xref: /openbmc/qemu/tcg/aarch64/tcg-target.c.inc (revision 74dbd36f1f87bd7fc4705644d63c5561a23b0567)
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_sar(TCGContext *s, TCGType ext,
1351                               TCGReg rd, TCGReg rn, unsigned int m)
1352{
1353    int max = ext ? 63 : 31;
1354    tcg_out_sbfm(s, ext, rd, rn, m & max, max);
1355}
1356
1357static inline void tcg_out_rotr(TCGContext *s, TCGType ext,
1358                                TCGReg rd, TCGReg rn, unsigned int m)
1359{
1360    int max = ext ? 63 : 31;
1361    tcg_out_extr(s, ext, rd, rn, rn, m & max);
1362}
1363
1364static inline void tcg_out_rotl(TCGContext *s, TCGType ext,
1365                                TCGReg rd, TCGReg rn, unsigned int m)
1366{
1367    int max = ext ? 63 : 31;
1368    tcg_out_extr(s, ext, rd, rn, rn, -m & max);
1369}
1370
1371static inline void tcg_out_dep(TCGContext *s, TCGType ext, TCGReg rd,
1372                               TCGReg rn, unsigned lsb, unsigned width)
1373{
1374    unsigned size = ext ? 64 : 32;
1375    unsigned a = (size - lsb) & (size - 1);
1376    unsigned b = width - 1;
1377    tcg_out_bfm(s, ext, rd, rn, a, b);
1378}
1379
1380static void tcg_out_cmp(TCGContext *s, TCGType ext, TCGCond cond, TCGReg a,
1381                        tcg_target_long b, bool const_b)
1382{
1383    if (is_tst_cond(cond)) {
1384        if (!const_b) {
1385            tcg_out_insn(s, 3510, ANDS, ext, TCG_REG_XZR, a, b);
1386        } else {
1387            tcg_out_logicali(s, I3404_ANDSI, ext, TCG_REG_XZR, a, b);
1388        }
1389    } else {
1390        if (!const_b) {
1391            tcg_out_insn(s, 3502, SUBS, ext, TCG_REG_XZR, a, b);
1392        } else if (b >= 0) {
1393            tcg_debug_assert(is_aimm(b));
1394            tcg_out_insn(s, 3401, SUBSI, ext, TCG_REG_XZR, a, b);
1395        } else {
1396            tcg_debug_assert(is_aimm(-b));
1397            tcg_out_insn(s, 3401, ADDSI, ext, TCG_REG_XZR, a, -b);
1398        }
1399    }
1400}
1401
1402static void tcg_out_goto(TCGContext *s, const tcg_insn_unit *target)
1403{
1404    ptrdiff_t offset = tcg_pcrel_diff(s, target) >> 2;
1405    tcg_debug_assert(offset == sextract64(offset, 0, 26));
1406    tcg_out_insn(s, 3206, B, offset);
1407}
1408
1409static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *target)
1410{
1411    ptrdiff_t offset = tcg_pcrel_diff(s, target) >> 2;
1412    if (offset == sextract64(offset, 0, 26)) {
1413        tcg_out_insn(s, 3206, BL, offset);
1414    } else {
1415        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_TMP0, (intptr_t)target);
1416        tcg_out_insn(s, 3207, BLR, TCG_REG_TMP0);
1417    }
1418}
1419
1420static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target,
1421                         const TCGHelperInfo *info)
1422{
1423    tcg_out_call_int(s, target);
1424}
1425
1426static inline void tcg_out_goto_label(TCGContext *s, TCGLabel *l)
1427{
1428    if (!l->has_value) {
1429        tcg_out_reloc(s, s->code_ptr, R_AARCH64_JUMP26, l, 0);
1430        tcg_out_insn(s, 3206, B, 0);
1431    } else {
1432        tcg_out_goto(s, l->u.value_ptr);
1433    }
1434}
1435
1436static void tcg_out_brcond(TCGContext *s, TCGType ext, TCGCond c, TCGArg a,
1437                           TCGArg b, bool b_const, TCGLabel *l)
1438{
1439    int tbit = -1;
1440    bool need_cmp = true;
1441
1442    switch (c) {
1443    case TCG_COND_EQ:
1444    case TCG_COND_NE:
1445        /* cmp xN,0; b.ne L -> cbnz xN,L */
1446        if (b_const && b == 0) {
1447            need_cmp = false;
1448        }
1449        break;
1450    case TCG_COND_LT:
1451    case TCG_COND_GE:
1452        /* cmp xN,0; b.mi L -> tbnz xN,63,L */
1453        if (b_const && b == 0) {
1454            c = (c == TCG_COND_LT ? TCG_COND_TSTNE : TCG_COND_TSTEQ);
1455            tbit = ext ? 63 : 31;
1456            need_cmp = false;
1457        }
1458        break;
1459    case TCG_COND_TSTEQ:
1460    case TCG_COND_TSTNE:
1461        /* tst xN,0xffffffff; b.ne L -> cbnz wN,L */
1462        if (b_const && b == UINT32_MAX) {
1463            c = tcg_tst_eqne_cond(c);
1464            ext = TCG_TYPE_I32;
1465            need_cmp = false;
1466            break;
1467        }
1468        /* tst xN,1<<B; b.ne L -> tbnz xN,B,L */
1469        if (b_const && is_power_of_2(b)) {
1470            tbit = ctz64(b);
1471            need_cmp = false;
1472        }
1473        break;
1474    default:
1475        break;
1476    }
1477
1478    if (need_cmp) {
1479        tcg_out_cmp(s, ext, c, a, b, b_const);
1480        tcg_out_reloc(s, s->code_ptr, R_AARCH64_CONDBR19, l, 0);
1481        tcg_out_insn(s, 3202, B_C, c, 0);
1482        return;
1483    }
1484
1485    if (tbit >= 0) {
1486        tcg_out_reloc(s, s->code_ptr, R_AARCH64_TSTBR14, l, 0);
1487        switch (c) {
1488        case TCG_COND_TSTEQ:
1489            tcg_out_insn(s, 3205, TBZ, a, tbit, 0);
1490            break;
1491        case TCG_COND_TSTNE:
1492            tcg_out_insn(s, 3205, TBNZ, a, tbit, 0);
1493            break;
1494        default:
1495            g_assert_not_reached();
1496        }
1497    } else {
1498        tcg_out_reloc(s, s->code_ptr, R_AARCH64_CONDBR19, l, 0);
1499        switch (c) {
1500        case TCG_COND_EQ:
1501            tcg_out_insn(s, 3201, CBZ, ext, a, 0);
1502            break;
1503        case TCG_COND_NE:
1504            tcg_out_insn(s, 3201, CBNZ, ext, a, 0);
1505            break;
1506        default:
1507            g_assert_not_reached();
1508        }
1509    }
1510}
1511
1512static inline void tcg_out_rev(TCGContext *s, int ext, MemOp s_bits,
1513                               TCGReg rd, TCGReg rn)
1514{
1515    /* REV, REV16, REV32 */
1516    tcg_out_insn_3507(s, I3507_REV | (s_bits << 10), ext, rd, rn);
1517}
1518
1519static inline void tcg_out_sxt(TCGContext *s, TCGType ext, MemOp s_bits,
1520                               TCGReg rd, TCGReg rn)
1521{
1522    /* Using ALIASes SXTB, SXTH, SXTW, of SBFM Xd, Xn, #0, #7|15|31 */
1523    int bits = (8 << s_bits) - 1;
1524    tcg_out_sbfm(s, ext, rd, rn, 0, bits);
1525}
1526
1527static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rn)
1528{
1529    tcg_out_sxt(s, type, MO_8, rd, rn);
1530}
1531
1532static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rn)
1533{
1534    tcg_out_sxt(s, type, MO_16, rd, rn);
1535}
1536
1537static void tcg_out_ext32s(TCGContext *s, TCGReg rd, TCGReg rn)
1538{
1539    tcg_out_sxt(s, TCG_TYPE_I64, MO_32, rd, rn);
1540}
1541
1542static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg rd, TCGReg rn)
1543{
1544    tcg_out_ext32s(s, rd, rn);
1545}
1546
1547static inline void tcg_out_uxt(TCGContext *s, MemOp s_bits,
1548                               TCGReg rd, TCGReg rn)
1549{
1550    /* Using ALIASes UXTB, UXTH of UBFM Wd, Wn, #0, #7|15 */
1551    int bits = (8 << s_bits) - 1;
1552    tcg_out_ubfm(s, 0, rd, rn, 0, bits);
1553}
1554
1555static void tcg_out_ext8u(TCGContext *s, TCGReg rd, TCGReg rn)
1556{
1557    tcg_out_uxt(s, MO_8, rd, rn);
1558}
1559
1560static void tcg_out_ext16u(TCGContext *s, TCGReg rd, TCGReg rn)
1561{
1562    tcg_out_uxt(s, MO_16, rd, rn);
1563}
1564
1565static void tcg_out_ext32u(TCGContext *s, TCGReg rd, TCGReg rn)
1566{
1567    tcg_out_movr(s, TCG_TYPE_I32, rd, rn);
1568}
1569
1570static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg rd, TCGReg rn)
1571{
1572    tcg_out_ext32u(s, rd, rn);
1573}
1574
1575static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rn)
1576{
1577    tcg_out_mov(s, TCG_TYPE_I32, rd, rn);
1578}
1579
1580static void tcg_out_addsub2(TCGContext *s, TCGType ext, TCGReg rl,
1581                            TCGReg rh, TCGReg al, TCGReg ah,
1582                            tcg_target_long bl, tcg_target_long bh,
1583                            bool const_bl, bool const_bh, bool sub)
1584{
1585    TCGReg orig_rl = rl;
1586    AArch64Insn insn;
1587
1588    if (rl == ah || (!const_bh && rl == bh)) {
1589        rl = TCG_REG_TMP0;
1590    }
1591
1592    if (const_bl) {
1593        if (bl < 0) {
1594            bl = -bl;
1595            insn = sub ? I3401_ADDSI : I3401_SUBSI;
1596        } else {
1597            insn = sub ? I3401_SUBSI : I3401_ADDSI;
1598        }
1599
1600        if (unlikely(al == TCG_REG_XZR)) {
1601            /* ??? We want to allow al to be zero for the benefit of
1602               negation via subtraction.  However, that leaves open the
