xref: /openbmc/qemu/tcg/arm/tcg-target.c.inc (revision 8ba0034ddffd14b1ceddeb223199c17becf0a02f)
1/*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2008 Andrzej Zaborowski
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25#include "elf.h"
26
27int arm_arch = __ARM_ARCH;
28
29#ifndef use_idiv_instructions
30bool use_idiv_instructions;
31#endif
32#ifndef use_neon_instructions
33bool use_neon_instructions;
34#endif
35
36/* Used for function call generation. */
37#define TCG_TARGET_STACK_ALIGN          8
38#define TCG_TARGET_CALL_STACK_OFFSET    0
39#define TCG_TARGET_CALL_ARG_I32         TCG_CALL_ARG_NORMAL
40#define TCG_TARGET_CALL_ARG_I64         TCG_CALL_ARG_EVEN
41#define TCG_TARGET_CALL_ARG_I128        TCG_CALL_ARG_EVEN
42#define TCG_TARGET_CALL_RET_I128        TCG_CALL_RET_BY_REF
43
44#ifdef CONFIG_DEBUG_TCG
45static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
46    "%r0",  "%r1",  "%r2",  "%r3",  "%r4",  "%r5",  "%r6",  "%r7",
47    "%r8",  "%r9",  "%r10", "%r11", "%r12", "%sp",  "%r14", "%pc",
48    "%q0",  "%q1",  "%q2",  "%q3",  "%q4",  "%q5",  "%q6",  "%q7",
49    "%q8",  "%q9",  "%q10", "%q11", "%q12", "%q13", "%q14", "%q15",
50};
51#endif
52
53static const int tcg_target_reg_alloc_order[] = {
54    TCG_REG_R4,
55    TCG_REG_R5,
56    TCG_REG_R6,
57    TCG_REG_R7,
58    TCG_REG_R8,
59    TCG_REG_R9,
60    TCG_REG_R10,
61    TCG_REG_R11,
62    TCG_REG_R13,
63    TCG_REG_R0,
64    TCG_REG_R1,
65    TCG_REG_R2,
66    TCG_REG_R3,
67    TCG_REG_R12,
68    TCG_REG_R14,
69
70    TCG_REG_Q0,
71    TCG_REG_Q1,
72    TCG_REG_Q2,
73    TCG_REG_Q3,
74    /* Q4 - Q7 are call-saved, and skipped. */
75    TCG_REG_Q8,
76    TCG_REG_Q9,
77    TCG_REG_Q10,
78    TCG_REG_Q11,
79    TCG_REG_Q12,
80    TCG_REG_Q13,
81    TCG_REG_Q14,
82    TCG_REG_Q15,
83};
84
85static const int tcg_target_call_iarg_regs[4] = {
86    TCG_REG_R0, TCG_REG_R1, TCG_REG_R2, TCG_REG_R3
87};
88
89static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
90{
91    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
92    tcg_debug_assert(slot >= 0 && slot <= 3);
93    return TCG_REG_R0 + slot;
94}
95
96#define TCG_REG_TMP  TCG_REG_R12
97#define TCG_VEC_TMP  TCG_REG_Q15
98#define TCG_REG_GUEST_BASE  TCG_REG_R11
99
100typedef enum {
101    COND_EQ = 0x0,
102    COND_NE = 0x1,
103    COND_CS = 0x2,	/* Unsigned greater or equal */
104    COND_CC = 0x3,	/* Unsigned less than */
105    COND_MI = 0x4,	/* Negative */
106    COND_PL = 0x5,	/* Zero or greater */
107    COND_VS = 0x6,	/* Overflow */
108    COND_VC = 0x7,	/* No overflow */
109    COND_HI = 0x8,	/* Unsigned greater than */
110    COND_LS = 0x9,	/* Unsigned less or equal */
111    COND_GE = 0xa,
112    COND_LT = 0xb,
113    COND_GT = 0xc,
114    COND_LE = 0xd,
115    COND_AL = 0xe,
116} ARMCond;
117
118#define TO_CPSR (1 << 20)
119
120#define SHIFT_IMM_LSL(im)	(((im) << 7) | 0x00)
121#define SHIFT_IMM_LSR(im)	(((im) << 7) | 0x20)
122#define SHIFT_IMM_ASR(im)	(((im) << 7) | 0x40)
123#define SHIFT_IMM_ROR(im)	(((im) << 7) | 0x60)
124#define SHIFT_REG_LSL(rs)	(((rs) << 8) | 0x10)
125#define SHIFT_REG_LSR(rs)	(((rs) << 8) | 0x30)
126#define SHIFT_REG_ASR(rs)	(((rs) << 8) | 0x50)
127#define SHIFT_REG_ROR(rs)	(((rs) << 8) | 0x70)
128
129typedef enum {
130    ARITH_AND = 0x0 << 21,
131    ARITH_EOR = 0x1 << 21,
132    ARITH_SUB = 0x2 << 21,
133    ARITH_RSB = 0x3 << 21,
134    ARITH_ADD = 0x4 << 21,
135    ARITH_ADC = 0x5 << 21,
136    ARITH_SBC = 0x6 << 21,
137    ARITH_RSC = 0x7 << 21,
138    ARITH_TST = 0x8 << 21 | TO_CPSR,
139    ARITH_CMP = 0xa << 21 | TO_CPSR,
140    ARITH_CMN = 0xb << 21 | TO_CPSR,
141    ARITH_ORR = 0xc << 21,
142    ARITH_MOV = 0xd << 21,
143    ARITH_BIC = 0xe << 21,
144    ARITH_MVN = 0xf << 21,
145
146    INSN_B         = 0x0a000000,
147
148    INSN_CLZ       = 0x016f0f10,
149    INSN_RBIT      = 0x06ff0f30,
150
151    INSN_LDMIA     = 0x08b00000,
152    INSN_STMDB     = 0x09200000,
153
154    INSN_LDR_IMM   = 0x04100000,
155    INSN_LDR_REG   = 0x06100000,
156    INSN_STR_IMM   = 0x04000000,
157    INSN_STR_REG   = 0x06000000,
158
159    INSN_LDRH_IMM  = 0x005000b0,
160    INSN_LDRH_REG  = 0x001000b0,
161    INSN_LDRSH_IMM = 0x005000f0,
162    INSN_LDRSH_REG = 0x001000f0,
163    INSN_STRH_IMM  = 0x004000b0,
164    INSN_STRH_REG  = 0x000000b0,
165
166    INSN_LDRB_IMM  = 0x04500000,
167    INSN_LDRB_REG  = 0x06500000,
168    INSN_LDRSB_IMM = 0x005000d0,
169    INSN_LDRSB_REG = 0x001000d0,
170    INSN_STRB_IMM  = 0x04400000,
171    INSN_STRB_REG  = 0x06400000,
172
173    INSN_LDRD_IMM  = 0x004000d0,
174    INSN_LDRD_REG  = 0x000000d0,
175    INSN_STRD_IMM  = 0x004000f0,
176    INSN_STRD_REG  = 0x000000f0,
177
178    INSN_DMB_ISH   = 0xf57ff05b,
179    INSN_DMB_MCR   = 0xee070fba,
180
181    INSN_MSRI_CPSR = 0x0360f000,
182
183    /* Architected nop introduced in v6k.  */
184    /* ??? This is an MSR (imm) 0,0,0 insn.  Anyone know if this
185       also Just So Happened to do nothing on pre-v6k so that we
186       don't need to conditionalize it?  */
187    INSN_NOP_v6k   = 0xe320f000,
188    /* Otherwise the assembler uses mov r0,r0 */
189    INSN_NOP_v4    = (COND_AL << 28) | ARITH_MOV,
190
191    INSN_VADD      = 0xf2000800,
192    INSN_VAND      = 0xf2000110,
193    INSN_VBIC      = 0xf2100110,
194    INSN_VEOR      = 0xf3000110,
195    INSN_VORN      = 0xf2300110,
196    INSN_VORR      = 0xf2200110,
197    INSN_VSUB      = 0xf3000800,
198    INSN_VMUL      = 0xf2000910,
199    INSN_VQADD     = 0xf2000010,
200    INSN_VQADD_U   = 0xf3000010,
201    INSN_VQSUB     = 0xf2000210,
202    INSN_VQSUB_U   = 0xf3000210,
203    INSN_VMAX      = 0xf2000600,
204    INSN_VMAX_U    = 0xf3000600,
205    INSN_VMIN      = 0xf2000610,
206    INSN_VMIN_U    = 0xf3000610,
207
208    INSN_VABS      = 0xf3b10300,
209    INSN_VMVN      = 0xf3b00580,
210    INSN_VNEG      = 0xf3b10380,
211
212    INSN_VCEQ0     = 0xf3b10100,
213    INSN_VCGT0     = 0xf3b10000,
214    INSN_VCGE0     = 0xf3b10080,
215    INSN_VCLE0     = 0xf3b10180,
216    INSN_VCLT0     = 0xf3b10200,
217
218    INSN_VCEQ      = 0xf3000810,
219    INSN_VCGE      = 0xf2000310,
220    INSN_VCGT      = 0xf2000300,
221    INSN_VCGE_U    = 0xf3000310,
222    INSN_VCGT_U    = 0xf3000300,
223
224    INSN_VSHLI     = 0xf2800510,  /* VSHL (immediate) */
225    INSN_VSARI     = 0xf2800010,  /* VSHR.S */
226    INSN_VSHRI     = 0xf3800010,  /* VSHR.U */
227    INSN_VSLI      = 0xf3800510,
228    INSN_VSHL_S    = 0xf2000400,  /* VSHL.S (register) */
229    INSN_VSHL_U    = 0xf3000400,  /* VSHL.U (register) */
230
231    INSN_VBSL      = 0xf3100110,
232    INSN_VBIT      = 0xf3200110,
233    INSN_VBIF      = 0xf3300110,
234
235    INSN_VTST      = 0xf2000810,
236
237    INSN_VDUP_G    = 0xee800b10,  /* VDUP (ARM core register) */
238    INSN_VDUP_S    = 0xf3b00c00,  /* VDUP (scalar) */
239    INSN_VLDR_D    = 0xed100b00,  /* VLDR.64 */
240    INSN_VLD1      = 0xf4200000,  /* VLD1 (multiple single elements) */
241    INSN_VLD1R     = 0xf4a00c00,  /* VLD1 (single element to all lanes) */
242    INSN_VST1      = 0xf4000000,  /* VST1 (multiple single elements) */
243    INSN_VMOVI     = 0xf2800010,  /* VMOV (immediate) */
244} ARMInsn;
245
246#define INSN_NOP   (use_armv7_instructions ? INSN_NOP_v6k : INSN_NOP_v4)
247
248static const uint8_t tcg_cond_to_arm_cond[] = {
249    [TCG_COND_EQ] = COND_EQ,
250    [TCG_COND_NE] = COND_NE,
251    [TCG_COND_LT] = COND_LT,
252    [TCG_COND_GE] = COND_GE,
253    [TCG_COND_LE] = COND_LE,
254    [TCG_COND_GT] = COND_GT,
255    /* unsigned */
256    [TCG_COND_LTU] = COND_CC,
257    [TCG_COND_GEU] = COND_CS,
258    [TCG_COND_LEU] = COND_LS,
259    [TCG_COND_GTU] = COND_HI,
260};
261
262static int encode_imm(uint32_t imm);
263
264/* TCG private relocation type: add with pc+imm8 */
265#define R_ARM_PC8  11
266
267/* TCG private relocation type: vldr with imm8 << 2 */
268#define R_ARM_PC11 12
269
270static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
271{
272    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
273    ptrdiff_t offset = (tcg_ptr_byte_diff(target, src_rx) - 8) >> 2;
274
275    if (offset == sextract32(offset, 0, 24)) {
276        *src_rw = deposit32(*src_rw, 0, 24, offset);
277        return true;
278    }
279    return false;
280}
281
282static bool reloc_pc13(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
283{
284    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
285    ptrdiff_t offset = tcg_ptr_byte_diff(target, src_rx) - 8;
286
287    if (offset >= -0xfff && offset <= 0xfff) {
288        tcg_insn_unit insn = *src_rw;
289        bool u = (offset >= 0);
290        if (!u) {
291            offset = -offset;
292        }
293        insn = deposit32(insn, 23, 1, u);
294        insn = deposit32(insn, 0, 12, offset);
295        *src_rw = insn;
296        return true;
297    }
298    return false;
299}
300
301static bool reloc_pc11(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
302{
303    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
304    ptrdiff_t offset = (tcg_ptr_byte_diff(target, src_rx) - 8) / 4;
305
306    if (offset >= -0xff && offset <= 0xff) {
307        tcg_insn_unit insn = *src_rw;
308        bool u = (offset >= 0);
309        if (!u) {
310            offset = -offset;
311        }
312        insn = deposit32(insn, 23, 1, u);
313        insn = deposit32(insn, 0, 8, offset);
314        *src_rw = insn;
315        return true;
316    }
317    return false;
318}
319
320static bool reloc_pc8(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
321{
322    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
323    ptrdiff_t offset = tcg_ptr_byte_diff(target, src_rx) - 8;
324    int imm12 = encode_imm(offset);
325
326    if (imm12 >= 0) {
327        *src_rw = deposit32(*src_rw, 0, 12, imm12);
328        return true;
329    }
330    return false;
331}
332
333static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
334                        intptr_t value, intptr_t addend)
335{
336    tcg_debug_assert(addend == 0);
337    switch (type) {
338    case R_ARM_PC24:
339        return reloc_pc24(code_ptr, (const tcg_insn_unit *)value);
340    case R_ARM_PC13:
341        return reloc_pc13(code_ptr, (const tcg_insn_unit *)value);
342    case R_ARM_PC11:
343        return reloc_pc11(code_ptr, (const tcg_insn_unit *)value);
344    case R_ARM_PC8:
345        return reloc_pc8(code_ptr, (const tcg_insn_unit *)value);
346    default:
347        g_assert_not_reached();
348    }
349}
350
351#define TCG_CT_CONST_ARM  0x100
352#define TCG_CT_CONST_INV  0x200
353#define TCG_CT_CONST_NEG  0x400
354#define TCG_CT_CONST_ZERO 0x800
355#define TCG_CT_CONST_ORRI 0x1000
356#define TCG_CT_CONST_ANDI 0x2000
357
358#define ALL_GENERAL_REGS  0xffffu
359#define ALL_VECTOR_REGS   0xffff0000u
360
361/*
362 * r0-r3 will be overwritten when reading the tlb entry (system-mode only);
363 * r14 will be overwritten by the BLNE branching to the slow path.
364 */
365#define ALL_QLDST_REGS \
366    (ALL_GENERAL_REGS & ~((tcg_use_softmmu ? 0xf : 0) | (1 << TCG_REG_R14)))
367
368/*
369 * ARM immediates for ALU instructions are made of an unsigned 8-bit
370 * right-rotated by an even amount between 0 and 30.
371 *
372 * Return < 0 if @imm cannot be encoded, else the entire imm12 field.
373 */
374static int encode_imm(uint32_t imm)
375{
376    uint32_t rot, imm8;
377
378    /* Simple case, no rotation required. */
379    if ((imm & ~0xff) == 0) {
380        return imm;
381    }
382
383    /* Next, try a simple even shift.  */
384    rot = ctz32(imm) & ~1;
385    imm8 = imm >> rot;
386    rot = 32 - rot;
387    if ((imm8 & ~0xff) == 0) {
388        goto found;
389    }
390
391    /*
392     * Finally, try harder with rotations.
393     * The ctz test above will have taken care of rotates >= 8.
394     */
395    for (rot = 2; rot < 8; rot += 2) {
396        imm8 = rol32(imm, rot);
397        if ((imm8 & ~0xff) == 0) {
398            goto found;
399        }
400    }
401    /* Fail: imm cannot be encoded. */
402    return -1;
403
404 found:
405    /* Note that rot is even, and we discard bit 0 by shifting by 7. */
406    return rot << 7 | imm8;
407}
408
409static int encode_imm_nofail(uint32_t imm)
410{
411    int ret = encode_imm(imm);
412    tcg_debug_assert(ret >= 0);
413    return ret;
414}
415
416static bool check_fit_imm(uint32_t imm)
417{
418    return encode_imm(imm) >= 0;
419}
420
421/* Return true if v16 is a valid 16-bit shifted immediate.  */
422static bool is_shimm16(uint16_t v16, int *cmode, int *imm8)
423{
424    if (v16 == (v16 & 0xff)) {
425        *cmode = 0x8;
426        *imm8 = v16 & 0xff;
427        return true;
428    } else if (v16 == (v16 & 0xff00)) {
429        *cmode = 0xa;
430        *imm8 = v16 >> 8;
431        return true;
432    }
433    return false;
434}
435
436/* Return true if v32 is a valid 32-bit shifted immediate.  */
437static bool is_shimm32(uint32_t v32, int *cmode, int *imm8)
438{
439    if (v32 == (v32 & 0xff)) {
440        *cmode = 0x0;
441        *imm8 = v32 & 0xff;
442        return true;
443    } else if (v32 == (v32 & 0xff00)) {
444        *cmode = 0x2;
445        *imm8 = (v32 >> 8) & 0xff;
446        return true;
447    } else if (v32 == (v32 & 0xff0000)) {
448        *cmode = 0x4;
449        *imm8 = (v32 >> 16) & 0xff;
450        return true;
451    } else if (v32 == (v32 & 0xff000000)) {
452        *cmode = 0x6;
453        *imm8 = v32 >> 24;
454        return true;
455    }
456    return false;
457}
458
459/* Return true if v32 is a valid 32-bit shifting ones immediate.  */
460static bool is_soimm32(uint32_t v32, int *cmode, int *imm8)
461{
462    if ((v32 & 0xffff00ff) == 0xff) {
463        *cmode = 0xc;
464        *imm8 = (v32 >> 8) & 0xff;
465        return true;
466    } else if ((v32 & 0xff00ffff) == 0xffff) {
467        *cmode = 0xd;
468        *imm8 = (v32 >> 16) & 0xff;
469        return true;
470    }
471    return false;
472}
473
474/*
475 * Return non-zero if v32 can be formed by MOVI+ORR.
