xref: /openbmc/qemu/tcg/arm/tcg-target.c.inc (revision f101c9fe)
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#include "../tcg-pool.c.inc"
27
28int arm_arch = __ARM_ARCH;
29
30#ifndef use_idiv_instructions
31bool use_idiv_instructions;
32#endif
33#ifndef use_neon_instructions
34bool use_neon_instructions;
35#endif
36
37/* ??? Ought to think about changing CONFIG_SOFTMMU to always defined.  */
38#ifdef CONFIG_SOFTMMU
39# define USING_SOFTMMU 1
40#else
41# define USING_SOFTMMU 0
42#endif
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};
88static const int tcg_target_call_oarg_regs[2] = {
89    TCG_REG_R0, TCG_REG_R1
90};
91
92#define TCG_REG_TMP  TCG_REG_R12
93#define TCG_VEC_TMP  TCG_REG_Q15
94
95enum arm_cond_code_e {
96    COND_EQ = 0x0,
97    COND_NE = 0x1,
98    COND_CS = 0x2,	/* Unsigned greater or equal */
99    COND_CC = 0x3,	/* Unsigned less than */
100    COND_MI = 0x4,	/* Negative */
101    COND_PL = 0x5,	/* Zero or greater */
102    COND_VS = 0x6,	/* Overflow */
103    COND_VC = 0x7,	/* No overflow */
104    COND_HI = 0x8,	/* Unsigned greater than */
105    COND_LS = 0x9,	/* Unsigned less or equal */
106    COND_GE = 0xa,
107    COND_LT = 0xb,
108    COND_GT = 0xc,
109    COND_LE = 0xd,
110    COND_AL = 0xe,
111};
112
113#define TO_CPSR (1 << 20)
114
115#define SHIFT_IMM_LSL(im)	(((im) << 7) | 0x00)
116#define SHIFT_IMM_LSR(im)	(((im) << 7) | 0x20)
117#define SHIFT_IMM_ASR(im)	(((im) << 7) | 0x40)
118#define SHIFT_IMM_ROR(im)	(((im) << 7) | 0x60)
119#define SHIFT_REG_LSL(rs)	(((rs) << 8) | 0x10)
120#define SHIFT_REG_LSR(rs)	(((rs) << 8) | 0x30)
121#define SHIFT_REG_ASR(rs)	(((rs) << 8) | 0x50)
122#define SHIFT_REG_ROR(rs)	(((rs) << 8) | 0x70)
123
124typedef enum {
125    ARITH_AND = 0x0 << 21,
126    ARITH_EOR = 0x1 << 21,
127    ARITH_SUB = 0x2 << 21,
128    ARITH_RSB = 0x3 << 21,
129    ARITH_ADD = 0x4 << 21,
130    ARITH_ADC = 0x5 << 21,
131    ARITH_SBC = 0x6 << 21,
132    ARITH_RSC = 0x7 << 21,
133    ARITH_TST = 0x8 << 21 | TO_CPSR,
134    ARITH_CMP = 0xa << 21 | TO_CPSR,
135    ARITH_CMN = 0xb << 21 | TO_CPSR,
136    ARITH_ORR = 0xc << 21,
137    ARITH_MOV = 0xd << 21,
138    ARITH_BIC = 0xe << 21,
139    ARITH_MVN = 0xf << 21,
140
141    INSN_CLZ       = 0x016f0f10,
142    INSN_RBIT      = 0x06ff0f30,
143
144    INSN_LDR_IMM   = 0x04100000,
145    INSN_LDR_REG   = 0x06100000,
146    INSN_STR_IMM   = 0x04000000,
147    INSN_STR_REG   = 0x06000000,
148
149    INSN_LDRH_IMM  = 0x005000b0,
150    INSN_LDRH_REG  = 0x001000b0,
151    INSN_LDRSH_IMM = 0x005000f0,
152    INSN_LDRSH_REG = 0x001000f0,
153    INSN_STRH_IMM  = 0x004000b0,
154    INSN_STRH_REG  = 0x000000b0,
155
156    INSN_LDRB_IMM  = 0x04500000,
157    INSN_LDRB_REG  = 0x06500000,
158    INSN_LDRSB_IMM = 0x005000d0,
159    INSN_LDRSB_REG = 0x001000d0,
160    INSN_STRB_IMM  = 0x04400000,
161    INSN_STRB_REG  = 0x06400000,
162
163    INSN_LDRD_IMM  = 0x004000d0,
164    INSN_LDRD_REG  = 0x000000d0,
165    INSN_STRD_IMM  = 0x004000f0,
166    INSN_STRD_REG  = 0x000000f0,
167
168    INSN_DMB_ISH   = 0xf57ff05b,
169    INSN_DMB_MCR   = 0xee070fba,
170
171    /* Architected nop introduced in v6k.  */
172    /* ??? This is an MSR (imm) 0,0,0 insn.  Anyone know if this
173       also Just So Happened to do nothing on pre-v6k so that we
174       don't need to conditionalize it?  */
175    INSN_NOP_v6k   = 0xe320f000,
176    /* Otherwise the assembler uses mov r0,r0 */
177    INSN_NOP_v4    = (COND_AL << 28) | ARITH_MOV,
178
179    INSN_VADD      = 0xf2000800,
180    INSN_VAND      = 0xf2000110,
181    INSN_VBIC      = 0xf2100110,
182    INSN_VEOR      = 0xf3000110,
183    INSN_VORN      = 0xf2300110,
184    INSN_VORR      = 0xf2200110,
185    INSN_VSUB      = 0xf3000800,
186    INSN_VMUL      = 0xf2000910,
187    INSN_VQADD     = 0xf2000010,
188    INSN_VQADD_U   = 0xf3000010,
189    INSN_VQSUB     = 0xf2000210,
190    INSN_VQSUB_U   = 0xf3000210,
191    INSN_VMAX      = 0xf2000600,
192    INSN_VMAX_U    = 0xf3000600,
193    INSN_VMIN      = 0xf2000610,
194    INSN_VMIN_U    = 0xf3000610,
195
196    INSN_VABS      = 0xf3b10300,
197    INSN_VMVN      = 0xf3b00580,
198    INSN_VNEG      = 0xf3b10380,
199
200    INSN_VCEQ0     = 0xf3b10100,
201    INSN_VCGT0     = 0xf3b10000,
202    INSN_VCGE0     = 0xf3b10080,
203    INSN_VCLE0     = 0xf3b10180,
204    INSN_VCLT0     = 0xf3b10200,
205
206    INSN_VCEQ      = 0xf3000810,
207    INSN_VCGE      = 0xf2000310,
208    INSN_VCGT      = 0xf2000300,
209    INSN_VCGE_U    = 0xf3000310,
210    INSN_VCGT_U    = 0xf3000300,
211
212    INSN_VSHLI     = 0xf2800510,  /* VSHL (immediate) */
213    INSN_VSARI     = 0xf2800010,  /* VSHR.S */
214    INSN_VSHRI     = 0xf3800010,  /* VSHR.U */
215    INSN_VSLI      = 0xf3800510,
216    INSN_VSHL_S    = 0xf2000400,  /* VSHL.S (register) */
217    INSN_VSHL_U    = 0xf3000400,  /* VSHL.U (register) */
218
219    INSN_VBSL      = 0xf3100110,
220    INSN_VBIT      = 0xf3200110,
221    INSN_VBIF      = 0xf3300110,
222
223    INSN_VTST      = 0xf2000810,
224
225    INSN_VDUP_G    = 0xee800b10,  /* VDUP (ARM core register) */
226    INSN_VDUP_S    = 0xf3b00c00,  /* VDUP (scalar) */
227    INSN_VLDR_D    = 0xed100b00,  /* VLDR.64 */
228    INSN_VLD1      = 0xf4200000,  /* VLD1 (multiple single elements) */
229    INSN_VLD1R     = 0xf4a00c00,  /* VLD1 (single element to all lanes) */
230    INSN_VST1      = 0xf4000000,  /* VST1 (multiple single elements) */
231    INSN_VMOVI     = 0xf2800010,  /* VMOV (immediate) */
232} ARMInsn;
233
234#define INSN_NOP   (use_armv7_instructions ? INSN_NOP_v6k : INSN_NOP_v4)
235
236static const uint8_t tcg_cond_to_arm_cond[] = {
237    [TCG_COND_EQ] = COND_EQ,
238    [TCG_COND_NE] = COND_NE,
239    [TCG_COND_LT] = COND_LT,
240    [TCG_COND_GE] = COND_GE,
241    [TCG_COND_LE] = COND_LE,
242    [TCG_COND_GT] = COND_GT,
243    /* unsigned */
244    [TCG_COND_LTU] = COND_CC,
245    [TCG_COND_GEU] = COND_CS,
246    [TCG_COND_LEU] = COND_LS,
247    [TCG_COND_GTU] = COND_HI,
248};
249
250static int encode_imm(uint32_t imm);
251
252/* TCG private relocation type: add with pc+imm8 */
253#define R_ARM_PC8  11
254
255/* TCG private relocation type: vldr with imm8 << 2 */
256#define R_ARM_PC11 12
257
258static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
259{
260    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
261    ptrdiff_t offset = (tcg_ptr_byte_diff(target, src_rx) - 8) >> 2;
262
263    if (offset == sextract32(offset, 0, 24)) {
264        *src_rw = deposit32(*src_rw, 0, 24, offset);
265        return true;
266    }
267    return false;
268}
269
270static bool reloc_pc13(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
271{
272    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
273    ptrdiff_t offset = tcg_ptr_byte_diff(target, src_rx) - 8;
274
275    if (offset >= -0xfff && offset <= 0xfff) {
276        tcg_insn_unit insn = *src_rw;
277        bool u = (offset >= 0);
278        if (!u) {
279            offset = -offset;
280        }
281        insn = deposit32(insn, 23, 1, u);
282        insn = deposit32(insn, 0, 12, offset);
283        *src_rw = insn;
284        return true;
285    }
286    return false;
287}
288
289static bool reloc_pc11(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
290{
291    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
292    ptrdiff_t offset = (tcg_ptr_byte_diff(target, src_rx) - 8) / 4;
293
294    if (offset >= -0xff && offset <= 0xff) {
295        tcg_insn_unit insn = *src_rw;
296        bool u = (offset >= 0);
297        if (!u) {
298            offset = -offset;
299        }
300        insn = deposit32(insn, 23, 1, u);
301        insn = deposit32(insn, 0, 8, offset);
302        *src_rw = insn;
303        return true;
304    }
305    return false;
306}
307
308static bool reloc_pc8(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
309{
310    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
311    ptrdiff_t offset = tcg_ptr_byte_diff(target, src_rx) - 8;
312    int rot = encode_imm(offset);
313
314    if (rot >= 0) {
315        *src_rw = deposit32(*src_rw, 0, 12, rol32(offset, rot) | (rot << 7));
316        return true;
317    }
318    return false;
319}
320
321static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
322                        intptr_t value, intptr_t addend)
323{
324    tcg_debug_assert(addend == 0);
325    switch (type) {
326    case R_ARM_PC24:
327        return reloc_pc24(code_ptr, (const tcg_insn_unit *)value);
328    case R_ARM_PC13:
329        return reloc_pc13(code_ptr, (const tcg_insn_unit *)value);
330    case R_ARM_PC11:
331        return reloc_pc11(code_ptr, (const tcg_insn_unit *)value);
332    case R_ARM_PC8:
333        return reloc_pc8(code_ptr, (const tcg_insn_unit *)value);
334    default:
335        g_assert_not_reached();
336    }
337}
338
339#define TCG_CT_CONST_ARM  0x100
340#define TCG_CT_CONST_INV  0x200
341#define TCG_CT_CONST_NEG  0x400
342#define TCG_CT_CONST_ZERO 0x800
343#define TCG_CT_CONST_ORRI 0x1000
344#define TCG_CT_CONST_ANDI 0x2000
345
346#define ALL_GENERAL_REGS  0xffffu
347#define ALL_VECTOR_REGS   0xffff0000u
348
349/*
350 * r0-r2 will be overwritten when reading the tlb entry (softmmu only)
351 * and r0-r1 doing the byte swapping, so don't use these.
352 * r3 is removed for softmmu to avoid clashes with helper arguments.
