xref: /openbmc/qemu/tcg/ppc/tcg-target.c.inc (revision 4748be5e9df56e13045c0f76fe9f60fa7655fed7)
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
27/*
28 * Standardize on the _CALL_FOO symbols used by GCC:
29 * Apple XCode does not define _CALL_DARWIN.
30 * Clang defines _CALL_ELF (64-bit) but not _CALL_SYSV or _CALL_AIX.
31 */
32#if TCG_TARGET_REG_BITS == 64
33# ifdef _CALL_AIX
34    /* ok */
35# elif defined(_CALL_ELF) && _CALL_ELF == 1
36#  define _CALL_AIX
37# elif defined(_CALL_ELF) && _CALL_ELF == 2
38    /* ok */
39# else
40#  error "Unknown ABI"
41# endif
42#else
43# if defined(_CALL_SYSV) || defined(_CALL_DARWIN)
44    /* ok */
45# elif defined(__APPLE__)
46#  define _CALL_DARWIN
47# elif defined(__ELF__)
48#  define _CALL_SYSV
49# else
50#  error "Unknown ABI"
51# endif
52#endif
53
54#if TCG_TARGET_REG_BITS == 64
55# define TCG_TARGET_CALL_ARG_I32   TCG_CALL_ARG_EXTEND
56# define TCG_TARGET_CALL_RET_I128  TCG_CALL_RET_NORMAL
57#else
58# define TCG_TARGET_CALL_ARG_I32   TCG_CALL_ARG_NORMAL
59# define TCG_TARGET_CALL_RET_I128  TCG_CALL_RET_BY_REF
60#endif
61#ifdef _CALL_SYSV
62# define TCG_TARGET_CALL_ARG_I64   TCG_CALL_ARG_EVEN
63# define TCG_TARGET_CALL_ARG_I128  TCG_CALL_ARG_BY_REF
64#else
65# define TCG_TARGET_CALL_ARG_I64   TCG_CALL_ARG_NORMAL
66# define TCG_TARGET_CALL_ARG_I128  TCG_CALL_ARG_NORMAL
67#endif
68
69/* For some memory operations, we need a scratch that isn't R0.  For the AIX
70   calling convention, we can re-use the TOC register since we'll be reloading
71   it at every call.  Otherwise R12 will do nicely as neither a call-saved
72   register nor a parameter register.  */
73#ifdef _CALL_AIX
74# define TCG_REG_TMP1   TCG_REG_R2
75#else
76# define TCG_REG_TMP1   TCG_REG_R12
77#endif
78#define TCG_REG_TMP2    TCG_REG_R11
79
80#define TCG_VEC_TMP1    TCG_REG_V0
81#define TCG_VEC_TMP2    TCG_REG_V1
82
83#define TCG_REG_TB     TCG_REG_R31
84#define USE_REG_TB     (TCG_TARGET_REG_BITS == 64 && !have_isa_3_00)
85
86/* Shorthand for size of a pointer.  Avoid promotion to unsigned.  */
87#define SZP  ((int)sizeof(void *))
88
89/* Shorthand for size of a register.  */
90#define SZR  (TCG_TARGET_REG_BITS / 8)
91
92#define TCG_CT_CONST_S16  0x100
93#define TCG_CT_CONST_U16  0x200
94#define TCG_CT_CONST_S32  0x400
95#define TCG_CT_CONST_U32  0x800
96#define TCG_CT_CONST_ZERO 0x1000
97#define TCG_CT_CONST_MONE 0x2000
98#define TCG_CT_CONST_WSZ  0x4000
99#define TCG_CT_CONST_CMP  0x8000
100
101#define ALL_GENERAL_REGS  0xffffffffu
102#define ALL_VECTOR_REGS   0xffffffff00000000ull
103
104#ifndef R_PPC64_PCREL34
105#define R_PPC64_PCREL34  132
106#endif
107
108#define have_isel  (cpuinfo & CPUINFO_ISEL)
109
110#define TCG_GUEST_BASE_REG  TCG_REG_R30
111
112#ifdef CONFIG_DEBUG_TCG
113static const char tcg_target_reg_names[TCG_TARGET_NB_REGS][4] = {
114    "r0",  "r1",  "r2",  "r3",  "r4",  "r5",  "r6",  "r7",
115    "r8",  "r9",  "r10", "r11", "r12", "r13", "r14", "r15",
116    "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
117    "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31",
118    "v0",  "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7",
119    "v8",  "v9",  "v10", "v11", "v12", "v13", "v14", "v15",
120    "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
121    "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",
122};
123#endif
124
125static const int tcg_target_reg_alloc_order[] = {
126    TCG_REG_R14,  /* call saved registers */
127    TCG_REG_R15,
128    TCG_REG_R16,
129    TCG_REG_R17,
130    TCG_REG_R18,
131    TCG_REG_R19,
132    TCG_REG_R20,
133    TCG_REG_R21,
134    TCG_REG_R22,
135    TCG_REG_R23,
136    TCG_REG_R24,
137    TCG_REG_R25,
138    TCG_REG_R26,
139    TCG_REG_R27,
140    TCG_REG_R28,
141    TCG_REG_R29,
142    TCG_REG_R30,
143    TCG_REG_R31,
144    TCG_REG_R12,  /* call clobbered, non-arguments */
145    TCG_REG_R11,
146    TCG_REG_R2,
147    TCG_REG_R13,
148    TCG_REG_R10,  /* call clobbered, arguments */
149    TCG_REG_R9,
150    TCG_REG_R8,
151    TCG_REG_R7,
152    TCG_REG_R6,
153    TCG_REG_R5,
154    TCG_REG_R4,
155    TCG_REG_R3,
156
157    /* V0 and V1 reserved as temporaries; V20 - V31 are call-saved */
158    TCG_REG_V2,   /* call clobbered, vectors */
159    TCG_REG_V3,
160    TCG_REG_V4,
161    TCG_REG_V5,
162    TCG_REG_V6,
163    TCG_REG_V7,
164    TCG_REG_V8,
165    TCG_REG_V9,
166    TCG_REG_V10,
167    TCG_REG_V11,
168    TCG_REG_V12,
169    TCG_REG_V13,
170    TCG_REG_V14,
171    TCG_REG_V15,
172    TCG_REG_V16,
173    TCG_REG_V17,
174    TCG_REG_V18,
175    TCG_REG_V19,
176};
177
178static const int tcg_target_call_iarg_regs[] = {
179    TCG_REG_R3,
180    TCG_REG_R4,
181    TCG_REG_R5,
182    TCG_REG_R6,
183    TCG_REG_R7,
184    TCG_REG_R8,
185    TCG_REG_R9,
186    TCG_REG_R10
187};
188
189static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
190{
191    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
192    tcg_debug_assert(slot >= 0 && slot <= 1);
193    return TCG_REG_R3 + slot;
194}
195
196static const int tcg_target_callee_save_regs[] = {
197#ifdef _CALL_DARWIN
198    TCG_REG_R11,
199#endif
200    TCG_REG_R14,
201    TCG_REG_R15,
202    TCG_REG_R16,
203    TCG_REG_R17,
204    TCG_REG_R18,
205    TCG_REG_R19,
206    TCG_REG_R20,
207    TCG_REG_R21,
208    TCG_REG_R22,
209    TCG_REG_R23,
210    TCG_REG_R24,
211    TCG_REG_R25,
212    TCG_REG_R26,
213    TCG_REG_R27, /* currently used for the global env */
214    TCG_REG_R28,
215    TCG_REG_R29,
216    TCG_REG_R30,
217    TCG_REG_R31
218};
219
220/* For PPC, we use TB+4 instead of TB as the base. */
221static inline ptrdiff_t ppc_tbrel_diff(TCGContext *s, const void *target)
222{
223    return tcg_tbrel_diff(s, target) - 4;
224}
225
226static inline bool in_range_b(tcg_target_long target)
227{
228    return target == sextract64(target, 0, 26);
229}
230
231static uint32_t reloc_pc24_val(const tcg_insn_unit *pc,
232                               const tcg_insn_unit *target)
233{
234    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
235    tcg_debug_assert(in_range_b(disp));
236    return disp & 0x3fffffc;
237}
238
239static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
240{
241    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
242    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
243
244    if (in_range_b(disp)) {
245        *src_rw = (*src_rw & ~0x3fffffc) | (disp & 0x3fffffc);
246        return true;
247    }
248    return false;
249}
250
251static uint16_t reloc_pc14_val(const tcg_insn_unit *pc,
252                               const tcg_insn_unit *target)
253{
254    ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
255    tcg_debug_assert(disp == (int16_t) disp);
256    return disp & 0xfffc;
257}
258
259static bool reloc_pc14(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
260{
261    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
262    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
263
264    if (disp == (int16_t) disp) {
265        *src_rw = (*src_rw & ~0xfffc) | (disp & 0xfffc);
266        return true;
267    }
268    return false;
269}
270
271static bool reloc_pc34(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
272{
273    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
274    ptrdiff_t disp = tcg_ptr_byte_diff(target, src_rx);
275
276    if (disp == sextract64(disp, 0, 34)) {
277        src_rw[0] = (src_rw[0] & ~0x3ffff) | ((disp >> 16) & 0x3ffff);
278        src_rw[1] = (src_rw[1] & ~0xffff) | (disp & 0xffff);
279        return true;
280    }
281    return false;
282}
283
284static bool mask_operand(uint32_t c, int *mb, int *me);
285static bool mask64_operand(uint64_t c, int *mb, int *me);
286
287/* test if a constant matches the constraint */
288static bool tcg_target_const_match(int64_t sval, int ct,
289                                   TCGType type, TCGCond cond, int vece)
290{
291    uint64_t uval = sval;
292    int mb, me;
293
294    if (ct & TCG_CT_CONST) {
295        return 1;
296    }
297
298    if (type == TCG_TYPE_I32) {
299        uval = (uint32_t)sval;
300        sval = (int32_t)sval;
301    }
302
303    if (ct & TCG_CT_CONST_CMP) {
304        switch (cond) {
305        case TCG_COND_EQ:
306        case TCG_COND_NE:
307            ct |= TCG_CT_CONST_S16 | TCG_CT_CONST_U16;
308            break;
309        case TCG_COND_LT:
310        case TCG_COND_GE:
311        case TCG_COND_LE:
312        case TCG_COND_GT:
313            ct |= TCG_CT_CONST_S16;
314            break;
315        case TCG_COND_LTU:
316        case TCG_COND_GEU:
317        case TCG_COND_LEU:
318        case TCG_COND_GTU:
319            ct |= TCG_CT_CONST_U16;
320            break;
321        case TCG_COND_TSTEQ:
322        case TCG_COND_TSTNE:
323            if ((uval & ~0xffff) == 0 || (uval & ~0xffff0000ull) == 0) {
324                return 1;
325            }
326            if (uval == (uint32_t)uval && mask_operand(uval, &mb, &me)) {
327                return 1;
328            }
329            if (TCG_TARGET_REG_BITS == 64 &&
330                mask64_operand(uval << clz64(uval), &mb, &me)) {
331                return 1;
332            }
333            return 0;
334        default:
335            g_assert_not_reached();
336        }
337    }
338
339    if ((ct & TCG_CT_CONST_S16) && sval == (int16_t)sval) {
340        return 1;
341    }
342    if ((ct & TCG_CT_CONST_U16) && uval == (uint16_t)uval) {
343        return 1;
344    }
345    if ((ct & TCG_CT_CONST_S32) && sval == (int32_t)sval) {
346        return 1;
347    }
348    if ((ct & TCG_CT_CONST_U32) && uval == (uint32_t)uval) {
349        return 1;
350    }
351    if ((ct & TCG_CT_CONST_ZERO) && sval == 0) {
352        return 1;
353    }
354    if ((ct & TCG_CT_CONST_MONE) && sval == -1) {
355        return 1;
356    }
357    if ((ct & TCG_CT_CONST_WSZ) && sval == (type == TCG_TYPE_I32 ? 32 : 64)) {
358        return 1;
359    }
360    return 0;
361}
362
363#define OPCD(opc) ((opc)<<26)
364#define XO19(opc) (OPCD(19)|((opc)<<1))
365#define MD30(opc) (OPCD(30)|((opc)<<2))
366#define MDS30(opc) (OPCD(30)|((opc)<<1))
367#define XO31(opc) (OPCD(31)|((opc)<<1))
368#define XO58(opc) (OPCD(58)|(opc))
369#define XO62(opc) (OPCD(62)|(opc))
370#define VX4(opc)  (OPCD(4)|(opc))
371
372#define B      OPCD( 18)
373#define BC     OPCD( 16)
374
375#define LBZ    OPCD( 34)
376#define LHZ    OPCD( 40)
377#define LHA    OPCD( 42)
378#define LWZ    OPCD( 32)
379#define LWZUX  XO31( 55)
380#define LD     XO58(  0)
381#define LDX    XO31( 21)
382#define LDU    XO58(  1)
383#define LDUX   XO31( 53)
384#define LWA    XO58(  2)
385#define LWAX   XO31(341)
386#define LQ     OPCD( 56)
387
388#define STB    OPCD( 38)
389#define STH    OPCD( 44)
390#define STW    OPCD( 36)
391#define STD    XO62(  0)
392#define STDU   XO62(  1)
393#define STDX   XO31(149)
394#define STQ    XO62(  2)
395
396#define PLWA   OPCD( 41)
397#define PLD    OPCD( 57)
398#define PLXSD  OPCD( 42)
399#define PLXV   OPCD(25 * 2 + 1)  /* force tx=1 */
400
401#define PSTD   OPCD( 61)
402#define PSTXSD OPCD( 46)
403#define PSTXV  OPCD(27 * 2 + 1)  /* force sx=1 */
404
405#define ADDIC  OPCD( 12)
406#define ADDI   OPCD( 14)
407#define ADDIS  OPCD( 15)
408#define ORI    OPCD( 24)
409#define ORIS   OPCD( 25)
410#define XORI   OPCD( 26)
411#define XORIS  OPCD( 27)
412#define ANDI   OPCD( 28)
413#define ANDIS  OPCD( 29)
414#define MULLI  OPCD(  7)
415#define CMPLI  OPCD( 10)
416#define CMPI   OPCD( 11)
417#define SUBFIC OPCD( 8)
418
419#define LWZU   OPCD( 33)
420#define STWU   OPCD( 37)
421
422#define RLWIMI OPCD( 20)
423#define RLWINM OPCD( 21)
424#define RLWNM  OPCD( 23)
425
426#define RLDICL MD30(  0)
427#define RLDICR MD30(  1)
428#define RLDIMI MD30(  3)
429#define RLDCL  MDS30( 8)
430
431#define BCLR   XO19( 16)
432#define BCCTR  XO19(528)
433#define CRAND  XO19(257)
434#define CRANDC XO19(129)
435#define CRNAND XO19(225)
436#define CROR   XO19(449)
437#define CRNOR  XO19( 33)
438#define ADDPCIS XO19( 2)
439
440#define EXTSB  XO31(954)
441#define EXTSH  XO31(922)
442#define EXTSW  XO31(986)
443#define ADD    XO31(266)
444#define ADDE   XO31(138)
445#define ADDME  XO31(234)
446#define ADDZE  XO31(202)
447#define ADDC   XO31( 10)
448#define AND    XO31( 28)
449#define SUBF   XO31( 40)
450#define SUBFC  XO31(  8)
451#define SUBFE  XO31(136)
452#define SUBFME XO31(232)
453#define SUBFZE XO31(200)
454#define OR     XO31(444)
455#define XOR    XO31(316)
456#define MULLW  XO31(235)
457#define MULHW  XO31( 75)
458#define MULHWU XO31( 11)
459#define DIVW   XO31(491)
460#define DIVWU  XO31(459)
461#define MODSW  XO31(779)
462#define MODUW  XO31(267)
463#define CMP    XO31(  0)
464#define CMPL   XO31( 32)
465#define LHBRX  XO31(790)
466#define LWBRX  XO31(534)
467#define LDBRX  XO31(532)
468#define STHBRX XO31(918)
469#define STWBRX XO31(662)
470#define STDBRX XO31(660)
471#define MFSPR  XO31(339)
472#define MTSPR  XO31(467)
473#define SRAWI  XO31(824)
474#define NEG    XO31(104)
475#define MFCR   XO31( 19)
476#define MFOCRF (MFCR | (1u << 20))
477#define NOR    XO31(124)
478#define CNTLZW XO31( 26)
479#define CNTLZD XO31( 58)
480#define CNTTZW XO31(538)
481#define CNTTZD XO31(570)
482#define CNTPOPW XO31(378)
483#define CNTPOPD XO31(506)
484#define ANDC   XO31( 60)
485#define ORC    XO31(412)
486#define EQV    XO31(284)
487#define NAND   XO31(476)
488#define ISEL   XO31( 15)
489
490#define MULLD  XO31(233)
491#define MULHD  XO31( 73)
492#define MULHDU XO31(  9)
493#define DIVD   XO31(489)
494#define DIVDU  XO31(457)
495#define MODSD  XO31(777)
496#define MODUD  XO31(265)
497
498#define LBZX   XO31( 87)
499#define LHZX   XO31(279)
500#define LHAX   XO31(343)
501#define LWZX   XO31( 23)
502#define STBX   XO31(215)
503#define STHX   XO31(407)
504#define STWX   XO31(151)
505
506#define EIEIO  XO31(854)
507#define HWSYNC XO31(598)
508#define LWSYNC (HWSYNC | (1u << 21))
509
510#define SPR(a, b) ((((a)<<5)|(b))<<11)
511#define LR     SPR(8, 0)
512#define CTR    SPR(9, 0)
513
514#define SLW    XO31( 24)
515#define SRW    XO31(536)
516#define SRAW   XO31(792)
517
518#define SLD    XO31( 27)
519#define SRD    XO31(539)
520#define SRAD   XO31(794)
521#define SRADI  XO31(413<<1)
522
523#define BRH    XO31(219)
524#define BRW    XO31(155)
525#define BRD    XO31(187)
526
527#define TW     XO31( 4)
528#define TRAP   (TW | TO(31))
529
530#define SETBC    XO31(384)  /* v3.10 */
531#define SETBCR   XO31(416)  /* v3.10 */
532#define SETNBC   XO31(448)  /* v3.10 */
533#define SETNBCR  XO31(480)  /* v3.10 */
534
535#define NOP    ORI  /* ori 0,0,0 */
536
537#define LVX        XO31(103)
538#define LVEBX      XO31(7)
539#define LVEHX      XO31(39)
540#define LVEWX      XO31(71)
541#define LXSDX      (XO31(588) | 1)  /* v2.06, force tx=1 */
542#define LXVDSX     (XO31(332) | 1)  /* v2.06, force tx=1 */
543#define LXSIWZX    (XO31(12) | 1)   /* v2.07, force tx=1 */
544#define LXV        (OPCD(61) | 8 | 1)  /* v3.00, force tx=1 */
545#define LXSD       (OPCD(57) | 2)   /* v3.00 */
546#define LXVWSX     (XO31(364) | 1)  /* v3.00, force tx=1 */
547
548#define STVX       XO31(231)
549#define STVEWX     XO31(199)
550#define STXSDX     (XO31(716) | 1)  /* v2.06, force sx=1 */
551#define STXSIWX    (XO31(140) | 1)  /* v2.07, force sx=1 */
552#define STXV       (OPCD(61) | 8 | 5) /* v3.00, force sx=1 */
553#define STXSD      (OPCD(61) | 2)   /* v3.00 */
554
555#define VADDSBS    VX4(768)
556#define VADDUBS    VX4(512)
557#define VADDUBM    VX4(0)
558#define VADDSHS    VX4(832)
559#define VADDUHS    VX4(576)
560#define VADDUHM    VX4(64)
561#define VADDSWS    VX4(896)
562#define VADDUWS    VX4(640)
563#define VADDUWM    VX4(128)
564#define VADDUDM    VX4(192)       /* v2.07 */
565
566#define VSUBSBS    VX4(1792)
567#define VSUBUBS    VX4(1536)
568#define VSUBUBM    VX4(1024)
569#define VSUBSHS    VX4(1856)
570#define VSUBUHS    VX4(1600)
571#define VSUBUHM    VX4(1088)
572#define VSUBSWS    VX4(1920)
573#define VSUBUWS    VX4(1664)
574#define VSUBUWM    VX4(1152)
575#define VSUBUDM    VX4(1216)      /* v2.07 */
576
577#define VNEGW      (VX4(1538) | (6 << 16))  /* v3.00 */
578#define VNEGD      (VX4(1538) | (7 << 16))  /* v3.00 */
579
580#define VMAXSB     VX4(258)
581#define VMAXSH     VX4(322)
582#define VMAXSW     VX4(386)
583#define VMAXSD     VX4(450)       /* v2.07 */
584#define VMAXUB     VX4(2)
585#define VMAXUH     VX4(66)
586#define VMAXUW     VX4(130)
587#define VMAXUD     VX4(194)       /* v2.07 */
588#define VMINSB     VX4(770)
589#define VMINSH     VX4(834)
590#define VMINSW     VX4(898)
591#define VMINSD     VX4(962)       /* v2.07 */
592#define VMINUB     VX4(514)
593#define VMINUH     VX4(578)
594#define VMINUW     VX4(642)
595#define VMINUD     VX4(706)       /* v2.07 */
596
597#define VCMPEQUB   VX4(6)
598#define VCMPEQUH   VX4(70)
599#define VCMPEQUW   VX4(134)
600#define VCMPEQUD   VX4(199)       /* v2.07 */
601#define VCMPGTSB   VX4(774)
602#define VCMPGTSH   VX4(838)
603#define VCMPGTSW   VX4(902)
604#define VCMPGTSD   VX4(967)       /* v2.07 */
605#define VCMPGTUB   VX4(518)
606#define VCMPGTUH   VX4(582)
607#define VCMPGTUW   VX4(646)
608#define VCMPGTUD   VX4(711)       /* v2.07 */
609#define VCMPNEB    VX4(7)         /* v3.00 */
610#define VCMPNEH    VX4(71)        /* v3.00 */
611#define VCMPNEW    VX4(135)       /* v3.00 */
612
613#define VSLB       VX4(260)
614#define VSLH       VX4(324)
615#define VSLW       VX4(388)
616#define VSLD       VX4(1476)      /* v2.07 */
617#define VSRB       VX4(516)
618#define VSRH       VX4(580)
619#define VSRW       VX4(644)
620#define VSRD       VX4(1732)      /* v2.07 */
621#define VSRAB      VX4(772)
622#define VSRAH      VX4(836)
623#define VSRAW      VX4(900)
624#define VSRAD      VX4(964)       /* v2.07 */
625#define VRLB       VX4(4)
626#define VRLH       VX4(68)
627#define VRLW       VX4(132)
628#define VRLD       VX4(196)       /* v2.07 */
629
630#define VMULEUB    VX4(520)
631#define VMULEUH    VX4(584)
632#define VMULEUW    VX4(648)       /* v2.07 */
633#define VMULOUB    VX4(8)
634#define VMULOUH    VX4(72)
