op_lo | SAI(src, dst, c));
1214        src = dst;
1215    }
1216}
1217
1218static void tcg_out_ori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1219{
1220    tcg_out_zori32(s, dst, src, c, ORI, ORIS);
1221}
1222
1223static void tcg_out_xori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1224{
1225    tcg_out_zori32(s, dst, src, c, XORI, XORIS);
1226}
1227
1228static void tcg_out_b(TCGContext *s, int mask, const tcg_insn_unit *target)
1229{
1230    ptrdiff_t disp = tcg_pcrel_diff(s, target);
1231    if (in_range_b(disp)) {
1232        tcg_out32(s, B | (disp & 0x3fffffc) | mask);
1233    } else {
1234        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, (uintptr_t)target);
1235        tcg_out32(s, MTSPR | RS(TCG_REG_R0) | CTR);
1236        tcg_out32(s, BCCTR | BO_ALWAYS | mask);
1237    }
1238}
1239
1240static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
1241                             TCGReg base, tcg_target_long offset)
1242{
1243    tcg_target_long orig = offset, l0, l1, extra = 0, align = 0;
1244    bool is_int_store = false;
1245    TCGReg rs = TCG_REG_TMP1;
1246
1247    switch (opi) {
1248    case LD: case LWA:
1249        align = 3;
1250        /* FALLTHRU */
1251    default:
1252        if (rt > TCG_REG_R0 && rt < TCG_REG_V0) {
1253            rs = rt;
1254            break;
1255        }
1256        break;
1257    case LXSD:
1258    case STXSD:
1259        align = 3;
1260        break;
1261    case LXV:
1262    case STXV:
1263        align = 15;
1264        break;
1265    case STD:
1266        align = 3;
1267        /* FALLTHRU */
1268    case STB: case STH: case STW:
1269        is_int_store = true;
1270        break;
1271    }
1272
1273    /* For unaligned, or very large offsets, use the indexed form.  */
1274    if (offset & align || offset != (int32_t)offset || opi == 0) {
1275        if (rs == base) {
1276            rs = TCG_REG_R0;
1277        }
1278        tcg_debug_assert(!is_int_store || rs != rt);
1279        tcg_out_movi(s, TCG_TYPE_PTR, rs, orig);
1280        tcg_out32(s, opx | TAB(rt & 31, base, rs));
1281        return;
1282    }
1283
1284    l0 = (int16_t)offset;
1285    offset = (offset - l0) >> 16;
1286    l1 = (int16_t)offset;
1287
1288    if (l1 < 0 && orig >= 0) {
1289        extra = 0x4000;
1290        l1 = (int16_t)(offset - 0x4000);
1291    }
1292    if (l1) {
1293        tcg_out32(s, ADDIS | TAI(rs, base, l1));
1294        base = rs;
1295    }
1296    if (extra) {
1297        tcg_out32(s, ADDIS | TAI(rs, base, extra));
1298        base = rs;
1299    }
1300    if (opi != ADDI || base != rt || l0 != 0) {
1301        tcg_out32(s, opi | TAI(rt & 31, base, l0));
1302    }
1303}
1304
1305static void tcg_out_vsldoi(TCGContext *s, TCGReg ret,
1306                           TCGReg va, TCGReg vb, int shb)
1307{
1308    tcg_out32(s, VSLDOI | VRT(ret) | VRA(va) | VRB(vb) | (shb << 6));
1309}
1310
1311static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
1312                       TCGReg base, intptr_t offset)
1313{
1314    int shift;
1315
1316    switch (type) {
1317    case TCG_TYPE_I32:
1318        if (ret < TCG_REG_V0) {
1319            tcg_out_mem_long(s, LWZ, LWZX, ret, base, offset);
1320            break;
1321        }
1322        if (have_isa_2_07 && have_vsx) {
1323            tcg_out_mem_long(s, 0, LXSIWZX, ret, base, offset);
1324            break;
1325        }
1326        tcg_debug_assert((offset & 3) == 0);
1327        tcg_out_mem_long(s, 0, LVEWX, ret, base, offset);
1328        shift = (offset - 4) & 0xc;
1329        if (shift) {
1330            tcg_out_vsldoi(s, ret, ret, ret, shift);
1331        }
1332        break;
1333    case TCG_TYPE_I64:
1334        if (ret < TCG_REG_V0) {
1335            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1336            tcg_out_mem_long(s, LD, LDX, ret, base, offset);
1337            break;
1338        }
1339        /* fallthru */
1340    case TCG_TYPE_V64:
1341        tcg_debug_assert(ret >= TCG_REG_V0);
1342        if (have_vsx) {
1343            tcg_out_mem_long(s, have_isa_3_00 ? LXSD : 0, LXSDX,
1344                             ret, base, offset);
1345            break;
1346        }
1347        tcg_debug_assert((offset & 7) == 0);
1348        tcg_out_mem_long(s, 0, LVX, ret, base, offset & -16);
1349        if (offset & 8) {
1350            tcg_out_vsldoi(s, ret, ret, ret, 8);
1351        }
1352        break;
1353    case TCG_TYPE_V128:
1354        tcg_debug_assert(ret >= TCG_REG_V0);
1355        tcg_debug_assert((offset & 15) == 0);
1356        tcg_out_mem_long(s, have_isa_3_00 ? LXV : 0,
1357                         LVX, ret, base, offset);
1358        break;
1359    default:
1360        g_assert_not_reached();
1361    }
1362}
1363
1364static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
1365                              TCGReg base, intptr_t offset)
1366{
1367    int shift;
1368
1369    switch (type) {
1370    case TCG_TYPE_I32:
1371        if (arg < TCG_REG_V0) {
1372            tcg_out_mem_long(s, STW, STWX, arg, base, offset);
1373            break;
1374        }
1375        if (have_isa_2_07 && have_vsx) {
1376            tcg_out_mem_long(s, 0, STXSIWX, arg, base, offset);
1377            break;
1378        }
1379        assert((offset & 3) == 0);
1380        tcg_debug_assert((offset & 3) == 0);
1381        shift = (offset - 4) & 0xc;
1382        if (shift) {
1383            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, shift);
1384            arg = TCG_VEC_TMP1;
1385        }
1386        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1387        break;
1388    case TCG_TYPE_I64:
1389        if (arg < TCG_REG_V0) {
1390            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1391            tcg_out_mem_long(s, STD, STDX, arg, base, offset);
1392            break;
1393        }
1394        /* fallthru */
1395    case TCG_TYPE_V64:
1396        tcg_debug_assert(arg >= TCG_REG_V0);
1397        if (have_vsx) {
1398            tcg_out_mem_long(s, have_isa_3_00 ? STXSD : 0,
1399                             STXSDX, arg, base, offset);
1400            break;
1401        }
1402        tcg_debug_assert((offset & 7) == 0);
1403        if (offset & 8) {
1404            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, 8);
1405            arg = TCG_VEC_TMP1;
1406        }
1407        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1408        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset + 4);
1409        break;
1410    case TCG_TYPE_V128:
1411        tcg_debug_assert(arg >= TCG_REG_V0);
1412        tcg_out_mem_long(s, have_isa_3_00 ? STXV : 0,
1413                         STVX, arg, base, offset);
1414        break;
1415    default:
1416        g_assert_not_reached();
1417    }
1418}
1419
1420static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1421                               TCGReg base, intptr_t ofs)
1422{
1423    return false;
1424}
1425
1426static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2,
1427                        int const_arg2, int cr, TCGType type)
1428{
1429    int imm;
1430    uint32_t op;
1431
1432    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1433
1434    /* Simplify the comparisons below wrt CMPI.  */
1435    if (type == TCG_TYPE_I32) {
1436        arg2 = (int32_t)arg2;
1437    }
1438
1439    switch (cond) {
1440    case TCG_COND_EQ:
1441    case TCG_COND_NE:
1442        if (const_arg2) {
1443            if ((int16_t) arg2 == arg2) {
1444                op = CMPI;
1445                imm = 1;
1446                break;
1447            } else if ((uint16_t) arg2 == arg2) {
1448                op = CMPLI;
1449                imm = 1;
1450                break;
1451            }
1452        }
1453        op = CMPL;
1454        imm = 0;
1455        break;
1456
1457    case TCG_COND_LT:
1458    case TCG_COND_GE:
1459    case TCG_COND_LE:
1460    case TCG_COND_GT:
1461        if (const_arg2) {
1462            if ((int16_t) arg2 == arg2) {
1463                op = CMPI;
1464                imm = 1;
1465                break;
1466            }
1467        }
1468        op = CMP;
1469        imm = 0;
1470        break;
1471
1472    case TCG_COND_LTU:
1473    case TCG_COND_GEU:
1474    case TCG_COND_LEU:
1475    case TCG_COND_GTU:
1476        if (const_arg2) {
1477            if ((uint16_t) arg2 == arg2) {
1478                op = CMPLI;
1479                imm = 1;
1480                break;
1481            }
1482        }
1483        op = CMPL;
1484        imm = 0;
1485        break;
1486
1487    default:
1488        tcg_abort();
1489    }
1490    op |= BF(cr) | ((type == TCG_TYPE_I64) << 21);
1491
1492    if (imm) {
1493        tcg_out32(s, op | RA(arg1) | (arg2 & 0xffff));
1494    } else {
1495        if (const_arg2) {
1496            tcg_out_movi(s, type, TCG_REG_R0, arg2);
1497            arg2 = TCG_REG_R0;
1498        }
1499        tcg_out32(s, op | RA(arg1) | RB(arg2));
1500    }
1501}
1502
1503static void tcg_out_setcond_eq0(TCGContext *s, TCGType type,
1504                                TCGReg dst, TCGReg src)
1505{
1506    if (type == TCG_TYPE_I32) {
1507        tcg_out32(s, CNTLZW | RS(src) | RA(dst));
1508        tcg_out_shri32(s, dst, dst, 5);
1509    } else {
1510        tcg_out32(s, CNTLZD | RS(src) | RA(dst));
1511        tcg_out_shri64(s, dst, dst, 6);
1512    }
1513}
1514
1515static void tcg_out_setcond_ne0(TCGContext *s, TCGReg dst, TCGReg src)
1516{
1517    /* X != 0 implies X + -1 generates a carry.  Extra addition
1518       trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C.  */
1519    if (dst != src) {
1520        tcg_out32(s, ADDIC | TAI(dst, src, -1));
1521        tcg_out32(s, SUBFE | TAB(dst, dst, src));
1522    } else {
1523        tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1));
1524        tcg_out32(s, SUBFE | TAB(dst, TCG_REG_R0, src));
1525    }
1526}
1527
1528static TCGReg tcg_gen_setcond_xor(TCGContext *s, TCGReg arg1, TCGArg arg2,
1529                                  bool const_arg2)
1530{
1531    if (const_arg2) {
1532        if ((uint32_t)arg2 == arg2) {
1533            tcg_out_xori32(s, TCG_REG_R0, arg1, arg2);
1534        } else {
1535            tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, arg2);
1536            tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, TCG_REG_R0));
1537        }
1538    } else {
1539        tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, arg2));
1540    }
1541    return TCG_REG_R0;
1542}
1543
1544static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
1545                            TCGArg arg0, TCGArg arg1, TCGArg arg2,
1546                            int const_arg2)
1547{
1548    int crop, sh;
1549
1550    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1551
1552    /* Ignore high bits of a potential constant arg2.  */
1553    if (type == TCG_TYPE_I32) {
1554        arg2 = (uint32_t)arg2;
1555    }
1556
1557    /* Handle common and trivial cases before handling anything else.  */
1558    if (arg2 == 0) {
1559        switch (cond) {
1560        case TCG_COND_EQ:
1561            tcg_out_setcond_eq0(s, type, arg0, arg1);
1562            return;
1563        case TCG_COND_NE:
1564            if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1565                tcg_out_ext32u(s, TCG_REG_R0, arg1);
1566                arg1 = TCG_REG_R0;
1567            }
1568            tcg_out_setcond_ne0(s, arg0, arg1);
1569            return;
1570        case TCG_COND_GE:
1571            tcg_out32(s, NOR | SAB(arg1, arg0, arg1));
1572            arg1 = arg0;
1573            /* FALLTHRU */
1574        case TCG_COND_LT:
1575            /* Extract the sign bit.  */
1576            if (type == TCG_TYPE_I32) {
1577                tcg_out_shri32(s, arg0, arg1, 31);
1578            } else {
1579                tcg_out_shri64(s, arg0, arg1, 63);
1580            }
1581            return;
1582        default:
1583            break;
1584        }
1585    }
1586
1587    /* If we have ISEL, we can implement everything with 3 or 4 insns.