476 * Place the parameters for MOVI in (cmode, imm8).
477 * Return the cmode for ORR; the imm8 can be had via extraction from v32.
478 */
479static int is_shimm32_pair(uint32_t v32, int *cmode, int *imm8)
480{
481    int i;
482
483    for (i = 6; i > 0; i -= 2) {
484        /* Mask out one byte we can add with ORR.  */
485        uint32_t tmp = v32 & ~(0xffu << (i * 4));
486        if (is_shimm32(tmp, cmode, imm8) ||
487            is_soimm32(tmp, cmode, imm8)) {
488            break;
489        }
490    }
491    return i;
492}
493
494/* Return true if V is a valid 16-bit or 32-bit shifted immediate.  */
495static bool is_shimm1632(uint32_t v32, int *cmode, int *imm8)
496{
497    if (v32 == deposit32(v32, 16, 16, v32)) {
498        return is_shimm16(v32, cmode, imm8);
499    } else {
500        return is_shimm32(v32, cmode, imm8);
501    }
502}
503
504/* Test if a constant matches the constraint.
505 * TODO: define constraints for:
506 *
507 * ldr/str offset:   between -0xfff and 0xfff
508 * ldrh/strh offset: between -0xff and 0xff
509 * mov operand2:     values represented with x << (2 * y), x < 0x100
510 * add, sub, eor...: ditto
511 */
512static bool tcg_target_const_match(int64_t val, int ct,
513                                   TCGType type, TCGCond cond, int vece)
514{
515    if (ct & TCG_CT_CONST) {
516        return 1;
517    } else if ((ct & TCG_CT_CONST_ARM) && check_fit_imm(val)) {
518        return 1;
519    } else if ((ct & TCG_CT_CONST_INV) && check_fit_imm(~val)) {
520        return 1;
521    } else if ((ct & TCG_CT_CONST_NEG) && check_fit_imm(-val)) {
522        return 1;
523    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
524        return 1;
525    }
526
527    switch (ct & (TCG_CT_CONST_ORRI | TCG_CT_CONST_ANDI)) {
528    case 0:
529        break;
530    case TCG_CT_CONST_ANDI:
531        val = ~val;
532        /* fallthru */
533    case TCG_CT_CONST_ORRI:
534        if (val == deposit64(val, 32, 32, val)) {
535            int cmode, imm8;
536            return is_shimm1632(val, &cmode, &imm8);
537        }
538        break;
539    default:
540        /* Both bits should not be set for the same insn.  */
541        g_assert_not_reached();
542    }
543
544    return 0;
545}
546
547static void tcg_out_b_imm(TCGContext *s, ARMCond cond, int32_t offset)
548{
549    tcg_out32(s, (cond << 28) | INSN_B |
550                    (((offset - 8) >> 2) & 0x00ffffff));
551}
552
553static void tcg_out_bl_imm(TCGContext *s, ARMCond cond, int32_t offset)
554{
555    tcg_out32(s, (cond << 28) | 0x0b000000 |
556                    (((offset - 8) >> 2) & 0x00ffffff));
557}
558
559static void tcg_out_blx_reg(TCGContext *s, ARMCond cond, TCGReg rn)
560{
561    tcg_out32(s, (cond << 28) | 0x012fff30 | rn);
562}
563
564static void tcg_out_blx_imm(TCGContext *s, int32_t offset)
565{
566    tcg_out32(s, 0xfa000000 | ((offset & 2) << 23) |
567                (((offset - 8) >> 2) & 0x00ffffff));
568}
569
570static void tcg_out_dat_reg(TCGContext *s, ARMCond cond, ARMInsn opc,
571                            TCGReg rd, TCGReg rn, TCGReg rm, int shift)
572{
573    tcg_out32(s, (cond << 28) | (0 << 25) | opc |
574                    (rn << 16) | (rd << 12) | shift | rm);
575}
576
577static void tcg_out_mov_reg(TCGContext *s, ARMCond cond, TCGReg rd, TCGReg rm)
578{
579    /* Simple reg-reg move, optimising out the 'do nothing' case */
580    if (rd != rm) {
581        tcg_out_dat_reg(s, cond, ARITH_MOV, rd, 0, rm, SHIFT_IMM_LSL(0));
582    }
583}
584
585static void tcg_out_bx_reg(TCGContext *s, ARMCond cond, TCGReg rn)
586{
587    tcg_out32(s, (cond << 28) | 0x012fff10 | rn);
588}
589
590static void tcg_out_b_reg(TCGContext *s, ARMCond cond, TCGReg rn)
591{
592    /*
593     * Unless the C portion of QEMU is compiled as thumb, we don't need
594     * true BX semantics; merely a branch to an address held in a register.
595     */
596    tcg_out_bx_reg(s, cond, rn);
597}
598
599static void tcg_out_dat_imm(TCGContext *s, ARMCond cond, ARMInsn opc,
600                            TCGReg rd, TCGReg rn, int im)
601{
602    tcg_out32(s, (cond << 28) | (1 << 25) | opc |
603                    (rn << 16) | (rd << 12) | im);
604}
605
606static void tcg_out_ldstm(TCGContext *s, ARMCond cond, ARMInsn opc,
607                          TCGReg rn, uint16_t mask)
608{
609    tcg_out32(s, (cond << 28) | opc | (rn << 16) | mask);
610}
611
612/* Note that this routine is used for both LDR and LDRH formats, so we do
613   not wish to include an immediate shift at this point.  */
614static void tcg_out_memop_r(TCGContext *s, ARMCond cond, ARMInsn opc, TCGReg rt,
615                            TCGReg rn, TCGReg rm, bool u, bool p, bool w)
616{
617    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24)
618              | (w << 21) | (rn << 16) | (rt << 12) | rm);
619}
620
621static void tcg_out_memop_8(TCGContext *s, ARMCond cond, ARMInsn opc, TCGReg rt,
622                            TCGReg rn, int imm8, bool p, bool w)
623{
624    bool u = 1;
625    if (imm8 < 0) {
626        imm8 = -imm8;
627        u = 0;
628    }
629    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
630              (rn << 16) | (rt << 12) | ((imm8 & 0xf0) << 4) | (imm8 & 0xf));
631}
632
633static void tcg_out_memop_12(TCGContext *s, ARMCond cond, ARMInsn opc,
634                             TCGReg rt, TCGReg rn, int imm12, bool p, bool w)
635{
636    bool u = 1;
637    if (imm12 < 0) {
638        imm12 = -imm12;
639        u = 0;
640    }
641    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
642              (rn << 16) | (rt << 12) | imm12);
643}
644
645static void tcg_out_ld32_12(TCGContext *s, ARMCond cond, TCGReg rt,
646                            TCGReg rn, int imm12)
647{
648    tcg_out_memop_12(s, cond, INSN_LDR_IMM, rt, rn, imm12, 1, 0);
649}
650
651static void tcg_out_st32_12(TCGContext *s, ARMCond cond, TCGReg rt,
652                            TCGReg rn, int imm12)
653{
654    tcg_out_memop_12(s, cond, INSN_STR_IMM, rt, rn, imm12, 1, 0);
655}
656
657static void tcg_out_ld32_r(TCGContext *s, ARMCond cond, TCGReg rt,
658                           TCGReg rn, TCGReg rm)
659{
660    tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 0);
661}
662
663static void tcg_out_st32_r(TCGContext *s, ARMCond cond, TCGReg rt,
664                           TCGReg rn, TCGReg rm)
665{
666    tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 0);
667}
668
669static void tcg_out_ldrd_8(TCGContext *s, ARMCond cond, TCGReg rt,
670                           TCGReg rn, int imm8)
671{
672    tcg_out_memop_8(s, cond, INSN_LDRD_IMM, rt, rn, imm8, 1, 0);
673}
674
675static void tcg_out_ldrd_r(TCGContext *s, ARMCond cond, TCGReg rt,
676                           TCGReg rn, TCGReg rm)
677{
678    tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 0);
679}
680
681static void tcg_out_strd_8(TCGContext *s, ARMCond cond, TCGReg rt,
682                           TCGReg rn, int imm8)
683{
684    tcg_out_memop_8(s, cond, INSN_STRD_IMM, rt, rn, imm8, 1, 0);
685}
686
687static void tcg_out_strd_r(TCGContext *s, ARMCond cond, TCGReg rt,
688                           TCGReg rn, TCGReg rm)
689{
690    tcg_out_memop_r(s, cond, INSN_STRD_REG, rt, rn, rm, 1, 1, 0);
691}
692
693/* Register pre-increment with base writeback.  */
694static void tcg_out_ld32_rwb(TCGContext *s, ARMCond cond, TCGReg rt,
695                             TCGReg rn, TCGReg rm)
696{
697    tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 1);
698}
699
700static void tcg_out_st32_rwb(TCGContext *s, ARMCond cond, TCGReg rt,
701                             TCGReg rn, TCGReg rm)
702{
703    tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 1);
704}
705
706static void tcg_out_ld16u_8(TCGContext *s, ARMCond cond, TCGReg rt,
707                            TCGReg rn, int imm8)
708{
709    tcg_out_memop_8(s, cond, INSN_LDRH_IMM, rt, rn, imm8, 1, 0);
710}
711
712static void tcg_out_st16_8(TCGContext *s, ARMCond cond, TCGReg rt,
713                           TCGReg rn, int imm8)
714{
715    tcg_out_memop_8(s, cond, INSN_STRH_IMM, rt, rn, imm8, 1, 0);
716}
717
718static void tcg_out_ld16u_r(TCGContext *s, ARMCond cond, TCGReg rt,
719                            TCGReg rn, TCGReg rm)
720{
721    tcg_out_memop_r(s, cond, INSN_LDRH_REG, rt, rn, rm, 1, 1, 0);
722}
723
724static void tcg_out_st16_r(TCGContext *s, ARMCond cond, TCGReg rt,
725                           TCGReg rn, TCGReg rm)
726{
727    tcg_out_memop_r(s, cond, INSN_STRH_REG, rt, rn, rm, 1, 1, 0);
728}
729
730static void tcg_out_ld16s_8(TCGContext *s, ARMCond cond, TCGReg rt,
731                            TCGReg rn, int imm8)
732{
733    tcg_out_memop_8(s, cond, INSN_LDRSH_IMM, rt, rn, imm8, 1, 0);
734}
735
736static void tcg_out_ld16s_r(TCGContext *s, ARMCond cond, TCGReg rt,
737                            TCGReg rn, TCGReg rm)
738{
739    tcg_out_memop_r(s, cond, INSN_LDRSH_REG, rt, rn, rm, 1, 1, 0);
740}
741
742static void tcg_out_ld8_12(TCGContext *s, ARMCond cond, TCGReg rt,
743                           TCGReg rn, int imm12)
744{
745    tcg_out_memop_12(s, cond, INSN_LDRB_IMM, rt, rn, imm12, 1, 0);
746}
747
748static void tcg_out_st8_12(TCGContext *s, ARMCond cond, TCGReg rt,
749                           TCGReg rn, int imm12)
750{
751    tcg_out_memop_12(s, cond, INSN_STRB_IMM, rt, rn, imm12, 1, 0);
752}
753
754static void tcg_out_ld8_r(TCGContext *s, ARMCond cond, TCGReg rt,
755                          TCGReg rn, TCGReg rm)
756{
757    tcg_out_memop_r(s, cond, INSN_LDRB_REG, rt, rn, rm, 1, 1, 0);
758}
759
760static void tcg_out_st8_r(TCGContext *s, ARMCond cond, TCGReg rt,
761                          TCGReg rn, TCGReg rm)
762{
763    tcg_out_memop_r(s, cond, INSN_STRB_REG, rt, rn, rm, 1, 1, 0);
764}
765
766static void tcg_out_ld8s_8(TCGContext *s, ARMCond cond, TCGReg rt,
767                           TCGReg rn, int imm8)
768{
769    tcg_out_memop_8(s, cond, INSN_LDRSB_IMM, rt, rn, imm8, 1, 0);
770}
771
772static void tcg_out_ld8s_r(TCGContext *s, ARMCond cond, TCGReg rt,
773                           TCGReg rn, TCGReg rm)
774{
775    tcg_out_memop_r(s, cond, INSN_LDRSB_REG, rt, rn, rm, 1, 1, 0);
776}
777
778static void tcg_out_movi_pool(TCGContext *s, ARMCond cond,
779                              TCGReg rd, uint32_t arg)
780{
781    new_pool_label(s, arg, R_ARM_PC13, s->code_ptr, 0);
782    tcg_out_ld32_12(s, cond, rd, TCG_REG_PC, 0);
783}
784
785static void tcg_out_movi32(TCGContext *s, ARMCond cond,
786                           TCGReg rd, uint32_t arg)
787{
788    int imm12, diff, opc, sh1, sh2;
789    uint32_t tt0, tt1, tt2;
790
791    /* Check a single MOV/MVN before anything else.  */
792    imm12 = encode_imm(arg);
793    if (imm12 >= 0) {
794        tcg_out_dat_imm(s, cond, ARITH_MOV, rd, 0, imm12);
795        return;
796    }
797    imm12 = encode_imm(~arg);
798    if (imm12 >= 0) {
799        tcg_out_dat_imm(s, cond, ARITH_MVN, rd, 0, imm12);
800        return;
801    }
802
803    /* Check for a pc-relative address.  This will usually be the TB,
804       or within the TB, which is immediately before the code block.  */
805    diff = tcg_pcrel_diff(s, (void *)arg) - 8;
806    if (diff >= 0) {
807        imm12 = encode_imm(diff);
808        if (imm12 >= 0) {
809            tcg_out_dat_imm(s, cond, ARITH_ADD, rd, TCG_REG_PC, imm12);
810            return;
811        }
812    } else {
813        imm12 = encode_imm(-diff);
814        if (imm12 >= 0) {
815            tcg_out_dat_imm(s, cond, ARITH_SUB, rd, TCG_REG_PC, imm12);
816            return;
817        }
818    }
819
820    /* Use movw + movt.  */
821    if (use_armv7_instructions) {
822        /* movw */
823        tcg_out32(s, (cond << 28) | 0x03000000 | (rd << 12)
824                  | ((arg << 4) & 0x000f0000) | (arg & 0xfff));
825        if (arg & 0xffff0000) {
826            /* movt */
827            tcg_out32(s, (cond << 28) | 0x03400000 | (rd << 12)
828                      | ((arg >> 12) & 0x000f0000) | ((arg >> 16) & 0xfff));
829        }
830        return;
831    }
832
833    /* Look for sequences of two insns.  If we have lots of 1's, we can
834       shorten the sequence by beginning with mvn and then clearing
835       higher bits with eor.  */
836    tt0 = arg;
837    opc = ARITH_MOV;
838    if (ctpop32(arg) > 16) {
839        tt0 = ~arg;
840        opc = ARITH_MVN;
841    }
842    sh1 = ctz32(tt0) & ~1;
843    tt1 = tt0 & ~(0xff << sh1);
844    sh2 = ctz32(tt1) & ~1;
845    tt2 = tt1 & ~(0xff << sh2);
846    if (tt2 == 0) {
847        int rot;
848
849        rot = ((32 - sh1) << 7) & 0xf00;
850        tcg_out_dat_imm(s, cond, opc, rd,  0, ((tt0 >> sh1) & 0xff) | rot);
851        rot = ((32 - sh2) << 7) & 0xf00;
852        tcg_out_dat_imm(s, cond, ARITH_EOR, rd, rd,
853                        ((tt0 >> sh2) & 0xff) | rot);
854        return;
855    }
856
857    /* Otherwise, drop it into the constant pool.  */
858    tcg_out_movi_pool(s, cond, rd, arg);
859}
860
861/*
862 * Emit either the reg,imm or reg,reg form of a data-processing insn.
863 * rhs must satisfy the "rI" constraint.
864 */
865static void tcg_out_dat_rI(TCGContext *s, ARMCond cond, ARMInsn opc,
866                           TCGReg dst, TCGReg lhs, TCGArg rhs, int rhs_is_const)
867{
868    if (rhs_is_const) {
869        tcg_out_dat_imm(s, cond, opc, dst, lhs, encode_imm_nofail(rhs));
870    } else {
871        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
872    }
873}
874
875/*
876 * Emit either the reg,imm or reg,reg form of a data-processing insn.
877 * rhs must satisfy the "rIK" constraint.