635#define VMULOUW    VX4(136)       /* v2.07 */
636#define VMULUWM    VX4(137)       /* v2.07 */
637#define VMULLD     VX4(457)       /* v3.10 */
638#define VMSUMUHM   VX4(38)
639
640#define VMRGHB     VX4(12)
641#define VMRGHH     VX4(76)
642#define VMRGHW     VX4(140)
643#define VMRGLB     VX4(268)
644#define VMRGLH     VX4(332)
645#define VMRGLW     VX4(396)
646
647#define VPKUHUM    VX4(14)
648#define VPKUWUM    VX4(78)
649
650#define VAND       VX4(1028)
651#define VANDC      VX4(1092)
652#define VNOR       VX4(1284)
653#define VOR        VX4(1156)
654#define VXOR       VX4(1220)
655#define VEQV       VX4(1668)      /* v2.07 */
656#define VNAND      VX4(1412)      /* v2.07 */
657#define VORC       VX4(1348)      /* v2.07 */
658
659#define VSPLTB     VX4(524)
660#define VSPLTH     VX4(588)
661#define VSPLTW     VX4(652)
662#define VSPLTISB   VX4(780)
663#define VSPLTISH   VX4(844)
664#define VSPLTISW   VX4(908)
665
666#define VSLDOI     VX4(44)
667
668#define XXPERMDI   (OPCD(60) | (10 << 3) | 7)  /* v2.06, force ax=bx=tx=1 */
669#define XXSEL      (OPCD(60) | (3 << 4) | 0xf) /* v2.06, force ax=bx=cx=tx=1 */
670#define XXSPLTIB   (OPCD(60) | (360 << 1) | 1) /* v3.00, force tx=1 */
671
672#define MFVSRD     (XO31(51) | 1)   /* v2.07, force sx=1 */
673#define MFVSRWZ    (XO31(115) | 1)  /* v2.07, force sx=1 */
674#define MTVSRD     (XO31(179) | 1)  /* v2.07, force tx=1 */
675#define MTVSRWZ    (XO31(243) | 1)  /* v2.07, force tx=1 */
676#define MTVSRDD    (XO31(435) | 1)  /* v3.00, force tx=1 */
677#define MTVSRWS    (XO31(403) | 1)  /* v3.00, force tx=1 */
678
679#define RT(r) ((r)<<21)
680#define RS(r) ((r)<<21)
681#define RA(r) ((r)<<16)
682#define RB(r) ((r)<<11)
683#define TO(t) ((t)<<21)
684#define SH(s) ((s)<<11)
685#define MB(b) ((b)<<6)
686#define ME(e) ((e)<<1)
687#define BO(o) ((o)<<21)
688#define MB64(b) ((b)<<5)
689#define FXM(b) (1 << (19 - (b)))
690
691#define VRT(r)  (((r) & 31) << 21)
692#define VRA(r)  (((r) & 31) << 16)
693#define VRB(r)  (((r) & 31) << 11)
694#define VRC(r)  (((r) & 31) <<  6)
695
696#define LK    1
697
698#define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
699#define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
700#define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
701#define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
702
703#define BF(n)    ((n)<<23)
704#define BI(n, c) (((c)+((n)*4))<<16)
705#define BT(n, c) (((c)+((n)*4))<<21)
706#define BA(n, c) (((c)+((n)*4))<<16)
707#define BB(n, c) (((c)+((n)*4))<<11)
708#define BC_(n, c) (((c)+((n)*4))<<6)
709
710#define BO_COND_TRUE  BO(12)
711#define BO_COND_FALSE BO( 4)
712#define BO_ALWAYS     BO(20)
713
714enum {
715    CR_LT,
716    CR_GT,
717    CR_EQ,
718    CR_SO
719};
720
721static const uint32_t tcg_to_bc[16] = {
722    [TCG_COND_EQ]  = BC | BI(0, CR_EQ) | BO_COND_TRUE,
723    [TCG_COND_NE]  = BC | BI(0, CR_EQ) | BO_COND_FALSE,
724    [TCG_COND_TSTEQ]  = BC | BI(0, CR_EQ) | BO_COND_TRUE,
725    [TCG_COND_TSTNE]  = BC | BI(0, CR_EQ) | BO_COND_FALSE,
726    [TCG_COND_LT]  = BC | BI(0, CR_LT) | BO_COND_TRUE,
727    [TCG_COND_GE]  = BC | BI(0, CR_LT) | BO_COND_FALSE,
728    [TCG_COND_LE]  = BC | BI(0, CR_GT) | BO_COND_FALSE,
729    [TCG_COND_GT]  = BC | BI(0, CR_GT) | BO_COND_TRUE,
730    [TCG_COND_LTU] = BC | BI(0, CR_LT) | BO_COND_TRUE,
731    [TCG_COND_GEU] = BC | BI(0, CR_LT) | BO_COND_FALSE,
732    [TCG_COND_LEU] = BC | BI(0, CR_GT) | BO_COND_FALSE,
733    [TCG_COND_GTU] = BC | BI(0, CR_GT) | BO_COND_TRUE,
734};
735
736/* The low bit here is set if the RA and RB fields must be inverted.  */
737static const uint32_t tcg_to_isel[16] = {
738    [TCG_COND_EQ]  = ISEL | BC_(0, CR_EQ),
739    [TCG_COND_NE]  = ISEL | BC_(0, CR_EQ) | 1,
740    [TCG_COND_TSTEQ] = ISEL | BC_(0, CR_EQ),
741    [TCG_COND_TSTNE] = ISEL | BC_(0, CR_EQ) | 1,
742    [TCG_COND_LT]  = ISEL | BC_(0, CR_LT),
743    [TCG_COND_GE]  = ISEL | BC_(0, CR_LT) | 1,
744    [TCG_COND_LE]  = ISEL | BC_(0, CR_GT) | 1,
745    [TCG_COND_GT]  = ISEL | BC_(0, CR_GT),
746    [TCG_COND_LTU] = ISEL | BC_(0, CR_LT),
747    [TCG_COND_GEU] = ISEL | BC_(0, CR_LT) | 1,
748    [TCG_COND_LEU] = ISEL | BC_(0, CR_GT) | 1,
749    [TCG_COND_GTU] = ISEL | BC_(0, CR_GT),
750};
751
752static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
753                        intptr_t value, intptr_t addend)
754{
755    const tcg_insn_unit *target;
756    int16_t lo;
757    int32_t hi;
758
759    value += addend;
760    target = (const tcg_insn_unit *)value;
761
762    switch (type) {
763    case R_PPC_REL14:
764        return reloc_pc14(code_ptr, target);
765    case R_PPC_REL24:
766        return reloc_pc24(code_ptr, target);
767    case R_PPC64_PCREL34:
768        return reloc_pc34(code_ptr, target);
769    case R_PPC_ADDR16:
770        /*
771         * We are (slightly) abusing this relocation type.  In particular,
772         * assert that the low 2 bits are zero, and do not modify them.
773         * That way we can use this with LD et al that have opcode bits
774         * in the low 2 bits of the insn.
775         */
776        if ((value & 3) || value != (int16_t)value) {
777            return false;
778        }
779        *code_ptr = (*code_ptr & ~0xfffc) | (value & 0xfffc);
780        break;
781    case R_PPC_ADDR32:
782        /*
783         * We are abusing this relocation type.  Again, this points to
784         * a pair of insns, lis + load.  This is an absolute address
785         * relocation for PPC32 so the lis cannot be removed.
786         */
787        lo = value;
788        hi = value - lo;
789        if (hi + lo != value) {
790            return false;
791        }
792        code_ptr[0] = deposit32(code_ptr[0], 0, 16, hi >> 16);
793        code_ptr[1] = deposit32(code_ptr[1], 0, 16, lo);
794        break;
795    default:
796        g_assert_not_reached();
797    }
798    return true;
799}
800
801/* Ensure that the prefixed instruction does not cross a 64-byte boundary. */
802static bool tcg_out_need_prefix_align(TCGContext *s)
803{
804    return ((uintptr_t)s->code_ptr & 0x3f) == 0x3c;
805}
806
807static void tcg_out_prefix_align(TCGContext *s)
808{
809    if (tcg_out_need_prefix_align(s)) {
810        tcg_out32(s, NOP);
811    }
812}
813
814static ptrdiff_t tcg_pcrel_diff_for_prefix(TCGContext *s, const void *target)
815{
816    return tcg_pcrel_diff(s, target) - (tcg_out_need_prefix_align(s) ? 4 : 0);
817}
818
819/* Output Type 00 Prefix - 8-Byte Load/Store Form (8LS:D) */
820static void tcg_out_8ls_d(TCGContext *s, tcg_insn_unit opc, unsigned rt,
821                          unsigned ra, tcg_target_long imm, bool r)
822{
823    tcg_insn_unit p, i;
824
825    p = OPCD(1) | (r << 20) | ((imm >> 16) & 0x3ffff);
826    i = opc | TAI(rt, ra, imm);
827
828    tcg_out_prefix_align(s);
829    tcg_out32(s, p);
830    tcg_out32(s, i);
831}
832
833/* Output Type 10 Prefix - Modified Load/Store Form (MLS:D) */
834static void tcg_out_mls_d(TCGContext *s, tcg_insn_unit opc, unsigned rt,
835                          unsigned ra, tcg_target_long imm, bool r)
836{
837    tcg_insn_unit p, i;
838
839    p = OPCD(1) | (2 << 24) | (r << 20) | ((imm >> 16) & 0x3ffff);
840    i = opc | TAI(rt, ra, imm);
841
842    tcg_out_prefix_align(s);
843    tcg_out32(s, p);
844    tcg_out32(s, i);
845}
846
847static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
848                             TCGReg base, tcg_target_long offset);
849
850static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
851{
852    if (ret == arg) {
853        return true;
854    }
855    switch (type) {
856    case TCG_TYPE_I64:
857        tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
858        /* fallthru */
859    case TCG_TYPE_I32:
860        if (ret < TCG_REG_V0) {
861            if (arg < TCG_REG_V0) {
862                tcg_out32(s, OR | SAB(arg, ret, arg));
863                break;
864            } else if (have_isa_2_07) {
865                tcg_out32(s, (type == TCG_TYPE_I32 ? MFVSRWZ : MFVSRD)
866                          | VRT(arg) | RA(ret));
867                break;
868            } else {
869                /* Altivec does not support vector->integer moves.  */
870                return false;
871            }
872        } else if (arg < TCG_REG_V0) {
873            if (have_isa_2_07) {
874                tcg_out32(s, (type == TCG_TYPE_I32 ? MTVSRWZ : MTVSRD)
875                          | VRT(ret) | RA(arg));
876                break;
877            } else {
878                /* Altivec does not support integer->vector moves.  */
879                return false;
880            }
881        }
882        /* fallthru */
883    case TCG_TYPE_V64:
884    case TCG_TYPE_V128:
885        tcg_debug_assert(ret >= TCG_REG_V0 && arg >= TCG_REG_V0);
886        tcg_out32(s, VOR | VRT(ret) | VRA(arg) | VRB(arg));
887        break;
888    default:
889        g_assert_not_reached();
890    }
891    return true;
892}
893
894static void tcg_out_rld_rc(TCGContext *s, int op, TCGReg ra, TCGReg rs,
895                           int sh, int mb, bool rc)
896{
897    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
898    sh = SH(sh & 0x1f) | (((sh >> 5) & 1) << 1);
899    mb = MB64((mb >> 5) | ((mb << 1) & 0x3f));
900    tcg_out32(s, op | RA(ra) | RS(rs) | sh | mb | rc);
901}
902
903static void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs,
904                        int sh, int mb)
905{
906    tcg_out_rld_rc(s, op, ra, rs, sh, mb, false);
907}
908
909static void tcg_out_rlw_rc(TCGContext *s, int op, TCGReg ra, TCGReg rs,
910                           int sh, int mb, int me, bool rc)
911{
912    tcg_debug_assert((mb & 0x1f) == mb);
913    tcg_debug_assert((me & 0x1f) == me);
914    tcg_out32(s, op | RA(ra) | RS(rs) | SH(sh & 0x1f) | MB(mb) | ME(me) | rc);
915}
916
917static void tcg_out_rlw(TCGContext *s, int op, TCGReg ra, TCGReg rs,
918                        int sh, int mb, int me)
919{
920    tcg_out_rlw_rc(s, op, ra, rs, sh, mb, me, false);
921}
922
923static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
924{
925    tcg_out32(s, EXTSB | RA(dst) | RS(src));
926}
927
928static void tcg_out_ext8u(TCGContext *s, TCGReg dst, TCGReg src)
929{
930    tcg_out32(s, ANDI | SAI(src, dst, 0xff));
931}
932
933static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
934{
935    tcg_out32(s, EXTSH | RA(dst) | RS(src));
936}
937
938static void tcg_out_ext16u(TCGContext *s, TCGReg dst, TCGReg src)
939{
940    tcg_out32(s, ANDI | SAI(src, dst, 0xffff));
941}
942
943static void tcg_out_ext32s(TCGContext *s, TCGReg dst, TCGReg src)
944{
945    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
946    tcg_out32(s, EXTSW | RA(dst) | RS(src));
947}
948
949static void tcg_out_ext32u(TCGContext *s, TCGReg dst, TCGReg src)
950{
951    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
952    tcg_out_rld(s, RLDICL, dst, src, 0, 32);
953}
954
955static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dst, TCGReg src)
956{
957    tcg_out_ext32s(s, dst, src);
958}
959
960static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dst, TCGReg src)
961{
962    tcg_out_ext32u(s, dst, src);
963}
964
965static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rn)
966{
967    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
968    tcg_out_mov(s, TCG_TYPE_I32, rd, rn);
969}
970
971static inline void tcg_out_shli32(TCGContext *s, TCGReg dst, TCGReg src, int c)
972{
973    tcg_out_rlw(s, RLWINM, dst, src, c, 0, 31 - c);
974}
975
976static inline void tcg_out_shli64(TCGContext *s, TCGReg dst, TCGReg src, int c)
977{
978    tcg_out_rld(s, RLDICR, dst, src, c, 63 - c);
979}
980
981static inline void tcg_out_sari32(TCGContext *s, TCGReg dst, TCGReg src, int c)
982{
983    /* Limit immediate shift count lest we create an illegal insn.  */
984    tcg_out32(s, SRAWI | RA(dst) | RS(src) | SH(c & 31));
985}
986
987static inline void tcg_out_shri32(TCGContext *s, TCGReg dst, TCGReg src, int c)
988{
989    tcg_out_rlw(s, RLWINM, dst, src, 32 - c, c, 31);
990}
991
992static inline void tcg_out_shri64(TCGContext *s, TCGReg dst, TCGReg src, int c)
993{
994    tcg_out_rld(s, RLDICL, dst, src, 64 - c, c);
995}
996
997static inline void tcg_out_sari64(TCGContext *s, TCGReg dst, TCGReg src, int c)
998{
999    tcg_out32(s, SRADI | RA(dst) | RS(src) | SH(c & 0x1f) | ((c >> 4) & 2));
1000}
1001
1002static void tcg_out_addpcis(TCGContext *s, TCGReg dst, intptr_t imm)
1003{
1004    uint32_t d0, d1, d2;
1005
1006    tcg_debug_assert((imm & 0xffff) == 0);
1007    tcg_debug_assert(imm == (int32_t)imm);
1008
1009    d2 = extract32(imm, 16, 1);
1010    d1 = extract32(imm, 17, 5);
1011    d0 = extract32(imm, 22, 10);
1012    tcg_out32(s, ADDPCIS | RT(dst) | (d1 << 16) | (d0 << 6) | d2);
1013}
1014
1015static void tcg_out_bswap16(TCGContext *s, TCGReg dst, TCGReg src, int flags)
1016{
1017    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
1018
1019    if (have_isa_3_10) {
1020        tcg_out32(s, BRH | RA(dst) | RS(src));
1021        if (flags & TCG_BSWAP_OS) {
1022            tcg_out_ext16s(s, TCG_TYPE_REG, dst, dst);
1023        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
1024            tcg_out_ext16u(s, dst, dst);
1025        }
1026        return;
1027    }
1028
1029    /*
1030     * In the following,
1031     *   dep(a, b, m) -> (a & ~m) | (b & m)
1032     *
1033     * Begin with:                              src = xxxxabcd
1034     */
1035    /* tmp = rol32(src, 24) & 0x000000ff            = 0000000c */
1036    tcg_out_rlw(s, RLWINM, tmp, src, 24, 24, 31);
1037    /* tmp = dep(tmp, rol32(src, 8), 0x0000ff00)    = 000000dc */
1038    tcg_out_rlw(s, RLWIMI, tmp, src, 8, 16, 23);
1039
1040    if (flags & TCG_BSWAP_OS) {
1041        tcg_out_ext16s(s, TCG_TYPE_REG, dst, tmp);
1042    } else {
1043        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
1044    }
1045}
1046
1047static void tcg_out_bswap32(TCGContext *s, TCGReg dst, TCGReg src, int flags)
1048{
1049    TCGReg tmp = dst == src ? TCG_REG_R0 : dst;
1050
1051    if (have_isa_3_10) {
1052        tcg_out32(s, BRW | RA(dst) | RS(src));
1053        if (flags & TCG_BSWAP_OS) {
1054            tcg_out_ext32s(s, dst, dst);
1055        } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
1056            tcg_out_ext32u(s, dst, dst);
1057        }
1058        return;
1059    }
1060
1061    /*
1062     * Stolen from gcc's builtin_bswap32.
1063     * In the following,
1064     *   dep(a, b, m) -> (a & ~m) | (b & m)
1065     *
1066     * Begin with:                              src = xxxxabcd
1067     */
1068    /* tmp = rol32(src, 8) & 0xffffffff             = 0000bcda */
1069    tcg_out_rlw(s, RLWINM, tmp, src, 8, 0, 31);
1070    /* tmp = dep(tmp, rol32(src, 24), 0xff000000)   = 0000dcda */
1071    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 0, 7);
1072    /* tmp = dep(tmp, rol32(src, 24), 0x0000ff00)   = 0000dcba */
1073    tcg_out_rlw(s, RLWIMI, tmp, src, 24, 16, 23);
1074
1075    if (flags & TCG_BSWAP_OS) {
1076        tcg_out_ext32s(s, dst, tmp);
1077    } else {
1078        tcg_out_mov(s, TCG_TYPE_REG, dst, tmp);
1079    }
1080}
1081
1082static void tcg_out_bswap64(TCGContext *s, TCGReg dst, TCGReg src)
1083{
1084    TCGReg t0 = dst == src ? TCG_REG_R0 : dst;
1085    TCGReg t1 = dst == src ? dst : TCG_REG_R0;
1086
1087    if (have_isa_3_10) {
1088        tcg_out32(s, BRD | RA(dst) | RS(src));
1089        return;
1090    }
1091
1092    /*
1093     * In the following,
1094     *   dep(a, b, m) -> (a & ~m) | (b & m)
1095     *
1096     * Begin with:                              src = abcdefgh
1097     */
1098    /* t0 = rol32(src, 8) & 0xffffffff              = 0000fghe */
1099    tcg_out_rlw(s, RLWINM, t0, src, 8, 0, 31);
1100    /* t0 = dep(t0, rol32(src, 24), 0xff000000)     = 0000hghe */
1101    tcg_out_rlw(s, RLWIMI, t0, src, 24, 0, 7);
1102    /* t0 = dep(t0, rol32(src, 24), 0x0000ff00)     = 0000hgfe */
1103    tcg_out_rlw(s, RLWIMI, t0, src, 24, 16, 23);
1104
1105    /* t0 = rol64(t0, 32)                           = hgfe0000 */
1106    tcg_out_rld(s, RLDICL, t0, t0, 32, 0);
1107    /* t1 = rol64(src, 32)                          = efghabcd */
1108    tcg_out_rld(s, RLDICL, t1, src, 32, 0);
1109
1110    /* t0 = dep(t0, rol32(t1, 24), 0xffffffff)      = hgfebcda */
1111    tcg_out_rlw(s, RLWIMI, t0, t1, 8, 0, 31);
1112    /* t0 = dep(t0, rol32(t1, 24), 0xff000000)      = hgfedcda */
1113    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 0, 7);
1114    /* t0 = dep(t0, rol32(t1, 24), 0x0000ff00)      = hgfedcba */
1115    tcg_out_rlw(s, RLWIMI, t0, t1, 24, 16, 23);
1116
1117    tcg_out_mov(s, TCG_TYPE_REG, dst, t0);
1118}
1119
1120/* Emit a move into ret of arg, if it can be done in one insn.  */
1121static bool tcg_out_movi_one(TCGContext *s, TCGReg ret, tcg_target_long arg)
1122{
1123    if (arg == (int16_t)arg) {
1124        tcg_out32(s, ADDI | TAI(ret, 0, arg));
1125        return true;
1126    }
1127    if (arg == (int32_t)arg && (arg & 0xffff) == 0) {
1128        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
1129        return true;
1130    }
1131    return false;
1132}
1133
1134static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
1135                             tcg_target_long arg, bool in_prologue)
1136{
1137    intptr_t tb_diff;
1138    tcg_target_long tmp;
1139    int shift;
1140
1141    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1142
1143    if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1144        arg = (int32_t)arg;
1145    }
1146
1147    /* Load 16-bit immediates with one insn.  */
1148    if (tcg_out_movi_one(s, ret, arg)) {
1149        return;
1150    }
1151
1152    /* Load addresses within the TB with one insn.  */
1153    tb_diff = ppc_tbrel_diff(s, (void *)arg);
1154    if (!in_prologue && USE_REG_TB && tb_diff == (int16_t)tb_diff) {
1155        tcg_out32(s, ADDI | TAI(ret, TCG_REG_TB, tb_diff));
1156        return;
1157    }
1158
1159    /*
1160     * Load values up to 34 bits, and pc-relative addresses,
1161     * with one prefixed insn.