353 */
354#ifdef CONFIG_SOFTMMU
355#define ALL_QLOAD_REGS \
356    (ALL_GENERAL_REGS & ~((1 << TCG_REG_R0) | (1 << TCG_REG_R1) | \
357                          (1 << TCG_REG_R2) | (1 << TCG_REG_R3) | \
358                          (1 << TCG_REG_R14)))
359#define ALL_QSTORE_REGS \
360    (ALL_GENERAL_REGS & ~((1 << TCG_REG_R0) | (1 << TCG_REG_R1) | \
361                          (1 << TCG_REG_R2) | (1 << TCG_REG_R14) | \
362                          ((TARGET_LONG_BITS == 64) << TCG_REG_R3)))
363#else
364#define ALL_QLOAD_REGS   ALL_GENERAL_REGS
365#define ALL_QSTORE_REGS \
366    (ALL_GENERAL_REGS & ~((1 << TCG_REG_R0) | (1 << TCG_REG_R1)))
367#endif
368
369static inline uint32_t rotl(uint32_t val, int n)
370{
371  return (val << n) | (val >> (32 - n));
372}
373
374/* ARM immediates for ALU instructions are made of an unsigned 8-bit
375   right-rotated by an even amount between 0 and 30. */
376static int encode_imm(uint32_t imm)
377{
378    int shift;
379
380    /* simple case, only lower bits */
381    if ((imm & ~0xff) == 0)
382        return 0;
383    /* then try a simple even shift */
384    shift = ctz32(imm) & ~1;
385    if (((imm >> shift) & ~0xff) == 0)
386        return 32 - shift;
387    /* now try harder with rotations */
388    if ((rotl(imm, 2) & ~0xff) == 0)
389        return 2;
390    if ((rotl(imm, 4) & ~0xff) == 0)
391        return 4;
392    if ((rotl(imm, 6) & ~0xff) == 0)
393        return 6;
394    /* imm can't be encoded */
395    return -1;
396}
397
398static inline int check_fit_imm(uint32_t imm)
399{
400    return encode_imm(imm) >= 0;
401}
402
403/* Return true if v16 is a valid 16-bit shifted immediate.  */
404static bool is_shimm16(uint16_t v16, int *cmode, int *imm8)
405{
406    if (v16 == (v16 & 0xff)) {
407        *cmode = 0x8;
408        *imm8 = v16 & 0xff;
409        return true;
410    } else if (v16 == (v16 & 0xff00)) {
411        *cmode = 0xa;
412        *imm8 = v16 >> 8;
413        return true;
414    }
415    return false;
416}
417
418/* Return true if v32 is a valid 32-bit shifted immediate.  */
419static bool is_shimm32(uint32_t v32, int *cmode, int *imm8)
420{
421    if (v32 == (v32 & 0xff)) {
422        *cmode = 0x0;
423        *imm8 = v32 & 0xff;
424        return true;
425    } else if (v32 == (v32 & 0xff00)) {
426        *cmode = 0x2;
427        *imm8 = (v32 >> 8) & 0xff;
428        return true;
429    } else if (v32 == (v32 & 0xff0000)) {
430        *cmode = 0x4;
431        *imm8 = (v32 >> 16) & 0xff;
432        return true;
433    } else if (v32 == (v32 & 0xff000000)) {
434        *cmode = 0x6;
435        *imm8 = v32 >> 24;
436        return true;
437    }
438    return false;
439}
440
441/* Return true if v32 is a valid 32-bit shifting ones immediate.  */
442static bool is_soimm32(uint32_t v32, int *cmode, int *imm8)
443{
444    if ((v32 & 0xffff00ff) == 0xff) {
445        *cmode = 0xc;
446        *imm8 = (v32 >> 8) & 0xff;
447        return true;
448    } else if ((v32 & 0xff00ffff) == 0xffff) {
449        *cmode = 0xd;
450        *imm8 = (v32 >> 16) & 0xff;
451        return true;
452    }
453    return false;
454}
455
456/*
457 * Return non-zero if v32 can be formed by MOVI+ORR.
458 * Place the parameters for MOVI in (cmode, imm8).
459 * Return the cmode for ORR; the imm8 can be had via extraction from v32.
460 */
461static int is_shimm32_pair(uint32_t v32, int *cmode, int *imm8)
462{
463    int i;
464
465    for (i = 6; i > 0; i -= 2) {
466        /* Mask out one byte we can add with ORR.  */
467        uint32_t tmp = v32 & ~(0xffu << (i * 4));
468        if (is_shimm32(tmp, cmode, imm8) ||
469            is_soimm32(tmp, cmode, imm8)) {
470            break;
471        }
472    }
473    return i;
474}
475
476/* Return true if V is a valid 16-bit or 32-bit shifted immediate.  */
477static bool is_shimm1632(uint32_t v32, int *cmode, int *imm8)
478{
479    if (v32 == deposit32(v32, 16, 16, v32)) {
480        return is_shimm16(v32, cmode, imm8);
481    } else {
482        return is_shimm32(v32, cmode, imm8);
483    }
484}
485
486/* Test if a constant matches the constraint.
487 * TODO: define constraints for:
488 *
489 * ldr/str offset:   between -0xfff and 0xfff
490 * ldrh/strh offset: between -0xff and 0xff
491 * mov operand2:     values represented with x << (2 * y), x < 0x100
492 * add, sub, eor...: ditto
493 */
494static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
495{
496    if (ct & TCG_CT_CONST) {
497        return 1;
498    } else if ((ct & TCG_CT_CONST_ARM) && check_fit_imm(val)) {
499        return 1;
500    } else if ((ct & TCG_CT_CONST_INV) && check_fit_imm(~val)) {
501        return 1;
502    } else if ((ct & TCG_CT_CONST_NEG) && check_fit_imm(-val)) {
503        return 1;
504    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
505        return 1;
506    }
507
508    switch (ct & (TCG_CT_CONST_ORRI | TCG_CT_CONST_ANDI)) {
509    case 0:
510        break;
511    case TCG_CT_CONST_ANDI:
512        val = ~val;
513        /* fallthru */
514    case TCG_CT_CONST_ORRI:
515        if (val == deposit64(val, 32, 32, val)) {
516            int cmode, imm8;
517            return is_shimm1632(val, &cmode, &imm8);
518        }
519        break;
520    default:
521        /* Both bits should not be set for the same insn.  */
522        g_assert_not_reached();
523    }
524
525    return 0;
526}
527
528static inline void tcg_out_b(TCGContext *s, int cond, int32_t offset)
529{
530    tcg_out32(s, (cond << 28) | 0x0a000000 |
531                    (((offset - 8) >> 2) & 0x00ffffff));
532}
533
534static inline void tcg_out_bl(TCGContext *s, int cond, int32_t offset)
535{
536    tcg_out32(s, (cond << 28) | 0x0b000000 |
537                    (((offset - 8) >> 2) & 0x00ffffff));
538}
539
540static inline void tcg_out_blx(TCGContext *s, int cond, int rn)
541{
542    tcg_out32(s, (cond << 28) | 0x012fff30 | rn);
543}
544
545static inline void tcg_out_blx_imm(TCGContext *s, int32_t offset)
546{
547    tcg_out32(s, 0xfa000000 | ((offset & 2) << 23) |
548                (((offset - 8) >> 2) & 0x00ffffff));
549}
550
551static inline void tcg_out_dat_reg(TCGContext *s,
552                int cond, int opc, int rd, int rn, int rm, int shift)
553{
554    tcg_out32(s, (cond << 28) | (0 << 25) | opc |
555                    (rn << 16) | (rd << 12) | shift | rm);
556}
557
558static inline void tcg_out_nop(TCGContext *s)
559{
560    tcg_out32(s, INSN_NOP);
561}
562
563static inline void tcg_out_mov_reg(TCGContext *s, int cond, int rd, int rm)
564{
565    /* Simple reg-reg move, optimising out the 'do nothing' case */
566    if (rd != rm) {
567        tcg_out_dat_reg(s, cond, ARITH_MOV, rd, 0, rm, SHIFT_IMM_LSL(0));
568    }
569}
570
571static inline void tcg_out_bx(TCGContext *s, int cond, TCGReg rn)
572{
573    /* Unless the C portion of QEMU is compiled as thumb, we don't
574       actually need true BX semantics; merely a branch to an address
575       held in a register.  */
576    if (use_armv5t_instructions) {
577        tcg_out32(s, (cond << 28) | 0x012fff10 | rn);
578    } else {
579        tcg_out_mov_reg(s, cond, TCG_REG_PC, rn);
580    }
581}
582
583static inline void tcg_out_dat_imm(TCGContext *s,
584                int cond, int opc, int rd, int rn, int im)
585{
586    tcg_out32(s, (cond << 28) | (1 << 25) | opc |
587                    (rn << 16) | (rd << 12) | im);
588}
589
590/* Note that this routine is used for both LDR and LDRH formats, so we do
591   not wish to include an immediate shift at this point.  */
592static void tcg_out_memop_r(TCGContext *s, int cond, ARMInsn opc, TCGReg rt,
593                            TCGReg rn, TCGReg rm, bool u, bool p, bool w)
594{
595    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24)
596              | (w << 21) | (rn << 16) | (rt << 12) | rm);
597}
598
599static void tcg_out_memop_8(TCGContext *s, int cond, ARMInsn opc, TCGReg rt,
600                            TCGReg rn, int imm8, bool p, bool w)
601{
602    bool u = 1;
603    if (imm8 < 0) {
604        imm8 = -imm8;
605        u = 0;
606    }
607    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
608              (rn << 16) | (rt << 12) | ((imm8 & 0xf0) << 4) | (imm8 & 0xf));
609}
610
611static void tcg_out_memop_12(TCGContext *s, int cond, ARMInsn opc, TCGReg rt,
612                             TCGReg rn, int imm12, bool p, bool w)
613{
614    bool u = 1;
615    if (imm12 < 0) {
616        imm12 = -imm12;
617        u = 0;
618    }
619    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
620              (rn << 16) | (rt << 12) | imm12);
621}
622
623static inline void tcg_out_ld32_12(TCGContext *s, int cond, TCGReg rt,
624                                   TCGReg rn, int imm12)
625{
626    tcg_out_memop_12(s, cond, INSN_LDR_IMM, rt, rn, imm12, 1, 0);
627}
628
629static inline void tcg_out_st32_12(TCGContext *s, int cond, TCGReg rt,
630                                   TCGReg rn, int imm12)
631{
632    tcg_out_memop_12(s, cond, INSN_STR_IMM, rt, rn, imm12, 1, 0);
633}
634
635static inline void tcg_out_ld32_r(TCGContext *s, int cond, TCGReg rt,
636                                  TCGReg rn, TCGReg rm)
637{
638    tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 0);
639}
640
641static inline void tcg_out_st32_r(TCGContext *s, int cond, TCGReg rt,
642                                  TCGReg rn, TCGReg rm)
643{
644    tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 0);
645}
646
647static inline void tcg_out_ldrd_8(TCGContext *s, int cond, TCGReg rt,
648                                   TCGReg rn, int imm8)
649{
650    tcg_out_memop_8(s, cond, INSN_LDRD_IMM, rt, rn, imm8, 1, 0);
651}
652
653static inline void tcg_out_ldrd_r(TCGContext *s, int cond, TCGReg rt,
654                                  TCGReg rn, TCGReg rm)
655{
656    tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 0);
657}
658
659static inline void tcg_out_ldrd_rwb(TCGContext *s, int cond, TCGReg rt,
660                                    TCGReg rn, TCGReg rm)
661{
662    tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 1);
663}
664
665static inline void tcg_out_strd_8(TCGContext *s, int cond, TCGReg rt,
666                                   TCGReg rn, int imm8)
667{
668    tcg_out_memop_8(s, cond, INSN_STRD_IMM, rt, rn, imm8, 1, 0);
669}
670
671static inline void tcg_out_strd_r(TCGContext *s, int cond, TCGReg rt,
672                                  TCGReg rn, TCGReg rm)
673{
674    tcg_out_memop_r(s, cond, INSN_STRD_REG, rt, rn, rm, 1, 1, 0);
675}
676
677/* Register pre-increment with base writeback.  */
678static inline void tcg_out_ld32_rwb(TCGContext *s, int cond, TCGReg rt,
679                                    TCGReg rn, TCGReg rm)
680{
681    tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 1);
682}
683
684static inline void tcg_out_st32_rwb(TCGContext *s, int cond, TCGReg rt,
685                                    TCGReg rn, TCGReg rm)
686{
687    tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 1);
688}
689
690static inline void tcg_out_ld16u_8(TCGContext *s, int cond, TCGReg rt,
691                                   TCGReg rn, int imm8)
692{
693    tcg_out_memop_8(s, cond, INSN_LDRH_IMM, rt, rn, imm8, 1, 0);
694}
695