878 */
879static void tcg_out_dat_IK(TCGContext *s, ARMCond cond, ARMInsn opc,
880                            ARMInsn opinv, TCGReg dst, TCGReg lhs, TCGArg rhs)
881{
882    int imm12 = encode_imm(rhs);
883    if (imm12 < 0) {
884        imm12 = encode_imm_nofail(~rhs);
885        opc = opinv;
886    }
887    tcg_out_dat_imm(s, cond, opc, dst, lhs, imm12);
888}
889
890static void tcg_out_dat_rIK(TCGContext *s, ARMCond cond, ARMInsn opc,
891                            ARMInsn opinv, TCGReg dst, TCGReg lhs, TCGArg rhs,
892                            bool rhs_is_const)
893{
894    if (rhs_is_const) {
895        tcg_out_dat_IK(s, cond, opc, opinv, dst, lhs, rhs);
896    } else {
897        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
898    }
899}
900
901static void tcg_out_dat_IN(TCGContext *s, ARMCond cond, ARMInsn opc,
902                           ARMInsn opneg, TCGReg dst, TCGReg lhs, TCGArg rhs)
903{
904    int imm12 = encode_imm(rhs);
905    if (imm12 < 0) {
906        imm12 = encode_imm_nofail(-rhs);
907        opc = opneg;
908    }
909    tcg_out_dat_imm(s, cond, opc, dst, lhs, imm12);
910}
911
912static void tcg_out_dat_rIN(TCGContext *s, ARMCond cond, ARMInsn opc,
913                            ARMInsn opneg, TCGReg dst, TCGReg lhs, TCGArg rhs,
914                            bool rhs_is_const)
915{
916    /* Emit either the reg,imm or reg,reg form of a data-processing insn.
917     * rhs must satisfy the "rIN" constraint.
918     */
919    if (rhs_is_const) {
920        tcg_out_dat_IN(s, cond, opc, opneg, dst, lhs, rhs);
921    } else {
922        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
923    }
924}
925
926static void tcg_out_ext8s(TCGContext *s, TCGType t, TCGReg rd, TCGReg rn)
927{
928    /* sxtb */
929    tcg_out32(s, 0x06af0070 | (COND_AL << 28) | (rd << 12) | rn);
930}
931
932static void tcg_out_ext8u(TCGContext *s, TCGReg rd, TCGReg rn)
933{
934    tcg_out_dat_imm(s, COND_AL, ARITH_AND, rd, rn, 0xff);
935}
936
937static void tcg_out_ext16s(TCGContext *s, TCGType t, TCGReg rd, TCGReg rn)
938{
939    /* sxth */
940    tcg_out32(s, 0x06bf0070 | (COND_AL << 28) | (rd << 12) | rn);
941}
942
943static void tcg_out_ext16u(TCGContext *s, TCGReg rd, TCGReg rn)
944{
945    /* uxth */
946    tcg_out32(s, 0x06ff0070 | (COND_AL << 28) | (rd << 12) | rn);
947}
948
949static void tcg_out_ext32s(TCGContext *s, TCGReg rd, TCGReg rn)
950{
951    g_assert_not_reached();
952}
953
954static void tcg_out_ext32u(TCGContext *s, TCGReg rd, TCGReg rn)
955{
956    g_assert_not_reached();
957}
958
959static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg rd, TCGReg rn)
960{
961    g_assert_not_reached();
962}
963
964static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg rd, TCGReg rn)
965{
966    g_assert_not_reached();
967}
968
969static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rn)
970{
971    g_assert_not_reached();
972}
973
974static void tgen_deposit(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1,
975                         TCGReg a2, unsigned ofs, unsigned len)
976{
977    /* bfi/bfc */
978    tcg_debug_assert(a0 == a1);
979    tcg_out32(s, 0x07c00010 | (COND_AL << 28) | (a0 << 12) | a2
980              | (ofs << 7) | ((ofs + len - 1) << 16));
981}
982
983static void tgen_depositi(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1,
984                          tcg_target_long a2, unsigned ofs, unsigned len)
985{
986    /* bfi becomes bfc with rn == 15.  */
987    tgen_deposit(s, type, a0, a1, 15, ofs, len);
988}
989
990static const TCGOutOpDeposit outop_deposit = {
991    .base.static_constraint = C_O1_I2(r, 0, rZ),
992    .out_rrr = tgen_deposit,
993    .out_rri = tgen_depositi,
994};
995
996static void tgen_extract(TCGContext *s, TCGType type, TCGReg rd, TCGReg rn,
997                         unsigned ofs, unsigned len)
998{
999    /* According to gcc, AND can be faster. */
1000    if (ofs == 0 && len <= 8) {
1001        tcg_out_dat_imm(s, COND_AL, ARITH_AND, rd, rn,
1002                        encode_imm_nofail((1 << len) - 1));
1003        return;
1004    }
1005
1006    if (use_armv7_instructions) {
1007        /* ubfx */
1008        tcg_out32(s, 0x07e00050 | (COND_AL << 28) | (rd << 12) | rn
1009                  | (ofs << 7) | ((len - 1) << 16));
1010        return;
1011    }
1012
1013    assert(ofs % 8 == 0);
1014    switch (len) {
1015    case 8:
1016        /* uxtb */
1017        tcg_out32(s, 0x06ef0070 | (COND_AL << 28) |
1018                  (rd << 12) | (ofs << 7) | rn);
1019        break;
1020    case 16:
1021        /* uxth */
1022        tcg_out32(s, 0x06ff0070 | (COND_AL << 28) |
1023                  (rd << 12) | (ofs << 7) | rn);
1024        break;
1025    default:
1026        g_assert_not_reached();
1027    }
1028}
1029
1030static const TCGOutOpExtract outop_extract = {
1031    .base.static_constraint = C_O1_I1(r, r),
1032    .out_rr = tgen_extract,
1033};
1034
1035static void tgen_sextract(TCGContext *s, TCGType type, TCGReg rd, TCGReg rn,
1036                          unsigned ofs, unsigned len)
1037{
1038    if (use_armv7_instructions) {
1039        /* sbfx */
1040        tcg_out32(s, 0x07a00050 | (COND_AL << 28) | (rd << 12) | rn
1041                  | (ofs << 7) | ((len - 1) << 16));
1042        return;
1043    }
1044
1045    assert(ofs % 8 == 0);
1046    switch (len) {
1047    case 8:
1048        /* sxtb */
1049        tcg_out32(s, 0x06af0070 | (COND_AL << 28) |
1050                  (rd << 12) | (ofs << 7) | rn);
1051        break;
1052    case 16:
1053        /* sxth */
1054        tcg_out32(s, 0x06bf0070 | (COND_AL << 28) |
1055                  (rd << 12) | (ofs << 7) | rn);
1056        break;
1057    default:
1058        g_assert_not_reached();
1059    }
1060}
1061
1062static const TCGOutOpExtract outop_sextract = {
1063    .base.static_constraint = C_O1_I1(r, r),
1064    .out_rr = tgen_sextract,
1065};
1066
1067
1068static void tcg_out_ld32u(TCGContext *s, ARMCond cond,
1069                          TCGReg rd, TCGReg rn, int32_t offset)
1070{
1071    if (offset > 0xfff || offset < -0xfff) {
1072        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1073        tcg_out_ld32_r(s, cond, rd, rn, TCG_REG_TMP);
1074    } else
1075        tcg_out_ld32_12(s, cond, rd, rn, offset);
1076}
1077
1078static void tcg_out_st32(TCGContext *s, ARMCond cond,
1079                         TCGReg rd, TCGReg rn, int32_t offset)
1080{
1081    if (offset > 0xfff || offset < -0xfff) {
1082        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1083        tcg_out_st32_r(s, cond, rd, rn, TCG_REG_TMP);
1084    } else
1085        tcg_out_st32_12(s, cond, rd, rn, offset);
1086}
1087
1088/*
1089 * The _goto case is normally between TBs within the same code buffer, and
1090 * with the code buffer limited to 16MB we wouldn't need the long case.
1091 * But we also use it for the tail-call to the qemu_ld/st helpers, which does.
1092 */
1093static void tcg_out_goto(TCGContext *s, ARMCond cond, const tcg_insn_unit *addr)
1094{
1095    intptr_t addri = (intptr_t)addr;
1096    ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1097    bool arm_mode = !(addri & 1);
1098
1099    if (arm_mode && disp - 8 < 0x01fffffd && disp - 8 > -0x01fffffd) {
1100        tcg_out_b_imm(s, cond, disp);
1101        return;
1102    }
1103
1104    /* LDR is interworking from v5t. */
1105    tcg_out_movi_pool(s, cond, TCG_REG_PC, addri);
1106}
1107
1108/*
1109 * The call case is mostly used for helpers - so it's not unreasonable
1110 * for them to be beyond branch range.
1111 */
1112static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *addr)
1113{
1114    intptr_t addri = (intptr_t)addr;
1115    ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1116    bool arm_mode = !(addri & 1);
1117
1118    if (disp - 8 < 0x02000000 && disp - 8 >= -0x02000000) {
1119        if (arm_mode) {
1120            tcg_out_bl_imm(s, COND_AL, disp);
1121        } else {
1122            tcg_out_blx_imm(s, disp);
1123        }
1124        return;
1125    }
1126
1127    tcg_out_movi32(s, COND_AL, TCG_REG_TMP, addri);
1128    tcg_out_blx_reg(s, COND_AL, TCG_REG_TMP);
1129}
1130
1131static void tcg_out_call(TCGContext *s, const tcg_insn_unit *addr,
1132                         const TCGHelperInfo *info)
1133{
1134    tcg_out_call_int(s, addr);
1135}
1136
1137static void tcg_out_goto_label(TCGContext *s, ARMCond cond, TCGLabel *l)
1138{
1139    if (l->has_value) {
1140        tcg_out_goto(s, cond, l->u.value_ptr);
1141    } else {
1142        tcg_out_reloc(s, s->code_ptr, R_ARM_PC24, l, 0);
1143        tcg_out_b_imm(s, cond, 0);
1144    }
1145}
1146
1147static void tcg_out_br(TCGContext *s, TCGLabel *l)
1148{
1149    tcg_out_goto_label(s, COND_AL, l);
1150}
1151
1152static void tcg_out_mb(TCGContext *s, unsigned a0)
1153{
1154    if (use_armv7_instructions) {
1155        tcg_out32(s, INSN_DMB_ISH);
1156    } else {
1157        tcg_out32(s, INSN_DMB_MCR);
1158    }
1159}
1160
1161static TCGCond tgen_cmp(TCGContext *s, TCGCond cond, TCGReg a, TCGReg b)
1162{
1163    if (is_tst_cond(cond)) {
1164        tcg_out_dat_reg(s, COND_AL, ARITH_TST, 0, a, b, SHIFT_IMM_LSL(0));
1165        return tcg_tst_eqne_cond(cond);
1166    }
1167    tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, a, b, SHIFT_IMM_LSL(0));
1168    return cond;
1169}
1170
1171static TCGCond tgen_cmpi(TCGContext *s, TCGCond cond, TCGReg a, TCGArg b)
1172{
1173    int imm12;
1174
1175    if (!is_tst_cond(cond)) {
1176        tcg_out_dat_IN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0, a, b);
1177        return cond;
1178    }
1179
1180    /*
1181     * The compare constraints allow rIN, but TST does not support N.
1182     * Be prepared to load the constant into a scratch register.
1183     */
1184    imm12 = encode_imm(b);
1185    if (imm12 >= 0) {
1186        tcg_out_dat_imm(s, COND_AL, ARITH_TST, 0, a, imm12);
1187    } else {
1188        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, b);
1189        tcg_out_dat_reg(s, COND_AL, ARITH_TST, 0,
1190                        a, TCG_REG_TMP, SHIFT_IMM_LSL(0));
1191    }
1192    return tcg_tst_eqne_cond(cond);
1193}
1194
1195static TCGCond tcg_out_cmp(TCGContext *s, TCGCond cond, TCGReg a,
1196                           TCGArg b, int b_const)
1197{
1198    if (b_const) {
1199        return tgen_cmpi(s, cond, a, b);
1200    } else {
1201        return tgen_cmp(s, cond, a, b);
1202    }
1203}
1204
1205static TCGCond tcg_out_cmp2(TCGContext *s, TCGCond cond, TCGReg al, TCGReg ah,
1206                            TCGArg bl, bool const_bl, TCGArg bh, bool const_bh)
1207{
1208    switch (cond) {
1209    case TCG_COND_EQ:
1210    case TCG_COND_NE:
1211    case TCG_COND_LTU:
1212    case TCG_COND_LEU:
1213    case TCG_COND_GTU:
1214    case TCG_COND_GEU:
1215        /*
1216         * We perform a conditional comparison.  If the high half is
1217         * equal, then overwrite the flags with the comparison of the
1218         * low half.  The resulting flags cover the whole.
1219         */
1220        tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0, ah, bh, const_bh);
1221        tcg_out_dat_rI(s, COND_EQ, ARITH_CMP, 0, al, bl, const_bl);
1222        return cond;
1223
1224    case TCG_COND_TSTEQ:
1225    case TCG_COND_TSTNE:
1226        /* Similar, but with TST instead of CMP. */
1227        tcg_out_dat_rI(s, COND_AL, ARITH_TST, 0, ah, bh, const_bh);
1228        tcg_out_dat_rI(s, COND_EQ, ARITH_TST, 0, al, bl, const_bl);
1229        return tcg_tst_eqne_cond(cond);
1230
1231    case TCG_COND_LT:
1232    case TCG_COND_GE:
1233        /* We perform a double-word subtraction and examine the result.
1234           We do not actually need the result of the subtract, so the
1235           low part "subtract" is a compare.  For the high half we have
1236           no choice but to compute into a temporary.  */
1237        tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0, al, bl, const_bl);
1238        tcg_out_dat_rI(s, COND_AL, ARITH_SBC | TO_CPSR,
1239                       TCG_REG_TMP, ah, bh, const_bh);
1240        return cond;
1241
1242    case TCG_COND_LE:
1243    case TCG_COND_GT:
1244        /* Similar, but with swapped arguments, via reversed subtract.  */
1245        tcg_out_dat_rI(s, COND_AL, ARITH_RSB | TO_CPSR,
1246                       TCG_REG_TMP, al, bl, const_bl);
1247        tcg_out_dat_rI(s, COND_AL, ARITH_RSC | TO_CPSR,
1248                       TCG_REG_TMP, ah, bh, const_bh);
1249        return tcg_swap_cond(cond);
1250
1251    default:
1252        g_assert_not_reached();
1253    }
1254}
1255
1256/*
1257 * Note that TCGReg references Q-registers.
1258 * Q-regno = 2 * D-regno, so shift left by 1 while inserting.