1162     */
1163    if (have_isa_3_10) {
1164        if (arg == sextract64(arg, 0, 34)) {
1165            /* pli ret,value = paddi ret,0,value,0 */
1166            tcg_out_mls_d(s, ADDI, ret, 0, arg, 0);
1167            return;
1168        }
1169
1170        tmp = tcg_pcrel_diff_for_prefix(s, (void *)arg);
1171        if (tmp == sextract64(tmp, 0, 34)) {
1172            /* pla ret,value = paddi ret,0,value,1 */
1173            tcg_out_mls_d(s, ADDI, ret, 0, tmp, 1);
1174            return;
1175        }
1176    }
1177
1178    /* Load 32-bit immediates with two insns.  Note that we've already
1179       eliminated bare ADDIS, so we know both insns are required.  */
1180    if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) {
1181        tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
1182        tcg_out32(s, ORI | SAI(ret, ret, arg));
1183        return;
1184    }
1185    if (arg == (uint32_t)arg && !(arg & 0x8000)) {
1186        tcg_out32(s, ADDI | TAI(ret, 0, arg));
1187        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
1188        return;
1189    }
1190
1191    /* Load masked 16-bit value.  */
1192    if (arg > 0 && (arg & 0x8000)) {
1193        tmp = arg | 0x7fff;
1194        if ((tmp & (tmp + 1)) == 0) {
1195            int mb = clz64(tmp + 1) + 1;
1196            tcg_out32(s, ADDI | TAI(ret, 0, arg));
1197            tcg_out_rld(s, RLDICL, ret, ret, 0, mb);
1198            return;
1199        }
1200    }
1201
1202    /* Load common masks with 2 insns.  */
1203    shift = ctz64(arg);
1204    tmp = arg >> shift;
1205    if (tmp == (int16_t)tmp) {
1206        tcg_out32(s, ADDI | TAI(ret, 0, tmp));
1207        tcg_out_shli64(s, ret, ret, shift);
1208        return;
1209    }
1210    shift = clz64(arg);
1211    if (tcg_out_movi_one(s, ret, arg << shift)) {
1212        tcg_out_shri64(s, ret, ret, shift);
1213        return;
1214    }
1215
1216    /* Load addresses within 2GB with 2 insns. */
1217    if (have_isa_3_00) {
1218        intptr_t hi = tcg_pcrel_diff(s, (void *)arg) - 4;
1219        int16_t lo = hi;
1220
1221        hi -= lo;
1222        if (hi == (int32_t)hi) {
1223            tcg_out_addpcis(s, TCG_REG_TMP2, hi);
1224            tcg_out32(s, ADDI | TAI(ret, TCG_REG_TMP2, lo));
1225            return;
1226        }
1227    }
1228
1229    /* Load addresses within 2GB of TB with 2 (or rarely 3) insns.  */
1230    if (!in_prologue && USE_REG_TB && tb_diff == (int32_t)tb_diff) {
1231        tcg_out_mem_long(s, ADDI, ADD, ret, TCG_REG_TB, tb_diff);
1232        return;
1233    }
1234
1235    /* Use the constant pool, if possible.  */
1236    if (!in_prologue && USE_REG_TB) {
1237        new_pool_label(s, arg, R_PPC_ADDR16, s->code_ptr,
1238                       ppc_tbrel_diff(s, NULL));
1239        tcg_out32(s, LD | TAI(ret, TCG_REG_TB, 0));
1240        return;
1241    }
1242    if (have_isa_3_10) {
1243        tcg_out_8ls_d(s, PLD, ret, 0, 0, 1);
1244        new_pool_label(s, arg, R_PPC64_PCREL34, s->code_ptr - 2, 0);
1245        return;
1246    }
1247    if (have_isa_3_00) {
1248        tcg_out_addpcis(s, TCG_REG_TMP2, 0);
1249        new_pool_label(s, arg, R_PPC_REL14, s->code_ptr, 0);
1250        tcg_out32(s, LD | TAI(ret, TCG_REG_TMP2, 0));
1251        return;
1252    }
1253
1254    tmp = arg >> 31 >> 1;
1255    tcg_out_movi(s, TCG_TYPE_I32, ret, tmp);
1256    if (tmp) {
1257        tcg_out_shli64(s, ret, ret, 32);
1258    }
1259    if (arg & 0xffff0000) {
1260        tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
1261    }
1262    if (arg & 0xffff) {
1263        tcg_out32(s, ORI | SAI(ret, ret, arg));
1264    }
1265}
1266
1267static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
1268                             TCGReg ret, int64_t val)
1269{
1270    uint32_t load_insn;
1271    int rel, low;
1272    intptr_t add;
1273
1274    switch (vece) {
1275    case MO_8:
1276        low = (int8_t)val;
1277        if (low >= -16 && low < 16) {
1278            tcg_out32(s, VSPLTISB | VRT(ret) | ((val & 31) << 16));
1279            return;
1280        }
1281        if (have_isa_3_00) {
1282            tcg_out32(s, XXSPLTIB | VRT(ret) | ((val & 0xff) << 11));
1283            return;
1284        }
1285        break;
1286
1287    case MO_16:
1288        low = (int16_t)val;
1289        if (low >= -16 && low < 16) {
1290            tcg_out32(s, VSPLTISH | VRT(ret) | ((val & 31) << 16));
1291            return;
1292        }
1293        break;
1294
1295    case MO_32:
1296        low = (int32_t)val;
1297        if (low >= -16 && low < 16) {
1298            tcg_out32(s, VSPLTISW | VRT(ret) | ((val & 31) << 16));
1299            return;
1300        }
1301        break;
1302    }
1303
1304    /*
1305     * Otherwise we must load the value from the constant pool.
1306     */
1307    if (USE_REG_TB) {
1308        rel = R_PPC_ADDR16;
1309        add = ppc_tbrel_diff(s, NULL);
1310    } else if (have_isa_3_10) {
1311        if (type == TCG_TYPE_V64) {
1312            tcg_out_8ls_d(s, PLXSD, ret & 31, 0, 0, 1);
1313            new_pool_label(s, val, R_PPC64_PCREL34, s->code_ptr - 2, 0);
1314        } else {
1315            tcg_out_8ls_d(s, PLXV, ret & 31, 0, 0, 1);
1316            new_pool_l2(s, R_PPC64_PCREL34, s->code_ptr - 2, 0, val, val);
1317        }
1318        return;
1319    } else if (have_isa_3_00) {
1320        tcg_out_addpcis(s, TCG_REG_TMP1, 0);
1321        rel = R_PPC_REL14;
1322        add = 0;
1323    } else {
1324        rel = R_PPC_ADDR32;
1325        add = 0;
1326    }
1327
1328    if (have_vsx) {
1329        load_insn = type == TCG_TYPE_V64 ? LXSDX : LXVDSX;
1330        load_insn |= VRT(ret) | RB(TCG_REG_TMP1);
1331        if (TCG_TARGET_REG_BITS == 64) {
1332            new_pool_label(s, val, rel, s->code_ptr, add);
1333        } else {
1334            new_pool_l2(s, rel, s->code_ptr, add, val >> 32, val);
1335        }
1336    } else {
1337        load_insn = LVX | VRT(ret) | RB(TCG_REG_TMP1);
1338        if (TCG_TARGET_REG_BITS == 64) {
1339            new_pool_l2(s, rel, s->code_ptr, add, val, val);
1340        } else {
1341            new_pool_l4(s, rel, s->code_ptr, add,
1342                        val >> 32, val, val >> 32, val);
1343        }
1344    }
1345
1346    if (USE_REG_TB) {
1347        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, 0, 0));
1348        load_insn |= RA(TCG_REG_TB);
1349    } else if (have_isa_3_00) {
1350        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, TCG_REG_TMP1, 0));
1351    } else {
1352        tcg_out32(s, ADDIS | TAI(TCG_REG_TMP1, 0, 0));
1353        tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, TCG_REG_TMP1, 0));
1354    }
1355    tcg_out32(s, load_insn);
1356}
1357
1358static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,
1359                         tcg_target_long arg)
1360{
1361    switch (type) {
1362    case TCG_TYPE_I32:
1363    case TCG_TYPE_I64:
1364        tcg_debug_assert(ret < TCG_REG_V0);
1365        tcg_out_movi_int(s, type, ret, arg, false);
1366        break;
1367
1368    default:
1369        g_assert_not_reached();
1370    }
1371}
1372
1373static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
1374{
1375    return false;
1376}
1377
1378static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
1379                             tcg_target_long imm)
1380{
1381    /* This function is only used for passing structs by reference. */
1382    g_assert_not_reached();
1383}
1384
1385static bool mask_operand(uint32_t c, int *mb, int *me)
1386{
1387    uint32_t lsb, test;
1388
1389    /* Accept a bit pattern like:
1390           0....01....1
1391           1....10....0
1392           0..01..10..0
1393       Keep track of the transitions.  */
1394    if (c == 0 || c == -1) {
1395        return false;
1396    }
1397    test = c;
1398    lsb = test & -test;
1399    test += lsb;
1400    if (test & (test - 1)) {
1401        return false;
1402    }
1403
1404    *me = clz32(lsb);
1405    *mb = test ? clz32(test & -test) + 1 : 0;
1406    return true;
1407}
1408
1409static bool mask64_operand(uint64_t c, int *mb, int *me)
1410{
1411    uint64_t lsb;
1412
1413    if (c == 0) {
1414        return false;
1415    }
1416
1417    lsb = c & -c;
1418    /* Accept 1..10..0.  */
1419    if (c == -lsb) {
1420        *mb = 0;
1421        *me = clz64(lsb);
1422        return true;
1423    }
1424    /* Accept 0..01..1.  */
1425    if (lsb == 1 && (c & (c + 1)) == 0) {
1426        *mb = clz64(c + 1) + 1;
1427        *me = 63;
1428        return true;
1429    }
1430    return false;
1431}
1432
1433static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1434{
1435    int mb, me;
1436
1437    if (mask_operand(c, &mb, &me)) {
1438        tcg_out_rlw(s, RLWINM, dst, src, 0, mb, me);
1439    } else if ((c & 0xffff) == c) {
1440        tcg_out32(s, ANDI | SAI(src, dst, c));
1441        return;
1442    } else if ((c & 0xffff0000) == c) {
1443        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1444        return;
1445    } else {
1446        tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R0, c);
1447        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1448    }
1449}
1450
1451static void tcg_out_andi64(TCGContext *s, TCGReg dst, TCGReg src, uint64_t c)
1452{
1453    int mb, me;
1454
1455    tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1456    if (mask64_operand(c, &mb, &me)) {
1457        if (mb == 0) {
1458            tcg_out_rld(s, RLDICR, dst, src, 0, me);
1459        } else {
1460            tcg_out_rld(s, RLDICL, dst, src, 0, mb);
1461        }
1462    } else if ((c & 0xffff) == c) {
1463        tcg_out32(s, ANDI | SAI(src, dst, c));
1464        return;
1465    } else if ((c & 0xffff0000) == c) {
1466        tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
1467        return;
1468    } else {
1469        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, c);
1470        tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
1471    }
1472}
1473
1474static void tcg_out_zori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c,
1475                           int op_lo, int op_hi)
1476{
1477    if (c >> 16) {
1478        tcg_out32(s, op_hi | SAI(src, dst, c >> 16));
1479        src = dst;
1480    }
1481    if (c & 0xffff) {
1482        tcg_out32(s, op_lo | SAI(src, dst, c));
1483        src = dst;
1484    }
1485}
1486
1487static void tcg_out_ori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1488{
1489    tcg_out_zori32(s, dst, src, c, ORI, ORIS);
1490}
1491
1492static void tcg_out_xori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
1493{
1494    tcg_out_zori32(s, dst, src, c, XORI, XORIS);
1495}
1496
1497static void tcg_out_b(TCGContext *s, int mask, const tcg_insn_unit *target)
1498{
1499    ptrdiff_t disp = tcg_pcrel_diff(s, target);
1500    if (in_range_b(disp)) {
1501        tcg_out32(s, B | (disp & 0x3fffffc) | mask);
1502    } else {
1503        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, (uintptr_t)target);
1504        tcg_out32(s, MTSPR | RS(TCG_REG_R0) | CTR);
1505        tcg_out32(s, BCCTR | BO_ALWAYS | mask);
1506    }
1507}
1508
1509static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
1510                             TCGReg base, tcg_target_long offset)
1511{
1512    tcg_target_long orig = offset, l0, l1, extra = 0, align = 0;
1513    bool is_int_store = false;
1514    TCGReg rs = TCG_REG_TMP1;
1515
1516    switch (opi) {
1517    case LD: case LWA:
1518        align = 3;
1519        /* FALLTHRU */
1520    default:
1521        if (rt > TCG_REG_R0 && rt < TCG_REG_V0) {
1522            rs = rt;
1523            break;
1524        }
1525        break;
1526    case LXSD:
1527    case STXSD:
1528        align = 3;
1529        break;
1530    case LXV:
1531    case STXV:
1532        align = 15;
1533        break;
1534    case STD:
1535        align = 3;
1536        /* FALLTHRU */
1537    case STB: case STH: case STW:
1538        is_int_store = true;
1539        break;
1540    }
1541
1542    /* For unaligned or large offsets, use the prefixed form. */
1543    if (have_isa_3_10
1544        && (offset != (int16_t)offset || (offset & align))
1545        && offset == sextract64(offset, 0, 34)) {
1546        /*
1547         * Note that the MLS:D insns retain their un-prefixed opcode,
1548         * while the 8LS:D insns use a different opcode space.
1549         */
1550        switch (opi) {
1551        case LBZ:
1552        case LHZ:
1553        case LHA:
1554        case LWZ:
1555        case STB:
1556        case STH:
1557        case STW:
1558        case ADDI:
1559            tcg_out_mls_d(s, opi, rt, base, offset, 0);
1560            return;
1561        case LWA:
1562            tcg_out_8ls_d(s, PLWA, rt, base, offset, 0);
1563            return;
1564        case LD:
1565            tcg_out_8ls_d(s, PLD, rt, base, offset, 0);
1566            return;
1567        case STD:
1568            tcg_out_8ls_d(s, PSTD, rt, base, offset, 0);
1569            return;
1570        case LXSD:
1571            tcg_out_8ls_d(s, PLXSD, rt & 31, base, offset, 0);
1572            return;
1573        case STXSD:
1574            tcg_out_8ls_d(s, PSTXSD, rt & 31, base, offset, 0);
1575            return;
1576        case LXV:
1577            tcg_out_8ls_d(s, PLXV, rt & 31, base, offset, 0);
1578            return;
1579        case STXV:
1580            tcg_out_8ls_d(s, PSTXV, rt & 31, base, offset, 0);
1581            return;
1582        }
1583    }
1584
1585    /* For unaligned, or very large offsets, use the indexed form.  */
1586    if (offset & align || offset != (int32_t)offset || opi == 0) {
1587        if (rs == base) {
1588            rs = TCG_REG_R0;
1589        }
1590        tcg_debug_assert(!is_int_store || rs != rt);
1591        tcg_out_movi(s, TCG_TYPE_PTR, rs, orig);
1592        tcg_out32(s, opx | TAB(rt & 31, base, rs));
1593        return;
1594    }
1595
1596    l0 = (int16_t)offset;
1597    offset = (offset - l0) >> 16;
1598    l1 = (int16_t)offset;
1599
1600    if (l1 < 0 && orig >= 0) {
1601        extra = 0x4000;
1602        l1 = (int16_t)(offset - 0x4000);
1603    }
1604    if (l1) {
1605        tcg_out32(s, ADDIS | TAI(rs, base, l1));
1606        base = rs;
1607    }
1608    if (extra) {
1609        tcg_out32(s, ADDIS | TAI(rs, base, extra));
1610        base = rs;
1611    }
1612    if (opi != ADDI || base != rt || l0 != 0) {
1613        tcg_out32(s, opi | TAI(rt & 31, base, l0));
1614    }
1615}
1616
1617static void tcg_out_vsldoi(TCGContext *s, TCGReg ret,
1618                           TCGReg va, TCGReg vb, int shb)
1619{
1620    tcg_out32(s, VSLDOI | VRT(ret) | VRA(va) | VRB(vb) | (shb << 6));
1621}
1622
1623static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
1624                       TCGReg base, intptr_t offset)
1625{
1626    int shift;
1627
1628    switch (type) {
1629    case TCG_TYPE_I32:
1630        if (ret < TCG_REG_V0) {
1631            tcg_out_mem_long(s, LWZ, LWZX, ret, base, offset);
1632            break;
1633        }
1634        if (have_isa_2_07 && have_vsx) {
1635            tcg_out_mem_long(s, 0, LXSIWZX, ret, base, offset);
1636            break;
1637        }
1638        tcg_debug_assert((offset & 3) == 0);
1639        tcg_out_mem_long(s, 0, LVEWX, ret, base, offset);
1640        shift = (offset - 4) & 0xc;
1641        if (shift) {
1642            tcg_out_vsldoi(s, ret, ret, ret, shift);
1643        }
1644        break;
1645    case TCG_TYPE_I64:
1646        if (ret < TCG_REG_V0) {
1647            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1648            tcg_out_mem_long(s, LD, LDX, ret, base, offset);
1649            break;
1650        }
1651        /* fallthru */
1652    case TCG_TYPE_V64:
1653        tcg_debug_assert(ret >= TCG_REG_V0);
1654        if (have_vsx) {
1655            tcg_out_mem_long(s, have_isa_3_00 ? LXSD : 0, LXSDX,
1656                             ret, base, offset);
1657            break;
1658        }
1659        tcg_debug_assert((offset & 7) == 0);
1660        tcg_out_mem_long(s, 0, LVX, ret, base, offset & -16);
1661        if (offset & 8) {
1662            tcg_out_vsldoi(s, ret, ret, ret, 8);
1663        }
1664        break;
1665    case TCG_TYPE_V128:
1666        tcg_debug_assert(ret >= TCG_REG_V0);
1667        tcg_debug_assert((offset & 15) == 0);
1668        tcg_out_mem_long(s, have_isa_3_00 ? LXV : 0,
1669                         LVX, ret, base, offset);
1670        break;
1671    default:
1672        g_assert_not_reached();
1673    }
1674}
1675
1676static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
1677                              TCGReg base, intptr_t offset)
1678{
1679    int shift;
1680
1681    switch (type) {
1682    case TCG_TYPE_I32:
1683        if (arg < TCG_REG_V0) {
1684            tcg_out_mem_long(s, STW, STWX, arg, base, offset);
1685            break;
1686        }
1687        if (have_isa_2_07 && have_vsx) {
1688            tcg_out_mem_long(s, 0, STXSIWX, arg, base, offset);
1689            break;
1690        }
1691        assert((offset & 3) == 0);
1692        tcg_debug_assert((offset & 3) == 0);
1693        shift = (offset - 4) & 0xc;
1694        if (shift) {
1695            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, shift);
1696            arg = TCG_VEC_TMP1;
1697        }
1698        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1699        break;
1700    case TCG_TYPE_I64:
1701        if (arg < TCG_REG_V0) {
1702            tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
1703            tcg_out_mem_long(s, STD, STDX, arg, base, offset);
1704            break;
1705        }
1706        /* fallthru */
1707    case TCG_TYPE_V64:
1708        tcg_debug_assert(arg >= TCG_REG_V0);
1709        if (have_vsx) {
1710            tcg_out_mem_long(s, have_isa_3_00 ? STXSD : 0,
1711                             STXSDX, arg, base, offset);
1712            break;
1713        }
1714        tcg_debug_assert((offset & 7) == 0);
1715        if (offset & 8) {
1716            tcg_out_vsldoi(s, TCG_VEC_TMP1, arg, arg, 8);
1717            arg = TCG_VEC_TMP1;
1718        }
1719        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset);
1720        tcg_out_mem_long(s, 0, STVEWX, arg, base, offset + 4);
1721        break;
1722    case TCG_TYPE_V128:
1723        tcg_debug_assert(arg >= TCG_REG_V0);
1724        tcg_out_mem_long(s, have_isa_3_00 ? STXV : 0,
1725                         STVX, arg, base, offset);
1726        break;
1727    default:
1728        g_assert_not_reached();
1729    }
1730}
1731
1732static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1733                               TCGReg base, intptr_t ofs)
1734{
1735    return false;
1736}
1737
1738/*
1739 * Set dest non-zero if and only if (arg1 & arg2) is non-zero.