696static inline void tcg_out_st16_8(TCGContext *s, int cond, TCGReg rt,
697                                  TCGReg rn, int imm8)
698{
699    tcg_out_memop_8(s, cond, INSN_STRH_IMM, rt, rn, imm8, 1, 0);
700}
701
702static inline void tcg_out_ld16u_r(TCGContext *s, int cond, TCGReg rt,
703                                   TCGReg rn, TCGReg rm)
704{
705    tcg_out_memop_r(s, cond, INSN_LDRH_REG, rt, rn, rm, 1, 1, 0);
706}
707
708static inline void tcg_out_st16_r(TCGContext *s, int cond, TCGReg rt,
709                                  TCGReg rn, TCGReg rm)
710{
711    tcg_out_memop_r(s, cond, INSN_STRH_REG, rt, rn, rm, 1, 1, 0);
712}
713
714static inline void tcg_out_ld16s_8(TCGContext *s, int cond, TCGReg rt,
715                                   TCGReg rn, int imm8)
716{
717    tcg_out_memop_8(s, cond, INSN_LDRSH_IMM, rt, rn, imm8, 1, 0);
718}
719
720static inline void tcg_out_ld16s_r(TCGContext *s, int cond, TCGReg rt,
721                                   TCGReg rn, TCGReg rm)
722{
723    tcg_out_memop_r(s, cond, INSN_LDRSH_REG, rt, rn, rm, 1, 1, 0);
724}
725
726static inline void tcg_out_ld8_12(TCGContext *s, int cond, TCGReg rt,
727                                  TCGReg rn, int imm12)
728{
729    tcg_out_memop_12(s, cond, INSN_LDRB_IMM, rt, rn, imm12, 1, 0);
730}
731
732static inline void tcg_out_st8_12(TCGContext *s, int cond, TCGReg rt,
733                                  TCGReg rn, int imm12)
734{
735    tcg_out_memop_12(s, cond, INSN_STRB_IMM, rt, rn, imm12, 1, 0);
736}
737
738static inline void tcg_out_ld8_r(TCGContext *s, int cond, TCGReg rt,
739                                 TCGReg rn, TCGReg rm)
740{
741    tcg_out_memop_r(s, cond, INSN_LDRB_REG, rt, rn, rm, 1, 1, 0);
742}
743
744static inline void tcg_out_st8_r(TCGContext *s, int cond, TCGReg rt,
745                                 TCGReg rn, TCGReg rm)
746{
747    tcg_out_memop_r(s, cond, INSN_STRB_REG, rt, rn, rm, 1, 1, 0);
748}
749
750static inline void tcg_out_ld8s_8(TCGContext *s, int cond, TCGReg rt,
751                                  TCGReg rn, int imm8)
752{
753    tcg_out_memop_8(s, cond, INSN_LDRSB_IMM, rt, rn, imm8, 1, 0);
754}
755
756static inline void tcg_out_ld8s_r(TCGContext *s, int cond, TCGReg rt,
757                                  TCGReg rn, TCGReg rm)
758{
759    tcg_out_memop_r(s, cond, INSN_LDRSB_REG, rt, rn, rm, 1, 1, 0);
760}
761
762static void tcg_out_movi_pool(TCGContext *s, int cond, int rd, uint32_t arg)
763{
764    new_pool_label(s, arg, R_ARM_PC13, s->code_ptr, 0);
765    tcg_out_ld32_12(s, cond, rd, TCG_REG_PC, 0);
766}
767
768static void tcg_out_movi32(TCGContext *s, int cond, int rd, uint32_t arg)
769{
770    int rot, diff, opc, sh1, sh2;
771    uint32_t tt0, tt1, tt2;
772
773    /* Check a single MOV/MVN before anything else.  */
774    rot = encode_imm(arg);
775    if (rot >= 0) {
776        tcg_out_dat_imm(s, cond, ARITH_MOV, rd, 0,
777                        rotl(arg, rot) | (rot << 7));
778        return;
779    }
780    rot = encode_imm(~arg);
781    if (rot >= 0) {
782        tcg_out_dat_imm(s, cond, ARITH_MVN, rd, 0,
783                        rotl(~arg, rot) | (rot << 7));
784        return;
785    }
786
787    /* Check for a pc-relative address.  This will usually be the TB,
788       or within the TB, which is immediately before the code block.  */
789    diff = tcg_pcrel_diff(s, (void *)arg) - 8;
790    if (diff >= 0) {
791        rot = encode_imm(diff);
792        if (rot >= 0) {
793            tcg_out_dat_imm(s, cond, ARITH_ADD, rd, TCG_REG_PC,
794                            rotl(diff, rot) | (rot << 7));
795            return;
796        }
797    } else {
798        rot = encode_imm(-diff);
799        if (rot >= 0) {
800            tcg_out_dat_imm(s, cond, ARITH_SUB, rd, TCG_REG_PC,
801                            rotl(-diff, rot) | (rot << 7));
802            return;
803        }
804    }
805
806    /* Use movw + movt.  */
807    if (use_armv7_instructions) {
808        /* movw */
809        tcg_out32(s, (cond << 28) | 0x03000000 | (rd << 12)
810                  | ((arg << 4) & 0x000f0000) | (arg & 0xfff));
811        if (arg & 0xffff0000) {
812            /* movt */
813            tcg_out32(s, (cond << 28) | 0x03400000 | (rd << 12)
814                      | ((arg >> 12) & 0x000f0000) | ((arg >> 16) & 0xfff));
815        }
816        return;
817    }
818
819    /* Look for sequences of two insns.  If we have lots of 1's, we can
820       shorten the sequence by beginning with mvn and then clearing
821       higher bits with eor.  */
822    tt0 = arg;
823    opc = ARITH_MOV;
824    if (ctpop32(arg) > 16) {
825        tt0 = ~arg;
826        opc = ARITH_MVN;
827    }
828    sh1 = ctz32(tt0) & ~1;
829    tt1 = tt0 & ~(0xff << sh1);
830    sh2 = ctz32(tt1) & ~1;
831    tt2 = tt1 & ~(0xff << sh2);
832    if (tt2 == 0) {
833        rot = ((32 - sh1) << 7) & 0xf00;
834        tcg_out_dat_imm(s, cond, opc, rd,  0, ((tt0 >> sh1) & 0xff) | rot);
835        rot = ((32 - sh2) << 7) & 0xf00;
836        tcg_out_dat_imm(s, cond, ARITH_EOR, rd, rd,
837                        ((tt0 >> sh2) & 0xff) | rot);
838        return;
839    }
840
841    /* Otherwise, drop it into the constant pool.  */
842    tcg_out_movi_pool(s, cond, rd, arg);
843}
844
845static inline void tcg_out_dat_rI(TCGContext *s, int cond, int opc, TCGArg dst,
846                                  TCGArg lhs, TCGArg rhs, int rhs_is_const)
847{
848    /* Emit either the reg,imm or reg,reg form of a data-processing insn.
849     * rhs must satisfy the "rI" constraint.
850     */
851    if (rhs_is_const) {
852        int rot = encode_imm(rhs);
853        tcg_debug_assert(rot >= 0);
854        tcg_out_dat_imm(s, cond, opc, dst, lhs, rotl(rhs, rot) | (rot << 7));
855    } else {
856        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
857    }
858}
859
860static void tcg_out_dat_rIK(TCGContext *s, int cond, int opc, int opinv,
861                            TCGReg dst, TCGReg lhs, TCGArg rhs,
862                            bool rhs_is_const)
863{
864    /* Emit either the reg,imm or reg,reg form of a data-processing insn.
865     * rhs must satisfy the "rIK" constraint.
866     */
867    if (rhs_is_const) {
868        int rot = encode_imm(rhs);
869        if (rot < 0) {
870            rhs = ~rhs;
871            rot = encode_imm(rhs);
872            tcg_debug_assert(rot >= 0);
873            opc = opinv;
874        }
875        tcg_out_dat_imm(s, cond, opc, dst, lhs, rotl(rhs, rot) | (rot << 7));
876    } else {
877        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
878    }
879}
880
881static void tcg_out_dat_rIN(TCGContext *s, int cond, int opc, int opneg,
882                            TCGArg dst, TCGArg lhs, TCGArg rhs,
883                            bool rhs_is_const)
884{
885    /* Emit either the reg,imm or reg,reg form of a data-processing insn.
886     * rhs must satisfy the "rIN" constraint.
887     */
888    if (rhs_is_const) {
889        int rot = encode_imm(rhs);
890        if (rot < 0) {
891            rhs = -rhs;
892            rot = encode_imm(rhs);
893            tcg_debug_assert(rot >= 0);
894            opc = opneg;
895        }
896        tcg_out_dat_imm(s, cond, opc, dst, lhs, rotl(rhs, rot) | (rot << 7));
897    } else {
898        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
899    }
900}
901
902static inline void tcg_out_mul32(TCGContext *s, int cond, TCGReg rd,
903                                 TCGReg rn, TCGReg rm)
904{
905    /* if ArchVersion() < 6 && d == n then UNPREDICTABLE;  */
906    if (!use_armv6_instructions && rd == rn) {
907        if (rd == rm) {
908            /* rd == rn == rm; copy an input to tmp first.  */
909            tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
910            rm = rn = TCG_REG_TMP;
911        } else {
912            rn = rm;
913            rm = rd;
914        }
915    }
916    /* mul */
917    tcg_out32(s, (cond << 28) | 0x90 | (rd << 16) | (rm << 8) | rn);
918}
919
920static inline void tcg_out_umull32(TCGContext *s, int cond, TCGReg rd0,
921                                   TCGReg rd1, TCGReg rn, TCGReg rm)
922{
923    /* if ArchVersion() < 6 && (dHi == n || dLo == n) then UNPREDICTABLE;  */
924    if (!use_armv6_instructions && (rd0 == rn || rd1 == rn)) {
925        if (rd0 == rm || rd1 == rm) {
926            tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
927            rn = TCG_REG_TMP;
928        } else {
929            TCGReg t = rn;
930            rn = rm;
931            rm = t;
932        }
933    }
934    /* umull */
935    tcg_out32(s, (cond << 28) | 0x00800090 |
936              (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
937}
938
939static inline void tcg_out_smull32(TCGContext *s, int cond, TCGReg rd0,
940                                   TCGReg rd1, TCGReg rn, TCGReg rm)
941{
942    /* if ArchVersion() < 6 && (dHi == n || dLo == n) then UNPREDICTABLE;  */
943    if (!use_armv6_instructions && (rd0 == rn || rd1 == rn)) {
944        if (rd0 == rm || rd1 == rm) {
945            tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
946            rn = TCG_REG_TMP;
947        } else {
948            TCGReg t = rn;
949            rn = rm;
950            rm = t;
951        }
952    }
953    /* smull */
954    tcg_out32(s, (cond << 28) | 0x00c00090 |
955              (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
956}
957
958static inline void tcg_out_sdiv(TCGContext *s, int cond, int rd, int rn, int rm)
959{
960    tcg_out32(s, 0x0710f010 | (cond << 28) | (rd << 16) | rn | (rm << 8));
961}
962
963static inline void tcg_out_udiv(TCGContext *s, int cond, int rd, int rn, int rm)
964{
965    tcg_out32(s, 0x0730f010 | (cond << 28) | (rd << 16) | rn | (rm << 8));
966}
967
968static inline void tcg_out_ext8s(TCGContext *s, int cond,
969                                 int rd, int rn)
970{
971    if (use_armv6_instructions) {
972        /* sxtb */
973        tcg_out32(s, 0x06af0070 | (cond << 28) | (rd << 12) | rn);
974    } else {
975        tcg_out_dat_reg(s, cond, ARITH_MOV,
976                        rd, 0, rn, SHIFT_IMM_LSL(24));
977        tcg_out_dat_reg(s, cond, ARITH_MOV,
978                        rd, 0, rd, SHIFT_IMM_ASR(24));
979    }
980}
981
982static inline void tcg_out_ext8u(TCGContext *s, int cond,
983                                 int rd, int rn)
984{
985    tcg_out_dat_imm(s, cond, ARITH_AND, rd, rn, 0xff);
986}
987
988static inline void tcg_out_ext16s(TCGContext *s, int cond,
989                                  int rd, int rn)
990{
991    if (use_armv6_instructions) {
992        /* sxth */
993        tcg_out32(s, 0x06bf0070 | (cond << 28) | (rd << 12) | rn);
994    } else {
995        tcg_out_dat_reg(s, cond, ARITH_MOV,
996                        rd, 0, rn, SHIFT_IMM_LSL(16));
997        tcg_out_dat_reg(s, cond, ARITH_MOV,
998                        rd, 0, rd, SHIFT_IMM_ASR(16));
999    }
1000}
1001
1002static inline void tcg_out_ext16u(TCGContext *s, int cond,
1003                                  int rd, int rn)
1004{
1005    if (use_armv6_instructions) {
1006        /* uxth */
1007        tcg_out32(s, 0x06ff0070 | (cond << 28) | (rd << 12) | rn);
1008    } else {