1259 */
1260static uint32_t encode_vd(TCGReg rd)
1261{
1262    tcg_debug_assert(rd >= TCG_REG_Q0);
1263    return (extract32(rd, 3, 1) << 22) | (extract32(rd, 0, 3) << 13);
1264}
1265
1266static uint32_t encode_vn(TCGReg rn)
1267{
1268    tcg_debug_assert(rn >= TCG_REG_Q0);
1269    return (extract32(rn, 3, 1) << 7) | (extract32(rn, 0, 3) << 17);
1270}
1271
1272static uint32_t encode_vm(TCGReg rm)
1273{
1274    tcg_debug_assert(rm >= TCG_REG_Q0);
1275    return (extract32(rm, 3, 1) << 5) | (extract32(rm, 0, 3) << 1);
1276}
1277
1278static void tcg_out_vreg2(TCGContext *s, ARMInsn insn, int q, int vece,
1279                          TCGReg d, TCGReg m)
1280{
1281    tcg_out32(s, insn | (vece << 18) | (q << 6) |
1282              encode_vd(d) | encode_vm(m));
1283}
1284
1285static void tcg_out_vreg3(TCGContext *s, ARMInsn insn, int q, int vece,
1286                          TCGReg d, TCGReg n, TCGReg m)
1287{
1288    tcg_out32(s, insn | (vece << 20) | (q << 6) |
1289              encode_vd(d) | encode_vn(n) | encode_vm(m));
1290}
1291
1292static void tcg_out_vmovi(TCGContext *s, TCGReg rd,
1293                          int q, int op, int cmode, uint8_t imm8)
1294{
1295    tcg_out32(s, INSN_VMOVI | encode_vd(rd) | (q << 6) | (op << 5)
1296              | (cmode << 8) | extract32(imm8, 0, 4)
1297              | (extract32(imm8, 4, 3) << 16)
1298              | (extract32(imm8, 7, 1) << 24));
1299}
1300
1301static void tcg_out_vshifti(TCGContext *s, ARMInsn insn, int q,
1302                            TCGReg rd, TCGReg rm, int l_imm6)
1303{
1304    tcg_out32(s, insn | (q << 6) | encode_vd(rd) | encode_vm(rm) |
1305              (extract32(l_imm6, 6, 1) << 7) |
1306              (extract32(l_imm6, 0, 6) << 16));
1307}
1308
1309static void tcg_out_vldst(TCGContext *s, ARMInsn insn,
1310                          TCGReg rd, TCGReg rn, int offset)
1311{
1312    if (offset != 0) {
1313        if (check_fit_imm(offset) || check_fit_imm(-offset)) {
1314            tcg_out_dat_rIN(s, COND_AL, ARITH_ADD, ARITH_SUB,
1315                            TCG_REG_TMP, rn, offset, true);
1316        } else {
1317            tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, offset);
1318            tcg_out_dat_reg(s, COND_AL, ARITH_ADD,
1319                            TCG_REG_TMP, TCG_REG_TMP, rn, 0);
1320        }
1321        rn = TCG_REG_TMP;
1322    }
1323    tcg_out32(s, insn | (rn << 16) | encode_vd(rd) | 0xf);
1324}
1325
1326typedef struct {
1327    ARMCond cond;
1328    TCGReg base;
1329    int index;
1330    bool index_scratch;
1331    TCGAtomAlign aa;
1332} HostAddress;
1333
1334bool tcg_target_has_memory_bswap(MemOp memop)
1335{
1336    return false;
1337}
1338
1339static TCGReg ldst_ra_gen(TCGContext *s, const TCGLabelQemuLdst *l, int arg)
1340{
1341    /* We arrive at the slow path via "BLNE", so R14 contains l->raddr. */
1342    return TCG_REG_R14;
1343}
1344
1345static const TCGLdstHelperParam ldst_helper_param = {
1346    .ra_gen = ldst_ra_gen,
1347    .ntmp = 1,
1348    .tmp = { TCG_REG_TMP },
1349};
1350
1351static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1352{
1353    MemOp opc = get_memop(lb->oi);
1354
1355    if (!reloc_pc24(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1356        return false;
1357    }
1358
1359    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1360    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1361    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1362
1363    tcg_out_goto(s, COND_AL, lb->raddr);
1364    return true;
1365}
1366
1367static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1368{
1369    MemOp opc = get_memop(lb->oi);
1370
1371    if (!reloc_pc24(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1372        return false;
1373    }
1374
1375    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1376
1377    /* Tail-call to the helper, which will return to the fast path.  */
1378    tcg_out_goto(s, COND_AL, qemu_st_helpers[opc & MO_SIZE]);
1379    return true;
1380}
1381
1382/* We expect to use an 9-bit sign-magnitude negative offset from ENV.  */
1383#define MIN_TLB_MASK_TABLE_OFS  -256
1384
1385static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1386                                           TCGReg addr, MemOpIdx oi, bool is_ld)
1387{
1388    TCGLabelQemuLdst *ldst = NULL;
1389    MemOp opc = get_memop(oi);
1390    unsigned a_mask;
1391
1392    if (tcg_use_softmmu) {
1393        *h = (HostAddress){
1394            .cond = COND_AL,
1395            .base = addr,
1396            .index = TCG_REG_R1,
1397            .index_scratch = true,
1398        };
1399    } else {
1400        *h = (HostAddress){
1401            .cond = COND_AL,
1402            .base = addr,
1403            .index = guest_base ? TCG_REG_GUEST_BASE : -1,
1404            .index_scratch = false,
1405        };
1406    }
1407
1408    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, false);
1409    a_mask = (1 << h->aa.align) - 1;
1410
1411    if (tcg_use_softmmu) {
1412        int mem_index = get_mmuidx(oi);
1413        int cmp_off = is_ld ? offsetof(CPUTLBEntry, addr_read)
1414                            : offsetof(CPUTLBEntry, addr_write);
1415        int fast_off = tlb_mask_table_ofs(s, mem_index);
1416        unsigned s_mask = (1 << (opc & MO_SIZE)) - 1;
1417        TCGReg t_addr;
1418
1419        ldst = new_ldst_label(s);
1420        ldst->is_ld = is_ld;
1421        ldst->oi = oi;
1422        ldst->addr_reg = addr;
1423
1424        /* Load cpu->neg.tlb.f[mmu_idx].{mask,table} into {r0,r1}.  */
1425        QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, mask) != 0);
1426        QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, table) != 4);
1427        tcg_out_ldrd_8(s, COND_AL, TCG_REG_R0, TCG_AREG0, fast_off);
1428
1429        /* Extract the tlb index from the address into R0.  */
1430        tcg_out_dat_reg(s, COND_AL, ARITH_AND, TCG_REG_R0, TCG_REG_R0, addr,
1431                        SHIFT_IMM_LSR(TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS));
1432
1433        /*
1434         * Add the tlb_table pointer, creating the CPUTLBEntry address in R1.
1435         * Load the tlb comparator into R2 and the fast path addend into R1.
1436         */
1437        if (cmp_off == 0) {
1438            tcg_out_ld32_rwb(s, COND_AL, TCG_REG_R2, TCG_REG_R1, TCG_REG_R0);
1439        } else {
1440            tcg_out_dat_reg(s, COND_AL, ARITH_ADD,
1441                            TCG_REG_R1, TCG_REG_R1, TCG_REG_R0, 0);
1442            tcg_out_ld32_12(s, COND_AL, TCG_REG_R2, TCG_REG_R1, cmp_off);
1443        }
1444
1445        /* Load the tlb addend.  */
1446        tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R1,
1447                        offsetof(CPUTLBEntry, addend));
1448
1449        /*
1450         * Check alignment, check comparators.
1451         * Do this in 2-4 insns.  Use MOVW for v7, if possible,
1452         * to reduce the number of sequential conditional instructions.
1453         * Almost all guests have at least 4k pages, which means that we need
1454         * to clear at least 9 bits even for an 8-byte memory, which means it
1455         * isn't worth checking for an immediate operand for BIC.
1456         *
1457         * For unaligned accesses, test the page of the last unit of alignment.
1458         * This leaves the least significant alignment bits unchanged, and of
1459         * course must be zero.
1460         */
1461        t_addr = addr;
1462        if (a_mask < s_mask) {
1463            t_addr = TCG_REG_R0;
1464            tcg_out_dat_imm(s, COND_AL, ARITH_ADD, t_addr,
1465                            addr, s_mask - a_mask);
1466        }
1467        if (use_armv7_instructions && TARGET_PAGE_BITS <= 16) {
1468            tcg_out_movi32(s, COND_AL, TCG_REG_TMP, ~(TARGET_PAGE_MASK | a_mask));
1469            tcg_out_dat_reg(s, COND_AL, ARITH_BIC, TCG_REG_TMP,
1470                            t_addr, TCG_REG_TMP, 0);
1471            tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1472                            TCG_REG_R2, TCG_REG_TMP, 0);
1473        } else {
1474            if (a_mask) {
1475                tcg_debug_assert(a_mask <= 0xff);
1476                tcg_out_dat_imm(s, COND_AL, ARITH_TST, 0, addr, a_mask);
1477            }
1478            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_TMP, 0, t_addr,
1479                            SHIFT_IMM_LSR(TARGET_PAGE_BITS));
1480            tcg_out_dat_reg(s, (a_mask ? COND_EQ : COND_AL), ARITH_CMP,
1481                            0, TCG_REG_R2, TCG_REG_TMP,
1482                            SHIFT_IMM_LSL(TARGET_PAGE_BITS));
1483        }
1484    } else if (a_mask) {
1485        ldst = new_ldst_label(s);
1486        ldst->is_ld = is_ld;
1487        ldst->oi = oi;
1488        ldst->addr_reg = addr;
1489
1490        /* We are expecting alignment to max out at 7 */
1491        tcg_debug_assert(a_mask <= 0xff);
1492        /* tst addr, #mask */
1493        tcg_out_dat_imm(s, COND_AL, ARITH_TST, 0, addr, a_mask);
1494    }
1495
1496    return ldst;
1497}
1498
1499static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg datalo,
1500                                   TCGReg datahi, HostAddress h)
1501{
1502    TCGReg base;
1503
1504    /* Byte swapping is left to middle-end expansion. */
1505    tcg_debug_assert((opc & MO_BSWAP) == 0);
1506
1507    switch (opc & MO_SSIZE) {
1508    case MO_UB:
1509        if (h.index < 0) {
1510            tcg_out_ld8_12(s, h.cond, datalo, h.base, 0);
1511        } else {
1512            tcg_out_ld8_r(s, h.cond, datalo, h.base, h.index);
1513        }
1514        break;
1515    case MO_SB:
1516        if (h.index < 0) {
1517            tcg_out_ld8s_8(s, h.cond, datalo, h.base, 0);
1518        } else {
1519            tcg_out_ld8s_r(s, h.cond, datalo, h.base, h.index);
1520        }
1521        break;
1522    case MO_UW:
1523        if (h.index < 0) {
1524            tcg_out_ld16u_8(s, h.cond, datalo, h.base, 0);
1525        } else {
1526            tcg_out_ld16u_r(s, h.cond, datalo, h.base, h.index);
1527        }
1528        break;
1529    case MO_SW:
1530        if (h.index < 0) {
1531            tcg_out_ld16s_8(s, h.cond, datalo, h.base, 0);
1532        } else {
1533            tcg_out_ld16s_r(s, h.cond, datalo, h.base, h.index);
1534        }
1535        break;
1536    case MO_UL:
1537        if (h.index < 0) {
1538            tcg_out_ld32_12(s, h.cond, datalo, h.base, 0);
1539        } else {
1540            tcg_out_ld32_r(s, h.cond, datalo, h.base, h.index);
1541        }
1542        break;
1543    case MO_UQ:
1544        /* We used pair allocation for datalo, so already should be aligned. */
1545        tcg_debug_assert((datalo & 1) == 0);
1546        tcg_debug_assert(datahi == datalo + 1);
1547        /* LDRD requires alignment; double-check that. */
1548        if (memop_alignment_bits(opc) >= MO_64) {
1549            if (h.index < 0) {
1550                tcg_out_ldrd_8(s, h.cond, datalo, h.base, 0);
1551                break;
1552            }
1553            /*
1554             * Rm (the second address op) must not overlap Rt or Rt + 1.
1555             * Since datalo is aligned, we can simplify the test via alignment.
1556             * Flip the two address arguments if that works.
1557             */
1558            if ((h.index & ~1) != datalo) {
1559                tcg_out_ldrd_r(s, h.cond, datalo, h.base, h.index);
1560                break;
1561            }
1562            if ((h.base & ~1) != datalo) {
1563                tcg_out_ldrd_r(s, h.cond, datalo, h.index, h.base);
1564                break;
1565            }
1566        }
1567        if (h.index < 0) {
1568            base = h.base;
1569            if (datalo == h.base) {
1570                tcg_out_mov_reg(s, h.cond, TCG_REG_TMP, base);
1571                base = TCG_REG_TMP;
1572            }
1573        } else if (h.index_scratch) {
1574            tcg_out_ld32_rwb(s, h.cond, datalo, h.index, h.base);
1575            tcg_out_ld32_12(s, h.cond, datahi, h.index, 4);
1576            break;
1577        } else {
1578            tcg_out_dat_reg(s, h.cond, ARITH_ADD, TCG_REG_TMP,
1579                            h.base, h.index, SHIFT_IMM_LSL(0));
1580            base = TCG_REG_TMP;
1581        }
1582        tcg_out_ld32_12(s, h.cond, datalo, base, 0);
1583        tcg_out_ld32_12(s, h.cond, datahi, base, 4);
1584        break;
1585    default:
1586        g_assert_not_reached();
1587    }
1588}
1589
1590static void tgen_qemu_ld(TCGContext *s, TCGType type, TCGReg data,
1591                         TCGReg addr, MemOpIdx oi)
1592{
1593    MemOp opc = get_memop(oi);
1594    TCGLabelQemuLdst *ldst;
1595    HostAddress h;
1596
1597    ldst = prepare_host_addr(s, &h, addr, oi, true);
1598    if (ldst) {
1599        ldst->type = type;
1600        ldst->datalo_reg = data;
1601        ldst->datahi_reg = -1;
1602
1603        /*
1604         * This a conditional BL only to load a pointer within this
1605         * opcode into LR for the slow path.  We will not be using
1606         * the value for a tail call.
1607         */
1608        ldst->label_ptr[0] = s->code_ptr;
1609        tcg_out_bl_imm(s, COND_NE, 0);
1610    }
1611
1612    tcg_out_qemu_ld_direct(s, opc, data, -1, h);
1613
1614    if (ldst) {
1615        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1616    }
1617}
1618
1619static const TCGOutOpQemuLdSt outop_qemu_ld = {
1620    .base.static_constraint = C_O1_I1(r, q),
1621    .out = tgen_qemu_ld,
1622};
1623
1624static void tgen_qemu_ld2(TCGContext *s, TCGType type, TCGReg datalo,
1625                          TCGReg datahi, TCGReg addr, MemOpIdx oi)
1626{
1627    MemOp opc = get_memop(oi);
1628    TCGLabelQemuLdst *ldst;
1629    HostAddress h;
1630
1631    ldst = prepare_host_addr(s, &h, addr, oi, true);
1632    if (ldst) {
1633        ldst->type = type;
1634        ldst->datalo_reg = datalo;
1635        ldst->datahi_reg = datahi;
1636
1637        /*
1638         * This a conditional BL only to load a pointer within this
1639         * opcode into LR for the slow path.  We will not be using
1640         * the value for a tail call.
1641         */
1642        ldst->label_ptr[0] = s->code_ptr;
1643        tcg_out_bl_imm(s, COND_NE, 0);
1644    }
1645
1646    tcg_out_qemu_ld_direct(s, opc, datalo, datahi, h);
1647
1648    if (ldst) {
1649        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1650    }
1651}
1652
1653static const TCGOutOpQemuLdSt2 outop_qemu_ld2 = {
1654    .base.static_constraint = C_O2_I1(e, p, q),
1655    .out = tgen_qemu_ld2,
1656};
1657
1658static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg datalo,
1659                                   TCGReg datahi, HostAddress h)
1660{
1661    /* Byte swapping is left to middle-end expansion. */
1662    tcg_debug_assert((opc & MO_BSWAP) == 0);
1663
1664    switch (opc & MO_SIZE) {
1665    case MO_8:
1666        if (h.index < 0) {
1667            tcg_out_st8_12(s, h.cond, datalo, h.base, 0);
1668        } else {
1669            tcg_out_st8_r(s, h.cond, datalo, h.base, h.index);
1670        }
1671        break;
1672    case MO_16:
1673        if (h.index < 0) {
1674            tcg_out_st16_8(s, h.cond, datalo, h.base, 0);
1675        } else {
1676            tcg_out_st16_r(s, h.cond, datalo, h.base, h.index);
1677        }
1678        break;
1679    case MO_32:
1680        if (h.index < 0) {
1681            tcg_out_st32_12(s, h.cond, datalo, h.base, 0);
1682        } else {
1683            tcg_out_st32_r(s, h.cond, datalo, h.base, h.index);
1684        }
1685        break;
1686    case MO_64:
1687        /* We used pair allocation for datalo, so already should be aligned. */
1688        tcg_debug_assert((datalo & 1) == 0);
1689        tcg_debug_assert(datahi == datalo + 1);
1690        /* STRD requires alignment; double-check that. */
1691        if (memop_alignment_bits(opc) >= MO_64) {
1692            if (h.index < 0) {
1693                tcg_out_strd_8(s, h.cond, datalo, h.base, 0);
1694            } else {
1695                tcg_out_strd_r(s, h.cond, datalo, h.base, h.index);
1696            }
1697        } else if (h.index < 0) {
1698            tcg_out_st32_12(s, h.cond, datalo, h.base, 0);
1699            tcg_out_st32_12(s, h.cond, datahi, h.base, 4);
1700        } else if (h.index_scratch) {
1701            tcg_out_st32_rwb(s, h.cond, datalo, h.index, h.base);
1702            tcg_out_st32_12(s, h.cond, datahi, h.index, 4);
1703        } else {
1704            tcg_out_dat_reg(s, h.cond, ARITH_ADD, TCG_REG_TMP,
1705                            h.base, h.index, SHIFT_IMM_LSL(0));
1706            tcg_out_st32_12(s, h.cond, datalo, TCG_REG_TMP, 0);
1707            tcg_out_st32_12(s, h.cond, datahi, TCG_REG_TMP, 4);
1708        }
1709        break;
1710    default:
1711        g_assert_not_reached();
1712    }
1713}
1714
1715static void tgen_qemu_st(TCGContext *s, TCGType type, TCGReg data,
1716                         TCGReg addr, MemOpIdx oi)
1717{
1718    MemOp opc = get_memop(oi);
1719    TCGLabelQemuLdst *ldst;
1720    HostAddress h;
1721
1722    ldst = prepare_host_addr(s, &h, addr, oi, false);
1723    if (ldst) {
1724        ldst->type = type;
1725        ldst->datalo_reg = data;
1726        ldst->datahi_reg = -1;
1727
1728        h.cond = COND_EQ;
1729        tcg_out_qemu_st_direct(s, opc, data, -1, h);
1730
1731        /* The conditional call is last, as we're going to return here. */
1732        ldst->label_ptr[0] = s->code_ptr;
1733        tcg_out_bl_imm(s, COND_NE, 0);
1734        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1735    } else {
1736        tcg_out_qemu_st_direct(s, opc, data, -1, h);
1737    }
1738}
1739
1740static const TCGOutOpQemuLdSt outop_qemu_st = {
1741    .base.static_constraint = C_O0_I2(q, q),
1742    .out = tgen_qemu_st,
1743};
1744
1745static void tgen_qemu_st2(TCGContext *s, TCGType type, TCGReg datalo,
1746                          TCGReg datahi, TCGReg addr, MemOpIdx oi)
1747{
1748    MemOp opc = get_memop(oi);
1749    TCGLabelQemuLdst *ldst;
1750    HostAddress h;
1751
1752    ldst = prepare_host_addr(s, &h, addr, oi, false);
1753    if (ldst) {
1754        ldst->type = type;
1755        ldst->datalo_reg = datalo;
1756        ldst->datahi_reg = datahi;
1757
1758        h.cond = COND_EQ;
1759        tcg_out_qemu_st_direct(s, opc, datalo, datahi, h);
1760
1761        /* The conditional call is last, as we're going to return here. */
1762        ldst->label_ptr[0] = s->code_ptr;
1763        tcg_out_bl_imm(s, COND_NE, 0);
1764        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1765    } else {
1766        tcg_out_qemu_st_direct(s, opc, datalo, datahi, h);
1767    }
1768}
1769
1770static const TCGOutOpQemuLdSt2 outop_qemu_st2 = {
1771    .base.static_constraint = C_O0_I3(Q, p, q),
1772    .out = tgen_qemu_st2,
1773};
1774
1775static void tcg_out_epilogue(TCGContext *s);
1776
1777static void tcg_out_exit_tb(TCGContext *s, uintptr_t arg)
1778{
1779    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, arg);
1780    tcg_out_epilogue(s);
1781}
1782
1783static void tcg_out_goto_tb(TCGContext *s, int which)
1784{
1785    uintptr_t i_addr;
1786    intptr_t i_disp;
1787
1788    /* Direct branch will be patched by tb_target_set_jmp_target. */
1789    set_jmp_insn_offset(s, which);
1790    tcg_out32(s, INSN_NOP);
1791
1792    /* When branch is out of range, fall through to indirect. */
1793    i_addr = get_jmp_target_addr(s, which);
1794    i_disp = tcg_pcrel_diff(s, (void *)i_addr) - 8;
1795    tcg_debug_assert(i_disp < 0);
1796    if (i_disp >= -0xfff) {
1797        tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, i_disp);
1798    } else {
1799        /*
1800         * The TB is close, but outside the 12 bits addressable by
1801         * the load.  We can extend this to 20 bits with a sub of a
1802         * shifted immediate from pc.