1740 * If RC, then also set RC0.
1741 */
1742static void tcg_out_test(TCGContext *s, TCGReg dest, TCGReg arg1, TCGArg arg2,
1743                         bool const_arg2, TCGType type, bool rc)
1744{
1745    int mb, me;
1746
1747    if (!const_arg2) {
1748        tcg_out32(s, AND | SAB(arg1, dest, arg2) | rc);
1749        return;
1750    }
1751
1752    if (type == TCG_TYPE_I32) {
1753        arg2 = (uint32_t)arg2;
1754    }
1755
1756    if ((arg2 & ~0xffff) == 0) {
1757        tcg_out32(s, ANDI | SAI(arg1, dest, arg2));
1758        return;
1759    }
1760    if ((arg2 & ~0xffff0000ull) == 0) {
1761        tcg_out32(s, ANDIS | SAI(arg1, dest, arg2 >> 16));
1762        return;
1763    }
1764    if (arg2 == (uint32_t)arg2 && mask_operand(arg2, &mb, &me)) {
1765        tcg_out_rlw_rc(s, RLWINM, dest, arg1, 0, mb, me, rc);
1766        return;
1767    }
1768    if (TCG_TARGET_REG_BITS == 64) {
1769        int sh = clz64(arg2);
1770        if (mask64_operand(arg2 << sh, &mb, &me)) {
1771            tcg_out_rld_rc(s, RLDICR, dest, arg1, sh, me, rc);
1772            return;
1773        }
1774    }
1775    /* Constraints should satisfy this. */
1776    g_assert_not_reached();
1777}
1778
1779static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2,
1780                        int const_arg2, int cr, TCGType type)
1781{
1782    int imm;
1783    uint32_t op;
1784
1785    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1786
1787    /*
1788     * Simplify the comparisons below wrt CMPI.
1789     * All of the tests are 16-bit, so a 32-bit sign extend always works.
1790     */
1791    if (type == TCG_TYPE_I32) {
1792        arg2 = (int32_t)arg2;
1793    }
1794
1795    switch (cond) {
1796    case TCG_COND_EQ:
1797    case TCG_COND_NE:
1798        if (const_arg2) {
1799            if ((int16_t) arg2 == arg2) {
1800                op = CMPI;
1801                imm = 1;
1802                break;
1803            } else if ((uint16_t) arg2 == arg2) {
1804                op = CMPLI;
1805                imm = 1;
1806                break;
1807            }
1808        }
1809        op = CMPL;
1810        imm = 0;
1811        break;
1812
1813    case TCG_COND_TSTEQ:
1814    case TCG_COND_TSTNE:
1815        tcg_debug_assert(cr == 0);
1816        tcg_out_test(s, TCG_REG_R0, arg1, arg2, const_arg2, type, true);
1817        return;
1818
1819    case TCG_COND_LT:
1820    case TCG_COND_GE:
1821    case TCG_COND_LE:
1822    case TCG_COND_GT:
1823        if (const_arg2) {
1824            if ((int16_t) arg2 == arg2) {
1825                op = CMPI;
1826                imm = 1;
1827                break;
1828            }
1829        }
1830        op = CMP;
1831        imm = 0;
1832        break;
1833
1834    case TCG_COND_LTU:
1835    case TCG_COND_GEU:
1836    case TCG_COND_LEU:
1837    case TCG_COND_GTU:
1838        if (const_arg2) {
1839            if ((uint16_t) arg2 == arg2) {
1840                op = CMPLI;
1841                imm = 1;
1842                break;
1843            }
1844        }
1845        op = CMPL;
1846        imm = 0;
1847        break;
1848
1849    default:
1850        g_assert_not_reached();
1851    }
1852    op |= BF(cr) | ((type == TCG_TYPE_I64) << 21);
1853
1854    if (imm) {
1855        tcg_out32(s, op | RA(arg1) | (arg2 & 0xffff));
1856    } else {
1857        if (const_arg2) {
1858            tcg_out_movi(s, type, TCG_REG_R0, arg2);
1859            arg2 = TCG_REG_R0;
1860        }
1861        tcg_out32(s, op | RA(arg1) | RB(arg2));
1862    }
1863}
1864
1865static void tcg_out_setcond_eq0(TCGContext *s, TCGType type,
1866                                TCGReg dst, TCGReg src, bool neg)
1867{
1868    if (neg && (TCG_TARGET_REG_BITS == 32 || type == TCG_TYPE_I64)) {
1869        /*
1870         * X != 0 implies X + -1 generates a carry.
1871         * RT = (~X + X) + CA
1872         *    = -1 + CA
1873         *    = CA ? 0 : -1
1874         */
1875        tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1));
1876        tcg_out32(s, SUBFE | TAB(dst, src, src));
1877        return;
1878    }
1879
1880    if (type == TCG_TYPE_I32) {
1881        tcg_out32(s, CNTLZW | RS(src) | RA(dst));
1882        tcg_out_shri32(s, dst, dst, 5);
1883    } else {
1884        tcg_out32(s, CNTLZD | RS(src) | RA(dst));
1885        tcg_out_shri64(s, dst, dst, 6);
1886    }
1887    if (neg) {
1888        tcg_out32(s, NEG | RT(dst) | RA(dst));
1889    }
1890}
1891
1892static void tcg_out_setcond_ne0(TCGContext *s, TCGType type,
1893                                TCGReg dst, TCGReg src, bool neg)
1894{
1895    if (!neg && (TCG_TARGET_REG_BITS == 32 || type == TCG_TYPE_I64)) {
1896        /*
1897         * X != 0 implies X + -1 generates a carry.  Extra addition
1898         * trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C.
1899         */
1900        tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1));
1901        tcg_out32(s, SUBFE | TAB(dst, TCG_REG_R0, src));
1902        return;
1903    }
1904    tcg_out_setcond_eq0(s, type, dst, src, false);
1905    if (neg) {
1906        tcg_out32(s, ADDI | TAI(dst, dst, -1));
1907    } else {
1908        tcg_out_xori32(s, dst, dst, 1);
1909    }
1910}
1911
1912static TCGReg tcg_gen_setcond_xor(TCGContext *s, TCGReg arg1, TCGArg arg2,
1913                                  bool const_arg2)
1914{
1915    if (const_arg2) {
1916        if ((uint32_t)arg2 == arg2) {
1917            tcg_out_xori32(s, TCG_REG_R0, arg1, arg2);
1918        } else {
1919            tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, arg2);
1920            tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, TCG_REG_R0));
1921        }
1922    } else {
1923        tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, arg2));
1924    }
1925    return TCG_REG_R0;
1926}
1927
1928static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
1929                            TCGArg arg0, TCGArg arg1, TCGArg arg2,
1930                            int const_arg2, bool neg)
1931{
1932    int sh;
1933    bool inv;
1934
1935    tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1936
1937    /* Ignore high bits of a potential constant arg2.  */
1938    if (type == TCG_TYPE_I32) {
1939        arg2 = (uint32_t)arg2;
1940    }
1941
1942    /* With SETBC/SETBCR, we can always implement with 2 insns. */
1943    if (have_isa_3_10) {
1944        tcg_insn_unit bi, opc;
1945
1946        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 0, type);
1947
1948        /* Re-use tcg_to_bc for BI and BO_COND_{TRUE,FALSE}. */
1949        bi = tcg_to_bc[cond] & (0x1f << 16);
1950        if (tcg_to_bc[cond] & BO(8)) {
1951            opc = neg ? SETNBC : SETBC;
1952        } else {
1953            opc = neg ? SETNBCR : SETBCR;
1954        }
1955        tcg_out32(s, opc | RT(arg0) | bi);
1956        return;
1957    }
1958
1959    /* Handle common and trivial cases before handling anything else.  */
1960    if (arg2 == 0) {
1961        switch (cond) {
1962        case TCG_COND_EQ:
1963            tcg_out_setcond_eq0(s, type, arg0, arg1, neg);
1964            return;
1965        case TCG_COND_NE:
1966            tcg_out_setcond_ne0(s, type, arg0, arg1, neg);
1967            return;
1968        case TCG_COND_GE:
1969            tcg_out32(s, NOR | SAB(arg1, arg0, arg1));
1970            arg1 = arg0;
1971            /* FALLTHRU */
1972        case TCG_COND_LT:
1973            /* Extract the sign bit.  */
1974            if (type == TCG_TYPE_I32) {
1975                if (neg) {
1976                    tcg_out_sari32(s, arg0, arg1, 31);
1977                } else {
1978                    tcg_out_shri32(s, arg0, arg1, 31);
1979                }
1980            } else {
1981                if (neg) {
1982                    tcg_out_sari64(s, arg0, arg1, 63);
1983                } else {
1984                    tcg_out_shri64(s, arg0, arg1, 63);
1985                }
1986            }
1987            return;
1988        default:
1989            break;
1990        }
1991    }
1992
1993    /* If we have ISEL, we can implement everything with 3 or 4 insns.
1994       All other cases below are also at least 3 insns, so speed up the
1995       code generator by not considering them and always using ISEL.  */
1996    if (have_isel) {
1997        int isel, tab;
1998
1999        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 0, type);
2000
2001        isel = tcg_to_isel[cond];
2002
2003        tcg_out_movi(s, type, arg0, neg ? -1 : 1);
2004        if (isel & 1) {
2005            /* arg0 = (bc ? 0 : 1) */
2006            tab = TAB(arg0, 0, arg0);
2007            isel &= ~1;
2008        } else {
2009            /* arg0 = (bc ? 1 : 0) */
2010            tcg_out_movi(s, type, TCG_REG_R0, 0);
2011            tab = TAB(arg0, arg0, TCG_REG_R0);
2012        }
2013        tcg_out32(s, isel | tab);
2014        return;
2015    }
2016
2017    inv = false;
2018    switch (cond) {
2019    case TCG_COND_EQ:
2020        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
2021        tcg_out_setcond_eq0(s, type, arg0, arg1, neg);
2022        break;
2023
2024    case TCG_COND_NE:
2025        arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
2026        tcg_out_setcond_ne0(s, type, arg0, arg1, neg);
2027        break;
2028
2029    case TCG_COND_TSTEQ:
2030        tcg_out_test(s, TCG_REG_R0, arg1, arg2, const_arg2, type, false);
2031        tcg_out_setcond_eq0(s, type, arg0, TCG_REG_R0, neg);
2032        break;
2033
2034    case TCG_COND_TSTNE:
2035        tcg_out_test(s, TCG_REG_R0, arg1, arg2, const_arg2, type, false);
2036        tcg_out_setcond_ne0(s, type, arg0, TCG_REG_R0, neg);
2037        break;
2038
2039    case TCG_COND_LE:
2040    case TCG_COND_LEU:
2041        inv = true;
2042        /* fall through */
2043    case TCG_COND_GT:
2044    case TCG_COND_GTU:
2045        sh = 30; /* CR7 CR_GT */
2046        goto crtest;
2047
2048    case TCG_COND_GE:
2049    case TCG_COND_GEU:
2050        inv = true;
2051        /* fall through */
2052    case TCG_COND_LT:
2053    case TCG_COND_LTU:
2054        sh = 29; /* CR7 CR_LT */
2055        goto crtest;
2056
2057    crtest:
2058        tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
2059        tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
2060        tcg_out_rlw(s, RLWINM, arg0, TCG_REG_R0, sh, 31, 31);
2061        if (neg && inv) {
2062            tcg_out32(s, ADDI | TAI(arg0, arg0, -1));
2063        } else if (neg) {
2064            tcg_out32(s, NEG | RT(arg0) | RA(arg0));
2065        } else if (inv) {
2066            tcg_out_xori32(s, arg0, arg0, 1);
2067        }
2068        break;
2069
2070    default:
2071        g_assert_not_reached();
2072    }
2073}
2074
2075static void tcg_out_bc(TCGContext *s, TCGCond cond, int bd)
2076{
2077    tcg_out32(s, tcg_to_bc[cond] | bd);
2078}
2079
2080static void tcg_out_bc_lab(TCGContext *s, TCGCond cond, TCGLabel *l)
2081{
2082    int bd = 0;
2083    if (l->has_value) {
2084        bd = reloc_pc14_val(tcg_splitwx_to_rx(s->code_ptr), l->u.value_ptr);
2085    } else {
2086        tcg_out_reloc(s, s->code_ptr, R_PPC_REL14, l, 0);
2087    }
2088    tcg_out_bc(s, cond, bd);
2089}
2090
2091static void tcg_out_brcond(TCGContext *s, TCGCond cond,
2092                           TCGArg arg1, TCGArg arg2, int const_arg2,
2093                           TCGLabel *l, TCGType type)
2094{
2095    tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 0, type);
2096    tcg_out_bc_lab(s, cond, l);
2097}
2098
2099static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
2100                            TCGArg dest, TCGArg c1, TCGArg c2, TCGArg v1,
2101                            TCGArg v2, bool const_c2)
2102{
2103    /* If for some reason both inputs are zero, don't produce bad code.  */
2104    if (v1 == 0 && v2 == 0) {
2105        tcg_out_movi(s, type, dest, 0);
2106        return;
2107    }
2108
2109    tcg_out_cmp(s, cond, c1, c2, const_c2, 0, type);
2110
2111    if (have_isel) {
2112        int isel = tcg_to_isel[cond];
2113
2114        /* Swap the V operands if the operation indicates inversion.  */
2115        if (isel & 1) {
2116            int t = v1;
2117            v1 = v2;
2118            v2 = t;
2119            isel &= ~1;
2120        }
2121        /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand.  */
2122        if (v2 == 0) {
2123            tcg_out_movi(s, type, TCG_REG_R0, 0);
2124        }
2125        tcg_out32(s, isel | TAB(dest, v1, v2));
2126    } else {
2127        if (dest == v2) {
2128            cond = tcg_invert_cond(cond);
2129            v2 = v1;
2130        } else if (dest != v1) {
2131            if (v1 == 0) {
2132                tcg_out_movi(s, type, dest, 0);
2133            } else {
2134                tcg_out_mov(s, type, dest, v1);
2135            }
2136        }
2137        /* Branch forward over one insn */
2138        tcg_out_bc(s, cond, 8);
2139        if (v2 == 0) {
2140            tcg_out_movi(s, type, dest, 0);
2141        } else {
2142            tcg_out_mov(s, type, dest, v2);
2143        }
2144    }
2145}
2146
2147static void tcg_out_cntxz(TCGContext *s, TCGType type, uint32_t opc,
2148                          TCGArg a0, TCGArg a1, TCGArg a2, bool const_a2)
2149{
2150    if (const_a2 && a2 == (type == TCG_TYPE_I32 ? 32 : 64)) {
2151        tcg_out32(s, opc | RA(a0) | RS(a1));
2152    } else {
2153        tcg_out_cmp(s, TCG_COND_EQ, a1, 0, 1, 0, type);
2154        /* Note that the only other valid constant for a2 is 0.  */
2155        if (have_isel) {
2156            tcg_out32(s, opc | RA(TCG_REG_R0) | RS(a1));
2157            tcg_out32(s, tcg_to_isel[TCG_COND_EQ] | TAB(a0, a2, TCG_REG_R0));
2158        } else if (!const_a2 && a0 == a2) {
2159            tcg_out_bc(s, TCG_COND_EQ, 8);
2160            tcg_out32(s, opc | RA(a0) | RS(a1));
2161        } else {
2162            tcg_out32(s, opc | RA(a0) | RS(a1));
2163            tcg_out_bc(s, TCG_COND_NE, 8);
2164            if (const_a2) {
2165                tcg_out_movi(s, type, a0, 0);
2166            } else {
2167                tcg_out_mov(s, type, a0, a2);
2168            }
2169        }
2170    }
2171}
2172
2173static void tcg_out_cmp2(TCGContext *s, const TCGArg *args,
2174                         const int *const_args)
2175{
2176    static const struct { uint8_t bit1, bit2; } bits[] = {
2177        [TCG_COND_LT ] = { CR_LT, CR_LT },
2178        [TCG_COND_LE ] = { CR_LT, CR_GT },
2179        [TCG_COND_GT ] = { CR_GT, CR_GT },
2180        [TCG_COND_GE ] = { CR_GT, CR_LT },
2181        [TCG_COND_LTU] = { CR_LT, CR_LT },
2182        [TCG_COND_LEU] = { CR_LT, CR_GT },
2183        [TCG_COND_GTU] = { CR_GT, CR_GT },
2184        [TCG_COND_GEU] = { CR_GT, CR_LT },
2185    };
2186
2187    TCGCond cond = args[4], cond2;
2188    TCGArg al, ah, bl, bh;
2189    int blconst, bhconst;
2190    int op, bit1, bit2;
2191
2192    al = args[0];
2193    ah = args[1];
2194    bl = args[2];
2195    bh = args[3];
2196    blconst = const_args[2];
2197    bhconst = const_args[3];
2198
2199    switch (cond) {
2200    case TCG_COND_EQ:
2201        op = CRAND;
2202        goto do_equality;
2203    case TCG_COND_NE:
2204        op = CRNAND;
2205    do_equality:
2206        tcg_out_cmp(s, cond, al, bl, blconst, 6, TCG_TYPE_I32);
2207        tcg_out_cmp(s, cond, ah, bh, bhconst, 7, TCG_TYPE_I32);
2208        tcg_out32(s, op | BT(0, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
2209        break;
2210
2211    case TCG_COND_TSTEQ:
2212    case TCG_COND_TSTNE:
2213        if (blconst) {
2214            tcg_out_andi32(s, TCG_REG_R0, al, bl);
2215        } else {
2216            tcg_out32(s, AND | SAB(al, TCG_REG_R0, bl));
2217        }
2218        if (bhconst) {
2219            tcg_out_andi32(s, TCG_REG_TMP1, ah, bh);
2220        } else {
2221            tcg_out32(s, AND | SAB(ah, TCG_REG_TMP1, bh));
2222        }
2223        tcg_out32(s, OR | SAB(TCG_REG_R0, TCG_REG_R0, TCG_REG_TMP1) | 1);
2224        break;
2225
2226    case TCG_COND_LT:
2227    case TCG_COND_LE:
2228    case TCG_COND_GT:
2229    case TCG_COND_GE:
2230    case TCG_COND_LTU:
2231    case TCG_COND_LEU:
2232    case TCG_COND_GTU:
2233    case TCG_COND_GEU:
2234        bit1 = bits[cond].bit1;
2235        bit2 = bits[cond].bit2;
2236        op = (bit1 != bit2 ? CRANDC : CRAND);
2237        cond2 = tcg_unsigned_cond(cond);
2238
2239        tcg_out_cmp(s, cond, ah, bh, bhconst, 6, TCG_TYPE_I32);
2240        tcg_out_cmp(s, cond2, al, bl, blconst, 7, TCG_TYPE_I32);
2241        tcg_out32(s, op | BT(0, CR_EQ) | BA(6, CR_EQ) | BB(7, bit2));
2242        tcg_out32(s, CROR | BT(0, CR_EQ) | BA(6, bit1) | BB(0, CR_EQ));
2243        break;
2244
2245    default:
2246        g_assert_not_reached();
2247    }
2248}
2249
2250static void tcg_out_setcond2(TCGContext *s, const TCGArg *args,
2251                             const int *const_args)
2252{
2253    tcg_out_cmp2(s, args + 1, const_args + 1);
2254    tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(0));
2255    tcg_out_rlw(s, RLWINM, args[0], TCG_REG_R0, CR_EQ + 0*4 + 1, 31, 31);
2256}
2257
2258static void tcg_out_brcond2(TCGContext *s, const TCGArg *args,
2259                            const int *const_args)
2260{
2261    tcg_out_cmp2(s, args, const_args);
2262    tcg_out_bc_lab(s, TCG_COND_EQ, arg_label(args[5]));
2263}
2264
2265static void tcg_out_mb(TCGContext *s, TCGArg a0)
2266{
2267    uint32_t insn;
2268
2269    if (a0 & TCG_MO_ST_LD) {
2270        insn = HWSYNC;
2271    } else {
2272        insn = LWSYNC;
2273    }
2274
2275    tcg_out32(s, insn);
2276}
2277
2278static void tcg_out_call_int(TCGContext *s, int lk,
2279                             const tcg_insn_unit *target)
2280{
2281#ifdef _CALL_AIX
2282    /* Look through the descriptor.  If the branch is in range, and we
2283       don't have to spend too much effort on building the toc.  */
2284    const void *tgt = ((const void * const *)target)[0];
2285    uintptr_t toc = ((const uintptr_t *)target)[1];
2286    intptr_t diff = tcg_pcrel_diff(s, tgt);
2287
2288    if (in_range_b(diff) && toc == (uint32_t)toc) {
2289        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, toc);
2290        tcg_out_b(s, lk, tgt);
2291    } else {
2292        /* Fold the low bits of the constant into the addresses below.  */
2293        intptr_t arg = (intptr_t)target;
2294        int ofs = (int16_t)arg;
2295
2296        if (ofs + 8 < 0x8000) {
2297            arg -= ofs;
2298        } else {
2299            ofs = 0;
2300        }
2301        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, arg);
2302        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_TMP1, ofs);
2303        tcg_out32(s, MTSPR | RA(TCG_REG_R0) | CTR);
2304        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_REG_TMP1, ofs + SZP);
2305        tcg_out32(s, BCCTR | BO_ALWAYS | lk);
2306    }
2307#elif defined(_CALL_ELF) && _CALL_ELF == 2
2308    intptr_t diff;
2309
2310    /* In the ELFv2 ABI, we have to set up r12 to contain the destination
2311       address, which the callee uses to compute its TOC address.  */
2312    /* FIXME: when the branch is in range, we could avoid r12 load if we
2313       knew that the destination uses the same TOC, and what its local
2314       entry point offset is.  */
2315    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R12, (intptr_t)target);
2316
2317    diff = tcg_pcrel_diff(s, target);
2318    if (in_range_b(diff)) {
2319        tcg_out_b(s, lk, target);
2320    } else {
2321        tcg_out32(s, MTSPR | RS(TCG_REG_R12) | CTR);
2322        tcg_out32(s, BCCTR | BO_ALWAYS | lk);
2323    }
2324#else
2325    tcg_out_b(s, lk, target);
2326#endif
2327}
2328
2329static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target,
2330                         const TCGHelperInfo *info)
2331{
2332    tcg_out_call_int(s, LK, target);
2333}
2334
2335static const uint32_t qemu_ldx_opc[(MO_SSIZE + MO_BSWAP) + 1] = {
2336    [MO_UB] = LBZX,
2337    [MO_UW] = LHZX,
2338    [MO_UL] = LWZX,
2339    [MO_UQ] = LDX,
2340    [MO_SW] = LHAX,
2341    [MO_SL] = LWAX,
2342    [MO_BSWAP | MO_UB] = LBZX,
2343    [MO_BSWAP | MO_UW] = LHBRX,
2344    [MO_BSWAP | MO_UL] = LWBRX,
2345    [MO_BSWAP | MO_UQ] = LDBRX,
2346};
2347
2348static const uint32_t qemu_stx_opc[(MO_SIZE + MO_BSWAP) + 1] = {
2349    [MO_UB] = STBX,
2350    [MO_UW] = STHX,
2351    [MO_UL] = STWX,
2352    [MO_UQ] = STDX,
2353    [MO_BSWAP | MO_UB] = STBX,
2354    [MO_BSWAP | MO_UW] = STHBRX,
2355    [MO_BSWAP | MO_UL] = STWBRX,
2356    [MO_BSWAP | MO_UQ] = STDBRX,
2357};
2358
2359static TCGReg ldst_ra_gen(TCGContext *s, const TCGLabelQemuLdst *l, int arg)
2360{
2361    if (arg < 0) {
2362        arg = TCG_REG_TMP1;
2363    }
2364    tcg_out32(s, MFSPR | RT(arg) | LR);
2365    return arg;
2366}
2367
2368/*
2369 * For the purposes of ppc32 sorting 4 input registers into 4 argument
2370 * registers, there is an outside chance we would require 3 temps.
