xref: /openbmc/qemu/tcg/aarch64/tcg-target.c.inc (revision 5a5bb0a5a0b879c8f110c6a9bde9146181ef840c)
1/*
2 * Initial TCG Implementation for aarch64
3 *
4 * Copyright (c) 2013 Huawei Technologies Duesseldorf GmbH
5 * Written by Claudio Fontana
6 *
7 * This work is licensed under the terms of the GNU GPL, version 2 or
8 * (at your option) any later version.
9 *
10 * See the COPYING file in the top-level directory for details.
11 */
12
13#include "qemu/bitops.h"
14
15/* Used for function call generation. */
16#define TCG_REG_CALL_STACK              TCG_REG_SP
17#define TCG_TARGET_STACK_ALIGN          16
18#define TCG_TARGET_CALL_STACK_OFFSET    0
19#define TCG_TARGET_CALL_ARG_I32         TCG_CALL_ARG_NORMAL
20#define TCG_TARGET_CALL_ARG_I64         TCG_CALL_ARG_NORMAL
21#ifdef CONFIG_DARWIN
22# define TCG_TARGET_CALL_ARG_I128       TCG_CALL_ARG_NORMAL
23#else
24# define TCG_TARGET_CALL_ARG_I128       TCG_CALL_ARG_EVEN
25#endif
26#define TCG_TARGET_CALL_RET_I128        TCG_CALL_RET_NORMAL
27
28/* We're going to re-use TCGType in setting of the SF bit, which controls
29   the size of the operation performed.  If we know the values match, it
30   makes things much cleaner.  */
31QEMU_BUILD_BUG_ON(TCG_TYPE_I32 != 0 || TCG_TYPE_I64 != 1);
32
33#ifdef CONFIG_DEBUG_TCG
34static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
35    "x0", "x1", "x2", "x3", "x4", "x5", "x6", "x7",
36    "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15",
37    "x16", "x17", "x18", "x19", "x20", "x21", "x22", "x23",
38    "x24", "x25", "x26", "x27", "x28", "fp", "x30", "sp",
39
40    "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
41    "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15",
42    "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
43    "v24", "v25", "v26", "v27", "v28", "fp", "v30", "v31",
44};
45#endif /* CONFIG_DEBUG_TCG */
46
47static const int tcg_target_reg_alloc_order[] = {
48    TCG_REG_X20, TCG_REG_X21, TCG_REG_X22, TCG_REG_X23,
49    TCG_REG_X24, TCG_REG_X25, TCG_REG_X26, TCG_REG_X27,
50    TCG_REG_X28, /* we will reserve this for guest_base if configured */
51
52    TCG_REG_X8, TCG_REG_X9, TCG_REG_X10, TCG_REG_X11,
53    TCG_REG_X12, TCG_REG_X13, TCG_REG_X14, TCG_REG_X15,
54
55    TCG_REG_X0, TCG_REG_X1, TCG_REG_X2, TCG_REG_X3,
56    TCG_REG_X4, TCG_REG_X5, TCG_REG_X6, TCG_REG_X7,
57
58    /* X16 reserved as temporary */
59    /* X17 reserved as temporary */
60    /* X18 reserved by system */
61    /* X19 reserved for AREG0 */
62    /* X29 reserved as fp */
63    /* X30 reserved as temporary */
64
65    TCG_REG_V0, TCG_REG_V1, TCG_REG_V2, TCG_REG_V3,
66    TCG_REG_V4, TCG_REG_V5, TCG_REG_V6, TCG_REG_V7,
67    /* V8 - V15 are call-saved, and skipped.  */
68    TCG_REG_V16, TCG_REG_V17, TCG_REG_V18, TCG_REG_V19,
69    TCG_REG_V20, TCG_REG_V21, TCG_REG_V22, TCG_REG_V23,
70    TCG_REG_V24, TCG_REG_V25, TCG_REG_V26, TCG_REG_V27,
71    TCG_REG_V28, TCG_REG_V29, TCG_REG_V30, TCG_REG_V31,
72};
73
74static const int tcg_target_call_iarg_regs[8] = {
75    TCG_REG_X0, TCG_REG_X1, TCG_REG_X2, TCG_REG_X3,
76    TCG_REG_X4, TCG_REG_X5, TCG_REG_X6, TCG_REG_X7
77};
78
79static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
80{
81    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
82    tcg_debug_assert(slot >= 0 && slot <= 1);
83    return TCG_REG_X0 + slot;
84}
85
86#define TCG_REG_TMP0 TCG_REG_X16
87#define TCG_REG_TMP1 TCG_REG_X17
88#define TCG_REG_TMP2 TCG_REG_X30
89#define TCG_VEC_TMP0 TCG_REG_V31
90
91#define TCG_REG_GUEST_BASE TCG_REG_X28
92
93static bool reloc_pc26(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
94{
95    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
96    ptrdiff_t offset = target - src_rx;
97
98    if (offset == sextract64(offset, 0, 26)) {
99        /* read instruction, mask away previous PC_REL26 parameter contents,
100           set the proper offset, then write back the instruction. */
101        *src_rw = deposit32(*src_rw, 0, 26, offset);
102        return true;
103    }
104    return false;
105}
106
107static bool reloc_pc19(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
108{
109    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
110    ptrdiff_t offset = target - src_rx;
111
112    if (offset == sextract64(offset, 0, 19)) {
113        *src_rw = deposit32(*src_rw, 5, 19, offset);
114        return true;
115    }
116    return false;
117}
118
119static bool reloc_pc14(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
120{
121    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
122    ptrdiff_t offset = target - src_rx;
123
124    if (offset == sextract64(offset, 0, 14)) {
125        *src_rw = deposit32(*src_rw, 5, 14, offset);
126        return true;
127    }
128    return false;
129}
130
131static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
132                        intptr_t value, intptr_t addend)
133{
134    tcg_debug_assert(addend == 0);
135    switch (type) {
136    case R_AARCH64_JUMP26:
137    case R_AARCH64_CALL26:
138        return reloc_pc26(code_ptr, (const tcg_insn_unit *)value);
139    case R_AARCH64_CONDBR19:
140        return reloc_pc19(code_ptr, (const tcg_insn_unit *)value);
141    case R_AARCH64_TSTBR14:
142        return reloc_pc14(code_ptr, (const tcg_insn_unit *)value);
143    default:
144        g_assert_not_reached();
145    }
146}
147
148#define TCG_CT_CONST_AIMM 0x100
149#define TCG_CT_CONST_LIMM 0x200
150#define TCG_CT_CONST_ZERO 0x400
151#define TCG_CT_CONST_MONE 0x800
152#define TCG_CT_CONST_ORRI 0x1000
153#define TCG_CT_CONST_ANDI 0x2000
154#define TCG_CT_CONST_CMP  0x4000
155
156#define ALL_GENERAL_REGS  0xffffffffu
157#define ALL_VECTOR_REGS   0xffffffff00000000ull
158
159/* Match a constant valid for addition (12-bit, optionally shifted).  */
160static inline bool is_aimm(uint64_t val)
161{
162    return (val & ~0xfff) == 0 || (val & ~0xfff000) == 0;
163}
164
165/* Match a constant valid for logical operations.  */
166static inline bool is_limm(uint64_t val)
167{
168    /* Taking a simplified view of the logical immediates for now, ignoring
169       the replication that can happen across the field.  Match bit patterns
170       of the forms
171           0....01....1
172           0..01..10..0
173       and their inverses.  */
174
175    /* Make things easier below, by testing the form with msb clear. */
176    if ((int64_t)val < 0) {
177        val = ~val;
178    }
179    if (val == 0) {
180        return false;
181    }
182    val += val & -val;
183    return (val & (val - 1)) == 0;
184}
185
186/* Return true if v16 is a valid 16-bit shifted immediate.  */
187static bool is_shimm16(uint16_t v16, int *cmode, int *imm8)
188{
189    if (v16 == (v16 & 0xff)) {
190        *cmode = 0x8;
191        *imm8 = v16 & 0xff;
192        return true;
193    } else if (v16 == (v16 & 0xff00)) {
194        *cmode = 0xa;
195        *imm8 = v16 >> 8;
196        return true;
197    }
198    return false;
199}
200
201/* Return true if v32 is a valid 32-bit shifted immediate.  */
202static bool is_shimm32(uint32_t v32, int *cmode, int *imm8)
203{
204    if (v32 == (v32 & 0xff)) {
205        *cmode = 0x0;
206        *imm8 = v32 & 0xff;
207        return true;
208    } else if (v32 == (v32 & 0xff00)) {
209        *cmode = 0x2;
210        *imm8 = (v32 >> 8) & 0xff;
211        return true;
212    } else if (v32 == (v32 & 0xff0000)) {
213        *cmode = 0x4;
214        *imm8 = (v32 >> 16) & 0xff;
215        return true;
216    } else if (v32 == (v32 & 0xff000000)) {
217        *cmode = 0x6;
218        *imm8 = v32 >> 24;
219        return true;
220    }
221    return false;
222}
223
224/* Return true if v32 is a valid 32-bit shifting ones immediate.  */
225static bool is_soimm32(uint32_t v32, int *cmode, int *imm8)
226{
227    if ((v32 & 0xffff00ff) == 0xff) {
228        *cmode = 0xc;
229        *imm8 = (v32 >> 8) & 0xff;
230        return true;
231    } else if ((v32 & 0xff00ffff) == 0xffff) {
232        *cmode = 0xd;
233        *imm8 = (v32 >> 16) & 0xff;
234        return true;
235    }
236    return false;
237}
238
239/* Return true if v32 is a valid float32 immediate.  */
240static bool is_fimm32(uint32_t v32, int *cmode, int *imm8)
241{
242    if (extract32(v32, 0, 19) == 0
243        && (extract32(v32, 25, 6) == 0x20
244            || extract32(v32, 25, 6) == 0x1f)) {
245        *cmode = 0xf;
246        *imm8 = (extract32(v32, 31, 1) << 7)
247              | (extract32(v32, 25, 1) << 6)
248              | extract32(v32, 19, 6);
249        return true;
250    }
251    return false;
252}
253
254/* Return true if v64 is a valid float64 immediate.  */
255static bool is_fimm64(uint64_t v64, int *cmode, int *imm8)
256{
257    if (extract64(v64, 0, 48) == 0
258        && (extract64(v64, 54, 9) == 0x100
259            || extract64(v64, 54, 9) == 0x0ff)) {
260        *cmode = 0xf;
261        *imm8 = (extract64(v64, 63, 1) << 7)
262              | (extract64(v64, 54, 1) << 6)
263              | extract64(v64, 48, 6);
264        return true;
265    }
266    return false;
267}
268
269/*
270 * Return non-zero if v32 can be formed by MOVI+ORR.
271 * Place the parameters for MOVI in (cmode, imm8).
272 * Return the cmode for ORR; the imm8 can be had via extraction from v32.
273 */
274static int is_shimm32_pair(uint32_t v32, int *cmode, int *imm8)
275{
276    int i;
277
278    for (i = 6; i > 0; i -= 2) {
279        /* Mask out one byte we can add with ORR.  */
280        uint32_t tmp = v32 & ~(0xffu << (i * 4));
281        if (is_shimm32(tmp, cmode, imm8) ||
282            is_soimm32(tmp, cmode, imm8)) {
283            break;
284        }
285    }
286    return i;
287}
288
289/* Return true if V is a valid 16-bit or 32-bit shifted immediate.  */
290static bool is_shimm1632(uint32_t v32, int *cmode, int *imm8)
291{
292    if (v32 == deposit32(v32, 16, 16, v32)) {
293        return is_shimm16(v32, cmode, imm8);
294    } else {
295        return is_shimm32(v32, cmode, imm8);
296    }
297}
298
299static bool tcg_target_const_match(int64_t val, int ct,
300                                   TCGType type, TCGCond cond, int vece)
301{
302    if (ct & TCG_CT_CONST) {
303        return 1;
304    }
305    if (type == TCG_TYPE_I32) {
306        val = (int32_t)val;
307    }
308
309    if (ct & TCG_CT_CONST_CMP) {
310        if (is_tst_cond(cond)) {
311            ct |= TCG_CT_CONST_LIMM;
312        } else {
313            ct |= TCG_CT_CONST_AIMM;
314        }
315    }
316
317    if ((ct & TCG_CT_CONST_AIMM) && (is_aimm(val) || is_aimm(-val))) {
318        return 1;
319    }
320    if ((ct & TCG_CT_CONST_LIMM) && is_limm(val)) {
321        return 1;
322    }
323    if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
324        return 1;
325    }
326    if ((ct & TCG_CT_CONST_MONE) && val == -1) {
327        return 1;
328    }
329
330    switch (ct & (TCG_CT_CONST_ORRI | TCG_CT_CONST_ANDI)) {
331    case 0:
332        break;
333    case TCG_CT_CONST_ANDI:
334        val = ~val;
335        /* fallthru */
336    case TCG_CT_CONST_ORRI:
337        if (val == deposit64(val, 32, 32, val)) {
338            int cmode, imm8;
339            return is_shimm1632(val, &cmode, &imm8);
340        }
341        break;
342    default:
343        /* Both bits should not be set for the same insn.  */
344        g_assert_not_reached();
345    }
346
347    return 0;
348}
349
350enum aarch64_cond_code {
351    COND_EQ = 0x0,
352    COND_NE = 0x1,
353    COND_CS = 0x2,     /* Unsigned greater or equal */
354    COND_HS = COND_CS, /* ALIAS greater or equal */
355    COND_CC = 0x3,     /* Unsigned less than */
356    COND_LO = COND_CC, /* ALIAS Lower */
357    COND_MI = 0x4,     /* Negative */
358    COND_PL = 0x5,     /* Zero or greater */
359    COND_VS = 0x6,     /* Overflow */
360    COND_VC = 0x7,     /* No overflow */
361    COND_HI = 0x8,     /* Unsigned greater than */
362    COND_LS = 0x9,     /* Unsigned less or equal */
363    COND_GE = 0xa,
364    COND_LT = 0xb,
365    COND_GT = 0xc,
366    COND_LE = 0xd,
367    COND_AL = 0xe,
368    COND_NV = 0xf, /* behaves like COND_AL here */
369};
370
371static const enum aarch64_cond_code tcg_cond_to_aarch64[] = {
372    [TCG_COND_EQ] = COND_EQ,
373    [TCG_COND_NE] = COND_NE,
374    [TCG_COND_LT] = COND_LT,
375    [TCG_COND_GE] = COND_GE,
376    [TCG_COND_LE] = COND_LE,
377    [TCG_COND_GT] = COND_GT,
378    /* unsigned */
379    [TCG_COND_LTU] = COND_LO,
380    [TCG_COND_GTU] = COND_HI,
381    [TCG_COND_GEU] = COND_HS,
382    [TCG_COND_LEU] = COND_LS,
383    /* bit test */
384    [TCG_COND_TSTEQ] = COND_EQ,
385    [TCG_COND_TSTNE] = COND_NE,
386};
387
388typedef enum {
389    LDST_ST = 0,    /* store */
390    LDST_LD = 1,    /* load */
391    LDST_LD_S_X = 2,  /* load and sign-extend into Xt */
392    LDST_LD_S_W = 3,  /* load and sign-extend into Wt */
393} AArch64LdstType;
394
395/* We encode the format of the insn into the beginning of the name, so that