1588       All other cases below are also at least 3 insns, so speed up the
1589       code generator by not considering them and always using ISEL.  */
1590    if (have_isel) {
1591        int isel, tab;
1592
1593        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1594
1595        isel = tcg_to_isel[cond];
1596
1597        tcg_out_movi(s, type, arg0, 1);
1598        if (isel & 1) {
1599            /* arg0 = (bc ? 0 : 1) */
1600            tab = TAB(arg0, 0, arg0);
1601            isel &= ~1;
1602        } else {
1603            /* arg0 = (bc ? 1 : 0) */
1604            tcg_out_movi(s, type, TCG_REG_R0, 0);
1605            tab = TAB(arg0, arg0, TCG_REG_R0);
1606        }
1607        tcg_out32(s, isel | tab);
1608        return;
1609    }
1610
1611    switch (cond) {
1612    case TCG_COND_EQ:
1613        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1614        tcg_out_setcond_eq0(s, type, arg0, arg1);
1615        return;
1616
1617    case TCG_COND_NE:
1618        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1619        /* Discard the high bits only once, rather than both inputs.  */
1620        if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1621            tcg_out_ext32u(s, TCG_REG_R0, arg1);
1622            arg1 = TCG_REG_R0;
1623        }
1624        tcg_out_setcond_ne0(s, arg0, arg1);
1625        return;
1626
1627    case TCG_COND_GT:
1628    case TCG_COND_GTU:
1629        sh = 30;
1630        crop = 0;
1631        goto crtest;
1632
1633    case TCG_COND_LT:
1634    case TCG_COND_LTU:
1635        sh = 29;
1636        crop = 0;
1637        goto crtest;
1638
1639    case TCG_COND_GE:
1640    case TCG_COND_GEU:
1641        sh = 31;
1642        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_LT) | BB(7, CR_LT);
1643        goto crtest;
1644
1645    case TCG_COND_LE:
1646    case TCG_COND_LEU:
1647        sh = 31;
1648        crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_GT) | BB(7, CR_GT);
1649    crtest:
1650        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1651        if (crop) {
1652            tcg_out32(s, crop);
1653        }
1654        tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1655        tcg_out_rlw(s, RLWINM, arg0, TCG_REG_R0, sh, 31, 31);
1656        break;
1657
1658    default:
1659        tcg_abort();
1660    }
1661}
1662
1663static void tcg_out_bc(TCGContext *s, int bc, TCGLabel *l)
1664{
1665    if (l->has_value) {
1666        bc |= reloc_pc14_val(tcg_splitwx_to_rx(s->code_ptr), l->u.value_ptr);
1667    } else {
1668        tcg_out_reloc(s, s->code_ptr, R_PPC_REL14, l, 0);
1669    }
1670    tcg_out32(s, bc);
1671}
1672
1673static void tcg_out_brcond(TCGContext *s, TCGCond cond,
1674                           TCGArg arg1, TCGArg arg2, int const_arg2,
1675                           TCGLabel *l, TCGType type)
1676{
1677    tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1678    tcg_out_bc(s, tcg_to_bc[cond], l);
1679}
1680
1681static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
1682                            TCGArg dest, TCGArg c1, TCGArg c2, TCGArg v1,
1683                            TCGArg v2, bool const_c2)
1684{
1685    /* If for some reason both inputs are zero, don't produce bad code.  */
1686    if (v1 == 0 && v2 == 0) {
1687        tcg_out_movi(s, type, dest, 0);
1688        return;
1689    }
1690
1691    tcg_out_cmp(s, cond, c1, c2, const_c2, 7, type);
1692
1693    if (have_isel) {
1694        int isel = tcg_to_isel[cond];
1695
1696        /* Swap the V operands if the operation indicates inversion.  */
1697        if (isel & 1) {
1698            int t = v1;
1699            v1 = v2;
1700            v2 = t;
1701            isel &= ~1;
1702        }
1703        /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand.  */
1704        if (v2 == 0) {
1705            tcg_out_movi(s, type, TCG_REG_R0, 0);
1706        }
1707        tcg_out32(s, isel | TAB(dest, v1, v2));
1708    } else {
1709        if (dest == v2) {
1710            cond = tcg_invert_cond(cond);
1711            v2 = v1;
1712        } else if (dest != v1) {
1713            if (v1 == 0) {
1714                tcg_out_movi(s, type, dest, 0);
1715            } else {
1716                tcg_out_mov(s, type, dest, v1);
1717            }
1718        }
1719        /* Branch forward over one insn */
1720        tcg_out32(s, tcg_to_bc[cond] | 8);
1721        if (v2 == 0) {
1722            tcg_out_movi(s, type, dest, 0);
1723        } else {
1724            tcg_out_mov(s, type, dest, v2);
1725        }
1726    }
1727}
1728
1729static void tcg_out_cntxz(TCGContext *s, TCGType type, uint32_t opc,
1730                          TCGArg a0, TCGArg a1, TCGArg a2, bool const_a2)
1731{
1732    if (const_a2 && a2 == (type == TCG_TYPE_I32 ? 32 : 64)) {
1733        tcg_out32(s, opc | RA(a0) | RS(a1));
1734    } else {
1735        tcg_out_cmp(s, TCG_COND_EQ, a1, 0, 1, 7, type);
1736        /* Note that the only other valid constant for a2 is 0.  */
1737        if (have_isel) {
1738            tcg_out32(s, opc | RA(TCG_REG_R0) | RS(a1));
1739            tcg_out32(s, tcg_to_isel[TCG_COND_EQ] | TAB(a0, a2, TCG_REG_R0));
1740        } else if (!const_a2 && a0 == a2) {
1741            tcg_out32(s, tcg_to_bc[TCG_COND_EQ] | 8);
1742            tcg_out32(s, opc | RA(a0) | RS(a1));
1743        } else {
1744            tcg_out32(s, opc | RA(a0) | RS(a1));
1745            tcg_out32(s, tcg_to_bc[TCG_COND_NE] | 8);
1746            if (const_a2) {
1747                tcg_out_movi(s, type, a0, 0);
1748            } else {
1749                tcg_out_mov(s, type, a0, a2);
1750            }
1751        }
1752    }
1753}
1754
1755static void tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1756                         const int *const_args)
1757{
1758    static const struct { uint8_t bit1, bit2; } bits[] = {
1759        [TCG_COND_LT ] = { CR_LT, CR_LT },
1760        [TCG_COND_LE ] = { CR_LT, CR_GT },
1761        [TCG_COND_GT ] = { CR_GT, CR_GT },
1762        [TCG_COND_GE ] = { CR_GT, CR_LT },
1763        [TCG_COND_LTU] = { CR_LT, CR_LT },
1764        [TCG_COND_LEU] = { CR_LT, CR_GT },
1765        [TCG_COND_GTU] = { CR_GT, CR_GT },
1766        [TCG_COND_GEU] = { CR_GT, CR_LT },
1767    };
1768
1769    TCGCond cond = args[4], cond2;
1770    TCGArg al, ah, bl, bh;
1771    int blconst, bhconst;
1772    int op, bit1, bit2;
1773
1774    al = args[0];
1775    ah = args[1];
1776    bl = args[2];
1777    bh = args[3];
1778    blconst = const_args[2];
1779    bhconst = const_args[3];
1780
1781    switch (cond) {
1782    case TCG_COND_EQ:
1783        op = CRAND;
1784        goto do_equality;
1785    case TCG_COND_NE:
1786        op = CRNAND;
1787    do_equality:
1788        tcg_out_cmp(s, cond, al, bl, blconst, 6, TCG_TYPE_I32);
1789        tcg_out_cmp(s, cond, ah, bh, bhconst, 7, TCG_TYPE_I32);
1790        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1791        break;
1792
1793    case TCG_COND_LT:
1794    case TCG_COND_LE:
1795    case TCG_COND_GT:
1796    case TCG_COND_GE:
1797    case TCG_COND_LTU:
1798    case TCG_COND_LEU:
1799    case TCG_COND_GTU:
1800    case TCG_COND_GEU:
1801        bit1 = bits[cond].bit1;
1802        bit2 = bits[cond].bit2;
1803        op = (bit1 != bit2 ? CRANDC : CRAND);
1804        cond2 = tcg_unsigned_cond(cond);
1805
1806        tcg_out_cmp(s, cond, ah, bh, bhconst, 6, TCG_TYPE_I32);
1807        tcg_out_cmp(s, cond2, al, bl, blconst, 7, TCG_TYPE_I32);
1808        tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, bit2));
1809        tcg_out32(s, CROR | BT(7, CR_EQ) | BA(6, bit1) | BB(7, CR_EQ));
1810        break;
1811
1812    default:
1813        tcg_abort();
1814    }
1815}
1816
1817static void tcg_out_setcond2(TCGContext *s, const TCGArg *args,
1818                             const int *const_args)
1819{
1820    tcg_out_cmp2(s, args + 1, const_args + 1);
1821    tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1822    tcg_out_rlw(s, RLWINM, args[0], TCG_REG_R0, 31, 31, 31);
1823}
1824
1825static void tcg_out_brcond2 (TCGContext *s, const TCGArg *args,
1826                             const int *const_args)
1827{
1828    tcg_out_cmp2(s, args, const_args);
1829    tcg_out_bc(s, BC | BI(7, CR_EQ) | BO_COND_TRUE, arg_label(args[5]));
1830}
1831
1832static void tcg_out_mb(TCGContext *s, TCGArg a0)
1833{
1834    uint32_t insn = HWSYNC;
1835    a0 &= TCG_MO_ALL;
1836    if (a0 == TCG_MO_LD_LD) {
1837        insn = LWSYNC;
1838    } else if (a0 == TCG_MO_ST_ST) {
1839        insn = EIEIO;
1840    }
1841    tcg_out32(s, insn);
1842}
1843
1844void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_rx,
1845                              uintptr_t jmp_rw, uintptr_t addr)
1846{
1847    if (TCG_TARGET_REG_BITS == 64) {
1848        tcg_insn_unit i1, i2;
1849        intptr_t tb_diff = addr - tc_ptr;
1850        intptr_t br_diff = addr - (jmp_rx + 4);
1851        uint64_t pair;
1852
1853        /* This does not exercise the range of the branch, but we do
1854           still need to be able to load the new value of TCG_REG_TB.
