xref: /openbmc/qemu/tcg/arm/tcg-target.c.inc (revision 3ad5d4ccb4bdebdff4e90957bb2b8a93e5e418e2)
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    /* Architected nop introduced in v6k.  */
182    /* ??? This is an MSR (imm) 0,0,0 insn.  Anyone know if this
183       also Just So Happened to do nothing on pre-v6k so that we
184       don't need to conditionalize it?  */
185    INSN_NOP_v6k   = 0xe320f000,
186    /* Otherwise the assembler uses mov r0,r0 */
187    INSN_NOP_v4    = (COND_AL << 28) | ARITH_MOV,
188
189    INSN_VADD      = 0xf2000800,
190    INSN_VAND      = 0xf2000110,
191    INSN_VBIC      = 0xf2100110,
192    INSN_VEOR      = 0xf3000110,
193    INSN_VORN      = 0xf2300110,
194    INSN_VORR      = 0xf2200110,
195    INSN_VSUB      = 0xf3000800,
196    INSN_VMUL      = 0xf2000910,
197    INSN_VQADD     = 0xf2000010,
198    INSN_VQADD_U   = 0xf3000010,
199    INSN_VQSUB     = 0xf2000210,
200    INSN_VQSUB_U   = 0xf3000210,
201    INSN_VMAX      = 0xf2000600,
202    INSN_VMAX_U    = 0xf3000600,
203    INSN_VMIN      = 0xf2000610,
204    INSN_VMIN_U    = 0xf3000610,
205
206    INSN_VABS      = 0xf3b10300,
207    INSN_VMVN      = 0xf3b00580,
208    INSN_VNEG      = 0xf3b10380,
209
210    INSN_VCEQ0     = 0xf3b10100,
211    INSN_VCGT0     = 0xf3b10000,
212    INSN_VCGE0     = 0xf3b10080,
213    INSN_VCLE0     = 0xf3b10180,
214    INSN_VCLT0     = 0xf3b10200,
215
216    INSN_VCEQ      = 0xf3000810,
217    INSN_VCGE      = 0xf2000310,
218    INSN_VCGT      = 0xf2000300,
219    INSN_VCGE_U    = 0xf3000310,
220    INSN_VCGT_U    = 0xf3000300,
221
222    INSN_VSHLI     = 0xf2800510,  /* VSHL (immediate) */
223    INSN_VSARI     = 0xf2800010,  /* VSHR.S */
224    INSN_VSHRI     = 0xf3800010,  /* VSHR.U */
225    INSN_VSLI      = 0xf3800510,
226    INSN_VSHL_S    = 0xf2000400,  /* VSHL.S (register) */
227    INSN_VSHL_U    = 0xf3000400,  /* VSHL.U (register) */
228
229    INSN_VBSL      = 0xf3100110,
230    INSN_VBIT      = 0xf3200110,
231    INSN_VBIF      = 0xf3300110,
232
233    INSN_VTST      = 0xf2000810,
234
235    INSN_VDUP_G    = 0xee800b10,  /* VDUP (ARM core register) */
236    INSN_VDUP_S    = 0xf3b00c00,  /* VDUP (scalar) */
237    INSN_VLDR_D    = 0xed100b00,  /* VLDR.64 */
238    INSN_VLD1      = 0xf4200000,  /* VLD1 (multiple single elements) */
239    INSN_VLD1R     = 0xf4a00c00,  /* VLD1 (single element to all lanes) */
240    INSN_VST1      = 0xf4000000,  /* VST1 (multiple single elements) */
241    INSN_VMOVI     = 0xf2800010,  /* VMOV (immediate) */
242} ARMInsn;
243
244#define INSN_NOP   (use_armv7_instructions ? INSN_NOP_v6k : INSN_NOP_v4)
245
246static const uint8_t tcg_cond_to_arm_cond[] = {
247    [TCG_COND_EQ] = COND_EQ,
248    [TCG_COND_NE] = COND_NE,
249    [TCG_COND_LT] = COND_LT,
250    [TCG_COND_GE] = COND_GE,
251    [TCG_COND_LE] = COND_LE,
252    [TCG_COND_GT] = COND_GT,
253    /* unsigned */
254    [TCG_COND_LTU] = COND_CC,
255    [TCG_COND_GEU] = COND_CS,
256    [TCG_COND_LEU] = COND_LS,
257    [TCG_COND_GTU] = COND_HI,
258};
259
260static int encode_imm(uint32_t imm);
261
262/* TCG private relocation type: add with pc+imm8 */
263#define R_ARM_PC8  11
264
265/* TCG private relocation type: vldr with imm8 << 2 */
266#define R_ARM_PC11 12
267
268static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
269{
270    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
271    ptrdiff_t offset = (tcg_ptr_byte_diff(target, src_rx) - 8) >> 2;
272
273    if (offset == sextract32(offset, 0, 24)) {
274        *src_rw = deposit32(*src_rw, 0, 24, offset);
275        return true;
276    }
277    return false;
278}
279
280static bool reloc_pc13(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
281{
282    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
283    ptrdiff_t offset = tcg_ptr_byte_diff(target, src_rx) - 8;
284
285    if (offset >= -0xfff && offset <= 0xfff) {
286        tcg_insn_unit insn = *src_rw;
287        bool u = (offset >= 0);
288        if (!u) {
289            offset = -offset;
290        }
291        insn = deposit32(insn, 23, 1, u);
292        insn = deposit32(insn, 0, 12, offset);
293        *src_rw = insn;
294        return true;
295    }
296    return false;
297}
298
299static bool reloc_pc11(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
300{
301    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
302    ptrdiff_t offset = (tcg_ptr_byte_diff(target, src_rx) - 8) / 4;
303
304    if (offset >= -0xff && offset <= 0xff) {
305        tcg_insn_unit insn = *src_rw;
306        bool u = (offset >= 0);
307        if (!u) {
308            offset = -offset;
309        }
310        insn = deposit32(insn, 23, 1, u);
311        insn = deposit32(insn, 0, 8, offset);
312        *src_rw = insn;
313        return true;
314    }
315    return false;
316}
317
318static bool reloc_pc8(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
319{
320    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
321    ptrdiff_t offset = tcg_ptr_byte_diff(target, src_rx) - 8;
322    int imm12 = encode_imm(offset);
323
324    if (imm12 >= 0) {
325        *src_rw = deposit32(*src_rw, 0, 12, imm12);
326        return true;
327    }
328    return false;
329}
330
331static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
332                        intptr_t value, intptr_t addend)
333{
334    tcg_debug_assert(addend == 0);
335    switch (type) {
336    case R_ARM_PC24:
337        return reloc_pc24(code_ptr, (const tcg_insn_unit *)value);
338    case R_ARM_PC13:
339        return reloc_pc13(code_ptr, (const tcg_insn_unit *)value);
340    case R_ARM_PC11:
341        return reloc_pc11(code_ptr, (const tcg_insn_unit *)value);
342    case R_ARM_PC8:
343        return reloc_pc8(code_ptr, (const tcg_insn_unit *)value);
344    default:
345        g_assert_not_reached();
346    }
347}
348
349#define TCG_CT_CONST_ARM  0x100
350#define TCG_CT_CONST_INV  0x200
351#define TCG_CT_CONST_NEG  0x400
352#define TCG_CT_CONST_ZERO 0x800
353#define TCG_CT_CONST_ORRI 0x1000
354#define TCG_CT_CONST_ANDI 0x2000
355
356#define ALL_GENERAL_REGS  0xffffu
357#define ALL_VECTOR_REGS   0xffff0000u
358
359/*
360 * r0-r3 will be overwritten when reading the tlb entry (system-mode only);
361 * r14 will be overwritten by the BLNE branching to the slow path.
362 */
363#define ALL_QLDST_REGS \
364    (ALL_GENERAL_REGS & ~((tcg_use_softmmu ? 0xf : 0) | (1 << TCG_REG_R14)))
365
366/*
367 * ARM immediates for ALU instructions are made of an unsigned 8-bit
368 * right-rotated by an even amount between 0 and 30.
369 *
370 * Return < 0 if @imm cannot be encoded, else the entire imm12 field.
371 */
372static int encode_imm(uint32_t imm)
373{
374    uint32_t rot, imm8;
375
376    /* Simple case, no rotation required. */
377    if ((imm & ~0xff) == 0) {
378        return imm;
379    }
380
381    /* Next, try a simple even shift.  */
382    rot = ctz32(imm) & ~1;
383    imm8 = imm >> rot;
384    rot = 32 - rot;
385    if ((imm8 & ~0xff) == 0) {
386        goto found;
387    }
388
389    /*
390     * Finally, try harder with rotations.
391     * The ctz test above will have taken care of rotates >= 8.
392     */
393    for (rot = 2; rot < 8; rot += 2) {
394        imm8 = rol32(imm, rot);
395        if ((imm8 & ~0xff) == 0) {
396            goto found;
397        }
398    }
399    /* Fail: imm cannot be encoded. */
400    return -1;
401
402 found:
403    /* Note that rot is even, and we discard bit 0 by shifting by 7. */
404    return rot << 7 | imm8;
405}
406
407static int encode_imm_nofail(uint32_t imm)
408{
409    int ret = encode_imm(imm);
410    tcg_debug_assert(ret >= 0);
411    return ret;
412}
413
414static bool check_fit_imm(uint32_t imm)
415{
416    return encode_imm(imm) >= 0;
417}
418
419/* Return true if v16 is a valid 16-bit shifted immediate.  */
420static bool is_shimm16(uint16_t v16, int *cmode, int *imm8)
421{
422    if (v16 == (v16 & 0xff)) {
423        *cmode = 0x8;
424        *imm8 = v16 & 0xff;
425        return true;
426    } else if (v16 == (v16 & 0xff00)) {
427        *cmode = 0xa;
428        *imm8 = v16 >> 8;
429        return true;
430    }
431    return false;
432}
433
434/* Return true if v32 is a valid 32-bit shifted immediate.  */
435static bool is_shimm32(uint32_t v32, int *cmode, int *imm8)
436{
437    if (v32 == (v32 & 0xff)) {
438        *cmode = 0x0;
439        *imm8 = v32 & 0xff;
440        return true;
441    } else if (v32 == (v32 & 0xff00)) {
442        *cmode = 0x2;
443        *imm8 = (v32 >> 8) & 0xff;
444        return true;
445    } else if (v32 == (v32 & 0xff0000)) {
446        *cmode = 0x4;
447        *imm8 = (v32 >> 16) & 0xff;
448        return true;
449    } else if (v32 == (v32 & 0xff000000)) {
450        *cmode = 0x6;
451        *imm8 = v32 >> 24;
452        return true;
453    }
454    return false;
455}
456
457/* Return true if v32 is a valid 32-bit shifting ones immediate.  */
458static bool is_soimm32(uint32_t v32, int *cmode, int *imm8)
459{
460    if ((v32 & 0xffff00ff) == 0xff) {
461        *cmode = 0xc;
462        *imm8 = (v32 >> 8) & 0xff;
463        return true;
464    } else if ((v32 & 0xff00ffff) == 0xffff) {
465        *cmode = 0xd;
466        *imm8 = (v32 >> 16) & 0xff;
467        return true;
468    }
469    return false;
470}
471
472/*
473 * Return non-zero if v32 can be formed by MOVI+ORR.
474 * Place the parameters for MOVI in (cmode, imm8).
475 * Return the cmode for ORR; the imm8 can be had via extraction from v32.
476 */
477static int is_shimm32_pair(uint32_t v32, int *cmode, int *imm8)
478{
479    int i;
480
481    for (i = 6; i > 0; i -= 2) {
482        /* Mask out one byte we can add with ORR.  */
483        uint32_t tmp = v32 & ~(0xffu << (i * 4));
484        if (is_shimm32(tmp, cmode, imm8) ||
485            is_soimm32(tmp, cmode, imm8)) {
486            break;
487        }
488    }
489    return i;
490}
491
492/* Return true if V is a valid 16-bit or 32-bit shifted immediate.  */
493static bool is_shimm1632(uint32_t v32, int *cmode, int *imm8)
494{
495    if (v32 == deposit32(v32, 16, 16, v32)) {
496        return is_shimm16(v32, cmode, imm8);
497    } else {
498        return is_shimm32(v32, cmode, imm8);
499    }
500}
501
502/* Test if a constant matches the constraint.
503 * TODO: define constraints for:
504 *
505 * ldr/str offset:   between -0xfff and 0xfff
506 * ldrh/strh offset: between -0xff and 0xff
507 * mov operand2:     values represented with x << (2 * y), x < 0x100
508 * add, sub, eor...: ditto
509 */
510static bool tcg_target_const_match(int64_t val, int ct,
511                                   TCGType type, TCGCond cond, int vece)
512{
513    if (ct & TCG_CT_CONST) {
514        return 1;
515    } else if ((ct & TCG_CT_CONST_ARM) && check_fit_imm(val)) {
516        return 1;
517    } else if ((ct & TCG_CT_CONST_INV) && check_fit_imm(~val)) {
518        return 1;
519    } else if ((ct & TCG_CT_CONST_NEG) && check_fit_imm(-val)) {
520        return 1;
521    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
522        return 1;
523    }
524
525    switch (ct & (TCG_CT_CONST_ORRI | TCG_CT_CONST_ANDI)) {
526    case 0:
527        break;
528    case TCG_CT_CONST_ANDI:
529        val = ~val;
530        /* fallthru */
531    case TCG_CT_CONST_ORRI:
532        if (val == deposit64(val, 32, 32, val)) {
533            int cmode, imm8;
534            return is_shimm1632(val, &cmode, &imm8);
535        }
536        break;
537    default:
538        /* Both bits should not be set for the same insn.  */
539        g_assert_not_reached();
540    }
541
542    return 0;
543}
544
545static void tcg_out_b_imm(TCGContext *s, ARMCond cond, int32_t offset)
546{
547    tcg_out32(s, (cond << 28) | INSN_B |
548                    (((offset - 8) >> 2) & 0x00ffffff));
549}
550
551static void tcg_out_bl_imm(TCGContext *s, ARMCond cond, int32_t offset)
552{
553    tcg_out32(s, (cond << 28) | 0x0b000000 |
554                    (((offset - 8) >> 2) & 0x00ffffff));
555}
556
557static void tcg_out_blx_reg(TCGContext *s, ARMCond cond, TCGReg rn)
558{
559    tcg_out32(s, (cond << 28) | 0x012fff30 | rn);
560}
561
562static void tcg_out_blx_imm(TCGContext *s, int32_t offset)
563{
564    tcg_out32(s, 0xfa000000 | ((offset & 2) << 23) |
565                (((offset - 8) >> 2) & 0x00ffffff));
566}
567
568static void tcg_out_dat_reg(TCGContext *s, ARMCond cond, ARMInsn opc,
569                            TCGReg rd, TCGReg rn, TCGReg rm, int shift)
570{
571    tcg_out32(s, (cond << 28) | (0 << 25) | opc |
572                    (rn << 16) | (rd << 12) | shift | rm);
573}
574
575static void tcg_out_mov_reg(TCGContext *s, ARMCond cond, TCGReg rd, TCGReg rm)
576{
577    /* Simple reg-reg move, optimising out the 'do nothing' case */
578    if (rd != rm) {
579        tcg_out_dat_reg(s, cond, ARITH_MOV, rd, 0, rm, SHIFT_IMM_LSL(0));
580    }
581}
582
583static void tcg_out_bx_reg(TCGContext *s, ARMCond cond, TCGReg rn)
584{
585    tcg_out32(s, (cond << 28) | 0x012fff10 | rn);
586}
587
588static void tcg_out_b_reg(TCGContext *s, ARMCond cond, TCGReg rn)
589{
590    /*
591     * Unless the C portion of QEMU is compiled as thumb, we don't need
592     * true BX semantics; merely a branch to an address held in a register.