396   we can have the preprocessor help "typecheck" the insn vs the output
397   function.  Arm didn't provide us with nice names for the formats, so we
398   use the section number of the architecture reference manual in which the
399   instruction group is described.  */
400typedef enum {
401    /* Compare and branch (immediate).  */
402    I3201_CBZ       = 0x34000000,
403    I3201_CBNZ      = 0x35000000,
404
405    /* Conditional branch (immediate).  */
406    I3202_B_C       = 0x54000000,
407
408    /* Test and branch (immediate).  */
409    I3205_TBZ       = 0x36000000,
410    I3205_TBNZ      = 0x37000000,
411
412    /* Unconditional branch (immediate).  */
413    I3206_B         = 0x14000000,
414    I3206_BL        = 0x94000000,
415
416    /* Unconditional branch (register).  */
417    I3207_BR        = 0xd61f0000,
418    I3207_BLR       = 0xd63f0000,
419    I3207_RET       = 0xd65f0000,
420
421    /* AdvSIMD load/store single structure.  */
422    I3303_LD1R      = 0x0d40c000,
423
424    /* Load literal for loading the address at pc-relative offset */
425    I3305_LDR       = 0x58000000,
426    I3305_LDR_v64   = 0x5c000000,
427    I3305_LDR_v128  = 0x9c000000,
428
429    /* Load/store exclusive. */
430    I3306_LDXP      = 0xc8600000,
431    I3306_STXP      = 0xc8200000,
432
433    /* Load/store register.  Described here as 3.3.12, but the helper
434       that emits them can transform to 3.3.10 or 3.3.13.  */
435    I3312_STRB      = 0x38000000 | LDST_ST << 22 | MO_8 << 30,
436    I3312_STRH      = 0x38000000 | LDST_ST << 22 | MO_16 << 30,
437    I3312_STRW      = 0x38000000 | LDST_ST << 22 | MO_32 << 30,
438    I3312_STRX      = 0x38000000 | LDST_ST << 22 | MO_64 << 30,
439
440    I3312_LDRB      = 0x38000000 | LDST_LD << 22 | MO_8 << 30,
441    I3312_LDRH      = 0x38000000 | LDST_LD << 22 | MO_16 << 30,
442    I3312_LDRW      = 0x38000000 | LDST_LD << 22 | MO_32 << 30,
443    I3312_LDRX      = 0x38000000 | LDST_LD << 22 | MO_64 << 30,
444
445    I3312_LDRSBW    = 0x38000000 | LDST_LD_S_W << 22 | MO_8 << 30,
446    I3312_LDRSHW    = 0x38000000 | LDST_LD_S_W << 22 | MO_16 << 30,
447
448    I3312_LDRSBX    = 0x38000000 | LDST_LD_S_X << 22 | MO_8 << 30,
449    I3312_LDRSHX    = 0x38000000 | LDST_LD_S_X << 22 | MO_16 << 30,
450    I3312_LDRSWX    = 0x38000000 | LDST_LD_S_X << 22 | MO_32 << 30,
451
452    I3312_LDRVS     = 0x3c000000 | LDST_LD << 22 | MO_32 << 30,
453    I3312_STRVS     = 0x3c000000 | LDST_ST << 22 | MO_32 << 30,
454
455    I3312_LDRVD     = 0x3c000000 | LDST_LD << 22 | MO_64 << 30,
456    I3312_STRVD     = 0x3c000000 | LDST_ST << 22 | MO_64 << 30,
457
458    I3312_LDRVQ     = 0x3c000000 | 3 << 22 | 0 << 30,
459    I3312_STRVQ     = 0x3c000000 | 2 << 22 | 0 << 30,
460
461    I3312_TO_I3310  = 0x00200800,
462    I3312_TO_I3313  = 0x01000000,
463
464    /* Load/store register pair instructions.  */
465    I3314_LDP       = 0x28400000,
466    I3314_STP       = 0x28000000,
467
468    /* Add/subtract immediate instructions.  */
469    I3401_ADDI      = 0x11000000,
470    I3401_ADDSI     = 0x31000000,
471    I3401_SUBI      = 0x51000000,
472    I3401_SUBSI     = 0x71000000,
473
474    /* Bitfield instructions.  */
475    I3402_BFM       = 0x33000000,
476    I3402_SBFM      = 0x13000000,
477    I3402_UBFM      = 0x53000000,
478
479    /* Extract instruction.  */
480    I3403_EXTR      = 0x13800000,
481
482    /* Logical immediate instructions.  */
483    I3404_ANDI      = 0x12000000,
484    I3404_ORRI      = 0x32000000,
485    I3404_EORI      = 0x52000000,
486    I3404_ANDSI     = 0x72000000,
487
488    /* Move wide immediate instructions.  */
489    I3405_MOVN      = 0x12800000,
490    I3405_MOVZ      = 0x52800000,
491    I3405_MOVK      = 0x72800000,
492
493    /* PC relative addressing instructions.  */
494    I3406_ADR       = 0x10000000,
495    I3406_ADRP      = 0x90000000,
496
497    /* Add/subtract extended register instructions. */
498    I3501_ADD       = 0x0b200000,
499
500    /* Add/subtract shifted register instructions (without a shift).  */
501    I3502_ADD       = 0x0b000000,
502    I3502_ADDS      = 0x2b000000,
503    I3502_SUB       = 0x4b000000,
504    I3502_SUBS      = 0x6b000000,
505
506    /* Add/subtract shifted register instructions (with a shift).  */
507    I3502S_ADD_LSL  = I3502_ADD,
508
509    /* Add/subtract with carry instructions.  */
510    I3503_ADC       = 0x1a000000,
511    I3503_SBC       = 0x5a000000,
512
513    /* Conditional select instructions.  */
514    I3506_CSEL      = 0x1a800000,
515    I3506_CSINC     = 0x1a800400,
516    I3506_CSINV     = 0x5a800000,
517    I3506_CSNEG     = 0x5a800400,
518
519    /* Data-processing (1 source) instructions.  */
520    I3507_CLZ       = 0x5ac01000,
521    I3507_RBIT      = 0x5ac00000,
522    I3507_REV       = 0x5ac00000, /* + size << 10 */
523
524    /* Data-processing (2 source) instructions.  */
525    I3508_LSLV      = 0x1ac02000,
526    I3508_LSRV      = 0x1ac02400,
527    I3508_ASRV      = 0x1ac02800,
528    I3508_RORV      = 0x1ac02c00,
529    I3508_SMULH     = 0x9b407c00,
530    I3508_UMULH     = 0x9bc07c00,
531    I3508_UDIV      = 0x1ac00800,
532    I3508_SDIV      = 0x1ac00c00,
533
534    /* Data-processing (3 source) instructions.  */
535    I3509_MADD      = 0x1b000000,
536    I3509_MSUB      = 0x1b008000,
537
538    /* Logical shifted register instructions (without a shift).  */
539    I3510_AND       = 0x0a000000,
540    I3510_BIC       = 0x0a200000,
541    I3510_ORR       = 0x2a000000,
542    I3510_ORN       = 0x2a200000,
543    I3510_EOR       = 0x4a000000,
544    I3510_EON       = 0x4a200000,
545    I3510_ANDS      = 0x6a000000,
546
547    /* Logical shifted register instructions (with a shift).  */
548    I3502S_AND_LSR  = I3510_AND | (1 << 22),
549
550    /* AdvSIMD copy */
551    I3605_DUP      = 0x0e000400,
552    I3605_INS      = 0x4e001c00,
553    I3605_UMOV     = 0x0e003c00,
554
555    /* AdvSIMD modified immediate */
556    I3606_MOVI      = 0x0f000400,
557    I3606_MVNI      = 0x2f000400,
558    I3606_BIC       = 0x2f001400,
559    I3606_ORR       = 0x0f001400,
560
561    /* AdvSIMD scalar shift by immediate */
562    I3609_SSHR      = 0x5f000400,
563    I3609_SSRA      = 0x5f001400,
564    I3609_SHL       = 0x5f005400,
565    I3609_USHR      = 0x7f000400,
566    I3609_USRA      = 0x7f001400,
567    I3609_SLI       = 0x7f005400,
568
569    /* AdvSIMD scalar three same */
570    I3611_SQADD     = 0x5e200c00,
571    I3611_SQSUB     = 0x5e202c00,
572    I3611_CMGT      = 0x5e203400,
573    I3611_CMGE      = 0x5e203c00,
574    I3611_SSHL      = 0x5e204400,
575    I3611_ADD       = 0x5e208400,
576    I3611_CMTST     = 0x5e208c00,
577    I3611_UQADD     = 0x7e200c00,
578    I3611_UQSUB     = 0x7e202c00,
579    I3611_CMHI      = 0x7e203400,
580    I3611_CMHS      = 0x7e203c00,
581    I3611_USHL      = 0x7e204400,
582    I3611_SUB       = 0x7e208400,
583    I3611_CMEQ      = 0x7e208c00,
584
585    /* AdvSIMD scalar two-reg misc */
586    I3612_CMGT0     = 0x5e208800,
587    I3612_CMEQ0     = 0x5e209800,
588    I3612_CMLT0     = 0x5e20a800,
589    I3612_ABS       = 0x5e20b800,
590    I3612_CMGE0     = 0x7e208800,
591    I3612_CMLE0     = 0x7e209800,
592    I3612_NEG       = 0x7e20b800,
593
594    /* AdvSIMD shift by immediate */
595    I3614_SSHR      = 0x0f000400,
596    I3614_SSRA      = 0x0f001400,
597    I3614_SHL       = 0x0f005400,
598    I3614_SLI       = 0x2f005400,
599    I3614_USHR      = 0x2f000400,
600    I3614_USRA      = 0x2f001400,
601
602    /* AdvSIMD three same.  */
603    I3616_ADD       = 0x0e208400,
604    I3616_AND       = 0x0e201c00,
605    I3616_BIC       = 0x0e601c00,
606    I3616_BIF       = 0x2ee01c00,
607    I3616_BIT       = 0x2ea01c00,
608    I3616_BSL       = 0x2e601c00,
609    I3616_EOR       = 0x2e201c00,
610    I3616_MUL       = 0x0e209c00,
611    I3616_ORR       = 0x0ea01c00,
612    I3616_ORN       = 0x0ee01c00,
613    I3616_SUB       = 0x2e208400,
614    I3616_CMGT      = 0x0e203400,
615    I3616_CMGE      = 0x0e203c00,
616    I3616_CMTST     = 0x0e208c00,
617    I3616_CMHI      = 0x2e203400,
618    I3616_CMHS      = 0x2e203c00,
619    I3616_CMEQ      = 0x2e208c00,
620    I3616_SMAX      = 0x0e206400,
621    I3616_SMIN      = 0x0e206c00,
622    I3616_SSHL      = 0x0e204400,
623    I3616_SQADD     = 0x0e200c00,
624    I3616_SQSUB     = 0x0e202c00,
625    I3616_UMAX      = 0x2e206400,
626    I3616_UMIN      = 0x2e206c00,
627    I3616_UQADD     = 0x2e200c00,
628    I3616_UQSUB     = 0x2e202c00,
629    I3616_USHL      = 0x2e204400,
630
631    /* AdvSIMD two-reg misc.  */
632    I3617_CMGT0     = 0x0e208800,
633    I3617_CMEQ0     = 0x0e209800,
634    I3617_CMLT0     = 0x0e20a800,
635    I3617_CMGE0     = 0x2e208800,
636    I3617_CMLE0     = 0x2e209800,
637    I3617_NOT       = 0x2e205800,
638    I3617_ABS       = 0x0e20b800,
639    I3617_NEG       = 0x2e20b800,
640
641    /* System instructions.  */
642    NOP             = 0xd503201f,
643    DMB_ISH         = 0xd50338bf,
644    DMB_LD          = 0x00000100,
645    DMB_ST          = 0x00000200,
646
647    BTI_C           = 0xd503245f,
648    BTI_J           = 0xd503249f,
649    BTI_JC          = 0xd50324df,
650} AArch64Insn;
651
652static inline uint32_t tcg_in32(TCGContext *s)
653{
654    uint32_t v = *(uint32_t *)s->code_ptr;
655    return v;
656}
657
658/* Emit an opcode with "type-checking" of the format.  */
659#define tcg_out_insn(S, FMT, OP, ...) \
660    glue(tcg_out_insn_,FMT)(S, glue(glue(glue(I,FMT),_),OP), ## __VA_ARGS__)
661
662static void tcg_out_insn_3303(TCGContext *s, AArch64Insn insn, bool q,
663                              TCGReg rt, TCGReg rn, unsigned size)
664{
665    tcg_out32(s, insn | (rt & 0x1f) | (rn << 5) | (size << 10) | (q << 30));
666}
667
668static void tcg_out_insn_3305(TCGContext *s, AArch64Insn insn,
669                              int imm19, TCGReg rt)
670{
671    tcg_out32(s, insn | (imm19 & 0x7ffff) << 5 | rt);
672}
673
674static void tcg_out_insn_3306(TCGContext *s, AArch64Insn insn, TCGReg rs,
675                              TCGReg rt, TCGReg rt2, TCGReg rn)
676{
677    tcg_out32(s, insn | rs << 16 | rt2 << 10 | rn << 5 | rt);
678}
679
680static void tcg_out_insn_3201(TCGContext *s, AArch64Insn insn, TCGType ext,
681                              TCGReg rt, int imm19)
682{
683    tcg_out32(s, insn | ext << 31 | (imm19 & 0x7ffff) << 5 | rt);
684}
685
686static void tcg_out_insn_3202(TCGContext *s, AArch64Insn insn,
687                              TCGCond c, int imm19)
688{
689    tcg_out32(s, insn | tcg_cond_to_aarch64[c] | (imm19 & 0x7ffff) << 5);
690}
691
692static void tcg_out_insn_3205(TCGContext *s, AArch64Insn insn,
693                              TCGReg rt, int imm6, int imm14)
694{
695    insn |= (imm6 & 0x20) << (31 - 5);
696    insn |= (imm6 & 0x1f) << 19;
697    tcg_out32(s, insn | (imm14 & 0x3fff) << 5 | rt);
698}
699
700static void tcg_out_insn_3206(TCGContext *s, AArch64Insn insn, int imm26)
701{
702    tcg_out32(s, insn | (imm26 & 0x03ffffff));
703}
704
705static void tcg_out_insn_3207(TCGContext *s, AArch64Insn insn, TCGReg rn)
706{
707    tcg_out32(s, insn | rn << 5);
708}
709
710static void tcg_out_insn_3314(TCGContext *s, AArch64Insn insn,
711                              TCGReg r1, TCGReg r2, TCGReg rn,
712                              tcg_target_long ofs, bool pre, bool w)
713{
714    insn |= 1u << 31; /* ext */
715    insn |= pre << 24;
716    insn |= w << 23;
717
718    tcg_debug_assert(ofs >= -0x200 && ofs < 0x200 && (ofs & 7) == 0);
719    insn |= (ofs & (0x7f << 3)) << (15 - 3);
720
721    tcg_out32(s, insn | r2 << 10 | rn << 5 | r1);
722}
723
724static void tcg_out_insn_3401(TCGContext *s, AArch64Insn insn, TCGType ext,
725                              TCGReg rd, TCGReg rn, uint64_t aimm)
726{
727    if (aimm > 0xfff) {
728        tcg_debug_assert((aimm & 0xfff) == 0);
729        aimm >>= 12;
730        tcg_debug_assert(aimm <= 0xfff);
731        aimm |= 1 << 12;  /* apply LSL 12 */
732    }
733    tcg_out32(s, insn | ext << 31 | aimm << 10 | rn << 5 | rd);
734}
735
736/* This function can be used for both 3.4.2 (Bitfield) and 3.4.4
737   (Logical immediate).  Both insn groups have N, IMMR and IMMS fields
738   that feed the DecodeBitMasks pseudo function.  */
739static void tcg_out_insn_3402(TCGContext *s, AArch64Insn insn, TCGType ext,
740                              TCGReg rd, TCGReg rn, int n, int immr, int imms)
741{
742    tcg_out32(s, insn | ext << 31 | n << 22 | immr << 16 | imms << 10
743              | rn << 5 | rd);
744}
745
746#define tcg_out_insn_3404  tcg_out_insn_3402
747
748static void tcg_out_insn_3403(TCGContext *s, AArch64Insn insn, TCGType ext,
749                              TCGReg rd, TCGReg rn, TCGReg rm, int imms)
750{
751    tcg_out32(s, insn | ext << 31 | ext << 22 | rm << 16 | imms << 10
752              | rn << 5 | rd);
753}
754
755/* This function is used for the Move (wide immediate) instruction group.