1603               possibility of adding 0+const in the low part, and the
1604               immediate add instructions encode XSP not XZR.  Don't try
1605               anything more elaborate here than loading another zero.  */
1606            al = TCG_REG_TMP0;
1607            tcg_out_movi(s, ext, al, 0);
1608        }
1609        tcg_out_insn_3401(s, insn, ext, rl, al, bl);
1610    } else {
1611        tcg_out_insn_3502(s, sub ? I3502_SUBS : I3502_ADDS, ext, rl, al, bl);
1612    }
1613
1614    insn = I3503_ADC;
1615    if (const_bh) {
1616        /* Note that the only two constants we support are 0 and -1, and
1617           that SBC = rn + ~rm + c, so adc -1 is sbc 0, and vice-versa.  */
1618        if ((bh != 0) ^ sub) {
1619            insn = I3503_SBC;
1620        }
1621        bh = TCG_REG_XZR;
1622    } else if (sub) {
1623        insn = I3503_SBC;
1624    }
1625    tcg_out_insn_3503(s, insn, ext, rh, ah, bh);
1626
1627    tcg_out_mov(s, ext, orig_rl, rl);
1628}
1629
1630static inline void tcg_out_mb(TCGContext *s, TCGArg a0)
1631{
1632    static const uint32_t sync[] = {
1633        [0 ... TCG_MO_ALL]            = DMB_ISH | DMB_LD | DMB_ST,
1634        [TCG_MO_ST_ST]                = DMB_ISH | DMB_ST,
1635        [TCG_MO_LD_LD]                = DMB_ISH | DMB_LD,
1636        [TCG_MO_LD_ST]                = DMB_ISH | DMB_LD,
1637        [TCG_MO_LD_ST | TCG_MO_LD_LD] = DMB_ISH | DMB_LD,
1638    };
1639    tcg_out32(s, sync[a0 & TCG_MO_ALL]);
1640}
1641
1642static void tcg_out_cltz(TCGContext *s, TCGType ext, TCGReg d,
1643                         TCGReg a0, TCGArg b, bool const_b, bool is_ctz)
1644{
1645    TCGReg a1 = a0;
1646    if (is_ctz) {
1647        a1 = TCG_REG_TMP0;
1648        tcg_out_insn(s, 3507, RBIT, ext, a1, a0);
1649    }
1650    if (const_b && b == (ext ? 64 : 32)) {
1651        tcg_out_insn(s, 3507, CLZ, ext, d, a1);
1652    } else {
1653        AArch64Insn sel = I3506_CSEL;
1654
1655        tcg_out_cmp(s, ext, TCG_COND_NE, a0, 0, 1);
1656        tcg_out_insn(s, 3507, CLZ, ext, TCG_REG_TMP0, a1);
1657
1658        if (const_b) {
1659            if (b == -1) {
1660                b = TCG_REG_XZR;
1661                sel = I3506_CSINV;
1662            } else if (b == 0) {
1663                b = TCG_REG_XZR;
1664            } else {
1665                tcg_out_movi(s, ext, d, b);
1666                b = d;
1667            }
1668        }
1669        tcg_out_insn_3506(s, sel, ext, d, TCG_REG_TMP0, b, TCG_COND_NE);
1670    }
1671}
1672
1673typedef struct {
1674    TCGReg base;
1675    TCGReg index;
1676    TCGType index_ext;
1677    TCGAtomAlign aa;
1678} HostAddress;
1679
1680bool tcg_target_has_memory_bswap(MemOp memop)
1681{
1682    return false;
1683}
1684
1685static const TCGLdstHelperParam ldst_helper_param = {
1686    .ntmp = 1, .tmp = { TCG_REG_TMP0 }
1687};
1688
1689static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1690{
1691    MemOp opc = get_memop(lb->oi);
1692
1693    if (!reloc_pc19(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1694        return false;
1695    }
1696
1697    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1698    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1699    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1700    tcg_out_goto(s, lb->raddr);
1701    return true;
1702}
1703
1704static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1705{
1706    MemOp opc = get_memop(lb->oi);
1707
1708    if (!reloc_pc19(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1709        return false;
1710    }
1711
1712    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1713    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE]);
1714    tcg_out_goto(s, lb->raddr);
1715    return true;
1716}
1717
1718/* We expect to use a 7-bit scaled negative offset from ENV.  */
1719#define MIN_TLB_MASK_TABLE_OFS  -512
1720
1721/*
1722 * For system-mode, perform the TLB load and compare.
1723 * For user-mode, perform any required alignment tests.
1724 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1725 * is required and fill in @h with the host address for the fast path.
1726 */
1727static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1728                                           TCGReg addr_reg, MemOpIdx oi,
1729                                           bool is_ld)
1730{
1731    TCGType addr_type = s->addr_type;
1732    TCGLabelQemuLdst *ldst = NULL;
1733    MemOp opc = get_memop(oi);
1734    MemOp s_bits = opc & MO_SIZE;
1735    unsigned a_mask;
1736
1737    h->aa = atom_and_align_for_opc(s, opc,
1738                                   have_lse2 ? MO_ATOM_WITHIN16
1739                                             : MO_ATOM_IFALIGN,
1740                                   s_bits == MO_128);
1741    a_mask = (1 << h->aa.align) - 1;
1742
1743    if (tcg_use_softmmu) {
1744        unsigned s_mask = (1u << s_bits) - 1;
1745        unsigned mem_index = get_mmuidx(oi);
1746        TCGReg addr_adj;
1747        TCGType mask_type;
1748        uint64_t compare_mask;
1749
1750        ldst = new_ldst_label(s);
1751        ldst->is_ld = is_ld;
1752        ldst->oi = oi;
1753        ldst->addr_reg = addr_reg;
1754
1755        mask_type = (s->page_bits + s->tlb_dyn_max_bits > 32
1756                     ? TCG_TYPE_I64 : TCG_TYPE_I32);
1757
1758        /* Load cpu->neg.tlb.f[mmu_idx].{mask,table} into {tmp0,tmp1}. */
1759        QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, mask) != 0);
1760        QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, table) != 8);
1761        tcg_out_insn(s, 3314, LDP, TCG_REG_TMP0, TCG_REG_TMP1, TCG_AREG0,
1762                     tlb_mask_table_ofs(s, mem_index), 1, 0);
1763
1764        /* Extract the TLB index from the address into X0.  */
1765        tcg_out_insn(s, 3502S, AND_LSR, mask_type == TCG_TYPE_I64,
1766                     TCG_REG_TMP0, TCG_REG_TMP0, addr_reg,
1767                     s->page_bits - CPU_TLB_ENTRY_BITS);
1768
1769        /* Add the tlb_table pointer, forming the CPUTLBEntry address. */
1770        tcg_out_insn(s, 3502, ADD, 1, TCG_REG_TMP1, TCG_REG_TMP1, TCG_REG_TMP0);
1771
1772        /* Load the tlb comparator into TMP0, and the fast path addend. */
1773        QEMU_BUILD_BUG_ON(HOST_BIG_ENDIAN);
1774        tcg_out_ld(s, addr_type, TCG_REG_TMP0, TCG_REG_TMP1,
1775                   is_ld ? offsetof(CPUTLBEntry, addr_read)
1776                         : offsetof(CPUTLBEntry, addr_write));
1777        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_REG_TMP1,
1778                   offsetof(CPUTLBEntry, addend));
1779
1780        /*
1781         * For aligned accesses, we check the first byte and include
1782         * the alignment bits within the address.  For unaligned access,
1783         * we check that we don't cross pages using the address of the
1784         * last byte of the access.
1785         */
1786        if (a_mask >= s_mask) {
1787            addr_adj = addr_reg;
1788        } else {
1789            addr_adj = TCG_REG_TMP2;
1790            tcg_out_insn(s, 3401, ADDI, addr_type,
1791                         addr_adj, addr_reg, s_mask - a_mask);
1792        }
1793        compare_mask = (uint64_t)s->page_mask | a_mask;
1794
1795        /* Store the page mask part of the address into TMP2.  */
1796        tcg_out_logicali(s, I3404_ANDI, addr_type, TCG_REG_TMP2,
1797                         addr_adj, compare_mask);
1798
1799        /* Perform the address comparison. */
1800        tcg_out_cmp(s, addr_type, TCG_COND_NE, TCG_REG_TMP0, TCG_REG_TMP2, 0);
1801
1802        /* If not equal, we jump to the slow path. */
1803        ldst->label_ptr[0] = s->code_ptr;
1804        tcg_out_insn(s, 3202, B_C, TCG_COND_NE, 0);
1805
1806        h->base = TCG_REG_TMP1;
1807        h->index = addr_reg;
1808        h->index_ext = addr_type;
1809    } else {
1810        if (a_mask) {
1811            ldst = new_ldst_label(s);
1812
1813            ldst->is_ld = is_ld;
1814            ldst->oi = oi;
1815            ldst->addr_reg = addr_reg;
1816
1817            /* tst addr, #mask */
1818            tcg_out_logicali(s, I3404_ANDSI, 0, TCG_REG_XZR, addr_reg, a_mask);
1819
1820            /* b.ne slow_path */
1821            ldst->label_ptr[0] = s->code_ptr;
1822            tcg_out_insn(s, 3202, B_C, TCG_COND_NE, 0);
1823        }
1824