593     */
594    tcg_out_bx_reg(s, cond, rn);
595}
596
597static void tcg_out_dat_imm(TCGContext *s, ARMCond cond, ARMInsn opc,
598                            TCGReg rd, TCGReg rn, int im)
599{
600    tcg_out32(s, (cond << 28) | (1 << 25) | opc |
601                    (rn << 16) | (rd << 12) | im);
602}
603
604static void tcg_out_ldstm(TCGContext *s, ARMCond cond, ARMInsn opc,
605                          TCGReg rn, uint16_t mask)
606{
607    tcg_out32(s, (cond << 28) | opc | (rn << 16) | mask);
608}
609
610/* Note that this routine is used for both LDR and LDRH formats, so we do
611   not wish to include an immediate shift at this point.  */
612static void tcg_out_memop_r(TCGContext *s, ARMCond cond, ARMInsn opc, TCGReg rt,
613                            TCGReg rn, TCGReg rm, bool u, bool p, bool w)
614{
615    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24)
616              | (w << 21) | (rn << 16) | (rt << 12) | rm);
617}
618
619static void tcg_out_memop_8(TCGContext *s, ARMCond cond, ARMInsn opc, TCGReg rt,
620                            TCGReg rn, int imm8, bool p, bool w)
621{
622    bool u = 1;
623    if (imm8 < 0) {
624        imm8 = -imm8;
625        u = 0;
626    }
627    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
628              (rn << 16) | (rt << 12) | ((imm8 & 0xf0) << 4) | (imm8 & 0xf));
629}
630
631static void tcg_out_memop_12(TCGContext *s, ARMCond cond, ARMInsn opc,
632                             TCGReg rt, TCGReg rn, int imm12, bool p, bool w)
633{
634    bool u = 1;
635    if (imm12 < 0) {
636        imm12 = -imm12;
637        u = 0;
638    }
639    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
640              (rn << 16) | (rt << 12) | imm12);
641}
642
643static void tcg_out_ld32_12(TCGContext *s, ARMCond cond, TCGReg rt,
644                            TCGReg rn, int imm12)
645{
646    tcg_out_memop_12(s, cond, INSN_LDR_IMM, rt, rn, imm12, 1, 0);
647}
648
649static void tcg_out_st32_12(TCGContext *s, ARMCond cond, TCGReg rt,
650                            TCGReg rn, int imm12)
651{
652    tcg_out_memop_12(s, cond, INSN_STR_IMM, rt, rn, imm12, 1, 0);
653}
654
655static void tcg_out_ld32_r(TCGContext *s, ARMCond cond, TCGReg rt,
656                           TCGReg rn, TCGReg rm)
657{
658    tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 0);
659}
660
661static void tcg_out_st32_r(TCGContext *s, ARMCond cond, TCGReg rt,
662                           TCGReg rn, TCGReg rm)
663{
664    tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 0);
665}
666
667static void tcg_out_ldrd_8(TCGContext *s, ARMCond cond, TCGReg rt,
668                           TCGReg rn, int imm8)
669{
670    tcg_out_memop_8(s, cond, INSN_LDRD_IMM, rt, rn, imm8, 1, 0);
671}
672
673static void tcg_out_ldrd_r(TCGContext *s, ARMCond cond, TCGReg rt,
674                           TCGReg rn, TCGReg rm)
675{
676    tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 0);
677}
678
679static void tcg_out_strd_8(TCGContext *s, ARMCond cond, TCGReg rt,
680                           TCGReg rn, int imm8)
681{
682    tcg_out_memop_8(s, cond, INSN_STRD_IMM, rt, rn, imm8, 1, 0);
683}
684
685static void tcg_out_strd_r(TCGContext *s, ARMCond cond, TCGReg rt,
686                           TCGReg rn, TCGReg rm)
687{
688    tcg_out_memop_r(s, cond, INSN_STRD_REG, rt, rn, rm, 1, 1, 0);
689}
690
691/* Register pre-increment with base writeback.  */
692static void tcg_out_ld32_rwb(TCGContext *s, ARMCond cond, TCGReg rt,
693                             TCGReg rn, TCGReg rm)
694{
695    tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 1);
696}
697
698static void tcg_out_st32_rwb(TCGContext *s, ARMCond cond, TCGReg rt,
699                             TCGReg rn, TCGReg rm)
700{
701    tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 1);
702}
703
704static void tcg_out_ld16u_8(TCGContext *s, ARMCond cond, TCGReg rt,
705                            TCGReg rn, int imm8)
706{
707    tcg_out_memop_8(s, cond, INSN_LDRH_IMM, rt, rn, imm8, 1, 0);
708}
709
710static void tcg_out_st16_8(TCGContext *s, ARMCond cond, TCGReg rt,
711                           TCGReg rn, int imm8)
712{
713    tcg_out_memop_8(s, cond, INSN_STRH_IMM, rt, rn, imm8, 1, 0);
714}
715
716static void tcg_out_ld16u_r(TCGContext *s, ARMCond cond, TCGReg rt,
717                            TCGReg rn, TCGReg rm)
718{
719    tcg_out_memop_r(s, cond, INSN_LDRH_REG, rt, rn, rm, 1, 1, 0);
720}
721
722static void tcg_out_st16_r(TCGContext *s, ARMCond cond, TCGReg rt,
723                           TCGReg rn, TCGReg rm)
724{
725    tcg_out_memop_r(s, cond, INSN_STRH_REG, rt, rn, rm, 1, 1, 0);
726}
727
728static void tcg_out_ld16s_8(TCGContext *s, ARMCond cond, TCGReg rt,
729                            TCGReg rn, int imm8)
730{
731    tcg_out_memop_8(s, cond, INSN_LDRSH_IMM, rt, rn, imm8, 1, 0);
732}
733
734static void tcg_out_ld16s_r(TCGContext *s, ARMCond cond, TCGReg rt,
735                            TCGReg rn, TCGReg rm)
736{
737    tcg_out_memop_r(s, cond, INSN_LDRSH_REG, rt, rn, rm, 1, 1, 0);
738}
739
740static void tcg_out_ld8_12(TCGContext *s, ARMCond cond, TCGReg rt,
741                           TCGReg rn, int imm12)
742{
743    tcg_out_memop_12(s, cond, INSN_LDRB_IMM, rt, rn, imm12, 1, 0);
744}
745
746static void tcg_out_st8_12(TCGContext *s, ARMCond cond, TCGReg rt,
747                           TCGReg rn, int imm12)
748{
749    tcg_out_memop_12(s, cond, INSN_STRB_IMM, rt, rn, imm12, 1, 0);
750}
751
752static void tcg_out_ld8_r(TCGContext *s, ARMCond cond, TCGReg rt,
753                          TCGReg rn, TCGReg rm)
754{
755    tcg_out_memop_r(s, cond, INSN_LDRB_REG, rt, rn, rm, 1, 1, 0);
756}
757
758static void tcg_out_st8_r(TCGContext *s, ARMCond cond, TCGReg rt,
759                          TCGReg rn, TCGReg rm)
760{
761    tcg_out_memop_r(s, cond, INSN_STRB_REG, rt, rn, rm, 1, 1, 0);
762}
763
764static void tcg_out_ld8s_8(TCGContext *s, ARMCond cond, TCGReg rt,
765                           TCGReg rn, int imm8)
766{
767    tcg_out_memop_8(s, cond, INSN_LDRSB_IMM, rt, rn, imm8, 1, 0);
768}
769
770static void tcg_out_ld8s_r(TCGContext *s, ARMCond cond, TCGReg rt,
771                           TCGReg rn, TCGReg rm)
772{
773    tcg_out_memop_r(s, cond, INSN_LDRSB_REG, rt, rn, rm, 1, 1, 0);
774}
775
776static void tcg_out_movi_pool(TCGContext *s, ARMCond cond,
777                              TCGReg rd, uint32_t arg)
778{
779    new_pool_label(s, arg, R_ARM_PC13, s->code_ptr, 0);
780    tcg_out_ld32_12(s, cond, rd, TCG_REG_PC, 0);
781}
782
783static void tcg_out_movi32(TCGContext *s, ARMCond cond,
784                           TCGReg rd, uint32_t arg)
785{
786    int imm12, diff, opc, sh1, sh2;
787    uint32_t tt0, tt1, tt2;
788
789    /* Check a single MOV/MVN before anything else.  */
790    imm12 = encode_imm(arg);
791    if (imm12 >= 0) {
792        tcg_out_dat_imm(s, cond, ARITH_MOV, rd, 0, imm12);
793        return;
794    }
795    imm12 = encode_imm(~arg);
796    if (imm12 >= 0) {
797        tcg_out_dat_imm(s, cond, ARITH_MVN, rd, 0, imm12);
798        return;
799    }
800
801    /* Check for a pc-relative address.  This will usually be the TB,
802       or within the TB, which is immediately before the code block.  */
803    diff = tcg_pcrel_diff(s, (void *)arg) - 8;
804    if (diff >= 0) {
805        imm12 = encode_imm(diff);
806        if (imm12 >= 0) {
807            tcg_out_dat_imm(s, cond, ARITH_ADD, rd, TCG_REG_PC, imm12);
808            return;
809        }
810    } else {
811        imm12 = encode_imm(-diff);
812        if (imm12 >= 0) {
813            tcg_out_dat_imm(s, cond, ARITH_SUB, rd, TCG_REG_PC, imm12);
814            return;
815        }
816    }
817
818    /* Use movw + movt.  */
819    if (use_armv7_instructions) {
820        /* movw */
821        tcg_out32(s, (cond << 28) | 0x03000000 | (rd << 12)
822                  | ((arg << 4) & 0x000f0000) | (arg & 0xfff));
823        if (arg & 0xffff0000) {
824            /* movt */
825            tcg_out32(s, (cond << 28) | 0x03400000 | (rd << 12)
826                      | ((arg >> 12) & 0x000f0000) | ((arg >> 16) & 0xfff));
827        }
828        return;
829    }
830
831    /* Look for sequences of two insns.  If we have lots of 1's, we can
832       shorten the sequence by beginning with mvn and then clearing
833       higher bits with eor.  */
834    tt0 = arg;
835    opc = ARITH_MOV;
836    if (ctpop32(arg) > 16) {
837        tt0 = ~arg;
838        opc = ARITH_MVN;
839    }
840    sh1 = ctz32(tt0) & ~1;
841    tt1 = tt0 & ~(0xff << sh1);
842    sh2 = ctz32(tt1) & ~1;
843    tt2 = tt1 & ~(0xff << sh2);
844    if (tt2 == 0) {
845        int rot;
846
847        rot = ((32 - sh1) << 7) & 0xf00;
848        tcg_out_dat_imm(s, cond, opc, rd,  0, ((tt0 >> sh1) & 0xff) | rot);
849        rot = ((32 - sh2) << 7) & 0xf00;
850        tcg_out_dat_imm(s, cond, ARITH_EOR, rd, rd,
851                        ((tt0 >> sh2) & 0xff) | rot);
852        return;
853    }
854
855    /* Otherwise, drop it into the constant pool.  */
856    tcg_out_movi_pool(s, cond, rd, arg);
857}
858
859/*
860 * Emit either the reg,imm or reg,reg form of a data-processing insn.
861 * rhs must satisfy the "rI" constraint.
862 */
863static void tcg_out_dat_rI(TCGContext *s, ARMCond cond, ARMInsn opc,
864                           TCGReg dst, TCGReg lhs, TCGArg rhs, int rhs_is_const)
865{
866    if (rhs_is_const) {
867        tcg_out_dat_imm(s, cond, opc, dst, lhs, encode_imm_nofail(rhs));
868    } else {
869        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
870    }
871}
872
873/*
874 * Emit either the reg,imm or reg,reg form of a data-processing insn.
875 * rhs must satisfy the "rIK" constraint.
876 */
877static void tcg_out_dat_IK(TCGContext *s, ARMCond cond, ARMInsn opc,
878                            ARMInsn opinv, TCGReg dst, TCGReg lhs, TCGArg rhs)
879{
880    int imm12 = encode_imm(rhs);
881    if (imm12 < 0) {
882        imm12 = encode_imm_nofail(~rhs);
883        opc = opinv;
884    }
885    tcg_out_dat_imm(s, cond, opc, dst, lhs, imm12);
886}
887
888static void tcg_out_dat_rIK(TCGContext *s, ARMCond cond, ARMInsn opc,
889                            ARMInsn opinv, TCGReg dst, TCGReg lhs, TCGArg rhs,
890                            bool rhs_is_const)
891{
892    if (rhs_is_const) {
893        tcg_out_dat_IK(s, cond, opc, opinv, dst, lhs, rhs);
894    } else {
895        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
896    }
897}
898
899static void tcg_out_dat_IN(TCGContext *s, ARMCond cond, ARMInsn opc,
900                           ARMInsn opneg, TCGReg dst, TCGReg lhs, TCGArg rhs)
901{
902    int imm12 = encode_imm(rhs);
903    if (imm12 < 0) {
904        imm12 = encode_imm_nofail(-rhs);
905        opc = opneg;
906    }
907    tcg_out_dat_imm(s, cond, opc, dst, lhs, imm12);
908}
909
910static void tcg_out_dat_rIN(TCGContext *s, ARMCond cond, ARMInsn opc,
911                            ARMInsn opneg, TCGReg dst, TCGReg lhs, TCGArg rhs,
912                            bool rhs_is_const)
913{
914    /* Emit either the reg,imm or reg,reg form of a data-processing insn.
915     * rhs must satisfy the "rIN" constraint.