1855           But this does still happen quite often.  */
1856        if (tb_diff == (int16_t)tb_diff) {
1857            i1 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, tb_diff);
1858            i2 = B | (br_diff & 0x3fffffc);
1859        } else {
1860            intptr_t lo = (int16_t)tb_diff;
1861            intptr_t hi = (int32_t)(tb_diff - lo);
1862            assert(tb_diff == hi + lo);
1863            i1 = ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, hi >> 16);
1864            i2 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, lo);
1865        }
1866#ifdef HOST_WORDS_BIGENDIAN
1867        pair = (uint64_t)i1 << 32 | i2;
1868#else
1869        pair = (uint64_t)i2 << 32 | i1;
1870#endif
1871
1872        /* As per the enclosing if, this is ppc64.  Avoid the _Static_assert
1873           within qatomic_set that would fail to build a ppc32 host.  */
1874        qatomic_set__nocheck((uint64_t *)jmp_rw, pair);
1875        flush_idcache_range(jmp_rx, jmp_rw, 8);
1876    } else {
1877        intptr_t diff = addr - jmp_rx;
1878        tcg_debug_assert(in_range_b(diff));
1879        qatomic_set((uint32_t *)jmp_rw, B | (diff & 0x3fffffc));
1880        flush_idcache_range(jmp_rx, jmp_rw, 4);
1881    }
1882}
1883
1884static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target)
1885{
1886#ifdef _CALL_AIX
1887    /* Look through the descriptor.  If the branch is in range, and we
1888       don't have to spend too much effort on building the toc.  */
1889    const void *tgt = ((const void * const *)target)[0];
1890    uintptr_t toc = ((const uintptr_t *)target)[1];
1891    intptr_t diff = tcg_pcrel_diff(s, tgt);
1892
1893    if (in_range_b(diff) && toc == (uint32_t)toc) {
1894        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, toc);
1895        tcg_out_b(s, LK, tgt);
1896    } else {
1897        /* Fold the low bits of the constant into the addresses below.  */
1898        intptr_t arg = (intptr_t)target;
1899        int ofs = (int16_t)arg;
1900
1901        if (ofs + 8 < 0x8000) {
1902            arg -= ofs;
1903        } else {
1904            ofs = 0;
1905        }
1906        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, arg);
1907        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_TMP1, ofs);
1908        tcg_out32(s, MTSPR | RA(TCG_REG_R0) | CTR);
1909        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_REG_TMP1, ofs + SZP);
1910        tcg_out32(s, BCCTR | BO_ALWAYS | LK);
1911    }
1912#elif defined(_CALL_ELF) && _CALL_ELF == 2
1913    intptr_t diff;
1914
1915    /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1916       address, which the callee uses to compute its TOC address.  */
1917    /* FIXME: when the branch is in range, we could avoid r12 load if we
1918       knew that the destination uses the same TOC, and what its local
1919       entry point offset is.  */
1920    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R12, (intptr_t)target);
1921
1922    diff = tcg_pcrel_diff(s, target);
1923    if (in_range_b(diff)) {
1924        tcg_out_b(s, LK, target);
1925    } else {
1926        tcg_out32(s, MTSPR | RS(TCG_REG_R12) | CTR);
1927        tcg_out32(s, BCCTR | BO_ALWAYS | LK);
1928    }
1929#else
1930    tcg_out_b(s, LK, target);
1931#endif
1932}
1933
1934static const uint32_t qemu_ldx_opc[(MO_SSIZE + MO_BSWAP) + 1] = {
1935    [MO_UB] = LBZX,
1936    [MO_UW] = LHZX,
1937    [MO_UL] = LWZX,
1938    [MO_Q]  = LDX,
1939    [MO_SW] = LHAX,
1940    [MO_SL] = LWAX,
1941    [MO_BSWAP | MO_UB] = LBZX,
1942    [MO_BSWAP | MO_UW] = LHBRX,
1943    [MO_BSWAP | MO_UL] = LWBRX,
1944    [MO_BSWAP | MO_Q]  = LDBRX,
1945};
1946
1947static const uint32_t qemu_stx_opc[(MO_SIZE + MO_BSWAP) + 1] = {
1948    [MO_UB] = STBX,
1949    [MO_UW] = STHX,
1950    [MO_UL] = STWX,
1951    [MO_Q]  = STDX,
1952    [MO_BSWAP | MO_UB] = STBX,
1953    [MO_BSWAP | MO_UW] = STHBRX,
1954    [MO_BSWAP | MO_UL] = STWBRX,
1955    [MO_BSWAP | MO_Q]  = STDBRX,
1956};
1957
1958static const uint32_t qemu_exts_opc[4] = {
1959    EXTSB, EXTSH, EXTSW, 0
1960};
1961
1962#if defined (CONFIG_SOFTMMU)
1963#include "../tcg-ldst.c.inc"
1964
1965/* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
1966 *                                 int mmu_idx, uintptr_t ra)
1967 */
1968static void * const qemu_ld_helpers[(MO_SIZE | MO_BSWAP) + 1] = {
1969    [MO_UB]   = helper_ret_ldub_mmu,
1970    [MO_LEUW] = helper_le_lduw_mmu,
1971    [MO_LEUL] = helper_le_ldul_mmu,
1972    [MO_LEQ]  = helper_le_ldq_mmu,
1973    [MO_BEUW] = helper_be_lduw_mmu,
1974    [MO_BEUL] = helper_be_ldul_mmu,
1975    [MO_BEQ]  = helper_be_ldq_mmu,
1976};
1977
1978/* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
1979 *                                 uintxx_t val, int mmu_idx, uintptr_t ra)
1980 */
1981static void * const qemu_st_helpers[(MO_SIZE | MO_BSWAP) + 1] = {
1982    [MO_UB]   = helper_ret_stb_mmu,
1983    [MO_LEUW] = helper_le_stw_mmu,
1984    [MO_LEUL] = helper_le_stl_mmu,
1985    [MO_LEQ]  = helper_le_stq_mmu,
1986    [MO_BEUW] = helper_be_stw_mmu,
1987    [MO_BEUL] = helper_be_stl_mmu,
1988    [MO_BEQ]  = helper_be_stq_mmu,
1989};
1990
1991/* We expect to use a 16-bit negative offset from ENV.  */
1992QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
1993QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -32768);
1994
1995/* Perform the TLB load and compare.  Places the result of the comparison
1996   in CR7, loads the addend of the TLB into R3, and returns the register
1997   containing the guest address (zero-extended into R4).  Clobbers R0 and R2. */
1998
1999static TCGReg tcg_out_tlb_read(TCGContext *s, MemOp opc,
2000                               TCGReg addrlo, TCGReg addrhi,
2001                               int mem_index, bool is_read)
2002{
2003    int cmp_off
2004        = (is_read
2005           ? offsetof(CPUTLBEntry, addr_read)
2006           : offsetof(CPUTLBEntry, addr_write));
2007    int fast_off = TLB_MASK_TABLE_OFS(mem_index);
2008    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
2009    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
2010    unsigned s_bits = opc & MO_SIZE;
2011    unsigned a_bits = get_alignment_bits(opc);
2012
2013    /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx].  */
2014    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_AREG0, mask_off);
2015    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R4, TCG_AREG0, table_off);
2016
2017    /* Extract the page index, shifted into place for tlb index.  */
2018    if (TCG_TARGET_REG_BITS == 32) {
2019        tcg_out_shri32(s, TCG_REG_TMP1, addrlo,
2020                       TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
2021    } else {
2022        tcg_out_shri64(s, TCG_REG_TMP1, addrlo,
2023                       TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
2024    }
2025    tcg_out32(s, AND | SAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_TMP1));
2026
2027    /* Load the TLB comparator.  */
2028    if (cmp_off == 0 && TCG_TARGET_REG_BITS >= TARGET_LONG_BITS) {
2029        uint32_t lxu = (TCG_TARGET_REG_BITS == 32 || TARGET_LONG_BITS == 32
2030                        ? LWZUX : LDUX);
2031        tcg_out32(s, lxu | TAB(TCG_REG_TMP1, TCG_REG_R3, TCG_REG_R4));
2032    } else {
2033        tcg_out32(s, ADD | TAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_R4));
2034        if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2035            tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP1, TCG_REG_R3, cmp_off + 4);
2036            tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R4, TCG_REG_R3, cmp_off);
2037        } else {
2038            tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP1, TCG_REG_R3, cmp_off);
2039        }
2040    }
2041
2042    /* Load the TLB addend for use on the fast path.  Do this asap
2043       to minimize any load use delay.  */
2044    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_REG_R3,
2045               offsetof(CPUTLBEntry, addend));
2046
2047    /* Clear the non-page, non-alignment bits from the address */
2048    if (TCG_TARGET_REG_BITS == 32) {
2049        /* We don't support unaligned accesses on 32-bits.
2050         * Preserve the bottom bits and thus trigger a comparison
2051         * failure on unaligned accesses.
2052         */
2053        if (a_bits < s_bits) {
2054            a_bits = s_bits;
2055        }
2056        tcg_out_rlw(s, RLWINM, TCG_REG_R0, addrlo, 0,
2057                    (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
2058    } else {
2059        TCGReg t = addrlo;
2060
2061        /* If the access is unaligned, we need to make sure we fail if we
2062         * cross a page boundary.  The trick is to add the access size-1
2063         * to the address before masking the low bits.  That will make the
2064         * address overflow to the next page if we cross a page boundary,
2065         * which will then force a mismatch of the TLB compare.
2066         */
2067        if (a_bits < s_bits) {
2068            unsigned a_mask = (1 << a_bits) - 1;
2069            unsigned s_mask = (1 << s_bits) - 1;
2070            tcg_out32(s, ADDI | TAI(TCG_REG_R0, t, s_mask - a_mask));
2071            t = TCG_REG_R0;
2072        }
2073
2074        /* Mask the address for the requested alignment.  */
2075        if (TARGET_LONG_BITS == 32) {
2076            tcg_out_rlw(s, RLWINM, TCG_REG_R0, t, 0,
2077                        (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
2078            /* Zero-extend the address for use in the final address.  */
2079            tcg_out_ext32u(s, TCG_REG_R4, addrlo);
2080            addrlo = TCG_REG_R4;
2081        } else if (a_bits == 0) {
2082            tcg_out_rld(s, RLDICR, TCG_REG_R0, t, 0, 63 - TARGET_PAGE_BITS);
2083        } else {
2084            tcg_out_rld(s, RLDICL, TCG_REG_R0, t,
2085                        64 - TARGET_PAGE_BITS, TARGET_PAGE_BITS - a_bits);
2086            tcg_out_rld(s, RLDICL, TCG_REG_R0, TCG_REG_R0, TARGET_PAGE_BITS, 0);
2087        }
2088    }
2089
2090    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2091        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
2092                    0, 7, TCG_TYPE_I32);
2093        tcg_out_cmp(s, TCG_COND_EQ, addrhi, TCG_REG_R4, 0, 6, TCG_TYPE_I32);
2094        tcg_out32(s, CRAND | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
2095    } else {
2096        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
2097                    0, 7, TCG_TYPE_TL);
2098    }
2099
2100    return addrlo;
2101}
2102
2103/* Record the context of a call to the out of line helper code for the slow
2104   path for a load or store, so that we can later generate the correct
2105   helper code.  */
2106static void add_qemu_ldst_label(TCGContext *s, bool is_ld, MemOpIdx oi,
2107                                TCGReg datalo_reg, TCGReg datahi_reg,
2108                                TCGReg addrlo_reg, TCGReg addrhi_reg,
2109                                tcg_insn_unit *raddr, tcg_insn_unit *lptr)
2110{
2111    TCGLabelQemuLdst *label = new_ldst_label(s);
2112
2113    label->is_ld = is_ld;
2114    label->oi = oi;
2115    label->datalo_reg = datalo_reg;
2116    label->datahi_reg = datahi_reg;
2117    label->addrlo_reg = addrlo_reg;
2118    label->addrhi_reg = addrhi_reg;
2119    label->raddr = tcg_splitwx_to_rx(raddr);
2120    label->label_ptr[0] = lptr;
2121}
2122
2123static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2124{
2125    MemOpIdx oi = lb->oi;
2126    MemOp opc = get_memop(oi);
2127    TCGReg hi, lo, arg = TCG_REG_R3;
2128
2129    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2130        return false;
2131    }
2132
2133    tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
2134
2135    lo = lb->addrlo_reg;
2136    hi = lb->addrhi_reg;
2137    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2138#ifdef TCG_TARGET_CALL_ALIGN_ARGS
2139        arg |= 1;
2140#endif
2141        tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2142        tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2143    } else {
2144        /* If the address needed to be zero-extended, we'll have already
2145           placed it in R4.  The only remaining case is 64-bit guest.  */
2146        tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
2147    }
2148
2149    tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
2150    tcg_out32(s, MFSPR | RT(arg) | LR);
2151
2152    tcg_out_call(s, qemu_ld_helpers[opc & (MO_BSWAP | MO_SIZE)]);
2153
2154    lo = lb->datalo_reg;
2155    hi = lb->datahi_reg;
2156    if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
2157        tcg_out_mov(s, TCG_TYPE_I32, lo, TCG_REG_R4);
2158        tcg_out_mov(s, TCG_TYPE_I32, hi, TCG_REG_R3);
2159    } else if (opc & MO_SIGN) {
2160        uint32_t insn = qemu_exts_opc[opc & MO_SIZE];
2161        tcg_out32(s, insn | RA(lo) | RS(TCG_REG_R3));
2162    } else {
2163        tcg_out_mov(s, TCG_TYPE_REG, lo, TCG_REG_R3);
2164    }
2165
2166    tcg_out_b(s, 0, lb->raddr);
2167    return true;
2168}
2169
2170static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2171{
2172    MemOpIdx oi = lb->oi;
2173    MemOp opc = get_memop(oi);
2174    MemOp s_bits = opc & MO_SIZE;
2175    TCGReg hi, lo, arg = TCG_REG_R3;
2176
2177    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2178        return false;
2179    }
2180
2181    tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
2182
2183    lo = lb->addrlo_reg;
2184    hi = lb->addrhi_reg;
2185    if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
2186#ifdef TCG_TARGET_CALL_ALIGN_ARGS
2187        arg |= 1;
2188#endif
2189        tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2190        tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2191    } else {
2192        /* If the address needed to be zero-extended, we'll have already
2193           placed it in R4.  The only remaining case is 64-bit guest.  */
2194        tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
2195    }
2196
2197    lo = lb->datalo_reg;
2198    hi = lb->datahi_reg;
2199    if (TCG_TARGET_REG_BITS == 32) {
2200        switch (s_bits) {
2201        case MO_64:
2202#ifdef TCG_TARGET_CALL_ALIGN_ARGS
2203            arg |= 1;
2204#endif