1825        if (guest_base || addr_type == TCG_TYPE_I32) {
1826            h->base = TCG_REG_GUEST_BASE;
1827            h->index = addr_reg;
1828            h->index_ext = addr_type;
1829        } else {
1830            h->base = addr_reg;
1831            h->index = TCG_REG_XZR;
1832            h->index_ext = TCG_TYPE_I64;
1833        }
1834    }
1835
1836    return ldst;
1837}
1838
1839static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp memop, TCGType ext,
1840                                   TCGReg data_r, HostAddress h)
1841{
1842    switch (memop & MO_SSIZE) {
1843    case MO_UB:
1844        tcg_out_ldst_r(s, I3312_LDRB, data_r, h.base, h.index_ext, h.index);
1845        break;
1846    case MO_SB:
1847        tcg_out_ldst_r(s, ext ? I3312_LDRSBX : I3312_LDRSBW,
1848                       data_r, h.base, h.index_ext, h.index);
1849        break;
1850    case MO_UW:
1851        tcg_out_ldst_r(s, I3312_LDRH, data_r, h.base, h.index_ext, h.index);
1852        break;
1853    case MO_SW:
1854        tcg_out_ldst_r(s, (ext ? I3312_LDRSHX : I3312_LDRSHW),
1855                       data_r, h.base, h.index_ext, h.index);
1856        break;
1857    case MO_UL:
1858        tcg_out_ldst_r(s, I3312_LDRW, data_r, h.base, h.index_ext, h.index);
1859        break;
1860    case MO_SL:
1861        tcg_out_ldst_r(s, I3312_LDRSWX, data_r, h.base, h.index_ext, h.index);
1862        break;
1863    case MO_UQ:
1864        tcg_out_ldst_r(s, I3312_LDRX, data_r, h.base, h.index_ext, h.index);
1865        break;
1866    default:
1867        g_assert_not_reached();
1868    }
1869}
1870
1871static void tcg_out_qemu_st_direct(TCGContext *s, MemOp memop,
1872                                   TCGReg data_r, HostAddress h)
1873{
1874    switch (memop & MO_SIZE) {
1875    case MO_8:
1876        tcg_out_ldst_r(s, I3312_STRB, data_r, h.base, h.index_ext, h.index);
1877        break;
1878    case MO_16:
1879        tcg_out_ldst_r(s, I3312_STRH, data_r, h.base, h.index_ext, h.index);
1880        break;
1881    case MO_32:
1882        tcg_out_ldst_r(s, I3312_STRW, data_r, h.base, h.index_ext, h.index);
1883        break;
1884    case MO_64:
1885        tcg_out_ldst_r(s, I3312_STRX, data_r, h.base, h.index_ext, h.index);
1886        break;
1887    default:
1888        g_assert_not_reached();
1889    }
1890}
1891
1892static void tcg_out_qemu_ld(TCGContext *s, TCGReg data_reg, TCGReg addr_reg,
1893                            MemOpIdx oi, TCGType data_type)
1894{
1895    TCGLabelQemuLdst *ldst;
1896    HostAddress h;
1897
1898    ldst = prepare_host_addr(s, &h, addr_reg, oi, true);
1899    tcg_out_qemu_ld_direct(s, get_memop(oi), data_type, data_reg, h);
1900
1901    if (ldst) {
1902        ldst->type = data_type;
1903        ldst->datalo_reg = data_reg;
1904        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1905    }
1906}
1907
1908static void tcg_out_qemu_st(TCGContext *s, TCGReg data_reg, TCGReg addr_reg,
1909                            MemOpIdx oi, TCGType data_type)
1910{
1911    TCGLabelQemuLdst *ldst;
1912    HostAddress h;
1913
1914    ldst = prepare_host_addr(s, &h, addr_reg, oi, false);
1915    tcg_out_qemu_st_direct(s, get_memop(oi), data_reg, h);
1916
1917    if (ldst) {
1918        ldst->type = data_type;
1919        ldst->datalo_reg = data_reg;
1920        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1921    }
1922}
1923
1924static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi,
1925                                   TCGReg addr_reg, MemOpIdx oi, bool is_ld)
1926{
1927    TCGLabelQemuLdst *ldst;
1928    HostAddress h;
1929    TCGReg base;
1930    bool use_pair;
1931
1932    ldst = prepare_host_addr(s, &h, addr_reg, oi, is_ld);
1933
1934    /* Compose the final address, as LDP/STP have no indexing. */
1935    if (h.index == TCG_REG_XZR) {
1936        base = h.base;
1937    } else {
1938        base = TCG_REG_TMP2;
1939        if (h.index_ext == TCG_TYPE_I32) {
1940            /* add base, base, index, uxtw */
1941            tcg_out_insn(s, 3501, ADD, TCG_TYPE_I64, base,
1942                         h.base, h.index, MO_32, 0);
1943        } else {
1944            /* add base, base, index */
1945            tcg_out_insn(s, 3502, ADD, 1, base, h.base, h.index);
1946        }
1947    }
1948
1949    use_pair = h.aa.atom < MO_128 || have_lse2;
1950
1951    if (!use_pair) {
1952        tcg_insn_unit *branch = NULL;
1953        TCGReg ll, lh, sl, sh;
1954
1955        /*
1956         * If we have already checked for 16-byte alignment, that's all
1957         * we need. Otherwise we have determined that misaligned atomicity
1958         * may be handled with two 8-byte loads.
1959         */
1960        if (h.aa.align < MO_128) {
1961            /*
1962             * TODO: align should be MO_64, so we only need test bit 3,
1963             * which means we could use TBNZ instead of ANDS+B_C.
1964             */
1965            tcg_out_logicali(s, I3404_ANDSI, 0, TCG_REG_XZR, addr_reg, 15);
1966            branch = s->code_ptr;
1967            tcg_out_insn(s, 3202, B_C, TCG_COND_NE, 0);
1968            use_pair = true;
1969        }
1970
1971        if (is_ld) {
1972            /*
1973             * 16-byte atomicity without LSE2 requires LDXP+STXP loop:
1974             *    ldxp lo, hi, [base]
1975             *    stxp t0, lo, hi, [base]
1976             *    cbnz t0, .-8
1977             * Require no overlap between data{lo,hi} and base.
1978             */
1979            if (datalo == base || datahi == base) {
1980                tcg_out_mov(s, TCG_TYPE_REG, TCG_REG_TMP2, base);
1981                base = TCG_REG_TMP2;
1982            }
1983            ll = sl = datalo;
1984            lh = sh = datahi;
1985        } else {
1986            /*
1987             * 16-byte atomicity without LSE2 requires LDXP+STXP loop:
1988             * 1: ldxp t0, t1, [base]
1989             *    stxp t0, lo, hi, [base]
1990             *    cbnz t0, 1b
1991             */
1992            tcg_debug_assert(base != TCG_REG_TMP0 && base != TCG_REG_TMP1);
1993            ll = TCG_REG_TMP0;
1994            lh = TCG_REG_TMP1;
1995            sl = datalo;
1996            sh = datahi;
1997        }
1998
1999        tcg_out_insn(s, 3306, LDXP, TCG_REG_XZR, ll, lh, base);
2000        tcg_out_insn(s, 3306, STXP, TCG_REG_TMP0, sl, sh, base);
2001        tcg_out_insn(s, 3201, CBNZ, 0, TCG_REG_TMP0, -2);
2002
2003        if (use_pair) {
2004            /* "b .+8", branching across the one insn of use_pair. */
2005            tcg_out_insn(s, 3206, B, 2);
2006            reloc_pc19(branch, tcg_splitwx_to_rx(s->code_ptr));
2007        }
2008    }
2009
2010    if (use_pair) {
2011        if (is_ld) {
2012            tcg_out_insn(s, 3314, LDP, datalo, datahi, base, 0, 1, 0);
2013        } else {
2014            tcg_out_insn(s, 3314, STP, datalo, datahi, base, 0, 1, 0);
2015        }
2016    }
2017
2018    if (ldst) {
2019        ldst->type = TCG_TYPE_I128;
2020        ldst->datalo_reg = datalo;
2021        ldst->datahi_reg = datahi;
2022        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2023    }
2024}
2025
2026static const tcg_insn_unit *tb_ret_addr;
2027
2028static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
2029{
2030    const tcg_insn_unit *target;
2031    ptrdiff_t offset;
2032
2033    /* Reuse the zeroing that exists for goto_ptr.  */
2034    if (a0 == 0) {
2035        target = tcg_code_gen_epilogue;
2036    } else {
2037        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_X0, a0);
2038        target = tb_ret_addr;
2039    }
2040
2041    offset = tcg_pcrel_diff(s, target) >> 2;
2042    if (offset == sextract64(offset, 0, 26)) {
2043        tcg_out_insn(s, 3206, B, offset);
2044    } else {
2045        /*
2046         * Only x16/x17 generate BTI type Jump (2),
2047         * other registers generate BTI type Jump|Call (3).
2048         */
2049        QEMU_BUILD_BUG_ON(TCG_REG_TMP0 != TCG_REG_X16);
2050        tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_TMP0, (intptr_t)target);
2051        tcg_out_insn(s, 3207, BR, TCG_REG_TMP0);
2052    }
2053}
2054
2055static void tcg_out_goto_tb(TCGContext *s, int which)
2056{
2057    /*
2058     * Direct branch, or indirect address load, will be patched
2059     * by tb_target_set_jmp_target.  Assert indirect load offset
2060     * in range early, regardless of direct branch distance.