2371 */
2372static const TCGLdstHelperParam ldst_helper_param = {
2373    .ra_gen = ldst_ra_gen,
2374    .ntmp = 3,
2375    .tmp = { TCG_REG_TMP1, TCG_REG_TMP2, TCG_REG_R0 }
2376};
2377
2378static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2379{
2380    MemOp opc = get_memop(lb->oi);
2381
2382    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2383        return false;
2384    }
2385
2386    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
2387    tcg_out_call_int(s, LK, qemu_ld_helpers[opc & MO_SIZE]);
2388    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
2389
2390    tcg_out_b(s, 0, lb->raddr);
2391    return true;
2392}
2393
2394static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
2395{
2396    MemOp opc = get_memop(lb->oi);
2397
2398    if (!reloc_pc14(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
2399        return false;
2400    }
2401
2402    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
2403    tcg_out_call_int(s, LK, qemu_st_helpers[opc & MO_SIZE]);
2404
2405    tcg_out_b(s, 0, lb->raddr);
2406    return true;
2407}
2408
2409typedef struct {
2410    TCGReg base;
2411    TCGReg index;
2412    TCGAtomAlign aa;
2413} HostAddress;
2414
2415bool tcg_target_has_memory_bswap(MemOp memop)
2416{
2417    TCGAtomAlign aa;
2418
2419    if ((memop & MO_SIZE) <= MO_64) {
2420        return true;
2421    }
2422
2423    /*
2424     * Reject 16-byte memop with 16-byte atomicity,
2425     * but do allow a pair of 64-bit operations.
2426     */
2427    aa = atom_and_align_for_opc(tcg_ctx, memop, MO_ATOM_IFALIGN, true);
2428    return aa.atom <= MO_64;
2429}
2430
2431/* We expect to use a 16-bit negative offset from ENV.  */
2432#define MIN_TLB_MASK_TABLE_OFS  -32768
2433
2434/*
2435 * For system-mode, perform the TLB load and compare.
2436 * For user-mode, perform any required alignment tests.
2437 * In both cases, return a TCGLabelQemuLdst structure if the slow path
2438 * is required and fill in @h with the host address for the fast path.
2439 */
2440static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
2441                                           TCGReg addr, MemOpIdx oi, bool is_ld)
2442{
2443    TCGType addr_type = s->addr_type;
2444    TCGLabelQemuLdst *ldst = NULL;
2445    MemOp opc = get_memop(oi);
2446    MemOp a_bits, s_bits;
2447
2448    /*
2449     * Book II, Section 1.4, Single-Copy Atomicity, specifies:
2450     *
2451     * Before 3.0, "An access that is not atomic is performed as a set of
2452     * smaller disjoint atomic accesses. In general, the number and alignment
2453     * of these accesses are implementation-dependent."  Thus MO_ATOM_IFALIGN.
2454     *
2455     * As of 3.0, "the non-atomic access is performed as described in
2456     * the corresponding list", which matches MO_ATOM_SUBALIGN.
2457     */
2458    s_bits = opc & MO_SIZE;
2459    h->aa = atom_and_align_for_opc(s, opc,
2460                                   have_isa_3_00 ? MO_ATOM_SUBALIGN
2461                                                 : MO_ATOM_IFALIGN,
2462                                   s_bits == MO_128);
2463    a_bits = h->aa.align;
2464
2465    if (tcg_use_softmmu) {
2466        int mem_index = get_mmuidx(oi);
2467        int cmp_off = is_ld ? offsetof(CPUTLBEntry, addr_read)
2468                            : offsetof(CPUTLBEntry, addr_write);
2469        int fast_off = tlb_mask_table_ofs(s, mem_index);
2470        int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
2471        int table_off = fast_off + offsetof(CPUTLBDescFast, table);
2472
2473        ldst = new_ldst_label(s);
2474        ldst->is_ld = is_ld;
2475        ldst->oi = oi;
2476        ldst->addr_reg = addr;
2477
2478        /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx].  */
2479        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_AREG0, mask_off);
2480        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP2, TCG_AREG0, table_off);
2481
2482        /* Extract the page index, shifted into place for tlb index.  */
2483        if (TCG_TARGET_REG_BITS == 32) {
2484            tcg_out_shri32(s, TCG_REG_R0, addr,
2485                           s->page_bits - CPU_TLB_ENTRY_BITS);
2486        } else {
2487            tcg_out_shri64(s, TCG_REG_R0, addr,
2488                           s->page_bits - CPU_TLB_ENTRY_BITS);
2489        }
2490        tcg_out32(s, AND | SAB(TCG_REG_TMP1, TCG_REG_TMP1, TCG_REG_R0));
2491
2492        /*
2493         * Load the TLB comparator into TMP2.
2494         * For 64-bit host, always load the entire 64-bit slot for simplicity.
2495         * We will ignore the high bits with tcg_out_cmp(..., addr_type).
2496         */
2497        if (cmp_off == 0) {
2498            tcg_out32(s, (TCG_TARGET_REG_BITS == 64 ? LDUX : LWZUX)
2499                      | TAB(TCG_REG_TMP2, TCG_REG_TMP1, TCG_REG_TMP2));
2500        } else {
2501            tcg_out32(s, ADD | TAB(TCG_REG_TMP1, TCG_REG_TMP1, TCG_REG_TMP2));
2502            tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP2, TCG_REG_TMP1, cmp_off);
2503        }
2504
2505        /*
2506         * Load the TLB addend for use on the fast path.
2507         * Do this asap to minimize any load use delay.
2508         */
2509        if (TCG_TARGET_REG_BITS == 64 || addr_type == TCG_TYPE_I32) {
2510            tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1,
2511                       offsetof(CPUTLBEntry, addend));
2512        }
2513
2514        /* Clear the non-page, non-alignment bits from the address in R0. */
2515        if (TCG_TARGET_REG_BITS == 32) {
2516            /*
2517             * We don't support unaligned accesses on 32-bits.
2518             * Preserve the bottom bits and thus trigger a comparison
2519             * failure on unaligned accesses.
2520             */
2521            if (a_bits < s_bits) {
2522                a_bits = s_bits;
2523            }
2524            tcg_out_rlw(s, RLWINM, TCG_REG_R0, addr, 0,
2525                        (32 - a_bits) & 31, 31 - s->page_bits);
2526        } else {
2527            TCGReg t = addr;
2528
2529            /*
2530             * If the access is unaligned, we need to make sure we fail if we
2531             * cross a page boundary.  The trick is to add the access size-1
2532             * to the address before masking the low bits.  That will make the
2533             * address overflow to the next page if we cross a page boundary,
2534             * which will then force a mismatch of the TLB compare.
2535             */
2536            if (a_bits < s_bits) {
2537                unsigned a_mask = (1 << a_bits) - 1;
2538                unsigned s_mask = (1 << s_bits) - 1;
2539                tcg_out32(s, ADDI | TAI(TCG_REG_R0, t, s_mask - a_mask));
2540                t = TCG_REG_R0;
2541            }
2542
2543            /* Mask the address for the requested alignment.  */
2544            if (addr_type == TCG_TYPE_I32) {
2545                tcg_out_rlw(s, RLWINM, TCG_REG_R0, t, 0,
2546                            (32 - a_bits) & 31, 31 - s->page_bits);
2547            } else if (a_bits == 0) {
2548                tcg_out_rld(s, RLDICR, TCG_REG_R0, t, 0, 63 - s->page_bits);
2549            } else {
2550                tcg_out_rld(s, RLDICL, TCG_REG_R0, t,
2551                            64 - s->page_bits, s->page_bits - a_bits);
2552                tcg_out_rld(s, RLDICL, TCG_REG_R0, TCG_REG_R0, s->page_bits, 0);
2553            }
2554        }
2555
2556        /* Full comparison into cr0. */
2557        tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP2, 0, 0, addr_type);
2558
2559        /* Load a pointer into the current opcode w/conditional branch-link. */
2560        ldst->label_ptr[0] = s->code_ptr;
2561        tcg_out_bc(s, TCG_COND_NE, LK);
2562
2563        h->base = TCG_REG_TMP1;
2564    } else {
2565        if (a_bits) {
2566            ldst = new_ldst_label(s);
2567            ldst->is_ld = is_ld;
2568            ldst->oi = oi;
2569            ldst->addr_reg = addr;
2570
2571            /* We are expecting a_bits to max out at 7, much lower than ANDI. */
2572            tcg_debug_assert(a_bits < 16);
2573            tcg_out32(s, ANDI | SAI(addr, TCG_REG_R0, (1 << a_bits) - 1));
2574
2575            ldst->label_ptr[0] = s->code_ptr;
2576            tcg_out32(s, BC | BI(0, CR_EQ) | BO_COND_FALSE | LK);
2577        }
2578
2579        h->base = guest_base ? TCG_GUEST_BASE_REG : 0;
2580    }
2581
2582    if (TCG_TARGET_REG_BITS == 64 && addr_type == TCG_TYPE_I32) {
2583        /* Zero-extend the guest address for use in the host address. */
2584        tcg_out_ext32u(s, TCG_REG_TMP2, addr);
2585        h->index = TCG_REG_TMP2;
2586    } else {
2587        h->index = addr;
2588    }
2589
2590    return ldst;
2591}
2592
2593static void tcg_out_qemu_ld(TCGContext *s, TCGReg datalo, TCGReg datahi,
2594                            TCGReg addr, MemOpIdx oi, TCGType data_type)
2595{
2596    MemOp opc = get_memop(oi);
2597    TCGLabelQemuLdst *ldst;
2598    HostAddress h;
2599
2600    ldst = prepare_host_addr(s, &h, addr, oi, true);
2601
2602    if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
2603        if (opc & MO_BSWAP) {
2604            tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4));
2605            tcg_out32(s, LWBRX | TAB(datalo, h.base, h.index));
2606            tcg_out32(s, LWBRX | TAB(datahi, h.base, TCG_REG_R0));
2607        } else if (h.base != 0) {
2608            tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4));
2609            tcg_out32(s, LWZX | TAB(datahi, h.base, h.index));
2610            tcg_out32(s, LWZX | TAB(datalo, h.base, TCG_REG_R0));
2611        } else if (h.index == datahi) {
2612            tcg_out32(s, LWZ | TAI(datalo, h.index, 4));
2613            tcg_out32(s, LWZ | TAI(datahi, h.index, 0));
2614        } else {
2615            tcg_out32(s, LWZ | TAI(datahi, h.index, 0));
2616            tcg_out32(s, LWZ | TAI(datalo, h.index, 4));
2617        }
2618    } else {
2619        uint32_t insn = qemu_ldx_opc[opc & (MO_BSWAP | MO_SSIZE)];
2620        if (!have_isa_2_06 && insn == LDBRX) {
2621            tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4));
2622            tcg_out32(s, LWBRX | TAB(datalo, h.base, h.index));
2623            tcg_out32(s, LWBRX | TAB(TCG_REG_R0, h.base, TCG_REG_R0));
2624            tcg_out_rld(s, RLDIMI, datalo, TCG_REG_R0, 32, 0);
2625        } else if (insn) {
2626            tcg_out32(s, insn | TAB(datalo, h.base, h.index));
2627        } else {
2628            insn = qemu_ldx_opc[opc & (MO_SIZE | MO_BSWAP)];
2629            tcg_out32(s, insn | TAB(datalo, h.base, h.index));
2630            tcg_out_movext(s, TCG_TYPE_REG, datalo,
2631                           TCG_TYPE_REG, opc & MO_SSIZE, datalo);
2632        }
2633    }
2634
2635    if (ldst) {
2636        ldst->type = data_type;
2637        ldst->datalo_reg = datalo;
2638        ldst->datahi_reg = datahi;
2639        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2640    }
2641}
2642
2643static void tcg_out_qemu_st(TCGContext *s, TCGReg datalo, TCGReg datahi,
2644                            TCGReg addr, MemOpIdx oi, TCGType data_type)
2645{
2646    MemOp opc = get_memop(oi);
2647    TCGLabelQemuLdst *ldst;
2648    HostAddress h;
2649
2650    ldst = prepare_host_addr(s, &h, addr, oi, false);
2651
2652    if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
2653        if (opc & MO_BSWAP) {
2654            tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4));
2655            tcg_out32(s, STWBRX | SAB(datalo, h.base, h.index));
2656            tcg_out32(s, STWBRX | SAB(datahi, h.base, TCG_REG_R0));
2657        } else if (h.base != 0) {
2658            tcg_out32(s, ADDI | TAI(TCG_REG_R0, h.index, 4));
2659            tcg_out32(s, STWX | SAB(datahi, h.base, h.index));
2660            tcg_out32(s, STWX | SAB(datalo, h.base, TCG_REG_R0));
2661        } else {
2662            tcg_out32(s, STW | TAI(datahi, h.index, 0));
2663            tcg_out32(s, STW | TAI(datalo, h.index, 4));
2664        }
2665    } else {
2666        uint32_t insn = qemu_stx_opc[opc & (MO_BSWAP | MO_SIZE)];
2667        if (!have_isa_2_06 && insn == STDBRX) {
2668            tcg_out32(s, STWBRX | SAB(datalo, h.base, h.index));
2669            tcg_out32(s, ADDI | TAI(TCG_REG_TMP2, h.index, 4));
2670            tcg_out_shri64(s, TCG_REG_R0, datalo, 32);
2671            tcg_out32(s, STWBRX | SAB(TCG_REG_R0, h.base, TCG_REG_TMP2));
2672        } else {
2673            tcg_out32(s, insn | SAB(datalo, h.base, h.index));
2674        }
2675    }
2676
2677    if (ldst) {
2678        ldst->type = data_type;
2679        ldst->datalo_reg = datalo;
2680        ldst->datahi_reg = datahi;
2681        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2682    }
2683}
2684
2685static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi,
2686                                   TCGReg addr_reg, MemOpIdx oi, bool is_ld)
2687{
2688    TCGLabelQemuLdst *ldst;
2689    HostAddress h;
2690    bool need_bswap;
2691    uint32_t insn;
2692    TCGReg index;
2693
2694    ldst = prepare_host_addr(s, &h, addr_reg, oi, is_ld);
2695
2696    /* Compose the final address, as LQ/STQ have no indexing. */
2697    index = h.index;
2698    if (h.base != 0) {
2699        index = TCG_REG_TMP1;
2700        tcg_out32(s, ADD | TAB(index, h.base, h.index));
2701    }
2702    need_bswap = get_memop(oi) & MO_BSWAP;
2703
2704    if (h.aa.atom == MO_128) {
2705        tcg_debug_assert(!need_bswap);
2706        tcg_debug_assert(datalo & 1);
2707        tcg_debug_assert(datahi == datalo - 1);
2708        tcg_debug_assert(!is_ld || datahi != index);
2709        insn = is_ld ? LQ : STQ;
2710        tcg_out32(s, insn | TAI(datahi, index, 0));
2711    } else {
2712        TCGReg d1, d2;
2713
2714        if (HOST_BIG_ENDIAN ^ need_bswap) {
2715            d1 = datahi, d2 = datalo;
2716        } else {
2717            d1 = datalo, d2 = datahi;
2718        }
2719
2720        if (need_bswap) {
2721            tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, 8);
2722            insn = is_ld ? LDBRX : STDBRX;
2723            tcg_out32(s, insn | TAB(d1, 0, index));
2724            tcg_out32(s, insn | TAB(d2, index, TCG_REG_R0));
2725        } else {
2726            insn = is_ld ? LD : STD;
2727            tcg_out32(s, insn | TAI(d1, index, 0));
2728            tcg_out32(s, insn | TAI(d2, index, 8));
2729        }
2730    }
2731
2732    if (ldst) {
2733        ldst->type = TCG_TYPE_I128;
2734        ldst->datalo_reg = datalo;
2735        ldst->datahi_reg = datahi;
2736        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2737    }
2738}
2739
2740static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
2741{
2742    int i;
2743    for (i = 0; i < count; ++i) {
2744        p[i] = NOP;
2745    }
2746}
2747
2748/* Parameters for function call generation, used in tcg.c.  */
2749#define TCG_TARGET_STACK_ALIGN       16
2750
2751#ifdef _CALL_AIX
2752# define LINK_AREA_SIZE                (6 * SZR)
2753# define LR_OFFSET                     (1 * SZR)
2754# define TCG_TARGET_CALL_STACK_OFFSET  (LINK_AREA_SIZE + 8 * SZR)
2755#elif defined(_CALL_DARWIN)
2756# define LINK_AREA_SIZE                (6 * SZR)
2757# define LR_OFFSET                     (2 * SZR)
2758#elif TCG_TARGET_REG_BITS == 64
2759# if defined(_CALL_ELF) && _CALL_ELF == 2
2760#  define LINK_AREA_SIZE               (4 * SZR)
2761#  define LR_OFFSET                    (1 * SZR)
2762# endif
2763#else /* TCG_TARGET_REG_BITS == 32 */
2764# if defined(_CALL_SYSV)
2765#  define LINK_AREA_SIZE               (2 * SZR)
2766#  define LR_OFFSET                    (1 * SZR)
2767# endif
2768#endif
2769#ifndef LR_OFFSET
2770# error "Unhandled abi"
2771#endif
2772#ifndef TCG_TARGET_CALL_STACK_OFFSET
2773# define TCG_TARGET_CALL_STACK_OFFSET  LINK_AREA_SIZE
2774#endif
2775
2776#define CPU_TEMP_BUF_SIZE  (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
2777#define REG_SAVE_SIZE      ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
2778
2779#define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET   \
2780                     + TCG_STATIC_CALL_ARGS_SIZE    \
2781                     + CPU_TEMP_BUF_SIZE            \
2782                     + REG_SAVE_SIZE                \
2783                     + TCG_TARGET_STACK_ALIGN - 1)  \
2784                    & -TCG_TARGET_STACK_ALIGN)
2785
2786#define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
2787
2788static void tcg_target_qemu_prologue(TCGContext *s)
2789{
2790    int i;
2791
2792#ifdef _CALL_AIX
2793    const void **desc = (const void **)s->code_ptr;
2794    desc[0] = tcg_splitwx_to_rx(desc + 2);  /* entry point */
2795    desc[1] = 0;                            /* environment pointer */
2796    s->code_ptr = (void *)(desc + 2);       /* skip over descriptor */
2797#endif
2798
2799    tcg_set_frame(s, TCG_REG_CALL_STACK, REG_SAVE_BOT - CPU_TEMP_BUF_SIZE,
2800                  CPU_TEMP_BUF_SIZE);
2801
2802    /* Prologue */
2803    tcg_out32(s, MFSPR | RT(TCG_REG_R0) | LR);
2804    tcg_out32(s, (SZR == 8 ? STDU : STWU)
2805              | SAI(TCG_REG_R1, TCG_REG_R1, -FRAME_SIZE));
2806
2807    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2808        tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2809                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2810    }
2811    tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2812
2813    if (!tcg_use_softmmu && guest_base) {
2814        tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true);
2815        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2816    }
2817
2818    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2819    tcg_out32(s, MTSPR | RS(tcg_target_call_iarg_regs[1]) | CTR);
2820    tcg_out32(s, BCCTR | BO_ALWAYS);
2821
2822    /* Epilogue */
2823    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
2824
2825    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
2826    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
2827        tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2828                   TCG_REG_R1, REG_SAVE_BOT + i * SZR);
2829    }
2830    tcg_out32(s, MTSPR | RS(TCG_REG_R0) | LR);
2831    tcg_out32(s, ADDI | TAI(TCG_REG_R1, TCG_REG_R1, FRAME_SIZE));
2832    tcg_out32(s, BCLR | BO_ALWAYS);
2833}
2834
2835static void tcg_out_tb_start(TCGContext *s)
2836{
2837    /* Load TCG_REG_TB. */
2838    if (USE_REG_TB) {
2839        if (have_isa_3_00) {
2840            /* lnia REG_TB */
2841            tcg_out_addpcis(s, TCG_REG_TB, 0);
2842        } else {
2843            /* bcl 20,31,$+4 (preferred form for getting nia) */
2844            tcg_out32(s, BC | BO_ALWAYS | BI(7, CR_SO) | 0x4 | LK);
2845            tcg_out32(s, MFSPR | RT(TCG_REG_TB) | LR);
2846        }
2847    }
2848}
2849
2850static void tcg_out_exit_tb(TCGContext *s, uintptr_t arg)
2851{
2852    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, arg);
2853    tcg_out_b(s, 0, tcg_code_gen_epilogue);
2854}
2855
2856static void tcg_out_goto_tb(TCGContext *s, int which)
2857{
2858    uintptr_t ptr = get_jmp_target_addr(s, which);
2859    int16_t lo;
2860
2861    /* Direct branch will be patched by tb_target_set_jmp_target. */
2862    set_jmp_insn_offset(s, which);
2863    tcg_out32(s, NOP);
2864
2865    /* When branch is out of range, fall through to indirect. */
2866    if (USE_REG_TB) {
2867        ptrdiff_t offset = ppc_tbrel_diff(s, (void *)ptr);
2868        tcg_out_mem_long(s, LD, LDX, TCG_REG_TMP1, TCG_REG_TB, offset);
2869    } else if (have_isa_3_10) {
2870        ptrdiff_t offset = tcg_pcrel_diff_for_prefix(s, (void *)ptr);
2871        tcg_out_8ls_d(s, PLD, TCG_REG_TMP1, 0, offset, 1);
2872    } else if (have_isa_3_00) {