756   Note that SHIFT is a full shift count, not the 2 bit HW field. */
757static void tcg_out_insn_3405(TCGContext *s, AArch64Insn insn, TCGType ext,
758                              TCGReg rd, uint16_t half, unsigned shift)
759{
760    tcg_debug_assert((shift & ~0x30) == 0);
761    tcg_out32(s, insn | ext << 31 | shift << (21 - 4) | half << 5 | rd);
762}
763
764static void tcg_out_insn_3406(TCGContext *s, AArch64Insn insn,
765                              TCGReg rd, int64_t disp)
766{
767    tcg_out32(s, insn | (disp & 3) << 29 | (disp & 0x1ffffc) << (5 - 2) | rd);
768}
769
770static inline void tcg_out_insn_3501(TCGContext *s, AArch64Insn insn,
771                                     TCGType sf, TCGReg rd, TCGReg rn,
772                                     TCGReg rm, int opt, int imm3)
773{
774    tcg_out32(s, insn | sf << 31 | rm << 16 | opt << 13 |
775              imm3 << 10 | rn << 5 | rd);
776}
777
778/* This function is for both 3.5.2 (Add/Subtract shifted register), for
779   the rare occasion when we actually want to supply a shift amount.  */
780static inline void tcg_out_insn_3502S(TCGContext *s, AArch64Insn insn,
781                                      TCGType ext, TCGReg rd, TCGReg rn,
782                                      TCGReg rm, int imm6)
783{
784    tcg_out32(s, insn | ext << 31 | rm << 16 | imm6 << 10 | rn << 5 | rd);
785}
786
787/* This function is for 3.5.2 (Add/subtract shifted register),
788   and 3.5.10 (Logical shifted register), for the vast majorty of cases
789   when we don't want to apply a shift.  Thus it can also be used for
790   3.5.3 (Add/subtract with carry) and 3.5.8 (Data processing 2 source).  */
791static void tcg_out_insn_3502(TCGContext *s, AArch64Insn insn, TCGType ext,
792                              TCGReg rd, TCGReg rn, TCGReg rm)
793{
794    tcg_out32(s, insn | ext << 31 | rm << 16 | rn << 5 | rd);
795}
796
797#define tcg_out_insn_3503  tcg_out_insn_3502
798#define tcg_out_insn_3508  tcg_out_insn_3502
799#define tcg_out_insn_3510  tcg_out_insn_3502
800
801static void tcg_out_insn_3506(TCGContext *s, AArch64Insn insn, TCGType ext,
802                              TCGReg rd, TCGReg rn, TCGReg rm, TCGCond c)
803{
804    tcg_out32(s, insn | ext << 31 | rm << 16 | rn << 5 | rd
805              | tcg_cond_to_aarch64[c] << 12);
806}
807
808static void tcg_out_insn_3507(TCGContext *s, AArch64Insn insn, TCGType ext,
809                              TCGReg rd, TCGReg rn)
810{
811    tcg_out32(s, insn | ext << 31 | rn << 5 | rd);
812}
813
814static void tcg_out_insn_3509(TCGContext *s, AArch64Insn insn, TCGType ext,
815                              TCGReg rd, TCGReg rn, TCGReg rm, TCGReg ra)
816{
817    tcg_out32(s, insn | ext << 31 | rm << 16 | ra << 10 | rn << 5 | rd);
818}
819
820static void tcg_out_insn_3605(TCGContext *s, AArch64Insn insn, bool q,
821                              TCGReg rd, TCGReg rn, int dst_idx, int src_idx)
822{
823    /* Note that bit 11 set means general register input.  Therefore
824       we can handle both register sets with one function.  */
825    tcg_out32(s, insn | q << 30 | (dst_idx << 16) | (src_idx << 11)
826              | (rd & 0x1f) | (~rn & 0x20) << 6 | (rn & 0x1f) << 5);
827}
828
829static void tcg_out_insn_3606(TCGContext *s, AArch64Insn insn, bool q,
830                              TCGReg rd, bool op, int cmode, uint8_t imm8)
831{
832    tcg_out32(s, insn | q << 30 | op << 29 | cmode << 12 | (rd & 0x1f)
833              | (imm8 & 0xe0) << (16 - 5) | (imm8 & 0x1f) << 5);
834}
835
836static void tcg_out_insn_3609(TCGContext *s, AArch64Insn insn,
837                              TCGReg rd, TCGReg rn, unsigned immhb)
838{
839    tcg_out32(s, insn | immhb << 16 | (rn & 0x1f) << 5 | (rd & 0x1f));
840}
841
842static void tcg_out_insn_3611(TCGContext *s, AArch64Insn insn,
843                              unsigned size, TCGReg rd, TCGReg rn, TCGReg rm)
844{
845    tcg_out32(s, insn | (size << 22) | (rm & 0x1f) << 16
846              | (rn & 0x1f) << 5 | (rd & 0x1f));
847}
848
849static void tcg_out_insn_3612(TCGContext *s, AArch64Insn insn,
850                              unsigned size, TCGReg rd, TCGReg rn)
851{
852    tcg_out32(s, insn | (size << 22) | (rn & 0x1f) << 5 | (rd & 0x1f));
853}
854
855static void tcg_out_insn_3614(TCGContext *s, AArch64Insn insn, bool q,
856                              TCGReg rd, TCGReg rn, unsigned immhb)
857{
858    tcg_out32(s, insn | q << 30 | immhb << 16
859              | (rn & 0x1f) << 5 | (rd & 0x1f));
860}
861
862static void tcg_out_insn_3616(TCGContext *s, AArch64Insn insn, bool q,
863                              unsigned size, TCGReg rd, TCGReg rn, TCGReg rm)
864{
865    tcg_out32(s, insn | q << 30 | (size << 22) | (rm & 0x1f) << 16
866              | (rn & 0x1f) << 5 | (rd & 0x1f));
867}
868
869static void tcg_out_insn_3617(TCGContext *s, AArch64Insn insn, bool q,
870                              unsigned size, TCGReg rd, TCGReg rn)
871{
872    tcg_out32(s, insn | q << 30 | (size << 22)
873              | (rn & 0x1f) << 5 | (rd & 0x1f));
874}
875
876static void tcg_out_insn_3310(TCGContext *s, AArch64Insn insn,
877                              TCGReg rd, TCGReg base, TCGType ext,
878                              TCGReg regoff)
879{
880    /* Note the AArch64Insn constants above are for C3.3.12.  Adjust.  */
881    tcg_out32(s, insn | I3312_TO_I3310 | regoff << 16 |
882              0x4000 | ext << 13 | base << 5 | (rd & 0x1f));
883}
884
885static void tcg_out_insn_3312(TCGContext *s, AArch64Insn insn,
886                              TCGReg rd, TCGReg rn, intptr_t offset)
887{
888    tcg_out32(s, insn | (offset & 0x1ff) << 12 | rn << 5 | (rd & 0x1f));
889}
890
891static void tcg_out_insn_3313(TCGContext *s, AArch64Insn insn,
892                              TCGReg rd, TCGReg rn, uintptr_t scaled_uimm)
893{
894    /* Note the AArch64Insn constants above are for C3.3.12.  Adjust.  */
895    tcg_out32(s, insn | I3312_TO_I3313 | scaled_uimm << 10
896              | rn << 5 | (rd & 0x1f));
897}
898
899static void tcg_out_bti(TCGContext *s, AArch64Insn insn)
900{
901    /*
902     * While BTI insns are nops on hosts without FEAT_BTI,
903     * there is no point in emitting them in that case either.
904     */
905    if (cpuinfo & CPUINFO_BTI) {
906        tcg_out32(s, insn);
907    }
908}
909
910/* Register to register move using ORR (shifted register with no shift). */
911static void tcg_out_movr(TCGContext *s, TCGType ext, TCGReg rd, TCGReg rm)
912{
913    tcg_out_insn(s, 3510, ORR, ext, rd, TCG_REG_XZR, rm);
914}
915
916/* Register to register move using ADDI (move to/from SP).  */
917static void tcg_out_movr_sp(TCGContext *s, TCGType ext, TCGReg rd, TCGReg rn)
918{
919    tcg_out_insn(s, 3401, ADDI, ext, rd, rn, 0);
920}
921
922/* This function is used for the Logical (immediate) instruction group.
923   The value of LIMM must satisfy IS_LIMM.  See the comment above about
924   only supporting simplified logical immediates.  */
925static void tcg_out_logicali(TCGContext *s, AArch64Insn insn, TCGType ext,
926                             TCGReg rd, TCGReg rn, uint64_t limm)
927{
928    unsigned h, l, r, c;
929
930    tcg_debug_assert(is_limm(limm));
931
932    h = clz64(limm);
933    l = ctz64(limm);
934    if (l == 0) {
935        r = 0;                  /* form 0....01....1 */
936        c = ctz64(~limm) - 1;
937        if (h == 0) {
938            r = clz64(~limm);   /* form 1..10..01..1 */
939            c += r;
940        }
941    } else {
942        r = 64 - l;             /* form 1....10....0 or 0..01..10..0 */
943        c = r - h - 1;
944    }
945    if (ext == TCG_TYPE_I32) {
946        r &= 31;
947        c &= 31;
948    }
949
950    tcg_out_insn_3404(s, insn, ext, rd, rn, ext, r, c);
951}
952
953static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
954                             TCGReg rd, int64_t v64)
955{
956    bool q = type == TCG_TYPE_V128;
957    int cmode, imm8, i;
958
959    /* Test all bytes equal first.  */
960    if (vece == MO_8) {
961        imm8 = (uint8_t)v64;
962        tcg_out_insn(s, 3606, MOVI, q, rd, 0, 0xe, imm8);
963        return;
964    }
965
966    /*
967     * Test all bytes 0x00 or 0xff second.  This can match cases that
968     * might otherwise take 2 or 3 insns for MO_16 or MO_32 below.
969     */
970    for (i = imm8 = 0; i < 8; i++) {
971        uint8_t byte = v64 >> (i * 8);
972        if (byte == 0xff) {
973            imm8 |= 1 << i;
974        } else if (byte != 0) {
975            goto fail_bytes;
976        }
977    }
978    tcg_out_insn(s, 3606, MOVI, q, rd, 1, 0xe, imm8);
979    return;
980 fail_bytes:
981
982    /*
983     * Tests for various replications.  For each element width, if we
984     * cannot find an expansion there's no point checking a larger
985     * width because we already know by replication it cannot match.
986     */
987    if (vece == MO_16) {
988        uint16_t v16 = v64;
989
990        if (is_shimm16(v16, &cmode, &imm8)) {
991            tcg_out_insn(s, 3606, MOVI, q, rd, 0, cmode, imm8);
992            return;
993        }
994        if (is_shimm16(~v16, &cmode, &imm8)) {
995            tcg_out_insn(s, 3606, MVNI, q, rd, 0, cmode, imm8);
996            return;
997        }
998
999        /*
1000         * Otherwise, all remaining constants can be loaded in two insns:
1001         * rd = v16 & 0xff, rd |= v16 & 0xff00.
1002         */
1003        tcg_out_insn(s, 3606, MOVI, q, rd, 0, 0x8, v16 & 0xff);
1004        tcg_out_insn(s, 3606, ORR, q, rd, 0, 0xa, v16 >> 8);
1005        return;
1006    } else if (vece == MO_32) {
1007        uint32_t v32 = v64;
1008        uint32_t n32 = ~v32;
1009
1010        if (is_shimm32(v32, &cmode, &imm8) ||
1011            is_soimm32(v32, &cmode, &imm8) ||
1012            is_fimm32(v32, &cmode, &imm8)) {
1013            tcg_out_insn(s, 3606, MOVI, q, rd, 0, cmode, imm8);
1014            return;
1015        }
1016        if (is_shimm32(n32, &cmode, &imm8) ||
1017            is_soimm32(n32, &cmode, &imm8)) {
1018            tcg_out_insn(s, 3606, MVNI, q, rd, 0, cmode, imm8);
1019            return;
1020        }
1021
1022        /*
1023         * Restrict the set of constants to those we can load with
1024         * two instructions.  Others we load from the pool.
1025         */
1026        i = is_shimm32_pair(v32, &cmode, &imm8);
1027        if (i) {
1028            tcg_out_insn(s, 3606, MOVI, q, rd, 0, cmode, imm8);
1029            tcg_out_insn(s, 3606, ORR, q, rd, 0, i, extract32(v32, i * 4, 8));
1030            return;
1031        }
1032        i = is_shimm32_pair(n32, &cmode, &imm8);
1033        if (i) {
1034            tcg_out_insn(s, 3606, MVNI, q, rd, 0, cmode, imm8);
1035            tcg_out_insn(s, 3606, BIC, q, rd, 0, i, extract32(n32, i * 4, 8));
1036            return;
1037        }
1038    } else if (is_fimm64(v64, &cmode, &imm8)) {
1039        tcg_out_insn(s, 3606, MOVI, q, rd, 1, cmode, imm8);
1040        return;
1041    }
1042
1043    /*
1044     * As a last resort, load from the constant pool.  Sadly there
1045     * is no LD1R (literal), so store the full 16-byte vector.
1046     */
1047    if (type == TCG_TYPE_V128) {
1048        new_pool_l2(s, R_AARCH64_CONDBR19, s->code_ptr, 0, v64, v64);
1049        tcg_out_insn(s, 3305, LDR_v128, 0, rd);
1050    } else {
1051        new_pool_label(s, v64, R_AARCH64_CONDBR19, s->code_ptr, 0);
1052        tcg_out_insn(s, 3305, LDR_v64, 0, rd);
1053    }
1054}
1055
1056static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
1057                            TCGReg rd, TCGReg rs)
1058{
1059    int is_q = type - TCG_TYPE_V64;
1060    tcg_out_insn(s, 3605, DUP, is_q, rd, rs, 1 << vece, 0);
1061    return true;
1062}
1063
1064static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
1065                             TCGReg r, TCGReg base, intptr_t offset)
1066{
1067    TCGReg temp = TCG_REG_TMP0;
1068
1069    if (offset < -0xffffff || offset > 0xffffff) {
1070        tcg_out_movi(s, TCG_TYPE_PTR, temp, offset);
1071        tcg_out_insn(s, 3502, ADD, 1, temp, temp, base);
1072        base = temp;
1073    } else {
1074        AArch64Insn add_insn = I3401_ADDI;
1075
1076        if (offset < 0) {
1077            add_insn = I3401_SUBI;
1078            offset = -offset;
1079        }
1080        if (offset & 0xfff000) {
1081            tcg_out_insn_3401(s, add_insn, 1, temp, base, offset & 0xfff000);
1082            base = temp;
1083        }
1084        if (offset & 0xfff) {
1085            tcg_out_insn_3401(s, add_insn, 1, temp, base, offset & 0xfff);
1086            base = temp;
1087        }
1088    }
1089    tcg_out_insn(s, 3303, LD1R, type == TCG_TYPE_V128, r, base, vece);
1090    return true;
1091}
1092
1093static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd,
1094                         tcg_target_long value)
1095{
1096    tcg_target_long svalue = value;
1097    tcg_target_long ivalue = ~value;
1098    tcg_target_long t0, t1, t2;
1099    int s0, s1;
1100    AArch64Insn opc;
1101
1102    switch (type) {
1103    case TCG_TYPE_I32:
1104    case TCG_TYPE_I64:
1105        tcg_debug_assert(rd < 32);
1106        break;
1107    default:
1108        g_assert_not_reached();
1109    }
1110
1111    /* For 32-bit values, discard potential garbage in value.  For 64-bit
1112       values within [2**31, 2**32-1], we can create smaller sequences by
1113       interpreting this as a negative 32-bit number, while ensuring that
1114       the high 32 bits are cleared by setting SF=0.  */
1115    if (type == TCG_TYPE_I32 || (value & ~0xffffffffull) == 0) {
1116        svalue = (int32_t)value;
1117        value = (uint32_t)value;
1118        ivalue = (uint32_t)ivalue;
1119        type = TCG_TYPE_I32;
1120    }
1121
1122    /* Speed things up by handling the common case of small positive
1123       and negative values specially.  */
1124    if ((value & ~0xffffull) == 0) {
1125        tcg_out_insn(s, 3405, MOVZ, type, rd, value, 0);
1126        return;
1127    } else if ((ivalue & ~0xffffull) == 0) {
1128        tcg_out_insn(s, 3405, MOVN, type, rd, ivalue, 0);
1129        return;
1130    }
1131
1132    /* Check for bitfield immediates.  For the benefit of 32-bit quantities,
1133       use the sign-extended value.  That lets us match rotated values such
1134       as 0xff0000ff with the same 64-bit logic matching 0xffffffffff0000ff. */
1135    if (is_limm(svalue)) {
1136        tcg_out_logicali(s, I3404_ORRI, type, rd, TCG_REG_XZR, svalue);
1137        return;
1138    }
1139
1140    /* Look for host pointer values within 4G of the PC.  This happens
1141       often when loading pointers to QEMU's own data structures.  */
1142    if (type == TCG_TYPE_I64) {
1143        intptr_t src_rx = (intptr_t)tcg_splitwx_to_rx(s->code_ptr);
1144        tcg_target_long disp = value - src_rx;
1145        if (disp == sextract64(disp, 0, 21)) {
1146            tcg_out_insn(s, 3406, ADR, rd, disp);
1147            return;
1148        }
1149        disp = (value >> 12) - (src_rx >> 12);
1150        if (disp == sextract64(disp, 0, 21)) {
1151            tcg_out_insn(s, 3406, ADRP, rd, disp);
1152            if (value & 0xfff) {
1153                tcg_out_insn(s, 3401, ADDI, type, rd, rd, value & 0xfff);
1154            }
1155            return;
1156        }
1157    }
1158
1159    /* Would it take fewer insns to begin with MOVN?  */
1160    if (ctpop64(value) >= 32) {
1161        t0 = ivalue;
1162        opc = I3405_MOVN;
1163    } else {
1164        t0 = value;
1165        opc = I3405_MOVZ;
1166    }
1167    s0 = ctz64(t0) & (63 & -16);
1168    t1 = t0 & ~(0xffffull << s0);
1169    s1 = ctz64(t1) & (63 & -16);
1170    t2 = t1 & ~(0xffffull << s1);
1171    if (t2 == 0) {
1172        tcg_out_insn_3405(s, opc, type, rd, t0 >> s0, s0);
1173        if (t1 != 0) {
1174            tcg_out_insn(s, 3405, MOVK, type, rd, value >> s1, s1);
1175        }
1176        return;
1177    }
1178
1179    /* For more than 2 insns, dump it into the constant pool.  */