2061     */
2062    intptr_t i_off = tcg_pcrel_diff(s, (void *)get_jmp_target_addr(s, which));
2063    tcg_debug_assert(i_off == sextract64(i_off, 0, 21));
2064
2065    set_jmp_insn_offset(s, which);
2066    tcg_out32(s, I3206_B);
2067    tcg_out_insn(s, 3207, BR, TCG_REG_TMP0);
2068    set_jmp_reset_offset(s, which);
2069    tcg_out_bti(s, BTI_J);
2070}
2071
2072void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
2073                              uintptr_t jmp_rx, uintptr_t jmp_rw)
2074{
2075    uintptr_t d_addr = tb->jmp_target_addr[n];
2076    ptrdiff_t d_offset = d_addr - jmp_rx;
2077    tcg_insn_unit insn;
2078
2079    /* Either directly branch, or indirect branch load. */
2080    if (d_offset == sextract64(d_offset, 0, 28)) {
2081        insn = deposit32(I3206_B, 0, 26, d_offset >> 2);
2082    } else {
2083        uintptr_t i_addr = (uintptr_t)&tb->jmp_target_addr[n];
2084        ptrdiff_t i_offset = i_addr - jmp_rx;
2085
2086        /* Note that we asserted this in range in tcg_out_goto_tb. */
2087        insn = deposit32(I3305_LDR | TCG_REG_TMP0, 5, 19, i_offset >> 2);
2088    }
2089    qatomic_set((uint32_t *)jmp_rw, insn);
2090    flush_idcache_range(jmp_rx, jmp_rw, 4);
2091}
2092
2093
2094static void tgen_add(TCGContext *s, TCGType type,
2095                     TCGReg a0, TCGReg a1, TCGReg a2)
2096{
2097    tcg_out_insn(s, 3502, ADD, type, a0, a1, a2);
2098}
2099
2100static void tgen_addi(TCGContext *s, TCGType type,
2101                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2102{
2103    if (a2 >= 0) {
2104        tcg_out_insn(s, 3401, ADDI, type, a0, a1, a2);
2105    } else {
2106        tcg_out_insn(s, 3401, SUBI, type, a0, a1, -a2);
2107    }
2108}
2109
2110static const TCGOutOpBinary outop_add = {
2111    .base.static_constraint = C_O1_I2(r, r, rA),
2112    .out_rrr = tgen_add,
2113    .out_rri = tgen_addi,
2114};
2115
2116static void tgen_and(TCGContext *s, TCGType type,
2117                     TCGReg a0, TCGReg a1, TCGReg a2)
2118{
2119    tcg_out_insn(s, 3510, AND, type, a0, a1, a2);
2120}
2121
2122static void tgen_andi(TCGContext *s, TCGType type,
2123                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2124{
2125    tcg_out_logicali(s, I3404_ANDI, type, a0, a1, a2);
2126}
2127
2128static const TCGOutOpBinary outop_and = {
2129    .base.static_constraint = C_O1_I2(r, r, rL),
2130    .out_rrr = tgen_and,
2131    .out_rri = tgen_andi,
2132};
2133
2134static void tgen_andc(TCGContext *s, TCGType type,
2135                      TCGReg a0, TCGReg a1, TCGReg a2)
2136{
2137    tcg_out_insn(s, 3510, BIC, type, a0, a1, a2);
2138}
2139
2140static const TCGOutOpBinary outop_andc = {
2141    .base.static_constraint = C_O1_I2(r, r, r),
2142    .out_rrr = tgen_andc,
2143};
2144
2145static void tgen_divs(TCGContext *s, TCGType type,
2146                      TCGReg a0, TCGReg a1, TCGReg a2)
2147{
2148    tcg_out_insn(s, 3508, SDIV, type, a0, a1, a2);
2149}
2150
2151static const TCGOutOpBinary outop_divs = {
2152    .base.static_constraint = C_O1_I2(r, r, r),
2153    .out_rrr = tgen_divs,
2154};
2155
2156static const TCGOutOpDivRem outop_divs2 = {
2157    .base.static_constraint = C_NotImplemented,
2158};
2159
2160static void tgen_divu(TCGContext *s, TCGType type,
2161                      TCGReg a0, TCGReg a1, TCGReg a2)
2162{
2163    tcg_out_insn(s, 3508, UDIV, type, a0, a1, a2);
2164}
2165
2166static const TCGOutOpBinary outop_divu = {
2167    .base.static_constraint = C_O1_I2(r, r, r),
2168    .out_rrr = tgen_divu,
2169};
2170
2171static const TCGOutOpDivRem outop_divu2 = {
2172    .base.static_constraint = C_NotImplemented,
2173};
2174
2175static void tgen_eqv(TCGContext *s, TCGType type,
2176                     TCGReg a0, TCGReg a1, TCGReg a2)
2177{
2178    tcg_out_insn(s, 3510, EON, type, a0, a1, a2);
2179}
2180
2181static const TCGOutOpBinary outop_eqv = {
2182    .base.static_constraint = C_O1_I2(r, r, r),
2183    .out_rrr = tgen_eqv,
2184};
2185
2186static void tgen_mul(TCGContext *s, TCGType type,
2187                     TCGReg a0, TCGReg a1, TCGReg a2)
2188{
2189    tcg_out_insn(s, 3509, MADD, type, a0, a1, a2, TCG_REG_XZR);
2190}
2191
2192static const TCGOutOpBinary outop_mul = {
2193    .base.static_constraint = C_O1_I2(r, r, r),
2194    .out_rrr = tgen_mul,
2195};
2196
2197static TCGConstraintSetIndex cset_mulh(TCGType type, unsigned flags)
2198{
2199    return type == TCG_TYPE_I64 ? C_O1_I2(r, r, r) : C_NotImplemented;
2200}
2201
2202static void tgen_mulsh(TCGContext *s, TCGType type,
2203                       TCGReg a0, TCGReg a1, TCGReg a2)
2204{
2205    tcg_out_insn(s, 3508, SMULH, TCG_TYPE_I64, a0, a1, a2);
2206}
2207
2208static const TCGOutOpBinary outop_mulsh = {
2209    .base.static_constraint = C_Dynamic,
2210    .base.dynamic_constraint = cset_mulh,
2211    .out_rrr = tgen_mulsh,
2212};
2213
2214static void tgen_muluh(TCGContext *s, TCGType type,
2215                       TCGReg a0, TCGReg a1, TCGReg a2)
2216{
2217    tcg_out_insn(s, 3508, UMULH, TCG_TYPE_I64, a0, a1, a2);
2218}
2219
2220static const TCGOutOpBinary outop_muluh = {
2221    .base.static_constraint = C_Dynamic,
2222    .base.dynamic_constraint = cset_mulh,
2223    .out_rrr = tgen_muluh,
2224};
2225
2226static const TCGOutOpBinary outop_nand = {
2227    .base.static_constraint = C_NotImplemented,
2228};
2229
2230static const TCGOutOpBinary outop_nor = {
2231    .base.static_constraint = C_NotImplemented,
2232};
2233
2234static void tgen_or(TCGContext *s, TCGType type,
2235                    TCGReg a0, TCGReg a1, TCGReg a2)
2236{
2237    tcg_out_insn(s, 3510, ORR, type, a0, a1, a2);
2238}
2239
2240static void tgen_ori(TCGContext *s, TCGType type,
2241                     TCGReg a0, TCGReg a1, tcg_target_long a2)
2242{
2243    tcg_out_logicali(s, I3404_ORRI, type, a0, a1, a2);
2244}
2245
2246static const TCGOutOpBinary outop_or = {
2247    .base.static_constraint = C_O1_I2(r, r, rL),
2248    .out_rrr = tgen_or,
2249    .out_rri = tgen_ori,
2250};
2251
2252static void tgen_orc(TCGContext *s, TCGType type,
2253                     TCGReg a0, TCGReg a1, TCGReg a2)
2254{
2255    tcg_out_insn(s, 3510, ORN, type, a0, a1, a2);
2256}
2257
2258static const TCGOutOpBinary outop_orc = {
2259    .base.static_constraint = C_O1_I2(r, r, r),
2260    .out_rrr = tgen_orc,
2261};
2262
2263static void tgen_rems(TCGContext *s, TCGType type,
2264                      TCGReg a0, TCGReg a1, TCGReg a2)
2265{
2266    tcg_out_insn(s, 3508, SDIV, type, TCG_REG_TMP0, a1, a2);
2267    tcg_out_insn(s, 3509, MSUB, type, a0, TCG_REG_TMP0, a2, a1);
2268}
2269
2270static const TCGOutOpBinary outop_rems = {
2271    .base.static_constraint = C_O1_I2(r, r, r),
2272    .out_rrr = tgen_rems,
2273};
2274
2275static void tgen_remu(TCGContext *s, TCGType type,
2276                      TCGReg a0, TCGReg a1, TCGReg a2)
2277{
2278    tcg_out_insn(s, 3508, UDIV, type, TCG_REG_TMP0, a1, a2);
2279    tcg_out_insn(s, 3509, MSUB, type, a0, TCG_REG_TMP0, a2, a1);
2280}
2281
2282static const TCGOutOpBinary outop_remu = {
2283    .base.static_constraint = C_O1_I2(r, r, r),
2284    .out_rrr = tgen_remu,
2285};
2286
2287static void tgen_shl(TCGContext *s, TCGType type,
2288                     TCGReg a0, TCGReg a1, TCGReg a2)
2289{
2290    tcg_out_insn(s, 3508, LSLV, type, a0, a1, a2);
2291}
2292
2293static void tgen_shli(TCGContext *s, TCGType type,
2294                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2295{
2296    int max = type == TCG_TYPE_I32 ? 31 : 63;
2297    tcg_out_ubfm(s, type, a0, a1, -a2 & max, ~a2 & max);
2298}
2299
2300static const TCGOutOpBinary outop_shl = {
2301    .base.static_constraint = C_O1_I2(r, r, ri),
2302    .out_rrr = tgen_shl,
2303    .out_rri = tgen_shli,
2304};
2305
2306static void tgen_shr(TCGContext *s, TCGType type,
2307                     TCGReg a0, TCGReg a1, TCGReg a2)
2308{
2309    tcg_out_insn(s, 3508, LSRV, type, a0, a1, a2);
2310}
2311
2312static void tgen_shri(TCGContext *s, TCGType type,
2313                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2314{
2315    int max = type == TCG_TYPE_I32 ? 31 : 63;
2316    tcg_out_ubfm(s, type, a0, a1, a2 & max, max);
2317}
2318
2319static const TCGOutOpBinary outop_shr = {
2320    .base.static_constraint = C_O1_I2(r, r, ri),
2321    .out_rrr = tgen_shr,
2322    .out_rri = tgen_shri,
2323};
2324
2325static void tgen_sub(TCGContext *s, TCGType type,
2326                     TCGReg a0, TCGReg a1, TCGReg a2)
2327{
2328    tcg_out_insn(s, 3502, SUB, type, a0, a1, a2);
2329}
2330
2331static const TCGOutOpSubtract outop_sub = {
2332    .base.static_constraint = C_O1_I2(r, r, r),
2333    .out_rrr = tgen_sub,
2334};
2335
2336static void tgen_xor(TCGContext *s, TCGType type,
2337                     TCGReg a0, TCGReg a1, TCGReg a2)
2338{
2339    tcg_out_insn(s, 3510, EOR, type, a0, a1, a2);
2340}
2341
2342static void tgen_xori(TCGContext *s, TCGType type,
2343                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2344{
2345    tcg_out_logicali(s, I3404_EORI, type, a0, a1, a2);
2346}
2347
2348static const TCGOutOpBinary outop_xor = {
2349    .base.static_constraint = C_O1_I2(r, r, rL),
2350    .out_rrr = tgen_xor,
2351    .out_rri = tgen_xori,
2352};
2353
2354
2355static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2356{