2205            tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
2206            /* FALLTHRU */
2207        case MO_32:
2208            tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
2209            break;
2210        default:
2211            tcg_out_rlw(s, RLWINM, arg++, lo, 0, 32 - (8 << s_bits), 31);
2212            break;
2213        }
2214    } else {
2215        if (s_bits == MO_64) {
2216            tcg_out_mov(s, TCG_TYPE_I64, arg++, lo);
2217        } else {
2218            tcg_out_rld(s, RLDICL, arg++, lo, 0, 64 - (8 << s_bits));
2219        }
2220    }
2221
2222    tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
2223    tcg_out32(s, MFSPR | RT(arg) | LR);
2224
2225    tcg_out_call(s, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
2226
2227    tcg_out_b(s, 0, lb->raddr);
2228    return true;
2229}
2230#endif /* SOFTMMU */
2231
2232static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64)
2233{
2234    TCGReg datalo, datahi, addrlo, rbase;
2235    TCGReg addrhi __attribute__((unused));
2236    MemOpIdx oi;
2237    MemOp opc, s_bits;
2238#ifdef CONFIG_SOFTMMU
2239    int mem_index;
2240    tcg_insn_unit *label_ptr;
2241#endif
2242
2243    datalo = *args++;
2244    datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
2245    addrlo = *args++;
2246    addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
2247    oi = *args++;
2248    opc = get_memop(oi);
2249    s_bits = opc & MO_SIZE;
2250
2251#ifdef CONFIG_SOFTMMU
2252    mem_index = get_mmuidx(oi);
2253    addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, true);
2254
2255    /* Load a pointer into the current opcode w/conditional branch-link. */
2256    label_ptr = s->code_ptr;
2257    tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
2258
2259    rbase = TCG_REG_R3;
2260#else  /* !CONFIG_SOFTMMU */
2261    rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
2262    if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
2263        tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
2264        addrlo = TCG_REG_TMP1;
2265    }
2266#endif
2267
2268    if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
2269        if (opc & MO_BSWAP) {
2270            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2271            tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
2272            tcg_out32(s, LWBRX | TAB(datahi, rbase, TCG_REG_R0));
2273        } else if (rbase != 0) {
2274            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2275            tcg_out32(s, LWZX | TAB(datahi, rbase, addrlo));
2276            tcg_out32(s, LWZX | TAB(datalo, rbase, TCG_REG_R0));
2277        } else if (addrlo == datahi) {
2278            tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
2279            tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
2280        } else {
2281            tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
2282            tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
2283        }
2284    } else {
2285        uint32_t insn = qemu_ldx_opc[opc & (MO_BSWAP | MO_SSIZE)];
2286        if (!have_isa_2_06 && insn == LDBRX) {
2287            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2288            tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
2289            tcg_out32(s, LWBRX | TAB(TCG_REG_R0, rbase, TCG_REG_R0));
2290            tcg_out_rld(s, RLDIMI, datalo, TCG_REG_R0, 32, 0);
2291        } else if (insn) {
2292            tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
2293        } else {
2294            insn = qemu_ldx_opc[opc & (MO_SIZE | MO_BSWAP)];
2295            tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
2296            insn = qemu_exts_opc[s_bits];
2297            tcg_out32(s, insn | RA(datalo) | RS(datalo));
2298        }
2299    }
2300
2301#ifdef CONFIG_SOFTMMU
2302    add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
2303                        s->code_ptr, label_ptr);
2304#endif
2305}
2306
2307static void tcg_out_qemu_st(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#endif
2317
2318    datalo = *args++;
2319    datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
2320    addrlo = *args++;
2321    addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
2322    oi = *args++;
2323    opc = get_memop(oi);
2324    s_bits = opc & MO_SIZE;
2325
2326#ifdef CONFIG_SOFTMMU
2327    mem_index = get_mmuidx(oi);
2328    addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, false);
2329
2330    /* Load a pointer into the current opcode w/conditional branch-link. */
2331    label_ptr = s->code_ptr;
2332    tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
2333
2334    rbase = TCG_REG_R3;
2335#else  /* !CONFIG_SOFTMMU */
2336    rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
2337    if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
2338        tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
2339        addrlo = TCG_REG_TMP1;
2340    }
2341#endif
2342
2343    if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
2344        if (opc & MO_BSWAP) {
2345            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2346            tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
2347            tcg_out32(s, STWBRX | SAB(datahi, rbase, TCG_REG_R0));
2348        } else if (rbase != 0) {
2349            tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
2350            tcg_out32(s, STWX | SAB(datahi, rbase, addrlo));
2351            tcg_out32(s, STWX | SAB(datalo, rbase, TCG_REG_R0));
2352        } else {
2353            tcg_out32(s, STW | TAI(datahi, addrlo, 0));
2354            tcg_out32(s, STW | TAI(datalo, addrlo, 4));
2355        }
2356    } else {
2357        uint32_t insn = qemu_stx_opc[opc & (MO_BSWAP | MO_SIZE)];
2358        if (!have_isa_2_06 && insn == STDBRX) {
2359            tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
2360            tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, addrlo, 4));
2361            tcg_out_shri64(s, TCG_REG_R0, datalo, 32);
2362            tcg_out32(s, STWBRX | SAB(TCG_REG_R0, rbase, TCG_REG_TMP1));
2363        } else {
2364            tcg_out32(s, insn | SAB(datalo, rbase, addrlo));
2365        }
2366    }
2367
2368#ifdef CONFIG_SOFTMMU
2369    add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
2370                        s->code_ptr, label_ptr);
2371#endif
2372}
2373
2374static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
2375{
2376    int i;
2377    for (i = 0; i < count; ++i) {
2378        p[i] = NOP;
2379    }
2380}
2381
2382/* Parameters for function call generation, used in tcg.c.  */
2383#define TCG_TARGET_STACK_ALIGN       16
2384#define TCG_TARGET_EXTEND_ARGS       1
2385
2386#ifdef _CALL_AIX
2387# define LINK_AREA_SIZE                (6 * SZR)
2388# define LR_OFFSET                     (1 * SZR)
2389# define TCG_TARGET_CALL_STACK_OFFSET  (LINK_AREA_SIZE + 8 * SZR)
2390#elif defined(_CALL_DARWIN)
2391# define LINK_AREA_SIZE                (6 * SZR)
2392# define LR_OFFSET                     (2 * SZR)
2393#elif TCG_TARGET_REG_BITS == 64
2394# if defined(_CALL_ELF) && _CALL_ELF == 2
2395#  define LINK_AREA_SIZE               (4 * SZR)
2396#  define LR_OFFSET                    (1 * SZR)
2397# endif
2398#else /* TCG_TARGET_REG_BITS == 32 */
2399# if defined(_CALL_SYSV)
2400#  define LINK_AREA_SIZE               (2 * SZR)
2401#  define LR_OFFSET                    (1 * SZR)
2402# endif
2403#endif
2404#ifndef LR_OFFSET
2405# error "Unhandled abi"
2406#endif
2407#ifndef TCG_TARGET_CALL_STACK_OFFSET
2408# define TCG_TARGET_CALL_STACK_OFFSET  LINK_AREA_SIZE
2409#endif
2410
2411#define CPU_TEMP_BUF_SIZE  (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
2412#define REG_SAVE_SIZE      ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
2413
2414#define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET   \
2415                     + TCG_STATIC_CALL_ARGS_SIZE    \
2416                     + CPU_TEMP_BUF_SIZE            \
2417                     + REG_SAVE_SIZE                \
2418                     + TCG_TARGET_STACK_ALIGN - 1)  \
2419                    & -TCG_TARGET_STACK_ALIGN)
2420
2421#define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
2422
2423static void tcg_target_qemu_prologue(TCGContext *s)
2424{
2425    int i;
2426
2427#ifdef _CALL_AIX
2428    const void **desc = (const void **)s->code_ptr;
2429    desc[0] = tcg_splitwx_to_rx(desc + 2);  /* entry point */
2430    desc[1] = 0;                            /* environment pointer */
2431    s->code_ptr = (void *)(desc + 2);       /* skip over descriptor */
2432#endif
2433
2434    tcg_set_frame(s, TCG_REG_CALL_STACK, REG_SAVE_BOT - CPU_TEMP_BUF_SIZE,
2435                  CPU_TEMP_BUF_SIZE);
2436
2437    /* Prologue */
2438    tcg_out32(s, MFSPR | RT(TCG_REG_R0) | LR);
2439    tcg_out32(s, (SZR == 8 ? STDU : STWU)
2440              | SAI(TCG_REG_R1, TCG_REG_R1, -FRAME_SIZE));
2441
2442    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2443        tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2444                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2445    }
2446    tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2447
2448#ifndef CONFIG_SOFTMMU
2449    if (guest_base) {
2450        tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true);
2451        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2452    }
2453#endif
2454
2455    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2456    tcg_out32(s, MTSPR | RS(tcg_target_call_iarg_regs[1]) | CTR);
2457    if (USE_REG_TB) {
2458        tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, tcg_target_call_iarg_regs[1]);
2459    }
2460    tcg_out32(s, BCCTR | BO_ALWAYS);
2461
2462    /* Epilogue */
2463    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
2464
2465    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2466    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2467        tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2468                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2469    }
2470    tcg_out32(s, MTSPR | RS(TCG_REG_R0) | LR);
2471    tcg_out32(s, ADDI | TAI(TCG_REG_R1, TCG_REG_R1, FRAME_SIZE));
2472    tcg_out32(s, BCLR | BO_ALWAYS);
2473}
2474
2475static void tcg_out_op(TCGContext *s, TCGOpcode opc,
2476                       const TCGArg args[TCG_MAX_OP_ARGS],
2477                       const int const_args[TCG_MAX_OP_ARGS])
2478{
2479    TCGArg a0, a1, a2;
2480
2481    switch (opc) {
2482    case INDEX_op_exit_tb:
2483        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, args[0]);
2484        tcg_out_b(s, 0, tcg_code_gen_epilogue);
2485        break;
2486    case INDEX_op_goto_tb:
2487        if (s->tb_jmp_insn_offset) {
2488            /* Direct jump. */
2489            if (TCG_TARGET_REG_BITS == 64) {
2490                /* Ensure the next insns are 8-byte aligned. */
2491                if ((uintptr_t)s->code_ptr & 7) {
2492                    tcg_out32(s, NOP);
2493                }
2494                s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2495                tcg_out32(s, ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2496                tcg_out32(s, ADDI | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2497            } else {
2498                s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2499                tcg_out32(s, B);
2500                s->tb_jmp_reset_offset[args[0]] = tcg_current_code_size(s);
2501                break;
2502            }
2503        } else {
2504            /* Indirect jump. */
2505            tcg_debug_assert(s->tb_jmp_insn_offset == NULL);
2506            tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, 0,
2507                       (intptr_t)(s->tb_jmp_insn_offset + args[0]));
2508        }
2509        tcg_out32(s, MTSPR | RS(TCG_REG_TB) | CTR);
2510        tcg_out32(s, BCCTR | BO_ALWAYS);
2511        set_jmp_reset_offset(s, args[0]);
2512        if (USE_REG_TB) {
2513            /* For the unlinked case, need to reset TCG_REG_TB.  */
2514            tcg_out_mem_long(s, ADDI, ADD, TCG_REG_TB, TCG_REG_TB,
2515                             -tcg_current_code_size(s));
2516        }
2517        break;
2518    case INDEX_op_goto_ptr:
2519        tcg_out32(s, MTSPR | RS(args[0]) | CTR);
2520        if (USE_REG_TB) {
2521            tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, args[0]);
2522        }
2523        tcg_out32(s, ADDI | TAI(TCG_REG_R3, 0, 0));
2524        tcg_out32(s, BCCTR | BO_ALWAYS);
2525        break;
2526    case INDEX_op_br:
2527        {
2528            TCGLabel *l = arg_label(args[0]);
2529            uint32_t insn = B;
2530
2531            if (l->has_value) {
2532                insn |= reloc_pc24_val(tcg_splitwx_to_rx(s->code_ptr),
2533                                       l->u.value_ptr);
2534            } else {
2535                tcg_out_reloc(s, s->code_ptr, R_PPC_REL24, l, 0);
2536            }
2537            tcg_out32(s, insn);
2538        }
2539        break;
2540    case INDEX_op_ld8u_i32:
2541    case INDEX_op_ld8u_i64:
2542        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2543        break;
2544    case INDEX_op_ld8s_i32:
2545    case INDEX_op_ld8s_i64:
2546        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2547        tcg_out_ext8s(s, args[0], args[0]);
2548        break;
2549    case INDEX_op_ld16u_i32:
2550    case INDEX_op_ld16u_i64:
2551        tcg_out_mem_long(s, LHZ, LHZX, args[0], args[1], args[2]);
2552        break;
2553    case INDEX_op_ld16s_i32:
2554    case INDEX_op_ld16s_i64:
2555        tcg_out_mem_long(s, LHA, LHAX, args[0], args[1], args[2]);
2556        break;
2557    case INDEX_op_ld_i32:
2558    case INDEX_op_ld32u_i64:
2559        tcg_out_mem_long(s, LWZ, LWZX, args[0], args[1], args[2]);
2560        break;
2561    case INDEX_op_ld32s_i64:
2562        tcg_out_mem_long(s, LWA, LWAX, args[0], args[1], args[2]);
2563        break;
2564    case INDEX_op_ld_i64:
2565        tcg_out_mem_long(s, LD, LDX, args[0], args[1], args[2]);
2566        break;
2567    case INDEX_op_st8_i32:
2568    case INDEX_op_st8_i64:
2569        tcg_out_mem_long(s, STB, STBX, args[0], args[1], args[2]);
2570        break;