1180    new_pool_label(s, value, R_AARCH64_CONDBR19, s->code_ptr, 0);
1181    tcg_out_insn(s, 3305, LDR, 0, rd);
1182}
1183
1184static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
1185{
1186    return false;
1187}
1188
1189static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
1190                             tcg_target_long imm)
1191{
1192    /* This function is only used for passing structs by reference. */
1193    g_assert_not_reached();
1194}
1195
1196/* Define something more legible for general use.  */
1197#define tcg_out_ldst_r  tcg_out_insn_3310
1198
1199static void tcg_out_ldst(TCGContext *s, AArch64Insn insn, TCGReg rd,
1200                         TCGReg rn, intptr_t offset, int lgsize)
1201{
1202    /* If the offset is naturally aligned and in range, then we can
1203       use the scaled uimm12 encoding */
1204    if (offset >= 0 && !(offset & ((1 << lgsize) - 1))) {
1205        uintptr_t scaled_uimm = offset >> lgsize;
1206        if (scaled_uimm <= 0xfff) {
1207            tcg_out_insn_3313(s, insn, rd, rn, scaled_uimm);
1208            return;
1209        }
1210    }
1211
1212    /* Small signed offsets can use the unscaled encoding.  */
1213    if (offset >= -256 && offset < 256) {
1214        tcg_out_insn_3312(s, insn, rd, rn, offset);
1215        return;
1216    }
1217
1218    /* Worst-case scenario, move offset to temp register, use reg offset.  */
1219    tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_TMP0, offset);
1220    tcg_out_ldst_r(s, insn, rd, rn, TCG_TYPE_I64, TCG_REG_TMP0);
1221}
1222
1223static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
1224{
1225    if (ret == arg) {
1226        return true;
1227    }
1228    switch (type) {
1229    case TCG_TYPE_I32:
1230    case TCG_TYPE_I64:
1231        if (ret < 32 && arg < 32) {
1232            tcg_out_movr(s, type, ret, arg);
1233            break;
1234        } else if (ret < 32) {
1235            tcg_out_insn(s, 3605, UMOV, type, ret, arg, 0, 0);
1236            break;
1237        } else if (arg < 32) {
1238            tcg_out_insn(s, 3605, INS, 0, ret, arg, 4 << type, 0);
1239            break;
1240        }
1241        /* FALLTHRU */
1242
1243    case TCG_TYPE_V64:
1244        tcg_debug_assert(ret >= 32 && arg >= 32);
1245        tcg_out_insn(s, 3616, ORR, 0, 0, ret, arg, arg);
1246        break;
1247    case TCG_TYPE_V128:
1248        tcg_debug_assert(ret >= 32 && arg >= 32);
1249        tcg_out_insn(s, 3616, ORR, 1, 0, ret, arg, arg);
1250        break;
1251
1252    default:
1253        g_assert_not_reached();
1254    }
1255    return true;
1256}
1257
1258static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
1259                       TCGReg base, intptr_t ofs)
1260{
1261    AArch64Insn insn;
1262    int lgsz;
1263
1264    switch (type) {
1265    case TCG_TYPE_I32:
1266        insn = (ret < 32 ? I3312_LDRW : I3312_LDRVS);
1267        lgsz = 2;
1268        break;
1269    case TCG_TYPE_I64:
1270        insn = (ret < 32 ? I3312_LDRX : I3312_LDRVD);
1271        lgsz = 3;
1272        break;
1273    case TCG_TYPE_V64:
1274        insn = I3312_LDRVD;
1275        lgsz = 3;
1276        break;
1277    case TCG_TYPE_V128:
1278        insn = I3312_LDRVQ;
1279        lgsz = 4;
1280        break;
1281    default:
1282        g_assert_not_reached();
1283    }
1284    tcg_out_ldst(s, insn, ret, base, ofs, lgsz);
1285}
1286
1287static void tcg_out_st(TCGContext *s, TCGType type, TCGReg src,
1288                       TCGReg base, intptr_t ofs)
1289{
1290    AArch64Insn insn;
1291    int lgsz;
1292
1293    switch (type) {
1294    case TCG_TYPE_I32:
1295        insn = (src < 32 ? I3312_STRW : I3312_STRVS);
1296        lgsz = 2;
1297        break;
1298    case TCG_TYPE_I64:
1299        insn = (src < 32 ? I3312_STRX : I3312_STRVD);
1300        lgsz = 3;
1301        break;
1302    case TCG_TYPE_V64:
1303        insn = I3312_STRVD;
1304        lgsz = 3;
1305        break;
1306    case TCG_TYPE_V128:
1307        insn = I3312_STRVQ;
1308        lgsz = 4;
1309        break;
1310    default:
1311        g_assert_not_reached();
1312    }
1313    tcg_out_ldst(s, insn, src, base, ofs, lgsz);
1314}
1315
1316static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1317                               TCGReg base, intptr_t ofs)
1318{
1319    if (type <= TCG_TYPE_I64 && val == 0) {
1320        tcg_out_st(s, type, TCG_REG_XZR, base, ofs);
1321        return true;
1322    }
1323    return false;
1324}
1325
1326static inline void tcg_out_bfm(TCGContext *s, TCGType ext, TCGReg rd,
1327                               TCGReg rn, unsigned int a, unsigned int b)
1328{
1329    tcg_out_insn(s, 3402, BFM, ext, rd, rn, ext, a, b);
1330}
1331
1332static inline void tcg_out_ubfm(TCGContext *s, TCGType ext, TCGReg rd,
1333                                TCGReg rn, unsigned int a, unsigned int b)
1334{
1335    tcg_out_insn(s, 3402, UBFM, ext, rd, rn, ext, a, b);
1336}
1337
1338static inline void tcg_out_sbfm(TCGContext *s, TCGType ext, TCGReg rd,
1339                                TCGReg rn, unsigned int a, unsigned int b)
1340{
1341    tcg_out_insn(s, 3402, SBFM, ext, rd, rn, ext, a, b);
1342}
1343
1344static inline void tcg_out_extr(TCGContext *s, TCGType ext, TCGReg rd,
1345                                TCGReg rn, TCGReg rm, unsigned int a)
1346{
1347    tcg_out_insn(s, 3403, EXTR, ext, rd, rn, rm, a);
1348}
1349
1350static inline void tcg_out_dep(TCGContext *s, TCGType ext, TCGReg rd,
1351                               TCGReg rn, unsigned lsb, unsigned width)
1352{
1353    unsigned size = ext ? 64 : 32;
1354    unsigned a = (size - lsb) & (size - 1);
1355    unsigned b = width - 1;
1356    tcg_out_bfm(s, ext, rd, rn, a, b);
1357}
1358
1359static void tcg_out_cmp(TCGContext *s, TCGType ext, TCGCond cond, TCGReg a,
1360                        tcg_target_long b, bool const_b)
1361{
1362    if (is_tst_cond(cond)) {
1363        if (!const_b) {
1364            tcg_out_insn(s, 3510, ANDS, ext, TCG_REG_XZR, a, b);
1365        } else {
1366            tcg_out_logicali(s, I3404_ANDSI, ext, TCG_REG_XZR, a, b);
1367        }
1368    } else {
1369        if (!const_b) {
1370            tcg_out_insn(s, 3502, SUBS, ext, TCG_REG_XZR, a, b);
1371        } else if (b >= 0) {
1372            tcg_debug_assert(is_aimm(b));
1373            tcg_out_insn(s, 3401, SUBSI, ext, TCG_REG_XZR, a, b);
1374        } else {
1375            tcg_debug_assert(is_aimm(-b));
1376            tcg_out_insn(s, 3401, ADDSI, ext, TCG_REG_XZR, a, -b);
1377        }
1378    }
1379}
1380
1381static void tcg_out_goto(TCGContext *s, const tcg_insn_unit *target)
1382{
1383    ptrdiff_t offset = tcg_pcrel_diff(s, target) >> 2;
1384    tcg_debug_assert(offset == sextract64(offset, 0, 26));
1385    tcg_out_insn(s, 3206, B, offset);
1386}
1387
1388static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *target)
1389{
1390    ptrdiff_t offset = tcg_pcrel_diff(s, target) >> 2;
1391    if (offset == sextract64(offset, 0, 26)) {
1392        tcg_out_insn(s, 3206, BL, offset);
1393    } else {
1394        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_TMP0, (intptr_t)target);
1395        tcg_out_insn(s, 3207, BLR, TCG_REG_TMP0);
1396    }
1397}
1398
1399static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target,
1400                         const TCGHelperInfo *info)
1401{
1402    tcg_out_call_int(s, target);
1403}
1404
1405static inline void tcg_out_goto_label(TCGContext *s, TCGLabel *l)
1406{
1407    if (!l->has_value) {
1408        tcg_out_reloc(s, s->code_ptr, R_AARCH64_JUMP26, l, 0);
1409        tcg_out_insn(s, 3206, B, 0);
1410    } else {
1411        tcg_out_goto(s, l->u.value_ptr);
1412    }
1413}
1414
1415static void tcg_out_brcond(TCGContext *s, TCGType ext, TCGCond c, TCGArg a,
1416                           TCGArg b, bool b_const, TCGLabel *l)
1417{
1418    int tbit = -1;
1419    bool need_cmp = true;
1420
1421    switch (c) {
1422    case TCG_COND_EQ:
1423    case TCG_COND_NE:
1424        /* cmp xN,0; b.ne L -> cbnz xN,L */
1425        if (b_const && b == 0) {
1426            need_cmp = false;
1427        }
1428        break;
1429    case TCG_COND_LT:
1430    case TCG_COND_GE:
1431        /* cmp xN,0; b.mi L -> tbnz xN,63,L */
1432        if (b_const && b == 0) {
1433            c = (c == TCG_COND_LT ? TCG_COND_TSTNE : TCG_COND_TSTEQ);
1434            tbit = ext ? 63 : 31;
1435            need_cmp = false;
1436        }
1437        break;
1438    case TCG_COND_TSTEQ:
1439    case TCG_COND_TSTNE:
1440        /* tst xN,0xffffffff; b.ne L -> cbnz wN,L */
1441        if (b_const && b == UINT32_MAX) {
1442            c = tcg_tst_eqne_cond(c);
1443            ext = TCG_TYPE_I32;
1444            need_cmp = false;
1445            break;
1446        }
1447        /* tst xN,1<<B; b.ne L -> tbnz xN,B,L */
1448        if (b_const && is_power_of_2(b)) {
1449            tbit = ctz64(b);
1450            need_cmp = false;
1451        }
1452        break;
1453    default:
1454        break;
1455    }
1456
1457    if (need_cmp) {
1458        tcg_out_cmp(s, ext, c, a, b, b_const);
1459        tcg_out_reloc(s, s->code_ptr, R_AARCH64_CONDBR19, l, 0);
1460        tcg_out_insn(s, 3202, B_C, c, 0);
1461        return;
1462    }
1463
1464    if (tbit >= 0) {
1465        tcg_out_reloc(s, s->code_ptr, R_AARCH64_TSTBR14, l, 0);
1466        switch (c) {
1467        case TCG_COND_TSTEQ:
1468            tcg_out_insn(s, 3205, TBZ, a, tbit, 0);
1469            break;
1470        case TCG_COND_TSTNE:
1471            tcg_out_insn(s, 3205, TBNZ, a, tbit, 0);
1472            break;
1473        default:
1474            g_assert_not_reached();
1475        }
1476    } else {
1477        tcg_out_reloc(s, s->code_ptr, R_AARCH64_CONDBR19, l, 0);
1478        switch (c) {
1479        case TCG_COND_EQ:
1480            tcg_out_insn(s, 3201, CBZ, ext, a, 0);
1481            break;
1482        case TCG_COND_NE:
1483            tcg_out_insn(s, 3201, CBNZ, ext, a, 0);
1484            break;
1485        default:
1486            g_assert_not_reached();
1487        }
1488    }
1489}
1490
1491static inline void tcg_out_rev(TCGContext *s, int ext, MemOp s_bits,
1492                               TCGReg rd, TCGReg rn)
1493{
1494    /* REV, REV16, REV32 */
1495    tcg_out_insn_3507(s, I3507_REV | (s_bits << 10), ext, rd, rn);
1496}
1497
1498static inline void tcg_out_sxt(TCGContext *s, TCGType ext, MemOp s_bits,
1499                               TCGReg rd, TCGReg rn)
1500{
1501    /* Using ALIASes SXTB, SXTH, SXTW, of SBFM Xd, Xn, #0, #7|15|31 */
1502    int bits = (8 << s_bits) - 1;
1503    tcg_out_sbfm(s, ext, rd, rn, 0, bits);
1504}
1505
1506static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rn)
1507{
1508    tcg_out_sxt(s, type, MO_8, rd, rn);
1509}
1510
1511static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rn)
1512{
1513    tcg_out_sxt(s, type, MO_16, rd, rn);
1514}
1515
1516static void tcg_out_ext32s(TCGContext *s, TCGReg rd, TCGReg rn)
1517{
1518    tcg_out_sxt(s, TCG_TYPE_I64, MO_32, rd, rn);
1519}
1520
1521static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg rd, TCGReg rn)
1522{
1523    tcg_out_ext32s(s, rd, rn);
1524}
1525
1526static inline void tcg_out_uxt(TCGContext *s, MemOp s_bits,
1527                               TCGReg rd, TCGReg rn)
1528{
1529    /* Using ALIASes UXTB, UXTH of UBFM Wd, Wn, #0, #7|15 */
1530    int bits = (8 << s_bits) - 1;
1531    tcg_out_ubfm(s, 0, rd, rn, 0, bits);
1532}
1533
1534static void tcg_out_ext8u(TCGContext *s, TCGReg rd, TCGReg rn)
1535{
1536    tcg_out_uxt(s, MO_8, rd, rn);
1537}
1538
1539static void tcg_out_ext16u(TCGContext *s, TCGReg rd, TCGReg rn)
1540{
1541    tcg_out_uxt(s, MO_16, rd, rn);
1542}
1543
1544static void tcg_out_ext32u(TCGContext *s, TCGReg rd, TCGReg rn)
1545{
1546    tcg_out_movr(s, TCG_TYPE_I32, rd, rn);
1547}
1548
1549static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg rd, TCGReg rn)
1550{
1551    tcg_out_ext32u(s, rd, rn);
1552}
1553
1554static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rn)
1555{
1556    tcg_out_mov(s, TCG_TYPE_I32, rd, rn);
1557}
1558
1559static void tcg_out_addsub2(TCGContext *s, TCGType ext, TCGReg rl,
1560                            TCGReg rh, TCGReg al, TCGReg ah,
1561                            tcg_target_long bl, tcg_target_long bh,
1562                            bool const_bl, bool const_bh, bool sub)
1563{
1564    TCGReg orig_rl = rl;
1565    AArch64Insn insn;
1566
1567    if (rl == ah || (!const_bh && rl == bh)) {
1568        rl = TCG_REG_TMP0;
1569    }
1570
1571    if (const_bl) {
1572        if (bl < 0) {
1573            bl = -bl;
1574            insn = sub ? I3401_ADDSI : I3401_SUBSI;
1575        } else {
1576            insn = sub ? I3401_SUBSI : I3401_ADDSI;
1577        }
1578
1579        if (unlikely(al == TCG_REG_XZR)) {
1580            /* ??? We want to allow al to be zero for the benefit of
1581               negation via subtraction.  However, that leaves open the
1582               possibility of adding 0+const in the low part, and the
1583               immediate add instructions encode XSP not XZR.  Don't try
1584               anything more elaborate here than loading another zero.  */
1585            al = TCG_REG_TMP0;
1586            tcg_out_movi(s, ext, al, 0);
1587        }
1588        tcg_out_insn_3401(s, insn, ext, rl, al, bl);
1589    } else {
1590        tcg_out_insn_3502(s, sub ? I3502_SUBS : I3502_ADDS, ext, rl, al, bl);
1591    }
1592
1593    insn = I3503_ADC;
1594    if (const_bh) {
1595        /* Note that the only two constants we support are 0 and -1, and
1596           that SBC = rn + ~rm + c, so adc -1 is sbc 0, and vice-versa.  */
1597        if ((bh != 0) ^ sub) {
1598            insn = I3503_SBC;
1599        }
1600        bh = TCG_REG_XZR;
1601    } else if (sub) {
1602        insn = I3503_SBC;
1603    }
1604    tcg_out_insn_3503(s, insn, ext, rh, ah, bh);
1605
1606    tcg_out_mov(s, ext, orig_rl, rl);
1607}
1608
1609static inline void tcg_out_mb(TCGContext *s, TCGArg a0)
1610{
1611    static const uint32_t sync[] = {
1612        [0 ... TCG_MO_ALL]            = DMB_ISH | DMB_LD | DMB_ST,
1613        [TCG_MO_ST_ST]                = DMB_ISH | DMB_ST,
1614        [TCG_MO_LD_LD]                = DMB_ISH | DMB_LD,
1615        [TCG_MO_LD_ST]                = DMB_ISH | DMB_LD,
1616        [TCG_MO_LD_ST | TCG_MO_LD_LD] = DMB_ISH | DMB_LD,
1617    };
1618    tcg_out32(s, sync[a0 & TCG_MO_ALL]);
1619}
1620
1621typedef struct {
1622    TCGReg base;
1623    TCGReg index;
1624    TCGType index_ext;
1625    TCGAtomAlign aa;
1626} HostAddress;
1627
1628bool tcg_target_has_memory_bswap(MemOp memop)
1629{
1630    return false;
1631}
1632
1633static const TCGLdstHelperParam ldst_helper_param = {
1634    .ntmp = 1, .tmp = { TCG_REG_TMP0 }
1635};
1636
1637static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1638{
1639    MemOp opc = get_memop(lb->oi);
1640
1641    if (!reloc_pc19(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1642        return false;
1643    }
1644
1645    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1646    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1647    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1648    tcg_out_goto(s, lb->raddr);
1649    return true;
1650}
1651
1652static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1653{
1654    MemOp opc = get_memop(lb->oi);
1655
1656    if (!reloc_pc19(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1657        return false;
1658    }
1659
1660    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1661    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE]);
1662    tcg_out_goto(s, lb->raddr);
1663    return true;
1664}
1665
1666/* We expect to use a 7-bit scaled negative offset from ENV.  */
1667#define MIN_TLB_MASK_TABLE_OFS  -512
1668
1669/*
1670 * For system-mode, perform the TLB load and compare.
1671 * For user-mode, perform any required alignment tests.
1672 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1673 * is required and fill in @h with the host address for the fast path.