2357    tgen_sub(s, type, a0, TCG_REG_XZR, a1);
2358}
2359
2360static const TCGOutOpUnary outop_neg = {
2361    .base.static_constraint = C_O1_I1(r, r),
2362    .out_rr = tgen_neg,
2363};
2364
2365static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2366{
2367    tgen_orc(s, type, a0, TCG_REG_XZR, a1);
2368}
2369
2370static const TCGOutOpUnary outop_not = {
2371    .base.static_constraint = C_O1_I1(r, r),
2372    .out_rr = tgen_not,
2373};
2374
2375
2376static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType ext,
2377                       const TCGArg args[TCG_MAX_OP_ARGS],
2378                       const int const_args[TCG_MAX_OP_ARGS])
2379{
2380    /* Hoist the loads of the most common arguments.  */
2381    TCGArg a0 = args[0];
2382    TCGArg a1 = args[1];
2383    TCGArg a2 = args[2];
2384    int c2 = const_args[2];
2385
2386    switch (opc) {
2387    case INDEX_op_goto_ptr:
2388        tcg_out_insn(s, 3207, BR, a0);
2389        break;
2390
2391    case INDEX_op_br:
2392        tcg_out_goto_label(s, arg_label(a0));
2393        break;
2394
2395    case INDEX_op_ld8u_i32:
2396    case INDEX_op_ld8u_i64:
2397        tcg_out_ldst(s, I3312_LDRB, a0, a1, a2, 0);
2398        break;
2399    case INDEX_op_ld8s_i32:
2400        tcg_out_ldst(s, I3312_LDRSBW, a0, a1, a2, 0);
2401        break;
2402    case INDEX_op_ld8s_i64:
2403        tcg_out_ldst(s, I3312_LDRSBX, a0, a1, a2, 0);
2404        break;
2405    case INDEX_op_ld16u_i32:
2406    case INDEX_op_ld16u_i64:
2407        tcg_out_ldst(s, I3312_LDRH, a0, a1, a2, 1);
2408        break;
2409    case INDEX_op_ld16s_i32:
2410        tcg_out_ldst(s, I3312_LDRSHW, a0, a1, a2, 1);
2411        break;
2412    case INDEX_op_ld16s_i64:
2413        tcg_out_ldst(s, I3312_LDRSHX, a0, a1, a2, 1);
2414        break;
2415    case INDEX_op_ld_i32:
2416    case INDEX_op_ld32u_i64:
2417        tcg_out_ldst(s, I3312_LDRW, a0, a1, a2, 2);
2418        break;
2419    case INDEX_op_ld32s_i64:
2420        tcg_out_ldst(s, I3312_LDRSWX, a0, a1, a2, 2);
2421        break;
2422    case INDEX_op_ld_i64:
2423        tcg_out_ldst(s, I3312_LDRX, a0, a1, a2, 3);
2424        break;
2425
2426    case INDEX_op_st8_i32:
2427    case INDEX_op_st8_i64:
2428        tcg_out_ldst(s, I3312_STRB, a0, a1, a2, 0);
2429        break;
2430    case INDEX_op_st16_i32:
2431    case INDEX_op_st16_i64:
2432        tcg_out_ldst(s, I3312_STRH, a0, a1, a2, 1);
2433        break;
2434    case INDEX_op_st_i32:
2435    case INDEX_op_st32_i64:
2436        tcg_out_ldst(s, I3312_STRW, a0, a1, a2, 2);
2437        break;
2438    case INDEX_op_st_i64:
2439        tcg_out_ldst(s, I3312_STRX, a0, a1, a2, 3);
2440        break;
2441
2442    case INDEX_op_sar_i64:
2443    case INDEX_op_sar_i32:
2444        if (c2) {
2445            tcg_out_sar(s, ext, a0, a1, a2);
2446        } else {
2447            tcg_out_insn(s, 3508, ASRV, ext, a0, a1, a2);
2448        }
2449        break;
2450
2451    case INDEX_op_rotr_i64:
2452    case INDEX_op_rotr_i32:
2453        if (c2) {
2454            tcg_out_rotr(s, ext, a0, a1, a2);
2455        } else {
2456            tcg_out_insn(s, 3508, RORV, ext, a0, a1, a2);
2457        }
2458        break;
2459
2460    case INDEX_op_rotl_i64:
2461    case INDEX_op_rotl_i32:
2462        if (c2) {
2463            tcg_out_rotl(s, ext, a0, a1, a2);
2464        } else {
2465            tcg_out_insn(s, 3502, SUB, 0, TCG_REG_TMP0, TCG_REG_XZR, a2);
2466            tcg_out_insn(s, 3508, RORV, ext, a0, a1, TCG_REG_TMP0);
2467        }
2468        break;
2469
2470    case INDEX_op_clz_i64:
2471    case INDEX_op_clz_i32:
2472        tcg_out_cltz(s, ext, a0, a1, a2, c2, false);
2473        break;
2474    case INDEX_op_ctz_i64:
2475    case INDEX_op_ctz_i32:
2476        tcg_out_cltz(s, ext, a0, a1, a2, c2, true);
2477        break;
2478
2479    case INDEX_op_brcond_i32:
2480        a1 = (int32_t)a1;
2481        /* FALLTHRU */
2482    case INDEX_op_brcond_i64:
2483        tcg_out_brcond(s, ext, a2, a0, a1, const_args[1], arg_label(args[3]));
2484        break;
2485
2486    case INDEX_op_setcond_i32:
2487        a2 = (int32_t)a2;
2488        /* FALLTHRU */
2489    case INDEX_op_setcond_i64:
2490        tcg_out_cmp(s, ext, args[3], a1, a2, c2);
2491        /* Use CSET alias of CSINC Wd, WZR, WZR, invert(cond).  */
2492        tcg_out_insn(s, 3506, CSINC, TCG_TYPE_I32, a0, TCG_REG_XZR,
2493                     TCG_REG_XZR, tcg_invert_cond(args[3]));
2494        break;
2495
2496    case INDEX_op_negsetcond_i32:
2497        a2 = (int32_t)a2;
2498        /* FALLTHRU */
2499    case INDEX_op_negsetcond_i64:
2500        tcg_out_cmp(s, ext, args[3], a1, a2, c2);
2501        /* Use CSETM alias of CSINV Wd, WZR, WZR, invert(cond).  */
2502        tcg_out_insn(s, 3506, CSINV, ext, a0, TCG_REG_XZR,
2503                     TCG_REG_XZR, tcg_invert_cond(args[3]));
2504        break;
2505
2506    case INDEX_op_movcond_i32:
2507        a2 = (int32_t)a2;
2508        /* FALLTHRU */
2509    case INDEX_op_movcond_i64:
2510        tcg_out_cmp(s, ext, args[5], a1, a2, c2);
2511        tcg_out_insn(s, 3506, CSEL, ext, a0, args[3], args[4], args[5]);
2512        break;
2513
2514    case INDEX_op_qemu_ld_i32:
2515    case INDEX_op_qemu_ld_i64:
2516        tcg_out_qemu_ld(s, a0, a1, a2, ext);
2517        break;
2518    case INDEX_op_qemu_st_i32:
2519    case INDEX_op_qemu_st_i64:
2520        tcg_out_qemu_st(s, a0, a1, a2, ext);
2521        break;
2522    case INDEX_op_qemu_ld_i128:
2523        tcg_out_qemu_ldst_i128(s, a0, a1, a2, args[3], true);
2524        break;
2525    case INDEX_op_qemu_st_i128:
2526        tcg_out_qemu_ldst_i128(s, a0, a1, a2, args[3], false);
2527        break;
2528
2529    case INDEX_op_bswap64_i64:
2530        tcg_out_rev(s, TCG_TYPE_I64, MO_64, a0, a1);
2531        break;
2532    case INDEX_op_bswap32_i64:
2533        tcg_out_rev(s, TCG_TYPE_I32, MO_32, a0, a1);
2534        if (a2 & TCG_BSWAP_OS) {
2535            tcg_out_ext32s(s, a0, a0);
2536        }
2537        break;
2538    case INDEX_op_bswap32_i32:
2539        tcg_out_rev(s, TCG_TYPE_I32, MO_32, a0, a1);
2540        break;
2541    case INDEX_op_bswap16_i64:
2542    case INDEX_op_bswap16_i32:
2543        tcg_out_rev(s, TCG_TYPE_I32, MO_16, a0, a1);
2544        if (a2 & TCG_BSWAP_OS) {
2545            /* Output must be sign-extended. */
2546            tcg_out_ext16s(s, ext, a0, a0);
2547        } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2548            /* Output must be zero-extended, but input isn't. */
2549            tcg_out_ext16u(s, a0, a0);
2550        }
2551        break;
2552
2553    case INDEX_op_deposit_i64:
2554    case INDEX_op_deposit_i32:
2555        tcg_out_dep(s, ext, a0, a2, args[3], args[4]);
2556        break;
2557
2558    case INDEX_op_extract_i64:
2559    case INDEX_op_extract_i32:
2560        if (a2 == 0) {
2561            uint64_t mask = MAKE_64BIT_MASK(0, args[3]);
2562            tcg_out_logicali(s, I3404_ANDI, ext, a0, a1, mask);
2563        } else {
2564            tcg_out_ubfm(s, ext, a0, a1, a2, a2 + args[3] - 1);
2565        }
2566        break;
2567
2568    case INDEX_op_sextract_i64:
2569    case INDEX_op_sextract_i32:
2570        tcg_out_sbfm(s, ext, a0, a1, a2, a2 + args[3] - 1);
2571        break;
2572
2573    case INDEX_op_extract2_i64:
2574    case INDEX_op_extract2_i32:
2575        tcg_out_extr(s, ext, a0, a2, a1, args[3]);
2576        break;
2577
2578    case INDEX_op_add2_i32:
2579        tcg_out_addsub2(s, TCG_TYPE_I32, a0, a1, a2, args[3],
2580                        (int32_t)args[4], args[5], const_args[4],
2581                        const_args[5], false);
2582        break;
2583    case INDEX_op_add2_i64:
2584        tcg_out_addsub2(s, TCG_TYPE_I64, a0, a1, a2, args[3], args[4],
2585                        args[5], const_args[4], const_args[5], false);
2586        break;
2587    case INDEX_op_sub2_i32:
2588        tcg_out_addsub2(s, TCG_TYPE_I32, a0, a1, a2, args[3],
2589                        (int32_t)args[4], args[5], const_args[4],
2590                        const_args[5], true);
2591        break;
2592    case INDEX_op_sub2_i64:
2593        tcg_out_addsub2(s, TCG_TYPE_I64, a0, a1, a2, args[3], args[4],
2594                        args[5], const_args[4], const_args[5], true);
2595        break;
2596
2597    case INDEX_op_mb:
2598        tcg_out_mb(s, a0);
2599        break;
2600
2601    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2602    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2603    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2604    case INDEX_op_ext_i32_i64:  /* Always emitted via tcg_reg_alloc_op.  */
2605    case INDEX_op_extu_i32_i64:
2606    case INDEX_op_extrl_i64_i32:
2607    default:
2608        g_assert_not_reached();
2609    }
2610}
2611
2612static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2613                           unsigned vecl, unsigned vece,
2614                           const TCGArg args[TCG_MAX_OP_ARGS],
2615                           const int const_args[TCG_MAX_OP_ARGS])
2616{
2617    static const AArch64Insn cmp_vec_insn[16] = {
2618        [TCG_COND_EQ] = I3616_CMEQ,
2619        [TCG_COND_GT] = I3616_CMGT,
2620        [TCG_COND_GE] = I3616_CMGE,
2621        [TCG_COND_GTU] = I3616_CMHI,
2622        [TCG_COND_GEU] = I3616_CMHS,