1803         */
1804        int h = -i_disp;
1805        int l = -(h & 0xfff);
1806
1807        h = encode_imm_nofail(h + l);
1808        tcg_out_dat_imm(s, COND_AL, ARITH_SUB, TCG_REG_R0, TCG_REG_PC, h);
1809        tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, l);
1810    }
1811    set_jmp_reset_offset(s, which);
1812}
1813
1814static void tcg_out_goto_ptr(TCGContext *s, TCGReg a0)
1815{
1816    tcg_out_b_reg(s, COND_AL, a0);
1817}
1818
1819void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1820                              uintptr_t jmp_rx, uintptr_t jmp_rw)
1821{
1822    uintptr_t addr = tb->jmp_target_addr[n];
1823    ptrdiff_t offset = addr - (jmp_rx + 8);
1824    tcg_insn_unit insn;
1825
1826    /* Either directly branch, or fall through to indirect branch. */
1827    if (offset == sextract64(offset, 0, 26)) {
1828        /* B <addr> */
1829        insn = deposit32((COND_AL << 28) | INSN_B, 0, 24, offset >> 2);
1830    } else {
1831        insn = INSN_NOP;
1832    }
1833
1834    qatomic_set((uint32_t *)jmp_rw, insn);
1835    flush_idcache_range(jmp_rx, jmp_rw, 4);
1836}
1837
1838
1839static void tgen_add(TCGContext *s, TCGType type,
1840                     TCGReg a0, TCGReg a1, TCGReg a2)
1841{
1842    tcg_out_dat_reg(s, COND_AL, ARITH_ADD, a0, a1, a2, SHIFT_IMM_LSL(0));
1843}
1844
1845static void tgen_addi(TCGContext *s, TCGType type,
1846                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1847{
1848    tcg_out_dat_IN(s, COND_AL, ARITH_ADD, ARITH_SUB, a0, a1, a2);
1849}
1850
1851static const TCGOutOpBinary outop_add = {
1852    .base.static_constraint = C_O1_I2(r, r, rIN),
1853    .out_rrr = tgen_add,
1854    .out_rri = tgen_addi,
1855};
1856
1857static void tgen_addco(TCGContext *s, TCGType type,
1858                       TCGReg a0, TCGReg a1, TCGReg a2)
1859{
1860    tcg_out_dat_reg(s, COND_AL, ARITH_ADD | TO_CPSR,
1861                    a0, a1, a2, SHIFT_IMM_LSL(0));
1862}
1863
1864static void tgen_addco_imm(TCGContext *s, TCGType type,
1865                           TCGReg a0, TCGReg a1, tcg_target_long a2)
1866{
1867    tcg_out_dat_IN(s, COND_AL, ARITH_ADD | TO_CPSR, ARITH_SUB | TO_CPSR,
1868                   a0, a1, a2);
1869}
1870
1871static const TCGOutOpBinary outop_addco = {
1872    .base.static_constraint = C_O1_I2(r, r, rIN),
1873    .out_rrr = tgen_addco,
1874    .out_rri = tgen_addco_imm,
1875};
1876
1877static void tgen_addci(TCGContext *s, TCGType type,
1878                       TCGReg a0, TCGReg a1, TCGReg a2)
1879{
1880    tcg_out_dat_reg(s, COND_AL, ARITH_ADC, a0, a1, a2, SHIFT_IMM_LSL(0));
1881}
1882
1883static void tgen_addci_imm(TCGContext *s, TCGType type,
1884                           TCGReg a0, TCGReg a1, tcg_target_long a2)
1885{
1886    tcg_out_dat_IK(s, COND_AL, ARITH_ADC, ARITH_SBC, a0, a1, a2);
1887}
1888
1889static const TCGOutOpAddSubCarry outop_addci = {
1890    .base.static_constraint = C_O1_I2(r, r, rIK),
1891    .out_rrr = tgen_addci,
1892    .out_rri = tgen_addci_imm,
1893};
1894
1895static void tgen_addcio(TCGContext *s, TCGType type,
1896                        TCGReg a0, TCGReg a1, TCGReg a2)
1897{
1898    tcg_out_dat_reg(s, COND_AL, ARITH_ADC | TO_CPSR,
1899                    a0, a1, a2, SHIFT_IMM_LSL(0));
1900}
1901
1902static void tgen_addcio_imm(TCGContext *s, TCGType type,
1903                            TCGReg a0, TCGReg a1, tcg_target_long a2)
1904{
1905    tcg_out_dat_IK(s, COND_AL, ARITH_ADC | TO_CPSR, ARITH_SBC | TO_CPSR,
1906                   a0, a1, a2);
1907}
1908
1909static const TCGOutOpBinary outop_addcio = {
1910    .base.static_constraint = C_O1_I2(r, r, rIK),
1911    .out_rrr = tgen_addcio,
1912    .out_rri = tgen_addcio_imm,
1913};
1914
1915/* Set C to @c; NZVQ all set to 0. */
1916static void tcg_out_movi_apsr_c(TCGContext *s, bool c)
1917{
1918    int imm12 = encode_imm_nofail(c << 29);
1919    tcg_out32(s, (COND_AL << 28) | INSN_MSRI_CPSR | 0x80000 | imm12);
1920}
1921
1922static void tcg_out_set_carry(TCGContext *s)
1923{
1924    tcg_out_movi_apsr_c(s, 1);
1925}
1926
1927static void tgen_and(TCGContext *s, TCGType type,
1928                     TCGReg a0, TCGReg a1, TCGReg a2)
1929{
1930    tcg_out_dat_reg(s, COND_AL, ARITH_AND, a0, a1, a2, SHIFT_IMM_LSL(0));
1931}
1932
1933static void tgen_andi(TCGContext *s, TCGType type,
1934                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1935{
1936    tcg_out_dat_IK(s, COND_AL, ARITH_AND, ARITH_BIC, a0, a1, a2);
1937}
1938
1939static const TCGOutOpBinary outop_and = {
1940    .base.static_constraint = C_O1_I2(r, r, rIK),
1941    .out_rrr = tgen_and,
1942    .out_rri = tgen_andi,
1943};
1944
1945static void tgen_andc(TCGContext *s, TCGType type,
1946                      TCGReg a0, TCGReg a1, TCGReg a2)
1947{
1948    tcg_out_dat_reg(s, COND_AL, ARITH_BIC, a0, a1, a2, SHIFT_IMM_LSL(0));
1949}
1950
1951static const TCGOutOpBinary outop_andc = {
1952    .base.static_constraint = C_O1_I2(r, r, r),
1953    .out_rrr = tgen_andc,
1954};
1955
1956static void tgen_clz(TCGContext *s, TCGType type,
1957                     TCGReg a0, TCGReg a1, TCGReg a2)
1958{
1959    tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0, a1, 0);
1960    tcg_out_dat_reg(s, COND_NE, INSN_CLZ, a0, 0, a1, 0);
1961    tcg_out_mov_reg(s, COND_EQ, a0, a2);
1962}
1963
1964static void tgen_clzi(TCGContext *s, TCGType type,
1965                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1966{
1967    if (a2 == 32) {
1968        tcg_out_dat_reg(s, COND_AL, INSN_CLZ, a0, 0, a1, 0);
1969    } else {
1970        tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0, a1, 0);
1971        tcg_out_dat_reg(s, COND_NE, INSN_CLZ, a0, 0, a1, 0);
1972        tcg_out_movi32(s, COND_EQ, a0, a2);
1973    }
1974}
1975
1976static const TCGOutOpBinary outop_clz = {
1977    .base.static_constraint = C_O1_I2(r, r, rIK),
1978    .out_rrr = tgen_clz,
1979    .out_rri = tgen_clzi,
1980};
1981
1982static const TCGOutOpUnary outop_ctpop = {
1983    .base.static_constraint = C_NotImplemented,
1984};
1985
1986static void tgen_ctz(TCGContext *s, TCGType type,
1987                     TCGReg a0, TCGReg a1, TCGReg a2)
1988{
1989    tcg_out_dat_reg(s, COND_AL, INSN_RBIT, TCG_REG_TMP, 0, a1, 0);
1990    tgen_clz(s, TCG_TYPE_I32, a0, TCG_REG_TMP, a2);
1991}
1992
1993static void tgen_ctzi(TCGContext *s, TCGType type,
1994                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1995{
1996    tcg_out_dat_reg(s, COND_AL, INSN_RBIT, TCG_REG_TMP, 0, a1, 0);
1997    tgen_clzi(s, TCG_TYPE_I32, a0, TCG_REG_TMP, a2);
1998}
1999
2000static TCGConstraintSetIndex cset_ctz(TCGType type, unsigned flags)
2001{
2002    return use_armv7_instructions ? C_O1_I2(r, r, rIK) : C_NotImplemented;
2003}
2004
2005static const TCGOutOpBinary outop_ctz = {
2006    .base.static_constraint = C_Dynamic,
2007    .base.dynamic_constraint = cset_ctz,
2008    .out_rrr = tgen_ctz,
2009    .out_rri = tgen_ctzi,
2010};
2011
2012static TCGConstraintSetIndex cset_idiv(TCGType type, unsigned flags)
2013{
2014    return use_idiv_instructions ? C_O1_I2(r, r, r) : C_NotImplemented;
2015}
2016
2017static void tgen_divs(TCGContext *s, TCGType type,
2018                      TCGReg a0, TCGReg a1, TCGReg a2)
2019{
2020    /* sdiv */
2021    tcg_out32(s, 0x0710f010 | (COND_AL << 28) | (a0 << 16) | a1 | (a2 << 8));
2022}
2023
2024static const TCGOutOpBinary outop_divs = {
2025    .base.static_constraint = C_Dynamic,
2026    .base.dynamic_constraint = cset_idiv,
2027    .out_rrr = tgen_divs,
2028};
2029
2030static const TCGOutOpDivRem outop_divs2 = {
2031    .base.static_constraint = C_NotImplemented,
2032};
2033
2034static void tgen_divu(TCGContext *s, TCGType type,
2035                      TCGReg a0, TCGReg a1, TCGReg a2)
2036{
2037    /* udiv */
2038    tcg_out32(s, 0x0730f010 | (COND_AL << 28) | (a0 << 16) | a1 | (a2 << 8));
2039}
2040
2041static const TCGOutOpBinary outop_divu = {
2042    .base.static_constraint = C_Dynamic,
2043    .base.dynamic_constraint = cset_idiv,
2044    .out_rrr = tgen_divu,
2045};
2046
2047static const TCGOutOpDivRem outop_divu2 = {
2048    .base.static_constraint = C_NotImplemented,
2049};
2050
2051static const TCGOutOpBinary outop_eqv = {
2052    .base.static_constraint = C_NotImplemented,
2053};
2054
2055static void tgen_mul(TCGContext *s, TCGType type,
2056                     TCGReg a0, TCGReg a1, TCGReg a2)
2057{
2058    /* mul */
2059    tcg_out32(s, (COND_AL << 28) | 0x90 | (a0 << 16) | (a1 << 8) | a2);
2060}
2061
2062static const TCGOutOpBinary outop_mul = {
2063    .base.static_constraint = C_O1_I2(r, r, r),
2064    .out_rrr = tgen_mul,
2065};
2066
2067static void tgen_muls2(TCGContext *s, TCGType type,
2068                       TCGReg rd0, TCGReg rd1, TCGReg rn, TCGReg rm)
2069{
2070    /* smull */
2071    tcg_out32(s, (COND_AL << 28) | 0x00c00090 |
2072              (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
2073}
2074
2075static const TCGOutOpMul2 outop_muls2 = {
2076    .base.static_constraint = C_O2_I2(r, r, r, r),
2077    .out_rrrr = tgen_muls2,
2078};
2079
2080static const TCGOutOpBinary outop_mulsh = {
2081    .base.static_constraint = C_NotImplemented,
2082};
2083
2084static void tgen_mulu2(TCGContext *s, TCGType type,
2085                       TCGReg rd0, TCGReg rd1, TCGReg rn, TCGReg rm)
2086{
2087    /* umull */
2088    tcg_out32(s, (COND_AL << 28) | 0x00800090 |
2089              (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
2090}
2091
2092static const TCGOutOpMul2 outop_mulu2 = {
2093    .base.static_constraint = C_O2_I2(r, r, r, r),
2094    .out_rrrr = tgen_mulu2,
2095};
2096
2097static const TCGOutOpBinary outop_muluh = {
2098    .base.static_constraint = C_NotImplemented,
2099};
2100
2101static const TCGOutOpBinary outop_nand = {
2102    .base.static_constraint = C_NotImplemented,
2103};
2104
2105static const TCGOutOpBinary outop_nor = {
2106    .base.static_constraint = C_NotImplemented,
2107};
2108
2109static void tgen_or(TCGContext *s, TCGType type,
2110                     TCGReg a0, TCGReg a1, TCGReg a2)
2111{
2112    tcg_out_dat_reg(s, COND_AL, ARITH_ORR, a0, a1, a2, SHIFT_IMM_LSL(0));
2113}
2114
2115static void tgen_ori(TCGContext *s, TCGType type,
2116                     TCGReg a0, TCGReg a1, tcg_target_long a2)
2117{
2118    tcg_out_dat_imm(s, COND_AL, ARITH_ORR, a0, a1, encode_imm_nofail(a2));
2119}
2120
2121static const TCGOutOpBinary outop_or = {
2122    .base.static_constraint = C_O1_I2(r, r, rI),
2123    .out_rrr = tgen_or,
2124    .out_rri = tgen_ori,
2125};
2126
2127static const TCGOutOpBinary outop_orc = {
2128    .base.static_constraint = C_NotImplemented,
2129};
2130
2131static const TCGOutOpBinary outop_rems = {
2132    .base.static_constraint = C_NotImplemented,
2133};
2134
2135static const TCGOutOpBinary outop_remu = {
2136    .base.static_constraint = C_NotImplemented,
2137};
2138
2139static const TCGOutOpBinary outop_rotl = {
2140    .base.static_constraint = C_NotImplemented,
2141};
2142
2143static void tgen_rotr(TCGContext *s, TCGType type,
2144                      TCGReg a0, TCGReg a1, TCGReg a2)
2145{
2146    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1, SHIFT_REG_ROR(a2));
2147}
2148
2149static void tgen_rotri(TCGContext *s, TCGType type,
2150                       TCGReg a0, TCGReg a1, tcg_target_long a2)
2151{
2152    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1, SHIFT_IMM_ROR(a2 & 0x1f));
2153}
2154
2155static const TCGOutOpBinary outop_rotr = {
2156    .base.static_constraint = C_O1_I2(r, r, ri),
2157    .out_rrr = tgen_rotr,
2158    .out_rri = tgen_rotri,
2159};
2160
2161static void tgen_sar(TCGContext *s, TCGType type,
2162                     TCGReg a0, TCGReg a1, TCGReg a2)
2163{
2164    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1, SHIFT_REG_ASR(a2));
2165}
2166
2167static void tgen_sari(TCGContext *s, TCGType type,
2168                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2169{
2170    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1,
2171                    SHIFT_IMM_ASR(a2 & 0x1f));
2172}
2173
2174static const TCGOutOpBinary outop_sar = {
2175    .base.static_constraint = C_O1_I2(r, r, ri),
2176    .out_rrr = tgen_sar,
2177    .out_rri = tgen_sari,
2178};
2179
2180static void tgen_shl(TCGContext *s, TCGType type,
2181                     TCGReg a0, TCGReg a1, TCGReg a2)
2182{
2183    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1, SHIFT_REG_LSL(a2));
2184}
2185
2186static void tgen_shli(TCGContext *s, TCGType type,
2187                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2188{
2189    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1,
2190                    SHIFT_IMM_LSL(a2 & 0x1f));
2191}
2192
2193static const TCGOutOpBinary outop_shl = {
2194    .base.static_constraint = C_O1_I2(r, r, ri),
2195    .out_rrr = tgen_shl,
2196    .out_rri = tgen_shli,
2197};
2198
2199static void tgen_shr(TCGContext *s, TCGType type,
2200                     TCGReg a0, TCGReg a1, TCGReg a2)
2201{
2202    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1, SHIFT_REG_LSR(a2));
2203}
2204
2205static void tgen_shri(TCGContext *s, TCGType type,
2206                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2207{
2208    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1,
2209                    SHIFT_IMM_LSR(a2 & 0x1f));
2210}
2211
2212static const TCGOutOpBinary outop_shr = {
2213    .base.static_constraint = C_O1_I2(r, r, ri),
2214    .out_rrr = tgen_shr,
2215    .out_rri = tgen_shri,
2216};
2217
2218static void tgen_sub(TCGContext *s, TCGType type,
2219                     TCGReg a0, TCGReg a1, TCGReg a2)
2220{
2221    tcg_out_dat_reg(s, COND_AL, ARITH_SUB, a0, a1, a2, SHIFT_IMM_LSL(0));
2222}
2223
2224static void tgen_subfi(TCGContext *s, TCGType type,
2225                       TCGReg a0, tcg_target_long a1, TCGReg a2)
2226{
2227    tcg_out_dat_imm(s, COND_AL, ARITH_RSB, a0, a2, encode_imm_nofail(a1));
2228}
2229
2230static const TCGOutOpSubtract outop_sub = {
2231    .base.static_constraint = C_O1_I2(r, rI, r),
2232    .out_rrr = tgen_sub,
2233    .out_rir = tgen_subfi,
2234};
2235
2236static void tgen_subbo_rrr(TCGContext *s, TCGType type,
2237                           TCGReg a0, TCGReg a1, TCGReg a2)
2238{
2239    tcg_out_dat_reg(s, COND_AL, ARITH_SUB | TO_CPSR,
2240                    a0, a1, a2, SHIFT_IMM_LSL(0));
2241}
2242
2243static void tgen_subbo_rri(TCGContext *s, TCGType type,
2244                           TCGReg a0, TCGReg a1, tcg_target_long a2)
2245{
2246    tcg_out_dat_IN(s, COND_AL, ARITH_SUB | TO_CPSR, ARITH_ADD | TO_CPSR,
2247                   a0, a1, a2);
2248}
2249
2250static void tgen_subbo_rir(TCGContext *s, TCGType type,
2251                           TCGReg a0, tcg_target_long a1, TCGReg a2)
2252{
2253    tcg_out_dat_imm(s, COND_AL, ARITH_RSB | TO_CPSR,
2254                    a0, a2, encode_imm_nofail(a1));
2255}
2256
2257static void tgen_subbo_rii(TCGContext *s, TCGType type,
2258                           TCGReg a0, tcg_target_long a1, tcg_target_long a2)