1674 */
1675static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1676                                           TCGReg addr_reg, MemOpIdx oi,
1677                                           bool is_ld)
1678{
1679    TCGType addr_type = s->addr_type;
1680    TCGLabelQemuLdst *ldst = NULL;
1681    MemOp opc = get_memop(oi);
1682    MemOp s_bits = opc & MO_SIZE;
1683    unsigned a_mask;
1684
1685    h->aa = atom_and_align_for_opc(s, opc,
1686                                   have_lse2 ? MO_ATOM_WITHIN16
1687                                             : MO_ATOM_IFALIGN,
1688                                   s_bits == MO_128);
1689    a_mask = (1 << h->aa.align) - 1;
1690
1691    if (tcg_use_softmmu) {
1692        unsigned s_mask = (1u << s_bits) - 1;
1693        unsigned mem_index = get_mmuidx(oi);
1694        TCGReg addr_adj;
1695        TCGType mask_type;
1696        uint64_t compare_mask;
1697
1698        ldst = new_ldst_label(s);
1699        ldst->is_ld = is_ld;
1700        ldst->oi = oi;
1701        ldst->addr_reg = addr_reg;
1702
1703        mask_type = (s->page_bits + s->tlb_dyn_max_bits > 32
1704                     ? TCG_TYPE_I64 : TCG_TYPE_I32);
1705
1706        /* Load cpu->neg.tlb.f[mmu_idx].{mask,table} into {tmp0,tmp1}. */
1707        QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, mask) != 0);
1708        QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, table) != 8);
1709        tcg_out_insn(s, 3314, LDP, TCG_REG_TMP0, TCG_REG_TMP1, TCG_AREG0,
1710                     tlb_mask_table_ofs(s, mem_index), 1, 0);
1711
1712        /* Extract the TLB index from the address into X0.  */
1713        tcg_out_insn(s, 3502S, AND_LSR, mask_type == TCG_TYPE_I64,
1714                     TCG_REG_TMP0, TCG_REG_TMP0, addr_reg,
1715                     s->page_bits - CPU_TLB_ENTRY_BITS);
1716
1717        /* Add the tlb_table pointer, forming the CPUTLBEntry address. */
1718        tcg_out_insn(s, 3502, ADD, 1, TCG_REG_TMP1, TCG_REG_TMP1, TCG_REG_TMP0);
1719
1720        /* Load the tlb comparator into TMP0, and the fast path addend. */
1721        QEMU_BUILD_BUG_ON(HOST_BIG_ENDIAN);
1722        tcg_out_ld(s, addr_type, TCG_REG_TMP0, TCG_REG_TMP1,
1723                   is_ld ? offsetof(CPUTLBEntry, addr_read)
1724                         : offsetof(CPUTLBEntry, addr_write));
1725        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1,
1726                   offsetof(CPUTLBEntry, addend));
1727
1728        /*
1729         * For aligned accesses, we check the first byte and include
1730         * the alignment bits within the address.  For unaligned access,
1731         * we check that we don't cross pages using the address of the
1732         * last byte of the access.
1733         */
1734        if (a_mask >= s_mask) {
1735            addr_adj = addr_reg;
1736        } else {
1737            addr_adj = TCG_REG_TMP2;
1738            tcg_out_insn(s, 3401, ADDI, addr_type,
1739                         addr_adj, addr_reg, s_mask - a_mask);
1740        }
1741        compare_mask = (uint64_t)s->page_mask | a_mask;
1742
1743        /* Store the page mask part of the address into TMP2.  */
1744        tcg_out_logicali(s, I3404_ANDI, addr_type, TCG_REG_TMP2,
1745                         addr_adj, compare_mask);
1746
1747        /* Perform the address comparison. */
1748        tcg_out_cmp(s, addr_type, TCG_COND_NE, TCG_REG_TMP0, TCG_REG_TMP2, 0);
1749
1750        /* If not equal, we jump to the slow path. */
1751        ldst->label_ptr[0] = s->code_ptr;
1752        tcg_out_insn(s, 3202, B_C, TCG_COND_NE, 0);
1753
1754        h->base = TCG_REG_TMP1;
1755        h->index = addr_reg;
1756        h->index_ext = addr_type;
1757    } else {
1758        if (a_mask) {
1759            ldst = new_ldst_label(s);
1760
1761            ldst->is_ld = is_ld;
1762            ldst->oi = oi;
1763            ldst->addr_reg = addr_reg;
1764
1765            /* tst addr, #mask */
1766            tcg_out_logicali(s, I3404_ANDSI, 0, TCG_REG_XZR, addr_reg, a_mask);
1767
1768            /* b.ne slow_path */
1769            ldst->label_ptr[0] = s->code_ptr;
1770            tcg_out_insn(s, 3202, B_C, TCG_COND_NE, 0);
1771        }
1772
1773        if (guest_base || addr_type == TCG_TYPE_I32) {
1774            h->base = TCG_REG_GUEST_BASE;
1775            h->index = addr_reg;
1776            h->index_ext = addr_type;
1777        } else {
1778            h->base = addr_reg;
1779            h->index = TCG_REG_XZR;
1780            h->index_ext = TCG_TYPE_I64;
1781        }
1782    }
1783
1784    return ldst;
1785}
1786
1787static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp memop, TCGType ext,
1788                                   TCGReg data_r, HostAddress h)
1789{
1790    switch (memop & MO_SSIZE) {
1791    case MO_UB:
1792        tcg_out_ldst_r(s, I3312_LDRB, data_r, h.base, h.index_ext, h.index);
1793        break;
1794    case MO_SB:
1795        tcg_out_ldst_r(s, ext ? I3312_LDRSBX : I3312_LDRSBW,
1796                       data_r, h.base, h.index_ext, h.index);
1797        break;
1798    case MO_UW:
1799        tcg_out_ldst_r(s, I3312_LDRH, data_r, h.base, h.index_ext, h.index);
1800        break;
1801    case MO_SW:
1802        tcg_out_ldst_r(s, (ext ? I3312_LDRSHX : I3312_LDRSHW),
1803                       data_r, h.base, h.index_ext, h.index);
1804        break;
1805    case MO_UL:
1806        tcg_out_ldst_r(s, I3312_LDRW, data_r, h.base, h.index_ext, h.index);
1807        break;
1808    case MO_SL:
1809        tcg_out_ldst_r(s, I3312_LDRSWX, data_r, h.base, h.index_ext, h.index);
1810        break;
1811    case MO_UQ:
1812        tcg_out_ldst_r(s, I3312_LDRX, data_r, h.base, h.index_ext, h.index);
1813        break;
1814    default:
1815        g_assert_not_reached();
1816    }
1817}
1818
1819static void tcg_out_qemu_st_direct(TCGContext *s, MemOp memop,
1820                                   TCGReg data_r, HostAddress h)
1821{
1822    switch (memop & MO_SIZE) {
1823    case MO_8:
1824        tcg_out_ldst_r(s, I3312_STRB, data_r, h.base, h.index_ext, h.index);
1825        break;
1826    case MO_16:
1827        tcg_out_ldst_r(s, I3312_STRH, data_r, h.base, h.index_ext, h.index);
1828        break;
1829    case MO_32:
1830        tcg_out_ldst_r(s, I3312_STRW, data_r, h.base, h.index_ext, h.index);
1831        break;
1832    case MO_64:
1833        tcg_out_ldst_r(s, I3312_STRX, data_r, h.base, h.index_ext, h.index);
1834        break;
1835    default:
1836        g_assert_not_reached();
1837    }
1838}
1839
1840static void tcg_out_qemu_ld(TCGContext *s, TCGReg data_reg, TCGReg addr_reg,
1841                            MemOpIdx oi, TCGType data_type)
1842{
1843    TCGLabelQemuLdst *ldst;
1844    HostAddress h;
1845
1846    ldst = prepare_host_addr(s, &h, addr_reg, oi, true);
1847    tcg_out_qemu_ld_direct(s, get_memop(oi), data_type, data_reg, h);
1848
1849    if (ldst) {
1850        ldst->type = data_type;
1851        ldst->datalo_reg = data_reg;
1852        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1853    }
1854}
1855
1856static void tcg_out_qemu_st(TCGContext *s, TCGReg data_reg, TCGReg addr_reg,
1857                            MemOpIdx oi, TCGType data_type)
1858{
1859    TCGLabelQemuLdst *ldst;
1860    HostAddress h;
1861
1862    ldst = prepare_host_addr(s, &h, addr_reg, oi, false);
1863    tcg_out_qemu_st_direct(s, get_memop(oi), data_reg, h);
1864
1865    if (ldst) {
1866        ldst->type = data_type;
1867        ldst->datalo_reg = data_reg;
1868        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1869    }
1870}
1871
1872static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi,
1873                                   TCGReg addr_reg, MemOpIdx oi, bool is_ld)
1874{
1875    TCGLabelQemuLdst *ldst;
1876    HostAddress h;
1877    TCGReg base;
1878    bool use_pair;
1879
1880    ldst = prepare_host_addr(s, &h, addr_reg, oi, is_ld);
1881
1882    /* Compose the final address, as LDP/STP have no indexing. */
1883    if (h.index == TCG_REG_XZR) {
1884        base = h.base;
1885    } else {
1886        base = TCG_REG_TMP2;
1887        if (h.index_ext == TCG_TYPE_I32) {
1888            /* add base, base, index, uxtw */
1889            tcg_out_insn(s, 3501, ADD, TCG_TYPE_I64, base,
1890                         h.base, h.index, MO_32, 0);
1891        } else {
1892            /* add base, base, index */
1893            tcg_out_insn(s, 3502, ADD, 1, base, h.base, h.index);
1894        }
1895    }
1896
1897    use_pair = h.aa.atom < MO_128 || have_lse2;
1898
1899    if (!use_pair) {
1900        tcg_insn_unit *branch = NULL;
1901        TCGReg ll, lh, sl, sh;
1902
1903        /*
1904         * If we have already checked for 16-byte alignment, that's all
1905         * we need. Otherwise we have determined that misaligned atomicity
1906         * may be handled with two 8-byte loads.
1907         */
1908        if (h.aa.align < MO_128) {
1909            /*
1910             * TODO: align should be MO_64, so we only need test bit 3,
1911             * which means we could use TBNZ instead of ANDS+B_C.
1912             */
1913            tcg_out_logicali(s, I3404_ANDSI, 0, TCG_REG_XZR, addr_reg, 15);
1914            branch = s->code_ptr;
1915            tcg_out_insn(s, 3202, B_C, TCG_COND_NE, 0);
1916            use_pair = true;
1917        }
1918
1919        if (is_ld) {
1920            /*
1921             * 16-byte atomicity without LSE2 requires LDXP+STXP loop:
1922             *    ldxp lo, hi, [base]
1923             *    stxp t0, lo, hi, [base]
1924             *    cbnz t0, .-8
1925             * Require no overlap between data{lo,hi} and base.
1926             */
1927            if (datalo == base || datahi == base) {
1928                tcg_out_mov(s, TCG_TYPE_REG, TCG_REG_TMP2, base);
1929                base = TCG_REG_TMP2;
1930            }
1931            ll = sl = datalo;
1932            lh = sh = datahi;
1933        } else {
1934            /*
1935             * 16-byte atomicity without LSE2 requires LDXP+STXP loop:
1936             * 1: ldxp t0, t1, [base]
1937             *    stxp t0, lo, hi, [base]
1938             *    cbnz t0, 1b
1939             */
1940            tcg_debug_assert(base != TCG_REG_TMP0 && base != TCG_REG_TMP1);
1941            ll = TCG_REG_TMP0;
1942            lh = TCG_REG_TMP1;
1943            sl = datalo;
1944            sh = datahi;
1945        }
1946
1947        tcg_out_insn(s, 3306, LDXP, TCG_REG_XZR, ll, lh, base);
1948        tcg_out_insn(s, 3306, STXP, TCG_REG_TMP0, sl, sh, base);
1949        tcg_out_insn(s, 3201, CBNZ, 0, TCG_REG_TMP0, -2);
1950
1951        if (use_pair) {
1952            /* "b .+8", branching across the one insn of use_pair. */
1953            tcg_out_insn(s, 3206, B, 2);
1954            reloc_pc19(branch, tcg_splitwx_to_rx(s->code_ptr));
1955        }
1956    }
1957
1958    if (use_pair) {
1959        if (is_ld) {
1960            tcg_out_insn(s, 3314, LDP, datalo, datahi, base, 0, 1, 0);
1961        } else {
1962            tcg_out_insn(s, 3314, STP, datalo, datahi, base, 0, 1, 0);
1963        }
1964    }
1965
1966    if (ldst) {
1967        ldst->type = TCG_TYPE_I128;
1968        ldst->datalo_reg = datalo;
1969        ldst->datahi_reg = datahi;
1970        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1971    }
1972}
1973
1974static const tcg_insn_unit *tb_ret_addr;
1975
1976static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1977{
1978    const tcg_insn_unit *target;
1979    ptrdiff_t offset;
1980
1981    /* Reuse the zeroing that exists for goto_ptr.  */
1982    if (a0 == 0) {
1983        target = tcg_code_gen_epilogue;
1984    } else {
1985        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_X0, a0);
1986        target = tb_ret_addr;
1987    }
1988
1989    offset = tcg_pcrel_diff(s, target) >> 2;
1990    if (offset == sextract64(offset, 0, 26)) {
1991        tcg_out_insn(s, 3206, B, offset);
1992    } else {
1993        /*
1994         * Only x16/x17 generate BTI type Jump (2),
1995         * other registers generate BTI type Jump|Call (3).
1996         */
1997        QEMU_BUILD_BUG_ON(TCG_REG_TMP0 != TCG_REG_X16);
1998        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_TMP0, (intptr_t)target);
1999        tcg_out_insn(s, 3207, BR, TCG_REG_TMP0);
2000    }
2001}
2002
2003static void tcg_out_goto_tb(TCGContext *s, int which)
2004{
2005    /*
2006     * Direct branch, or indirect address load, will be patched
2007     * by tb_target_set_jmp_target.  Assert indirect load offset
2008     * in range early, regardless of direct branch distance.