1009        tcg_out_dat_reg(s, cond, ARITH_MOV,
1010                        rd, 0, rn, SHIFT_IMM_LSL(16));
1011        tcg_out_dat_reg(s, cond, ARITH_MOV,
1012                        rd, 0, rd, SHIFT_IMM_LSR(16));
1013    }
1014}
1015
1016static inline void tcg_out_bswap16s(TCGContext *s, int cond, int rd, int rn)
1017{
1018    if (use_armv6_instructions) {
1019        /* revsh */
1020        tcg_out32(s, 0x06ff0fb0 | (cond << 28) | (rd << 12) | rn);
1021    } else {
1022        tcg_out_dat_reg(s, cond, ARITH_MOV,
1023                        TCG_REG_TMP, 0, rn, SHIFT_IMM_LSL(24));
1024        tcg_out_dat_reg(s, cond, ARITH_MOV,
1025                        TCG_REG_TMP, 0, TCG_REG_TMP, SHIFT_IMM_ASR(16));
1026        tcg_out_dat_reg(s, cond, ARITH_ORR,
1027                        rd, TCG_REG_TMP, rn, SHIFT_IMM_LSR(8));
1028    }
1029}
1030
1031static inline void tcg_out_bswap16(TCGContext *s, int cond, int rd, int rn)
1032{
1033    if (use_armv6_instructions) {
1034        /* rev16 */
1035        tcg_out32(s, 0x06bf0fb0 | (cond << 28) | (rd << 12) | rn);
1036    } else {
1037        tcg_out_dat_reg(s, cond, ARITH_MOV,
1038                        TCG_REG_TMP, 0, rn, SHIFT_IMM_LSL(24));
1039        tcg_out_dat_reg(s, cond, ARITH_MOV,
1040                        TCG_REG_TMP, 0, TCG_REG_TMP, SHIFT_IMM_LSR(16));
1041        tcg_out_dat_reg(s, cond, ARITH_ORR,
1042                        rd, TCG_REG_TMP, rn, SHIFT_IMM_LSR(8));
1043    }
1044}
1045
1046/* swap the two low bytes assuming that the two high input bytes and the
1047   two high output bit can hold any value. */
1048static inline void tcg_out_bswap16st(TCGContext *s, int cond, int rd, int rn)
1049{
1050    if (use_armv6_instructions) {
1051        /* rev16 */
1052        tcg_out32(s, 0x06bf0fb0 | (cond << 28) | (rd << 12) | rn);
1053    } else {
1054        tcg_out_dat_reg(s, cond, ARITH_MOV,
1055                        TCG_REG_TMP, 0, rn, SHIFT_IMM_LSR(8));
1056        tcg_out_dat_imm(s, cond, ARITH_AND, TCG_REG_TMP, TCG_REG_TMP, 0xff);
1057        tcg_out_dat_reg(s, cond, ARITH_ORR,
1058                        rd, TCG_REG_TMP, rn, SHIFT_IMM_LSL(8));
1059    }
1060}
1061
1062static inline void tcg_out_bswap32(TCGContext *s, int cond, int rd, int rn)
1063{
1064    if (use_armv6_instructions) {
1065        /* rev */
1066        tcg_out32(s, 0x06bf0f30 | (cond << 28) | (rd << 12) | rn);
1067    } else {
1068        tcg_out_dat_reg(s, cond, ARITH_EOR,
1069                        TCG_REG_TMP, rn, rn, SHIFT_IMM_ROR(16));
1070        tcg_out_dat_imm(s, cond, ARITH_BIC,
1071                        TCG_REG_TMP, TCG_REG_TMP, 0xff | 0x800);
1072        tcg_out_dat_reg(s, cond, ARITH_MOV,
1073                        rd, 0, rn, SHIFT_IMM_ROR(8));
1074        tcg_out_dat_reg(s, cond, ARITH_EOR,
1075                        rd, rd, TCG_REG_TMP, SHIFT_IMM_LSR(8));
1076    }
1077}
1078
1079static inline void tcg_out_deposit(TCGContext *s, int cond, TCGReg rd,
1080                                   TCGArg a1, int ofs, int len, bool const_a1)
1081{
1082    if (const_a1) {
1083        /* bfi becomes bfc with rn == 15.  */
1084        a1 = 15;
1085    }
1086    /* bfi/bfc */
1087    tcg_out32(s, 0x07c00010 | (cond << 28) | (rd << 12) | a1
1088              | (ofs << 7) | ((ofs + len - 1) << 16));
1089}
1090
1091static inline void tcg_out_extract(TCGContext *s, int cond, TCGReg rd,
1092                                   TCGArg a1, int ofs, int len)
1093{
1094    /* ubfx */
1095    tcg_out32(s, 0x07e00050 | (cond << 28) | (rd << 12) | a1
1096              | (ofs << 7) | ((len - 1) << 16));
1097}
1098
1099static inline void tcg_out_sextract(TCGContext *s, int cond, TCGReg rd,
1100                                    TCGArg a1, int ofs, int len)
1101{
1102    /* sbfx */
1103    tcg_out32(s, 0x07a00050 | (cond << 28) | (rd << 12) | a1
1104              | (ofs << 7) | ((len - 1) << 16));
1105}
1106
1107static inline void tcg_out_ld32u(TCGContext *s, int cond,
1108                int rd, int rn, int32_t offset)
1109{
1110    if (offset > 0xfff || offset < -0xfff) {
1111        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1112        tcg_out_ld32_r(s, cond, rd, rn, TCG_REG_TMP);
1113    } else
1114        tcg_out_ld32_12(s, cond, rd, rn, offset);
1115}
1116
1117static inline void tcg_out_st32(TCGContext *s, int cond,
1118                int rd, int rn, int32_t offset)
1119{
1120    if (offset > 0xfff || offset < -0xfff) {
1121        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1122        tcg_out_st32_r(s, cond, rd, rn, TCG_REG_TMP);
1123    } else
1124        tcg_out_st32_12(s, cond, rd, rn, offset);
1125}
1126
1127static inline void tcg_out_ld16u(TCGContext *s, int cond,
1128                int rd, int rn, int32_t offset)
1129{
1130    if (offset > 0xff || offset < -0xff) {
1131        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1132        tcg_out_ld16u_r(s, cond, rd, rn, TCG_REG_TMP);
1133    } else
1134        tcg_out_ld16u_8(s, cond, rd, rn, offset);
1135}
1136
1137static inline void tcg_out_ld16s(TCGContext *s, int cond,
1138                int rd, int rn, int32_t offset)
1139{
1140    if (offset > 0xff || offset < -0xff) {
1141        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1142        tcg_out_ld16s_r(s, cond, rd, rn, TCG_REG_TMP);
1143    } else
1144        tcg_out_ld16s_8(s, cond, rd, rn, offset);
1145}
1146
1147static inline void tcg_out_st16(TCGContext *s, int cond,
1148                int rd, int rn, int32_t offset)
1149{
1150    if (offset > 0xff || offset < -0xff) {
1151        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1152        tcg_out_st16_r(s, cond, rd, rn, TCG_REG_TMP);
1153    } else
1154        tcg_out_st16_8(s, cond, rd, rn, offset);
1155}
1156
1157static inline void tcg_out_ld8u(TCGContext *s, int cond,
1158                int rd, int rn, int32_t offset)
1159{
1160    if (offset > 0xfff || offset < -0xfff) {
1161        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1162        tcg_out_ld8_r(s, cond, rd, rn, TCG_REG_TMP);
1163    } else
1164        tcg_out_ld8_12(s, cond, rd, rn, offset);
1165}
1166
1167static inline void tcg_out_ld8s(TCGContext *s, int cond,
1168                int rd, int rn, int32_t offset)
1169{
1170    if (offset > 0xff || offset < -0xff) {
1171        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1172        tcg_out_ld8s_r(s, cond, rd, rn, TCG_REG_TMP);
1173    } else
1174        tcg_out_ld8s_8(s, cond, rd, rn, offset);
1175}
1176
1177static inline void tcg_out_st8(TCGContext *s, int cond,
1178                int rd, int rn, int32_t offset)
1179{
1180    if (offset > 0xfff || offset < -0xfff) {
1181        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1182        tcg_out_st8_r(s, cond, rd, rn, TCG_REG_TMP);
1183    } else
1184        tcg_out_st8_12(s, cond, rd, rn, offset);
1185}
1186
1187/* The _goto case is normally between TBs within the same code buffer, and
1188 * with the code buffer limited to 16MB we wouldn't need the long case.
1189 * But we also use it for the tail-call to the qemu_ld/st helpers, which does.
1190 */
1191static void tcg_out_goto(TCGContext *s, int cond, const tcg_insn_unit *addr)
1192{
1193    intptr_t addri = (intptr_t)addr;
1194    ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1195
1196    if ((addri & 1) == 0 && disp - 8 < 0x01fffffd && disp - 8 > -0x01fffffd) {
1197        tcg_out_b(s, cond, disp);
1198        return;
1199    }
1200    tcg_out_movi_pool(s, cond, TCG_REG_PC, addri);
1201}
1202
1203/* The call case is mostly used for helpers - so it's not unreasonable
1204 * for them to be beyond branch range */
1205static void tcg_out_call(TCGContext *s, const tcg_insn_unit *addr)
1206{
1207    intptr_t addri = (intptr_t)addr;
1208    ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1209
1210    if (disp - 8 < 0x02000000 && disp - 8 >= -0x02000000) {
1211        if (addri & 1) {
1212            /* Use BLX if the target is in Thumb mode */
1213            if (!use_armv5t_instructions) {
1214                tcg_abort();
1215            }
1216            tcg_out_blx_imm(s, disp);
1217        } else {
1218            tcg_out_bl(s, COND_AL, disp);
1219        }
1220    } else if (use_armv7_instructions) {
1221        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, addri);
1222        tcg_out_blx(s, COND_AL, TCG_REG_TMP);
1223    } else {
1224        /* ??? Know that movi_pool emits exactly 1 insn.  */
1225        tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R14, TCG_REG_PC, 0);
1226        tcg_out_movi_pool(s, COND_AL, TCG_REG_PC, addri);
1227    }
1228}
1229
1230static inline void tcg_out_goto_label(TCGContext *s, int cond, TCGLabel *l)
1231{
1232    if (l->has_value) {
1233        tcg_out_goto(s, cond, l->u.value_ptr);
1234    } else {
1235        tcg_out_reloc(s, s->code_ptr, R_ARM_PC24, l, 0);
1236        tcg_out_b(s, cond, 0);
1237    }
1238}
1239
1240static inline void tcg_out_mb(TCGContext *s, TCGArg a0)
1241{
1242    if (use_armv7_instructions) {
1243        tcg_out32(s, INSN_DMB_ISH);
1244    } else if (use_armv6_instructions) {
1245        tcg_out32(s, INSN_DMB_MCR);
1246    }
1247}
1248
1249static TCGCond tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1250                            const int *const_args)
1251{
1252    TCGReg al = args[0];
1253    TCGReg ah = args[1];
1254    TCGArg bl = args[2];
1255    TCGArg bh = args[3];
1256    TCGCond cond = args[4];
1257    int const_bl = const_args[2];
1258    int const_bh = const_args[3];
1259
1260    switch (cond) {
1261    case TCG_COND_EQ:
1262    case TCG_COND_NE:
1263    case TCG_COND_LTU:
1264    case TCG_COND_LEU:
1265    case TCG_COND_GTU:
1266    case TCG_COND_GEU:
1267        /* We perform a conditional comparision.  If the high half is
1268           equal, then overwrite the flags with the comparison of the
1269           low half.  The resulting flags cover the whole.  */
1270        tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0, ah, bh, const_bh);
1271        tcg_out_dat_rI(s, COND_EQ, ARITH_CMP, 0, al, bl, const_bl);
1272        return cond;
1273
1274    case TCG_COND_LT:
1275    case TCG_COND_GE:
1276        /* We perform a double-word subtraction and examine the result.
1277           We do not actually need the result of the subtract, so the
1278           low part "subtract" is a compare.  For the high half we have
1279           no choice but to compute into a temporary.  */
1280        tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0, al, bl, const_bl);
1281        tcg_out_dat_rI(s, COND_AL, ARITH_SBC | TO_CPSR,
1282                       TCG_REG_TMP, ah, bh, const_bh);
1283        return cond;
1284
1285    case TCG_COND_LE:
1286    case TCG_COND_GT:
1287        /* Similar, but with swapped arguments, via reversed subtract.  */
1288        tcg_out_dat_rI(s, COND_AL, ARITH_RSB | TO_CPSR,
1289                       TCG_REG_TMP, al, bl, const_bl);
1290        tcg_out_dat_rI(s, COND_AL, ARITH_RSC | TO_CPSR,
1291                       TCG_REG_TMP, ah, bh, const_bh);
1292        return tcg_swap_cond(cond);
1293
1294    default:
1295        g_assert_not_reached();
1296    }
1297}
1298
1299/*
1300 * Note that TCGReg references Q-registers.