2873        ptrdiff_t offset = tcg_pcrel_diff(s, (void *)ptr) - 4;
2874        lo = offset;
2875        tcg_out_addpcis(s, TCG_REG_TMP1, offset - lo);
2876        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1, lo);
2877    } else {
2878        lo = ptr;
2879        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, ptr - lo);
2880        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1, lo);
2881    }
2882
2883    tcg_out32(s, MTSPR | RS(TCG_REG_TMP1) | CTR);
2884    tcg_out32(s, BCCTR | BO_ALWAYS);
2885    set_jmp_reset_offset(s, which);
2886}
2887
2888void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
2889                              uintptr_t jmp_rx, uintptr_t jmp_rw)
2890{
2891    uintptr_t addr = tb->jmp_target_addr[n];
2892    intptr_t diff = addr - jmp_rx;
2893    tcg_insn_unit insn;
2894
2895    if (in_range_b(diff)) {
2896        insn = B | (diff & 0x3fffffc);
2897    } else {
2898        insn = NOP;
2899    }
2900
2901    qatomic_set((uint32_t *)jmp_rw, insn);
2902    flush_idcache_range(jmp_rx, jmp_rw, 4);
2903}
2904
2905static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type,
2906                       const TCGArg args[TCG_MAX_OP_ARGS],
2907                       const int const_args[TCG_MAX_OP_ARGS])
2908{
2909    TCGArg a0, a1, a2;
2910
2911    switch (opc) {
2912    case INDEX_op_goto_ptr:
2913        tcg_out32(s, MTSPR | RS(args[0]) | CTR);
2914        tcg_out32(s, ADDI | TAI(TCG_REG_R3, 0, 0));
2915        tcg_out32(s, BCCTR | BO_ALWAYS);
2916        break;
2917    case INDEX_op_br:
2918        {
2919            TCGLabel *l = arg_label(args[0]);
2920            uint32_t insn = B;
2921
2922            if (l->has_value) {
2923                insn |= reloc_pc24_val(tcg_splitwx_to_rx(s->code_ptr),
2924                                       l->u.value_ptr);
2925            } else {
2926                tcg_out_reloc(s, s->code_ptr, R_PPC_REL24, l, 0);
2927            }
2928            tcg_out32(s, insn);
2929        }
2930        break;
2931    case INDEX_op_ld8u_i32:
2932    case INDEX_op_ld8u_i64:
2933        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2934        break;
2935    case INDEX_op_ld8s_i32:
2936    case INDEX_op_ld8s_i64:
2937        tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2938        tcg_out_ext8s(s, TCG_TYPE_REG, args[0], args[0]);
2939        break;
2940    case INDEX_op_ld16u_i32:
2941    case INDEX_op_ld16u_i64:
2942        tcg_out_mem_long(s, LHZ, LHZX, args[0], args[1], args[2]);
2943        break;
2944    case INDEX_op_ld16s_i32:
2945    case INDEX_op_ld16s_i64:
2946        tcg_out_mem_long(s, LHA, LHAX, args[0], args[1], args[2]);
2947        break;
2948    case INDEX_op_ld_i32:
2949    case INDEX_op_ld32u_i64:
2950        tcg_out_mem_long(s, LWZ, LWZX, args[0], args[1], args[2]);
2951        break;
2952    case INDEX_op_ld32s_i64:
2953        tcg_out_mem_long(s, LWA, LWAX, args[0], args[1], args[2]);
2954        break;
2955    case INDEX_op_ld_i64:
2956        tcg_out_mem_long(s, LD, LDX, args[0], args[1], args[2]);
2957        break;
2958    case INDEX_op_st8_i32:
2959    case INDEX_op_st8_i64:
2960        tcg_out_mem_long(s, STB, STBX, args[0], args[1], args[2]);
2961        break;
2962    case INDEX_op_st16_i32:
2963    case INDEX_op_st16_i64:
2964        tcg_out_mem_long(s, STH, STHX, args[0], args[1], args[2]);
2965        break;
2966    case INDEX_op_st_i32:
2967    case INDEX_op_st32_i64:
2968        tcg_out_mem_long(s, STW, STWX, args[0], args[1], args[2]);
2969        break;
2970    case INDEX_op_st_i64:
2971        tcg_out_mem_long(s, STD, STDX, args[0], args[1], args[2]);
2972        break;
2973
2974    case INDEX_op_add_i32:
2975        a0 = args[0], a1 = args[1], a2 = args[2];
2976        if (const_args[2]) {
2977        do_addi_32:
2978            tcg_out_mem_long(s, ADDI, ADD, a0, a1, (int32_t)a2);
2979        } else {
2980            tcg_out32(s, ADD | TAB(a0, a1, a2));
2981        }
2982        break;
2983    case INDEX_op_sub_i32:
2984        a0 = args[0], a1 = args[1], a2 = args[2];
2985        if (const_args[1]) {
2986            if (const_args[2]) {
2987                tcg_out_movi(s, TCG_TYPE_I32, a0, a1 - a2);
2988            } else {
2989                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2990            }
2991        } else if (const_args[2]) {
2992            a2 = -a2;
2993            goto do_addi_32;
2994        } else {
2995            tcg_out32(s, SUBF | TAB(a0, a2, a1));
2996        }
2997        break;
2998
2999    case INDEX_op_and_i32:
3000        a0 = args[0], a1 = args[1], a2 = args[2];
3001        if (const_args[2]) {
3002            tcg_out_andi32(s, a0, a1, a2);
3003        } else {
3004            tcg_out32(s, AND | SAB(a1, a0, a2));
3005        }
3006        break;
3007    case INDEX_op_and_i64:
3008        a0 = args[0], a1 = args[1], a2 = args[2];
3009        if (const_args[2]) {
3010            tcg_out_andi64(s, a0, a1, a2);
3011        } else {
3012            tcg_out32(s, AND | SAB(a1, a0, a2));
3013        }
3014        break;
3015    case INDEX_op_or_i64:
3016    case INDEX_op_or_i32:
3017        a0 = args[0], a1 = args[1], a2 = args[2];
3018        if (const_args[2]) {
3019            tcg_out_ori32(s, a0, a1, a2);
3020        } else {
3021            tcg_out32(s, OR | SAB(a1, a0, a2));
3022        }
3023        break;
3024    case INDEX_op_xor_i64:
3025    case INDEX_op_xor_i32:
3026        a0 = args[0], a1 = args[1], a2 = args[2];
3027        if (const_args[2]) {
3028            tcg_out_xori32(s, a0, a1, a2);
3029        } else {
3030            tcg_out32(s, XOR | SAB(a1, a0, a2));
3031        }
3032        break;
3033    case INDEX_op_andc_i32:
3034        a0 = args[0], a1 = args[1], a2 = args[2];
3035        if (const_args[2]) {
3036            tcg_out_andi32(s, a0, a1, ~a2);
3037        } else {
3038            tcg_out32(s, ANDC | SAB(a1, a0, a2));
3039        }
3040        break;
3041    case INDEX_op_andc_i64:
3042        a0 = args[0], a1 = args[1], a2 = args[2];
3043        if (const_args[2]) {
3044            tcg_out_andi64(s, a0, a1, ~a2);
3045        } else {
3046            tcg_out32(s, ANDC | SAB(a1, a0, a2));
3047        }
3048        break;
3049    case INDEX_op_orc_i32:
3050        if (const_args[2]) {
3051            tcg_out_ori32(s, args[0], args[1], ~args[2]);
3052            break;
3053        }
3054        /* FALLTHRU */
3055    case INDEX_op_orc_i64:
3056        tcg_out32(s, ORC | SAB(args[1], args[0], args[2]));
3057        break;
3058    case INDEX_op_eqv_i32:
3059        if (const_args[2]) {
3060            tcg_out_xori32(s, args[0], args[1], ~args[2]);
3061            break;
3062        }
3063        /* FALLTHRU */
3064    case INDEX_op_eqv_i64:
3065        tcg_out32(s, EQV | SAB(args[1], args[0], args[2]));
3066        break;
3067    case INDEX_op_nand_i32:
3068    case INDEX_op_nand_i64:
3069        tcg_out32(s, NAND | SAB(args[1], args[0], args[2]));
3070        break;
3071    case INDEX_op_nor_i32:
3072    case INDEX_op_nor_i64:
3073        tcg_out32(s, NOR | SAB(args[1], args[0], args[2]));
3074        break;
3075
3076    case INDEX_op_clz_i32:
3077        tcg_out_cntxz(s, TCG_TYPE_I32, CNTLZW, args[0], args[1],
3078                      args[2], const_args[2]);
3079        break;
3080    case INDEX_op_ctz_i32:
3081        tcg_out_cntxz(s, TCG_TYPE_I32, CNTTZW, args[0], args[1],
3082                      args[2], const_args[2]);
3083        break;
3084    case INDEX_op_ctpop_i32:
3085        tcg_out32(s, CNTPOPW | SAB(args[1], args[0], 0));
3086        break;
3087
3088    case INDEX_op_clz_i64:
3089        tcg_out_cntxz(s, TCG_TYPE_I64, CNTLZD, args[0], args[1],
3090                      args[2], const_args[2]);
3091        break;
3092    case INDEX_op_ctz_i64:
3093        tcg_out_cntxz(s, TCG_TYPE_I64, CNTTZD, args[0], args[1],
3094                      args[2], const_args[2]);
3095        break;
3096    case INDEX_op_ctpop_i64:
3097        tcg_out32(s, CNTPOPD | SAB(args[1], args[0], 0));
3098        break;
3099
3100    case INDEX_op_mul_i32:
3101        a0 = args[0], a1 = args[1], a2 = args[2];
3102        if (const_args[2]) {
3103            tcg_out32(s, MULLI | TAI(a0, a1, a2));
3104        } else {
3105            tcg_out32(s, MULLW | TAB(a0, a1, a2));
3106        }
3107        break;
3108
3109    case INDEX_op_div_i32:
3110        tcg_out32(s, DIVW | TAB(args[0], args[1], args[2]));
3111        break;
3112
3113    case INDEX_op_divu_i32:
3114        tcg_out32(s, DIVWU | TAB(args[0], args[1], args[2]));
3115        break;
3116
3117    case INDEX_op_rem_i32:
3118        tcg_out32(s, MODSW | TAB(args[0], args[1], args[2]));
3119        break;
3120
3121    case INDEX_op_remu_i32:
3122        tcg_out32(s, MODUW | TAB(args[0], args[1], args[2]));
3123        break;
3124
3125    case INDEX_op_shl_i32:
3126        if (const_args[2]) {
3127            /* Limit immediate shift count lest we create an illegal insn.  */
3128            tcg_out_shli32(s, args[0], args[1], args[2] & 31);
3129        } else {
3130            tcg_out32(s, SLW | SAB(args[1], args[0], args[2]));
3131        }
3132        break;
3133    case INDEX_op_shr_i32:
3134        if (const_args[2]) {
3135            /* Limit immediate shift count lest we create an illegal insn.  */
3136            tcg_out_shri32(s, args[0], args[1], args[2] & 31);
3137        } else {
3138            tcg_out32(s, SRW | SAB(args[1], args[0], args[2]));
3139        }
3140        break;
3141    case INDEX_op_sar_i32:
3142        if (const_args[2]) {
3143            tcg_out_sari32(s, args[0], args[1], args[2]);
3144        } else {
3145            tcg_out32(s, SRAW | SAB(args[1], args[0], args[2]));
3146        }
3147        break;
3148    case INDEX_op_rotl_i32:
3149        if (const_args[2]) {
3150            tcg_out_rlw(s, RLWINM, args[0], args[1], args[2], 0, 31);
3151        } else {
3152            tcg_out32(s, RLWNM | SAB(args[1], args[0], args[2])
3153                         | MB(0) | ME(31));
3154        }
3155        break;
3156    case INDEX_op_rotr_i32:
3157        if (const_args[2]) {
3158            tcg_out_rlw(s, RLWINM, args[0], args[1], 32 - args[2], 0, 31);
3159        } else {
3160            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 32));
3161            tcg_out32(s, RLWNM | SAB(args[1], args[0], TCG_REG_R0)
3162                         | MB(0) | ME(31));
3163        }
3164        break;
3165
3166    case INDEX_op_brcond_i32:
3167        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
3168                       arg_label(args[3]), TCG_TYPE_I32);
3169        break;
3170    case INDEX_op_brcond_i64:
3171        tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
3172                       arg_label(args[3]), TCG_TYPE_I64);
3173        break;
3174    case INDEX_op_brcond2_i32:
3175        tcg_out_brcond2(s, args, const_args);
3176        break;
3177
3178    case INDEX_op_neg_i32:
3179    case INDEX_op_neg_i64:
3180        tcg_out32(s, NEG | RT(args[0]) | RA(args[1]));
3181        break;
3182
3183    case INDEX_op_not_i32:
3184    case INDEX_op_not_i64:
3185        tcg_out32(s, NOR | SAB(args[1], args[0], args[1]));
3186        break;
3187
3188    case INDEX_op_add_i64:
3189        a0 = args[0], a1 = args[1], a2 = args[2];
3190        if (const_args[2]) {
3191        do_addi_64:
3192            tcg_out_mem_long(s, ADDI, ADD, a0, a1, a2);
3193        } else {
3194            tcg_out32(s, ADD | TAB(a0, a1, a2));
3195        }
3196        break;
3197    case INDEX_op_sub_i64:
3198        a0 = args[0], a1 = args[1], a2 = args[2];
3199        if (const_args[1]) {
3200            if (const_args[2]) {
3201                tcg_out_movi(s, TCG_TYPE_I64, a0, a1 - a2);
3202            } else {
3203                tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
3204            }
3205        } else if (const_args[2]) {
3206            a2 = -a2;
3207            goto do_addi_64;
3208        } else {
3209            tcg_out32(s, SUBF | TAB(a0, a2, a1));
3210        }
3211        break;
3212
3213    case INDEX_op_shl_i64:
3214        if (const_args[2]) {
3215            /* Limit immediate shift count lest we create an illegal insn.  */
3216            tcg_out_shli64(s, args[0], args[1], args[2] & 63);
3217        } else {
3218            tcg_out32(s, SLD | SAB(args[1], args[0], args[2]));
3219        }
3220        break;
3221    case INDEX_op_shr_i64:
3222        if (const_args[2]) {
3223            /* Limit immediate shift count lest we create an illegal insn.  */
3224            tcg_out_shri64(s, args[0], args[1], args[2] & 63);
3225        } else {
3226            tcg_out32(s, SRD | SAB(args[1], args[0], args[2]));
3227        }
3228        break;
3229    case INDEX_op_sar_i64:
3230        if (const_args[2]) {
3231            tcg_out_sari64(s, args[0], args[1], args[2]);
3232        } else {
3233            tcg_out32(s, SRAD | SAB(args[1], args[0], args[2]));
3234        }
3235        break;
3236    case INDEX_op_rotl_i64:
3237        if (const_args[2]) {
3238            tcg_out_rld(s, RLDICL, args[0], args[1], args[2], 0);
3239        } else {
3240            tcg_out32(s, RLDCL | SAB(args[1], args[0], args[2]) | MB64(0));
3241        }
3242        break;
3243    case INDEX_op_rotr_i64:
3244        if (const_args[2]) {
3245            tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 0);
3246        } else {
3247            tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 64));
3248            tcg_out32(s, RLDCL | SAB(args[1], args[0], TCG_REG_R0) | MB64(0));
3249        }
3250        break;
3251
3252    case INDEX_op_mul_i64:
3253        a0 = args[0], a1 = args[1], a2 = args[2];
3254        if (const_args[2]) {
3255            tcg_out32(s, MULLI | TAI(a0, a1, a2));
3256        } else {
3257            tcg_out32(s, MULLD | TAB(a0, a1, a2));
3258        }
3259        break;
3260    case INDEX_op_div_i64:
3261        tcg_out32(s, DIVD | TAB(args[0], args[1], args[2]));
3262        break;
3263    case INDEX_op_divu_i64:
3264        tcg_out32(s, DIVDU | TAB(args[0], args[1], args[2]));
3265        break;
3266    case INDEX_op_rem_i64:
3267        tcg_out32(s, MODSD | TAB(args[0], args[1], args[2]));
3268        break;
3269    case INDEX_op_remu_i64:
3270        tcg_out32(s, MODUD | TAB(args[0], args[1], args[2]));
3271        break;
3272
3273    case INDEX_op_qemu_ld_i32:
3274        tcg_out_qemu_ld(s, args[0], -1, args[1], args[2], TCG_TYPE_I32);
3275        break;
3276    case INDEX_op_qemu_ld_i64:
3277        if (TCG_TARGET_REG_BITS == 64) {
3278            tcg_out_qemu_ld(s, args[0], -1, args[1], args[2], TCG_TYPE_I64);
3279        } else {
3280            tcg_out_qemu_ld(s, args[0], args[1], args[2],
3281                            args[3], TCG_TYPE_I64);
3282        }
3283        break;
3284    case INDEX_op_qemu_ld_i128:
3285        tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
3286        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], true);
3287        break;
3288
3289    case INDEX_op_qemu_st_i32:
3290        tcg_out_qemu_st(s, args[0], -1, args[1], args[2], TCG_TYPE_I32);
3291        break;
3292    case INDEX_op_qemu_st_i64:
3293        if (TCG_TARGET_REG_BITS == 64) {
3294            tcg_out_qemu_st(s, args[0], -1, args[1], args[2], TCG_TYPE_I64);
3295        } else {
3296            tcg_out_qemu_st(s, args[0], args[1], args[2],
3297                            args[3], TCG_TYPE_I64);
3298        }
3299        break;
3300    case INDEX_op_qemu_st_i128:
3301        tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
3302        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], false);
3303        break;
3304
3305    case INDEX_op_setcond_i32:
3306        tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2],
3307                        const_args[2], false);
3308        break;
3309    case INDEX_op_setcond_i64:
3310        tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2],
3311                        const_args[2], false);
3312        break;
3313    case INDEX_op_negsetcond_i32:
3314        tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2],
3315                        const_args[2], true);
3316        break;
3317    case INDEX_op_negsetcond_i64:
3318        tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2],
3319                        const_args[2], true);
3320        break;
3321    case INDEX_op_setcond2_i32:
3322        tcg_out_setcond2(s, args, const_args);
3323        break;
3324
3325    case INDEX_op_bswap16_i32:
3326    case INDEX_op_bswap16_i64:
3327        tcg_out_bswap16(s, args[0], args[1], args[2]);
3328        break;
3329    case INDEX_op_bswap32_i32:
3330        tcg_out_bswap32(s, args[0], args[1], 0);
3331        break;
3332    case INDEX_op_bswap32_i64:
3333        tcg_out_bswap32(s, args[0], args[1], args[2]);
3334        break;
3335    case INDEX_op_bswap64_i64:
3336        tcg_out_bswap64(s, args[0], args[1]);
3337        break;
3338
3339    case INDEX_op_deposit_i32:
3340        if (const_args[2]) {
3341            uint32_t mask = ((2u << (args[4] - 1)) - 1) << args[3];
3342            tcg_out_andi32(s, args[0], args[0], ~mask);
3343        } else {
3344            tcg_out_rlw(s, RLWIMI, args[0], args[2], args[3],
3345                        32 - args[3] - args[4], 31 - args[3]);
3346        }
3347        break;
3348    case INDEX_op_deposit_i64:
3349        if (const_args[2]) {
3350            uint64_t mask = ((2ull << (args[4] - 1)) - 1) << args[3];
3351            tcg_out_andi64(s, args[0], args[0], ~mask);
3352        } else {
3353            tcg_out_rld(s, RLDIMI, args[0], args[2], args[3],
3354                        64 - args[3] - args[4]);
3355        }
3356        break;
3357
3358    case INDEX_op_extract_i32:
3359        if (args[2] == 0 && args[3] <= 16) {
3360            tcg_out32(s, ANDI | SAI(args[1], args[0], (1 << args[3]) - 1));
3361            break;
3362        }
3363        tcg_out_rlw(s, RLWINM, args[0], args[1],
3364                    32 - args[2], 32 - args[3], 31);
3365        break;
3366    case INDEX_op_extract_i64:
3367        if (args[2] == 0 && args[3] <= 16) {
3368            tcg_out32(s, ANDI | SAI(args[1], args[0], (1 << args[3]) - 1));
3369            break;
3370        }
3371        tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 64 - args[3]);
3372        break;
3373
3374    case INDEX_op_sextract_i64:
3375        if (args[2] + args[3] == 32) {
3376            if (args[2] == 0) {
3377                tcg_out_ext32s(s, args[0], args[1]);
3378            } else {
3379                tcg_out_sari32(s, args[0], args[1], args[2]);
3380            }
3381            break;
3382        }
3383        /* FALLTHRU */
3384    case INDEX_op_sextract_i32:
3385        if (args[2] == 0 && args[3] == 8) {
3386            tcg_out_ext8s(s, TCG_TYPE_I32, args[0], args[1]);
3387        } else if (args[2] == 0 && args[3] == 16) {
3388            tcg_out_ext16s(s, TCG_TYPE_I32, args[0], args[1]);
3389        } else {
3390            g_assert_not_reached();
3391        }
3392        break;
3393
3394    case INDEX_op_movcond_i32:
3395        tcg_out_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], args[2],
3396                        args[3], args[4], const_args[2]);
3397        break;
3398    case INDEX_op_movcond_i64:
3399        tcg_out_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], args[2],
3400                        args[3], args[4], const_args[2]);
3401        break;
3402
3403#if TCG_TARGET_REG_BITS == 64
3404    case INDEX_op_add2_i64:
3405#else
3406    case INDEX_op_add2_i32:
3407#endif
3408        /* Note that the CA bit is defined based on the word size of the
3409           environment.  So in 64-bit mode it's always carry-out of bit 63.