2571    case INDEX_op_st16_i32:
2572    case INDEX_op_st16_i64:
2573        tcg_out_mem_long(s, STH, STHX, args[0], args[1], args[2]);
2574        break;
2575    case INDEX_op_st_i32:
2576    case INDEX_op_st32_i64:
2577        tcg_out_mem_long(s, STW, STWX, args[0], args[1], args[2]);
2578        break;
2579    case INDEX_op_st_i64:
2580        tcg_out_mem_long(s, STD, STDX, args[0], args[1], args[2]);
2581        break;
2582
2583    case INDEX_op_add_i32:
2584        a0 = args[0], a1 = args[1], a2 = args[2];
2585        if (const_args[2]) {
2586        do_addi_32:
2587            tcg_out_mem_long(s, ADDI, ADD, a0, a1, (int32_t)a2);
2588        } else {
2589            tcg_out32(s, ADD | TAB(a0, a1, a2));
2590        }
2591        break;
2592    case INDEX_op_sub_i32:
2593        a0 = args[0], a1 = args[1], a2 = args[2];
2594        if (const_args[1]) {
2595            if (const_args[2]) {
2596                tcg_out_movi(s, TCG_TYPE_I32, a0, a1 - a2);
2597            } else {
2598                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2599            }
2600        } else if (const_args[2]) {
2601            a2 = -a2;
2602            goto do_addi_32;
2603        } else {
2604            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2605        }
2606        break;
2607
2608    case INDEX_op_and_i32:
2609        a0 = args[0], a1 = args[1], a2 = args[2];
2610        if (const_args[2]) {
2611            tcg_out_andi32(s, a0, a1, a2);
2612        } else {
2613            tcg_out32(s, AND | SAB(a1, a0, a2));
2614        }
2615        break;
2616    case INDEX_op_and_i64:
2617        a0 = args[0], a1 = args[1], a2 = args[2];
2618        if (const_args[2]) {
2619            tcg_out_andi64(s, a0, a1, a2);
2620        } else {
2621            tcg_out32(s, AND | SAB(a1, a0, a2));
2622        }
2623        break;
2624    case INDEX_op_or_i64:
2625    case INDEX_op_or_i32:
2626        a0 = args[0], a1 = args[1], a2 = args[2];
2627        if (const_args[2]) {
2628            tcg_out_ori32(s, a0, a1, a2);
2629        } else {
2630            tcg_out32(s, OR | SAB(a1, a0, a2));
2631        }
2632        break;
2633    case INDEX_op_xor_i64:
2634    case INDEX_op_xor_i32:
2635        a0 = args[0], a1 = args[1], a2 = args[2];
2636        if (const_args[2]) {
2637            tcg_out_xori32(s, a0, a1, a2);
2638        } else {
2639            tcg_out32(s, XOR | SAB(a1, a0, a2));
2640        }
2641        break;
2642    case INDEX_op_andc_i32:
2643        a0 = args[0], a1 = args[1], a2 = args[2];
2644        if (const_args[2]) {
2645            tcg_out_andi32(s, a0, a1, ~a2);
2646        } else {
2647            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2648        }
2649        break;
2650    case INDEX_op_andc_i64:
2651        a0 = args[0], a1 = args[1], a2 = args[2];
2652        if (const_args[2]) {
2653            tcg_out_andi64(s, a0, a1, ~a2);
2654        } else {
2655            tcg_out32(s, ANDC | SAB(a1, a0, a2));
2656        }
2657        break;
2658    case INDEX_op_orc_i32:
2659        if (const_args[2]) {
2660            tcg_out_ori32(s, args[0], args[1], ~args[2]);
2661            break;
2662        }
2663        /* FALLTHRU */
2664    case INDEX_op_orc_i64:
2665        tcg_out32(s, ORC | SAB(args[1], args[0], args[2]));
2666        break;
2667    case INDEX_op_eqv_i32:
2668        if (const_args[2]) {
2669            tcg_out_xori32(s, args[0], args[1], ~args[2]);
2670            break;
2671        }
2672        /* FALLTHRU */
2673    case INDEX_op_eqv_i64:
2674        tcg_out32(s, EQV | SAB(args[1], args[0], args[2]));
2675        break;
2676    case INDEX_op_nand_i32:
2677    case INDEX_op_nand_i64:
2678        tcg_out32(s, NAND | SAB(args[1], args[0], args[2]));
2679        break;
2680    case INDEX_op_nor_i32:
2681    case INDEX_op_nor_i64:
2682        tcg_out32(s, NOR | SAB(args[1], args[0], args[2]));
2683        break;
2684
2685    case INDEX_op_clz_i32:
2686        tcg_out_cntxz(s, TCG_TYPE_I32, CNTLZW, args[0], args[1],
2687                      args[2], const_args[2]);
2688        break;
2689    case INDEX_op_ctz_i32:
2690        tcg_out_cntxz(s, TCG_TYPE_I32, CNTTZW, args[0], args[1],
2691                      args[2], const_args[2]);
2692        break;
2693    case INDEX_op_ctpop_i32:
2694        tcg_out32(s, CNTPOPW | SAB(args[1], args[0], 0));
2695        break;
2696
2697    case INDEX_op_clz_i64:
2698        tcg_out_cntxz(s, TCG_TYPE_I64, CNTLZD, args[0], args[1],
2699                      args[2], const_args[2]);
2700        break;
2701    case INDEX_op_ctz_i64:
2702        tcg_out_cntxz(s, TCG_TYPE_I64, CNTTZD, args[0], args[1],
2703                      args[2], const_args[2]);
2704        break;
2705    case INDEX_op_ctpop_i64:
2706        tcg_out32(s, CNTPOPD | SAB(args[1], args[0], 0));
2707        break;
2708
2709    case INDEX_op_mul_i32:
2710        a0 = args[0], a1 = args[1], a2 = args[2];
2711        if (const_args[2]) {
2712            tcg_out32(s, MULLI | TAI(a0, a1, a2));
2713        } else {
2714            tcg_out32(s, MULLW | TAB(a0, a1, a2));
2715        }
2716        break;
2717
2718    case INDEX_op_div_i32:
2719        tcg_out32(s, DIVW | TAB(args[0], args[1], args[2]));
2720        break;
2721
2722    case INDEX_op_divu_i32:
2723        tcg_out32(s, DIVWU | TAB(args[0], args[1], args[2]));
2724        break;
2725
2726    case INDEX_op_shl_i32:
2727        if (const_args[2]) {
2728            /* Limit immediate shift count lest we create an illegal insn.  */
2729            tcg_out_shli32(s, args[0], args[1], args[2] & 31);
2730        } else {
2731            tcg_out32(s, SLW | SAB(args[1], args[0], args[2]));
2732        }
2733        break;
2734    case INDEX_op_shr_i32:
2735        if (const_args[2]) {
2736            /* Limit immediate shift count lest we create an illegal insn.  */
2737            tcg_out_shri32(s, args[0], args[1], args[2] & 31);
2738        } else {
2739            tcg_out32(s, SRW | SAB(args[1], args[0], args[2]));
2740        }
2741        break;
2742    case INDEX_op_sar_i32:
2743        if (const_args[2]) {
2744            tcg_out_sari32(s, args[0], args[1], args[2]);
2745        } else {
2746            tcg_out32(s, SRAW | SAB(args[1], args[0], args[2]));
2747        }
2748        break;
2749    case INDEX_op_rotl_i32:
2750        if (const_args[2]) {
2751            tcg_out_rlw(s, RLWINM, args[0], args[1], args[2], 0, 31);
2752        } else {
2753            tcg_out32(s, RLWNM | SAB(args[1], args[0], args[2])
2754                         | MB(0) | ME(31));
2755        }
2756        break;
2757    case INDEX_op_rotr_i32:
2758        if (const_args[2]) {
2759            tcg_out_rlw(s, RLWINM, args[0], args[1], 32 - args[2], 0, 31);
2760        } else {
2761            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 32));
2762            tcg_out32(s, RLWNM | SAB(args[1], args[0], TCG_REG_R0)
2763                         | MB(0) | ME(31));
2764        }
2765        break;
2766
2767    case INDEX_op_brcond_i32:
2768        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2769                       arg_label(args[3]), TCG_TYPE_I32);
2770        break;
2771    case INDEX_op_brcond_i64:
2772        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2773                       arg_label(args[3]), TCG_TYPE_I64);
2774        break;
2775    case INDEX_op_brcond2_i32:
2776        tcg_out_brcond2(s, args, const_args);
2777        break;
2778
2779    case INDEX_op_neg_i32:
2780    case INDEX_op_neg_i64:
2781        tcg_out32(s, NEG | RT(args[0]) | RA(args[1]));
2782        break;
2783
2784    case INDEX_op_not_i32:
2785    case INDEX_op_not_i64:
2786        tcg_out32(s, NOR | SAB(args[1], args[0], args[1]));
2787        break;
2788
2789    case INDEX_op_add_i64:
2790        a0 = args[0], a1 = args[1], a2 = args[2];
2791        if (const_args[2]) {
2792        do_addi_64:
2793            tcg_out_mem_long(s, ADDI, ADD, a0, a1, a2);
2794        } else {
2795            tcg_out32(s, ADD | TAB(a0, a1, a2));
2796        }
2797        break;
2798    case INDEX_op_sub_i64:
2799        a0 = args[0], a1 = args[1], a2 = args[2];
2800        if (const_args[1]) {
2801            if (const_args[2]) {
2802                tcg_out_movi(s, TCG_TYPE_I64, a0, a1 - a2);
2803            } else {
2804                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2805            }
2806        } else if (const_args[2]) {
2807            a2 = -a2;
2808            goto do_addi_64;
2809        } else {
2810            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2811        }
2812        break;
2813
2814    case INDEX_op_shl_i64:
2815        if (const_args[2]) {
2816            /* Limit immediate shift count lest we create an illegal insn.  */
2817            tcg_out_shli64(s, args[0], args[1], args[2] & 63);
2818        } else {
2819            tcg_out32(s, SLD | SAB(args[1], args[0], args[2]));
2820        }
2821        break;
2822    case INDEX_op_shr_i64:
2823        if (const_args[2]) {
2824            /* Limit immediate shift count lest we create an illegal insn.  */
2825            tcg_out_shri64(s, args[0], args[1], args[2] & 63);
2826        } else {
2827            tcg_out32(s, SRD | SAB(args[1], args[0], args[2]));
2828        }
2829        break;
2830    case INDEX_op_sar_i64:
2831        if (const_args[2]) {
2832            tcg_out_sari64(s, args[0], args[1], args[2]);
2833        } else {
2834            tcg_out32(s, SRAD | SAB(args[1], args[0], args[2]));
2835        }
2836        break;
2837    case INDEX_op_rotl_i64:
2838        if (const_args[2]) {
2839            tcg_out_rld(s, RLDICL, args[0], args[1], args[2], 0);
2840        } else {
2841            tcg_out32(s, RLDCL | SAB(args[1], args[0], args[2]) | MB64(0));
2842        }
2843        break;
2844    case INDEX_op_rotr_i64:
2845        if (const_args[2]) {
2846            tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 0);
2847        } else {
2848            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 64));
2849            tcg_out32(s, RLDCL | SAB(args[1], args[0], TCG_REG_R0) | MB64(0));
2850        }
2851        break;
2852
2853    case INDEX_op_mul_i64:
2854        a0 = args[0], a1 = args[1], a2 = args[2];
2855        if (const_args[2]) {
2856            tcg_out32(s, MULLI | TAI(a0, a1, a2));
2857        } else {
2858            tcg_out32(s, MULLD | TAB(a0, a1, a2));
2859        }
2860        break;
2861    case INDEX_op_div_i64:
2862        tcg_out32(s, DIVD | TAB(args[0], args[1], args[2]));
2863        break;
2864    case INDEX_op_divu_i64:
2865        tcg_out32(s, DIVDU | TAB(args[0], args[1], args[2]));
2866        break;
2867
2868    case INDEX_op_qemu_ld_i32:
2869        tcg_out_qemu_ld(s, args, false);
2870        break;
2871    case INDEX_op_qemu_ld_i64:
2872        tcg_out_qemu_ld(s, args, true);
2873        break;
2874    case INDEX_op_qemu_st_i32:
2875        tcg_out_qemu_st(s, args, false);
2876        break;
2877    case INDEX_op_qemu_st_i64:
2878        tcg_out_qemu_st(s, args, true);
2879        break;
2880
2881    case INDEX_op_ext8s_i32:
2882    case INDEX_op_ext8s_i64:
2883        tcg_out_ext8s(s, args[0], args[1]);
2884        break;
2885    case INDEX_op_ext16s_i32:
2886    case INDEX_op_ext16s_i64:
2887        tcg_out_ext16s(s, args[0], args[1]);
2888        break;
2889    case INDEX_op_ext_i32_i64:
2890    case INDEX_op_ext32s_i64:
2891        tcg_out_ext32s(s, args[0], args[1]);
2892        break;
2893    case INDEX_op_extu_i32_i64:
2894        tcg_out_ext32u(s, args[0], args[1]);
2895        break;
2896
2897    case INDEX_op_setcond_i32:
2898        tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2],
2899                        const_args[2]);
2900        break;
2901    case INDEX_op_setcond_i64:
2902        tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2],
2903                        const_args[2]);
2904        break;
2905    case INDEX_op_setcond2_i32:
2906        tcg_out_setcond2(s, args, const_args);
2907        break;
2908
2909    case INDEX_op_bswap16_i32:
2910    case INDEX_op_bswap16_i64:
2911        tcg_out_bswap16(s, args[0], args[1], args[2]);
2912        break;
2913    case INDEX_op_bswap32_i32:
2914        tcg_out_bswap32(s, args[0], args[1], 0);
2915        break;
2916    case INDEX_op_bswap32_i64:
2917        tcg_out_bswap32(s, args[0], args[1], args[2]);
2918        break;
2919    case INDEX_op_bswap64_i64:
2920        tcg_out_bswap64(s, args[0], args[1]);
2921        break;
2922
2923    case INDEX_op_deposit_i32:
2924        if (const_args[2]) {
2925            uint32_t mask = ((2u << (args[4] - 1)) - 1) << args[3];
2926            tcg_out_andi32(s, args[0], args[0], ~mask);
2927        } else {
2928            tcg_out_rlw(s, RLWIMI, args[0], args[2], args[3],
2929                        32 - args[3] - args[4], 31 - args[3]);
2930        }
2931        break;
2932    case INDEX_op_deposit_i64:
2933        if (const_args[2]) {
2934            uint64_t mask = ((2ull << (args[4] - 1)) - 1) << args[3];
2935            tcg_out_andi64(s, args[0], args[0], ~mask);
2936        } else {
2937            tcg_out_rld(s, RLDIMI, args[0], args[2], args[3],
2938                        64 - args[3] - args[4]);
2939        }
2940        break;
2941
2942    case INDEX_op_extract_i32:
2943        tcg_out_rlw(s, RLWINM, args[0], args[1],
2944                    32 - args[2], 32 - args[3], 31);
2945        break;
2946    case INDEX_op_extract_i64:
2947        tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 64 - args[3]);
2948        break;
2949
2950    case INDEX_op_movcond_i32:
2951        tcg_out_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], args[2],
2952                        args[3], args[4], const_args[2]);
2953        break;
2954    case INDEX_op_movcond_i64:
2955        tcg_out_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], args[2],
2956                        args[3], args[4], const_args[2]);
2957        break;
2958
2959#if TCG_TARGET_REG_BITS == 64
2960    case INDEX_op_add2_i64:
2961#else
2962    case INDEX_op_add2_i32:
2963#endif
2964        /* Note that the CA bit is defined based on the word size of the
2965           environment.  So in 64-bit mode it's always carry-out of bit 63.