916     */
917    if (rhs_is_const) {
918        tcg_out_dat_IN(s, cond, opc, opneg, dst, lhs, rhs);
919    } else {
920        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
921    }
922}
923
924static void tcg_out_ext8s(TCGContext *s, TCGType t, TCGReg rd, TCGReg rn)
925{
926    /* sxtb */
927    tcg_out32(s, 0x06af0070 | (COND_AL << 28) | (rd << 12) | rn);
928}
929
930static void tcg_out_ext8u(TCGContext *s, TCGReg rd, TCGReg rn)
931{
932    tcg_out_dat_imm(s, COND_AL, ARITH_AND, rd, rn, 0xff);
933}
934
935static void tcg_out_ext16s(TCGContext *s, TCGType t, TCGReg rd, TCGReg rn)
936{
937    /* sxth */
938    tcg_out32(s, 0x06bf0070 | (COND_AL << 28) | (rd << 12) | rn);
939}
940
941static void tcg_out_ext16u(TCGContext *s, TCGReg rd, TCGReg rn)
942{
943    /* uxth */
944    tcg_out32(s, 0x06ff0070 | (COND_AL << 28) | (rd << 12) | rn);
945}
946
947static void tcg_out_ext32s(TCGContext *s, TCGReg rd, TCGReg rn)
948{
949    g_assert_not_reached();
950}
951
952static void tcg_out_ext32u(TCGContext *s, TCGReg rd, TCGReg rn)
953{
954    g_assert_not_reached();
955}
956
957static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg rd, TCGReg rn)
958{
959    g_assert_not_reached();
960}
961
962static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg rd, TCGReg rn)
963{
964    g_assert_not_reached();
965}
966
967static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rn)
968{
969    g_assert_not_reached();
970}
971
972static void tcg_out_deposit(TCGContext *s, ARMCond cond, TCGReg rd,
973                            TCGArg a1, int ofs, int len, bool const_a1)
974{
975    if (const_a1) {
976        /* bfi becomes bfc with rn == 15.  */
977        a1 = 15;
978    }
979    /* bfi/bfc */
980    tcg_out32(s, 0x07c00010 | (cond << 28) | (rd << 12) | a1
981              | (ofs << 7) | ((ofs + len - 1) << 16));
982}
983
984static void tcg_out_extract(TCGContext *s, ARMCond cond, TCGReg rd,
985                            TCGReg rn, int ofs, int len)
986{
987    /* According to gcc, AND can be faster. */
988    if (ofs == 0 && len <= 8) {
989        tcg_out_dat_imm(s, cond, ARITH_AND, rd, rn,
990                        encode_imm_nofail((1 << len) - 1));
991        return;
992    }
993
994    if (use_armv7_instructions) {
995        /* ubfx */
996        tcg_out32(s, 0x07e00050 | (cond << 28) | (rd << 12) | rn
997                  | (ofs << 7) | ((len - 1) << 16));
998        return;
999    }
1000
1001    assert(ofs % 8 == 0);
1002    switch (len) {
1003    case 8:
1004        /* uxtb */
1005        tcg_out32(s, 0x06ef0070 | (cond << 28) | (rd << 12) | (ofs << 7) | rn);
1006        break;
1007    case 16:
1008        /* uxth */
1009        tcg_out32(s, 0x06ff0070 | (cond << 28) | (rd << 12) | (ofs << 7) | rn);
1010        break;
1011    default:
1012        g_assert_not_reached();
1013    }
1014}
1015
1016static void tcg_out_sextract(TCGContext *s, ARMCond cond, TCGReg rd,
1017                             TCGReg rn, int ofs, int len)
1018{
1019    if (use_armv7_instructions) {
1020        /* sbfx */
1021        tcg_out32(s, 0x07a00050 | (cond << 28) | (rd << 12) | rn
1022                  | (ofs << 7) | ((len - 1) << 16));
1023        return;
1024    }
1025
1026    assert(ofs % 8 == 0);
1027    switch (len) {
1028    case 8:
1029        /* sxtb */
1030        tcg_out32(s, 0x06af0070 | (cond << 28) | (rd << 12) | (ofs << 7) | rn);
1031        break;
1032    case 16:
1033        /* sxth */
1034        tcg_out32(s, 0x06bf0070 | (cond << 28) | (rd << 12) | (ofs << 7) | rn);
1035        break;
1036    default:
1037        g_assert_not_reached();
1038    }
1039}
1040
1041
1042static void tcg_out_ld32u(TCGContext *s, ARMCond cond,
1043                          TCGReg rd, TCGReg rn, int32_t offset)
1044{
1045    if (offset > 0xfff || offset < -0xfff) {
1046        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1047        tcg_out_ld32_r(s, cond, rd, rn, TCG_REG_TMP);
1048    } else
1049        tcg_out_ld32_12(s, cond, rd, rn, offset);
1050}
1051
1052static void tcg_out_st32(TCGContext *s, ARMCond cond,
1053                         TCGReg rd, TCGReg rn, int32_t offset)
1054{
1055    if (offset > 0xfff || offset < -0xfff) {
1056        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1057        tcg_out_st32_r(s, cond, rd, rn, TCG_REG_TMP);
1058    } else
1059        tcg_out_st32_12(s, cond, rd, rn, offset);
1060}
1061
1062static void tcg_out_ld16u(TCGContext *s, ARMCond cond,
1063                          TCGReg rd, TCGReg rn, int32_t offset)
1064{
1065    if (offset > 0xff || offset < -0xff) {
1066        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1067        tcg_out_ld16u_r(s, cond, rd, rn, TCG_REG_TMP);
1068    } else
1069        tcg_out_ld16u_8(s, cond, rd, rn, offset);
1070}
1071
1072static void tcg_out_ld16s(TCGContext *s, ARMCond cond,
1073                          TCGReg rd, TCGReg rn, int32_t offset)
1074{
1075    if (offset > 0xff || offset < -0xff) {
1076        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1077        tcg_out_ld16s_r(s, cond, rd, rn, TCG_REG_TMP);
1078    } else
1079        tcg_out_ld16s_8(s, cond, rd, rn, offset);
1080}
1081
1082static void tcg_out_st16(TCGContext *s, ARMCond cond,
1083                         TCGReg rd, TCGReg rn, int32_t offset)
1084{
1085    if (offset > 0xff || offset < -0xff) {
1086        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1087        tcg_out_st16_r(s, cond, rd, rn, TCG_REG_TMP);
1088    } else
1089        tcg_out_st16_8(s, cond, rd, rn, offset);
1090}
1091
1092static void tcg_out_ld8u(TCGContext *s, ARMCond cond,
1093                         TCGReg rd, TCGReg rn, int32_t offset)
1094{
1095    if (offset > 0xfff || offset < -0xfff) {
1096        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1097        tcg_out_ld8_r(s, cond, rd, rn, TCG_REG_TMP);
1098    } else
1099        tcg_out_ld8_12(s, cond, rd, rn, offset);
1100}
1101
1102static void tcg_out_ld8s(TCGContext *s, ARMCond cond,
1103                         TCGReg rd, TCGReg rn, int32_t offset)
1104{
1105    if (offset > 0xff || offset < -0xff) {
1106        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1107        tcg_out_ld8s_r(s, cond, rd, rn, TCG_REG_TMP);
1108    } else
1109        tcg_out_ld8s_8(s, cond, rd, rn, offset);
1110}
1111
1112static void tcg_out_st8(TCGContext *s, ARMCond cond,
1113                        TCGReg rd, TCGReg rn, int32_t offset)
1114{
1115    if (offset > 0xfff || offset < -0xfff) {
1116        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1117        tcg_out_st8_r(s, cond, rd, rn, TCG_REG_TMP);
1118    } else
1119        tcg_out_st8_12(s, cond, rd, rn, offset);
1120}
1121
1122/*
1123 * The _goto case is normally between TBs within the same code buffer, and
1124 * with the code buffer limited to 16MB we wouldn't need the long case.
1125 * But we also use it for the tail-call to the qemu_ld/st helpers, which does.
1126 */
1127static void tcg_out_goto(TCGContext *s, ARMCond cond, const tcg_insn_unit *addr)
1128{
1129    intptr_t addri = (intptr_t)addr;
1130    ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1131    bool arm_mode = !(addri & 1);
1132
1133    if (arm_mode && disp - 8 < 0x01fffffd && disp - 8 > -0x01fffffd) {
1134        tcg_out_b_imm(s, cond, disp);
1135        return;
1136    }
1137
1138    /* LDR is interworking from v5t. */
1139    tcg_out_movi_pool(s, cond, TCG_REG_PC, addri);
1140}
1141
1142/*
1143 * The call case is mostly used for helpers - so it's not unreasonable
1144 * for them to be beyond branch range.
1145 */
1146static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *addr)
1147{
1148    intptr_t addri = (intptr_t)addr;
1149    ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1150    bool arm_mode = !(addri & 1);
1151
1152    if (disp - 8 < 0x02000000 && disp - 8 >= -0x02000000) {
1153        if (arm_mode) {
1154            tcg_out_bl_imm(s, COND_AL, disp);
1155        } else {
1156            tcg_out_blx_imm(s, disp);
1157        }
1158        return;
1159    }
1160
1161    tcg_out_movi32(s, COND_AL, TCG_REG_TMP, addri);
1162    tcg_out_blx_reg(s, COND_AL, TCG_REG_TMP);
1163}
1164
1165static void tcg_out_call(TCGContext *s, const tcg_insn_unit *addr,
1166                         const TCGHelperInfo *info)
1167{
1168    tcg_out_call_int(s, addr);
1169}
1170
1171static void tcg_out_goto_label(TCGContext *s, ARMCond cond, TCGLabel *l)
1172{
1173    if (l->has_value) {
1174        tcg_out_goto(s, cond, l->u.value_ptr);
1175    } else {
1176        tcg_out_reloc(s, s->code_ptr, R_ARM_PC24, l, 0);
1177        tcg_out_b_imm(s, cond, 0);
1178    }
1179}
1180
1181static void tcg_out_mb(TCGContext *s, TCGArg a0)
1182{
1183    if (use_armv7_instructions) {
1184        tcg_out32(s, INSN_DMB_ISH);
1185    } else {
1186        tcg_out32(s, INSN_DMB_MCR);
1187    }
1188}
1189
1190static TCGCond tgen_cmp(TCGContext *s, TCGCond cond, TCGReg a, TCGReg b)
1191{
1192    if (is_tst_cond(cond)) {
1193        tcg_out_dat_reg(s, COND_AL, ARITH_TST, 0, a, b, SHIFT_IMM_LSL(0));
1194        return tcg_tst_eqne_cond(cond);
1195    }
1196    tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, a, b, SHIFT_IMM_LSL(0));
1197    return cond;
1198}
1199
1200static TCGCond tgen_cmpi(TCGContext *s, TCGCond cond, TCGReg a, TCGArg b)
1201{
1202    int imm12;
1203
1204    if (!is_tst_cond(cond)) {
1205        tcg_out_dat_IN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0, a, b);
1206        return cond;
1207    }
1208
1209    /*
1210     * The compare constraints allow rIN, but TST does not support N.
1211     * Be prepared to load the constant into a scratch register.
1212     */
1213    imm12 = encode_imm(b);
1214    if (imm12 >= 0) {
1215        tcg_out_dat_imm(s, COND_AL, ARITH_TST, 0, a, imm12);
1216    } else {
1217        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, b);
1218        tcg_out_dat_reg(s, COND_AL, ARITH_TST, 0,
1219                        a, TCG_REG_TMP, SHIFT_IMM_LSL(0));
1220    }
1221    return tcg_tst_eqne_cond(cond);
1222}
1223
1224static TCGCond tcg_out_cmp(TCGContext *s, TCGCond cond, TCGReg a,
1225                           TCGArg b, int b_const)
1226{
1227    if (b_const) {
1228        return tgen_cmpi(s, cond, a, b);
1229    } else {
1230        return tgen_cmp(s, cond, a, b);
1231    }
1232}
1233
1234static TCGCond tcg_out_cmp2(TCGContext *s, TCGCond cond, TCGReg al, TCGReg ah,
1235                            TCGArg bl, bool const_bl, TCGArg bh, bool const_bh)
1236{
1237    switch (cond) {
1238    case TCG_COND_EQ:
1239    case TCG_COND_NE:
1240    case TCG_COND_LTU:
1241    case TCG_COND_LEU:
1242    case TCG_COND_GTU:
1243    case TCG_COND_GEU:
1244        /*
1245         * We perform a conditional comparison.  If the high half is
1246         * equal, then overwrite the flags with the comparison of the
1247         * low half.  The resulting flags cover the whole.
1248         */
1249        tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0, ah, bh, const_bh);
1250        tcg_out_dat_rI(s, COND_EQ, ARITH_CMP, 0, al, bl, const_bl);
1251        return cond;
1252
1253    case TCG_COND_TSTEQ:
1254    case TCG_COND_TSTNE:
1255        /* Similar, but with TST instead of CMP. */
1256        tcg_out_dat_rI(s, COND_AL, ARITH_TST, 0, ah, bh, const_bh);
1257        tcg_out_dat_rI(s, COND_EQ, ARITH_TST, 0, al, bl, const_bl);
1258        return tcg_tst_eqne_cond(cond);
1259
1260    case TCG_COND_LT:
1261    case TCG_COND_GE:
1262        /* We perform a double-word subtraction and examine the result.
1263           We do not actually need the result of the subtract, so the
1264           low part "subtract" is a compare.  For the high half we have
1265           no choice but to compute into a temporary.  */
1266        tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0, al, bl, const_bl);
1267        tcg_out_dat_rI(s, COND_AL, ARITH_SBC | TO_CPSR,
1268                       TCG_REG_TMP, ah, bh, const_bh);
1269        return cond;
1270
1271    case TCG_COND_LE:
1272    case TCG_COND_GT:
1273        /* Similar, but with swapped arguments, via reversed subtract.  */
1274        tcg_out_dat_rI(s, COND_AL, ARITH_RSB | TO_CPSR,
1275                       TCG_REG_TMP, al, bl, const_bl);
1276        tcg_out_dat_rI(s, COND_AL, ARITH_RSC | TO_CPSR,
1277                       TCG_REG_TMP, ah, bh, const_bh);
1278        return tcg_swap_cond(cond);
1279
1280    default:
1281        g_assert_not_reached();
1282    }
1283}
1284
1285/*
1286 * Note that TCGReg references Q-registers.
1287 * Q-regno = 2 * D-regno, so shift left by 1 while inserting.
1288 */
1289static uint32_t encode_vd(TCGReg rd)
1290{
1291    tcg_debug_assert(rd >= TCG_REG_Q0);
1292    return (extract32(rd, 3, 1) << 22) | (extract32(rd, 0, 3) << 13);
1293}
1294
1295static uint32_t encode_vn(TCGReg rn)
1296{
1297    tcg_debug_assert(rn >= TCG_REG_Q0);
1298    return (extract32(rn, 3, 1) << 7) | (extract32(rn, 0, 3) << 17);
1299}
1300
1301static uint32_t encode_vm(TCGReg rm)
1302{
1303    tcg_debug_assert(rm >= TCG_REG_Q0);
1304    return (extract32(rm, 3, 1) << 5) | (extract32(rm, 0, 3) << 1);
1305}
1306
1307static void tcg_out_vreg2(TCGContext *s, ARMInsn insn, int q, int vece,
1308                          TCGReg d, TCGReg m)
1309{
1310    tcg_out32(s, insn | (vece << 18) | (q << 6) |
1311              encode_vd(d) | encode_vm(m));
1312}
1313
1314static void tcg_out_vreg3(TCGContext *s, ARMInsn insn, int q, int vece,
1315                          TCGReg d, TCGReg n, TCGReg m)
1316{
1317    tcg_out32(s, insn | (vece << 20) | (q << 6) |
1318              encode_vd(d) | encode_vn(n) | encode_vm(m));
1319}
1320
1321static void tcg_out_vmovi(TCGContext *s, TCGReg rd,
1322                          int q, int op, int cmode, uint8_t imm8)
1323{
1324    tcg_out32(s, INSN_VMOVI | encode_vd(rd) | (q << 6) | (op << 5)
1325              | (cmode << 8) | extract32(imm8, 0, 4)
1326              | (extract32(imm8, 4, 3) << 16)
1327              | (extract32(imm8, 7, 1) << 24));
1328}
1329
1330static void tcg_out_vshifti(TCGContext *s, ARMInsn insn, int q,
1331                            TCGReg rd, TCGReg rm, int l_imm6)
1332{
1333    tcg_out32(s, insn | (q << 6) | encode_vd(rd) | encode_vm(rm) |
1334              (extract32(l_imm6, 6, 1) << 7) |
1335              (extract32(l_imm6, 0, 6) << 16));
1336}
1337
1338static void tcg_out_vldst(TCGContext *s, ARMInsn insn,
1339                          TCGReg rd, TCGReg rn, int offset)
1340{
1341    if (offset != 0) {
1342        if (check_fit_imm(offset) || check_fit_imm(-offset)) {
1343            tcg_out_dat_rIN(s, COND_AL, ARITH_ADD, ARITH_SUB,
1344                            TCG_REG_TMP, rn, offset, true);
1345        } else {
1346            tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, offset);
1347            tcg_out_dat_reg(s, COND_AL, ARITH_ADD,
1348                            TCG_REG_TMP, TCG_REG_TMP, rn, 0);
1349        }
1350        rn = TCG_REG_TMP;
1351    }
1352    tcg_out32(s, insn | (rn << 16) | encode_vd(rd) | 0xf);
1353}
1354
1355typedef struct {
1356    ARMCond cond;
1357    TCGReg base;
1358    int index;
1359    bool index_scratch;
1360    TCGAtomAlign aa;
1361} HostAddress;
1362
1363bool tcg_target_has_memory_bswap(MemOp memop)
1364{
1365    return false;
1366}
1367
1368static TCGReg ldst_ra_gen(TCGContext *s, const TCGLabelQemuLdst *l, int arg)
1369{
1370    /* We arrive at the slow path via "BLNE", so R14 contains l->raddr. */
1371    return TCG_REG_R14;
1372}
1373
1374static const TCGLdstHelperParam ldst_helper_param = {
1375    .ra_gen = ldst_ra_gen,
1376    .ntmp = 1,
1377    .tmp = { TCG_REG_TMP },
1378};
1379
1380static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1381{
1382    MemOp opc = get_memop(lb->oi);
1383
1384    if (!reloc_pc24(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1385        return false;
1386    }
1387
1388    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1389    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1390    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1391
1392    tcg_out_goto(s, COND_AL, lb->raddr);
1393    return true;
1394}
1395
1396static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1397{
1398    MemOp opc = get_memop(lb->oi);
1399
1400    if (!reloc_pc24(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1401        return false;
1402    }
1403
1404    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1405
1406    /* Tail-call to the helper, which will return to the fast path.  */
1407    tcg_out_goto(s, COND_AL, qemu_st_helpers[opc & MO_SIZE]);
1408    return true;
1409}
1410
1411/* We expect to use an 9-bit sign-magnitude negative offset from ENV.  */
1412#define MIN_TLB_MASK_TABLE_OFS  -256
1413
1414static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1415                                           TCGReg addr, MemOpIdx oi, bool is_ld)
1416{
1417    TCGLabelQemuLdst *ldst = NULL;
1418    MemOp opc = get_memop(oi);
1419    unsigned a_mask;
1420
1421    if (tcg_use_softmmu) {
1422        *h = (HostAddress){
1423            .cond = COND_AL,
1424            .base = addr,
1425            .index = TCG_REG_R1,
1426            .index_scratch = true,
1427        };
1428    } else {
1429        *h = (HostAddress){
1430            .cond = COND_AL,
1431            .base = addr,
1432            .index = guest_base ? TCG_REG_GUEST_BASE : -1,
1433            .index_scratch = false,
1434        };
1435    }
1436
1437    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, false);
1438    a_mask = (1 << h->aa.align) - 1;
1439
1440    if (tcg_use_softmmu) {
1441        int mem_index = get_mmuidx(oi);
1442        int cmp_off = is_ld ? offsetof(CPUTLBEntry, addr_read)
1443                            : offsetof(CPUTLBEntry, addr_write);
1444        int fast_off = tlb_mask_table_ofs(s, mem_index);
1445        unsigned s_mask = (1 << (opc & MO_SIZE)) - 1;
1446        TCGReg t_addr;
1447
1448        ldst = new_ldst_label(s);
1449        ldst->is_ld = is_ld;
1450        ldst->oi = oi;
1451        ldst->addr_reg = addr;
1452
1453        /* Load cpu->neg.tlb.f[mmu_idx].{mask,table} into {r0,r1}.  */
1454        QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, mask) != 0);
1455        QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, table) != 4);
1456        tcg_out_ldrd_8(s, COND_AL, TCG_REG_R0, TCG_AREG0, fast_off);
1457
1458        /* Extract the tlb index from the address into R0.  */
1459        tcg_out_dat_reg(s, COND_AL, ARITH_AND, TCG_REG_R0, TCG_REG_R0, addr,
1460                        SHIFT_IMM_LSR(s->page_bits - CPU_TLB_ENTRY_BITS));
1461
1462        /*
1463         * Add the tlb_table pointer, creating the CPUTLBEntry address in R1.