1301 * Q-regno = 2 * D-regno, so shift left by 1 whlie inserting.
1302 */
1303static uint32_t encode_vd(TCGReg rd)
1304{
1305    tcg_debug_assert(rd >= TCG_REG_Q0);
1306    return (extract32(rd, 3, 1) << 22) | (extract32(rd, 0, 3) << 13);
1307}
1308
1309static uint32_t encode_vn(TCGReg rn)
1310{
1311    tcg_debug_assert(rn >= TCG_REG_Q0);
1312    return (extract32(rn, 3, 1) << 7) | (extract32(rn, 0, 3) << 17);
1313}
1314
1315static uint32_t encode_vm(TCGReg rm)
1316{
1317    tcg_debug_assert(rm >= TCG_REG_Q0);
1318    return (extract32(rm, 3, 1) << 5) | (extract32(rm, 0, 3) << 1);
1319}
1320
1321static void tcg_out_vreg2(TCGContext *s, ARMInsn insn, int q, int vece,
1322                          TCGReg d, TCGReg m)
1323{
1324    tcg_out32(s, insn | (vece << 18) | (q << 6) |
1325              encode_vd(d) | encode_vm(m));
1326}
1327
1328static void tcg_out_vreg3(TCGContext *s, ARMInsn insn, int q, int vece,
1329                          TCGReg d, TCGReg n, TCGReg m)
1330{
1331    tcg_out32(s, insn | (vece << 20) | (q << 6) |
1332              encode_vd(d) | encode_vn(n) | encode_vm(m));
1333}
1334
1335static void tcg_out_vmovi(TCGContext *s, TCGReg rd,
1336                          int q, int op, int cmode, uint8_t imm8)
1337{
1338    tcg_out32(s, INSN_VMOVI | encode_vd(rd) | (q << 6) | (op << 5)
1339              | (cmode << 8) | extract32(imm8, 0, 4)
1340              | (extract32(imm8, 4, 3) << 16)
1341              | (extract32(imm8, 7, 1) << 24));
1342}
1343
1344static void tcg_out_vshifti(TCGContext *s, ARMInsn insn, int q,
1345                            TCGReg rd, TCGReg rm, int l_imm6)
1346{
1347    tcg_out32(s, insn | (q << 6) | encode_vd(rd) | encode_vm(rm) |
1348              (extract32(l_imm6, 6, 1) << 7) |
1349              (extract32(l_imm6, 0, 6) << 16));
1350}
1351
1352static void tcg_out_vldst(TCGContext *s, ARMInsn insn,
1353                          TCGReg rd, TCGReg rn, int offset)
1354{
1355    if (offset != 0) {
1356        if (check_fit_imm(offset) || check_fit_imm(-offset)) {
1357            tcg_out_dat_rIN(s, COND_AL, ARITH_ADD, ARITH_SUB,
1358                            TCG_REG_TMP, rn, offset, true);
1359        } else {
1360            tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, offset);
1361            tcg_out_dat_reg(s, COND_AL, ARITH_ADD,
1362                            TCG_REG_TMP, TCG_REG_TMP, rn, 0);
1363        }
1364        rn = TCG_REG_TMP;
1365    }
1366    tcg_out32(s, insn | (rn << 16) | encode_vd(rd) | 0xf);
1367}
1368
1369#ifdef CONFIG_SOFTMMU
1370#include "../tcg-ldst.c.inc"
1371
1372/* helper signature: helper_ret_ld_mmu(CPUState *env, target_ulong addr,
1373 *                                     int mmu_idx, uintptr_t ra)
1374 */
1375static void * const qemu_ld_helpers[16] = {
1376    [MO_UB]   = helper_ret_ldub_mmu,
1377    [MO_SB]   = helper_ret_ldsb_mmu,
1378
1379    [MO_LEUW] = helper_le_lduw_mmu,
1380    [MO_LEUL] = helper_le_ldul_mmu,
1381    [MO_LEQ]  = helper_le_ldq_mmu,
1382    [MO_LESW] = helper_le_ldsw_mmu,
1383    [MO_LESL] = helper_le_ldul_mmu,
1384
1385    [MO_BEUW] = helper_be_lduw_mmu,
1386    [MO_BEUL] = helper_be_ldul_mmu,
1387    [MO_BEQ]  = helper_be_ldq_mmu,
1388    [MO_BESW] = helper_be_ldsw_mmu,
1389    [MO_BESL] = helper_be_ldul_mmu,
1390};
1391
1392/* helper signature: helper_ret_st_mmu(CPUState *env, target_ulong addr,
1393 *                                     uintxx_t val, int mmu_idx, uintptr_t ra)
1394 */
1395static void * const qemu_st_helpers[16] = {
1396    [MO_UB]   = helper_ret_stb_mmu,
1397    [MO_LEUW] = helper_le_stw_mmu,
1398    [MO_LEUL] = helper_le_stl_mmu,
1399    [MO_LEQ]  = helper_le_stq_mmu,
1400    [MO_BEUW] = helper_be_stw_mmu,
1401    [MO_BEUL] = helper_be_stl_mmu,
1402    [MO_BEQ]  = helper_be_stq_mmu,
1403};
1404
1405/* Helper routines for marshalling helper function arguments into
1406 * the correct registers and stack.
1407 * argreg is where we want to put this argument, arg is the argument itself.
1408 * Return value is the updated argreg ready for the next call.
1409 * Note that argreg 0..3 is real registers, 4+ on stack.
1410 *
1411 * We provide routines for arguments which are: immediate, 32 bit
1412 * value in register, 16 and 8 bit values in register (which must be zero
1413 * extended before use) and 64 bit value in a lo:hi register pair.
1414 */
1415#define DEFINE_TCG_OUT_ARG(NAME, ARGTYPE, MOV_ARG, EXT_ARG)                \
1416static TCGReg NAME(TCGContext *s, TCGReg argreg, ARGTYPE arg)              \
1417{                                                                          \
1418    if (argreg < 4) {                                                      \
1419        MOV_ARG(s, COND_AL, argreg, arg);                                  \
1420    } else {                                                               \
1421        int ofs = (argreg - 4) * 4;                                        \
1422        EXT_ARG;                                                           \
1423        tcg_debug_assert(ofs + 4 <= TCG_STATIC_CALL_ARGS_SIZE);            \
1424        tcg_out_st32_12(s, COND_AL, arg, TCG_REG_CALL_STACK, ofs);         \
1425    }                                                                      \
1426    return argreg + 1;                                                     \
1427}
1428
1429DEFINE_TCG_OUT_ARG(tcg_out_arg_imm32, uint32_t, tcg_out_movi32,
1430    (tcg_out_movi32(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1431DEFINE_TCG_OUT_ARG(tcg_out_arg_reg8, TCGReg, tcg_out_ext8u,
1432    (tcg_out_ext8u(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1433DEFINE_TCG_OUT_ARG(tcg_out_arg_reg16, TCGReg, tcg_out_ext16u,
1434    (tcg_out_ext16u(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1435DEFINE_TCG_OUT_ARG(tcg_out_arg_reg32, TCGReg, tcg_out_mov_reg, )
1436
1437static TCGReg tcg_out_arg_reg64(TCGContext *s, TCGReg argreg,
1438                                TCGReg arglo, TCGReg arghi)
1439{
1440    /* 64 bit arguments must go in even/odd register pairs
1441     * and in 8-aligned stack slots.
1442     */
1443    if (argreg & 1) {
1444        argreg++;
1445    }
1446    if (use_armv6_instructions && argreg >= 4
1447        && (arglo & 1) == 0 && arghi == arglo + 1) {
1448        tcg_out_strd_8(s, COND_AL, arglo,
1449                       TCG_REG_CALL_STACK, (argreg - 4) * 4);
1450        return argreg + 2;
1451    } else {
1452        argreg = tcg_out_arg_reg32(s, argreg, arglo);
1453        argreg = tcg_out_arg_reg32(s, argreg, arghi);
1454        return argreg;
1455    }
1456}
1457
1458#define TLB_SHIFT	(CPU_TLB_ENTRY_BITS + CPU_TLB_BITS)
1459
1460/* We expect to use an 9-bit sign-magnitude negative offset from ENV.  */
1461QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
1462QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -256);
1463
1464/* These offsets are built into the LDRD below.  */
1465QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, mask) != 0);
1466QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, table) != 4);
1467
1468/* Load and compare a TLB entry, leaving the flags set.  Returns the register
1469   containing the addend of the tlb entry.  Clobbers R0, R1, R2, TMP.  */
1470
1471static TCGReg tcg_out_tlb_read(TCGContext *s, TCGReg addrlo, TCGReg addrhi,
1472                               MemOp opc, int mem_index, bool is_load)
1473{
1474    int cmp_off = (is_load ? offsetof(CPUTLBEntry, addr_read)
1475                   : offsetof(CPUTLBEntry, addr_write));
1476    int fast_off = TLB_MASK_TABLE_OFS(mem_index);
1477    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1478    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1479    unsigned s_bits = opc & MO_SIZE;
1480    unsigned a_bits = get_alignment_bits(opc);
1481
1482    /*
1483     * We don't support inline unaligned acceses, but we can easily
1484     * support overalignment checks.
1485     */
1486    if (a_bits < s_bits) {
1487        a_bits = s_bits;
1488    }
1489
1490    /* Load env_tlb(env)->f[mmu_idx].{mask,table} into {r0,r1}.  */
1491    if (use_armv6_instructions) {
1492        tcg_out_ldrd_8(s, COND_AL, TCG_REG_R0, TCG_AREG0, fast_off);
1493    } else {
1494        tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R0, TCG_AREG0, mask_off);
1495        tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R1, TCG_AREG0, table_off);
1496    }
1497
1498    /* Extract the tlb index from the address into R0.  */
1499    tcg_out_dat_reg(s, COND_AL, ARITH_AND, TCG_REG_R0, TCG_REG_R0, addrlo,
1500                    SHIFT_IMM_LSR(TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS));
1501
1502    /*
1503     * Add the tlb_table pointer, creating the CPUTLBEntry address in R1.
1504     * Load the tlb comparator into R2/R3 and the fast path addend into R1.
1505     */
1506    if (cmp_off == 0) {
1507        if (use_armv6_instructions && TARGET_LONG_BITS == 64) {
1508            tcg_out_ldrd_rwb(s, COND_AL, TCG_REG_R2, TCG_REG_R1, TCG_REG_R0);
1509        } else {
1510            tcg_out_ld32_rwb(s, COND_AL, TCG_REG_R2, TCG_REG_R1, TCG_REG_R0);
1511        }
1512    } else {
1513        tcg_out_dat_reg(s, COND_AL, ARITH_ADD,
1514                        TCG_REG_R1, TCG_REG_R1, TCG_REG_R0, 0);
1515        if (use_armv6_instructions && TARGET_LONG_BITS == 64) {
1516            tcg_out_ldrd_8(s, COND_AL, TCG_REG_R2, TCG_REG_R1, cmp_off);
1517        } else {
1518            tcg_out_ld32_12(s, COND_AL, TCG_REG_R2, TCG_REG_R1, cmp_off);
1519        }
1520    }
1521    if (!use_armv6_instructions && TARGET_LONG_BITS == 64) {
1522        tcg_out_ld32_12(s, COND_AL, TCG_REG_R3, TCG_REG_R1, cmp_off + 4);
1523    }
1524
1525    /* Load the tlb addend.  */
1526    tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R1,
1527                    offsetof(CPUTLBEntry, addend));
1528
1529    /*
1530     * Check alignment, check comparators.
1531     * Do this in no more than 3 insns.  Use MOVW for v7, if possible,
1532     * to reduce the number of sequential conditional instructions.
1533     * Almost all guests have at least 4k pages, which means that we need
1534     * to clear at least 9 bits even for an 8-byte memory, which means it
1535     * isn't worth checking for an immediate operand for BIC.