2259{
2260    tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_TMP, a2);
2261    tgen_subbo_rir(s, TCG_TYPE_I32, a0, a1, TCG_REG_TMP);
2262}
2263
2264static const TCGOutOpAddSubCarry outop_subbo = {
2265    .base.static_constraint = C_O1_I2(r, rI, rIN),
2266    .out_rrr = tgen_subbo_rrr,
2267    .out_rri = tgen_subbo_rri,
2268    .out_rir = tgen_subbo_rir,
2269    .out_rii = tgen_subbo_rii,
2270};
2271
2272static void tgen_subbi_rrr(TCGContext *s, TCGType type,
2273                           TCGReg a0, TCGReg a1, TCGReg a2)
2274{
2275    tcg_out_dat_reg(s, COND_AL, ARITH_SBC,
2276                    a0, a1, a2, SHIFT_IMM_LSL(0));
2277}
2278
2279static void tgen_subbi_rri(TCGContext *s, TCGType type,
2280                           TCGReg a0, TCGReg a1, tcg_target_long a2)
2281{
2282    tcg_out_dat_IK(s, COND_AL, ARITH_SBC, ARITH_ADC, a0, a1, a2);
2283}
2284
2285static void tgen_subbi_rir(TCGContext *s, TCGType type,
2286                           TCGReg a0, tcg_target_long a1, TCGReg a2)
2287{
2288    tcg_out_dat_imm(s, COND_AL, ARITH_RSC, a0, a2, encode_imm_nofail(a1));
2289}
2290
2291static void tgen_subbi_rii(TCGContext *s, TCGType type,
2292                           TCGReg a0, tcg_target_long a1, tcg_target_long a2)
2293{
2294    tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_TMP, a2);
2295    tgen_subbi_rir(s, TCG_TYPE_I32, a0, a1, TCG_REG_TMP);
2296}
2297
2298static const TCGOutOpAddSubCarry outop_subbi = {
2299    .base.static_constraint = C_O1_I2(r, rI, rIK),
2300    .out_rrr = tgen_subbi_rrr,
2301    .out_rri = tgen_subbi_rri,
2302    .out_rir = tgen_subbi_rir,
2303    .out_rii = tgen_subbi_rii,
2304};
2305
2306static void tgen_subbio_rrr(TCGContext *s, TCGType type,
2307                            TCGReg a0, TCGReg a1, TCGReg a2)
2308{
2309    tcg_out_dat_reg(s, COND_AL, ARITH_SBC | TO_CPSR,
2310                    a0, a1, a2, SHIFT_IMM_LSL(0));
2311}
2312
2313static void tgen_subbio_rri(TCGContext *s, TCGType type,
2314                            TCGReg a0, TCGReg a1, tcg_target_long a2)
2315{
2316    tcg_out_dat_IK(s, COND_AL, ARITH_SBC | TO_CPSR, ARITH_ADC | TO_CPSR,
2317                   a0, a1, a2);
2318}
2319
2320static void tgen_subbio_rir(TCGContext *s, TCGType type,
2321                            TCGReg a0, tcg_target_long a1, TCGReg a2)
2322{
2323    tcg_out_dat_imm(s, COND_AL, ARITH_RSC | TO_CPSR,
2324                    a0, a2, encode_imm_nofail(a1));
2325}
2326
2327static void tgen_subbio_rii(TCGContext *s, TCGType type,
2328                            TCGReg a0, tcg_target_long a1, tcg_target_long a2)
2329{
2330    tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_TMP, a2);
2331    tgen_subbio_rir(s, TCG_TYPE_I32, a0, a1, TCG_REG_TMP);
2332}
2333
2334static const TCGOutOpAddSubCarry outop_subbio = {
2335    .base.static_constraint = C_O1_I2(r, rI, rIK),
2336    .out_rrr = tgen_subbio_rrr,
2337    .out_rri = tgen_subbio_rri,
2338    .out_rir = tgen_subbio_rir,
2339    .out_rii = tgen_subbio_rii,
2340};
2341
2342static void tcg_out_set_borrow(TCGContext *s)
2343{
2344    tcg_out_movi_apsr_c(s, 0);  /* borrow = !carry */
2345}
2346
2347static void tgen_xor(TCGContext *s, TCGType type,
2348                     TCGReg a0, TCGReg a1, TCGReg a2)
2349{
2350    tcg_out_dat_reg(s, COND_AL, ARITH_EOR, a0, a1, a2, SHIFT_IMM_LSL(0));
2351}
2352
2353static void tgen_xori(TCGContext *s, TCGType type,
2354                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2355{
2356    tcg_out_dat_imm(s, COND_AL, ARITH_EOR, a0, a1, encode_imm_nofail(a2));
2357}
2358
2359static const TCGOutOpBinary outop_xor = {
2360    .base.static_constraint = C_O1_I2(r, r, rI),
2361    .out_rrr = tgen_xor,
2362    .out_rri = tgen_xori,
2363};
2364
2365static void tgen_bswap16(TCGContext *s, TCGType type,
2366                         TCGReg rd, TCGReg rn, unsigned flags)
2367{
2368    if (flags & TCG_BSWAP_OS) {
2369        /* revsh */
2370        tcg_out32(s, 0x06ff0fb0 | (COND_AL << 28) | (rd << 12) | rn);
2371        return;
2372    }
2373
2374    /* rev16 */
2375    tcg_out32(s, 0x06bf0fb0 | (COND_AL << 28) | (rd << 12) | rn);
2376    if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2377        tcg_out_ext16u(s, rd, rd);
2378    }
2379}
2380
2381static const TCGOutOpBswap outop_bswap16 = {
2382    .base.static_constraint = C_O1_I1(r, r),
2383    .out_rr = tgen_bswap16,
2384};
2385
2386static void tgen_bswap32(TCGContext *s, TCGType type,
2387                         TCGReg rd, TCGReg rn, unsigned flags)
2388{
2389    /* rev */
2390    tcg_out32(s, 0x06bf0f30 | (COND_AL << 28) | (rd << 12) | rn);
2391}
2392
2393static const TCGOutOpBswap outop_bswap32 = {
2394    .base.static_constraint = C_O1_I1(r, r),
2395    .out_rr = tgen_bswap32,
2396};
2397
2398static const TCGOutOpUnary outop_bswap64 = {
2399    .base.static_constraint = C_NotImplemented,
2400};
2401
2402static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2403{
2404    tgen_subfi(s, type, a0, 0, a1);
2405}
2406
2407static const TCGOutOpUnary outop_neg = {
2408    .base.static_constraint = C_O1_I1(r, r),
2409    .out_rr = tgen_neg,
2410};
2411
2412static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2413{
2414    tcg_out_dat_reg(s, COND_AL, ARITH_MVN, a0, 0, a1, SHIFT_IMM_LSL(0));
2415}
2416
2417static const TCGOutOpUnary outop_not = {
2418    .base.static_constraint = C_O1_I1(r, r),
2419    .out_rr = tgen_not,
2420};
2421
2422static void tgen_brcond(TCGContext *s, TCGType type, TCGCond cond,
2423                        TCGReg a0, TCGReg a1, TCGLabel *l)
2424{
2425    cond = tgen_cmp(s, cond, a0, a1);
2426    tcg_out_goto_label(s, tcg_cond_to_arm_cond[cond], l);
2427}
2428
2429static void tgen_brcondi(TCGContext *s, TCGType type, TCGCond cond,
2430                         TCGReg a0, tcg_target_long a1, TCGLabel *l)
2431{
2432    cond = tgen_cmpi(s, cond, a0, a1);
2433    tcg_out_goto_label(s, tcg_cond_to_arm_cond[cond], l);
2434}
2435
2436static const TCGOutOpBrcond outop_brcond = {
2437    .base.static_constraint = C_O0_I2(r, rIN),
2438    .out_rr = tgen_brcond,
2439    .out_ri = tgen_brcondi,
2440};
2441
2442static void finish_setcond(TCGContext *s, TCGCond cond, TCGReg ret, bool neg)
2443{
2444    tcg_out_movi32(s, tcg_cond_to_arm_cond[tcg_invert_cond(cond)], ret, 0);
2445    tcg_out_movi32(s, tcg_cond_to_arm_cond[cond], ret, neg ? -1 : 1);
2446}
2447
2448static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
2449                         TCGReg a0, TCGReg a1, TCGReg a2)
2450{
2451    cond = tgen_cmp(s, cond, a1, a2);
2452    finish_setcond(s, cond, a0, false);
2453}
2454
2455static void tgen_setcondi(TCGContext *s, TCGType type, TCGCond cond,
2456                          TCGReg a0, TCGReg a1, tcg_target_long a2)
2457{
2458    cond = tgen_cmpi(s, cond, a1, a2);
2459    finish_setcond(s, cond, a0, false);
2460}
2461
2462static const TCGOutOpSetcond outop_setcond = {
2463    .base.static_constraint = C_O1_I2(r, r, rIN),
2464    .out_rrr = tgen_setcond,
2465    .out_rri = tgen_setcondi,
2466};
2467
2468static void tgen_negsetcond(TCGContext *s, TCGType type, TCGCond cond,
2469                            TCGReg a0, TCGReg a1, TCGReg a2)
2470{
2471    cond = tgen_cmp(s, cond, a1, a2);
2472    finish_setcond(s, cond, a0, true);
2473}
2474
2475static void tgen_negsetcondi(TCGContext *s, TCGType type, TCGCond cond,
2476                             TCGReg a0, TCGReg a1, tcg_target_long a2)
2477{
2478    cond = tgen_cmpi(s, cond, a1, a2);
2479    finish_setcond(s, cond, a0, true);
2480}
2481
2482static const TCGOutOpSetcond outop_negsetcond = {
2483    .base.static_constraint = C_O1_I2(r, r, rIN),
2484    .out_rrr = tgen_negsetcond,
2485    .out_rri = tgen_negsetcondi,
2486};
2487
2488static void tgen_movcond(TCGContext *s, TCGType type, TCGCond cond,
2489                         TCGReg ret, TCGReg c1, TCGArg c2, bool const_c2,
2490                         TCGArg vt, bool const_vt, TCGArg vf, bool consf_vf)
2491{
2492    cond = tcg_out_cmp(s, cond, c1, c2, const_c2);
2493    tcg_out_dat_rIK(s, tcg_cond_to_arm_cond[cond], ARITH_MOV, ARITH_MVN,
2494                    ret, 0, vt, const_vt);
2495}
2496
2497static const TCGOutOpMovcond outop_movcond = {
2498    .base.static_constraint = C_O1_I4(r, r, rIN, rIK, 0),
2499    .out = tgen_movcond,
2500};
2501
2502static void tgen_brcond2(TCGContext *s, TCGCond cond, TCGReg al, TCGReg ah,
2503                         TCGArg bl, bool const_bl, TCGArg bh, bool const_bh,
2504                         TCGLabel *l)
2505{
2506    cond = tcg_out_cmp2(s, cond, al, ah, bl, const_bl, bh, const_bh);
2507    tcg_out_goto_label(s, tcg_cond_to_arm_cond[cond], l);
2508}
2509
2510static const TCGOutOpBrcond2 outop_brcond2 = {
2511    .base.static_constraint = C_O0_I4(r, r, rI, rI),
2512    .out = tgen_brcond2,
2513};
2514
2515static void tgen_setcond2(TCGContext *s, TCGCond cond, TCGReg ret,
2516                          TCGReg al, TCGReg ah,
2517                          TCGArg bl, bool const_bl,
2518                          TCGArg bh, bool const_bh)
2519{
2520    cond = tcg_out_cmp2(s, cond, al, ah, bl, const_bl, bh, const_bh);
2521    finish_setcond(s, cond, ret, false);
2522}
2523
2524static const TCGOutOpSetcond2 outop_setcond2 = {
2525    .base.static_constraint = C_O1_I4(r, r, r, rI, rI),
2526    .out = tgen_setcond2,
2527};
2528
2529static void tgen_extract2(TCGContext *s, TCGType type, TCGReg a0,
2530                          TCGReg a1, TCGReg a2, unsigned shr)
2531{
2532    /* We can do extract2 in 2 insns, vs the 3 required otherwise.  */
2533    tgen_shli(s, TCG_TYPE_I32, TCG_REG_TMP, a2, 32 - shr);
2534    tcg_out_dat_reg(s, COND_AL, ARITH_ORR, a0, TCG_REG_TMP,
2535                    a1, SHIFT_IMM_LSR(shr));
2536}
2537
2538static const TCGOutOpExtract2 outop_extract2 = {
2539    .base.static_constraint = C_O1_I2(r, r, r),
2540    .out_rrr = tgen_extract2,
2541};
2542
2543static void tgen_ld8u(TCGContext *s, TCGType type, TCGReg rd,
2544                      TCGReg rn, ptrdiff_t offset)
2545{
2546    if (offset > 0xfff || offset < -0xfff) {
2547        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, offset);
2548        tcg_out_ld8_r(s, COND_AL, rd, rn, TCG_REG_TMP);
2549    } else {
2550        tcg_out_ld8_12(s, COND_AL, rd, rn, offset);
2551    }
2552}
2553
2554static const TCGOutOpLoad outop_ld8u = {
2555    .base.static_constraint = C_O1_I1(r, r),
2556    .out = tgen_ld8u,
2557};
2558
2559static void tgen_ld8s(TCGContext *s, TCGType type, TCGReg rd,
2560                      TCGReg rn, ptrdiff_t offset)
2561{
2562    if (offset > 0xff || offset < -0xff) {
2563        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, offset);
2564        tcg_out_ld8s_r(s, COND_AL, rd, rn, TCG_REG_TMP);
2565    } else {
2566        tcg_out_ld8s_8(s, COND_AL, rd, rn, offset);
2567    }
2568}
2569
2570static const TCGOutOpLoad outop_ld8s = {
2571    .base.static_constraint = C_O1_I1(r, r),
2572    .out = tgen_ld8s,
2573};
2574
2575static void tgen_ld16u(TCGContext *s, TCGType type, TCGReg rd,
2576                       TCGReg rn, ptrdiff_t offset)
2577{
2578    if (offset > 0xff || offset < -0xff) {
2579        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, offset);
2580        tcg_out_ld16u_r(s, COND_AL, rd, rn, TCG_REG_TMP);
2581    } else {
2582        tcg_out_ld16u_8(s, COND_AL, rd, rn, offset);
2583    }
2584}
2585
2586static const TCGOutOpLoad outop_ld16u = {
2587    .base.static_constraint = C_O1_I1(r, r),
2588    .out = tgen_ld16u,
2589};
2590
2591static void tgen_ld16s(TCGContext *s, TCGType type, TCGReg rd,
2592                       TCGReg rn, ptrdiff_t offset)
2593{
2594    if (offset > 0xff || offset < -0xff) {
2595        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, offset);
2596        tcg_out_ld16s_r(s, COND_AL, rd, rn, TCG_REG_TMP);
2597    } else {
2598        tcg_out_ld16s_8(s, COND_AL, rd, rn, offset);
2599    }
2600}
2601
2602static const TCGOutOpLoad outop_ld16s = {
2603    .base.static_constraint = C_O1_I1(r, r),
2604    .out = tgen_ld16s,
2605};
2606
2607static void tgen_st8(TCGContext *s, TCGType type, TCGReg rd,
2608                     TCGReg rn, ptrdiff_t offset)
2609{
2610    if (offset > 0xfff || offset < -0xfff) {
2611        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, offset);
2612        tcg_out_st8_r(s, COND_AL, rd, rn, TCG_REG_TMP);
2613    } else {
2614        tcg_out_st8_12(s, COND_AL, rd, rn, offset);
2615    }
2616}
2617
2618static const TCGOutOpStore outop_st8 = {
2619    .base.static_constraint = C_O0_I2(r, r),
2620    .out_r = tgen_st8,
2621};
2622
2623static void tgen_st16(TCGContext *s, TCGType type, TCGReg rd,
2624                      TCGReg rn, ptrdiff_t offset)
2625{
2626    if (offset > 0xff || offset < -0xff) {
2627        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, offset);
2628        tcg_out_st16_r(s, COND_AL, rd, rn, TCG_REG_TMP);
2629    } else {
2630        tcg_out_st16_8(s, COND_AL, rd, rn, offset);
2631    }
2632}
2633
2634static const TCGOutOpStore outop_st16 = {
2635    .base.static_constraint = C_O0_I2(r, r),
2636    .out_r = tgen_st16,
2637};
2638
2639static const TCGOutOpStore outop_st = {
2640    .base.static_constraint = C_O0_I2(r, r),
2641    .out_r = tcg_out_st,
2642};
2643
2644static TCGConstraintSetIndex
2645tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
2646{
2647    switch (op) {
2648    case INDEX_op_st_vec:
2649        return C_O0_I2(w, r);
2650    case INDEX_op_ld_vec:
2651    case INDEX_op_dupm_vec:
2652        return C_O1_I1(w, r);
2653    case INDEX_op_dup_vec:
2654        return C_O1_I1(w, wr);
2655    case INDEX_op_abs_vec:
2656    case INDEX_op_neg_vec:
2657    case INDEX_op_not_vec:
2658    case INDEX_op_shli_vec:
2659    case INDEX_op_shri_vec:
2660    case INDEX_op_sari_vec:
2661        return C_O1_I1(w, w);
2662    case INDEX_op_dup2_vec:
2663    case INDEX_op_add_vec:
2664    case INDEX_op_mul_vec:
2665    case INDEX_op_smax_vec:
2666    case INDEX_op_smin_vec:
2667    case INDEX_op_ssadd_vec:
2668    case INDEX_op_sssub_vec:
2669    case INDEX_op_sub_vec:
2670    case INDEX_op_umax_vec:
2671    case INDEX_op_umin_vec:
2672    case INDEX_op_usadd_vec:
2673    case INDEX_op_ussub_vec:
2674    case INDEX_op_xor_vec:
2675    case INDEX_op_arm_sshl_vec:
2676    case INDEX_op_arm_ushl_vec:
2677        return C_O1_I2(w, w, w);
2678    case INDEX_op_arm_sli_vec:
2679        return C_O1_I2(w, 0, w);
2680    case INDEX_op_or_vec:
2681    case INDEX_op_andc_vec:
2682        return C_O1_I2(w, w, wO);
2683    case INDEX_op_and_vec:
2684    case INDEX_op_orc_vec:
2685        return C_O1_I2(w, w, wV);
2686    case INDEX_op_cmp_vec:
2687        return C_O1_I2(w, w, wZ);
2688    case INDEX_op_bitsel_vec:
2689        return C_O1_I3(w, w, w, w);
2690    default:
2691        return C_NotImplemented;
2692    }
2693}
2694
2695static void tcg_target_init(TCGContext *s)
2696{
2697    /*
2698     * Only probe for the platform and capabilities if we haven't already
2699     * determined maximum values at compile time.