2009     */
2010    intptr_t i_off = tcg_pcrel_diff(s, (void *)get_jmp_target_addr(s, which));
2011    tcg_debug_assert(i_off == sextract64(i_off, 0, 21));
2012
2013    set_jmp_insn_offset(s, which);
2014    tcg_out32(s, I3206_B);
2015    tcg_out_insn(s, 3207, BR, TCG_REG_TMP0);
2016    set_jmp_reset_offset(s, which);
2017    tcg_out_bti(s, BTI_J);
2018}
2019
2020void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
2021                              uintptr_t jmp_rx, uintptr_t jmp_rw)
2022{
2023    uintptr_t d_addr = tb->jmp_target_addr[n];
2024    ptrdiff_t d_offset = d_addr - jmp_rx;
2025    tcg_insn_unit insn;
2026
2027    /* Either directly branch, or indirect branch load. */
2028    if (d_offset == sextract64(d_offset, 0, 28)) {
2029        insn = deposit32(I3206_B, 0, 26, d_offset >> 2);
2030    } else {
2031        uintptr_t i_addr = (uintptr_t)&tb->jmp_target_addr[n];
2032        ptrdiff_t i_offset = i_addr - jmp_rx;
2033
2034        /* Note that we asserted this in range in tcg_out_goto_tb. */
2035        insn = deposit32(I3305_LDR | TCG_REG_TMP0, 5, 19, i_offset >> 2);
2036    }
2037    qatomic_set((uint32_t *)jmp_rw, insn);
2038    flush_idcache_range(jmp_rx, jmp_rw, 4);
2039}
2040
2041
2042static void tgen_add(TCGContext *s, TCGType type,
2043                     TCGReg a0, TCGReg a1, TCGReg a2)
2044{
2045    tcg_out_insn(s, 3502, ADD, type, a0, a1, a2);
2046}
2047
2048static void tgen_addi(TCGContext *s, TCGType type,
2049                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2050{
2051    if (a2 >= 0) {
2052        tcg_out_insn(s, 3401, ADDI, type, a0, a1, a2);
2053    } else {
2054        tcg_out_insn(s, 3401, SUBI, type, a0, a1, -a2);
2055    }
2056}
2057
2058static const TCGOutOpBinary outop_add = {
2059    .base.static_constraint = C_O1_I2(r, r, rA),
2060    .out_rrr = tgen_add,
2061    .out_rri = tgen_addi,
2062};
2063
2064static void tgen_and(TCGContext *s, TCGType type,
2065                     TCGReg a0, TCGReg a1, TCGReg a2)
2066{
2067    tcg_out_insn(s, 3510, AND, type, a0, a1, a2);
2068}
2069
2070static void tgen_andi(TCGContext *s, TCGType type,
2071                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2072{
2073    tcg_out_logicali(s, I3404_ANDI, type, a0, a1, a2);
2074}
2075
2076static const TCGOutOpBinary outop_and = {
2077    .base.static_constraint = C_O1_I2(r, r, rL),
2078    .out_rrr = tgen_and,
2079    .out_rri = tgen_andi,
2080};
2081
2082static void tgen_andc(TCGContext *s, TCGType type,
2083                      TCGReg a0, TCGReg a1, TCGReg a2)
2084{
2085    tcg_out_insn(s, 3510, BIC, type, a0, a1, a2);
2086}
2087
2088static const TCGOutOpBinary outop_andc = {
2089    .base.static_constraint = C_O1_I2(r, r, r),
2090    .out_rrr = tgen_andc,
2091};
2092
2093static void tgen_clz(TCGContext *s, TCGType type,
2094                     TCGReg a0, TCGReg a1, TCGReg a2)
2095{
2096    tcg_out_cmp(s, type, TCG_COND_NE, a1, 0, true);
2097    tcg_out_insn(s, 3507, CLZ, type, TCG_REG_TMP0, a1);
2098    tcg_out_insn(s, 3506, CSEL, type, a0, TCG_REG_TMP0, a2, TCG_COND_NE);
2099}
2100
2101static void tgen_clzi(TCGContext *s, TCGType type,
2102                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2103{
2104    if (a2 == (type == TCG_TYPE_I32 ? 32 : 64)) {
2105        tcg_out_insn(s, 3507, CLZ, type, a0, a1);
2106        return;
2107    }
2108
2109    tcg_out_cmp(s, type, TCG_COND_NE, a1, 0, true);
2110    tcg_out_insn(s, 3507, CLZ, type, a0, a1);
2111
2112    switch (a2) {
2113    case -1:
2114        tcg_out_insn(s, 3506, CSINV, type, a0, a0, TCG_REG_XZR, TCG_COND_NE);
2115        break;
2116    case 0:
2117        tcg_out_insn(s, 3506, CSEL, type, a0, a0, TCG_REG_XZR, TCG_COND_NE);
2118        break;
2119    default:
2120        tcg_out_movi(s, type, TCG_REG_TMP0, a2);
2121        tcg_out_insn(s, 3506, CSEL, type, a0, a0, TCG_REG_TMP0, TCG_COND_NE);
2122        break;
2123    }
2124}
2125
2126static const TCGOutOpBinary outop_clz = {
2127    .base.static_constraint = C_O1_I2(r, r, rAL),
2128    .out_rrr = tgen_clz,
2129    .out_rri = tgen_clzi,
2130};
2131
2132static void tgen_divs(TCGContext *s, TCGType type,
2133                      TCGReg a0, TCGReg a1, TCGReg a2)
2134{
2135    tcg_out_insn(s, 3508, SDIV, type, a0, a1, a2);
2136}
2137
2138static const TCGOutOpBinary outop_divs = {
2139    .base.static_constraint = C_O1_I2(r, r, r),
2140    .out_rrr = tgen_divs,
2141};
2142
2143static const TCGOutOpDivRem outop_divs2 = {
2144    .base.static_constraint = C_NotImplemented,
2145};
2146
2147static void tgen_divu(TCGContext *s, TCGType type,
2148                      TCGReg a0, TCGReg a1, TCGReg a2)
2149{
2150    tcg_out_insn(s, 3508, UDIV, type, a0, a1, a2);
2151}
2152
2153static const TCGOutOpBinary outop_divu = {
2154    .base.static_constraint = C_O1_I2(r, r, r),
2155    .out_rrr = tgen_divu,
2156};
2157
2158static const TCGOutOpDivRem outop_divu2 = {
2159    .base.static_constraint = C_NotImplemented,
2160};
2161
2162static void tgen_eqv(TCGContext *s, TCGType type,
2163                     TCGReg a0, TCGReg a1, TCGReg a2)
2164{
2165    tcg_out_insn(s, 3510, EON, type, a0, a1, a2);
2166}
2167
2168static const TCGOutOpBinary outop_eqv = {
2169    .base.static_constraint = C_O1_I2(r, r, r),
2170    .out_rrr = tgen_eqv,
2171};
2172
2173static void tgen_mul(TCGContext *s, TCGType type,
2174                     TCGReg a0, TCGReg a1, TCGReg a2)
2175{
2176    tcg_out_insn(s, 3509, MADD, type, a0, a1, a2, TCG_REG_XZR);
2177}
2178
2179static const TCGOutOpBinary outop_mul = {
2180    .base.static_constraint = C_O1_I2(r, r, r),
2181    .out_rrr = tgen_mul,
2182};
2183
2184static TCGConstraintSetIndex cset_mulh(TCGType type, unsigned flags)
2185{
2186    return type == TCG_TYPE_I64 ? C_O1_I2(r, r, r) : C_NotImplemented;
2187}
2188
2189static void tgen_mulsh(TCGContext *s, TCGType type,
2190                       TCGReg a0, TCGReg a1, TCGReg a2)
2191{
2192    tcg_out_insn(s, 3508, SMULH, TCG_TYPE_I64, a0, a1, a2);
2193}
2194
2195static const TCGOutOpBinary outop_mulsh = {
2196    .base.static_constraint = C_Dynamic,
2197    .base.dynamic_constraint = cset_mulh,
2198    .out_rrr = tgen_mulsh,
2199};
2200
2201static void tgen_muluh(TCGContext *s, TCGType type,
2202                       TCGReg a0, TCGReg a1, TCGReg a2)
2203{
2204    tcg_out_insn(s, 3508, UMULH, TCG_TYPE_I64, a0, a1, a2);
2205}
2206
2207static const TCGOutOpBinary outop_muluh = {
2208    .base.static_constraint = C_Dynamic,
2209    .base.dynamic_constraint = cset_mulh,
2210    .out_rrr = tgen_muluh,
2211};
2212
2213static const TCGOutOpBinary outop_nand = {
2214    .base.static_constraint = C_NotImplemented,
2215};
2216
2217static const TCGOutOpBinary outop_nor = {
2218    .base.static_constraint = C_NotImplemented,
2219};
2220
2221static void tgen_or(TCGContext *s, TCGType type,
2222                    TCGReg a0, TCGReg a1, TCGReg a2)
2223{
2224    tcg_out_insn(s, 3510, ORR, type, a0, a1, a2);
2225}
2226
2227static void tgen_ori(TCGContext *s, TCGType type,
2228                     TCGReg a0, TCGReg a1, tcg_target_long a2)
2229{
2230    tcg_out_logicali(s, I3404_ORRI, type, a0, a1, a2);
2231}
2232
2233static const TCGOutOpBinary outop_or = {
2234    .base.static_constraint = C_O1_I2(r, r, rL),
2235    .out_rrr = tgen_or,
2236    .out_rri = tgen_ori,
2237};
2238
2239static void tgen_orc(TCGContext *s, TCGType type,
2240                     TCGReg a0, TCGReg a1, TCGReg a2)
2241{
2242    tcg_out_insn(s, 3510, ORN, type, a0, a1, a2);
2243}
2244
2245static const TCGOutOpBinary outop_orc = {
2246    .base.static_constraint = C_O1_I2(r, r, r),
2247    .out_rrr = tgen_orc,
2248};
2249
2250static void tgen_rems(TCGContext *s, TCGType type,
2251                      TCGReg a0, TCGReg a1, TCGReg a2)
2252{
2253    tcg_out_insn(s, 3508, SDIV, type, TCG_REG_TMP0, a1, a2);
2254    tcg_out_insn(s, 3509, MSUB, type, a0, TCG_REG_TMP0, a2, a1);
2255}
2256
2257static const TCGOutOpBinary outop_rems = {
2258    .base.static_constraint = C_O1_I2(r, r, r),
2259    .out_rrr = tgen_rems,
2260};
2261
2262static void tgen_remu(TCGContext *s, TCGType type,
2263                      TCGReg a0, TCGReg a1, TCGReg a2)
2264{
2265    tcg_out_insn(s, 3508, UDIV, type, TCG_REG_TMP0, a1, a2);
2266    tcg_out_insn(s, 3509, MSUB, type, a0, TCG_REG_TMP0, a2, a1);
2267}
2268
2269static const TCGOutOpBinary outop_remu = {
2270    .base.static_constraint = C_O1_I2(r, r, r),
2271    .out_rrr = tgen_remu,
2272};
2273
2274static const TCGOutOpBinary outop_rotl = {
2275    .base.static_constraint = C_NotImplemented,
2276};
2277
2278static void tgen_rotr(TCGContext *s, TCGType type,
2279                      TCGReg a0, TCGReg a1, TCGReg a2)
2280{
2281    tcg_out_insn(s, 3508, RORV, type, a0, a1, a2);
2282}
2283
2284static void tgen_rotri(TCGContext *s, TCGType type,
2285                       TCGReg a0, TCGReg a1, tcg_target_long a2)
2286{
2287    int max = type == TCG_TYPE_I32 ? 31 : 63;
2288    tcg_out_extr(s, type, a0, a1, a1, a2 & max);
2289}
2290
2291static const TCGOutOpBinary outop_rotr = {
2292    .base.static_constraint = C_O1_I2(r, r, ri),
2293    .out_rrr = tgen_rotr,
2294    .out_rri = tgen_rotri,
2295};
2296
2297static void tgen_sar(TCGContext *s, TCGType type,
2298                     TCGReg a0, TCGReg a1, TCGReg a2)
2299{
2300    tcg_out_insn(s, 3508, ASRV, type, a0, a1, a2);
2301}
2302
2303static void tgen_sari(TCGContext *s, TCGType type,
2304                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2305{
2306    int max = type == TCG_TYPE_I32 ? 31 : 63;
2307    tcg_out_sbfm(s, type, a0, a1, a2 & max, max);
2308}
2309
2310static const TCGOutOpBinary outop_sar = {
2311    .base.static_constraint = C_O1_I2(r, r, ri),
2312    .out_rrr = tgen_sar,
2313    .out_rri = tgen_sari,
2314};
2315
2316static void tgen_shl(TCGContext *s, TCGType type,
2317                     TCGReg a0, TCGReg a1, TCGReg a2)
2318{
2319    tcg_out_insn(s, 3508, LSLV, type, a0, a1, a2);
2320}
2321
2322static void tgen_shli(TCGContext *s, TCGType type,
2323                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2324{
2325    int max = type == TCG_TYPE_I32 ? 31 : 63;
2326    tcg_out_ubfm(s, type, a0, a1, -a2 & max, ~a2 & max);
2327}
2328
2329static const TCGOutOpBinary outop_shl = {
2330    .base.static_constraint = C_O1_I2(r, r, ri),
2331    .out_rrr = tgen_shl,
2332    .out_rri = tgen_shli,
2333};
2334
2335static void tgen_shr(TCGContext *s, TCGType type,
2336                     TCGReg a0, TCGReg a1, TCGReg a2)
2337{
2338    tcg_out_insn(s, 3508, LSRV, type, a0, a1, a2);
2339}
2340
2341static void tgen_shri(TCGContext *s, TCGType type,
2342                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2343{
2344    int max = type == TCG_TYPE_I32 ? 31 : 63;
2345    tcg_out_ubfm(s, type, a0, a1, a2 & max, max);
2346}
2347
2348static const TCGOutOpBinary outop_shr = {
2349    .base.static_constraint = C_O1_I2(r, r, ri),
2350    .out_rrr = tgen_shr,
2351    .out_rri = tgen_shri,
2352};
2353
2354static void tgen_sub(TCGContext *s, TCGType type,
2355                     TCGReg a0, TCGReg a1, TCGReg a2)
2356{
2357    tcg_out_insn(s, 3502, SUB, type, a0, a1, a2);
2358}
2359
2360static const TCGOutOpSubtract outop_sub = {
2361    .base.static_constraint = C_O1_I2(r, r, r),
2362    .out_rrr = tgen_sub,
2363};
2364
2365static void tgen_xor(TCGContext *s, TCGType type,
2366                     TCGReg a0, TCGReg a1, TCGReg a2)
2367{
2368    tcg_out_insn(s, 3510, EOR, type, a0, a1, a2);
2369}
2370
2371static void tgen_xori(TCGContext *s, TCGType type,
2372                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2373{
2374    tcg_out_logicali(s, I3404_EORI, type, a0, a1, a2);
2375}
2376
2377static const TCGOutOpBinary outop_xor = {
2378    .base.static_constraint = C_O1_I2(r, r, rL),
2379    .out_rrr = tgen_xor,
2380    .out_rri = tgen_xori,
2381};
2382
2383
2384static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2385{
2386    tgen_sub(s, type, a0, TCG_REG_XZR, a1);
2387}
2388
2389static const TCGOutOpUnary outop_neg = {
2390    .base.static_constraint = C_O1_I1(r, r),
2391    .out_rr = tgen_neg,
2392};
2393
2394static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2395{
2396    tgen_orc(s, type, a0, TCG_REG_XZR, a1);
2397}
2398
2399static const TCGOutOpUnary outop_not = {
2400    .base.static_constraint = C_O1_I1(r, r),
2401    .out_rr = tgen_not,
2402};
2403
2404
2405static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType ext,
2406                       const TCGArg args[TCG_MAX_OP_ARGS],
2407                       const int const_args[TCG_MAX_OP_ARGS])
2408{
2409    /* Hoist the loads of the most common arguments.  */
2410    TCGArg a0 = args[0];
2411    TCGArg a1 = args[1];
2412    TCGArg a2 = args[2];
2413    int c2 = const_args[2];
2414
2415    switch (opc) {
2416    case INDEX_op_goto_ptr:
2417        tcg_out_insn(s, 3207, BR, a0);
2418        break;
2419
2420    case INDEX_op_br:
2421        tcg_out_goto_label(s, arg_label(a0));
2422        break;
2423
2424    case INDEX_op_ld8u_i32:
2425    case INDEX_op_ld8u_i64:
2426        tcg_out_ldst(s, I3312_LDRB, a0, a1, a2, 0);
2427        break;
2428    case INDEX_op_ld8s_i32:
2429        tcg_out_ldst(s, I3312_LDRSBW, a0, a1, a2, 0);
2430        break;
2431    case INDEX_op_ld8s_i64:
2432        tcg_out_ldst(s, I3312_LDRSBX, a0, a1, a2, 0);
2433        break;
2434    case INDEX_op_ld16u_i32:
2435    case INDEX_op_ld16u_i64:
2436        tcg_out_ldst(s, I3312_LDRH, a0, a1, a2, 1);
2437        break;
2438    case INDEX_op_ld16s_i32:
2439        tcg_out_ldst(s, I3312_LDRSHW, a0, a1, a2, 1);
2440        break;
2441    case INDEX_op_ld16s_i64:
2442        tcg_out_ldst(s, I3312_LDRSHX, a0, a1, a2, 1);
2443        break;
2444    case INDEX_op_ld_i32:
2445    case INDEX_op_ld32u_i64:
2446        tcg_out_ldst(s, I3312_LDRW, a0, a1, a2, 2);
2447        break;
2448    case INDEX_op_ld32s_i64:
2449        tcg_out_ldst(s, I3312_LDRSWX, a0, a1, a2, 2);
2450        break;
2451    case INDEX_op_ld_i64:
2452        tcg_out_ldst(s, I3312_LDRX, a0, a1, a2, 3);
2453        break;
2454
2455    case INDEX_op_st8_i32:
2456    case INDEX_op_st8_i64:
2457        tcg_out_ldst(s, I3312_STRB, a0, a1, a2, 0);
2458        break;
2459    case INDEX_op_st16_i32:
2460    case INDEX_op_st16_i64:
2461        tcg_out_ldst(s, I3312_STRH, a0, a1, a2, 1);
2462        break;
2463    case INDEX_op_st_i32:
2464    case INDEX_op_st32_i64:
2465        tcg_out_ldst(s, I3312_STRW, a0, a1, a2, 2);
2466        break;
2467    case INDEX_op_st_i64:
2468        tcg_out_ldst(s, I3312_STRX, a0, a1, a2, 3);
2469        break;
2470
2471    case INDEX_op_ctz_i64:
2472    case INDEX_op_ctz_i32:
2473        tcg_out_insn(s, 3507, RBIT, ext, TCG_REG_TMP0, a1);
2474        if (c2) {
2475            tgen_clzi(s, ext, a0, TCG_REG_TMP0, a2);
2476        } else {
2477            tgen_clz(s, ext, a0, TCG_REG_TMP0, a2);
2478        }
2479        break;
2480
2481    case INDEX_op_brcond_i32:
2482        a1 = (int32_t)a1;
2483        /* FALLTHRU */
2484    case INDEX_op_brcond_i64:
2485        tcg_out_brcond(s, ext, a2, a0, a1, const_args[1], arg_label(args[3]));
2486        break;
2487
2488    case INDEX_op_setcond_i32:
2489        a2 = (int32_t)a2;
2490        /* FALLTHRU */
2491    case INDEX_op_setcond_i64:
2492        tcg_out_cmp(s, ext, args[3], a1, a2, c2);
2493        /* Use CSET alias of CSINC Wd, WZR, WZR, invert(cond).  */
2494        tcg_out_insn(s, 3506, CSINC, TCG_TYPE_I32, a0, TCG_REG_XZR,
2495                     TCG_REG_XZR, tcg_invert_cond(args[3]));
2496        break;
2497
2498    case INDEX_op_negsetcond_i32:
2499        a2 = (int32_t)a2;
2500        /* FALLTHRU */
2501    case INDEX_op_negsetcond_i64:
2502        tcg_out_cmp(s, ext, args[3], a1, a2, c2);
2503        /* Use CSETM alias of CSINV Wd, WZR, WZR, invert(cond).  */
2504        tcg_out_insn(s, 3506, CSINV, ext, a0, TCG_REG_XZR,
2505                     TCG_REG_XZR, tcg_invert_cond(args[3]));
2506        break;
2507
2508    case INDEX_op_movcond_i32:
2509        a2 = (int32_t)a2;
2510        /* FALLTHRU */
2511    case INDEX_op_movcond_i64:
2512        tcg_out_cmp(s, ext, args[5], a1, a2, c2);
2513        tcg_out_insn(s, 3506, CSEL, ext, a0, args[3], args[4], args[5]);
2514        break;
2515
2516    case INDEX_op_qemu_ld_i32:
2517    case INDEX_op_qemu_ld_i64:
2518        tcg_out_qemu_ld(s, a0, a1, a2, ext);
2519        break;
2520    case INDEX_op_qemu_st_i32:
2521    case INDEX_op_qemu_st_i64:
2522        tcg_out_qemu_st(s, a0, a1, a2, ext);
2523        break;
2524    case INDEX_op_qemu_ld_i128:
2525        tcg_out_qemu_ldst_i128(s, a0, a1, a2, args[3], true);
2526        break;