1464         * Load the tlb comparator into R2 and the fast path addend into R1.
1465         */
1466        if (cmp_off == 0) {
1467            tcg_out_ld32_rwb(s, COND_AL, TCG_REG_R2, TCG_REG_R1, TCG_REG_R0);
1468        } else {
1469            tcg_out_dat_reg(s, COND_AL, ARITH_ADD,
1470                            TCG_REG_R1, TCG_REG_R1, TCG_REG_R0, 0);
1471            tcg_out_ld32_12(s, COND_AL, TCG_REG_R2, TCG_REG_R1, cmp_off);
1472        }
1473
1474        /* Load the tlb addend.  */
1475        tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R1,
1476                        offsetof(CPUTLBEntry, addend));
1477
1478        /*
1479         * Check alignment, check comparators.
1480         * Do this in 2-4 insns.  Use MOVW for v7, if possible,
1481         * to reduce the number of sequential conditional instructions.
1482         * Almost all guests have at least 4k pages, which means that we need
1483         * to clear at least 9 bits even for an 8-byte memory, which means it
1484         * isn't worth checking for an immediate operand for BIC.
1485         *
1486         * For unaligned accesses, test the page of the last unit of alignment.
1487         * This leaves the least significant alignment bits unchanged, and of
1488         * course must be zero.
1489         */
1490        t_addr = addr;
1491        if (a_mask < s_mask) {
1492            t_addr = TCG_REG_R0;
1493            tcg_out_dat_imm(s, COND_AL, ARITH_ADD, t_addr,
1494                            addr, s_mask - a_mask);
1495        }
1496        if (use_armv7_instructions && s->page_bits <= 16) {
1497            tcg_out_movi32(s, COND_AL, TCG_REG_TMP, ~(s->page_mask | a_mask));
1498            tcg_out_dat_reg(s, COND_AL, ARITH_BIC, TCG_REG_TMP,
1499                            t_addr, TCG_REG_TMP, 0);
1500            tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1501                            TCG_REG_R2, TCG_REG_TMP, 0);
1502        } else {
1503            if (a_mask) {
1504                tcg_debug_assert(a_mask <= 0xff);
1505                tcg_out_dat_imm(s, COND_AL, ARITH_TST, 0, addr, a_mask);
1506            }
1507            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_TMP, 0, t_addr,
1508                            SHIFT_IMM_LSR(s->page_bits));
1509            tcg_out_dat_reg(s, (a_mask ? COND_EQ : COND_AL), ARITH_CMP,
1510                            0, TCG_REG_R2, TCG_REG_TMP,
1511                            SHIFT_IMM_LSL(s->page_bits));
1512        }
1513    } else if (a_mask) {
1514        ldst = new_ldst_label(s);
1515        ldst->is_ld = is_ld;
1516        ldst->oi = oi;
1517        ldst->addr_reg = addr;
1518
1519        /* We are expecting alignment to max out at 7 */
1520        tcg_debug_assert(a_mask <= 0xff);
1521        /* tst addr, #mask */
1522        tcg_out_dat_imm(s, COND_AL, ARITH_TST, 0, addr, a_mask);
1523    }
1524
1525    return ldst;
1526}
1527
1528static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg datalo,
1529                                   TCGReg datahi, HostAddress h)
1530{
1531    TCGReg base;
1532
1533    /* Byte swapping is left to middle-end expansion. */
1534    tcg_debug_assert((opc & MO_BSWAP) == 0);
1535
1536    switch (opc & MO_SSIZE) {
1537    case MO_UB:
1538        if (h.index < 0) {
1539            tcg_out_ld8_12(s, h.cond, datalo, h.base, 0);
1540        } else {
1541            tcg_out_ld8_r(s, h.cond, datalo, h.base, h.index);
1542        }
1543        break;
1544    case MO_SB:
1545        if (h.index < 0) {
1546            tcg_out_ld8s_8(s, h.cond, datalo, h.base, 0);
1547        } else {
1548            tcg_out_ld8s_r(s, h.cond, datalo, h.base, h.index);
1549        }
1550        break;
1551    case MO_UW:
1552        if (h.index < 0) {
1553            tcg_out_ld16u_8(s, h.cond, datalo, h.base, 0);
1554        } else {
1555            tcg_out_ld16u_r(s, h.cond, datalo, h.base, h.index);
1556        }
1557        break;
1558    case MO_SW:
1559        if (h.index < 0) {
1560            tcg_out_ld16s_8(s, h.cond, datalo, h.base, 0);
1561        } else {
1562            tcg_out_ld16s_r(s, h.cond, datalo, h.base, h.index);
1563        }
1564        break;
1565    case MO_UL:
1566        if (h.index < 0) {
1567            tcg_out_ld32_12(s, h.cond, datalo, h.base, 0);
1568        } else {
1569            tcg_out_ld32_r(s, h.cond, datalo, h.base, h.index);
1570        }
1571        break;
1572    case MO_UQ:
1573        /* We used pair allocation for datalo, so already should be aligned. */
1574        tcg_debug_assert((datalo & 1) == 0);
1575        tcg_debug_assert(datahi == datalo + 1);
1576        /* LDRD requires alignment; double-check that. */
1577        if (memop_alignment_bits(opc) >= MO_64) {
1578            if (h.index < 0) {
1579                tcg_out_ldrd_8(s, h.cond, datalo, h.base, 0);
1580                break;
1581            }
1582            /*
1583             * Rm (the second address op) must not overlap Rt or Rt + 1.
1584             * Since datalo is aligned, we can simplify the test via alignment.
1585             * Flip the two address arguments if that works.
1586             */
1587            if ((h.index & ~1) != datalo) {
1588                tcg_out_ldrd_r(s, h.cond, datalo, h.base, h.index);
1589                break;
1590            }
1591            if ((h.base & ~1) != datalo) {
1592                tcg_out_ldrd_r(s, h.cond, datalo, h.index, h.base);
1593                break;
1594            }
1595        }
1596        if (h.index < 0) {
1597            base = h.base;
1598            if (datalo == h.base) {
1599                tcg_out_mov_reg(s, h.cond, TCG_REG_TMP, base);
1600                base = TCG_REG_TMP;
1601            }
1602        } else if (h.index_scratch) {
1603            tcg_out_ld32_rwb(s, h.cond, datalo, h.index, h.base);
1604            tcg_out_ld32_12(s, h.cond, datahi, h.index, 4);
1605            break;
1606        } else {
1607            tcg_out_dat_reg(s, h.cond, ARITH_ADD, TCG_REG_TMP,
1608                            h.base, h.index, SHIFT_IMM_LSL(0));
1609            base = TCG_REG_TMP;
1610        }
1611        tcg_out_ld32_12(s, h.cond, datalo, base, 0);
1612        tcg_out_ld32_12(s, h.cond, datahi, base, 4);
1613        break;
1614    default:
1615        g_assert_not_reached();
1616    }
1617}
1618
1619static void tcg_out_qemu_ld(TCGContext *s, TCGReg datalo, TCGReg datahi,
1620                            TCGReg addr, MemOpIdx oi, TCGType data_type)
1621{
1622    MemOp opc = get_memop(oi);
1623    TCGLabelQemuLdst *ldst;
1624    HostAddress h;
1625
1626    ldst = prepare_host_addr(s, &h, addr, oi, true);
1627    if (ldst) {
1628        ldst->type = data_type;
1629        ldst->datalo_reg = datalo;
1630        ldst->datahi_reg = datahi;
1631
1632        /*
1633         * This a conditional BL only to load a pointer within this
1634         * opcode into LR for the slow path.  We will not be using
1635         * the value for a tail call.
1636         */
1637        ldst->label_ptr[0] = s->code_ptr;
1638        tcg_out_bl_imm(s, COND_NE, 0);
1639
1640        tcg_out_qemu_ld_direct(s, opc, datalo, datahi, h);
1641        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1642    } else {
1643        tcg_out_qemu_ld_direct(s, opc, datalo, datahi, h);
1644    }
1645}
1646
1647static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg datalo,
1648                                   TCGReg datahi, HostAddress h)
1649{
1650    /* Byte swapping is left to middle-end expansion. */
1651    tcg_debug_assert((opc & MO_BSWAP) == 0);
1652
1653    switch (opc & MO_SIZE) {
1654    case MO_8:
1655        if (h.index < 0) {
1656            tcg_out_st8_12(s, h.cond, datalo, h.base, 0);
1657        } else {
1658            tcg_out_st8_r(s, h.cond, datalo, h.base, h.index);
1659        }
1660        break;
1661    case MO_16:
1662        if (h.index < 0) {
1663            tcg_out_st16_8(s, h.cond, datalo, h.base, 0);
1664        } else {
1665            tcg_out_st16_r(s, h.cond, datalo, h.base, h.index);
1666        }
1667        break;
1668    case MO_32:
1669        if (h.index < 0) {
1670            tcg_out_st32_12(s, h.cond, datalo, h.base, 0);
1671        } else {
1672            tcg_out_st32_r(s, h.cond, datalo, h.base, h.index);
1673        }
1674        break;
1675    case MO_64:
1676        /* We used pair allocation for datalo, so already should be aligned. */
1677        tcg_debug_assert((datalo & 1) == 0);
1678        tcg_debug_assert(datahi == datalo + 1);
1679        /* STRD requires alignment; double-check that. */
1680        if (memop_alignment_bits(opc) >= MO_64) {
1681            if (h.index < 0) {
1682                tcg_out_strd_8(s, h.cond, datalo, h.base, 0);
1683            } else {
1684                tcg_out_strd_r(s, h.cond, datalo, h.base, h.index);
1685            }
1686        } else if (h.index < 0) {
1687            tcg_out_st32_12(s, h.cond, datalo, h.base, 0);
1688            tcg_out_st32_12(s, h.cond, datahi, h.base, 4);
1689        } else if (h.index_scratch) {
1690            tcg_out_st32_rwb(s, h.cond, datalo, h.index, h.base);
1691            tcg_out_st32_12(s, h.cond, datahi, h.index, 4);
1692        } else {
1693            tcg_out_dat_reg(s, h.cond, ARITH_ADD, TCG_REG_TMP,
1694                            h.base, h.index, SHIFT_IMM_LSL(0));
1695            tcg_out_st32_12(s, h.cond, datalo, TCG_REG_TMP, 0);
1696            tcg_out_st32_12(s, h.cond, datahi, TCG_REG_TMP, 4);
1697        }
1698        break;
1699    default:
1700        g_assert_not_reached();
1701    }
1702}
1703
1704static void tcg_out_qemu_st(TCGContext *s, TCGReg datalo, TCGReg datahi,
1705                            TCGReg addr, MemOpIdx oi, TCGType data_type)
1706{
1707    MemOp opc = get_memop(oi);
1708    TCGLabelQemuLdst *ldst;
1709    HostAddress h;
1710
1711    ldst = prepare_host_addr(s, &h, addr, oi, false);
1712    if (ldst) {
1713        ldst->type = data_type;
1714        ldst->datalo_reg = datalo;
1715        ldst->datahi_reg = datahi;
1716
1717        h.cond = COND_EQ;
1718        tcg_out_qemu_st_direct(s, opc, datalo, datahi, h);
1719
1720        /* The conditional call is last, as we're going to return here. */
1721        ldst->label_ptr[0] = s->code_ptr;
1722        tcg_out_bl_imm(s, COND_NE, 0);
1723        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1724    } else {
1725        tcg_out_qemu_st_direct(s, opc, datalo, datahi, h);
1726    }
1727}
1728
1729static void tcg_out_epilogue(TCGContext *s);
1730
1731static void tcg_out_exit_tb(TCGContext *s, uintptr_t arg)
1732{
1733    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, arg);
1734    tcg_out_epilogue(s);
1735}
1736
1737static void tcg_out_goto_tb(TCGContext *s, int which)
1738{
1739    uintptr_t i_addr;
1740    intptr_t i_disp;
1741
1742    /* Direct branch will be patched by tb_target_set_jmp_target. */
1743    set_jmp_insn_offset(s, which);
1744    tcg_out32(s, INSN_NOP);
1745
1746    /* When branch is out of range, fall through to indirect. */
1747    i_addr = get_jmp_target_addr(s, which);
1748    i_disp = tcg_pcrel_diff(s, (void *)i_addr) - 8;
1749    tcg_debug_assert(i_disp < 0);
1750    if (i_disp >= -0xfff) {
1751        tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, i_disp);
1752    } else {
1753        /*
1754         * The TB is close, but outside the 12 bits addressable by
1755         * the load.  We can extend this to 20 bits with a sub of a
1756         * shifted immediate from pc.