2700     */
2701#if !defined(use_idiv_instructions) || !defined(use_neon_instructions)
2702    {
2703        unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2704#ifndef use_idiv_instructions
2705        use_idiv_instructions = (hwcap & HWCAP_ARM_IDIVA) != 0;
2706#endif
2707#ifndef use_neon_instructions
2708        use_neon_instructions = (hwcap & HWCAP_ARM_NEON) != 0;
2709#endif
2710    }
2711#endif
2712
2713    if (__ARM_ARCH < 7) {
2714        const char *pl = (const char *)qemu_getauxval(AT_PLATFORM);
2715        if (pl != NULL && pl[0] == 'v' && pl[1] >= '4' && pl[1] <= '9') {
2716            arm_arch = pl[1] - '0';
2717        }
2718
2719        if (arm_arch < 6) {
2720            error_report("TCG: ARMv%d is unsupported; exiting", arm_arch);
2721            exit(EXIT_FAILURE);
2722        }
2723    }
2724
2725    tcg_target_available_regs[TCG_TYPE_I32] = ALL_GENERAL_REGS;
2726
2727    tcg_target_call_clobber_regs = 0;
2728    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
2729    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
2730    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
2731    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
2732    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
2733    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
2734
2735    if (use_neon_instructions) {
2736        tcg_target_available_regs[TCG_TYPE_V64]  = ALL_VECTOR_REGS;
2737        tcg_target_available_regs[TCG_TYPE_V128] = ALL_VECTOR_REGS;
2738
2739        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q0);
2740        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q1);
2741        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q2);
2742        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q3);
2743        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q8);
2744        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q9);
2745        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q10);
2746        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q11);
2747        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q12);
2748        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q13);
2749        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q14);
2750        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q15);
2751    }
2752
2753    s->reserved_regs = 0;
2754    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
2755    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP);
2756    tcg_regset_set_reg(s->reserved_regs, TCG_REG_PC);
2757    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP);
2758}
2759
2760static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
2761                       TCGReg arg1, intptr_t arg2)
2762{
2763    switch (type) {
2764    case TCG_TYPE_I32:
2765        tcg_out_ld32u(s, COND_AL, arg, arg1, arg2);
2766        return;
2767    case TCG_TYPE_V64:
2768        /* regs 1; size 8; align 8 */
2769        tcg_out_vldst(s, INSN_VLD1 | 0x7d0, arg, arg1, arg2);
2770        return;
2771    case TCG_TYPE_V128:
2772        /*
2773         * We have only 8-byte alignment for the stack per the ABI.
2774         * Rather than dynamically re-align the stack, it's easier
2775         * to simply not request alignment beyond that.  So:
2776         * regs 2; size 8; align 8
2777         */
2778        tcg_out_vldst(s, INSN_VLD1 | 0xad0, arg, arg1, arg2);
2779        return;
2780    default:
2781        g_assert_not_reached();
2782    }
2783}
2784
2785static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
2786                       TCGReg arg1, intptr_t arg2)
2787{
2788    switch (type) {
2789    case TCG_TYPE_I32:
2790        tcg_out_st32(s, COND_AL, arg, arg1, arg2);
2791        return;
2792    case TCG_TYPE_V64:
2793        /* regs 1; size 8; align 8 */
2794        tcg_out_vldst(s, INSN_VST1 | 0x7d0, arg, arg1, arg2);
2795        return;
2796    case TCG_TYPE_V128:
2797        /* See tcg_out_ld re alignment: regs 2; size 8; align 8 */
2798        tcg_out_vldst(s, INSN_VST1 | 0xad0, arg, arg1, arg2);
2799        return;
2800    default:
2801        g_assert_not_reached();
2802    }
2803}
2804
2805static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
2806                        TCGReg base, intptr_t ofs)
2807{
2808    return false;
2809}
2810
2811static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
2812{
2813    if (ret == arg) {
2814        return true;
2815    }
2816    switch (type) {
2817    case TCG_TYPE_I32:
2818        if (ret < TCG_REG_Q0 && arg < TCG_REG_Q0) {
2819            tcg_out_mov_reg(s, COND_AL, ret, arg);
2820            return true;
2821        }
2822        return false;
2823
2824    case TCG_TYPE_V64:
2825    case TCG_TYPE_V128:
2826        /* "VMOV D,N" is an alias for "VORR D,N,N". */
2827        tcg_out_vreg3(s, INSN_VORR, type - TCG_TYPE_V64, 0, ret, arg, arg);
2828        return true;
2829
2830    default:
2831        g_assert_not_reached();
2832    }
2833}
2834
2835static void tcg_out_movi(TCGContext *s, TCGType type,
2836                         TCGReg ret, tcg_target_long arg)
2837{
2838    tcg_debug_assert(type == TCG_TYPE_I32);
2839    tcg_debug_assert(ret < TCG_REG_Q0);
2840    tcg_out_movi32(s, COND_AL, ret, arg);
2841}
2842
2843static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
2844{
2845    return false;
2846}
2847
2848static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
2849                             tcg_target_long imm)
2850{
2851    int enc, opc = ARITH_ADD;
2852
2853    /* All of the easiest immediates to encode are positive. */
2854    if (imm < 0) {
2855        imm = -imm;
2856        opc = ARITH_SUB;
2857    }
2858    enc = encode_imm(imm);
2859    if (enc >= 0) {
2860        tcg_out_dat_imm(s, COND_AL, opc, rd, rs, enc);
2861    } else {
2862        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, imm);
2863        tcg_out_dat_reg(s, COND_AL, opc, rd, rs,
2864                        TCG_REG_TMP, SHIFT_IMM_LSL(0));
2865    }
2866}
2867
2868/* Type is always V128, with I64 elements.  */
2869static void tcg_out_dup2_vec(TCGContext *s, TCGReg rd, TCGReg rl, TCGReg rh)
2870{
2871    /* Move high element into place first. */
2872    /* VMOV Dd+1, Ds */
2873    tcg_out_vreg3(s, INSN_VORR | (1 << 12), 0, 0, rd, rh, rh);
2874    /* Move low element into place; tcg_out_mov will check for nop. */
2875    tcg_out_mov(s, TCG_TYPE_V64, rd, rl);
2876}
2877
2878static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
2879                            TCGReg rd, TCGReg rs)
2880{
2881    int q = type - TCG_TYPE_V64;
2882
2883    if (vece == MO_64) {
2884        if (type == TCG_TYPE_V128) {
2885            tcg_out_dup2_vec(s, rd, rs, rs);
2886        } else {
2887            tcg_out_mov(s, TCG_TYPE_V64, rd, rs);
2888        }
2889    } else if (rs < TCG_REG_Q0) {
2890        int b = (vece == MO_8);
2891        int e = (vece == MO_16);
2892        tcg_out32(s, INSN_VDUP_G | (b << 22) | (q << 21) | (e << 5) |
2893                  encode_vn(rd) | (rs << 12));
2894    } else {
2895        int imm4 = 1 << vece;
2896        tcg_out32(s, INSN_VDUP_S | (imm4 << 16) | (q << 6) |
2897                  encode_vd(rd) | encode_vm(rs));
2898    }
2899    return true;
2900}
2901
2902static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
2903                             TCGReg rd, TCGReg base, intptr_t offset)
2904{
2905    if (vece == MO_64) {
2906        tcg_out_ld(s, TCG_TYPE_V64, rd, base, offset);
2907        if (type == TCG_TYPE_V128) {
2908            tcg_out_dup2_vec(s, rd, rd, rd);
2909        }
2910    } else {
2911        int q = type - TCG_TYPE_V64;
2912        tcg_out_vldst(s, INSN_VLD1R | (vece << 6) | (q << 5),
2913                      rd, base, offset);
2914    }
2915    return true;
2916}
2917
2918static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
2919                             TCGReg rd, int64_t v64)
2920{
2921    int q = type - TCG_TYPE_V64;
2922    int cmode, imm8, i;
2923
2924    /* Test all bytes equal first.  */
2925    if (vece == MO_8) {
2926        tcg_out_vmovi(s, rd, q, 0, 0xe, v64);
2927        return;
2928    }
2929
2930    /*
2931     * Test all bytes 0x00 or 0xff second.  This can match cases that
2932     * might otherwise take 2 or 3 insns for MO_16 or MO_32 below.
2933     */
2934    for (i = imm8 = 0; i < 8; i++) {
2935        uint8_t byte = v64 >> (i * 8);
2936        if (byte == 0xff) {
2937            imm8 |= 1 << i;
2938        } else if (byte != 0) {
2939            goto fail_bytes;
2940        }
2941    }
2942    tcg_out_vmovi(s, rd, q, 1, 0xe, imm8);
2943    return;
2944 fail_bytes:
2945
2946    /*
2947     * Tests for various replications.  For each element width, if we
2948     * cannot find an expansion there's no point checking a larger
2949     * width because we already know by replication it cannot match.
2950     */
2951    if (vece == MO_16) {
2952        uint16_t v16 = v64;
2953
2954        if (is_shimm16(v16, &cmode, &imm8)) {
2955            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2956            return;
2957        }
2958        if (is_shimm16(~v16, &cmode, &imm8)) {
2959            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2960            return;
2961        }
2962
2963        /*
2964         * Otherwise, all remaining constants can be loaded in two insns:
2965         * rd = v16 & 0xff, rd |= v16 & 0xff00.
2966         */
2967        tcg_out_vmovi(s, rd, q, 0, 0x8, v16 & 0xff);
2968        tcg_out_vmovi(s, rd, q, 0, 0xb, v16 >> 8);   /* VORRI */
2969        return;
2970    }
2971
2972    if (vece == MO_32) {
2973        uint32_t v32 = v64;
2974
2975        if (is_shimm32(v32, &cmode, &imm8) ||
2976            is_soimm32(v32, &cmode, &imm8)) {
2977            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2978            return;
2979        }
2980        if (is_shimm32(~v32, &cmode, &imm8) ||
2981            is_soimm32(~v32, &cmode, &imm8)) {
2982            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2983            return;
2984        }
2985
2986        /*
2987         * Restrict the set of constants to those we can load with
2988         * two instructions.  Others we load from the pool.
2989         */
2990        i = is_shimm32_pair(v32, &cmode, &imm8);
2991        if (i) {
2992            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2993            tcg_out_vmovi(s, rd, q, 0, i | 1, extract32(v32, i * 4, 8));
2994            return;
2995        }
2996        i = is_shimm32_pair(~v32, &cmode, &imm8);
2997        if (i) {
2998            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2999            tcg_out_vmovi(s, rd, q, 1, i | 1, extract32(~v32, i * 4, 8));
3000            return;
3001        }
3002    }
3003
3004    /*
3005     * As a last resort, load from the constant pool.
3006     */
3007    if (!q || vece == MO_64) {
3008        new_pool_l2(s, R_ARM_PC11, s->code_ptr, 0, v64, v64 >> 32);
3009        /* VLDR Dd, [pc + offset] */
3010        tcg_out32(s, INSN_VLDR_D | encode_vd(rd) | (0xf << 16));
3011        if (q) {
3012            tcg_out_dup2_vec(s, rd, rd, rd);
3013        }
3014    } else {
3015        new_pool_label(s, (uint32_t)v64, R_ARM_PC8, s->code_ptr, 0);
3016        /* add tmp, pc, offset */
3017        tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_TMP, TCG_REG_PC, 0);
3018        tcg_out_dupm_vec(s, type, MO_32, rd, TCG_REG_TMP, 0);
3019    }
3020}
3021
3022static const ARMInsn vec_cmp_insn[16] = {
3023    [TCG_COND_EQ] = INSN_VCEQ,
3024    [TCG_COND_GT] = INSN_VCGT,
3025    [TCG_COND_GE] = INSN_VCGE,
3026    [TCG_COND_GTU] = INSN_VCGT_U,
3027    [TCG_COND_GEU] = INSN_VCGE_U,
3028};
3029
3030static const ARMInsn vec_cmp0_insn[16] = {
3031    [TCG_COND_EQ] = INSN_VCEQ0,
3032    [TCG_COND_GT] = INSN_VCGT0,
3033    [TCG_COND_GE] = INSN_VCGE0,
3034    [TCG_COND_LT] = INSN_VCLT0,
3035    [TCG_COND_LE] = INSN_VCLE0,
3036};
3037
3038static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
3039                           unsigned vecl, unsigned vece,
3040                           const TCGArg args[TCG_MAX_OP_ARGS],
3041                           const int const_args[TCG_MAX_OP_ARGS])
3042{
3043    TCGType type = vecl + TCG_TYPE_V64;
3044    unsigned q = vecl;
3045    TCGArg a0, a1, a2, a3;
3046    int cmode, imm8;
3047
3048    a0 = args[0];
3049    a1 = args[1];
3050    a2 = args[2];
3051
3052    switch (opc) {
3053    case INDEX_op_ld_vec:
3054        tcg_out_ld(s, type, a0, a1, a2);
3055        return;
3056    case INDEX_op_st_vec:
3057        tcg_out_st(s, type, a0, a1, a2);
3058        return;
3059    case INDEX_op_dupm_vec:
3060        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
3061        return;
3062    case INDEX_op_dup2_vec:
3063        tcg_out_dup2_vec(s, a0, a1, a2);
3064        return;
3065    case INDEX_op_abs_vec:
3066        tcg_out_vreg2(s, INSN_VABS, q, vece, a0, a1);
3067        return;
3068    case INDEX_op_neg_vec:
3069        tcg_out_vreg2(s, INSN_VNEG, q, vece, a0, a1);
3070        return;
3071    case INDEX_op_not_vec:
3072        tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a1);
3073        return;
3074    case INDEX_op_add_vec:
3075        tcg_out_vreg3(s, INSN_VADD, q, vece, a0, a1, a2);
3076        return;
3077    case INDEX_op_mul_vec:
3078        tcg_out_vreg3(s, INSN_VMUL, q, vece, a0, a1, a2);
3079        return;
3080    case INDEX_op_smax_vec:
3081        tcg_out_vreg3(s, INSN_VMAX, q, vece, a0, a1, a2);
3082        return;
3083    case INDEX_op_smin_vec:
3084        tcg_out_vreg3(s, INSN_VMIN, q, vece, a0, a1, a2);
3085        return;
3086    case INDEX_op_sub_vec:
3087        tcg_out_vreg3(s, INSN_VSUB, q, vece, a0, a1, a2);
3088        return;
3089    case INDEX_op_ssadd_vec:
3090        tcg_out_vreg3(s, INSN_VQADD, q, vece, a0, a1, a2);
3091        return;
3092    case INDEX_op_sssub_vec:
3093        tcg_out_vreg3(s, INSN_VQSUB, q, vece, a0, a1, a2);
3094        return;
3095    case INDEX_op_umax_vec:
3096        tcg_out_vreg3(s, INSN_VMAX_U, q, vece, a0, a1, a2);
3097        return;
3098    case INDEX_op_umin_vec:
3099        tcg_out_vreg3(s, INSN_VMIN_U, q, vece, a0, a1, a2);
3100        return;
3101    case INDEX_op_usadd_vec:
3102        tcg_out_vreg3(s, INSN_VQADD_U, q, vece, a0, a1, a2);
3103        return;
3104    case INDEX_op_ussub_vec:
3105        tcg_out_vreg3(s, INSN_VQSUB_U, q, vece, a0, a1, a2);
3106        return;
3107    case INDEX_op_xor_vec:
3108        tcg_out_vreg3(s, INSN_VEOR, q, 0, a0, a1, a2);
3109        return;
3110    case INDEX_op_arm_sshl_vec:
3111        /*
3112         * Note that Vm is the data and Vn is the shift count,
3113         * therefore the arguments appear reversed.