1536     */
1537    if (use_armv7_instructions && TARGET_PAGE_BITS <= 16) {
1538        tcg_target_ulong mask = ~(TARGET_PAGE_MASK | ((1 << a_bits) - 1));
1539
1540        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, mask);
1541        tcg_out_dat_reg(s, COND_AL, ARITH_BIC, TCG_REG_TMP,
1542                        addrlo, TCG_REG_TMP, 0);
1543        tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, TCG_REG_R2, TCG_REG_TMP, 0);
1544    } else {
1545        if (a_bits) {
1546            tcg_out_dat_imm(s, COND_AL, ARITH_TST, 0, addrlo,
1547                            (1 << a_bits) - 1);
1548        }
1549        tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_TMP, 0, addrlo,
1550                        SHIFT_IMM_LSR(TARGET_PAGE_BITS));
1551        tcg_out_dat_reg(s, (a_bits ? COND_EQ : COND_AL), ARITH_CMP,
1552                        0, TCG_REG_R2, TCG_REG_TMP,
1553                        SHIFT_IMM_LSL(TARGET_PAGE_BITS));
1554    }
1555
1556    if (TARGET_LONG_BITS == 64) {
1557        tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0, TCG_REG_R3, addrhi, 0);
1558    }
1559
1560    return TCG_REG_R1;
1561}
1562
1563/* Record the context of a call to the out of line helper code for the slow
1564   path for a load or store, so that we can later generate the correct
1565   helper code.  */
1566static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
1567                                TCGReg datalo, TCGReg datahi, TCGReg addrlo,
1568                                TCGReg addrhi, tcg_insn_unit *raddr,
1569                                tcg_insn_unit *label_ptr)
1570{
1571    TCGLabelQemuLdst *label = new_ldst_label(s);
1572
1573    label->is_ld = is_ld;
1574    label->oi = oi;
1575    label->datalo_reg = datalo;
1576    label->datahi_reg = datahi;
1577    label->addrlo_reg = addrlo;
1578    label->addrhi_reg = addrhi;
1579    label->raddr = tcg_splitwx_to_rx(raddr);
1580    label->label_ptr[0] = label_ptr;
1581}
1582
1583static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1584{
1585    TCGReg argreg, datalo, datahi;
1586    TCGMemOpIdx oi = lb->oi;
1587    MemOp opc = get_memop(oi);
1588    void *func;
1589
1590    if (!reloc_pc24(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1591        return false;
1592    }
1593
1594    argreg = tcg_out_arg_reg32(s, TCG_REG_R0, TCG_AREG0);
1595    if (TARGET_LONG_BITS == 64) {
1596        argreg = tcg_out_arg_reg64(s, argreg, lb->addrlo_reg, lb->addrhi_reg);
1597    } else {
1598        argreg = tcg_out_arg_reg32(s, argreg, lb->addrlo_reg);
1599    }
1600    argreg = tcg_out_arg_imm32(s, argreg, oi);
1601    argreg = tcg_out_arg_reg32(s, argreg, TCG_REG_R14);
1602
1603    /* For armv6 we can use the canonical unsigned helpers and minimize
1604       icache usage.  For pre-armv6, use the signed helpers since we do
1605       not have a single insn sign-extend.  */
1606    if (use_armv6_instructions) {
1607        func = qemu_ld_helpers[opc & (MO_BSWAP | MO_SIZE)];
1608    } else {
1609        func = qemu_ld_helpers[opc & (MO_BSWAP | MO_SSIZE)];
1610        if (opc & MO_SIGN) {
1611            opc = MO_UL;
1612        }
1613    }
1614    tcg_out_call(s, func);
1615
1616    datalo = lb->datalo_reg;
1617    datahi = lb->datahi_reg;
1618    switch (opc & MO_SSIZE) {
1619    case MO_SB:
1620        tcg_out_ext8s(s, COND_AL, datalo, TCG_REG_R0);
1621        break;
1622    case MO_SW:
1623        tcg_out_ext16s(s, COND_AL, datalo, TCG_REG_R0);
1624        break;
1625    default:
1626        tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1627        break;
1628    case MO_Q:
1629        if (datalo != TCG_REG_R1) {
1630            tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1631            tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1632        } else if (datahi != TCG_REG_R0) {
1633            tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1634            tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1635        } else {
1636            tcg_out_mov_reg(s, COND_AL, TCG_REG_TMP, TCG_REG_R0);
1637            tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1638            tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_TMP);
1639        }
1640        break;
1641    }
1642
1643    tcg_out_goto(s, COND_AL, lb->raddr);
1644    return true;
1645}
1646
1647static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1648{
1649    TCGReg argreg, datalo, datahi;
1650    TCGMemOpIdx oi = lb->oi;
1651    MemOp opc = get_memop(oi);
1652
1653    if (!reloc_pc24(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1654        return false;
1655    }
1656
1657    argreg = TCG_REG_R0;
1658    argreg = tcg_out_arg_reg32(s, argreg, TCG_AREG0);
1659    if (TARGET_LONG_BITS == 64) {
1660        argreg = tcg_out_arg_reg64(s, argreg, lb->addrlo_reg, lb->addrhi_reg);
1661    } else {
1662        argreg = tcg_out_arg_reg32(s, argreg, lb->addrlo_reg);
1663    }
1664
1665    datalo = lb->datalo_reg;
1666    datahi = lb->datahi_reg;
1667    switch (opc & MO_SIZE) {
1668    case MO_8:
1669        argreg = tcg_out_arg_reg8(s, argreg, datalo);
1670        break;
1671    case MO_16:
1672        argreg = tcg_out_arg_reg16(s, argreg, datalo);
1673        break;
1674    case MO_32:
1675    default:
1676        argreg = tcg_out_arg_reg32(s, argreg, datalo);
1677        break;
1678    case MO_64:
1679        argreg = tcg_out_arg_reg64(s, argreg, datalo, datahi);
1680        break;
1681    }
1682
1683    argreg = tcg_out_arg_imm32(s, argreg, oi);
1684    argreg = tcg_out_arg_reg32(s, argreg, TCG_REG_R14);
1685
1686    /* Tail-call to the helper, which will return to the fast path.  */
1687    tcg_out_goto(s, COND_AL, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
1688    return true;
1689}
1690#endif /* SOFTMMU */
1691
1692static inline void tcg_out_qemu_ld_index(TCGContext *s, MemOp opc,
1693                                         TCGReg datalo, TCGReg datahi,
1694                                         TCGReg addrlo, TCGReg addend)
1695{
1696    MemOp bswap = opc & MO_BSWAP;
1697
1698    switch (opc & MO_SSIZE) {
1699    case MO_UB:
1700        tcg_out_ld8_r(s, COND_AL, datalo, addrlo, addend);
1701        break;
1702    case MO_SB:
1703        tcg_out_ld8s_r(s, COND_AL, datalo, addrlo, addend);
1704        break;
1705    case MO_UW:
1706        tcg_out_ld16u_r(s, COND_AL, datalo, addrlo, addend);
1707        if (bswap) {
1708            tcg_out_bswap16(s, COND_AL, datalo, datalo);
1709        }
1710        break;
1711    case MO_SW:
1712        if (bswap) {
1713            tcg_out_ld16u_r(s, COND_AL, datalo, addrlo, addend);
1714            tcg_out_bswap16s(s, COND_AL, datalo, datalo);
1715        } else {
1716            tcg_out_ld16s_r(s, COND_AL, datalo, addrlo, addend);
1717        }
1718        break;
1719    case MO_UL:
1720    default:
1721        tcg_out_ld32_r(s, COND_AL, datalo, addrlo, addend);
1722        if (bswap) {
1723            tcg_out_bswap32(s, COND_AL, datalo, datalo);
1724        }
1725        break;
1726    case MO_Q:
1727        {
1728            TCGReg dl = (bswap ? datahi : datalo);
1729            TCGReg dh = (bswap ? datalo : datahi);
1730
1731            /* Avoid ldrd for user-only emulation, to handle unaligned.  */
1732            if (USING_SOFTMMU && use_armv6_instructions
1733                && (dl & 1) == 0 && dh == dl + 1) {
1734                tcg_out_ldrd_r(s, COND_AL, dl, addrlo, addend);
1735            } else if (dl != addend) {
1736                tcg_out_ld32_rwb(s, COND_AL, dl, addend, addrlo);
1737                tcg_out_ld32_12(s, COND_AL, dh, addend, 4);
1738            } else {
1739                tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_TMP,
1740                                addend, addrlo, SHIFT_IMM_LSL(0));
1741                tcg_out_ld32_12(s, COND_AL, dl, TCG_REG_TMP, 0);
1742                tcg_out_ld32_12(s, COND_AL, dh, TCG_REG_TMP, 4);
1743            }
1744            if (bswap) {
1745                tcg_out_bswap32(s, COND_AL, dl, dl);
1746                tcg_out_bswap32(s, COND_AL, dh, dh);
1747            }
1748        }
1749        break;
1750    }
1751}
1752
1753static inline void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc,
1754                                          TCGReg datalo, TCGReg datahi,
1755                                          TCGReg addrlo)
1756{
1757    MemOp bswap = opc & MO_BSWAP;
1758
1759    switch (opc & MO_SSIZE) {
1760    case MO_UB:
1761        tcg_out_ld8_12(s, COND_AL, datalo, addrlo, 0);
1762        break;
1763    case MO_SB:
1764        tcg_out_ld8s_8(s, COND_AL, datalo, addrlo, 0);
1765        break;
1766    case MO_UW:
1767        tcg_out_ld16u_8(s, COND_AL, datalo, addrlo, 0);
1768        if (bswap) {
1769            tcg_out_bswap16(s, COND_AL, datalo, datalo);
1770        }
1771        break;
1772    case MO_SW:
1773        if (bswap) {
1774            tcg_out_ld16u_8(s, COND_AL, datalo, addrlo, 0);
1775            tcg_out_bswap16s(s, COND_AL, datalo, datalo);
1776        } else {
1777            tcg_out_ld16s_8(s, COND_AL, datalo, addrlo, 0);
1778        }
1779        break;
1780    case MO_UL:
1781    default:
1782        tcg_out_ld32_12(s, COND_AL, datalo, addrlo, 0);
1783        if (bswap) {
1784            tcg_out_bswap32(s, COND_AL, datalo, datalo);
1785        }
1786        break;
1787    case MO_Q:
1788        {
1789            TCGReg dl = (bswap ? datahi : datalo);
1790            TCGReg dh = (bswap ? datalo : datahi);
1791
1792            /* Avoid ldrd for user-only emulation, to handle unaligned.  */
1793            if (USING_SOFTMMU && use_armv6_instructions
1794                && (dl & 1) == 0 && dh == dl + 1) {
1795                tcg_out_ldrd_8(s, COND_AL, dl, addrlo, 0);
1796            } else if (dl == addrlo) {
1797                tcg_out_ld32_12(s, COND_AL, dh, addrlo, bswap ? 0 : 4);
1798                tcg_out_ld32_12(s, COND_AL, dl, addrlo, bswap ? 4 : 0);
1799            } else {
1800                tcg_out_ld32_12(s, COND_AL, dl, addrlo, bswap ? 4 : 0);
1801                tcg_out_ld32_12(s, COND_AL, dh, addrlo, bswap ? 0 : 4);
1802            }
1803            if (bswap) {
1804                tcg_out_bswap32(s, COND_AL, dl, dl);
1805                tcg_out_bswap32(s, COND_AL, dh, dh);
1806            }
1807        }
1808        break;
1809    }
1810}
1811
1812static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is64)
1813{
1814    TCGReg addrlo, datalo, datahi, addrhi __attribute__((unused));
1815    TCGMemOpIdx oi;
1816    MemOp opc;
1817#ifdef CONFIG_SOFTMMU
1818    int mem_index;
1819    TCGReg addend;
1820    tcg_insn_unit *label_ptr;
1821#endif
1822
1823    datalo = *args++;
1824    datahi = (is64 ? *args++ : 0);
1825    addrlo = *args++;
1826    addrhi = (TARGET_LONG_BITS == 64 ? *args++ : 0);
1827    oi = *args++;
1828    opc = get_memop(oi);
1829
1830#ifdef CONFIG_SOFTMMU
1831    mem_index = get_mmuidx(oi);
1832    addend = tcg_out_tlb_read(s, addrlo, addrhi, opc, mem_index, 1);
1833
1834    /* This a conditional BL only to load a pointer within this opcode into LR
1835       for the slow path.  We will not be using the value for a tail call.  */
1836    label_ptr = s->code_ptr;
1837    tcg_out_bl(s, COND_NE, 0);
1838
1839    tcg_out_qemu_ld_index(s, opc, datalo, datahi, addrlo, addend);
1840
1841    add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
1842                        s->code_ptr, label_ptr);
1843#else /* !CONFIG_SOFTMMU */
1844    if (guest_base) {
1845        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, guest_base);
1846        tcg_out_qemu_ld_index(s, opc, datalo, datahi, addrlo, TCG_REG_TMP);
1847    } else {
1848        tcg_out_qemu_ld_direct(s, opc, datalo, datahi, addrlo);
1849    }
1850#endif
1851}
1852
1853static inline void tcg_out_qemu_st_index(TCGContext *s, int cond, MemOp opc,
1854                                         TCGReg datalo, TCGReg datahi,