2527    case INDEX_op_qemu_st_i128:
2528        tcg_out_qemu_ldst_i128(s, a0, a1, a2, args[3], false);
2529        break;
2530
2531    case INDEX_op_bswap64_i64:
2532        tcg_out_rev(s, TCG_TYPE_I64, MO_64, a0, a1);
2533        break;
2534    case INDEX_op_bswap32_i64:
2535        tcg_out_rev(s, TCG_TYPE_I32, MO_32, a0, a1);
2536        if (a2 & TCG_BSWAP_OS) {
2537            tcg_out_ext32s(s, a0, a0);
2538        }
2539        break;
2540    case INDEX_op_bswap32_i32:
2541        tcg_out_rev(s, TCG_TYPE_I32, MO_32, a0, a1);
2542        break;
2543    case INDEX_op_bswap16_i64:
2544    case INDEX_op_bswap16_i32:
2545        tcg_out_rev(s, TCG_TYPE_I32, MO_16, a0, a1);
2546        if (a2 & TCG_BSWAP_OS) {
2547            /* Output must be sign-extended. */
2548            tcg_out_ext16s(s, ext, a0, a0);
2549        } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2550            /* Output must be zero-extended, but input isn't. */
2551            tcg_out_ext16u(s, a0, a0);
2552        }
2553        break;
2554
2555    case INDEX_op_deposit_i64:
2556    case INDEX_op_deposit_i32:
2557        tcg_out_dep(s, ext, a0, a2, args[3], args[4]);
2558        break;
2559
2560    case INDEX_op_extract_i64:
2561    case INDEX_op_extract_i32:
2562        if (a2 == 0) {
2563            uint64_t mask = MAKE_64BIT_MASK(0, args[3]);
2564            tcg_out_logicali(s, I3404_ANDI, ext, a0, a1, mask);
2565        } else {
2566            tcg_out_ubfm(s, ext, a0, a1, a2, a2 + args[3] - 1);
2567        }
2568        break;
2569
2570    case INDEX_op_sextract_i64:
2571    case INDEX_op_sextract_i32:
2572        tcg_out_sbfm(s, ext, a0, a1, a2, a2 + args[3] - 1);
2573        break;
2574
2575    case INDEX_op_extract2_i64:
2576    case INDEX_op_extract2_i32:
2577        tcg_out_extr(s, ext, a0, a2, a1, args[3]);
2578        break;
2579
2580    case INDEX_op_add2_i32:
2581        tcg_out_addsub2(s, TCG_TYPE_I32, a0, a1, a2, args[3],
2582                        (int32_t)args[4], args[5], const_args[4],
2583                        const_args[5], false);
2584        break;
2585    case INDEX_op_add2_i64:
2586        tcg_out_addsub2(s, TCG_TYPE_I64, a0, a1, a2, args[3], args[4],
2587                        args[5], const_args[4], const_args[5], false);
2588        break;
2589    case INDEX_op_sub2_i32:
2590        tcg_out_addsub2(s, TCG_TYPE_I32, a0, a1, a2, args[3],
2591                        (int32_t)args[4], args[5], const_args[4],
2592                        const_args[5], true);
2593        break;
2594    case INDEX_op_sub2_i64:
2595        tcg_out_addsub2(s, TCG_TYPE_I64, a0, a1, a2, args[3], args[4],
2596                        args[5], const_args[4], const_args[5], true);
2597        break;
2598
2599    case INDEX_op_mb:
2600        tcg_out_mb(s, a0);
2601        break;
2602
2603    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2604    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2605    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2606    case INDEX_op_ext_i32_i64:  /* Always emitted via tcg_reg_alloc_op.  */
2607    case INDEX_op_extu_i32_i64:
2608    case INDEX_op_extrl_i64_i32:
2609    default:
2610        g_assert_not_reached();
2611    }
2612}
2613
2614static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2615                           unsigned vecl, unsigned vece,
2616                           const TCGArg args[TCG_MAX_OP_ARGS],
2617                           const int const_args[TCG_MAX_OP_ARGS])
2618{
2619    static const AArch64Insn cmp_vec_insn[16] = {
2620        [TCG_COND_EQ] = I3616_CMEQ,
2621        [TCG_COND_GT] = I3616_CMGT,
2622        [TCG_COND_GE] = I3616_CMGE,
2623        [TCG_COND_GTU] = I3616_CMHI,
2624        [TCG_COND_GEU] = I3616_CMHS,
2625    };
2626    static const AArch64Insn cmp_scalar_insn[16] = {
2627        [TCG_COND_EQ] = I3611_CMEQ,
2628        [TCG_COND_GT] = I3611_CMGT,
2629        [TCG_COND_GE] = I3611_CMGE,
2630        [TCG_COND_GTU] = I3611_CMHI,
2631        [TCG_COND_GEU] = I3611_CMHS,
2632    };
2633    static const AArch64Insn cmp0_vec_insn[16] = {
2634        [TCG_COND_EQ] = I3617_CMEQ0,
2635        [TCG_COND_GT] = I3617_CMGT0,
2636        [TCG_COND_GE] = I3617_CMGE0,
2637        [TCG_COND_LT] = I3617_CMLT0,
2638        [TCG_COND_LE] = I3617_CMLE0,
2639    };
2640    static const AArch64Insn cmp0_scalar_insn[16] = {
2641        [TCG_COND_EQ] = I3612_CMEQ0,
2642        [TCG_COND_GT] = I3612_CMGT0,
2643        [TCG_COND_GE] = I3612_CMGE0,
2644        [TCG_COND_LT] = I3612_CMLT0,
2645        [TCG_COND_LE] = I3612_CMLE0,
2646    };
2647
2648    TCGType type = vecl + TCG_TYPE_V64;
2649    unsigned is_q = vecl;
2650    bool is_scalar = !is_q && vece == MO_64;
2651    TCGArg a0, a1, a2, a3;
2652    int cmode, imm8;
2653
2654    a0 = args[0];
2655    a1 = args[1];
2656    a2 = args[2];
2657
2658    switch (opc) {
2659    case INDEX_op_ld_vec:
2660        tcg_out_ld(s, type, a0, a1, a2);
2661        break;
2662    case INDEX_op_st_vec:
2663        tcg_out_st(s, type, a0, a1, a2);
2664        break;
2665    case INDEX_op_dupm_vec:
2666        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2667        break;
2668    case INDEX_op_add_vec:
2669        if (is_scalar) {
2670            tcg_out_insn(s, 3611, ADD, vece, a0, a1, a2);
2671        } else {
2672            tcg_out_insn(s, 3616, ADD, is_q, vece, a0, a1, a2);
2673        }
2674        break;
2675    case INDEX_op_sub_vec:
2676        if (is_scalar) {
2677            tcg_out_insn(s, 3611, SUB, vece, a0, a1, a2);
2678        } else {
2679            tcg_out_insn(s, 3616, SUB, is_q, vece, a0, a1, a2);
2680        }
2681        break;
2682    case INDEX_op_mul_vec:
2683        tcg_out_insn(s, 3616, MUL, is_q, vece, a0, a1, a2);
2684        break;
2685    case INDEX_op_neg_vec:
2686        if (is_scalar) {
2687            tcg_out_insn(s, 3612, NEG, vece, a0, a1);
2688        } else {
2689            tcg_out_insn(s, 3617, NEG, is_q, vece, a0, a1);
2690        }
2691        break;
2692    case INDEX_op_abs_vec:
2693        if (is_scalar) {
2694            tcg_out_insn(s, 3612, ABS, vece, a0, a1);
2695        } else {
2696            tcg_out_insn(s, 3617, ABS, is_q, vece, a0, a1);
2697        }
2698        break;
2699    case INDEX_op_and_vec:
2700        if (const_args[2]) {
2701            is_shimm1632(~a2, &cmode, &imm8);
2702            if (a0 == a1) {
2703                tcg_out_insn(s, 3606, BIC, is_q, a0, 0, cmode, imm8);
2704                return;
2705            }
2706            tcg_out_insn(s, 3606, MVNI, is_q, a0, 0, cmode, imm8);
2707            a2 = a0;
2708        }
2709        tcg_out_insn(s, 3616, AND, is_q, 0, a0, a1, a2);
2710        break;
2711    case INDEX_op_or_vec:
2712        if (const_args[2]) {
2713            is_shimm1632(a2, &cmode, &imm8);
2714            if (a0 == a1) {
2715                tcg_out_insn(s, 3606, ORR, is_q, a0, 0, cmode, imm8);
2716                return;
2717            }
2718            tcg_out_insn(s, 3606, MOVI, is_q, a0, 0, cmode, imm8);
2719            a2 = a0;
2720        }
2721        tcg_out_insn(s, 3616, ORR, is_q, 0, a0, a1, a2);
2722        break;
2723    case INDEX_op_andc_vec:
2724        if (const_args[2]) {
2725            is_shimm1632(a2, &cmode, &imm8);
2726            if (a0 == a1) {
2727                tcg_out_insn(s, 3606, BIC, is_q, a0, 0, cmode, imm8);
2728                return;
2729            }
2730            tcg_out_insn(s, 3606, MOVI, is_q, a0, 0, cmode, imm8);
2731            a2 = a0;
2732        }
2733        tcg_out_insn(s, 3616, BIC, is_q, 0, a0, a1, a2);
2734        break;
2735    case INDEX_op_orc_vec:
2736        if (const_args[2]) {
2737            is_shimm1632(~a2, &cmode, &imm8);
2738            if (a0 == a1) {
2739                tcg_out_insn(s, 3606, ORR, is_q, a0, 0, cmode, imm8);
2740                return;
2741            }
2742            tcg_out_insn(s, 3606, MVNI, is_q, a0, 0, cmode, imm8);
2743            a2 = a0;
2744        }
2745        tcg_out_insn(s, 3616, ORN, is_q, 0, a0, a1, a2);
2746        break;
2747    case INDEX_op_xor_vec:
2748        tcg_out_insn(s, 3616, EOR, is_q, 0, a0, a1, a2);
2749        break;
2750    case INDEX_op_ssadd_vec:
2751        if (is_scalar) {
2752            tcg_out_insn(s, 3611, SQADD, vece, a0, a1, a2);
2753        } else {
2754            tcg_out_insn(s, 3616, SQADD, is_q, vece, a0, a1, a2);
2755        }
2756        break;
2757    case INDEX_op_sssub_vec:
2758        if (is_scalar) {
2759            tcg_out_insn(s, 3611, SQSUB, vece, a0, a1, a2);
2760        } else {
2761            tcg_out_insn(s, 3616, SQSUB, is_q, vece, a0, a1, a2);
2762        }
2763        break;
2764    case INDEX_op_usadd_vec:
2765        if (is_scalar) {
2766            tcg_out_insn(s, 3611, UQADD, vece, a0, a1, a2);
2767        } else {
2768            tcg_out_insn(s, 3616, UQADD, is_q, vece, a0, a1, a2);
2769        }
2770        break;
2771    case INDEX_op_ussub_vec:
2772        if (is_scalar) {
2773            tcg_out_insn(s, 3611, UQSUB, vece, a0, a1, a2);
2774        } else {
2775            tcg_out_insn(s, 3616, UQSUB, is_q, vece, a0, a1, a2);
2776        }
2777        break;
2778    case INDEX_op_smax_vec:
2779        tcg_out_insn(s, 3616, SMAX, is_q, vece, a0, a1, a2);
2780        break;
2781    case INDEX_op_smin_vec:
2782        tcg_out_insn(s, 3616, SMIN, is_q, vece, a0, a1, a2);
2783        break;
2784    case INDEX_op_umax_vec:
2785        tcg_out_insn(s, 3616, UMAX, is_q, vece, a0, a1, a2);
2786        break;
2787    case INDEX_op_umin_vec:
2788        tcg_out_insn(s, 3616, UMIN, is_q, vece, a0, a1, a2);
2789        break;
2790    case INDEX_op_not_vec:
2791        tcg_out_insn(s, 3617, NOT, is_q, 0, a0, a1);
2792        break;
2793    case INDEX_op_shli_vec:
2794        if (is_scalar) {
2795            tcg_out_insn(s, 3609, SHL, a0, a1, a2 + (8 << vece));
2796        } else {
2797            tcg_out_insn(s, 3614, SHL, is_q, a0, a1, a2 + (8 << vece));
2798        }
2799        break;
2800    case INDEX_op_shri_vec:
2801        if (is_scalar) {
2802            tcg_out_insn(s, 3609, USHR, a0, a1, (16 << vece) - a2);
2803        } else {
2804            tcg_out_insn(s, 3614, USHR, is_q, a0, a1, (16 << vece) - a2);
2805        }
2806        break;
2807    case INDEX_op_sari_vec:
2808        if (is_scalar) {
2809            tcg_out_insn(s, 3609, SSHR, a0, a1, (16 << vece) - a2);
2810        } else {
2811            tcg_out_insn(s, 3614, SSHR, is_q, a0, a1, (16 << vece) - a2);
2812        }
2813        break;
2814    case INDEX_op_aa64_sli_vec:
2815        if (is_scalar) {
2816            tcg_out_insn(s, 3609, SLI, a0, a2, args[3] + (8 << vece));
2817        } else {
2818            tcg_out_insn(s, 3614, SLI, is_q, a0, a2, args[3] + (8 << vece));
2819        }
2820        break;
2821    case INDEX_op_shlv_vec:
2822        if (is_scalar) {
2823            tcg_out_insn(s, 3611, USHL, vece, a0, a1, a2);
2824        } else {
2825            tcg_out_insn(s, 3616, USHL, is_q, vece, a0, a1, a2);
2826        }
2827        break;
2828    case INDEX_op_aa64_sshl_vec:
2829        if (is_scalar) {
2830            tcg_out_insn(s, 3611, SSHL, vece, a0, a1, a2);
2831        } else {
2832            tcg_out_insn(s, 3616, SSHL, is_q, vece, a0, a1, a2);
2833        }
2834        break;
2835    case INDEX_op_cmp_vec:
2836        {
2837            TCGCond cond = args[3];
2838            AArch64Insn insn;
2839
2840            switch (cond) {
2841            case TCG_COND_NE:
2842                if (const_args[2]) {
2843                    if (is_scalar) {
2844                        tcg_out_insn(s, 3611, CMTST, vece, a0, a1, a1);
2845                    } else {
2846                        tcg_out_insn(s, 3616, CMTST, is_q, vece, a0, a1, a1);
2847                    }
2848                } else {
2849                    if (is_scalar) {
2850                        tcg_out_insn(s, 3611, CMEQ, vece, a0, a1, a2);
2851                    } else {
2852                        tcg_out_insn(s, 3616, CMEQ, is_q, vece, a0, a1, a2);
2853                    }
2854                    tcg_out_insn(s, 3617, NOT, is_q, 0, a0, a0);
2855                }
2856                break;
2857
2858            case TCG_COND_TSTNE:
2859            case TCG_COND_TSTEQ:
2860                if (const_args[2]) {
2861                    /* (x & 0) == 0 */
2862                    tcg_out_dupi_vec(s, type, MO_8, a0,
2863                                     -(cond == TCG_COND_TSTEQ));
2864                    break;
2865                }
2866                if (is_scalar) {
2867                    tcg_out_insn(s, 3611, CMTST, vece, a0, a1, a2);
2868                } else {
2869                    tcg_out_insn(s, 3616, CMTST, is_q, vece, a0, a1, a2);
2870                }
2871                if (cond == TCG_COND_TSTEQ) {
2872                    tcg_out_insn(s, 3617, NOT, is_q, 0, a0, a0);
2873                }
2874                break;
2875
2876            default:
2877                if (const_args[2]) {
2878                    if (is_scalar) {
2879                        insn = cmp0_scalar_insn[cond];
2880                        if (insn) {
2881                            tcg_out_insn_3612(s, insn, vece, a0, a1);
2882                            break;
2883                        }
2884                    } else {
2885                        insn = cmp0_vec_insn[cond];
2886                        if (insn) {
2887                            tcg_out_insn_3617(s, insn, is_q, vece, a0, a1);
2888                            break;
2889                        }
2890                    }
2891                    tcg_out_dupi_vec(s, type, MO_8, TCG_VEC_TMP0, 0);
2892                    a2 = TCG_VEC_TMP0;
2893                }
2894                if (is_scalar) {
2895                    insn = cmp_scalar_insn[cond];
2896                    if (insn == 0) {
2897                        TCGArg t;
2898                        t = a1, a1 = a2, a2 = t;
2899                        cond = tcg_swap_cond(cond);
2900                        insn = cmp_scalar_insn[cond];
2901                        tcg_debug_assert(insn != 0);
2902                    }
2903                    tcg_out_insn_3611(s, insn, vece, a0, a1, a2);
2904                } else {
2905                    insn = cmp_vec_insn[cond];
2906                    if (insn == 0) {
2907                        TCGArg t;
2908                        t = a1, a1 = a2, a2 = t;
2909                        cond = tcg_swap_cond(cond);
2910                        insn = cmp_vec_insn[cond];
2911                        tcg_debug_assert(insn != 0);
2912                    }
2913                    tcg_out_insn_3616(s, insn, is_q, vece, a0, a1, a2);
2914                }
2915                break;
2916            }
2917        }
2918        break;
2919
2920    case INDEX_op_bitsel_vec:
2921        a3 = args[3];
2922        if (a0 == a3) {
2923            tcg_out_insn(s, 3616, BIT, is_q, 0, a0, a2, a1);
2924        } else if (a0 == a2) {
2925            tcg_out_insn(s, 3616, BIF, is_q, 0, a0, a3, a1);
2926        } else {
2927            if (a0 != a1) {
2928                tcg_out_mov(s, type, a0, a1);
2929            }
2930            tcg_out_insn(s, 3616, BSL, is_q, 0, a0, a2, a3);
2931        }
2932        break;
2933
2934    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
2935    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
2936    default:
2937        g_assert_not_reached();
2938    }
2939}
2940
2941int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
2942{
2943    switch (opc) {
2944    case INDEX_op_add_vec:
2945    case INDEX_op_sub_vec:
2946    case INDEX_op_and_vec:
2947    case INDEX_op_or_vec:
2948    case INDEX_op_xor_vec:
2949    case INDEX_op_andc_vec:
2950    case INDEX_op_orc_vec:
2951    case INDEX_op_neg_vec:
2952    case INDEX_op_abs_vec:
2953    case INDEX_op_not_vec:
2954    case INDEX_op_cmp_vec:
2955    case INDEX_op_shli_vec:
2956    case INDEX_op_shri_vec:
2957    case INDEX_op_sari_vec:
2958    case INDEX_op_ssadd_vec:
2959    case INDEX_op_sssub_vec:
2960    case INDEX_op_usadd_vec:
2961    case INDEX_op_ussub_vec:
2962    case INDEX_op_shlv_vec:
2963    case INDEX_op_bitsel_vec:
2964        return 1;
2965    case INDEX_op_rotli_vec:
2966    case INDEX_op_shrv_vec:
2967    case INDEX_op_sarv_vec:
2968    case INDEX_op_rotlv_vec:
2969    case INDEX_op_rotrv_vec:
2970        return -1;
2971    case INDEX_op_mul_vec:
2972    case INDEX_op_smax_vec:
2973    case INDEX_op_smin_vec:
2974    case INDEX_op_umax_vec:
2975    case INDEX_op_umin_vec:
2976        return vece < MO_64;
2977
2978    default:
2979        return 0;