1757         */
1758        int h = -i_disp;
1759        int l = -(h & 0xfff);
1760
1761        h = encode_imm_nofail(h + l);
1762        tcg_out_dat_imm(s, COND_AL, ARITH_SUB, TCG_REG_R0, TCG_REG_PC, h);
1763        tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, l);
1764    }
1765    set_jmp_reset_offset(s, which);
1766}
1767
1768void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1769                              uintptr_t jmp_rx, uintptr_t jmp_rw)
1770{
1771    uintptr_t addr = tb->jmp_target_addr[n];
1772    ptrdiff_t offset = addr - (jmp_rx + 8);
1773    tcg_insn_unit insn;
1774
1775    /* Either directly branch, or fall through to indirect branch. */
1776    if (offset == sextract64(offset, 0, 26)) {
1777        /* B <addr> */
1778        insn = deposit32((COND_AL << 28) | INSN_B, 0, 24, offset >> 2);
1779    } else {
1780        insn = INSN_NOP;
1781    }
1782
1783    qatomic_set((uint32_t *)jmp_rw, insn);
1784    flush_idcache_range(jmp_rx, jmp_rw, 4);
1785}
1786
1787
1788static void tgen_add(TCGContext *s, TCGType type,
1789                     TCGReg a0, TCGReg a1, TCGReg a2)
1790{
1791    tcg_out_dat_reg(s, COND_AL, ARITH_ADD, a0, a1, a2, SHIFT_IMM_LSL(0));
1792}
1793
1794static void tgen_addi(TCGContext *s, TCGType type,
1795                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1796{
1797    tcg_out_dat_IN(s, COND_AL, ARITH_ADD, ARITH_SUB, a0, a1, a2);
1798}
1799
1800static const TCGOutOpBinary outop_add = {
1801    .base.static_constraint = C_O1_I2(r, r, rIN),
1802    .out_rrr = tgen_add,
1803    .out_rri = tgen_addi,
1804};
1805
1806static void tgen_and(TCGContext *s, TCGType type,
1807                     TCGReg a0, TCGReg a1, TCGReg a2)
1808{
1809    tcg_out_dat_reg(s, COND_AL, ARITH_AND, a0, a1, a2, SHIFT_IMM_LSL(0));
1810}
1811
1812static void tgen_andi(TCGContext *s, TCGType type,
1813                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1814{
1815    tcg_out_dat_IK(s, COND_AL, ARITH_AND, ARITH_BIC, a0, a1, a2);
1816}
1817
1818static const TCGOutOpBinary outop_and = {
1819    .base.static_constraint = C_O1_I2(r, r, rIK),
1820    .out_rrr = tgen_and,
1821    .out_rri = tgen_andi,
1822};
1823
1824static void tgen_andc(TCGContext *s, TCGType type,
1825                      TCGReg a0, TCGReg a1, TCGReg a2)
1826{
1827    tcg_out_dat_reg(s, COND_AL, ARITH_BIC, a0, a1, a2, SHIFT_IMM_LSL(0));
1828}
1829
1830static const TCGOutOpBinary outop_andc = {
1831    .base.static_constraint = C_O1_I2(r, r, r),
1832    .out_rrr = tgen_andc,
1833};
1834
1835static void tgen_clz(TCGContext *s, TCGType type,
1836                     TCGReg a0, TCGReg a1, TCGReg a2)
1837{
1838    tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0, a1, 0);
1839    tcg_out_dat_reg(s, COND_NE, INSN_CLZ, a0, 0, a1, 0);
1840    tcg_out_mov_reg(s, COND_EQ, a0, a2);
1841}
1842
1843static void tgen_clzi(TCGContext *s, TCGType type,
1844                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1845{
1846    if (a2 == 32) {
1847        tcg_out_dat_reg(s, COND_AL, INSN_CLZ, a0, 0, a1, 0);
1848    } else {
1849        tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0, a1, 0);
1850        tcg_out_dat_reg(s, COND_NE, INSN_CLZ, a0, 0, a1, 0);
1851        tcg_out_movi32(s, COND_EQ, a0, a2);
1852    }
1853}
1854
1855static const TCGOutOpBinary outop_clz = {
1856    .base.static_constraint = C_O1_I2(r, r, rIK),
1857    .out_rrr = tgen_clz,
1858    .out_rri = tgen_clzi,
1859};
1860
1861static const TCGOutOpUnary outop_ctpop = {
1862    .base.static_constraint = C_NotImplemented,
1863};
1864
1865static void tgen_ctz(TCGContext *s, TCGType type,
1866                     TCGReg a0, TCGReg a1, TCGReg a2)
1867{
1868    tcg_out_dat_reg(s, COND_AL, INSN_RBIT, TCG_REG_TMP, 0, a1, 0);
1869    tgen_clz(s, TCG_TYPE_I32, a0, TCG_REG_TMP, a2);
1870}
1871
1872static void tgen_ctzi(TCGContext *s, TCGType type,
1873                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1874{
1875    tcg_out_dat_reg(s, COND_AL, INSN_RBIT, TCG_REG_TMP, 0, a1, 0);
1876    tgen_clzi(s, TCG_TYPE_I32, a0, TCG_REG_TMP, a2);
1877}
1878
1879static TCGConstraintSetIndex cset_ctz(TCGType type, unsigned flags)
1880{
1881    return use_armv7_instructions ? C_O1_I2(r, r, rIK) : C_NotImplemented;
1882}
1883
1884static const TCGOutOpBinary outop_ctz = {
1885    .base.static_constraint = C_Dynamic,
1886    .base.dynamic_constraint = cset_ctz,
1887    .out_rrr = tgen_ctz,
1888    .out_rri = tgen_ctzi,
1889};
1890
1891static TCGConstraintSetIndex cset_idiv(TCGType type, unsigned flags)
1892{
1893    return use_idiv_instructions ? C_O1_I2(r, r, r) : C_NotImplemented;
1894}
1895
1896static void tgen_divs(TCGContext *s, TCGType type,
1897                      TCGReg a0, TCGReg a1, TCGReg a2)
1898{
1899    /* sdiv */
1900    tcg_out32(s, 0x0710f010 | (COND_AL << 28) | (a0 << 16) | a1 | (a2 << 8));
1901}
1902
1903static const TCGOutOpBinary outop_divs = {
1904    .base.static_constraint = C_Dynamic,
1905    .base.dynamic_constraint = cset_idiv,
1906    .out_rrr = tgen_divs,
1907};
1908
1909static const TCGOutOpDivRem outop_divs2 = {
1910    .base.static_constraint = C_NotImplemented,
1911};
1912
1913static void tgen_divu(TCGContext *s, TCGType type,
1914                      TCGReg a0, TCGReg a1, TCGReg a2)
1915{
1916    /* udiv */
1917    tcg_out32(s, 0x0730f010 | (COND_AL << 28) | (a0 << 16) | a1 | (a2 << 8));
1918}
1919
1920static const TCGOutOpBinary outop_divu = {
1921    .base.static_constraint = C_Dynamic,
1922    .base.dynamic_constraint = cset_idiv,
1923    .out_rrr = tgen_divu,
1924};
1925
1926static const TCGOutOpDivRem outop_divu2 = {
1927    .base.static_constraint = C_NotImplemented,
1928};
1929
1930static const TCGOutOpBinary outop_eqv = {
1931    .base.static_constraint = C_NotImplemented,
1932};
1933
1934static void tgen_mul(TCGContext *s, TCGType type,
1935                     TCGReg a0, TCGReg a1, TCGReg a2)
1936{
1937    /* mul */
1938    tcg_out32(s, (COND_AL << 28) | 0x90 | (a0 << 16) | (a1 << 8) | a2);
1939}
1940
1941static const TCGOutOpBinary outop_mul = {
1942    .base.static_constraint = C_O1_I2(r, r, r),
1943    .out_rrr = tgen_mul,
1944};
1945
1946static void tgen_muls2(TCGContext *s, TCGType type,
1947                       TCGReg rd0, TCGReg rd1, TCGReg rn, TCGReg rm)
1948{
1949    /* smull */
1950    tcg_out32(s, (COND_AL << 28) | 0x00c00090 |
1951              (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
1952}
1953
1954static const TCGOutOpMul2 outop_muls2 = {
1955    .base.static_constraint = C_O2_I2(r, r, r, r),
1956    .out_rrrr = tgen_muls2,
1957};
1958
1959static const TCGOutOpBinary outop_mulsh = {
1960    .base.static_constraint = C_NotImplemented,
1961};
1962
1963static void tgen_mulu2(TCGContext *s, TCGType type,
1964                       TCGReg rd0, TCGReg rd1, TCGReg rn, TCGReg rm)
1965{
1966    /* umull */
1967    tcg_out32(s, (COND_AL << 28) | 0x00800090 |
1968              (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
1969}
1970
1971static const TCGOutOpMul2 outop_mulu2 = {
1972    .base.static_constraint = C_O2_I2(r, r, r, r),
1973    .out_rrrr = tgen_mulu2,
1974};
1975
1976static const TCGOutOpBinary outop_muluh = {
1977    .base.static_constraint = C_NotImplemented,
1978};
1979
1980static const TCGOutOpBinary outop_nand = {
1981    .base.static_constraint = C_NotImplemented,
1982};
1983
1984static const TCGOutOpBinary outop_nor = {
1985    .base.static_constraint = C_NotImplemented,
1986};
1987
1988static void tgen_or(TCGContext *s, TCGType type,
1989                     TCGReg a0, TCGReg a1, TCGReg a2)
1990{
1991    tcg_out_dat_reg(s, COND_AL, ARITH_ORR, a0, a1, a2, SHIFT_IMM_LSL(0));
1992}
1993
1994static void tgen_ori(TCGContext *s, TCGType type,
1995                     TCGReg a0, TCGReg a1, tcg_target_long a2)
1996{
1997    tcg_out_dat_imm(s, COND_AL, ARITH_ORR, a0, a1, encode_imm_nofail(a2));
1998}
1999
2000static const TCGOutOpBinary outop_or = {
2001    .base.static_constraint = C_O1_I2(r, r, rI),
2002    .out_rrr = tgen_or,
2003    .out_rri = tgen_ori,
2004};
2005
2006static const TCGOutOpBinary outop_orc = {
2007    .base.static_constraint = C_NotImplemented,
2008};
2009
2010static const TCGOutOpBinary outop_rems = {
2011    .base.static_constraint = C_NotImplemented,
2012};
2013
2014static const TCGOutOpBinary outop_remu = {
2015    .base.static_constraint = C_NotImplemented,
2016};
2017
2018static const TCGOutOpBinary outop_rotl = {
2019    .base.static_constraint = C_NotImplemented,
2020};
2021
2022static void tgen_rotr(TCGContext *s, TCGType type,
2023                      TCGReg a0, TCGReg a1, TCGReg a2)
2024{
2025    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1, SHIFT_REG_ROR(a2));
2026}
2027
2028static void tgen_rotri(TCGContext *s, TCGType type,
2029                       TCGReg a0, TCGReg a1, tcg_target_long a2)
2030{
2031    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1, SHIFT_IMM_ROR(a2 & 0x1f));
2032}
2033
2034static const TCGOutOpBinary outop_rotr = {
2035    .base.static_constraint = C_O1_I2(r, r, ri),
2036    .out_rrr = tgen_rotr,
2037    .out_rri = tgen_rotri,
2038};
2039
2040static void tgen_sar(TCGContext *s, TCGType type,
2041                     TCGReg a0, TCGReg a1, TCGReg a2)
2042{
2043    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1, SHIFT_REG_ASR(a2));
2044}
2045
2046static void tgen_sari(TCGContext *s, TCGType type,
2047                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2048{
2049    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1,
2050                    SHIFT_IMM_ASR(a2 & 0x1f));
2051}
2052
2053static const TCGOutOpBinary outop_sar = {
2054    .base.static_constraint = C_O1_I2(r, r, ri),
2055    .out_rrr = tgen_sar,
2056    .out_rri = tgen_sari,
2057};
2058
2059static void tgen_shl(TCGContext *s, TCGType type,
2060                     TCGReg a0, TCGReg a1, TCGReg a2)
2061{
2062    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1, SHIFT_REG_LSL(a2));
2063}
2064
2065static void tgen_shli(TCGContext *s, TCGType type,
2066                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2067{
2068    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1,
2069                    SHIFT_IMM_LSL(a2 & 0x1f));
2070}
2071
2072static const TCGOutOpBinary outop_shl = {
2073    .base.static_constraint = C_O1_I2(r, r, ri),
2074    .out_rrr = tgen_shl,
2075    .out_rri = tgen_shli,
2076};
2077
2078static void tgen_shr(TCGContext *s, TCGType type,
2079                     TCGReg a0, TCGReg a1, TCGReg a2)
2080{
2081    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1, SHIFT_REG_LSR(a2));
2082}
2083
2084static void tgen_shri(TCGContext *s, TCGType type,
2085                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2086{
2087    tcg_out_dat_reg(s, COND_AL, ARITH_MOV, a0, 0, a1,
2088                    SHIFT_IMM_LSR(a2 & 0x1f));
2089}
2090
2091static const TCGOutOpBinary outop_shr = {
2092    .base.static_constraint = C_O1_I2(r, r, ri),
2093    .out_rrr = tgen_shr,
2094    .out_rri = tgen_shri,
2095};
2096
2097static void tgen_sub(TCGContext *s, TCGType type,
2098                     TCGReg a0, TCGReg a1, TCGReg a2)
2099{
2100    tcg_out_dat_reg(s, COND_AL, ARITH_SUB, a0, a1, a2, SHIFT_IMM_LSL(0));
2101}
2102
2103static void tgen_subfi(TCGContext *s, TCGType type,
2104                       TCGReg a0, tcg_target_long a1, TCGReg a2)
2105{
2106    tcg_out_dat_imm(s, COND_AL, ARITH_RSB, a0, a2, encode_imm_nofail(a1));
2107}
2108
2109static const TCGOutOpSubtract outop_sub = {
2110    .base.static_constraint = C_O1_I2(r, rI, r),
2111    .out_rrr = tgen_sub,
2112    .out_rir = tgen_subfi,
2113};
2114
2115static void tgen_xor(TCGContext *s, TCGType type,
2116                     TCGReg a0, TCGReg a1, TCGReg a2)
2117{
2118    tcg_out_dat_reg(s, COND_AL, ARITH_EOR, a0, a1, a2, SHIFT_IMM_LSL(0));
2119}
2120
2121static void tgen_xori(TCGContext *s, TCGType type,
2122                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2123{
2124    tcg_out_dat_imm(s, COND_AL, ARITH_EOR, a0, a1, encode_imm_nofail(a2));
2125}
2126
2127static const TCGOutOpBinary outop_xor = {
2128    .base.static_constraint = C_O1_I2(r, r, rI),
2129    .out_rrr = tgen_xor,
2130    .out_rri = tgen_xori,
2131};
2132
2133static void tgen_bswap16(TCGContext *s, TCGType type,
2134                         TCGReg rd, TCGReg rn, unsigned flags)
2135{
2136    if (flags & TCG_BSWAP_OS) {
2137        /* revsh */
2138        tcg_out32(s, 0x06ff0fb0 | (COND_AL << 28) | (rd << 12) | rn);
2139        return;
2140    }
2141
2142    /* rev16 */
2143    tcg_out32(s, 0x06bf0fb0 | (COND_AL << 28) | (rd << 12) | rn);
2144    if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2145        tcg_out_ext16u(s, rd, rd);
2146    }
2147}
2148
2149static const TCGOutOpBswap outop_bswap16 = {
2150    .base.static_constraint = C_O1_I1(r, r),
2151    .out_rr = tgen_bswap16,
2152};
2153
2154static void tgen_bswap32(TCGContext *s, TCGType type,
2155                         TCGReg rd, TCGReg rn, unsigned flags)
2156{
2157    /* rev */
2158    tcg_out32(s, 0x06bf0f30 | (COND_AL << 28) | (rd << 12) | rn);
2159}
2160