1855                                         TCGReg addrlo, TCGReg addend)
1856{
1857    MemOp bswap = opc & MO_BSWAP;
1858
1859    switch (opc & MO_SIZE) {
1860    case MO_8:
1861        tcg_out_st8_r(s, cond, datalo, addrlo, addend);
1862        break;
1863    case MO_16:
1864        if (bswap) {
1865            tcg_out_bswap16st(s, cond, TCG_REG_R0, datalo);
1866            tcg_out_st16_r(s, cond, TCG_REG_R0, addrlo, addend);
1867        } else {
1868            tcg_out_st16_r(s, cond, datalo, addrlo, addend);
1869        }
1870        break;
1871    case MO_32:
1872    default:
1873        if (bswap) {
1874            tcg_out_bswap32(s, cond, TCG_REG_R0, datalo);
1875            tcg_out_st32_r(s, cond, TCG_REG_R0, addrlo, addend);
1876        } else {
1877            tcg_out_st32_r(s, cond, datalo, addrlo, addend);
1878        }
1879        break;
1880    case MO_64:
1881        /* Avoid strd for user-only emulation, to handle unaligned.  */
1882        if (bswap) {
1883            tcg_out_bswap32(s, cond, TCG_REG_R0, datahi);
1884            tcg_out_st32_rwb(s, cond, TCG_REG_R0, addend, addrlo);
1885            tcg_out_bswap32(s, cond, TCG_REG_R0, datalo);
1886            tcg_out_st32_12(s, cond, TCG_REG_R0, addend, 4);
1887        } else if (USING_SOFTMMU && use_armv6_instructions
1888                   && (datalo & 1) == 0 && datahi == datalo + 1) {
1889            tcg_out_strd_r(s, cond, datalo, addrlo, addend);
1890        } else {
1891            tcg_out_st32_rwb(s, cond, datalo, addend, addrlo);
1892            tcg_out_st32_12(s, cond, datahi, addend, 4);
1893        }
1894        break;
1895    }
1896}
1897
1898static inline void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc,
1899                                          TCGReg datalo, TCGReg datahi,
1900                                          TCGReg addrlo)
1901{
1902    MemOp bswap = opc & MO_BSWAP;
1903
1904    switch (opc & MO_SIZE) {
1905    case MO_8:
1906        tcg_out_st8_12(s, COND_AL, datalo, addrlo, 0);
1907        break;
1908    case MO_16:
1909        if (bswap) {
1910            tcg_out_bswap16st(s, COND_AL, TCG_REG_R0, datalo);
1911            tcg_out_st16_8(s, COND_AL, TCG_REG_R0, addrlo, 0);
1912        } else {
1913            tcg_out_st16_8(s, COND_AL, datalo, addrlo, 0);
1914        }
1915        break;
1916    case MO_32:
1917    default:
1918        if (bswap) {
1919            tcg_out_bswap32(s, COND_AL, TCG_REG_R0, datalo);
1920            tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addrlo, 0);
1921        } else {
1922            tcg_out_st32_12(s, COND_AL, datalo, addrlo, 0);
1923        }
1924        break;
1925    case MO_64:
1926        /* Avoid strd for user-only emulation, to handle unaligned.  */
1927        if (bswap) {
1928            tcg_out_bswap32(s, COND_AL, TCG_REG_R0, datahi);
1929            tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addrlo, 0);
1930            tcg_out_bswap32(s, COND_AL, TCG_REG_R0, datalo);
1931            tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addrlo, 4);
1932        } else if (USING_SOFTMMU && use_armv6_instructions
1933                   && (datalo & 1) == 0 && datahi == datalo + 1) {
1934            tcg_out_strd_8(s, COND_AL, datalo, addrlo, 0);
1935        } else {
1936            tcg_out_st32_12(s, COND_AL, datalo, addrlo, 0);
1937            tcg_out_st32_12(s, COND_AL, datahi, addrlo, 4);
1938        }
1939        break;
1940    }
1941}
1942
1943static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is64)
1944{
1945    TCGReg addrlo, datalo, datahi, addrhi __attribute__((unused));
1946    TCGMemOpIdx oi;
1947    MemOp opc;
1948#ifdef CONFIG_SOFTMMU
1949    int mem_index;
1950    TCGReg addend;
1951    tcg_insn_unit *label_ptr;
1952#endif
1953
1954    datalo = *args++;
1955    datahi = (is64 ? *args++ : 0);
1956    addrlo = *args++;
1957    addrhi = (TARGET_LONG_BITS == 64 ? *args++ : 0);
1958    oi = *args++;
1959    opc = get_memop(oi);
1960
1961#ifdef CONFIG_SOFTMMU
1962    mem_index = get_mmuidx(oi);
1963    addend = tcg_out_tlb_read(s, addrlo, addrhi, opc, mem_index, 0);
1964
1965    tcg_out_qemu_st_index(s, COND_EQ, opc, datalo, datahi, addrlo, addend);
1966
1967    /* The conditional call must come last, as we're going to return here.  */
1968    label_ptr = s->code_ptr;
1969    tcg_out_bl(s, COND_NE, 0);
1970
1971    add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
1972                        s->code_ptr, label_ptr);
1973#else /* !CONFIG_SOFTMMU */
1974    if (guest_base) {
1975        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, guest_base);
1976        tcg_out_qemu_st_index(s, COND_AL, opc, datalo,
1977                              datahi, addrlo, TCG_REG_TMP);
1978    } else {
1979        tcg_out_qemu_st_direct(s, opc, datalo, datahi, addrlo);
1980    }
1981#endif
1982}
1983
1984static void tcg_out_epilogue(TCGContext *s);
1985
1986static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
1987                const TCGArg args[TCG_MAX_OP_ARGS],
1988                const int const_args[TCG_MAX_OP_ARGS])
1989{
1990    TCGArg a0, a1, a2, a3, a4, a5;
1991    int c;
1992
1993    switch (opc) {
1994    case INDEX_op_exit_tb:
1995        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, args[0]);
1996        tcg_out_epilogue(s);
1997        break;
1998    case INDEX_op_goto_tb:
1999        {
2000            /* Indirect jump method */
2001            intptr_t ptr, dif, dil;
2002            TCGReg base = TCG_REG_PC;
2003
2004            tcg_debug_assert(s->tb_jmp_insn_offset == 0);
2005            ptr = (intptr_t)tcg_splitwx_to_rx(s->tb_jmp_target_addr + args[0]);
2006            dif = tcg_pcrel_diff(s, (void *)ptr) - 8;
2007            dil = sextract32(dif, 0, 12);
2008            if (dif != dil) {
2009                /* The TB is close, but outside the 12 bits addressable by
2010                   the load.  We can extend this to 20 bits with a sub of a
2011                   shifted immediate from pc.  In the vastly unlikely event
2012                   the code requires more than 1MB, we'll use 2 insns and
2013                   be no worse off.  */
2014                base = TCG_REG_R0;
2015                tcg_out_movi32(s, COND_AL, base, ptr - dil);
2016            }
2017            tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, base, dil);
2018            set_jmp_reset_offset(s, args[0]);
2019        }
2020        break;
2021    case INDEX_op_goto_ptr:
2022        tcg_out_bx(s, COND_AL, args[0]);
2023        break;
2024    case INDEX_op_br:
2025        tcg_out_goto_label(s, COND_AL, arg_label(args[0]));
2026        break;
2027
2028    case INDEX_op_ld8u_i32:
2029        tcg_out_ld8u(s, COND_AL, args[0], args[1], args[2]);
2030        break;
2031    case INDEX_op_ld8s_i32:
2032        tcg_out_ld8s(s, COND_AL, args[0], args[1], args[2]);
2033        break;
2034    case INDEX_op_ld16u_i32:
2035        tcg_out_ld16u(s, COND_AL, args[0], args[1], args[2]);
2036        break;
2037    case INDEX_op_ld16s_i32:
2038        tcg_out_ld16s(s, COND_AL, args[0], args[1], args[2]);
2039        break;
2040    case INDEX_op_ld_i32:
2041        tcg_out_ld32u(s, COND_AL, args[0], args[1], args[2]);
2042        break;
2043    case INDEX_op_st8_i32:
2044        tcg_out_st8(s, COND_AL, args[0], args[1], args[2]);
2045        break;
2046    case INDEX_op_st16_i32:
2047        tcg_out_st16(s, COND_AL, args[0], args[1], args[2]);
2048        break;
2049    case INDEX_op_st_i32:
2050        tcg_out_st32(s, COND_AL, args[0], args[1], args[2]);
2051        break;
2052
2053    case INDEX_op_movcond_i32:
2054        /* Constraints mean that v2 is always in the same register as dest,
2055         * so we only need to do "if condition passed, move v1 to dest".
2056         */
2057        tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
2058                        args[1], args[2], const_args[2]);
2059        tcg_out_dat_rIK(s, tcg_cond_to_arm_cond[args[5]], ARITH_MOV,
2060                        ARITH_MVN, args[0], 0, args[3], const_args[3]);
2061        break;
2062    case INDEX_op_add_i32:
2063        tcg_out_dat_rIN(s, COND_AL, ARITH_ADD, ARITH_SUB,
2064                        args[0], args[1], args[2], const_args[2]);
2065        break;
2066    case INDEX_op_sub_i32:
2067        if (const_args[1]) {
2068            if (const_args[2]) {
2069                tcg_out_movi32(s, COND_AL, args[0], args[1] - args[2]);
2070            } else {
2071                tcg_out_dat_rI(s, COND_AL, ARITH_RSB,
2072                               args[0], args[2], args[1], 1);
2073            }
2074        } else {
2075            tcg_out_dat_rIN(s, COND_AL, ARITH_SUB, ARITH_ADD,
2076                            args[0], args[1], args[2], const_args[2]);
2077        }
2078        break;
2079    case INDEX_op_and_i32:
2080        tcg_out_dat_rIK(s, COND_AL, ARITH_AND, ARITH_BIC,
2081                        args[0], args[1], args[2], const_args[2]);
2082        break;
2083    case INDEX_op_andc_i32:
2084        tcg_out_dat_rIK(s, COND_AL, ARITH_BIC, ARITH_AND,
2085                        args[0], args[1], args[2], const_args[2]);
2086        break;
2087    case INDEX_op_or_i32:
2088        c = ARITH_ORR;
2089        goto gen_arith;
2090    case INDEX_op_xor_i32:
2091        c = ARITH_EOR;
2092        /* Fall through.  */
2093    gen_arith:
2094        tcg_out_dat_rI(s, COND_AL, c, args[0], args[1], args[2], const_args[2]);
2095        break;
2096    case INDEX_op_add2_i32:
2097        a0 = args[0], a1 = args[1], a2 = args[2];
2098        a3 = args[3], a4 = args[4], a5 = args[5];
2099        if (a0 == a3 || (a0 == a5 && !const_args[5])) {
2100            a0 = TCG_REG_TMP;
2101        }
2102        tcg_out_dat_rIN(s, COND_AL, ARITH_ADD | TO_CPSR, ARITH_SUB | TO_CPSR,
2103                        a0, a2, a4, const_args[4]);
2104        tcg_out_dat_rIK(s, COND_AL, ARITH_ADC, ARITH_SBC,
2105                        a1, a3, a5, const_args[5]);
2106        tcg_out_mov_reg(s, COND_AL, args[0], a0);
2107        break;
2108    case INDEX_op_sub2_i32:
2109        a0 = args[0], a1 = args[1], a2 = args[2];
2110        a3 = args[3], a4 = args[4], a5 = args[5];
2111        if ((a0 == a3 && !const_args[3]) || (a0 == a5 && !const_args[5])) {
2112            a0 = TCG_REG_TMP;
2113        }
2114        if (const_args[2]) {
2115            if (const_args[4]) {
2116                tcg_out_movi32(s, COND_AL, a0, a4);
2117                a4 = a0;
2118            }
2119            tcg_out_dat_rI(s, COND_AL, ARITH_RSB | TO_CPSR, a0, a4, a2, 1);
2120        } else {
2121            tcg_out_dat_rIN(s, COND_AL, ARITH_SUB | TO_CPSR,
2122                            ARITH_ADD | TO_CPSR, a0, a2, a4, const_args[4]);
2123        }
2124        if (const_args[3]) {
2125            if (const_args[5]) {
2126                tcg_out_movi32(s, COND_AL, a1, a5);
2127                a5 = a1;
2128            }
2129            tcg_out_dat_rI(s, COND_AL, ARITH_RSC, a1, a5, a3, 1);
2130        } else {
2131            tcg_out_dat_rIK(s, COND_AL, ARITH_SBC, ARITH_ADC,
2132                            a1, a3, a5, const_args[5]);
2133        }
2134        tcg_out_mov_reg(s, COND_AL, args[0], a0);
2135        break;
2136    case INDEX_op_neg_i32:
2137        tcg_out_dat_imm(s, COND_AL, ARITH_RSB, args[0], args[1], 0);
2138        break;
2139    case INDEX_op_not_i32:
2140        tcg_out_dat_reg(s, COND_AL,
2141                        ARITH_MVN, args[0], 0, args[1], SHIFT_IMM_LSL(0));
2142        break;
2143    case INDEX_op_mul_i32:
2144        tcg_out_mul32(s, COND_AL, args[0], args[1], args[2]);
2145        break;
2146    case INDEX_op_mulu2_i32:
2147        tcg_out_umull32(s, COND_AL, args[0], args[1], args[2], args[3]);
2148        break;
2149    case INDEX_op_muls2_i32:
2150        tcg_out_smull32(s, COND_AL, args[0], args[1], args[2], args[3]);
2151        break;
2152    /* XXX: Perhaps args[2] & 0x1f is wrong */
2153    case INDEX_op_shl_i32:
2154        c = const_args[2] ?