3410           The fallback code using deposit works just as well for 32-bit.  */
3411        a0 = args[0], a1 = args[1];
3412        if (a0 == args[3] || (!const_args[5] && a0 == args[5])) {
3413            a0 = TCG_REG_R0;
3414        }
3415        if (const_args[4]) {
3416            tcg_out32(s, ADDIC | TAI(a0, args[2], args[4]));
3417        } else {
3418            tcg_out32(s, ADDC | TAB(a0, args[2], args[4]));
3419        }
3420        if (const_args[5]) {
3421            tcg_out32(s, (args[5] ? ADDME : ADDZE) | RT(a1) | RA(args[3]));
3422        } else {
3423            tcg_out32(s, ADDE | TAB(a1, args[3], args[5]));
3424        }
3425        if (a0 != args[0]) {
3426            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
3427        }
3428        break;
3429
3430#if TCG_TARGET_REG_BITS == 64
3431    case INDEX_op_sub2_i64:
3432#else
3433    case INDEX_op_sub2_i32:
3434#endif
3435        a0 = args[0], a1 = args[1];
3436        if (a0 == args[5] || (!const_args[3] && a0 == args[3])) {
3437            a0 = TCG_REG_R0;
3438        }
3439        if (const_args[2]) {
3440            tcg_out32(s, SUBFIC | TAI(a0, args[4], args[2]));
3441        } else {
3442            tcg_out32(s, SUBFC | TAB(a0, args[4], args[2]));
3443        }
3444        if (const_args[3]) {
3445            tcg_out32(s, (args[3] ? SUBFME : SUBFZE) | RT(a1) | RA(args[5]));
3446        } else {
3447            tcg_out32(s, SUBFE | TAB(a1, args[5], args[3]));
3448        }
3449        if (a0 != args[0]) {
3450            tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
3451        }
3452        break;
3453
3454    case INDEX_op_muluh_i32:
3455        tcg_out32(s, MULHWU | TAB(args[0], args[1], args[2]));
3456        break;
3457    case INDEX_op_mulsh_i32:
3458        tcg_out32(s, MULHW | TAB(args[0], args[1], args[2]));
3459        break;
3460    case INDEX_op_muluh_i64:
3461        tcg_out32(s, MULHDU | TAB(args[0], args[1], args[2]));
3462        break;
3463    case INDEX_op_mulsh_i64:
3464        tcg_out32(s, MULHD | TAB(args[0], args[1], args[2]));
3465        break;
3466
3467    case INDEX_op_mb:
3468        tcg_out_mb(s, args[0]);
3469        break;
3470
3471    case INDEX_op_mov_i32:   /* Always emitted via tcg_out_mov.  */
3472    case INDEX_op_mov_i64:
3473    case INDEX_op_call:      /* Always emitted via tcg_out_call.  */
3474    case INDEX_op_exit_tb:   /* Always emitted via tcg_out_exit_tb.  */
3475    case INDEX_op_goto_tb:   /* Always emitted via tcg_out_goto_tb.  */
3476    case INDEX_op_ext8s_i32:  /* Always emitted via tcg_reg_alloc_op.  */
3477    case INDEX_op_ext8s_i64:
3478    case INDEX_op_ext8u_i32:
3479    case INDEX_op_ext8u_i64:
3480    case INDEX_op_ext16s_i32:
3481    case INDEX_op_ext16s_i64:
3482    case INDEX_op_ext16u_i32:
3483    case INDEX_op_ext16u_i64:
3484    case INDEX_op_ext32s_i64:
3485    case INDEX_op_ext32u_i64:
3486    case INDEX_op_ext_i32_i64:
3487    case INDEX_op_extu_i32_i64:
3488    case INDEX_op_extrl_i64_i32:
3489    default:
3490        g_assert_not_reached();
3491    }
3492}
3493
3494int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3495{
3496    switch (opc) {
3497    case INDEX_op_and_vec:
3498    case INDEX_op_or_vec:
3499    case INDEX_op_xor_vec:
3500    case INDEX_op_andc_vec:
3501    case INDEX_op_not_vec:
3502    case INDEX_op_nor_vec:
3503    case INDEX_op_eqv_vec:
3504    case INDEX_op_nand_vec:
3505        return 1;
3506    case INDEX_op_orc_vec:
3507        return have_isa_2_07;
3508    case INDEX_op_add_vec:
3509    case INDEX_op_sub_vec:
3510    case INDEX_op_smax_vec:
3511    case INDEX_op_smin_vec:
3512    case INDEX_op_umax_vec:
3513    case INDEX_op_umin_vec:
3514    case INDEX_op_shlv_vec:
3515    case INDEX_op_shrv_vec:
3516    case INDEX_op_sarv_vec:
3517    case INDEX_op_rotlv_vec:
3518        return vece <= MO_32 || have_isa_2_07;
3519    case INDEX_op_ssadd_vec:
3520    case INDEX_op_sssub_vec:
3521    case INDEX_op_usadd_vec:
3522    case INDEX_op_ussub_vec:
3523        return vece <= MO_32;
3524    case INDEX_op_shli_vec:
3525    case INDEX_op_shri_vec:
3526    case INDEX_op_sari_vec:
3527    case INDEX_op_rotli_vec:
3528        return vece <= MO_32 || have_isa_2_07 ? -1 : 0;
3529    case INDEX_op_cmp_vec:
3530    case INDEX_op_cmpsel_vec:
3531        return vece <= MO_32 || have_isa_2_07 ? 1 : 0;
3532    case INDEX_op_neg_vec:
3533        return vece >= MO_32 && have_isa_3_00;
3534    case INDEX_op_mul_vec:
3535        switch (vece) {
3536        case MO_8:
3537        case MO_16:
3538            return -1;
3539        case MO_32:
3540            return have_isa_2_07 ? 1 : -1;
3541        case MO_64:
3542            return have_isa_3_10;
3543        }
3544        return 0;
3545    case INDEX_op_bitsel_vec:
3546        return have_vsx;
3547    case INDEX_op_rotrv_vec:
3548        return -1;
3549    default:
3550        return 0;
3551    }
3552}
3553
3554static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
3555                            TCGReg dst, TCGReg src)
3556{
3557    tcg_debug_assert(dst >= TCG_REG_V0);
3558
3559    /* Splat from integer reg allowed via constraints for v3.00.  */
3560    if (src < TCG_REG_V0) {
3561        tcg_debug_assert(have_isa_3_00);
3562        switch (vece) {
3563        case MO_64:
3564            tcg_out32(s, MTVSRDD | VRT(dst) | RA(src) | RB(src));
3565            return true;
3566        case MO_32:
3567            tcg_out32(s, MTVSRWS | VRT(dst) | RA(src));
3568            return true;
3569        default:
3570            /* Fail, so that we fall back on either dupm or mov+dup.  */
3571            return false;
3572        }
3573    }
3574
3575    /*
3576     * Recall we use (or emulate) VSX integer loads, so the integer is
3577     * right justified within the left (zero-index) double-word.
3578     */
3579    switch (vece) {
3580    case MO_8:
3581        tcg_out32(s, VSPLTB | VRT(dst) | VRB(src) | (7 << 16));
3582        break;
3583    case MO_16:
3584        tcg_out32(s, VSPLTH | VRT(dst) | VRB(src) | (3 << 16));
3585        break;
3586    case MO_32:
3587        tcg_out32(s, VSPLTW | VRT(dst) | VRB(src) | (1 << 16));
3588        break;
3589    case MO_64:
3590        if (have_vsx) {
3591            tcg_out32(s, XXPERMDI | VRT(dst) | VRA(src) | VRB(src));
3592            break;
3593        }
3594        tcg_out_vsldoi(s, TCG_VEC_TMP1, src, src, 8);
3595        tcg_out_vsldoi(s, dst, TCG_VEC_TMP1, src, 8);
3596        break;
3597    default:
3598        g_assert_not_reached();
3599    }
3600    return true;
3601}
3602
3603static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
3604                             TCGReg out, TCGReg base, intptr_t offset)
3605{
3606    int elt;
3607
3608    tcg_debug_assert(out >= TCG_REG_V0);
3609    switch (vece) {
3610    case MO_8:
3611        if (have_isa_3_00) {
3612            tcg_out_mem_long(s, LXV, LVX, out, base, offset & -16);
3613        } else {
3614            tcg_out_mem_long(s, 0, LVEBX, out, base, offset);
3615        }
3616        elt = extract32(offset, 0, 4);
3617#if !HOST_BIG_ENDIAN
3618        elt ^= 15;
3619#endif
3620        tcg_out32(s, VSPLTB | VRT(out) | VRB(out) | (elt << 16));
3621        break;
3622    case MO_16:
3623        tcg_debug_assert((offset & 1) == 0);
3624        if (have_isa_3_00) {
3625            tcg_out_mem_long(s, LXV | 8, LVX, out, base, offset & -16);
3626        } else {
3627            tcg_out_mem_long(s, 0, LVEHX, out, base, offset);
3628        }
3629        elt = extract32(offset, 1, 3);
3630#if !HOST_BIG_ENDIAN
3631        elt ^= 7;
3632#endif
3633        tcg_out32(s, VSPLTH | VRT(out) | VRB(out) | (elt << 16));
3634        break;
3635    case MO_32:
3636        if (have_isa_3_00) {
3637            tcg_out_mem_long(s, 0, LXVWSX, out, base, offset);
3638            break;
3639        }
3640        tcg_debug_assert((offset & 3) == 0);
3641        tcg_out_mem_long(s, 0, LVEWX, out, base, offset);
3642        elt = extract32(offset, 2, 2);
3643#if !HOST_BIG_ENDIAN
3644        elt ^= 3;
3645#endif
3646        tcg_out32(s, VSPLTW | VRT(out) | VRB(out) | (elt << 16));
3647        break;
3648    case MO_64:
3649        if (have_vsx) {
3650            tcg_out_mem_long(s, 0, LXVDSX, out, base, offset);
3651            break;
3652        }
3653        tcg_debug_assert((offset & 7) == 0);
3654        tcg_out_mem_long(s, 0, LVX, out, base, offset & -16);
3655        tcg_out_vsldoi(s, TCG_VEC_TMP1, out, out, 8);
3656        elt = extract32(offset, 3, 1);
3657#if !HOST_BIG_ENDIAN
3658        elt = !elt;
3659#endif
3660        if (elt) {
3661            tcg_out_vsldoi(s, out, out, TCG_VEC_TMP1, 8);
3662        } else {
3663            tcg_out_vsldoi(s, out, TCG_VEC_TMP1, out, 8);
3664        }
3665        break;
3666    default:
3667        g_assert_not_reached();
3668    }
3669    return true;
3670}
3671
3672static void tcg_out_not_vec(TCGContext *s, TCGReg a0, TCGReg a1)
3673{
3674    tcg_out32(s, VNOR | VRT(a0) | VRA(a1) | VRB(a1));
3675}
3676
3677static void tcg_out_or_vec(TCGContext *s, TCGReg a0, TCGReg a1, TCGReg a2)
3678{
3679    tcg_out32(s, VOR | VRT(a0) | VRA(a1) | VRB(a2));
3680}
3681
3682static void tcg_out_orc_vec(TCGContext *s, TCGReg a0, TCGReg a1, TCGReg a2)
3683{
3684    tcg_out32(s, VORC | VRT(a0) | VRA(a1) | VRB(a2));
3685}
3686
3687static void tcg_out_and_vec(TCGContext *s, TCGReg a0, TCGReg a1, TCGReg a2)
3688{
3689    tcg_out32(s, VAND | VRT(a0) | VRA(a1) | VRB(a2));
3690}
3691
3692static void tcg_out_andc_vec(TCGContext *s, TCGReg a0, TCGReg a1, TCGReg a2)
3693{
3694    tcg_out32(s, VANDC | VRT(a0) | VRA(a1) | VRB(a2));
3695}
3696
3697static void tcg_out_bitsel_vec(TCGContext *s, TCGReg d,
3698                               TCGReg c, TCGReg t, TCGReg f)
3699{
3700    if (TCG_TARGET_HAS_bitsel_vec) {
3701        tcg_out32(s, XXSEL | VRT(d) | VRC(c) | VRB(t) | VRA(f));
3702    } else {
3703        tcg_out_and_vec(s, TCG_VEC_TMP2, t, c);
3704        tcg_out_andc_vec(s, d, f, c);
3705        tcg_out_or_vec(s, d, d, TCG_VEC_TMP2);
3706    }
3707}
3708
3709static bool tcg_out_cmp_vec_noinv(TCGContext *s, unsigned vece, TCGReg a0,
3710                                  TCGReg a1, TCGReg a2, TCGCond cond)
3711{
3712    static const uint32_t
3713        eq_op[4]  = { VCMPEQUB, VCMPEQUH, VCMPEQUW, VCMPEQUD },
3714        ne_op[4]  = { VCMPNEB, VCMPNEH, VCMPNEW, 0 },
3715        gts_op[4] = { VCMPGTSB, VCMPGTSH, VCMPGTSW, VCMPGTSD },
3716        gtu_op[4] = { VCMPGTUB, VCMPGTUH, VCMPGTUW, VCMPGTUD };
3717    uint32_t insn;
3718
3719    bool need_swap = false, need_inv = false;
3720
3721    tcg_debug_assert(vece <= MO_32 || have_isa_2_07);
3722
3723    switch (cond) {
3724    case TCG_COND_EQ:
3725    case TCG_COND_GT:
3726    case TCG_COND_GTU:
3727        break;
3728    case TCG_COND_NE:
3729        if (have_isa_3_00 && vece <= MO_32) {
3730            break;
3731        }
3732        /* fall through */
3733    case TCG_COND_LE:
3734    case TCG_COND_LEU:
3735        need_inv = true;
3736        break;
3737    case TCG_COND_LT:
3738    case TCG_COND_LTU:
3739        need_swap = true;
3740        break;
3741    case TCG_COND_GE:
3742    case TCG_COND_GEU:
3743        need_swap = need_inv = true;
3744        break;
3745    default:
3746        g_assert_not_reached();
3747    }
3748
3749    if (need_inv) {
3750        cond = tcg_invert_cond(cond);
3751    }
3752    if (need_swap) {
3753        TCGReg swap = a1;
3754        a1 = a2;
3755        a2 = swap;
3756        cond = tcg_swap_cond(cond);
3757    }
3758
3759    switch (cond) {
3760    case TCG_COND_EQ:
3761        insn = eq_op[vece];
3762        break;
3763    case TCG_COND_NE:
3764        insn = ne_op[vece];
3765        break;
3766    case TCG_COND_GT:
3767        insn = gts_op[vece];
3768        break;
3769    case TCG_COND_GTU:
3770        insn = gtu_op[vece];
3771        break;
3772    default:
3773        g_assert_not_reached();
3774    }
3775    tcg_out32(s, insn | VRT(a0) | VRA(a1) | VRB(a2));
3776
3777    return need_inv;
3778}
3779
3780static void tcg_out_cmp_vec(TCGContext *s, unsigned vece, TCGReg a0,
3781                            TCGReg a1, TCGReg a2, TCGCond cond)
3782{
3783    if (tcg_out_cmp_vec_noinv(s, vece, a0, a1, a2, cond)) {
3784        tcg_out_not_vec(s, a0, a0);
3785    }
3786}
3787
3788static void tcg_out_cmpsel_vec(TCGContext *s, unsigned vece, TCGReg a0,
3789                               TCGReg c1, TCGReg c2, TCGArg v3, int const_v3,
3790                               TCGReg v4, TCGCond cond)
3791{
3792    bool inv = tcg_out_cmp_vec_noinv(s, vece, TCG_VEC_TMP1, c1, c2, cond);
3793
3794    if (!const_v3) {
3795        if (inv) {
3796            tcg_out_bitsel_vec(s, a0, TCG_VEC_TMP1, v4, v3);
3797        } else {
3798            tcg_out_bitsel_vec(s, a0, TCG_VEC_TMP1, v3, v4);
3799        }
3800    } else if (v3) {
3801        if (inv) {
3802            tcg_out_orc_vec(s, a0, v4, TCG_VEC_TMP1);
3803        } else {
3804            tcg_out_or_vec(s, a0, v4, TCG_VEC_TMP1);
3805        }
3806    } else {
3807        if (inv) {
3808            tcg_out_and_vec(s, a0, v4, TCG_VEC_TMP1);
3809        } else {
3810            tcg_out_andc_vec(s, a0, v4, TCG_VEC_TMP1);
3811        }
3812    }
3813}
3814
3815static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
3816                           unsigned vecl, unsigned vece,
3817                           const TCGArg args[TCG_MAX_OP_ARGS],
3818                           const int const_args[TCG_MAX_OP_ARGS])
3819{
3820    static const uint32_t
3821        add_op[4] = { VADDUBM, VADDUHM, VADDUWM, VADDUDM },
3822        sub_op[4] = { VSUBUBM, VSUBUHM, VSUBUWM, VSUBUDM },
3823        mul_op[4] = { 0, 0, VMULUWM, VMULLD },
3824        neg_op[4] = { 0, 0, VNEGW, VNEGD },
3825        ssadd_op[4] = { VADDSBS, VADDSHS, VADDSWS, 0 },
3826        usadd_op[4] = { VADDUBS, VADDUHS, VADDUWS, 0 },
3827        sssub_op[4] = { VSUBSBS, VSUBSHS, VSUBSWS, 0 },
3828        ussub_op[4] = { VSUBUBS, VSUBUHS, VSUBUWS, 0 },
3829        umin_op[4] = { VMINUB, VMINUH, VMINUW, VMINUD },
3830        smin_op[4] = { VMINSB, VMINSH, VMINSW, VMINSD },
3831        umax_op[4] = { VMAXUB, VMAXUH, VMAXUW, VMAXUD },
3832        smax_op[4] = { VMAXSB, VMAXSH, VMAXSW, VMAXSD },
3833        shlv_op[4] = { VSLB, VSLH, VSLW, VSLD },
3834        shrv_op[4] = { VSRB, VSRH, VSRW, VSRD },
3835        sarv_op[4] = { VSRAB, VSRAH, VSRAW, VSRAD },
3836        mrgh_op[4] = { VMRGHB, VMRGHH, VMRGHW, 0 },
3837        mrgl_op[4] = { VMRGLB, VMRGLH, VMRGLW, 0 },
3838        muleu_op[4] = { VMULEUB, VMULEUH, VMULEUW, 0 },
3839        mulou_op[4] = { VMULOUB, VMULOUH, VMULOUW, 0 },
3840        pkum_op[4] = { VPKUHUM, VPKUWUM, 0, 0 },
3841        rotl_op[4] = { VRLB, VRLH, VRLW, VRLD };
3842
3843    TCGType type = vecl + TCG_TYPE_V64;
3844    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
3845    uint32_t insn;
3846
3847    switch (opc) {
3848    case INDEX_op_ld_vec:
3849        tcg_out_ld(s, type, a0, a1, a2);
3850        return;
3851    case INDEX_op_st_vec:
3852        tcg_out_st(s, type, a0, a1, a2);
3853        return;
3854    case INDEX_op_dupm_vec:
3855        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
3856        return;
3857
3858    case INDEX_op_add_vec:
3859        insn = add_op[vece];
3860        break;
3861    case INDEX_op_sub_vec:
3862        insn = sub_op[vece];
3863        break;
3864    case INDEX_op_neg_vec:
3865        insn = neg_op[vece];
3866        a2 = a1;
3867        a1 = 0;
3868        break;
3869    case INDEX_op_mul_vec:
3870        insn = mul_op[vece];
3871        break;
3872    case INDEX_op_ssadd_vec:
3873        insn = ssadd_op[vece];
3874        break;
3875    case INDEX_op_sssub_vec:
3876        insn = sssub_op[vece];
3877        break;
3878    case INDEX_op_usadd_vec:
3879        insn = usadd_op[vece];
3880        break;
3881    case INDEX_op_ussub_vec:
3882        insn = ussub_op[vece];
3883        break;
3884    case INDEX_op_smin_vec:
3885        insn = smin_op[vece];
3886        break;
3887    case INDEX_op_umin_vec:
3888        insn = umin_op[vece];
3889        break;
3890    case INDEX_op_smax_vec:
3891        insn = smax_op[vece];
3892        break;
3893    case INDEX_op_umax_vec:
3894        insn = umax_op[vece];
3895        break;
3896    case INDEX_op_shlv_vec:
3897        insn = shlv_op[vece];
3898        break;
3899    case INDEX_op_shrv_vec:
3900        insn = shrv_op[vece];
3901        break;
3902    case INDEX_op_sarv_vec:
3903        insn = sarv_op[vece];
3904        break;
3905    case INDEX_op_and_vec:
3906        tcg_out_and_vec(s, a0, a1, a2);
3907        return;
3908    case INDEX_op_or_vec:
3909        tcg_out_or_vec(s, a0, a1, a2);
3910        return;
3911    case INDEX_op_xor_vec:
3912        insn = VXOR;
3913        break;
3914    case INDEX_op_andc_vec:
3915        tcg_out_andc_vec(s, a0, a1, a2);
3916        return;
3917    case INDEX_op_not_vec:
3918        tcg_out_not_vec(s, a0, a1);
3919        return;
3920    case INDEX_op_orc_vec:
3921        tcg_out_orc_vec(s, a0, a1, a2);
3922        return;
3923    case INDEX_op_nand_vec:
3924        insn = VNAND;
3925        break;
3926    case INDEX_op_nor_vec:
3927        insn = VNOR;
3928        break;
3929    case INDEX_op_eqv_vec:
3930        insn = VEQV;
3931        break;
3932
3933    case INDEX_op_cmp_vec:
3934        tcg_out_cmp_vec(s, vece, a0, a1, a2, args[3]);
3935        return;
3936    case INDEX_op_cmpsel_vec:
3937        tcg_out_cmpsel_vec(s, vece, a0, a1, a2,
3938                           args[3], const_args[3], args[4], args[5]);
3939        return;
3940    case INDEX_op_bitsel_vec:
3941        tcg_out_bitsel_vec(s, a0, a1, a2, args[3]);
3942        return;
3943
3944    case INDEX_op_dup2_vec:
3945        assert(TCG_TARGET_REG_BITS == 32);
3946        /* With inputs a1 = xLxx, a2 = xHxx  */
3947        tcg_out32(s, VMRGHW | VRT(a0) | VRA(a2) | VRB(a1));  /* a0  = xxHL */
3948        tcg_out_vsldoi(s, TCG_VEC_TMP1, a0, a0, 8);          /* tmp = HLxx */
3949        tcg_out_vsldoi(s, a0, a0, TCG_VEC_TMP1, 8);          /* a0  = HLHL */
3950        return;
3951
3952    case INDEX_op_ppc_mrgh_vec:
3953        insn = mrgh_op[vece];
3954        break;
3955    case INDEX_op_ppc_mrgl_vec:
3956        insn = mrgl_op[vece];
3957        break;
3958    case INDEX_op_ppc_muleu_vec:
3959        insn = muleu_op[vece];
3960        break;
3961    case INDEX_op_ppc_mulou_vec:
3962        insn = mulou_op[vece];
3963        break;
3964    case INDEX_op_ppc_pkum_vec:
3965        insn = pkum_op[vece];
3966        break;
3967    case INDEX_op_rotlv_vec:
3968        insn = rotl_op[vece];
3969        break;
3970    case INDEX_op_ppc_msum_vec:
3971        tcg_debug_assert(vece == MO_16);
3972        tcg_out32(s, VMSUMUHM | VRT(a0) | VRA(a1) | VRB(a2) | VRC(args[3]));
3973        return;
3974
3975    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
3976    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
3977    default:
3978        g_assert_not_reached();
3979    }
3980
3981    tcg_debug_assert(insn != 0);
3982    tcg_out32(s, insn | VRT(a0) | VRA(a1) | VRB(a2));
3983}
3984
3985static void expand_vec_shi(TCGType type, unsigned vece, TCGv_vec v0,
3986                           TCGv_vec v1, TCGArg imm, TCGOpcode opci)
3987{
3988    TCGv_vec t1;
3989
3990    if (vece == MO_32) {
3991        /*
3992         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
3993         * So using negative numbers gets us the 4th bit easily.