2161static const TCGOutOpBswap outop_bswap32 = {
2162    .base.static_constraint = C_O1_I1(r, r),
2163    .out_rr = tgen_bswap32,
2164};
2165
2166static const TCGOutOpUnary outop_bswap64 = {
2167    .base.static_constraint = C_NotImplemented,
2168};
2169
2170static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2171{
2172    tgen_subfi(s, type, a0, 0, a1);
2173}
2174
2175static const TCGOutOpUnary outop_neg = {
2176    .base.static_constraint = C_O1_I1(r, r),
2177    .out_rr = tgen_neg,
2178};
2179
2180static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2181{
2182    tcg_out_dat_reg(s, COND_AL, ARITH_MVN, a0, 0, a1, SHIFT_IMM_LSL(0));
2183}
2184
2185static const TCGOutOpUnary outop_not = {
2186    .base.static_constraint = C_O1_I1(r, r),
2187    .out_rr = tgen_not,
2188};
2189
2190static void tgen_brcond(TCGContext *s, TCGType type, TCGCond cond,
2191                        TCGReg a0, TCGReg a1, TCGLabel *l)
2192{
2193    cond = tgen_cmp(s, cond, a0, a1);
2194    tcg_out_goto_label(s, tcg_cond_to_arm_cond[cond], l);
2195}
2196
2197static void tgen_brcondi(TCGContext *s, TCGType type, TCGCond cond,
2198                         TCGReg a0, tcg_target_long a1, TCGLabel *l)
2199{
2200    cond = tgen_cmpi(s, cond, a0, a1);
2201    tcg_out_goto_label(s, tcg_cond_to_arm_cond[cond], l);
2202}
2203
2204static const TCGOutOpBrcond outop_brcond = {
2205    .base.static_constraint = C_O0_I2(r, rIN),
2206    .out_rr = tgen_brcond,
2207    .out_ri = tgen_brcondi,
2208};
2209
2210static void finish_setcond(TCGContext *s, TCGCond cond, TCGReg ret, bool neg)
2211{
2212    tcg_out_movi32(s, tcg_cond_to_arm_cond[tcg_invert_cond(cond)], ret, 0);
2213    tcg_out_movi32(s, tcg_cond_to_arm_cond[cond], ret, neg ? -1 : 1);
2214}
2215
2216static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
2217                         TCGReg a0, TCGReg a1, TCGReg a2)
2218{
2219    cond = tgen_cmp(s, cond, a1, a2);
2220    finish_setcond(s, cond, a0, false);
2221}
2222
2223static void tgen_setcondi(TCGContext *s, TCGType type, TCGCond cond,
2224                          TCGReg a0, TCGReg a1, tcg_target_long a2)
2225{
2226    cond = tgen_cmpi(s, cond, a1, a2);
2227    finish_setcond(s, cond, a0, false);
2228}
2229
2230static const TCGOutOpSetcond outop_setcond = {
2231    .base.static_constraint = C_O1_I2(r, r, rIN),
2232    .out_rrr = tgen_setcond,
2233    .out_rri = tgen_setcondi,
2234};
2235
2236static void tgen_negsetcond(TCGContext *s, TCGType type, TCGCond cond,
2237                            TCGReg a0, TCGReg a1, TCGReg a2)
2238{
2239    cond = tgen_cmp(s, cond, a1, a2);
2240    finish_setcond(s, cond, a0, true);
2241}
2242
2243static void tgen_negsetcondi(TCGContext *s, TCGType type, TCGCond cond,
2244                             TCGReg a0, TCGReg a1, tcg_target_long a2)
2245{
2246    cond = tgen_cmpi(s, cond, a1, a2);
2247    finish_setcond(s, cond, a0, true);
2248}
2249
2250static const TCGOutOpSetcond outop_negsetcond = {
2251    .base.static_constraint = C_O1_I2(r, r, rIN),
2252    .out_rrr = tgen_negsetcond,
2253    .out_rri = tgen_negsetcondi,
2254};
2255
2256static void tgen_movcond(TCGContext *s, TCGType type, TCGCond cond,
2257                         TCGReg ret, TCGReg c1, TCGArg c2, bool const_c2,
2258                         TCGArg vt, bool const_vt, TCGArg vf, bool consf_vf)
2259{
2260    cond = tcg_out_cmp(s, cond, c1, c2, const_c2);
2261    tcg_out_dat_rIK(s, tcg_cond_to_arm_cond[cond], ARITH_MOV, ARITH_MVN,
2262                    ret, 0, vt, const_vt);
2263}
2264
2265static const TCGOutOpMovcond outop_movcond = {
2266    .base.static_constraint = C_O1_I4(r, r, rIN, rIK, 0),
2267    .out = tgen_movcond,
2268};
2269
2270static void tgen_brcond2(TCGContext *s, TCGCond cond, TCGReg al, TCGReg ah,
2271                         TCGArg bl, bool const_bl, TCGArg bh, bool const_bh,
2272                         TCGLabel *l)
2273{
2274    cond = tcg_out_cmp2(s, cond, al, ah, bl, const_bl, bh, const_bh);
2275    tcg_out_goto_label(s, tcg_cond_to_arm_cond[cond], l);
2276}
2277
2278static const TCGOutOpBrcond2 outop_brcond2 = {
2279    .base.static_constraint = C_O0_I4(r, r, rI, rI),
2280    .out = tgen_brcond2,
2281};
2282
2283static void tgen_setcond2(TCGContext *s, TCGCond cond, TCGReg ret,
2284                          TCGReg al, TCGReg ah,
2285                          TCGArg bl, bool const_bl,
2286                          TCGArg bh, bool const_bh)
2287{
2288    cond = tcg_out_cmp2(s, cond, al, ah, bl, const_bl, bh, const_bh);
2289    finish_setcond(s, cond, ret, false);
2290}
2291
2292static const TCGOutOpSetcond2 outop_setcond2 = {
2293    .base.static_constraint = C_O1_I4(r, r, r, rI, rI),
2294    .out = tgen_setcond2,
2295};
2296
2297static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type,
2298                       const TCGArg args[TCG_MAX_OP_ARGS],
2299                       const int const_args[TCG_MAX_OP_ARGS])
2300{
2301    TCGArg a0, a1, a2, a3, a4, a5;
2302
2303    switch (opc) {
2304    case INDEX_op_goto_ptr:
2305        tcg_out_b_reg(s, COND_AL, args[0]);
2306        break;
2307    case INDEX_op_br:
2308        tcg_out_goto_label(s, COND_AL, arg_label(args[0]));
2309        break;
2310
2311    case INDEX_op_ld8u_i32:
2312        tcg_out_ld8u(s, COND_AL, args[0], args[1], args[2]);
2313        break;
2314    case INDEX_op_ld8s_i32:
2315        tcg_out_ld8s(s, COND_AL, args[0], args[1], args[2]);
2316        break;
2317    case INDEX_op_ld16u_i32:
2318        tcg_out_ld16u(s, COND_AL, args[0], args[1], args[2]);
2319        break;
2320    case INDEX_op_ld16s_i32:
2321        tcg_out_ld16s(s, COND_AL, args[0], args[1], args[2]);
2322        break;
2323    case INDEX_op_ld_i32:
2324        tcg_out_ld32u(s, COND_AL, args[0], args[1], args[2]);
2325        break;
2326    case INDEX_op_st8_i32:
2327        tcg_out_st8(s, COND_AL, args[0], args[1], args[2]);
2328        break;
2329    case INDEX_op_st16_i32:
2330        tcg_out_st16(s, COND_AL, args[0], args[1], args[2]);
2331        break;
2332    case INDEX_op_st_i32:
2333        tcg_out_st32(s, COND_AL, args[0], args[1], args[2]);
2334        break;
2335
2336    case INDEX_op_add2_i32:
2337        a0 = args[0], a1 = args[1], a2 = args[2];
2338        a3 = args[3], a4 = args[4], a5 = args[5];
2339        if (a0 == a3 || (a0 == a5 && !const_args[5])) {
2340            a0 = TCG_REG_TMP;
2341        }
2342        tcg_out_dat_rIN(s, COND_AL, ARITH_ADD | TO_CPSR, ARITH_SUB | TO_CPSR,
2343                        a0, a2, a4, const_args[4]);
2344        tcg_out_dat_rIK(s, COND_AL, ARITH_ADC, ARITH_SBC,
2345                        a1, a3, a5, const_args[5]);
2346        tcg_out_mov_reg(s, COND_AL, args[0], a0);
2347        break;
2348    case INDEX_op_sub2_i32:
2349        a0 = args[0], a1 = args[1], a2 = args[2];
2350        a3 = args[3], a4 = args[4], a5 = args[5];
2351        if ((a0 == a3 && !const_args[3]) || (a0 == a5 && !const_args[5])) {
2352            a0 = TCG_REG_TMP;
2353        }
2354        if (const_args[2]) {
2355            if (const_args[4]) {
2356                tcg_out_movi32(s, COND_AL, a0, a4);
2357                a4 = a0;
2358            }
2359            tcg_out_dat_rI(s, COND_AL, ARITH_RSB | TO_CPSR, a0, a4, a2, 1);
2360        } else {
2361            tcg_out_dat_rIN(s, COND_AL, ARITH_SUB | TO_CPSR,
2362                            ARITH_ADD | TO_CPSR, a0, a2, a4, const_args[4]);
2363        }
2364        if (const_args[3]) {
2365            if (const_args[5]) {
2366                tcg_out_movi32(s, COND_AL, a1, a5);
2367                a5 = a1;
2368            }
2369            tcg_out_dat_rI(s, COND_AL, ARITH_RSC, a1, a5, a3, 1);
2370        } else {
2371            tcg_out_dat_rIK(s, COND_AL, ARITH_SBC, ARITH_ADC,
2372                            a1, a3, a5, const_args[5]);
2373        }
2374        tcg_out_mov_reg(s, COND_AL, args[0], a0);
2375        break;
2376
2377    case INDEX_op_qemu_ld_i32:
2378        tcg_out_qemu_ld(s, args[0], -1, args[1], args[2], TCG_TYPE_I32);
2379        break;
2380    case INDEX_op_qemu_ld_i64:
2381        tcg_out_qemu_ld(s, args[0], args[1], args[2], args[3], TCG_TYPE_I64);
2382        break;
2383
2384    case INDEX_op_qemu_st_i32:
2385        tcg_out_qemu_st(s, args[0], -1, args[1], args[2], TCG_TYPE_I32);
2386        break;
2387    case INDEX_op_qemu_st_i64:
2388        tcg_out_qemu_st(s, args[0], args[1], args[2], args[3], TCG_TYPE_I64);
2389        break;
2390
2391    case INDEX_op_deposit_i32:
2392        tcg_out_deposit(s, COND_AL, args[0], args[2],
2393                        args[3], args[4], const_args[2]);
2394        break;
2395    case INDEX_op_extract_i32:
2396        tcg_out_extract(s, COND_AL, args[0], args[1], args[2], args[3]);
2397        break;
2398    case INDEX_op_sextract_i32:
2399        tcg_out_sextract(s, COND_AL, args[0], args[1], args[2], args[3]);
2400        break;
2401    case INDEX_op_extract2_i32:
2402        /* ??? These optimization vs zero should be generic.  */
2403        /* ??? But we can't substitute 2 for 1 in the opcode stream yet.  */
2404        if (const_args[1]) {
2405            if (const_args[2]) {
2406                tcg_out_movi(s, TCG_TYPE_REG, args[0], 0);
2407            } else {
2408                tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0,
2409                                args[2], SHIFT_IMM_LSL(32 - args[3]));
2410            }
2411        } else if (const_args[2]) {
2412            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0,
2413                            args[1], SHIFT_IMM_LSR(args[3]));
2414        } else {
2415            /* We can do extract2 in 2 insns, vs the 3 required otherwise.  */
2416            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_TMP, 0,
2417                            args[2], SHIFT_IMM_LSL(32 - args[3]));
2418            tcg_out_dat_reg(s, COND_AL, ARITH_ORR, args[0], TCG_REG_TMP,
2419                            args[1], SHIFT_IMM_LSR(args[3]));
2420        }
2421        break;
2422
2423    case INDEX_op_mb:
2424        tcg_out_mb(s, args[0]);
2425        break;
2426
2427    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2428    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2429    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2430    default:
2431        g_assert_not_reached();
2432    }
2433}
2434
2435static TCGConstraintSetIndex
2436tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
2437{
2438    switch (op) {
2439    case INDEX_op_goto_ptr:
2440        return C_O0_I1(r);
2441
2442    case INDEX_op_ld8u_i32:
2443    case INDEX_op_ld8s_i32:
2444    case INDEX_op_ld16u_i32:
2445    case INDEX_op_ld16s_i32:
2446    case INDEX_op_ld_i32:
2447    case INDEX_op_extract_i32:
2448    case INDEX_op_sextract_i32:
2449        return C_O1_I1(r, r);
2450
2451    case INDEX_op_st8_i32:
2452    case INDEX_op_st16_i32:
2453    case INDEX_op_st_i32:
2454        return C_O0_I2(r, r);
2455
2456    case INDEX_op_deposit_i32:
2457        return C_O1_I2(r, 0, rZ);
2458    case INDEX_op_extract2_i32:
2459        return C_O1_I2(r, rZ, rZ);
2460    case INDEX_op_add2_i32:
2461        return C_O2_I4(r, r, r, r, rIN, rIK);
2462    case INDEX_op_sub2_i32:
2463        return C_O2_I4(r, r, rI, rI, rIN, rIK);
2464    case INDEX_op_qemu_ld_i32:
2465        return C_O1_I1(r, q);
2466    case INDEX_op_qemu_ld_i64:
2467        return C_O2_I1(e, p, q);
2468    case INDEX_op_qemu_st_i32:
2469        return C_O0_I2(q, q);
2470    case INDEX_op_qemu_st_i64:
2471        return C_O0_I3(Q, p, q);
2472
2473    case INDEX_op_st_vec:
2474        return C_O0_I2(w, r);
2475    case INDEX_op_ld_vec:
2476    case INDEX_op_dupm_vec:
2477        return C_O1_I1(w, r);
2478    case INDEX_op_dup_vec:
2479        return C_O1_I1(w, wr);
2480    case INDEX_op_abs_vec:
2481    case INDEX_op_neg_vec:
2482    case INDEX_op_not_vec:
2483    case INDEX_op_shli_vec:
2484    case INDEX_op_shri_vec:
2485    case INDEX_op_sari_vec:
2486        return C_O1_I1(w, w);
2487    case INDEX_op_dup2_vec:
2488    case INDEX_op_add_vec:
2489    case INDEX_op_mul_vec:
2490    case INDEX_op_smax_vec:
2491    case INDEX_op_smin_vec:
2492    case INDEX_op_ssadd_vec:
2493    case INDEX_op_sssub_vec:
2494    case INDEX_op_sub_vec:
2495    case INDEX_op_umax_vec:
2496    case INDEX_op_umin_vec:
2497    case INDEX_op_usadd_vec:
2498    case INDEX_op_ussub_vec:
2499    case INDEX_op_xor_vec:
2500    case INDEX_op_arm_sshl_vec:
2501    case INDEX_op_arm_ushl_vec:
2502        return C_O1_I2(w, w, w);
2503    case INDEX_op_arm_sli_vec:
2504        return C_O1_I2(w, 0, w);
2505    case INDEX_op_or_vec:
2506    case INDEX_op_andc_vec:
2507        return C_O1_I2(w, w, wO);
2508    case INDEX_op_and_vec:
2509    case INDEX_op_orc_vec:
2510        return C_O1_I2(w, w, wV);
2511    case INDEX_op_cmp_vec:
2512        return C_O1_I2(w, w, wZ);
2513    case INDEX_op_bitsel_vec:
2514        return C_O1_I3(w, w, w, w);
2515    default:
2516        return C_NotImplemented;
2517    }
2518}
2519
2520static void tcg_target_init(TCGContext *s)
2521{
2522    /*
2523     * Only probe for the platform and capabilities if we haven't already
2524     * determined maximum values at compile time.