2623    };
2624    static const AArch64Insn cmp_scalar_insn[16] = {
2625        [TCG_COND_EQ] = I3611_CMEQ,
2626        [TCG_COND_GT] = I3611_CMGT,
2627        [TCG_COND_GE] = I3611_CMGE,
2628        [TCG_COND_GTU] = I3611_CMHI,
2629        [TCG_COND_GEU] = I3611_CMHS,
2630    };
2631    static const AArch64Insn cmp0_vec_insn[16] = {
2632        [TCG_COND_EQ] = I3617_CMEQ0,
2633        [TCG_COND_GT] = I3617_CMGT0,
2634        [TCG_COND_GE] = I3617_CMGE0,
2635        [TCG_COND_LT] = I3617_CMLT0,
2636        [TCG_COND_LE] = I3617_CMLE0,
2637    };
2638    static const AArch64Insn cmp0_scalar_insn[16] = {
2639        [TCG_COND_EQ] = I3612_CMEQ0,
2640        [TCG_COND_GT] = I3612_CMGT0,
2641        [TCG_COND_GE] = I3612_CMGE0,
2642        [TCG_COND_LT] = I3612_CMLT0,
2643        [TCG_COND_LE] = I3612_CMLE0,
2644    };
2645
2646    TCGType type = vecl + TCG_TYPE_V64;
2647    unsigned is_q = vecl;
2648    bool is_scalar = !is_q && vece == MO_64;
2649    TCGArg a0, a1, a2, a3;
2650    int cmode, imm8;
2651
2652    a0 = args[0];
2653    a1 = args[1];
2654    a2 = args[2];
2655
2656    switch (opc) {
2657    case INDEX_op_ld_vec:
2658        tcg_out_ld(s, type, a0, a1, a2);
2659        break;
2660    case INDEX_op_st_vec:
2661        tcg_out_st(s, type, a0, a1, a2);
2662        break;
2663    case INDEX_op_dupm_vec:
2664        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2665        break;
2666    case INDEX_op_add_vec:
2667        if (is_scalar) {
2668            tcg_out_insn(s, 3611, ADD, vece, a0, a1, a2);
2669        } else {
2670            tcg_out_insn(s, 3616, ADD, is_q, vece, a0, a1, a2);
2671        }
2672        break;
2673    case INDEX_op_sub_vec:
2674        if (is_scalar) {
2675            tcg_out_insn(s, 3611, SUB, vece, a0, a1, a2);
2676        } else {
2677            tcg_out_insn(s, 3616, SUB, is_q, vece, a0, a1, a2);
2678        }
2679        break;
2680    case INDEX_op_mul_vec:
2681        tcg_out_insn(s, 3616, MUL, is_q, vece, a0, a1, a2);
2682        break;
2683    case INDEX_op_neg_vec:
2684        if (is_scalar) {
2685            tcg_out_insn(s, 3612, NEG, vece, a0, a1);
2686        } else {
2687            tcg_out_insn(s, 3617, NEG, is_q, vece, a0, a1);
2688        }
2689        break;
2690    case INDEX_op_abs_vec:
2691        if (is_scalar) {
2692            tcg_out_insn(s, 3612, ABS, vece, a0, a1);
2693        } else {
2694            tcg_out_insn(s, 3617, ABS, is_q, vece, a0, a1);
2695        }
2696        break;
2697    case INDEX_op_and_vec:
2698        if (const_args[2]) {
2699            is_shimm1632(~a2, &cmode, &imm8);
2700            if (a0 == a1) {
2701                tcg_out_insn(s, 3606, BIC, is_q, a0, 0, cmode, imm8);
2702                return;
2703            }
2704            tcg_out_insn(s, 3606, MVNI, is_q, a0, 0, cmode, imm8);
2705            a2 = a0;
2706        }
2707        tcg_out_insn(s, 3616, AND, is_q, 0, a0, a1, a2);
2708        break;
2709    case INDEX_op_or_vec:
2710        if (const_args[2]) {
2711            is_shimm1632(a2, &cmode, &imm8);
2712            if (a0 == a1) {
2713                tcg_out_insn(s, 3606, ORR, is_q, a0, 0, cmode, imm8);
2714                return;
2715            }
2716            tcg_out_insn(s, 3606, MOVI, is_q, a0, 0, cmode, imm8);
2717            a2 = a0;
2718        }
2719        tcg_out_insn(s, 3616, ORR, is_q, 0, a0, a1, a2);
2720        break;
2721    case INDEX_op_andc_vec:
2722        if (const_args[2]) {
2723            is_shimm1632(a2, &cmode, &imm8);
2724            if (a0 == a1) {
2725                tcg_out_insn(s, 3606, BIC, is_q, a0, 0, cmode, imm8);
2726                return;
2727            }
2728            tcg_out_insn(s, 3606, MOVI, is_q, a0, 0, cmode, imm8);
2729            a2 = a0;
2730        }
2731        tcg_out_insn(s, 3616, BIC, is_q, 0, a0, a1, a2);
2732        break;
2733    case INDEX_op_orc_vec:
2734        if (const_args[2]) {
2735            is_shimm1632(~a2, &cmode, &imm8);
2736            if (a0 == a1) {
2737                tcg_out_insn(s, 3606, ORR, is_q, a0, 0, cmode, imm8);
2738                return;
2739            }
2740            tcg_out_insn(s, 3606, MVNI, is_q, a0, 0, cmode, imm8);
2741            a2 = a0;
2742        }
2743        tcg_out_insn(s, 3616, ORN, is_q, 0, a0, a1, a2);
2744        break;
2745    case INDEX_op_xor_vec:
2746        tcg_out_insn(s, 3616, EOR, is_q, 0, a0, a1, a2);
2747        break;
2748    case INDEX_op_ssadd_vec:
2749        if (is_scalar) {
2750            tcg_out_insn(s, 3611, SQADD, vece, a0, a1, a2);
2751        } else {
2752            tcg_out_insn(s, 3616, SQADD, is_q, vece, a0, a1, a2);
2753        }
2754        break;
2755    case INDEX_op_sssub_vec:
2756        if (is_scalar) {
2757            tcg_out_insn(s, 3611, SQSUB, vece, a0, a1, a2);
2758        } else {
2759            tcg_out_insn(s, 3616, SQSUB, is_q, vece, a0, a1, a2);
2760        }
2761        break;
2762    case INDEX_op_usadd_vec:
2763        if (is_scalar) {
2764            tcg_out_insn(s, 3611, UQADD, vece, a0, a1, a2);
2765        } else {
2766            tcg_out_insn(s, 3616, UQADD, is_q, vece, a0, a1, a2);
2767        }
2768        break;
2769    case INDEX_op_ussub_vec:
2770        if (is_scalar) {
2771            tcg_out_insn(s, 3611, UQSUB, vece, a0, a1, a2);
2772        } else {
2773            tcg_out_insn(s, 3616, UQSUB, is_q, vece, a0, a1, a2);
2774        }
2775        break;
2776    case INDEX_op_smax_vec:
2777        tcg_out_insn(s, 3616, SMAX, is_q, vece, a0, a1, a2);
2778        break;
2779    case INDEX_op_smin_vec:
2780        tcg_out_insn(s, 3616, SMIN, is_q, vece, a0, a1, a2);
2781        break;
2782    case INDEX_op_umax_vec:
2783        tcg_out_insn(s, 3616, UMAX, is_q, vece, a0, a1, a2);
2784        break;
2785    case INDEX_op_umin_vec:
2786        tcg_out_insn(s, 3616, UMIN, is_q, vece, a0, a1, a2);
2787        break;
2788    case INDEX_op_not_vec:
2789        tcg_out_insn(s, 3617, NOT, is_q, 0, a0, a1);
2790        break;
2791    case INDEX_op_shli_vec:
2792        if (is_scalar) {
2793            tcg_out_insn(s, 3609, SHL, a0, a1, a2 + (8 << vece));
2794        } else {
2795            tcg_out_insn(s, 3614, SHL, is_q, a0, a1, a2 + (8 << vece));
2796        }
2797        break;
2798    case INDEX_op_shri_vec:
2799        if (is_scalar) {
2800            tcg_out_insn(s, 3609, USHR, a0, a1, (16 << vece) - a2);
2801        } else {
2802            tcg_out_insn(s, 3614, USHR, is_q, a0, a1, (16 << vece) - a2);
2803        }
2804        break;
2805    case INDEX_op_sari_vec:
2806        if (is_scalar) {
2807            tcg_out_insn(s, 3609, SSHR, a0, a1, (16 << vece) - a2);
2808        } else {
2809            tcg_out_insn(s, 3614, SSHR, is_q, a0, a1, (16 << vece) - a2);
2810        }
2811        break;
2812    case INDEX_op_aa64_sli_vec:
2813        if (is_scalar) {
2814            tcg_out_insn(s, 3609, SLI, a0, a2, args[3] + (8 << vece));
2815        } else {
2816            tcg_out_insn(s, 3614, SLI, is_q, a0, a2, args[3] + (8 << vece));
2817        }
2818        break;
2819    case INDEX_op_shlv_vec:
2820        if (is_scalar) {
2821            tcg_out_insn(s, 3611, USHL, vece, a0, a1, a2);
2822        } else {
2823            tcg_out_insn(s, 3616, USHL, is_q, vece, a0, a1, a2);
2824        }
2825        break;
2826    case INDEX_op_aa64_sshl_vec:
2827        if (is_scalar) {
2828            tcg_out_insn(s, 3611, SSHL, vece, a0, a1, a2);
2829        } else {
2830            tcg_out_insn(s, 3616, SSHL, is_q, vece, a0, a1, a2);
2831        }
2832        break;
2833    case INDEX_op_cmp_vec:
2834        {
2835            TCGCond cond = args[3];
2836            AArch64Insn insn;
2837
2838            switch (cond) {
2839            case TCG_COND_NE:
2840                if (const_args[2]) {
2841                    if (is_scalar) {
2842                        tcg_out_insn(s, 3611, CMTST, vece, a0, a1, a1);
2843                    } else {
2844                        tcg_out_insn(s, 3616, CMTST, is_q, vece, a0, a1, a1);
2845                    }
2846                } else {
2847                    if (is_scalar) {
2848                        tcg_out_insn(s, 3611, CMEQ, vece, a0, a1, a2);
2849                    } else {
2850                        tcg_out_insn(s, 3616, CMEQ, is_q, vece, a0, a1, a2);
2851                    }
2852                    tcg_out_insn(s, 3617, NOT, is_q, 0, a0, a0);
2853                }
2854                break;
2855
2856            case TCG_COND_TSTNE:
2857            case TCG_COND_TSTEQ:
2858                if (const_args[2]) {
2859                    /* (x & 0) == 0 */
2860                    tcg_out_dupi_vec(s, type, MO_8, a0,
2861                                     -(cond == TCG_COND_TSTEQ));
2862                    break;
2863                }
2864                if (is_scalar) {
2865                    tcg_out_insn(s, 3611, CMTST, vece, a0, a1, a2);
2866                } else {
2867                    tcg_out_insn(s, 3616, CMTST, is_q, vece, a0, a1, a2);
2868                }
2869                if (cond == TCG_COND_TSTEQ) {
2870                    tcg_out_insn(s, 3617, NOT, is_q, 0, a0, a0);
2871                }
2872                break;
2873
2874            default:
2875                if (const_args[2]) {
2876                    if (is_scalar) {