3994         */
3995        imm = sextract32(imm, 0, 5);
3996    } else {
3997        imm &= (8 << vece) - 1;
3998    }
3999
4000    /* Splat w/bytes for xxspltib when 2.07 allows MO_64. */
4001    t1 = tcg_constant_vec(type, MO_8, imm);
4002    vec_gen_3(opci, type, vece, tcgv_vec_arg(v0),
4003              tcgv_vec_arg(v1), tcgv_vec_arg(t1));
4004}
4005
4006static void expand_vec_mul(TCGType type, unsigned vece, TCGv_vec v0,
4007                           TCGv_vec v1, TCGv_vec v2)
4008{
4009    TCGv_vec t1 = tcg_temp_new_vec(type);
4010    TCGv_vec t2 = tcg_temp_new_vec(type);
4011    TCGv_vec c0, c16;
4012
4013    switch (vece) {
4014    case MO_8:
4015    case MO_16:
4016        vec_gen_3(INDEX_op_ppc_muleu_vec, type, vece, tcgv_vec_arg(t1),
4017                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
4018        vec_gen_3(INDEX_op_ppc_mulou_vec, type, vece, tcgv_vec_arg(t2),
4019                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
4020        vec_gen_3(INDEX_op_ppc_mrgh_vec, type, vece + 1, tcgv_vec_arg(v0),
4021                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
4022        vec_gen_3(INDEX_op_ppc_mrgl_vec, type, vece + 1, tcgv_vec_arg(t1),
4023                  tcgv_vec_arg(t1), tcgv_vec_arg(t2));
4024        vec_gen_3(INDEX_op_ppc_pkum_vec, type, vece, tcgv_vec_arg(v0),
4025                  tcgv_vec_arg(v0), tcgv_vec_arg(t1));
4026        break;
4027
4028    case MO_32:
4029        tcg_debug_assert(!have_isa_2_07);
4030        /*
4031         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
4032         * So using -16 is a quick way to represent 16.
4033         */
4034        c16 = tcg_constant_vec(type, MO_8, -16);
4035        c0 = tcg_constant_vec(type, MO_8, 0);
4036
4037        vec_gen_3(INDEX_op_rotlv_vec, type, MO_32, tcgv_vec_arg(t1),
4038                  tcgv_vec_arg(v2), tcgv_vec_arg(c16));
4039        vec_gen_3(INDEX_op_ppc_mulou_vec, type, MO_16, tcgv_vec_arg(t2),
4040                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
4041        vec_gen_4(INDEX_op_ppc_msum_vec, type, MO_16, tcgv_vec_arg(t1),
4042                  tcgv_vec_arg(v1), tcgv_vec_arg(t1), tcgv_vec_arg(c0));
4043        vec_gen_3(INDEX_op_shlv_vec, type, MO_32, tcgv_vec_arg(t1),
4044                  tcgv_vec_arg(t1), tcgv_vec_arg(c16));
4045        tcg_gen_add_vec(MO_32, v0, t1, t2);
4046        break;
4047
4048    default:
4049        g_assert_not_reached();
4050    }
4051    tcg_temp_free_vec(t1);
4052    tcg_temp_free_vec(t2);
4053}
4054
4055void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
4056                       TCGArg a0, ...)
4057{
4058    va_list va;
4059    TCGv_vec v0, v1, v2, t0;
4060    TCGArg a2;
4061
4062    va_start(va, a0);
4063    v0 = temp_tcgv_vec(arg_temp(a0));
4064    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
4065    a2 = va_arg(va, TCGArg);
4066
4067    switch (opc) {
4068    case INDEX_op_shli_vec:
4069        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shlv_vec);
4070        break;
4071    case INDEX_op_shri_vec:
4072        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_shrv_vec);
4073        break;
4074    case INDEX_op_sari_vec:
4075        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_sarv_vec);
4076        break;
4077    case INDEX_op_rotli_vec:
4078        expand_vec_shi(type, vece, v0, v1, a2, INDEX_op_rotlv_vec);
4079        break;
4080    case INDEX_op_mul_vec:
4081        v2 = temp_tcgv_vec(arg_temp(a2));
4082        expand_vec_mul(type, vece, v0, v1, v2);
4083        break;
4084    case INDEX_op_rotlv_vec:
4085        v2 = temp_tcgv_vec(arg_temp(a2));
4086        t0 = tcg_temp_new_vec(type);
4087        tcg_gen_neg_vec(vece, t0, v2);
4088        tcg_gen_rotlv_vec(vece, v0, v1, t0);
4089        tcg_temp_free_vec(t0);
4090        break;
4091    default:
4092        g_assert_not_reached();
4093    }
4094    va_end(va);
4095}
4096
4097static TCGConstraintSetIndex
4098tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
4099{
4100    switch (op) {
4101    case INDEX_op_goto_ptr:
4102        return C_O0_I1(r);
4103
4104    case INDEX_op_ld8u_i32:
4105    case INDEX_op_ld8s_i32:
4106    case INDEX_op_ld16u_i32:
4107    case INDEX_op_ld16s_i32:
4108    case INDEX_op_ld_i32:
4109    case INDEX_op_ctpop_i32:
4110    case INDEX_op_neg_i32:
4111    case INDEX_op_not_i32:
4112    case INDEX_op_ext8s_i32:
4113    case INDEX_op_ext16s_i32:
4114    case INDEX_op_bswap16_i32:
4115    case INDEX_op_bswap32_i32:
4116    case INDEX_op_extract_i32:
4117    case INDEX_op_sextract_i32:
4118    case INDEX_op_ld8u_i64:
4119    case INDEX_op_ld8s_i64:
4120    case INDEX_op_ld16u_i64:
4121    case INDEX_op_ld16s_i64:
4122    case INDEX_op_ld32u_i64:
4123    case INDEX_op_ld32s_i64:
4124    case INDEX_op_ld_i64:
4125    case INDEX_op_ctpop_i64:
4126    case INDEX_op_neg_i64:
4127    case INDEX_op_not_i64:
4128    case INDEX_op_ext8s_i64:
4129    case INDEX_op_ext16s_i64:
4130    case INDEX_op_ext32s_i64:
4131    case INDEX_op_ext_i32_i64:
4132    case INDEX_op_extu_i32_i64:
4133    case INDEX_op_bswap16_i64:
4134    case INDEX_op_bswap32_i64:
4135    case INDEX_op_bswap64_i64:
4136    case INDEX_op_extract_i64:
4137    case INDEX_op_sextract_i64:
4138        return C_O1_I1(r, r);
4139
4140    case INDEX_op_st8_i32:
4141    case INDEX_op_st16_i32:
4142    case INDEX_op_st_i32:
4143    case INDEX_op_st8_i64:
4144    case INDEX_op_st16_i64:
4145    case INDEX_op_st32_i64:
4146    case INDEX_op_st_i64:
4147        return C_O0_I2(r, r);
4148
4149    case INDEX_op_add_i32:
4150    case INDEX_op_and_i32:
4151    case INDEX_op_or_i32:
4152    case INDEX_op_xor_i32:
4153    case INDEX_op_andc_i32:
4154    case INDEX_op_orc_i32:
4155    case INDEX_op_eqv_i32:
4156    case INDEX_op_shl_i32:
4157    case INDEX_op_shr_i32:
4158    case INDEX_op_sar_i32:
4159    case INDEX_op_rotl_i32:
4160    case INDEX_op_rotr_i32:
4161    case INDEX_op_and_i64:
4162    case INDEX_op_andc_i64:
4163    case INDEX_op_shl_i64:
4164    case INDEX_op_shr_i64:
4165    case INDEX_op_sar_i64:
4166    case INDEX_op_rotl_i64:
4167    case INDEX_op_rotr_i64:
4168        return C_O1_I2(r, r, ri);
4169
4170    case INDEX_op_mul_i32:
4171    case INDEX_op_mul_i64:
4172        return C_O1_I2(r, r, rI);
4173
4174    case INDEX_op_div_i32:
4175    case INDEX_op_divu_i32:
4176    case INDEX_op_rem_i32:
4177    case INDEX_op_remu_i32:
4178    case INDEX_op_nand_i32:
4179    case INDEX_op_nor_i32:
4180    case INDEX_op_muluh_i32:
4181    case INDEX_op_mulsh_i32:
4182    case INDEX_op_orc_i64:
4183    case INDEX_op_eqv_i64:
4184    case INDEX_op_nand_i64:
4185    case INDEX_op_nor_i64:
4186    case INDEX_op_div_i64:
4187    case INDEX_op_divu_i64:
4188    case INDEX_op_rem_i64:
4189    case INDEX_op_remu_i64:
4190    case INDEX_op_mulsh_i64:
4191    case INDEX_op_muluh_i64:
4192        return C_O1_I2(r, r, r);
4193
4194    case INDEX_op_sub_i32:
4195        return C_O1_I2(r, rI, ri);
4196    case INDEX_op_add_i64:
4197        return C_O1_I2(r, r, rT);
4198    case INDEX_op_or_i64:
4199    case INDEX_op_xor_i64:
4200        return C_O1_I2(r, r, rU);
4201    case INDEX_op_sub_i64:
4202        return C_O1_I2(r, rI, rT);
4203    case INDEX_op_clz_i32:
4204    case INDEX_op_ctz_i32:
4205    case INDEX_op_clz_i64:
4206    case INDEX_op_ctz_i64:
4207        return C_O1_I2(r, r, rZW);
4208
4209    case INDEX_op_brcond_i32:
4210    case INDEX_op_brcond_i64:
4211        return C_O0_I2(r, rC);
4212    case INDEX_op_setcond_i32:
4213    case INDEX_op_setcond_i64:
4214    case INDEX_op_negsetcond_i32:
4215    case INDEX_op_negsetcond_i64:
4216        return C_O1_I2(r, r, rC);
4217    case INDEX_op_movcond_i32:
4218    case INDEX_op_movcond_i64:
4219        return C_O1_I4(r, r, rC, rZ, rZ);
4220
4221    case INDEX_op_deposit_i32:
4222    case INDEX_op_deposit_i64:
4223        return C_O1_I2(r, 0, rZ);
4224    case INDEX_op_brcond2_i32:
4225        return C_O0_I4(r, r, ri, ri);
4226    case INDEX_op_setcond2_i32:
4227        return C_O1_I4(r, r, r, ri, ri);
4228    case INDEX_op_add2_i64:
4229    case INDEX_op_add2_i32:
4230        return C_O2_I4(r, r, r, r, rI, rZM);
4231    case INDEX_op_sub2_i64:
4232    case INDEX_op_sub2_i32:
4233        return C_O2_I4(r, r, rI, rZM, r, r);
4234
4235    case INDEX_op_qemu_ld_i32:
4236        return C_O1_I1(r, r);
4237    case INDEX_op_qemu_ld_i64:
4238        return TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, r) : C_O2_I1(r, r, r);
4239
4240    case INDEX_op_qemu_st_i32:
4241        return C_O0_I2(r, r);
4242    case INDEX_op_qemu_st_i64:
4243        return TCG_TARGET_REG_BITS == 64 ? C_O0_I2(r, r) : C_O0_I3(r, r, r);
4244
4245    case INDEX_op_qemu_ld_i128:
4246        return C_N1O1_I1(o, m, r);
4247    case INDEX_op_qemu_st_i128:
4248        return C_O0_I3(o, m, r);
4249
4250    case INDEX_op_add_vec:
4251    case INDEX_op_sub_vec:
4252    case INDEX_op_mul_vec:
4253    case INDEX_op_and_vec:
4254    case INDEX_op_or_vec:
4255    case INDEX_op_xor_vec:
4256    case INDEX_op_andc_vec:
4257    case INDEX_op_orc_vec:
4258    case INDEX_op_nor_vec:
4259    case INDEX_op_eqv_vec:
4260    case INDEX_op_nand_vec:
4261    case INDEX_op_cmp_vec:
4262    case INDEX_op_ssadd_vec:
4263    case INDEX_op_sssub_vec:
4264    case INDEX_op_usadd_vec:
4265    case INDEX_op_ussub_vec:
4266    case INDEX_op_smax_vec:
4267    case INDEX_op_smin_vec:
4268    case INDEX_op_umax_vec:
4269    case INDEX_op_umin_vec:
4270    case INDEX_op_shlv_vec:
4271    case INDEX_op_shrv_vec:
4272    case INDEX_op_sarv_vec:
4273    case INDEX_op_rotlv_vec:
4274    case INDEX_op_rotrv_vec:
4275    case INDEX_op_ppc_mrgh_vec:
4276    case INDEX_op_ppc_mrgl_vec:
4277    case INDEX_op_ppc_muleu_vec:
4278    case INDEX_op_ppc_mulou_vec:
4279    case INDEX_op_ppc_pkum_vec:
4280    case INDEX_op_dup2_vec:
4281        return C_O1_I2(v, v, v);
4282
4283    case INDEX_op_not_vec:
4284    case INDEX_op_neg_vec:
4285        return C_O1_I1(v, v);
4286
4287    case INDEX_op_dup_vec:
4288        return have_isa_3_00 ? C_O1_I1(v, vr) : C_O1_I1(v, v);
4289
4290    case INDEX_op_ld_vec:
4291    case INDEX_op_dupm_vec:
4292        return C_O1_I1(v, r);
4293
4294    case INDEX_op_st_vec:
4295        return C_O0_I2(v, r);
4296
4297    case INDEX_op_bitsel_vec:
4298    case INDEX_op_ppc_msum_vec:
4299        return C_O1_I3(v, v, v, v);
4300    case INDEX_op_cmpsel_vec:
4301        return C_O1_I4(v, v, v, vZM, v);
4302
4303    default:
4304        return C_NotImplemented;
4305    }
4306}
4307
4308static void tcg_target_init(TCGContext *s)
4309{
4310    tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
4311    tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
4312    if (have_altivec) {
4313        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
4314        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
4315    }
4316
4317    tcg_target_call_clobber_regs = 0;
4318    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
4319    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
4320    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
4321    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
4322    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
4323    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
4324    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R7);
4325    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R8);
4326    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R9);
4327    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R10);
4328    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R11);
4329    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
4330
4331    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
4332    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
4333    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
4334    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
4335    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
4336    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
4337    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
4338    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
4339    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V8);
4340    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V9);
4341    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V10);
4342    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V11);
4343    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V12);
4344    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V13);
4345    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V14);
4346    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V15);
4347    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
4348    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
4349    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
4350    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
4351
4352    s->reserved_regs = 0;
4353    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); /* tcg temp */
4354    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R1); /* stack pointer */
4355#if defined(_CALL_SYSV)
4356    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R2); /* toc pointer */
4357#endif
4358#if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
4359    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R13); /* thread pointer */
4360#endif
4361    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1);
4362    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP2);
4363    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP1);
4364    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP2);
4365    if (USE_REG_TB) {
4366        tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);  /* tb->tc_ptr */
4367    }
4368}
4369
4370#ifdef __ELF__
4371typedef struct {
4372    DebugFrameCIE cie;
4373    DebugFrameFDEHeader fde;
4374    uint8_t fde_def_cfa[4];
4375    uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2 + 3];
4376} DebugFrame;
4377
4378/* We're expecting a 2 byte uleb128 encoded value.  */
4379QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
4380
4381#if TCG_TARGET_REG_BITS == 64
4382# define ELF_HOST_MACHINE EM_PPC64
4383#else
4384# define ELF_HOST_MACHINE EM_PPC
4385#endif
4386
4387static DebugFrame debug_frame = {
4388    .cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
4389    .cie.id = -1,
4390    .cie.version = 1,
4391    .cie.code_align = 1,
4392    .cie.data_align = (-SZR & 0x7f),         /* sleb128 -SZR */
4393    .cie.return_column = 65,
4394
4395    /* Total FDE size does not include the "len" member.  */
4396    .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset),
4397
4398    .fde_def_cfa = {
4399        12, TCG_REG_R1,                 /* DW_CFA_def_cfa r1, ... */
4400        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
4401        (FRAME_SIZE >> 7)
4402    },
4403    .fde_reg_ofs = {
4404        /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
4405        0x11, 65, (LR_OFFSET / -SZR) & 0x7f,
4406    }
4407};
4408
4409void tcg_register_jit(const void *buf, size_t buf_size)
4410{
4411    uint8_t *p = &debug_frame.fde_reg_ofs[3];
4412    int i;
4413
4414    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i, p += 2) {
4415        p[0] = 0x80 + tcg_target_callee_save_regs[i];
4416        p[1] = (FRAME_SIZE - (REG_SAVE_BOT + i * SZR)) / SZR;
4417    }
4418
4419    debug_frame.fde.func_start = (uintptr_t)buf;
4420    debug_frame.fde.func_len = buf_size;
4421
4422    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
4423}
4424#endif /* __ELF__ */
4425#undef VMULEUB
4426#undef VMULEUH
4427#undef VMULEUW
4428#undef VMULOUB
4429#undef VMULOUH
4430#undef VMULOUW
4431#undef VMSUMUHM
4432