2980    }
2981}
2982
2983void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
2984                       TCGArg a0, ...)
2985{
2986    va_list va;
2987    TCGv_vec v0, v1, v2, t1, t2, c1;
2988    TCGArg a2;
2989
2990    va_start(va, a0);
2991    v0 = temp_tcgv_vec(arg_temp(a0));
2992    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
2993    a2 = va_arg(va, TCGArg);
2994    va_end(va);
2995
2996    switch (opc) {
2997    case INDEX_op_rotli_vec:
2998        t1 = tcg_temp_new_vec(type);
2999        tcg_gen_shri_vec(vece, t1, v1, -a2 & ((8 << vece) - 1));
3000        vec_gen_4(INDEX_op_aa64_sli_vec, type, vece,
3001                  tcgv_vec_arg(v0), tcgv_vec_arg(t1), tcgv_vec_arg(v1), a2);
3002        tcg_temp_free_vec(t1);
3003        break;
3004
3005    case INDEX_op_shrv_vec:
3006    case INDEX_op_sarv_vec:
3007        /* Right shifts are negative left shifts for AArch64.  */
3008        v2 = temp_tcgv_vec(arg_temp(a2));
3009        t1 = tcg_temp_new_vec(type);
3010        tcg_gen_neg_vec(vece, t1, v2);
3011        opc = (opc == INDEX_op_shrv_vec
3012               ? INDEX_op_shlv_vec : INDEX_op_aa64_sshl_vec);
3013        vec_gen_3(opc, type, vece, tcgv_vec_arg(v0),
3014                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3015        tcg_temp_free_vec(t1);
3016        break;
3017
3018    case INDEX_op_rotlv_vec:
3019        v2 = temp_tcgv_vec(arg_temp(a2));
3020        t1 = tcg_temp_new_vec(type);
3021        c1 = tcg_constant_vec(type, vece, 8 << vece);
3022        tcg_gen_sub_vec(vece, t1, v2, c1);
3023        /* Right shifts are negative left shifts for AArch64.  */
3024        vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(t1),
3025                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3026        vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(v0),
3027                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3028        tcg_gen_or_vec(vece, v0, v0, t1);
3029        tcg_temp_free_vec(t1);
3030        break;
3031
3032    case INDEX_op_rotrv_vec:
3033        v2 = temp_tcgv_vec(arg_temp(a2));
3034        t1 = tcg_temp_new_vec(type);
3035        t2 = tcg_temp_new_vec(type);
3036        c1 = tcg_constant_vec(type, vece, 8 << vece);
3037        tcg_gen_neg_vec(vece, t1, v2);
3038        tcg_gen_sub_vec(vece, t2, c1, v2);
3039        /* Right shifts are negative left shifts for AArch64.  */
3040        vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(t1),
3041                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3042        vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(t2),
3043                  tcgv_vec_arg(v1), tcgv_vec_arg(t2));
3044        tcg_gen_or_vec(vece, v0, t1, t2);
3045        tcg_temp_free_vec(t1);
3046        tcg_temp_free_vec(t2);
3047        break;
3048
3049    default:
3050        g_assert_not_reached();
3051    }
3052}
3053
3054static TCGConstraintSetIndex
3055tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
3056{
3057    switch (op) {
3058    case INDEX_op_goto_ptr:
3059        return C_O0_I1(r);
3060
3061    case INDEX_op_ld8u_i32:
3062    case INDEX_op_ld8s_i32:
3063    case INDEX_op_ld16u_i32:
3064    case INDEX_op_ld16s_i32:
3065    case INDEX_op_ld_i32:
3066    case INDEX_op_ld8u_i64:
3067    case INDEX_op_ld8s_i64:
3068    case INDEX_op_ld16u_i64:
3069    case INDEX_op_ld16s_i64:
3070    case INDEX_op_ld32u_i64:
3071    case INDEX_op_ld32s_i64:
3072    case INDEX_op_ld_i64:
3073    case INDEX_op_bswap16_i32:
3074    case INDEX_op_bswap32_i32:
3075    case INDEX_op_bswap16_i64:
3076    case INDEX_op_bswap32_i64:
3077    case INDEX_op_bswap64_i64:
3078    case INDEX_op_ext_i32_i64:
3079    case INDEX_op_extu_i32_i64:
3080    case INDEX_op_extract_i32:
3081    case INDEX_op_extract_i64:
3082    case INDEX_op_sextract_i32:
3083    case INDEX_op_sextract_i64:
3084        return C_O1_I1(r, r);
3085
3086    case INDEX_op_st8_i32:
3087    case INDEX_op_st16_i32:
3088    case INDEX_op_st_i32:
3089    case INDEX_op_st8_i64:
3090    case INDEX_op_st16_i64:
3091    case INDEX_op_st32_i64:
3092    case INDEX_op_st_i64:
3093        return C_O0_I2(rz, r);
3094
3095    case INDEX_op_setcond_i32:
3096    case INDEX_op_setcond_i64:
3097    case INDEX_op_negsetcond_i32:
3098    case INDEX_op_negsetcond_i64:
3099        return C_O1_I2(r, r, rC);
3100
3101    case INDEX_op_ctz_i32:
3102    case INDEX_op_ctz_i64:
3103        return C_O1_I2(r, r, rAL);
3104
3105    case INDEX_op_brcond_i32:
3106    case INDEX_op_brcond_i64:
3107        return C_O0_I2(r, rC);
3108
3109    case INDEX_op_movcond_i32:
3110    case INDEX_op_movcond_i64:
3111        return C_O1_I4(r, r, rC, rz, rz);
3112
3113    case INDEX_op_qemu_ld_i32:
3114    case INDEX_op_qemu_ld_i64:
3115        return C_O1_I1(r, r);
3116    case INDEX_op_qemu_ld_i128:
3117        return C_O2_I1(r, r, r);
3118    case INDEX_op_qemu_st_i32:
3119    case INDEX_op_qemu_st_i64:
3120        return C_O0_I2(rz, r);
3121    case INDEX_op_qemu_st_i128:
3122        return C_O0_I3(rz, rz, r);
3123
3124    case INDEX_op_deposit_i32:
3125    case INDEX_op_deposit_i64:
3126        return C_O1_I2(r, 0, rz);
3127
3128    case INDEX_op_extract2_i32:
3129    case INDEX_op_extract2_i64:
3130        return C_O1_I2(r, rz, rz);
3131
3132    case INDEX_op_add2_i32:
3133    case INDEX_op_add2_i64:
3134    case INDEX_op_sub2_i32:
3135    case INDEX_op_sub2_i64:
3136        return C_O2_I4(r, r, rz, rz, rA, rMZ);
3137
3138    case INDEX_op_add_vec:
3139    case INDEX_op_sub_vec:
3140    case INDEX_op_mul_vec:
3141    case INDEX_op_xor_vec:
3142    case INDEX_op_ssadd_vec:
3143    case INDEX_op_sssub_vec:
3144    case INDEX_op_usadd_vec:
3145    case INDEX_op_ussub_vec:
3146    case INDEX_op_smax_vec:
3147    case INDEX_op_smin_vec:
3148    case INDEX_op_umax_vec:
3149    case INDEX_op_umin_vec:
3150    case INDEX_op_shlv_vec:
3151    case INDEX_op_shrv_vec:
3152    case INDEX_op_sarv_vec:
3153    case INDEX_op_aa64_sshl_vec:
3154        return C_O1_I2(w, w, w);
3155    case INDEX_op_not_vec:
3156    case INDEX_op_neg_vec:
3157    case INDEX_op_abs_vec:
3158    case INDEX_op_shli_vec:
3159    case INDEX_op_shri_vec:
3160    case INDEX_op_sari_vec:
3161        return C_O1_I1(w, w);
3162    case INDEX_op_ld_vec:
3163    case INDEX_op_dupm_vec:
3164        return C_O1_I1(w, r);
3165    case INDEX_op_st_vec:
3166        return C_O0_I2(w, r);
3167    case INDEX_op_dup_vec:
3168        return C_O1_I1(w, wr);
3169    case INDEX_op_or_vec:
3170    case INDEX_op_andc_vec:
3171        return C_O1_I2(w, w, wO);
3172    case INDEX_op_and_vec:
3173    case INDEX_op_orc_vec:
3174        return C_O1_I2(w, w, wN);
3175    case INDEX_op_cmp_vec:
3176        return C_O1_I2(w, w, wZ);
3177    case INDEX_op_bitsel_vec:
3178        return C_O1_I3(w, w, w, w);
3179    case INDEX_op_aa64_sli_vec:
3180        return C_O1_I2(w, 0, w);
3181
3182    default:
3183        return C_NotImplemented;
3184    }
3185}
3186
3187static void tcg_target_init(TCGContext *s)
3188{
3189    tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffffu;
3190    tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffffu;
3191    tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3192    tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3193
3194    tcg_target_call_clobber_regs = -1ull;
3195    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X19);
3196    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X20);
3197    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X21);
3198    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X22);
3199    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X23);
3200    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X24);
3201    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X25);
3202    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X26);
3203    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X27);
3204    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X28);
3205    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X29);
3206    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V8);
3207    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V9);
3208    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V10);
3209    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V11);
3210    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V12);
3211    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V13);
3212    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V14);
3213    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V15);
3214
3215    s->reserved_regs = 0;
3216    tcg_regset_set_reg(s->reserved_regs, TCG_REG_SP);
3217    tcg_regset_set_reg(s->reserved_regs, TCG_REG_FP);
3218    tcg_regset_set_reg(s->reserved_regs, TCG_REG_X18); /* platform register */
3219    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP0);
3220    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1);
3221    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP2);
3222    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP0);
3223}
3224
3225/* Saving pairs: (X19, X20) .. (X27, X28), (X29(fp), X30(lr)).  */
3226#define PUSH_SIZE  ((30 - 19 + 1) * 8)
3227
3228#define FRAME_SIZE \
3229    ((PUSH_SIZE \
3230      + TCG_STATIC_CALL_ARGS_SIZE \
3231      + CPU_TEMP_BUF_NLONGS * sizeof(long) \
3232      + TCG_TARGET_STACK_ALIGN - 1) \
3233     & ~(TCG_TARGET_STACK_ALIGN - 1))
3234
3235/* We're expecting a 2 byte uleb128 encoded value.  */
3236QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3237
3238/* We're expecting to use a single ADDI insn.  */
3239QEMU_BUILD_BUG_ON(FRAME_SIZE - PUSH_SIZE > 0xfff);
3240
3241static void tcg_target_qemu_prologue(TCGContext *s)
3242{
3243    TCGReg r;
3244
3245    tcg_out_bti(s, BTI_C);
3246
3247    /* Push (FP, LR) and allocate space for all saved registers.  */
3248    tcg_out_insn(s, 3314, STP, TCG_REG_FP, TCG_REG_LR,
3249                 TCG_REG_SP, -PUSH_SIZE, 1, 1);
3250
3251    /* Set up frame pointer for canonical unwinding.  */
3252    tcg_out_movr_sp(s, TCG_TYPE_I64, TCG_REG_FP, TCG_REG_SP);
3253
3254    /* Store callee-preserved regs x19..x28.  */
3255    for (r = TCG_REG_X19; r <= TCG_REG_X27; r += 2) {
3256        int ofs = (r - TCG_REG_X19 + 2) * 8;
3257        tcg_out_insn(s, 3314, STP, r, r + 1, TCG_REG_SP, ofs, 1, 0);
3258    }
3259
3260    /* Make stack space for TCG locals.  */
3261    tcg_out_insn(s, 3401, SUBI, TCG_TYPE_I64, TCG_REG_SP, TCG_REG_SP,
3262                 FRAME_SIZE - PUSH_SIZE);
3263
3264    /* Inform TCG about how to find TCG locals with register, offset, size.  */
3265    tcg_set_frame(s, TCG_REG_SP, TCG_STATIC_CALL_ARGS_SIZE,
3266                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3267
3268    if (!tcg_use_softmmu) {
3269        /*
3270         * Note that XZR cannot be encoded in the address base register slot,
3271         * as that actually encodes SP.  Depending on the guest, we may need
3272         * to zero-extend the guest address via the address index register slot,
3273         * therefore we need to load even a zero guest base into a register.
3274         */
3275        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_GUEST_BASE, guest_base);
3276        tcg_regset_set_reg(s->reserved_regs, TCG_REG_GUEST_BASE);
3277    }
3278
3279    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3280    tcg_out_insn(s, 3207, BR, tcg_target_call_iarg_regs[1]);
3281
3282    /*
3283     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3284     * and fall through to the rest of the epilogue.
3285     */
3286    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3287    tcg_out_bti(s, BTI_J);
3288    tcg_out_movi(s, TCG_TYPE_REG, TCG_REG_X0, 0);
3289
3290    /* TB epilogue */
3291    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
3292    tcg_out_bti(s, BTI_J);
3293
3294    /* Remove TCG locals stack space.  */
3295    tcg_out_insn(s, 3401, ADDI, TCG_TYPE_I64, TCG_REG_SP, TCG_REG_SP,
3296                 FRAME_SIZE - PUSH_SIZE);
3297
3298    /* Restore registers x19..x28.  */
3299    for (r = TCG_REG_X19; r <= TCG_REG_X27; r += 2) {
3300        int ofs = (r - TCG_REG_X19 + 2) * 8;
3301        tcg_out_insn(s, 3314, LDP, r, r + 1, TCG_REG_SP, ofs, 1, 0);
3302    }
3303
3304    /* Pop (FP, LR), restore SP to previous frame.  */
3305    tcg_out_insn(s, 3314, LDP, TCG_REG_FP, TCG_REG_LR,
3306                 TCG_REG_SP, PUSH_SIZE, 0, 1);
3307    tcg_out_insn(s, 3207, RET, TCG_REG_LR);
3308}
3309
3310static void tcg_out_tb_start(TCGContext *s)
3311{
3312    tcg_out_bti(s, BTI_J);
3313}
3314
3315static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3316{
3317    int i;
3318    for (i = 0; i < count; ++i) {
3319        p[i] = NOP;
3320    }
3321}
3322
3323typedef struct {
3324    DebugFrameHeader h;
3325    uint8_t fde_def_cfa[4];
3326    uint8_t fde_reg_ofs[24];
3327} DebugFrame;
3328
3329#define ELF_HOST_MACHINE EM_AARCH64
3330
3331static const DebugFrame debug_frame = {
3332    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3333    .h.cie.id = -1,
3334    .h.cie.version = 1,
3335    .h.cie.code_align = 1,
3336    .h.cie.data_align = 0x78,             /* sleb128 -8 */
3337    .h.cie.return_column = TCG_REG_LR,
3338
3339    /* Total FDE size does not include the "len" member.  */
3340    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3341
3342    .fde_def_cfa = {
3343        12, TCG_REG_SP,                 /* DW_CFA_def_cfa sp, ... */
3344        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3345        (FRAME_SIZE >> 7)
3346    },
3347    .fde_reg_ofs = {
3348        0x80 + 28, 1,                   /* DW_CFA_offset, x28,  -8 */
3349        0x80 + 27, 2,                   /* DW_CFA_offset, x27, -16 */
3350        0x80 + 26, 3,                   /* DW_CFA_offset, x26, -24 */
3351        0x80 + 25, 4,                   /* DW_CFA_offset, x25, -32 */
3352        0x80 + 24, 5,                   /* DW_CFA_offset, x24, -40 */
3353        0x80 + 23, 6,                   /* DW_CFA_offset, x23, -48 */
3354        0x80 + 22, 7,                   /* DW_CFA_offset, x22, -56 */
3355        0x80 + 21, 8,                   /* DW_CFA_offset, x21, -64 */
3356        0x80 + 20, 9,                   /* DW_CFA_offset, x20, -72 */
3357        0x80 + 19, 10,                  /* DW_CFA_offset, x1p, -80 */
3358        0x80 + 30, 11,                  /* DW_CFA_offset,  lr, -88 */
3359        0x80 + 29, 12,                  /* DW_CFA_offset,  fp, -96 */
3360    }
3361};
3362
3363void tcg_register_jit(const void *buf, size_t buf_size)
3364{
3365    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3366}
3367