2966           The fallback code using deposit works just as well for 32-bit.  */
2967        a0 = args[0], a1 = args[1];
2968        if (a0 == args[3] || (!const_args[5] && a0 == args[5])) {
2969            a0 = TCG_REG_R0;
2970        }
2971        if (const_args[4]) {
2972            tcg_out32(s, ADDIC | TAI(a0, args[2], args[4]));
2973        } else {
2974            tcg_out32(s, ADDC | TAB(a0, args[2], args[4]));
2975        }
2976        if (const_args[5]) {
2977            tcg_out32(s, (args[5] ? ADDME : ADDZE) | RT(a1) | RA(args[3]));
2978        } else {
2979            tcg_out32(s, ADDE | TAB(a1, args[3], args[5]));
2980        }
2981        if (a0 != args[0]) {
2982            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2983        }
2984        break;
2985
2986#if TCG_TARGET_REG_BITS == 64
2987    case INDEX_op_sub2_i64:
2988#else
2989    case INDEX_op_sub2_i32:
2990#endif
2991        a0 = args[0], a1 = args[1];
2992        if (a0 == args[5] || (!const_args[3] && a0 == args[3])) {
2993            a0 = TCG_REG_R0;
2994        }
2995        if (const_args[2]) {
2996            tcg_out32(s, SUBFIC | TAI(a0, args[4], args[2]));
2997        } else {
2998            tcg_out32(s, SUBFC | TAB(a0, args[4], args[2]));
2999        }
3000        if (const_args[3]) {
3001            tcg_out32(s, (args[3] ? SUBFME : SUBFZE) | RT(a1) | RA(args[5]));
3002        } else {
3003            tcg_out32(s, SUBFE | TAB(a1, args[5], args[3]));
3004        }
3005        if (a0 != args[0]) {
3006            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
3007        }
3008        break;
3009
3010    case INDEX_op_muluh_i32:
3011        tcg_out32(s, MULHWU | TAB(args[0], args[1], args[2]));
3012        break;
3013    case INDEX_op_mulsh_i32:
3014        tcg_out32(s, MULHW | TAB(args[0], args[1], args[2]));
3015        break;
3016    case INDEX_op_muluh_i64:
3017        tcg_out32(s, MULHDU | TAB(args[0], args[1], args[2]));
3018        break;
3019    case INDEX_op_mulsh_i64:
3020        tcg_out32(s, MULHD | TAB(args[0], args[1], args[2]));
3021        break;
3022
3023    case INDEX_op_mb:
3024        tcg_out_mb(s, args[0]);
3025        break;
3026
3027    case INDEX_op_mov_i32:   /* Always emitted via tcg_out_mov.  */
3028    case INDEX_op_mov_i64:
3029    case INDEX_op_call:      /* Always emitted via tcg_out_call.  */
3030    default:
3031        tcg_abort();
3032    }
3033}
3034
3035int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3036{
3037    switch (opc) {
3038    case INDEX_op_and_vec:
3039    case INDEX_op_or_vec:
3040    case INDEX_op_xor_vec:
3041    case INDEX_op_andc_vec:
3042    case INDEX_op_not_vec:
3043        return 1;
3044    case INDEX_op_orc_vec:
3045        return have_isa_2_07;
3046    case INDEX_op_add_vec:
3047    case INDEX_op_sub_vec:
3048    case INDEX_op_smax_vec:
3049    case INDEX_op_smin_vec:
3050    case INDEX_op_umax_vec:
3051    case INDEX_op_umin_vec:
3052    case INDEX_op_shlv_vec:
3053    case INDEX_op_shrv_vec:
3054    case INDEX_op_sarv_vec:
3055    case INDEX_op_rotlv_vec:
3056        return vece <= MO_32 || have_isa_2_07;
3057    case INDEX_op_ssadd_vec:
3058    case INDEX_op_sssub_vec:
3059    case INDEX_op_usadd_vec:
3060    case INDEX_op_ussub_vec:
3061        return vece <= MO_32;
3062    case INDEX_op_cmp_vec:
3063    case INDEX_op_shli_vec:
3064    case INDEX_op_shri_vec:
3065    case INDEX_op_sari_vec:
3066    case INDEX_op_rotli_vec:
3067        return vece <= MO_32 || have_isa_2_07 ? -1 : 0;
3068    case INDEX_op_neg_vec:
3069        return vece >= MO_32 && have_isa_3_00;
3070    case INDEX_op_mul_vec:
3071        switch (vece) {
3072        case MO_8:
3073        case MO_16:
3074            return -1;
3075        case MO_32:
3076            return have_isa_2_07 ? 1 : -1;
3077        case MO_64:
3078            return have_isa_3_10;
3079        }
3080        return 0;
3081    case INDEX_op_bitsel_vec:
3082        return have_vsx;
3083    case INDEX_op_rotrv_vec:
3084        return -1;
3085    default:
3086        return 0;
3087    }
3088}
3089
3090static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
3091                            TCGReg dst, TCGReg src)
3092{
3093    tcg_debug_assert(dst >= TCG_REG_V0);
3094
3095    /* Splat from integer reg allowed via constraints for v3.00.  */
3096    if (src < TCG_REG_V0) {
3097        tcg_debug_assert(have_isa_3_00);
3098        switch (vece) {
3099        case MO_64:
3100            tcg_out32(s, MTVSRDD | VRT(dst) | RA(src) | RB(src));
3101            return true;
3102        case MO_32:
3103            tcg_out32(s, MTVSRWS | VRT(dst) | RA(src));
3104            return true;
3105        default:
3106            /* Fail, so that we fall back on either dupm or mov+dup.  */
3107            return false;
3108        }
3109    }
3110
3111    /*
3112     * Recall we use (or emulate) VSX integer loads, so the integer is
3113     * right justified within the left (zero-index) double-word.
3114     */
3115    switch (vece) {
3116    case MO_8:
3117        tcg_out32(s, VSPLTB | VRT(dst) | VRB(src) | (7 << 16));
3118        break;
3119    case MO_16:
3120        tcg_out32(s, VSPLTH | VRT(dst) | VRB(src) | (3 << 16));
3121        break;
3122    case MO_32:
3123        tcg_out32(s, VSPLTW | VRT(dst) | VRB(src) | (1 << 16));
3124        break;
3125    case MO_64:
3126        if (have_vsx) {
3127            tcg_out32(s, XXPERMDI | VRT(dst) | VRA(src) | VRB(src));
3128            break;
3129        }
3130        tcg_out_vsldoi(s, TCG_VEC_TMP1, src, src, 8);
3131        tcg_out_vsldoi(s, dst, TCG_VEC_TMP1, src, 8);
3132        break;
3133    default:
3134        g_assert_not_reached();
3135    }
3136    return true;
3137}
3138
3139static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
3140                             TCGReg out, TCGReg base, intptr_t offset)
3141{
3142    int elt;
3143
3144    tcg_debug_assert(out >= TCG_REG_V0);
3145    switch (vece) {
3146    case MO_8:
3147        if (have_isa_3_00) {
3148            tcg_out_mem_long(s, LXV, LVX, out, base, offset & -16);
3149        } else {
3150            tcg_out_mem_long(s, 0, LVEBX, out, base, offset);
3151        }
3152        elt = extract32(offset, 0, 4);
3153#ifndef HOST_WORDS_BIGENDIAN
3154        elt ^= 15;
3155#endif
3156        tcg_out32(s, VSPLTB | VRT(out) | VRB(out) | (elt << 16));
3157        break;
3158    case MO_16:
3159        tcg_debug_assert((offset & 1) == 0);
3160        if (have_isa_3_00) {
3161            tcg_out_mem_long(s, LXV | 8, LVX, out, base, offset & -16);
3162        } else {
3163            tcg_out_mem_long(s, 0, LVEHX, out, base, offset);
3164        }
3165        elt = extract32(offset, 1, 3);
3166#ifndef HOST_WORDS_BIGENDIAN
3167        elt ^= 7;
3168#endif
3169        tcg_out32(s, VSPLTH | VRT(out) | VRB(out) | (elt << 16));
3170        break;
3171    case MO_32:
3172        if (have_isa_3_00) {
3173            tcg_out_mem_long(s, 0, LXVWSX, out, base, offset);
3174            break;
3175        }
3176        tcg_debug_assert((offset & 3) == 0);
3177        tcg_out_mem_long(s, 0, LVEWX, out, base, offset);
3178        elt = extract32(offset, 2, 2);
3179#ifndef HOST_WORDS_BIGENDIAN
3180        elt ^= 3;
3181#endif
3182        tcg_out32(s, VSPLTW | VRT(out) | VRB(out) | (elt << 16));
3183        break;
3184    case MO_64:
3185        if (have_vsx) {
3186            tcg_out_mem_long(s, 0, LXVDSX, out, base, offset);
3187            break;
3188        }
3189        tcg_debug_assert((offset & 7) == 0);
3190        tcg_out_mem_long(s, 0, LVX, out, base, offset & -16);
3191        tcg_out_vsldoi(s, TCG_VEC_TMP1, out, out, 8);
3192        elt = extract32(offset, 3, 1);
3193#ifndef HOST_WORDS_BIGENDIAN
3194        elt = !elt;
3195#endif
3196        if (elt) {
3197            tcg_out_vsldoi(s, out, out, TCG_VEC_TMP1, 8);
3198        } else {
3199            tcg_out_vsldoi(s, out, TCG_VEC_TMP1, out, 8);
3200        }
3201        break;
3202    default:
3203        g_assert_not_reached();
3204    }
3205    return true;
3206}
3207
3208static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
3209                           unsigned vecl, unsigned vece,
3210                           const TCGArg args[TCG_MAX_OP_ARGS],
3211                           const int const_args[TCG_MAX_OP_ARGS])
3212{
3213    static const uint32_t
3214        add_op[4] = { VADDUBM, VADDUHM, VADDUWM, VADDUDM },
3215        sub_op[4] = { VSUBUBM, VSUBUHM, VSUBUWM, VSUBUDM },
3216        mul_op[4] = { 0, 0, VMULUWM, VMULLD },
3217        neg_op[4] = { 0, 0, VNEGW, VNEGD },
3218        eq_op[4]  = { VCMPEQUB, VCMPEQUH, VCMPEQUW, VCMPEQUD },
3219        ne_op[4]  = { VCMPNEB, VCMPNEH, VCMPNEW, 0 },
3220        gts_op[4] = { VCMPGTSB, VCMPGTSH, VCMPGTSW, VCMPGTSD },
3221        gtu_op[4] = { VCMPGTUB, VCMPGTUH, VCMPGTUW, VCMPGTUD },
3222        ssadd_op[4] = { VADDSBS, VADDSHS, VADDSWS, 0 },
3223        usadd_op[4] = { VADDUBS, VADDUHS, VADDUWS, 0 },
3224        sssub_op[4] = { VSUBSBS, VSUBSHS, VSUBSWS, 0 },
3225        ussub_op[4] = { VSUBUBS, VSUBUHS, VSUBUWS, 0 },
3226        umin_op[4] = { VMINUB, VMINUH, VMINUW, VMINUD },
3227        smin_op[4] = { VMINSB, VMINSH, VMINSW, VMINSD },
3228        umax_op[4] = { VMAXUB, VMAXUH, VMAXUW, VMAXUD },