2525     */
2526#if !defined(use_idiv_instructions) || !defined(use_neon_instructions)
2527    {
2528        unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2529#ifndef use_idiv_instructions
2530        use_idiv_instructions = (hwcap & HWCAP_ARM_IDIVA) != 0;
2531#endif
2532#ifndef use_neon_instructions
2533        use_neon_instructions = (hwcap & HWCAP_ARM_NEON) != 0;
2534#endif
2535    }
2536#endif
2537
2538    if (__ARM_ARCH < 7) {
2539        const char *pl = (const char *)qemu_getauxval(AT_PLATFORM);
2540        if (pl != NULL && pl[0] == 'v' && pl[1] >= '4' && pl[1] <= '9') {
2541            arm_arch = pl[1] - '0';
2542        }
2543
2544        if (arm_arch < 6) {
2545            error_report("TCG: ARMv%d is unsupported; exiting", arm_arch);
2546            exit(EXIT_FAILURE);
2547        }
2548    }
2549
2550    tcg_target_available_regs[TCG_TYPE_I32] = ALL_GENERAL_REGS;
2551
2552    tcg_target_call_clobber_regs = 0;
2553    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
2554    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
2555    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
2556    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
2557    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
2558    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
2559
2560    if (use_neon_instructions) {
2561        tcg_target_available_regs[TCG_TYPE_V64]  = ALL_VECTOR_REGS;
2562        tcg_target_available_regs[TCG_TYPE_V128] = ALL_VECTOR_REGS;
2563
2564        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q0);
2565        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q1);
2566        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q2);
2567        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q3);
2568        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q8);
2569        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q9);
2570        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q10);
2571        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q11);
2572        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q12);
2573        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q13);
2574        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q14);
2575        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q15);
2576    }
2577
2578    s->reserved_regs = 0;
2579    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
2580    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP);
2581    tcg_regset_set_reg(s->reserved_regs, TCG_REG_PC);
2582    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP);
2583}
2584
2585static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
2586                       TCGReg arg1, intptr_t arg2)
2587{
2588    switch (type) {
2589    case TCG_TYPE_I32:
2590        tcg_out_ld32u(s, COND_AL, arg, arg1, arg2);
2591        return;
2592    case TCG_TYPE_V64:
2593        /* regs 1; size 8; align 8 */
2594        tcg_out_vldst(s, INSN_VLD1 | 0x7d0, arg, arg1, arg2);
2595        return;
2596    case TCG_TYPE_V128:
2597        /*
2598         * We have only 8-byte alignment for the stack per the ABI.
2599         * Rather than dynamically re-align the stack, it's easier
2600         * to simply not request alignment beyond that.  So:
2601         * regs 2; size 8; align 8
2602         */
2603        tcg_out_vldst(s, INSN_VLD1 | 0xad0, arg, arg1, arg2);
2604        return;
2605    default:
2606        g_assert_not_reached();
2607    }
2608}
2609
2610static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
2611                       TCGReg arg1, intptr_t arg2)
2612{
2613    switch (type) {
2614    case TCG_TYPE_I32:
2615        tcg_out_st32(s, COND_AL, arg, arg1, arg2);
2616        return;
2617    case TCG_TYPE_V64:
2618        /* regs 1; size 8; align 8 */
2619        tcg_out_vldst(s, INSN_VST1 | 0x7d0, arg, arg1, arg2);
2620        return;
2621    case TCG_TYPE_V128:
2622        /* See tcg_out_ld re alignment: regs 2; size 8; align 8 */
2623        tcg_out_vldst(s, INSN_VST1 | 0xad0, arg, arg1, arg2);
2624        return;
2625    default:
2626        g_assert_not_reached();
2627    }
2628}
2629
2630static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
2631                        TCGReg base, intptr_t ofs)
2632{
2633    return false;
2634}
2635
2636static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
2637{
2638    if (ret == arg) {
2639        return true;
2640    }
2641    switch (type) {
2642    case TCG_TYPE_I32:
2643        if (ret < TCG_REG_Q0 && arg < TCG_REG_Q0) {
2644            tcg_out_mov_reg(s, COND_AL, ret, arg);
2645            return true;
2646        }
2647        return false;
2648
2649    case TCG_TYPE_V64:
2650    case TCG_TYPE_V128:
2651        /* "VMOV D,N" is an alias for "VORR D,N,N". */
2652        tcg_out_vreg3(s, INSN_VORR, type - TCG_TYPE_V64, 0, ret, arg, arg);
2653        return true;
2654
2655    default:
2656        g_assert_not_reached();
2657    }
2658}
2659
2660static void tcg_out_movi(TCGContext *s, TCGType type,
2661                         TCGReg ret, tcg_target_long arg)
2662{
2663    tcg_debug_assert(type == TCG_TYPE_I32);
2664    tcg_debug_assert(ret < TCG_REG_Q0);
2665    tcg_out_movi32(s, COND_AL, ret, arg);
2666}
2667
2668static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
2669{
2670    return false;
2671}
2672
2673static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
2674                             tcg_target_long imm)
2675{
2676    int enc, opc = ARITH_ADD;
2677
2678    /* All of the easiest immediates to encode are positive. */
2679    if (imm < 0) {
2680        imm = -imm;
2681        opc = ARITH_SUB;
2682    }
2683    enc = encode_imm(imm);
2684    if (enc >= 0) {
2685        tcg_out_dat_imm(s, COND_AL, opc, rd, rs, enc);
2686    } else {
2687        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, imm);
2688        tcg_out_dat_reg(s, COND_AL, opc, rd, rs,
2689                        TCG_REG_TMP, SHIFT_IMM_LSL(0));
2690    }
2691}
2692
2693/* Type is always V128, with I64 elements.  */
2694static void tcg_out_dup2_vec(TCGContext *s, TCGReg rd, TCGReg rl, TCGReg rh)
2695{
2696    /* Move high element into place first. */
2697    /* VMOV Dd+1, Ds */
2698    tcg_out_vreg3(s, INSN_VORR | (1 << 12), 0, 0, rd, rh, rh);
2699    /* Move low element into place; tcg_out_mov will check for nop. */
2700    tcg_out_mov(s, TCG_TYPE_V64, rd, rl);
2701}
2702
2703static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
2704                            TCGReg rd, TCGReg rs)
2705{
2706    int q = type - TCG_TYPE_V64;
2707
2708    if (vece == MO_64) {
2709        if (type == TCG_TYPE_V128) {
2710            tcg_out_dup2_vec(s, rd, rs, rs);
2711        } else {
2712            tcg_out_mov(s, TCG_TYPE_V64, rd, rs);
2713        }
2714    } else if (rs < TCG_REG_Q0) {
2715        int b = (vece == MO_8);
2716        int e = (vece == MO_16);
2717        tcg_out32(s, INSN_VDUP_G | (b << 22) | (q << 21) | (e << 5) |
2718                  encode_vn(rd) | (rs << 12));
2719    } else {
2720        int imm4 = 1 << vece;
2721        tcg_out32(s, INSN_VDUP_S | (imm4 << 16) | (q << 6) |
2722                  encode_vd(rd) | encode_vm(rs));
2723    }
2724    return true;
2725}
2726
2727static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
2728                             TCGReg rd, TCGReg base, intptr_t offset)
2729{
2730    if (vece == MO_64) {
2731        tcg_out_ld(s, TCG_TYPE_V64, rd, base, offset);
2732        if (type == TCG_TYPE_V128) {
2733            tcg_out_dup2_vec(s, rd, rd, rd);
2734        }
2735    } else {
2736        int q = type - TCG_TYPE_V64;
2737        tcg_out_vldst(s, INSN_VLD1R | (vece << 6) | (q << 5),
2738                      rd, base, offset);
2739    }
2740    return true;
2741}
2742
2743static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
2744                             TCGReg rd, int64_t v64)
2745{
2746    int q = type - TCG_TYPE_V64;
2747    int cmode, imm8, i;
2748
2749    /* Test all bytes equal first.  */
2750    if (vece == MO_8) {
2751        tcg_out_vmovi(s, rd, q, 0, 0xe, v64);
2752        return;
2753    }
2754
2755    /*
2756     * Test all bytes 0x00 or 0xff second.  This can match cases that
2757     * might otherwise take 2 or 3 insns for MO_16 or MO_32 below.
2758     */
2759    for (i = imm8 = 0; i < 8; i++) {
2760        uint8_t byte = v64 >> (i * 8);
2761        if (byte == 0xff) {
2762            imm8 |= 1 << i;
2763        } else if (byte != 0) {
2764            goto fail_bytes;
2765        }
2766    }
2767    tcg_out_vmovi(s, rd, q, 1, 0xe, imm8);
2768    return;
2769 fail_bytes:
2770
2771    /*
2772     * Tests for various replications.  For each element width, if we
2773     * cannot find an expansion there's no point checking a larger
2774     * width because we already know by replication it cannot match.
2775     */
2776    if (vece == MO_16) {
2777        uint16_t v16 = v64;
2778
2779        if (is_shimm16(v16, &cmode, &imm8)) {
2780            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2781            return;
2782        }
2783        if (is_shimm16(~v16, &cmode, &imm8)) {
2784            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2785            return;
2786        }
2787
2788        /*
2789         * Otherwise, all remaining constants can be loaded in two insns:
2790         * rd = v16 & 0xff, rd |= v16 & 0xff00.
2791         */
2792        tcg_out_vmovi(s, rd, q, 0, 0x8, v16 & 0xff);
2793        tcg_out_vmovi(s, rd, q, 0, 0xb, v16 >> 8);   /* VORRI */
2794        return;
2795    }
2796
2797    if (vece == MO_32) {
2798        uint32_t v32 = v64;
2799
2800        if (is_shimm32(v32, &cmode, &imm8) ||
2801            is_soimm32(v32, &cmode, &imm8)) {
2802            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2803            return;
2804        }
2805        if (is_shimm32(~v32, &cmode, &imm8) ||
2806            is_soimm32(~v32, &cmode, &imm8)) {
2807            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2808            return;
2809        }
2810
2811        /*
2812         * Restrict the set of constants to those we can load with
2813         * two instructions.  Others we load from the pool.
2814         */
2815        i = is_shimm32_pair(v32, &cmode, &imm8);
2816        if (i) {
2817            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2818            tcg_out_vmovi(s, rd, q, 0, i | 1, extract32(v32, i * 4, 8));
2819            return;
2820        }
2821        i = is_shimm32_pair(~v32, &cmode, &imm8);
2822        if (i) {
2823            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2824            tcg_out_vmovi(s, rd, q, 1, i | 1, extract32(~v32, i * 4, 8));
2825            return;
2826        }
2827    }
2828
2829    /*
2830     * As a last resort, load from the constant pool.
2831     */
2832    if (!q || vece == MO_64) {
2833        new_pool_l2(s, R_ARM_PC11, s->code_ptr, 0, v64, v64 >> 32);
2834        /* VLDR Dd, [pc + offset] */
2835        tcg_out32(s, INSN_VLDR_D | encode_vd(rd) | (0xf << 16));
2836        if (q) {
2837            tcg_out_dup2_vec(s, rd, rd, rd);
2838        }
2839    } else {
2840        new_pool_label(s, (uint32_t)v64, R_ARM_PC8, s->code_ptr, 0);
2841        /* add tmp, pc, offset */
2842        tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_TMP, TCG_REG_PC, 0);
2843        tcg_out_dupm_vec(s, type, MO_32, rd, TCG_REG_TMP, 0);
2844    }
2845}
2846
2847static const ARMInsn vec_cmp_insn[16] = {
2848    [TCG_COND_EQ] = INSN_VCEQ,
2849    [TCG_COND_GT] = INSN_VCGT,
2850    [TCG_COND_GE] = INSN_VCGE,
2851    [TCG_COND_GTU] = INSN_VCGT_U,
2852    [TCG_COND_GEU] = INSN_VCGE_U,
2853};
2854
2855static const ARMInsn vec_cmp0_insn[16] = {
2856    [TCG_COND_EQ] = INSN_VCEQ0,
2857    [TCG_COND_GT] = INSN_VCGT0,
2858    [TCG_COND_GE] = INSN_VCGE0,
2859    [TCG_COND_LT] = INSN_VCLT0,
2860    [TCG_COND_LE] = INSN_VCLE0,
2861};
2862
2863static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2864                           unsigned vecl, unsigned vece,
2865                           const TCGArg args[TCG_MAX_OP_ARGS],
2866                           const int const_args[TCG_MAX_OP_ARGS])
2867{
2868    TCGType type = vecl + TCG_TYPE_V64;
2869    unsigned q = vecl;
2870    TCGArg a0, a1, a2, a3;
2871    int cmode, imm8;
2872
2873    a0 = args[0];
2874    a1 = args[1];
2875    a2 = args[2];
2876
2877    switch (opc) {
2878    case INDEX_op_ld_vec:
2879        tcg_out_ld(s, type, a0, a1, a2);
2880        return;
2881    case INDEX_op_st_vec:
2882        tcg_out_st(s, type, a0, a1, a2);
2883        return;
2884    case INDEX_op_dupm_vec:
2885        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2886        return;
2887    case INDEX_op_dup2_vec:
2888        tcg_out_dup2_vec(s, a0, a1, a2);
2889        return;
2890    case INDEX_op_abs_vec:
2891        tcg_out_vreg2(s, INSN_VABS, q, vece, a0, a1);
2892        return;
2893    case INDEX_op_neg_vec:
2894        tcg_out_vreg2(s, INSN_VNEG, q, vece, a0, a1);
2895        return;
2896    case INDEX_op_not_vec:
2897        tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a1);
2898        return;
2899    case INDEX_op_add_vec:
2900        tcg_out_vreg3(s, INSN_VADD, q, vece, a0, a1, a2);
2901        return;
2902    case INDEX_op_mul_vec:
2903        tcg_out_vreg3(s, INSN_VMUL, q, vece, a0, a1, a2);
2904        return;
2905    case INDEX_op_smax_vec:
2906        tcg_out_vreg3(s, INSN_VMAX, q, vece, a0, a1, a2);
2907        return;
2908    case INDEX_op_smin_vec:
2909        tcg_out_vreg3(s, INSN_VMIN, q, vece, a0, a1, a2);
2910        return;
2911    case INDEX_op_sub_vec:
2912        tcg_out_vreg3(s, INSN_VSUB, q, vece, a0, a1, a2);
2913        return;
2914    case INDEX_op_ssadd_vec:
2915        tcg_out_vreg3(s, INSN_VQADD, q, vece, a0, a1, a2);
2916        return;
2917    case INDEX_op_sssub_vec:
2918        tcg_out_vreg3(s, INSN_VQSUB, q, vece, a0, a1, a2);
2919        return;
2920    case INDEX_op_umax_vec:
2921        tcg_out_vreg3(s, INSN_VMAX_U, q, vece, a0, a1, a2);
2922        return;
2923    case INDEX_op_umin_vec:
2924        tcg_out_vreg3(s, INSN_VMIN_U, q, vece, a0, a1, a2);
2925        return;
2926    case INDEX_op_usadd_vec:
2927        tcg_out_vreg3(s, INSN_VQADD_U, q, vece, a0, a1, a2);
2928        return;
2929    case INDEX_op_ussub_vec:
2930        tcg_out_vreg3(s, INSN_VQSUB_U, q, vece, a0, a1, a2);
2931        return;
2932    case INDEX_op_xor_vec:
2933        tcg_out_vreg3(s, INSN_VEOR, q, 0, a0, a1, a2);
2934        return;
2935    case INDEX_op_arm_sshl_vec:
2936        /*
2937         * Note that Vm is the data and Vn is the shift count,
2938         * therefore the arguments appear reversed.