3114         */
3115        tcg_out_vreg3(s, INSN_VSHL_S, q, vece, a0, a2, a1);
3116        return;
3117    case INDEX_op_arm_ushl_vec:
3118        /* See above. */
3119        tcg_out_vreg3(s, INSN_VSHL_U, q, vece, a0, a2, a1);
3120        return;
3121    case INDEX_op_shli_vec:
3122        tcg_out_vshifti(s, INSN_VSHLI, q, a0, a1, a2 + (8 << vece));
3123        return;
3124    case INDEX_op_shri_vec:
3125        tcg_out_vshifti(s, INSN_VSHRI, q, a0, a1, (16 << vece) - a2);
3126        return;
3127    case INDEX_op_sari_vec:
3128        tcg_out_vshifti(s, INSN_VSARI, q, a0, a1, (16 << vece) - a2);
3129        return;
3130    case INDEX_op_arm_sli_vec:
3131        tcg_out_vshifti(s, INSN_VSLI, q, a0, a2, args[3] + (8 << vece));
3132        return;
3133
3134    case INDEX_op_andc_vec:
3135        if (!const_args[2]) {
3136            tcg_out_vreg3(s, INSN_VBIC, q, 0, a0, a1, a2);
3137            return;
3138        }
3139        a2 = ~a2;
3140        /* fall through */
3141    case INDEX_op_and_vec:
3142        if (const_args[2]) {
3143            is_shimm1632(~a2, &cmode, &imm8);
3144            if (a0 == a1) {
3145                tcg_out_vmovi(s, a0, q, 1, cmode | 1, imm8); /* VBICI */
3146                return;
3147            }
3148            tcg_out_vmovi(s, a0, q, 1, cmode, imm8); /* VMVNI */
3149            a2 = a0;
3150        }
3151        tcg_out_vreg3(s, INSN_VAND, q, 0, a0, a1, a2);
3152        return;
3153
3154    case INDEX_op_orc_vec:
3155        if (!const_args[2]) {
3156            tcg_out_vreg3(s, INSN_VORN, q, 0, a0, a1, a2);
3157            return;
3158        }
3159        a2 = ~a2;
3160        /* fall through */
3161    case INDEX_op_or_vec:
3162        if (const_args[2]) {
3163            is_shimm1632(a2, &cmode, &imm8);
3164            if (a0 == a1) {
3165                tcg_out_vmovi(s, a0, q, 0, cmode | 1, imm8); /* VORRI */
3166                return;
3167            }
3168            tcg_out_vmovi(s, a0, q, 0, cmode, imm8); /* VMOVI */
3169            a2 = a0;
3170        }
3171        tcg_out_vreg3(s, INSN_VORR, q, 0, a0, a1, a2);
3172        return;
3173
3174    case INDEX_op_cmp_vec:
3175        {
3176            TCGCond cond = args[3];
3177            ARMInsn insn;
3178
3179            switch (cond) {
3180            case TCG_COND_NE:
3181                if (const_args[2]) {
3182                    tcg_out_vreg3(s, INSN_VTST, q, vece, a0, a1, a1);
3183                } else {
3184                    tcg_out_vreg3(s, INSN_VCEQ, q, vece, a0, a1, a2);
3185                    tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a0);
3186                }
3187                break;
3188
3189            case TCG_COND_TSTNE:
3190            case TCG_COND_TSTEQ:
3191                if (const_args[2]) {
3192                    /* (x & 0) == 0 */
3193                    tcg_out_dupi_vec(s, type, MO_8, a0,
3194                                     -(cond == TCG_COND_TSTEQ));
3195                    break;
3196                }
3197                tcg_out_vreg3(s, INSN_VTST, q, vece, a0, a1, a2);
3198                if (cond == TCG_COND_TSTEQ) {
3199                    tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a0);
3200                }
3201                break;
3202
3203            default:
3204                if (const_args[2]) {
3205                    insn = vec_cmp0_insn[cond];
3206                    if (insn) {
3207                        tcg_out_vreg2(s, insn, q, vece, a0, a1);
3208                        return;
3209                    }
3210                    tcg_out_dupi_vec(s, type, MO_8, TCG_VEC_TMP, 0);
3211                    a2 = TCG_VEC_TMP;
3212                }
3213                insn = vec_cmp_insn[cond];
3214                if (insn == 0) {
3215                    TCGArg t;
3216                    t = a1, a1 = a2, a2 = t;
3217                    cond = tcg_swap_cond(cond);
3218                    insn = vec_cmp_insn[cond];
3219                    tcg_debug_assert(insn != 0);
3220                }
3221                tcg_out_vreg3(s, insn, q, vece, a0, a1, a2);
3222                break;
3223            }
3224        }
3225        return;
3226
3227    case INDEX_op_bitsel_vec:
3228        a3 = args[3];
3229        if (a0 == a3) {
3230            tcg_out_vreg3(s, INSN_VBIT, q, 0, a0, a2, a1);
3231        } else if (a0 == a2) {
3232            tcg_out_vreg3(s, INSN_VBIF, q, 0, a0, a3, a1);
3233        } else {
3234            tcg_out_mov(s, type, a0, a1);
3235            tcg_out_vreg3(s, INSN_VBSL, q, 0, a0, a2, a3);
3236        }
3237        return;
3238
3239    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
3240    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
3241    default:
3242        g_assert_not_reached();
3243    }
3244}
3245
3246int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3247{
3248    switch (opc) {
3249    case INDEX_op_add_vec:
3250    case INDEX_op_sub_vec:
3251    case INDEX_op_and_vec:
3252    case INDEX_op_andc_vec:
3253    case INDEX_op_or_vec:
3254    case INDEX_op_orc_vec:
3255    case INDEX_op_xor_vec:
3256    case INDEX_op_not_vec:
3257    case INDEX_op_shli_vec:
3258    case INDEX_op_shri_vec:
3259    case INDEX_op_sari_vec:
3260    case INDEX_op_ssadd_vec:
3261    case INDEX_op_sssub_vec:
3262    case INDEX_op_usadd_vec:
3263    case INDEX_op_ussub_vec:
3264    case INDEX_op_bitsel_vec:
3265        return 1;
3266    case INDEX_op_abs_vec:
3267    case INDEX_op_cmp_vec:
3268    case INDEX_op_mul_vec:
3269    case INDEX_op_neg_vec:
3270    case INDEX_op_smax_vec:
3271    case INDEX_op_smin_vec:
3272    case INDEX_op_umax_vec:
3273    case INDEX_op_umin_vec:
3274        return vece < MO_64;
3275    case INDEX_op_shlv_vec:
3276    case INDEX_op_shrv_vec:
3277    case INDEX_op_sarv_vec:
3278    case INDEX_op_rotli_vec:
3279    case INDEX_op_rotlv_vec:
3280    case INDEX_op_rotrv_vec:
3281        return -1;
3282    default:
3283        return 0;
3284    }
3285}
3286
3287void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3288                       TCGArg a0, ...)
3289{
3290    va_list va;
3291    TCGv_vec v0, v1, v2, t1, t2, c1;
3292    TCGArg a2;
3293
3294    va_start(va, a0);
3295    v0 = temp_tcgv_vec(arg_temp(a0));
3296    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3297    a2 = va_arg(va, TCGArg);
3298    va_end(va);
3299
3300    switch (opc) {
3301    case INDEX_op_shlv_vec:
3302        /*
3303         * Merely propagate shlv_vec to arm_ushl_vec.
3304         * In this way we don't set TCG_TARGET_HAS_shv_vec
3305         * because everything is done via expansion.
3306         */
3307        v2 = temp_tcgv_vec(arg_temp(a2));
3308        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(v0),
3309                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3310        break;
3311
3312    case INDEX_op_shrv_vec:
3313    case INDEX_op_sarv_vec:
3314        /* Right shifts are negative left shifts for NEON.  */
3315        v2 = temp_tcgv_vec(arg_temp(a2));
3316        t1 = tcg_temp_new_vec(type);
3317        tcg_gen_neg_vec(vece, t1, v2);
3318        if (opc == INDEX_op_shrv_vec) {
3319            opc = INDEX_op_arm_ushl_vec;
3320        } else {
3321            opc = INDEX_op_arm_sshl_vec;
3322        }
3323        vec_gen_3(opc, type, vece, tcgv_vec_arg(v0),
3324                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3325        tcg_temp_free_vec(t1);
3326        break;
3327
3328    case INDEX_op_rotli_vec:
3329        t1 = tcg_temp_new_vec(type);
3330        tcg_gen_shri_vec(vece, t1, v1, -a2 & ((8 << vece) - 1));
3331        vec_gen_4(INDEX_op_arm_sli_vec, type, vece,
3332                  tcgv_vec_arg(v0), tcgv_vec_arg(t1), tcgv_vec_arg(v1), a2);
3333        tcg_temp_free_vec(t1);
3334        break;
3335
3336    case INDEX_op_rotlv_vec:
3337        v2 = temp_tcgv_vec(arg_temp(a2));
3338        t1 = tcg_temp_new_vec(type);
3339        c1 = tcg_constant_vec(type, vece, 8 << vece);
3340        tcg_gen_sub_vec(vece, t1, v2, c1);
3341        /* Right shifts are negative left shifts for NEON.  */
3342        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t1),
3343                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3344        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(v0),
3345                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3346        tcg_gen_or_vec(vece, v0, v0, t1);
3347        tcg_temp_free_vec(t1);
3348        break;
3349
3350    case INDEX_op_rotrv_vec:
3351        v2 = temp_tcgv_vec(arg_temp(a2));
3352        t1 = tcg_temp_new_vec(type);
3353        t2 = tcg_temp_new_vec(type);
3354        c1 = tcg_constant_vec(type, vece, 8 << vece);
3355        tcg_gen_neg_vec(vece, t1, v2);
3356        tcg_gen_sub_vec(vece, t2, c1, v2);
3357        /* Right shifts are negative left shifts for NEON.  */
3358        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t1),
3359                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3360        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t2),
3361                  tcgv_vec_arg(v1), tcgv_vec_arg(t2));
3362        tcg_gen_or_vec(vece, v0, t1, t2);
3363        tcg_temp_free_vec(t1);
3364        tcg_temp_free_vec(t2);
3365        break;
3366
3367    default:
3368        g_assert_not_reached();
3369    }
3370}
3371
3372static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3373{
3374    int i;
3375    for (i = 0; i < count; ++i) {
3376        p[i] = INSN_NOP;
3377    }
3378}
3379
3380/* Compute frame size via macros, to share between tcg_target_qemu_prologue
3381   and tcg_register_jit.  */
3382
3383#define PUSH_SIZE  ((11 - 4 + 1 + 1) * sizeof(tcg_target_long))
3384
3385#define FRAME_SIZE \
3386    ((PUSH_SIZE \
3387      + TCG_STATIC_CALL_ARGS_SIZE \
3388      + CPU_TEMP_BUF_NLONGS * sizeof(long) \
3389      + TCG_TARGET_STACK_ALIGN - 1) \
3390     & -TCG_TARGET_STACK_ALIGN)
3391
3392#define STACK_ADDEND  (FRAME_SIZE - PUSH_SIZE)
3393
3394static void tcg_target_qemu_prologue(TCGContext *s)
3395{
3396    /* Calling convention requires us to save r4-r11 and lr.  */
3397    /* stmdb sp!, { r4 - r11, lr } */
3398    tcg_out_ldstm(s, COND_AL, INSN_STMDB, TCG_REG_CALL_STACK,
3399                  (1 << TCG_REG_R4) | (1 << TCG_REG_R5) | (1 << TCG_REG_R6) |
3400                  (1 << TCG_REG_R7) | (1 << TCG_REG_R8) | (1 << TCG_REG_R9) |
3401                  (1 << TCG_REG_R10) | (1 << TCG_REG_R11) | (1 << TCG_REG_R14));
3402
3403    /* Reserve callee argument and tcg temp space.  */
3404    tcg_out_dat_rI(s, COND_AL, ARITH_SUB, TCG_REG_CALL_STACK,
3405                   TCG_REG_CALL_STACK, STACK_ADDEND, 1);
3406    tcg_set_frame(s, TCG_REG_CALL_STACK, TCG_STATIC_CALL_ARGS_SIZE,
3407                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3408
3409    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3410
3411    if (!tcg_use_softmmu && guest_base) {
3412        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_GUEST_BASE, guest_base);
3413        tcg_regset_set_reg(s->reserved_regs, TCG_REG_GUEST_BASE);
3414    }
3415
3416    tcg_out_b_reg(s, COND_AL, tcg_target_call_iarg_regs[1]);
3417
3418    /*
3419     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3420     * and fall through to the rest of the epilogue.
3421     */
3422    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3423    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, 0);
3424    tcg_out_epilogue(s);
3425}
3426
3427static void tcg_out_epilogue(TCGContext *s)
3428{
3429    /* Release local stack frame.  */
3430    tcg_out_dat_rI(s, COND_AL, ARITH_ADD, TCG_REG_CALL_STACK,
3431                   TCG_REG_CALL_STACK, STACK_ADDEND, 1);
3432
3433    /* ldmia sp!, { r4 - r11, pc } */
3434    tcg_out_ldstm(s, COND_AL, INSN_LDMIA, TCG_REG_CALL_STACK,
3435                  (1 << TCG_REG_R4) | (1 << TCG_REG_R5) | (1 << TCG_REG_R6) |
3436                  (1 << TCG_REG_R7) | (1 << TCG_REG_R8) | (1 << TCG_REG_R9) |
3437                  (1 << TCG_REG_R10) | (1 << TCG_REG_R11) | (1 << TCG_REG_PC));
3438}
3439
3440static void tcg_out_tb_start(TCGContext *s)
3441{
3442    /* nothing to do */
3443}
3444
3445typedef struct {
3446    DebugFrameHeader h;
3447    uint8_t fde_def_cfa[4];
3448    uint8_t fde_reg_ofs[18];
3449} DebugFrame;
3450
3451#define ELF_HOST_MACHINE EM_ARM
3452
3453/* We're expecting a 2 byte uleb128 encoded value.  */
3454QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3455
3456static const DebugFrame debug_frame = {
3457    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3458    .h.cie.id = -1,
3459    .h.cie.version = 1,
3460    .h.cie.code_align = 1,
3461    .h.cie.data_align = 0x7c,             /* sleb128 -4 */
3462    .h.cie.return_column = 14,
3463
3464    /* Total FDE size does not include the "len" member.  */
3465    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3466
3467    .fde_def_cfa = {
3468        12, 13,                         /* DW_CFA_def_cfa sp, ... */
3469        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3470        (FRAME_SIZE >> 7)
3471    },
3472    .fde_reg_ofs = {
3473        /* The following must match the stmdb in the prologue.  */
3474        0x8e, 1,                        /* DW_CFA_offset, lr, -4 */
3475        0x8b, 2,                        /* DW_CFA_offset, r11, -8 */
3476        0x8a, 3,                        /* DW_CFA_offset, r10, -12 */
3477        0x89, 4,                        /* DW_CFA_offset, r9, -16 */
3478        0x88, 5,                        /* DW_CFA_offset, r8, -20 */
3479        0x87, 6,                        /* DW_CFA_offset, r7, -24 */
3480        0x86, 7,                        /* DW_CFA_offset, r6, -28 */
3481        0x85, 8,                        /* DW_CFA_offset, r5, -32 */
3482        0x84, 9,                        /* DW_CFA_offset, r4, -36 */
3483    }
3484};
3485
3486void tcg_register_jit(const void *buf, size_t buf_size)
3487{
3488    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3489}
3490