3229        smax_op[4] = { VMAXSB, VMAXSH, VMAXSW, VMAXSD },
3230        shlv_op[4] = { VSLB, VSLH, VSLW, VSLD },
3231        shrv_op[4] = { VSRB, VSRH, VSRW, VSRD },
3232        sarv_op[4] = { VSRAB, VSRAH, VSRAW, VSRAD },
3233        mrgh_op[4] = { VMRGHB, VMRGHH, VMRGHW, 0 },
3234        mrgl_op[4] = { VMRGLB, VMRGLH, VMRGLW, 0 },
3235        muleu_op[4] = { VMULEUB, VMULEUH, VMULEUW, 0 },
3236        mulou_op[4] = { VMULOUB, VMULOUH, VMULOUW, 0 },
3237        pkum_op[4] = { VPKUHUM, VPKUWUM, 0, 0 },
3238        rotl_op[4] = { VRLB, VRLH, VRLW, VRLD };
3239
3240    TCGType type = vecl + TCG_TYPE_V64;
3241    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
3242    uint32_t insn;
3243
3244    switch (opc) {
3245    case INDEX_op_ld_vec:
3246        tcg_out_ld(s, type, a0, a1, a2);
3247        return;
3248    case INDEX_op_st_vec:
3249        tcg_out_st(s, type, a0, a1, a2);
3250        return;
3251    case INDEX_op_dupm_vec:
3252        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
3253        return;
3254
3255    case INDEX_op_add_vec:
3256        insn = add_op[vece];
3257        break;
3258    case INDEX_op_sub_vec:
3259        insn = sub_op[vece];
3260        break;
3261    case INDEX_op_neg_vec:
3262        insn = neg_op[vece];
3263        a2 = a1;
3264        a1 = 0;
3265        break;
3266    case INDEX_op_mul_vec:
3267        insn = mul_op[vece];
3268        break;
3269    case INDEX_op_ssadd_vec:
3270        insn = ssadd_op[vece];
3271        break;
3272    case INDEX_op_sssub_vec:
3273        insn = sssub_op[vece];
3274        break;
3275    case INDEX_op_usadd_vec:
3276        insn = usadd_op[vece];
3277        break;
3278    case INDEX_op_ussub_vec:
3279        insn = ussub_op[vece];
3280        break;
3281    case INDEX_op_smin_vec:
3282        insn = smin_op[vece];
3283        break;
3284    case INDEX_op_umin_vec:
3285        insn = umin_op[vece];
3286        break;
3287    case INDEX_op_smax_vec:
3288        insn = smax_op[vece];
3289        break;
3290    case INDEX_op_umax_vec:
3291        insn = umax_op[vece];
3292        break;
3293    case INDEX_op_shlv_vec:
3294        insn = shlv_op[vece];
3295        break;
3296    case INDEX_op_shrv_vec:
3297        insn = shrv_op[vece];
3298        break;
3299    case INDEX_op_sarv_vec:
3300        insn = sarv_op[vece];
3301        break;
3302    case INDEX_op_and_vec:
3303        insn = VAND;
3304        break;
3305    case INDEX_op_or_vec:
3306        insn = VOR;
3307        break;
3308    case INDEX_op_xor_vec:
3309        insn = VXOR;
3310        break;
3311    case INDEX_op_andc_vec:
3312        insn = VANDC;
3313        break;
3314    case INDEX_op_not_vec:
3315        insn = VNOR;
3316        a2 = a1;
3317        break;
3318    case INDEX_op_orc_vec:
3319        insn = VORC;
3320        break;
3321
3322    case INDEX_op_cmp_vec:
3323        switch (args[3]) {
3324        case TCG_COND_EQ:
3325            insn = eq_op[vece];
3326            break;
3327        case TCG_COND_NE:
3328            insn = ne_op[vece];
3329            break;
3330        case TCG_COND_GT:
3331            insn = gts_op[vece];
3332            break;
3333        case TCG_COND_GTU:
3334            insn = gtu_op[vece];
3335            break;
3336        default:
3337            g_assert_not_reached();
3338        }
3339        break;
3340
3341    case INDEX_op_bitsel_vec:
3342        tcg_out32(s, XXSEL | VRT(a0) | VRC(a1) | VRB(a2) | VRA(args[3]));
3343        return;
3344
3345    case INDEX_op_dup2_vec:
3346        assert(TCG_TARGET_REG_BITS == 32);
3347        /* With inputs a1 = xLxx, a2 = xHxx  */
3348        tcg_out32(s, VMRGHW | VRT(a0) | VRA(a2) | VRB(a1));  /* a0  = xxHL */
3349        tcg_out_vsldoi(s, TCG_VEC_TMP1, a0, a0, 8);          /* tmp = HLxx */
3350        tcg_out_vsldoi(s, a0, a0, TCG_VEC_TMP1, 8);          /* a0  = HLHL */
3351        return;
3352
3353    case INDEX_op_ppc_mrgh_vec:
3354        insn = mrgh_op[vece];
3355        break;
3356    case INDEX_op_ppc_mrgl_vec:
3357        insn = mrgl_op[vece];
3358        break;
3359    case INDEX_op_ppc_muleu_vec:
3360        insn = muleu_op[vece];
3361        break;
3362    case INDEX_op_ppc_mulou_vec:
3363        insn = mulou_op[vece];
3364        break;
3365    case INDEX_op_ppc_pkum_vec:
3366        insn = pkum_op[vece];
3367        break;
3368    case INDEX_op_rotlv_vec:
3369        insn = rotl_op[vece];
3370        break;
3371    case INDEX_op_ppc_msum_vec:
3372        tcg_debug_assert(vece == MO_16);
3373        tcg_out32(s, VMSUMUHM | VRT(a0) | VRA(a1) | VRB(a2) | VRC(args[3]));
3374        return;
3375
3376    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
3377    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
3378    default:
3379        g_assert_not_reached();
3380    }
3381
3382    tcg_debug_assert(insn != 0);
3383    tcg_out32(s, insn | VRT(a0) | VRA(a1) | VRB(a2));
3384}
3385
3386static void expand_vec_shi(TCGType type, unsigned vece, TCGv_vec v0,
3387                           TCGv_vec v1, TCGArg imm, TCGOpcode opci)
3388{
3389    TCGv_vec t1;
3390
3391    if (vece == MO_32) {
3392        /*
3393         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
3394         * So using negative numbers gets us the 4th bit easily.
3395         */
3396        imm = sextract32(imm, 0, 5);
3397    } else {
3398        imm &= (8 << vece) - 1;
3399    }
3400
3401    /* Splat w/bytes for xxspltib when 2.07 allows MO_64. */
3402    t1 = tcg_constant_vec(type, MO_8, imm);
3403    vec_gen_3(opci, type, vece, tcgv_vec_arg(v0),
3404              tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3405}
3406
3407static void expand_vec_cmp(TCGType type, unsigned vece, TCGv_vec v0,
3408                           TCGv_vec v1, TCGv_vec v2, TCGCond cond)
3409{
3410    bool need_swap = false, need_inv = false;
3411
3412    tcg_debug_assert(vece <= MO_32 || have_isa_2_07);
3413
3414    switch (cond) {
3415    case TCG_COND_EQ:
3416    case TCG_COND_GT:
3417    case TCG_COND_GTU:
3418        break;
3419    case TCG_COND_NE:
3420        if (have_isa_3_00 && vece <= MO_32) {
3421            break;
3422        }
3423        /* fall through */
3424    case TCG_COND_LE:
3425    case TCG_COND_LEU:
3426        need_inv = true;
3427        break;
3428    case TCG_COND_LT:
3429    case TCG_COND_LTU:
3430        need_swap = true;
3431        break;
3432    case TCG_COND_GE:
3433    case TCG_COND_GEU:
3434        need_swap = need_inv = true;
3435        break;
3436    default:
3437        g_assert_not_reached();
3438    }
3439
3440    if (need_inv) {
3441        cond = tcg_invert_cond(cond);
3442    }
3443    if (need_swap) {
3444        TCGv_vec t1;
3445        t1 = v1, v1 = v2, v2 = t1;
3446        cond = tcg_swap_cond(cond);
3447    }
3448
3449    vec_gen_4(INDEX_op_cmp_vec, type, vece, tcgv_vec_arg(v0),
3450              tcgv_vec_arg(v1), tcgv_vec_arg(v2), cond);
3451
3452    if (need_inv) {
3453        tcg_gen_not_vec(vece, v0, v0);
3454    }
3455}
3456
3457static void expand_vec_mul(TCGType type, unsigned vece, TCGv_vec v0,
3458                           TCGv_vec v1, TCGv_vec v2)
3459{
3460    TCGv_vec t1 = tcg_temp_new_vec(type);
3461    TCGv_vec t2 = tcg_temp_new_vec(type);
3462    TCGv_vec c0, c16;
3463
3464    switch (vece) {
3465    case MO_8:
3466    case MO_16:
3467        vec_gen_3(INDEX_op_ppc_muleu_vec, type, vece, tcgv_vec_arg(t1),
3468                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3469        vec_gen_3(INDEX_op_ppc_mulou_vec, type, vece, tcgv_vec_arg(t2),
3470                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3471        vec_gen_3(INDEX_op_ppc_mrgh_vec, type, vece + 1, tcgv_vec_arg(v0),
3472                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3473        vec_gen_3(INDEX_op_ppc_mrgl_vec, type, vece + 1, tcgv_vec_arg(t1),
3474                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
3475        vec_gen_3(INDEX_op_ppc_pkum_vec, type, vece, tcgv_vec_arg(v0),
3476                  tcgv_vec_arg(v0), tcgv_vec_arg(t1));
3477	break;
3478
3479    case MO_32:
3480        tcg_debug_assert(!have_isa_2_07);
3481        /*
3482         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
3483         * So using -16 is a quick way to represent 16.
3484         */
3485        c16 = tcg_constant_vec(type, MO_8, -16);
3486        c0 = tcg_constant_vec(type, MO_8, 0);
3487
3488        vec_gen_3(INDEX_op_rotlv_vec, type, MO_32, tcgv_vec_arg(t1),
3489                  tcgv_vec_arg(v2), tcgv_vec_arg(c16));
3490        vec_gen_3(INDEX_op_ppc_mulou_vec, type, MO_16, tcgv_vec_arg(t2),
3491                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3492        vec_gen_4(INDEX_op_ppc_msum_vec, type, MO_16, tcgv_vec_arg(t1),
3493                  tcgv_vec_arg(v1), tcgv_vec_arg(t1), tcgv_vec_arg(c0));
3494        vec_gen_3(INDEX_op_shlv_vec, type, MO_32, tcgv_vec_arg(t1),
3495                  tcgv_vec_arg(t1), tcgv_vec_arg(c16));
3496        tcg_gen_add_vec(MO_32, v0, t1, t2);
3497        break;
3498
3499    default:
3500        g_assert_not_reached();
3501    }
3502    tcg_temp_free_vec(t1);
3503    tcg_temp_free_vec(t2);
3504}
3505
3506void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3507                       TCGArg a0, ...)