2939         */
2940        tcg_out_vreg3(s, INSN_VSHL_S, q, vece, a0, a2, a1);
2941        return;
2942    case INDEX_op_arm_ushl_vec:
2943        /* See above. */
2944        tcg_out_vreg3(s, INSN_VSHL_U, q, vece, a0, a2, a1);
2945        return;
2946    case INDEX_op_shli_vec:
2947        tcg_out_vshifti(s, INSN_VSHLI, q, a0, a1, a2 + (8 << vece));
2948        return;
2949    case INDEX_op_shri_vec:
2950        tcg_out_vshifti(s, INSN_VSHRI, q, a0, a1, (16 << vece) - a2);
2951        return;
2952    case INDEX_op_sari_vec:
2953        tcg_out_vshifti(s, INSN_VSARI, q, a0, a1, (16 << vece) - a2);
2954        return;
2955    case INDEX_op_arm_sli_vec:
2956        tcg_out_vshifti(s, INSN_VSLI, q, a0, a2, args[3] + (8 << vece));
2957        return;
2958
2959    case INDEX_op_andc_vec:
2960        if (!const_args[2]) {
2961            tcg_out_vreg3(s, INSN_VBIC, q, 0, a0, a1, a2);
2962            return;
2963        }
2964        a2 = ~a2;
2965        /* fall through */
2966    case INDEX_op_and_vec:
2967        if (const_args[2]) {
2968            is_shimm1632(~a2, &cmode, &imm8);
2969            if (a0 == a1) {
2970                tcg_out_vmovi(s, a0, q, 1, cmode | 1, imm8); /* VBICI */
2971                return;
2972            }
2973            tcg_out_vmovi(s, a0, q, 1, cmode, imm8); /* VMVNI */
2974            a2 = a0;
2975        }
2976        tcg_out_vreg3(s, INSN_VAND, q, 0, a0, a1, a2);
2977        return;
2978
2979    case INDEX_op_orc_vec:
2980        if (!const_args[2]) {
2981            tcg_out_vreg3(s, INSN_VORN, q, 0, a0, a1, a2);
2982            return;
2983        }
2984        a2 = ~a2;
2985        /* fall through */
2986    case INDEX_op_or_vec:
2987        if (const_args[2]) {
2988            is_shimm1632(a2, &cmode, &imm8);
2989            if (a0 == a1) {
2990                tcg_out_vmovi(s, a0, q, 0, cmode | 1, imm8); /* VORRI */
2991                return;
2992            }
2993            tcg_out_vmovi(s, a0, q, 0, cmode, imm8); /* VMOVI */
2994            a2 = a0;
2995        }
2996        tcg_out_vreg3(s, INSN_VORR, q, 0, a0, a1, a2);
2997        return;
2998
2999    case INDEX_op_cmp_vec:
3000        {
3001            TCGCond cond = args[3];
3002            ARMInsn insn;
3003
3004            switch (cond) {
3005            case TCG_COND_NE:
3006                if (const_args[2]) {
3007                    tcg_out_vreg3(s, INSN_VTST, q, vece, a0, a1, a1);
3008                } else {
3009                    tcg_out_vreg3(s, INSN_VCEQ, q, vece, a0, a1, a2);
3010                    tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a0);
3011                }
3012                break;
3013
3014            case TCG_COND_TSTNE:
3015            case TCG_COND_TSTEQ:
3016                if (const_args[2]) {
3017                    /* (x & 0) == 0 */
3018                    tcg_out_dupi_vec(s, type, MO_8, a0,
3019                                     -(cond == TCG_COND_TSTEQ));
3020                    break;
3021                }
3022                tcg_out_vreg3(s, INSN_VTST, q, vece, a0, a1, a2);
3023                if (cond == TCG_COND_TSTEQ) {
3024                    tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a0);
3025                }
3026                break;
3027
3028            default:
3029                if (const_args[2]) {
3030                    insn = vec_cmp0_insn[cond];
3031                    if (insn) {
3032                        tcg_out_vreg2(s, insn, q, vece, a0, a1);
3033                        return;
3034                    }
3035                    tcg_out_dupi_vec(s, type, MO_8, TCG_VEC_TMP, 0);
3036                    a2 = TCG_VEC_TMP;
3037                }
3038                insn = vec_cmp_insn[cond];
3039                if (insn == 0) {
3040                    TCGArg t;
3041                    t = a1, a1 = a2, a2 = t;
3042                    cond = tcg_swap_cond(cond);
3043                    insn = vec_cmp_insn[cond];
3044                    tcg_debug_assert(insn != 0);
3045                }
3046                tcg_out_vreg3(s, insn, q, vece, a0, a1, a2);
3047                break;
3048            }
3049        }
3050        return;
3051
3052    case INDEX_op_bitsel_vec:
3053        a3 = args[3];
3054        if (a0 == a3) {
3055            tcg_out_vreg3(s, INSN_VBIT, q, 0, a0, a2, a1);
3056        } else if (a0 == a2) {
3057            tcg_out_vreg3(s, INSN_VBIF, q, 0, a0, a3, a1);
3058        } else {
3059            tcg_out_mov(s, type, a0, a1);
3060            tcg_out_vreg3(s, INSN_VBSL, q, 0, a0, a2, a3);
3061        }
3062        return;
3063
3064    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
3065    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
3066    default:
3067        g_assert_not_reached();
3068    }
3069}
3070
3071int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3072{
3073    switch (opc) {
3074    case INDEX_op_add_vec:
3075    case INDEX_op_sub_vec:
3076    case INDEX_op_and_vec:
3077    case INDEX_op_andc_vec:
3078    case INDEX_op_or_vec:
3079    case INDEX_op_orc_vec:
3080    case INDEX_op_xor_vec:
3081    case INDEX_op_not_vec:
3082    case INDEX_op_shli_vec:
3083    case INDEX_op_shri_vec:
3084    case INDEX_op_sari_vec:
3085    case INDEX_op_ssadd_vec:
3086    case INDEX_op_sssub_vec:
3087    case INDEX_op_usadd_vec:
3088    case INDEX_op_ussub_vec:
3089    case INDEX_op_bitsel_vec:
3090        return 1;
3091    case INDEX_op_abs_vec:
3092    case INDEX_op_cmp_vec:
3093    case INDEX_op_mul_vec:
3094    case INDEX_op_neg_vec:
3095    case INDEX_op_smax_vec:
3096    case INDEX_op_smin_vec:
3097    case INDEX_op_umax_vec:
3098    case INDEX_op_umin_vec:
3099        return vece < MO_64;
3100    case INDEX_op_shlv_vec:
3101    case INDEX_op_shrv_vec:
3102    case INDEX_op_sarv_vec:
3103    case INDEX_op_rotli_vec:
3104    case INDEX_op_rotlv_vec:
3105    case INDEX_op_rotrv_vec:
3106        return -1;
3107    default:
3108        return 0;
3109    }
3110}
3111
3112void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3113                       TCGArg a0, ...)
3114{
3115    va_list va;
3116    TCGv_vec v0, v1, v2, t1, t2, c1;
3117    TCGArg a2;
3118
3119    va_start(va, a0);
3120    v0 = temp_tcgv_vec(arg_temp(a0));
3121    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3122    a2 = va_arg(va, TCGArg);
3123    va_end(va);
3124
3125    switch (opc) {
3126    case INDEX_op_shlv_vec:
3127        /*
3128         * Merely propagate shlv_vec to arm_ushl_vec.
3129         * In this way we don't set TCG_TARGET_HAS_shv_vec
3130         * because everything is done via expansion.
3131         */
3132        v2 = temp_tcgv_vec(arg_temp(a2));
3133        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(v0),
3134                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3135        break;
3136
3137    case INDEX_op_shrv_vec:
3138    case INDEX_op_sarv_vec:
3139        /* Right shifts are negative left shifts for NEON.  */
3140        v2 = temp_tcgv_vec(arg_temp(a2));
3141        t1 = tcg_temp_new_vec(type);
3142        tcg_gen_neg_vec(vece, t1, v2);
3143        if (opc == INDEX_op_shrv_vec) {
3144            opc = INDEX_op_arm_ushl_vec;
3145        } else {
3146            opc = INDEX_op_arm_sshl_vec;
3147        }
3148        vec_gen_3(opc, type, vece, tcgv_vec_arg(v0),
3149                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3150        tcg_temp_free_vec(t1);
3151        break;
3152
3153    case INDEX_op_rotli_vec:
3154        t1 = tcg_temp_new_vec(type);
3155        tcg_gen_shri_vec(vece, t1, v1, -a2 & ((8 << vece) - 1));
3156        vec_gen_4(INDEX_op_arm_sli_vec, type, vece,
3157                  tcgv_vec_arg(v0), tcgv_vec_arg(t1), tcgv_vec_arg(v1), a2);
3158        tcg_temp_free_vec(t1);
3159        break;
3160
3161    case INDEX_op_rotlv_vec:
3162        v2 = temp_tcgv_vec(arg_temp(a2));
3163        t1 = tcg_temp_new_vec(type);
3164        c1 = tcg_constant_vec(type, vece, 8 << vece);
3165        tcg_gen_sub_vec(vece, t1, v2, c1);
3166        /* Right shifts are negative left shifts for NEON.  */
3167        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t1),
3168                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3169        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(v0),
3170                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3171        tcg_gen_or_vec(vece, v0, v0, t1);
3172        tcg_temp_free_vec(t1);
3173        break;
3174
3175    case INDEX_op_rotrv_vec:
3176        v2 = temp_tcgv_vec(arg_temp(a2));
3177        t1 = tcg_temp_new_vec(type);
3178        t2 = tcg_temp_new_vec(type);
3179        c1 = tcg_constant_vec(type, vece, 8 << vece);
3180        tcg_gen_neg_vec(vece, t1, v2);
3181        tcg_gen_sub_vec(vece, t2, c1, v2);
3182        /* Right shifts are negative left shifts for NEON.  */
3183        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t1),
3184                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3185        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t2),
3186                  tcgv_vec_arg(v1), tcgv_vec_arg(t2));
3187        tcg_gen_or_vec(vece, v0, t1, t2);
3188        tcg_temp_free_vec(t1);
3189        tcg_temp_free_vec(t2);
3190        break;
3191
3192    default:
3193        g_assert_not_reached();
3194    }
3195}
3196
3197static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3198{
3199    int i;
3200    for (i = 0; i < count; ++i) {
3201        p[i] = INSN_NOP;
3202    }
3203}
3204
3205/* Compute frame size via macros, to share between tcg_target_qemu_prologue
3206   and tcg_register_jit.  */
3207
3208#define PUSH_SIZE  ((11 - 4 + 1 + 1) * sizeof(tcg_target_long))
3209
3210#define FRAME_SIZE \
3211    ((PUSH_SIZE \
3212      + TCG_STATIC_CALL_ARGS_SIZE \
3213      + CPU_TEMP_BUF_NLONGS * sizeof(long) \
3214      + TCG_TARGET_STACK_ALIGN - 1) \
3215     & -TCG_TARGET_STACK_ALIGN)
3216
3217#define STACK_ADDEND  (FRAME_SIZE - PUSH_SIZE)
3218
3219static void tcg_target_qemu_prologue(TCGContext *s)
3220{
3221    /* Calling convention requires us to save r4-r11 and lr.  */
3222    /* stmdb sp!, { r4 - r11, lr } */
3223    tcg_out_ldstm(s, COND_AL, INSN_STMDB, TCG_REG_CALL_STACK,
3224                  (1 << TCG_REG_R4) | (1 << TCG_REG_R5) | (1 << TCG_REG_R6) |
3225                  (1 << TCG_REG_R7) | (1 << TCG_REG_R8) | (1 << TCG_REG_R9) |
3226                  (1 << TCG_REG_R10) | (1 << TCG_REG_R11) | (1 << TCG_REG_R14));
3227
3228    /* Reserve callee argument and tcg temp space.  */
3229    tcg_out_dat_rI(s, COND_AL, ARITH_SUB, TCG_REG_CALL_STACK,
3230                   TCG_REG_CALL_STACK, STACK_ADDEND, 1);
3231    tcg_set_frame(s, TCG_REG_CALL_STACK, TCG_STATIC_CALL_ARGS_SIZE,
3232                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3233
3234    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3235
3236    if (!tcg_use_softmmu && guest_base) {
3237        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_GUEST_BASE, guest_base);
3238        tcg_regset_set_reg(s->reserved_regs, TCG_REG_GUEST_BASE);
3239    }
3240
3241    tcg_out_b_reg(s, COND_AL, tcg_target_call_iarg_regs[1]);
3242
3243    /*
3244     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3245     * and fall through to the rest of the epilogue.
3246     */
3247    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3248    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, 0);
3249    tcg_out_epilogue(s);
3250}
3251
3252static void tcg_out_epilogue(TCGContext *s)
3253{
3254    /* Release local stack frame.  */
3255    tcg_out_dat_rI(s, COND_AL, ARITH_ADD, TCG_REG_CALL_STACK,
3256                   TCG_REG_CALL_STACK, STACK_ADDEND, 1);
3257
3258    /* ldmia sp!, { r4 - r11, pc } */
3259    tcg_out_ldstm(s, COND_AL, INSN_LDMIA, TCG_REG_CALL_STACK,
3260                  (1 << TCG_REG_R4) | (1 << TCG_REG_R5) | (1 << TCG_REG_R6) |
3261                  (1 << TCG_REG_R7) | (1 << TCG_REG_R8) | (1 << TCG_REG_R9) |
3262                  (1 << TCG_REG_R10) | (1 << TCG_REG_R11) | (1 << TCG_REG_PC));
3263}
3264
3265static void tcg_out_tb_start(TCGContext *s)
3266{
3267    /* nothing to do */
3268}
3269
3270typedef struct {
3271    DebugFrameHeader h;
3272    uint8_t fde_def_cfa[4];
3273    uint8_t fde_reg_ofs[18];
3274} DebugFrame;
3275
3276#define ELF_HOST_MACHINE EM_ARM
3277
3278/* We're expecting a 2 byte uleb128 encoded value.  */
3279QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3280
3281static const DebugFrame debug_frame = {
3282    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3283    .h.cie.id = -1,
3284    .h.cie.version = 1,
3285    .h.cie.code_align = 1,
3286    .h.cie.data_align = 0x7c,             /* sleb128 -4 */
3287    .h.cie.return_column = 14,
3288
3289    /* Total FDE size does not include the "len" member.  */
3290    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3291
3292    .fde_def_cfa = {
3293        12, 13,                         /* DW_CFA_def_cfa sp, ... */
3294        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3295        (FRAME_SIZE >> 7)
3296    },
3297    .fde_reg_ofs = {
3298        /* The following must match the stmdb in the prologue.  */
3299        0x8e, 1,                        /* DW_CFA_offset, lr, -4 */
3300        0x8b, 2,                        /* DW_CFA_offset, r11, -8 */
3301        0x8a, 3,                        /* DW_CFA_offset, r10, -12 */
3302        0x89, 4,                        /* DW_CFA_offset, r9, -16 */
3303        0x88, 5,                        /* DW_CFA_offset, r8, -20 */
3304        0x87, 6,                        /* DW_CFA_offset, r7, -24 */
3305        0x86, 7,                        /* DW_CFA_offset, r6, -28 */
3306        0x85, 8,                        /* DW_CFA_offset, r5, -32 */
3307        0x84, 9,                        /* DW_CFA_offset, r4, -36 */
3308    }
3309};
3310
3311void tcg_register_jit(const void *buf, size_t buf_size)
3312{
3313    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3314}
3315