2877                        insn = cmp0_scalar_insn[cond];
2878                        if (insn) {
2879                            tcg_out_insn_3612(s, insn, vece, a0, a1);
2880                            break;
2881                        }
2882                    } else {
2883                        insn = cmp0_vec_insn[cond];
2884                        if (insn) {
2885                            tcg_out_insn_3617(s, insn, is_q, vece, a0, a1);
2886                            break;
2887                        }
2888                    }
2889                    tcg_out_dupi_vec(s, type, MO_8, TCG_VEC_TMP0, 0);
2890                    a2 = TCG_VEC_TMP0;
2891                }
2892                if (is_scalar) {
2893                    insn = cmp_scalar_insn[cond];
2894                    if (insn == 0) {
2895                        TCGArg t;
2896                        t = a1, a1 = a2, a2 = t;
2897                        cond = tcg_swap_cond(cond);
2898                        insn = cmp_scalar_insn[cond];
2899                        tcg_debug_assert(insn != 0);
2900                    }
2901                    tcg_out_insn_3611(s, insn, vece, a0, a1, a2);
2902                } else {
2903                    insn = cmp_vec_insn[cond];
2904                    if (insn == 0) {
2905                        TCGArg t;
2906                        t = a1, a1 = a2, a2 = t;
2907                        cond = tcg_swap_cond(cond);
2908                        insn = cmp_vec_insn[cond];
2909                        tcg_debug_assert(insn != 0);
2910                    }
2911                    tcg_out_insn_3616(s, insn, is_q, vece, a0, a1, a2);
2912                }
2913                break;
2914            }
2915        }
2916        break;
2917
2918    case INDEX_op_bitsel_vec:
2919        a3 = args[3];
2920        if (a0 == a3) {
2921            tcg_out_insn(s, 3616, BIT, is_q, 0, a0, a2, a1);
2922        } else if (a0 == a2) {
2923            tcg_out_insn(s, 3616, BIF, is_q, 0, a0, a3, a1);
2924        } else {
2925            if (a0 != a1) {
2926                tcg_out_mov(s, type, a0, a1);
2927            }
2928            tcg_out_insn(s, 3616, BSL, is_q, 0, a0, a2, a3);
2929        }
2930        break;
2931
2932    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
2933    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
2934    default:
2935        g_assert_not_reached();
2936    }
2937}
2938
2939int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
2940{
2941    switch (opc) {
2942    case INDEX_op_add_vec:
2943    case INDEX_op_sub_vec:
2944    case INDEX_op_and_vec:
2945    case INDEX_op_or_vec:
2946    case INDEX_op_xor_vec:
2947    case INDEX_op_andc_vec:
2948    case INDEX_op_orc_vec:
2949    case INDEX_op_neg_vec:
2950    case INDEX_op_abs_vec:
2951    case INDEX_op_not_vec:
2952    case INDEX_op_cmp_vec:
2953    case INDEX_op_shli_vec:
2954    case INDEX_op_shri_vec:
2955    case INDEX_op_sari_vec:
2956    case INDEX_op_ssadd_vec:
2957    case INDEX_op_sssub_vec:
2958    case INDEX_op_usadd_vec:
2959    case INDEX_op_ussub_vec:
2960    case INDEX_op_shlv_vec:
2961    case INDEX_op_bitsel_vec:
2962        return 1;
2963    case INDEX_op_rotli_vec:
2964    case INDEX_op_shrv_vec:
2965    case INDEX_op_sarv_vec:
2966    case INDEX_op_rotlv_vec:
2967    case INDEX_op_rotrv_vec:
2968        return -1;
2969    case INDEX_op_mul_vec:
2970    case INDEX_op_smax_vec:
2971    case INDEX_op_smin_vec:
2972    case INDEX_op_umax_vec:
2973    case INDEX_op_umin_vec:
2974        return vece < MO_64;
2975
2976    default:
2977        return 0;
2978    }
2979}
2980
2981void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
2982                       TCGArg a0, ...)
2983{
2984    va_list va;
2985    TCGv_vec v0, v1, v2, t1, t2, c1;
2986    TCGArg a2;
2987
2988    va_start(va, a0);
2989    v0 = temp_tcgv_vec(arg_temp(a0));
2990    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
2991    a2 = va_arg(va, TCGArg);
2992    va_end(va);
2993
2994    switch (opc) {
2995    case INDEX_op_rotli_vec:
2996        t1 = tcg_temp_new_vec(type);
2997        tcg_gen_shri_vec(vece, t1, v1, -a2 & ((8 << vece) - 1));
2998        vec_gen_4(INDEX_op_aa64_sli_vec, type, vece,
2999                  tcgv_vec_arg(v0), tcgv_vec_arg(t1), tcgv_vec_arg(v1), a2);
3000        tcg_temp_free_vec(t1);
3001        break;
3002
3003    case INDEX_op_shrv_vec:
3004    case INDEX_op_sarv_vec:
3005        /* Right shifts are negative left shifts for AArch64.  */
3006        v2 = temp_tcgv_vec(arg_temp(a2));
3007        t1 = tcg_temp_new_vec(type);
3008        tcg_gen_neg_vec(vece, t1, v2);
3009        opc = (opc == INDEX_op_shrv_vec
3010               ? INDEX_op_shlv_vec : INDEX_op_aa64_sshl_vec);
3011        vec_gen_3(opc, type, vece, tcgv_vec_arg(v0),
3012                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3013        tcg_temp_free_vec(t1);
3014        break;
3015
3016    case INDEX_op_rotlv_vec:
3017        v2 = temp_tcgv_vec(arg_temp(a2));
3018        t1 = tcg_temp_new_vec(type);
3019        c1 = tcg_constant_vec(type, vece, 8 << vece);
3020        tcg_gen_sub_vec(vece, t1, v2, c1);
3021        /* Right shifts are negative left shifts for AArch64.  */
3022        vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(t1),
3023                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3024        vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(v0),
3025                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3026        tcg_gen_or_vec(vece, v0, v0, t1);
3027        tcg_temp_free_vec(t1);
3028        break;
3029
3030    case INDEX_op_rotrv_vec:
3031        v2 = temp_tcgv_vec(arg_temp(a2));
3032        t1 = tcg_temp_new_vec(type);
3033        t2 = tcg_temp_new_vec(type);
3034        c1 = tcg_constant_vec(type, vece, 8 << vece);
3035        tcg_gen_neg_vec(vece, t1, v2);
3036        tcg_gen_sub_vec(vece, t2, c1, v2);
3037        /* Right shifts are negative left shifts for AArch64.  */
3038        vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(t1),
3039                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3040        vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(t2),
3041                  tcgv_vec_arg(v1), tcgv_vec_arg(t2));
3042        tcg_gen_or_vec(vece, v0, t1, t2);
3043        tcg_temp_free_vec(t1);
3044        tcg_temp_free_vec(t2);
3045        break;
3046
3047    default:
3048        g_assert_not_reached();
3049    }
3050}
3051
3052static TCGConstraintSetIndex
3053tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
3054{
3055    switch (op) {
3056    case INDEX_op_goto_ptr:
3057        return C_O0_I1(r);
3058
3059    case INDEX_op_ld8u_i32:
3060    case INDEX_op_ld8s_i32:
3061    case INDEX_op_ld16u_i32:
3062    case INDEX_op_ld16s_i32:
3063    case INDEX_op_ld_i32:
3064    case INDEX_op_ld8u_i64:
3065    case INDEX_op_ld8s_i64:
3066    case INDEX_op_ld16u_i64:
3067    case INDEX_op_ld16s_i64:
3068    case INDEX_op_ld32u_i64:
3069    case INDEX_op_ld32s_i64:
3070    case INDEX_op_ld_i64:
3071    case INDEX_op_bswap16_i32:
3072    case INDEX_op_bswap32_i32:
3073    case INDEX_op_bswap16_i64:
3074    case INDEX_op_bswap32_i64:
3075    case INDEX_op_bswap64_i64:
3076    case INDEX_op_ext_i32_i64:
3077    case INDEX_op_extu_i32_i64:
3078    case INDEX_op_extract_i32:
3079    case INDEX_op_extract_i64:
3080    case INDEX_op_sextract_i32:
3081    case INDEX_op_sextract_i64:
3082        return C_O1_I1(r, r);
3083
3084    case INDEX_op_st8_i32:
3085    case INDEX_op_st16_i32:
3086    case INDEX_op_st_i32:
3087    case INDEX_op_st8_i64:
3088    case INDEX_op_st16_i64:
3089    case INDEX_op_st32_i64:
3090    case INDEX_op_st_i64:
3091        return C_O0_I2(rz, r);
3092
3093    case INDEX_op_setcond_i32:
3094    case INDEX_op_setcond_i64:
3095    case INDEX_op_negsetcond_i32:
3096    case INDEX_op_negsetcond_i64:
3097        return C_O1_I2(r, r, rC);
3098
3099    case INDEX_op_sar_i32:
3100    case INDEX_op_rotl_i32:
3101    case INDEX_op_rotr_i32:
3102    case INDEX_op_sar_i64:
3103    case INDEX_op_rotl_i64:
3104    case INDEX_op_rotr_i64:
3105        return C_O1_I2(r, r, ri);
3106
3107    case INDEX_op_clz_i32:
3108    case INDEX_op_ctz_i32:
3109    case INDEX_op_clz_i64:
3110    case INDEX_op_ctz_i64:
3111        return C_O1_I2(r, r, rAL);
3112
3113    case INDEX_op_brcond_i32:
3114    case INDEX_op_brcond_i64:
3115        return C_O0_I2(r, rC);
3116
3117    case INDEX_op_movcond_i32:
3118    case INDEX_op_movcond_i64:
3119        return C_O1_I4(r, r, rC, rz, rz);
3120
3121    case INDEX_op_qemu_ld_i32:
3122    case INDEX_op_qemu_ld_i64:
3123        return C_O1_I1(r, r);
3124    case INDEX_op_qemu_ld_i128:
3125        return C_O2_I1(r, r, r);
3126    case INDEX_op_qemu_st_i32:
3127    case INDEX_op_qemu_st_i64:
3128        return C_O0_I2(rz, r);
3129    case INDEX_op_qemu_st_i128:
3130        return C_O0_I3(rz, rz, r);
3131
3132    case INDEX_op_deposit_i32:
3133    case INDEX_op_deposit_i64:
3134        return C_O1_I2(r, 0, rz);
3135
3136    case INDEX_op_extract2_i32:
3137    case INDEX_op_extract2_i64:
3138        return C_O1_I2(r, rz, rz);
3139
3140    case INDEX_op_add2_i32:
3141    case INDEX_op_add2_i64:
3142    case INDEX_op_sub2_i32:
3143    case INDEX_op_sub2_i64:
3144        return C_O2_I4(r, r, rz, rz, rA, rMZ);
3145
3146    case INDEX_op_add_vec:
3147    case INDEX_op_sub_vec:
3148    case INDEX_op_mul_vec:
3149    case INDEX_op_xor_vec:
3150    case INDEX_op_ssadd_vec:
3151    case INDEX_op_sssub_vec:
3152    case INDEX_op_usadd_vec:
3153    case INDEX_op_ussub_vec:
3154    case INDEX_op_smax_vec:
3155    case INDEX_op_smin_vec:
3156    case INDEX_op_umax_vec:
3157    case INDEX_op_umin_vec:
3158    case INDEX_op_shlv_vec:
3159    case INDEX_op_shrv_vec:
3160    case INDEX_op_sarv_vec:
3161    case INDEX_op_aa64_sshl_vec:
3162        return C_O1_I2(w, w, w);
3163    case INDEX_op_not_vec:
3164    case INDEX_op_neg_vec:
3165    case INDEX_op_abs_vec:
3166    case INDEX_op_shli_vec:
3167    case INDEX_op_shri_vec:
3168    case INDEX_op_sari_vec:
3169        return C_O1_I1(w, w);
3170    case INDEX_op_ld_vec:
3171    case INDEX_op_dupm_vec:
3172        return C_O1_I1(w, r);
3173    case INDEX_op_st_vec:
3174        return C_O0_I2(w, r);
3175    case INDEX_op_dup_vec:
3176        return C_O1_I1(w, wr);
3177    case INDEX_op_or_vec:
3178    case INDEX_op_andc_vec:
3179        return C_O1_I2(w, w, wO);
3180    case INDEX_op_and_vec:
3181    case INDEX_op_orc_vec:
3182        return C_O1_I2(w, w, wN);
3183    case INDEX_op_cmp_vec:
3184        return C_O1_I2(w, w, wZ);
3185    case INDEX_op_bitsel_vec:
3186        return C_O1_I3(w, w, w, w);
3187    case INDEX_op_aa64_sli_vec:
3188        return C_O1_I2(w, 0, w);
3189
3190    default:
3191        return C_NotImplemented;
3192    }
3193}
3194
3195static void tcg_target_init(TCGContext *s)
3196{
3197    tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffffu;
3198    tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffffu;
3199    tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3200    tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3201
3202    tcg_target_call_clobber_regs = -1ull;
3203    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X19);
3204    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X20);
3205    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X21);
3206    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X22);
3207    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X23);
3208    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X24);
3209    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X25);
3210    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X26);
3211    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X27);
3212    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X28);
3213    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_X29);
3214    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V8);
3215    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V9);
3216    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V10);
3217    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V11);
3218    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V12);
3219    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V13);
3220    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V14);
3221    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_V15);
3222
3223    s->reserved_regs = 0;
3224    tcg_regset_set_reg(s->reserved_regs, TCG_REG_SP);
3225    tcg_regset_set_reg(s->reserved_regs, TCG_REG_FP);
3226    tcg_regset_set_reg(s->reserved_regs, TCG_REG_X18); /* platform register */
3227    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP0);
3228    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1);
3229    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP2);
3230    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP0);
3231}
3232
3233/* Saving pairs: (X19, X20) .. (X27, X28), (X29(fp), X30(lr)).  */
3234#define PUSH_SIZE  ((30 - 19 + 1) * 8)
3235
3236#define FRAME_SIZE \
3237    ((PUSH_SIZE \
3238      + TCG_STATIC_CALL_ARGS_SIZE \
3239      + CPU_TEMP_BUF_NLONGS * sizeof(long) \
3240      + TCG_TARGET_STACK_ALIGN - 1) \
3241     & ~(TCG_TARGET_STACK_ALIGN - 1))
3242
3243/* We're expecting a 2 byte uleb128 encoded value.  */
3244QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3245
3246/* We're expecting to use a single ADDI insn.  */
3247QEMU_BUILD_BUG_ON(FRAME_SIZE - PUSH_SIZE > 0xfff);
3248
3249static void tcg_target_qemu_prologue(TCGContext *s)
3250{
3251    TCGReg r;
3252
3253    tcg_out_bti(s, BTI_C);
3254
3255    /* Push (FP, LR) and allocate space for all saved registers.  */
3256    tcg_out_insn(s, 3314, STP, TCG_REG_FP, TCG_REG_LR,
3257                 TCG_REG_SP, -PUSH_SIZE, 1, 1);
3258
3259    /* Set up frame pointer for canonical unwinding.  */
3260    tcg_out_movr_sp(s, TCG_TYPE_I64, TCG_REG_FP, TCG_REG_SP);
3261
3262    /* Store callee-preserved regs x19..x28.  */
3263    for (r = TCG_REG_X19; r <= TCG_REG_X27; r += 2) {
3264        int ofs = (r - TCG_REG_X19 + 2) * 8;
3265        tcg_out_insn(s, 3314, STP, r, r + 1, TCG_REG_SP, ofs, 1, 0);
3266    }
3267
3268    /* Make stack space for TCG locals.  */
3269    tcg_out_insn(s, 3401, SUBI, TCG_TYPE_I64, TCG_REG_SP, TCG_REG_SP,
3270                 FRAME_SIZE - PUSH_SIZE);
3271
3272    /* Inform TCG about how to find TCG locals with register, offset, size.  */
3273    tcg_set_frame(s, TCG_REG_SP, TCG_STATIC_CALL_ARGS_SIZE,
3274                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3275
3276    if (!tcg_use_softmmu) {
3277        /*
3278         * Note that XZR cannot be encoded in the address base register slot,
3279         * as that actually encodes SP.  Depending on the guest, we may need
3280         * to zero-extend the guest address via the address index register slot,
3281         * therefore we need to load even a zero guest base into a register.
3282         */
3283        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_GUEST_BASE, guest_base);
3284        tcg_regset_set_reg(s->reserved_regs, TCG_REG_GUEST_BASE);
3285    }
3286
3287    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3288    tcg_out_insn(s, 3207, BR, tcg_target_call_iarg_regs[1]);
3289
3290    /*
3291     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3292     * and fall through to the rest of the epilogue.
3293     */
3294    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3295    tcg_out_bti(s, BTI_J);
3296    tcg_out_movi(s, TCG_TYPE_REG, TCG_REG_X0, 0);
3297
3298    /* TB epilogue */
3299    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
3300    tcg_out_bti(s, BTI_J);
3301
3302    /* Remove TCG locals stack space.  */
3303    tcg_out_insn(s, 3401, ADDI, TCG_TYPE_I64, TCG_REG_SP, TCG_REG_SP,
3304                 FRAME_SIZE - PUSH_SIZE);
3305
3306    /* Restore registers x19..x28.  */
3307    for (r = TCG_REG_X19; r <= TCG_REG_X27; r += 2) {
3308        int ofs = (r - TCG_REG_X19 + 2) * 8;
3309        tcg_out_insn(s, 3314, LDP, r, r + 1, TCG_REG_SP, ofs, 1, 0);
3310    }
3311
3312    /* Pop (FP, LR), restore SP to previous frame.  */
3313    tcg_out_insn(s, 3314, LDP, TCG_REG_FP, TCG_REG_LR,
3314                 TCG_REG_SP, PUSH_SIZE, 0, 1);
3315    tcg_out_insn(s, 3207, RET, TCG_REG_LR);
3316}
3317
3318static void tcg_out_tb_start(TCGContext *s)
3319{
3320    tcg_out_bti(s, BTI_J);
3321}
3322
3323static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3324{
3325    int i;
3326    for (i = 0; i < count; ++i) {
3327        p[i] = NOP;
3328    }
3329}
3330
3331typedef struct {
3332    DebugFrameHeader h;
3333    uint8_t fde_def_cfa[4];
3334    uint8_t fde_reg_ofs[24];
3335} DebugFrame;
3336
3337#define ELF_HOST_MACHINE EM_AARCH64
3338
3339static const DebugFrame debug_frame = {
3340    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3341    .h.cie.id = -1,
3342    .h.cie.version = 1,
3343    .h.cie.code_align = 1,
3344    .h.cie.data_align = 0x78,             /* sleb128 -8 */
3345    .h.cie.return_column = TCG_REG_LR,
3346
3347    /* Total FDE size does not include the "len" member.  */
3348    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3349
3350    .fde_def_cfa = {
3351        12, TCG_REG_SP,                 /* DW_CFA_def_cfa sp, ... */
3352        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3353        (FRAME_SIZE >> 7)
3354    },
3355    .fde_reg_ofs = {
3356        0x80 + 28, 1,                   /* DW_CFA_offset, x28,  -8 */
3357        0x80 + 27, 2,                   /* DW_CFA_offset, x27, -16 */
3358        0x80 + 26, 3,                   /* DW_CFA_offset, x26, -24 */
3359        0x80 + 25, 4,                   /* DW_CFA_offset, x25, -32 */
3360        0x80 + 24, 5,                   /* DW_CFA_offset, x24, -40 */
3361        0x80 + 23, 6,                   /* DW_CFA_offset, x23, -48 */
3362        0x80 + 22, 7,                   /* DW_CFA_offset, x22, -56 */
3363        0x80 + 21, 8,                   /* DW_CFA_offset, x21, -64 */
3364        0x80 + 20, 9,                   /* DW_CFA_offset, x20, -72 */
3365        0x80 + 19, 10,                  /* DW_CFA_offset, x1p, -80 */
3366        0x80 + 30, 11,                  /* DW_CFA_offset,  lr, -88 */
3367        0x80 + 29, 12,                  /* DW_CFA_offset,  fp, -96 */
3368    }
3369};
3370
3371void tcg_register_jit(const void *buf, size_t buf_size)
3372{
3373    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3374}
3375