3508{
3509    va_list va;
3510    TCGv_vec v0, v1, v2, t0;
3511    TCGArg a2;
3512
3513    va_start(va, a0);
3514    v0 = temp_tcgv_vec(arg_temp(a0));
3515    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3516    a2 = va_arg(va, TCGArg);
3517
3518    switch (opc) {
3519    case INDEX_op_shli_vec:
3520        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shlv_vec);
3521        break;
3522    case INDEX_op_shri_vec:
3523        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shrv_vec);
3524        break;
3525    case INDEX_op_sari_vec:
3526        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_sarv_vec);
3527        break;
3528    case INDEX_op_rotli_vec:
3529        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_rotlv_vec);
3530        break;
3531    case INDEX_op_cmp_vec:
3532        v2 = temp_tcgv_vec(arg_temp(a2));
3533        expand_vec_cmp(type, vece, v0, v1, v2, va_arg(va, TCGArg));
3534        break;
3535    case INDEX_op_mul_vec:
3536        v2 = temp_tcgv_vec(arg_temp(a2));
3537        expand_vec_mul(type, vece, v0, v1, v2);
3538        break;
3539    case INDEX_op_rotlv_vec:
3540        v2 = temp_tcgv_vec(arg_temp(a2));
3541        t0 = tcg_temp_new_vec(type);
3542        tcg_gen_neg_vec(vece, t0, v2);
3543        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3544        tcg_temp_free_vec(t0);
3545        break;
3546    default:
3547        g_assert_not_reached();
3548    }
3549    va_end(va);
3550}
3551
3552static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
3553{
3554    switch (op) {
3555    case INDEX_op_goto_ptr:
3556        return C_O0_I1(r);
3557
3558    case INDEX_op_ld8u_i32:
3559    case INDEX_op_ld8s_i32:
3560    case INDEX_op_ld16u_i32:
3561    case INDEX_op_ld16s_i32:
3562    case INDEX_op_ld_i32:
3563    case INDEX_op_ctpop_i32:
3564    case INDEX_op_neg_i32:
3565    case INDEX_op_not_i32:
3566    case INDEX_op_ext8s_i32:
3567    case INDEX_op_ext16s_i32:
3568    case INDEX_op_bswap16_i32:
3569    case INDEX_op_bswap32_i32:
3570    case INDEX_op_extract_i32:
3571    case INDEX_op_ld8u_i64:
3572    case INDEX_op_ld8s_i64:
3573    case INDEX_op_ld16u_i64:
3574    case INDEX_op_ld16s_i64:
3575    case INDEX_op_ld32u_i64:
3576    case INDEX_op_ld32s_i64:
3577    case INDEX_op_ld_i64:
3578    case INDEX_op_ctpop_i64:
3579    case INDEX_op_neg_i64:
3580    case INDEX_op_not_i64:
3581    case INDEX_op_ext8s_i64:
3582    case INDEX_op_ext16s_i64:
3583    case INDEX_op_ext32s_i64:
3584    case INDEX_op_ext_i32_i64:
3585    case INDEX_op_extu_i32_i64:
3586    case INDEX_op_bswap16_i64:
3587    case INDEX_op_bswap32_i64:
3588    case INDEX_op_bswap64_i64:
3589    case INDEX_op_extract_i64:
3590        return C_O1_I1(r, r);
3591
3592    case INDEX_op_st8_i32:
3593    case INDEX_op_st16_i32:
3594    case INDEX_op_st_i32:
3595    case INDEX_op_st8_i64:
3596    case INDEX_op_st16_i64:
3597    case INDEX_op_st32_i64:
3598    case INDEX_op_st_i64:
3599        return C_O0_I2(r, r);
3600
3601    case INDEX_op_add_i32:
3602    case INDEX_op_and_i32:
3603    case INDEX_op_or_i32:
3604    case INDEX_op_xor_i32:
3605    case INDEX_op_andc_i32:
3606    case INDEX_op_orc_i32:
3607    case INDEX_op_eqv_i32:
3608    case INDEX_op_shl_i32:
3609    case INDEX_op_shr_i32:
3610    case INDEX_op_sar_i32:
3611    case INDEX_op_rotl_i32:
3612    case INDEX_op_rotr_i32:
3613    case INDEX_op_setcond_i32:
3614    case INDEX_op_and_i64:
3615    case INDEX_op_andc_i64:
3616    case INDEX_op_shl_i64:
3617    case INDEX_op_shr_i64:
3618    case INDEX_op_sar_i64:
3619    case INDEX_op_rotl_i64:
3620    case INDEX_op_rotr_i64:
3621    case INDEX_op_setcond_i64:
3622        return C_O1_I2(r, r, ri);
3623
3624    case INDEX_op_mul_i32:
3625    case INDEX_op_mul_i64:
3626        return C_O1_I2(r, r, rI);
3627
3628    case INDEX_op_div_i32:
3629    case INDEX_op_divu_i32:
3630    case INDEX_op_nand_i32:
3631    case INDEX_op_nor_i32:
3632    case INDEX_op_muluh_i32:
3633    case INDEX_op_mulsh_i32:
3634    case INDEX_op_orc_i64:
3635    case INDEX_op_eqv_i64:
3636    case INDEX_op_nand_i64:
3637    case INDEX_op_nor_i64:
3638    case INDEX_op_div_i64:
3639    case INDEX_op_divu_i64:
3640    case INDEX_op_mulsh_i64:
3641    case INDEX_op_muluh_i64:
3642        return C_O1_I2(r, r, r);
3643
3644    case INDEX_op_sub_i32:
3645        return C_O1_I2(r, rI, ri);
3646    case INDEX_op_add_i64:
3647        return C_O1_I2(r, r, rT);
3648    case INDEX_op_or_i64:
3649    case INDEX_op_xor_i64:
3650        return C_O1_I2(r, r, rU);
3651    case INDEX_op_sub_i64:
3652        return C_O1_I2(r, rI, rT);
3653    case INDEX_op_clz_i32:
3654    case INDEX_op_ctz_i32:
3655    case INDEX_op_clz_i64:
3656    case INDEX_op_ctz_i64:
3657        return C_O1_I2(r, r, rZW);
3658
3659    case INDEX_op_brcond_i32:
3660    case INDEX_op_brcond_i64:
3661        return C_O0_I2(r, ri);
3662
3663    case INDEX_op_movcond_i32:
3664    case INDEX_op_movcond_i64:
3665        return C_O1_I4(r, r, ri, rZ, rZ);
3666    case INDEX_op_deposit_i32:
3667    case INDEX_op_deposit_i64:
3668        return C_O1_I2(r, 0, rZ);
3669    case INDEX_op_brcond2_i32:
3670        return C_O0_I4(r, r, ri, ri);
3671    case INDEX_op_setcond2_i32:
3672        return C_O1_I4(r, r, r, ri, ri);
3673    case INDEX_op_add2_i64:
3674    case INDEX_op_add2_i32:
3675        return C_O2_I4(r, r, r, r, rI, rZM);
3676    case INDEX_op_sub2_i64:
3677    case INDEX_op_sub2_i32:
3678        return C_O2_I4(r, r, rI, rZM, r, r);
3679
3680    case INDEX_op_qemu_ld_i32:
3681        return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
3682                ? C_O1_I1(r, L)
3683                : C_O1_I2(r, L, L));
3684
3685    case INDEX_op_qemu_st_i32:
3686        return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
3687                ? C_O0_I2(S, S)
3688                : C_O0_I3(S, S, S));
3689
3690    case INDEX_op_qemu_ld_i64:
3691        return (TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, L)
3692                : TARGET_LONG_BITS == 32 ? C_O2_I1(L, L, L)
3693                : C_O2_I2(L, L, L, L));
3694
3695    case INDEX_op_qemu_st_i64:
3696        return (TCG_TARGET_REG_BITS == 64 ? C_O0_I2(S, S)
3697                : TARGET_LONG_BITS == 32 ? C_O0_I3(S, S, S)
3698                : C_O0_I4(S, S, S, S));
3699
3700    case INDEX_op_add_vec:
3701    case INDEX_op_sub_vec:
3702    case INDEX_op_mul_vec:
3703    case INDEX_op_and_vec:
3704    case INDEX_op_or_vec:
3705    case INDEX_op_xor_vec:
3706    case INDEX_op_andc_vec:
3707    case INDEX_op_orc_vec:
3708    case INDEX_op_cmp_vec:
3709    case INDEX_op_ssadd_vec:
3710    case INDEX_op_sssub_vec:
3711    case INDEX_op_usadd_vec:
3712    case INDEX_op_ussub_vec:
3713    case INDEX_op_smax_vec:
3714    case INDEX_op_smin_vec:
3715    case INDEX_op_umax_vec:
3716    case INDEX_op_umin_vec:
3717    case INDEX_op_shlv_vec:
3718    case INDEX_op_shrv_vec:
3719    case INDEX_op_sarv_vec:
3720    case INDEX_op_rotlv_vec:
3721    case INDEX_op_rotrv_vec:
3722    case INDEX_op_ppc_mrgh_vec:
3723    case INDEX_op_ppc_mrgl_vec:
3724    case INDEX_op_ppc_muleu_vec:
3725    case INDEX_op_ppc_mulou_vec:
3726    case INDEX_op_ppc_pkum_vec:
3727    case INDEX_op_dup2_vec:
3728        return C_O1_I2(v, v, v);
3729
3730    case INDEX_op_not_vec:
3731    case INDEX_op_neg_vec:
3732        return C_O1_I1(v, v);
3733
3734    case INDEX_op_dup_vec:
3735        return have_isa_3_00 ? C_O1_I1(v, vr) : C_O1_I1(v, v);
3736
3737    case INDEX_op_ld_vec:
3738    case INDEX_op_dupm_vec:
3739        return C_O1_I1(v, r);
3740
3741    case INDEX_op_st_vec:
3742        return C_O0_I2(v, r);
3743
3744    case INDEX_op_bitsel_vec:
3745    case INDEX_op_ppc_msum_vec:
3746        return C_O1_I3(v, v, v, v);
3747
3748    default:
3749        g_assert_not_reached();
3750    }
3751}
3752
3753static void tcg_target_init(TCGContext *s)
3754{
3755    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3756    unsigned long hwcap2 = qemu_getauxval(AT_HWCAP2);
3757
3758    have_isa = tcg_isa_base;
3759    if (hwcap & PPC_FEATURE_ARCH_2_06) {
3760        have_isa = tcg_isa_2_06;
3761    }
3762#ifdef PPC_FEATURE2_ARCH_2_07
3763    if (hwcap2 & PPC_FEATURE2_ARCH_2_07) {
3764        have_isa = tcg_isa_2_07;
3765    }
3766#endif
3767#ifdef PPC_FEATURE2_ARCH_3_00
3768    if (hwcap2 & PPC_FEATURE2_ARCH_3_00) {
3769        have_isa = tcg_isa_3_00;
3770    }
3771#endif
3772#ifdef PPC_FEATURE2_ARCH_3_10
3773    if (hwcap2 & PPC_FEATURE2_ARCH_3_10) {
3774        have_isa = tcg_isa_3_10;
3775    }
3776#endif
3777
3778#ifdef PPC_FEATURE2_HAS_ISEL
3779    /* Prefer explicit instruction from the kernel. */
3780    have_isel = (hwcap2 & PPC_FEATURE2_HAS_ISEL) != 0;
3781#else
3782    /* Fall back to knowing Power7 (2.06) has ISEL. */
3783    have_isel = have_isa_2_06;
3784#endif
3785
3786    if (hwcap & PPC_FEATURE_HAS_ALTIVEC) {
3787        have_altivec = true;
3788        /* We only care about the portion of VSX that overlaps Altivec. */
3789        if (hwcap & PPC_FEATURE_HAS_VSX) {
3790            have_vsx = true;
3791        }
3792    }
3793
3794    tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
3795    tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
3796    if (have_altivec) {
3797        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3798        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3799    }
3800
3801    tcg_target_call_clobber_regs = 0;
3802    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3803    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3804    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3805    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3806    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3807    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3808    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R7);
3809    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R8);
3810    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R9);
3811    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R10);
3812    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R11);
3813    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
3814
3815    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3816    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3817    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3818    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3819    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3820    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3821    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3822    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3823    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V8);
3824    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V9);
3825    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V10);
3826    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V11);
3827    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V12);
3828    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V13);
3829    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V14);
3830    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V15);
3831    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3832    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3833    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3834    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3835
3836    s->reserved_regs = 0;
3837    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); /* tcg temp */
3838    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R1); /* stack pointer */
3839#if defined(_CALL_SYSV)
3840    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R2); /* toc pointer */
3841#endif
3842#if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
3843    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R13); /* thread pointer */
3844#endif
3845    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1); /* mem temp */
3846    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP1);
3847    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP2);
3848    if (USE_REG_TB) {
3849        tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);  /* tb->tc_ptr */
3850    }
3851}
3852
3853#ifdef __ELF__
3854typedef struct {
3855    DebugFrameCIE cie;
3856    DebugFrameFDEHeader fde;
3857    uint8_t fde_def_cfa[4];
3858    uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2 + 3];
3859} DebugFrame;
3860
3861/* We're expecting a 2 byte uleb128 encoded value.  */
3862QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3863
3864#if TCG_TARGET_REG_BITS == 64
3865# define ELF_HOST_MACHINE EM_PPC64
3866#else
3867# define ELF_HOST_MACHINE EM_PPC
3868#endif
3869
3870static DebugFrame debug_frame = {
3871    .cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3872    .cie.id = -1,
3873    .cie.version = 1,
3874    .cie.code_align = 1,
3875    .cie.data_align = (-SZR & 0x7f),         /* sleb128 -SZR */
3876    .cie.return_column = 65,
3877
3878    /* Total FDE size does not include the "len" member.  */
3879    .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset),
3880
3881    .fde_def_cfa = {
3882        12, TCG_REG_R1,                 /* DW_CFA_def_cfa r1, ... */
3883        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3884        (FRAME_SIZE >> 7)
3885    },
3886    .fde_reg_ofs = {
3887        /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
3888        0x11, 65, (LR_OFFSET / -SZR) & 0x7f,
3889    }
3890};
3891
3892void tcg_register_jit(const void *buf, size_t buf_size)
3893{
3894    uint8_t *p = &debug_frame.fde_reg_ofs[3];
3895    int i;
3896
3897    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i, p += 2) {
3898        p[0] = 0x80 + tcg_target_callee_save_regs[i];
3899        p[1] = (FRAME_SIZE - (REG_SAVE_BOT + i * SZR)) / SZR;
3900    }
3901
3902    debug_frame.fde.func_start = (uintptr_t)buf;
3903    debug_frame.fde.func_len = buf_size;
3904
3905    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3906}
3907#endif /* __ELF__ */
3908