2155                SHIFT_IMM_LSL(args[2] & 0x1f) : SHIFT_REG_LSL(args[2]);
2156        goto gen_shift32;
2157    case INDEX_op_shr_i32:
2158        c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_LSR(args[2] & 0x1f) :
2159                SHIFT_IMM_LSL(0) : SHIFT_REG_LSR(args[2]);
2160        goto gen_shift32;
2161    case INDEX_op_sar_i32:
2162        c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ASR(args[2] & 0x1f) :
2163                SHIFT_IMM_LSL(0) : SHIFT_REG_ASR(args[2]);
2164        goto gen_shift32;
2165    case INDEX_op_rotr_i32:
2166        c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ROR(args[2] & 0x1f) :
2167                SHIFT_IMM_LSL(0) : SHIFT_REG_ROR(args[2]);
2168        /* Fall through.  */
2169    gen_shift32:
2170        tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1], c);
2171        break;
2172
2173    case INDEX_op_rotl_i32:
2174        if (const_args[2]) {
2175            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
2176                            ((0x20 - args[2]) & 0x1f) ?
2177                            SHIFT_IMM_ROR((0x20 - args[2]) & 0x1f) :
2178                            SHIFT_IMM_LSL(0));
2179        } else {
2180            tcg_out_dat_imm(s, COND_AL, ARITH_RSB, TCG_REG_TMP, args[2], 0x20);
2181            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
2182                            SHIFT_REG_ROR(TCG_REG_TMP));
2183        }
2184        break;
2185
2186    case INDEX_op_ctz_i32:
2187        tcg_out_dat_reg(s, COND_AL, INSN_RBIT, TCG_REG_TMP, 0, args[1], 0);
2188        a1 = TCG_REG_TMP;
2189        goto do_clz;
2190
2191    case INDEX_op_clz_i32:
2192        a1 = args[1];
2193    do_clz:
2194        a0 = args[0];
2195        a2 = args[2];
2196        c = const_args[2];
2197        if (c && a2 == 32) {
2198            tcg_out_dat_reg(s, COND_AL, INSN_CLZ, a0, 0, a1, 0);
2199            break;
2200        }
2201        tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0, a1, 0);
2202        tcg_out_dat_reg(s, COND_NE, INSN_CLZ, a0, 0, a1, 0);
2203        if (c || a0 != a2) {
2204            tcg_out_dat_rIK(s, COND_EQ, ARITH_MOV, ARITH_MVN, a0, 0, a2, c);
2205        }
2206        break;
2207
2208    case INDEX_op_brcond_i32:
2209        tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
2210                       args[0], args[1], const_args[1]);
2211        tcg_out_goto_label(s, tcg_cond_to_arm_cond[args[2]],
2212                           arg_label(args[3]));
2213        break;
2214    case INDEX_op_setcond_i32:
2215        tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
2216                        args[1], args[2], const_args[2]);
2217        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[args[3]],
2218                        ARITH_MOV, args[0], 0, 1);
2219        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(args[3])],
2220                        ARITH_MOV, args[0], 0, 0);
2221        break;
2222
2223    case INDEX_op_brcond2_i32:
2224        c = tcg_out_cmp2(s, args, const_args);
2225        tcg_out_goto_label(s, tcg_cond_to_arm_cond[c], arg_label(args[5]));
2226        break;
2227    case INDEX_op_setcond2_i32:
2228        c = tcg_out_cmp2(s, args + 1, const_args + 1);
2229        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[c], ARITH_MOV, args[0], 0, 1);
2230        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(c)],
2231                        ARITH_MOV, args[0], 0, 0);
2232        break;
2233
2234    case INDEX_op_qemu_ld_i32:
2235        tcg_out_qemu_ld(s, args, 0);
2236        break;
2237    case INDEX_op_qemu_ld_i64:
2238        tcg_out_qemu_ld(s, args, 1);
2239        break;
2240    case INDEX_op_qemu_st_i32:
2241        tcg_out_qemu_st(s, args, 0);
2242        break;
2243    case INDEX_op_qemu_st_i64:
2244        tcg_out_qemu_st(s, args, 1);
2245        break;
2246
2247    case INDEX_op_bswap16_i32:
2248        tcg_out_bswap16(s, COND_AL, args[0], args[1]);
2249        break;
2250    case INDEX_op_bswap32_i32:
2251        tcg_out_bswap32(s, COND_AL, args[0], args[1]);
2252        break;
2253
2254    case INDEX_op_ext8s_i32:
2255        tcg_out_ext8s(s, COND_AL, args[0], args[1]);
2256        break;
2257    case INDEX_op_ext16s_i32:
2258        tcg_out_ext16s(s, COND_AL, args[0], args[1]);
2259        break;
2260    case INDEX_op_ext16u_i32:
2261        tcg_out_ext16u(s, COND_AL, args[0], args[1]);
2262        break;
2263
2264    case INDEX_op_deposit_i32:
2265        tcg_out_deposit(s, COND_AL, args[0], args[2],
2266                        args[3], args[4], const_args[2]);
2267        break;
2268    case INDEX_op_extract_i32:
2269        tcg_out_extract(s, COND_AL, args[0], args[1], args[2], args[3]);
2270        break;
2271    case INDEX_op_sextract_i32:
2272        tcg_out_sextract(s, COND_AL, args[0], args[1], args[2], args[3]);
2273        break;
2274    case INDEX_op_extract2_i32:
2275        /* ??? These optimization vs zero should be generic.  */
2276        /* ??? But we can't substitute 2 for 1 in the opcode stream yet.  */
2277        if (const_args[1]) {
2278            if (const_args[2]) {
2279                tcg_out_movi(s, TCG_TYPE_REG, args[0], 0);
2280            } else {
2281                tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0,
2282                                args[2], SHIFT_IMM_LSL(32 - args[3]));
2283            }
2284        } else if (const_args[2]) {
2285            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0,
2286                            args[1], SHIFT_IMM_LSR(args[3]));
2287        } else {
2288            /* We can do extract2 in 2 insns, vs the 3 required otherwise.  */
2289            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_TMP, 0,
2290                            args[2], SHIFT_IMM_LSL(32 - args[3]));
2291            tcg_out_dat_reg(s, COND_AL, ARITH_ORR, args[0], TCG_REG_TMP,
2292                            args[1], SHIFT_IMM_LSR(args[3]));
2293        }
2294        break;
2295
2296    case INDEX_op_div_i32:
2297        tcg_out_sdiv(s, COND_AL, args[0], args[1], args[2]);
2298        break;
2299    case INDEX_op_divu_i32:
2300        tcg_out_udiv(s, COND_AL, args[0], args[1], args[2]);
2301        break;
2302
2303    case INDEX_op_mb:
2304        tcg_out_mb(s, args[0]);
2305        break;
2306
2307    case INDEX_op_mov_i32:  /* Always emitted via tcg_out_mov.  */
2308    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2309    default:
2310        tcg_abort();
2311    }
2312}
2313
2314static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
2315{
2316    switch (op) {
2317    case INDEX_op_goto_ptr:
2318        return C_O0_I1(r);
2319
2320    case INDEX_op_ld8u_i32:
2321    case INDEX_op_ld8s_i32:
2322    case INDEX_op_ld16u_i32:
2323    case INDEX_op_ld16s_i32:
2324    case INDEX_op_ld_i32:
2325    case INDEX_op_neg_i32:
2326    case INDEX_op_not_i32:
2327    case INDEX_op_bswap16_i32:
2328    case INDEX_op_bswap32_i32:
2329    case INDEX_op_ext8s_i32:
2330    case INDEX_op_ext16s_i32:
2331    case INDEX_op_ext16u_i32:
2332    case INDEX_op_extract_i32:
2333    case INDEX_op_sextract_i32:
2334        return C_O1_I1(r, r);
2335
2336    case INDEX_op_st8_i32:
2337    case INDEX_op_st16_i32:
2338    case INDEX_op_st_i32:
2339        return C_O0_I2(r, r);
2340
2341    case INDEX_op_add_i32:
2342    case INDEX_op_sub_i32:
2343    case INDEX_op_setcond_i32:
2344        return C_O1_I2(r, r, rIN);
2345
2346    case INDEX_op_and_i32:
2347    case INDEX_op_andc_i32:
2348    case INDEX_op_clz_i32:
2349    case INDEX_op_ctz_i32:
2350        return C_O1_I2(r, r, rIK);
2351
2352    case INDEX_op_mul_i32:
2353    case INDEX_op_div_i32:
2354    case INDEX_op_divu_i32:
2355        return C_O1_I2(r, r, r);
2356
2357    case INDEX_op_mulu2_i32:
2358    case INDEX_op_muls2_i32:
2359        return C_O2_I2(r, r, r, r);
2360
2361    case INDEX_op_or_i32:
2362    case INDEX_op_xor_i32:
2363        return C_O1_I2(r, r, rI);
2364
2365    case INDEX_op_shl_i32:
2366    case INDEX_op_shr_i32:
2367    case INDEX_op_sar_i32:
2368    case INDEX_op_rotl_i32:
2369    case INDEX_op_rotr_i32:
2370        return C_O1_I2(r, r, ri);
2371
2372    case INDEX_op_brcond_i32:
2373        return C_O0_I2(r, rIN);
2374    case INDEX_op_deposit_i32:
2375        return C_O1_I2(r, 0, rZ);
2376    case INDEX_op_extract2_i32:
2377        return C_O1_I2(r, rZ, rZ);
2378    case INDEX_op_movcond_i32:
2379        return C_O1_I4(r, r, rIN, rIK, 0);
2380    case INDEX_op_add2_i32:
2381        return C_O2_I4(r, r, r, r, rIN, rIK);
2382    case INDEX_op_sub2_i32:
2383        return C_O2_I4(r, r, rI, rI, rIN, rIK);
2384    case INDEX_op_brcond2_i32:
2385        return C_O0_I4(r, r, rI, rI);
2386    case INDEX_op_setcond2_i32:
2387        return C_O1_I4(r, r, r, rI, rI);
2388
2389    case INDEX_op_qemu_ld_i32:
2390        return TARGET_LONG_BITS == 32 ? C_O1_I1(r, l) : C_O1_I2(r, l, l);
2391    case INDEX_op_qemu_ld_i64:
2392        return TARGET_LONG_BITS == 32 ? C_O2_I1(r, r, l) : C_O2_I2(r, r, l, l);
2393    case INDEX_op_qemu_st_i32:
2394        return TARGET_LONG_BITS == 32 ? C_O0_I2(s, s) : C_O0_I3(s, s, s);
2395    case INDEX_op_qemu_st_i64:
2396        return TARGET_LONG_BITS == 32 ? C_O0_I3(s, s, s) : C_O0_I4(s, s, s, s);
2397
2398    case INDEX_op_st_vec:
2399        return C_O0_I2(w, r);
2400    case INDEX_op_ld_vec:
2401    case INDEX_op_dupm_vec:
2402        return C_O1_I1(w, r);
2403    case INDEX_op_dup_vec:
2404        return C_O1_I1(w, wr);
2405    case INDEX_op_abs_vec:
2406    case INDEX_op_neg_vec:
2407    case INDEX_op_not_vec:
2408    case INDEX_op_shli_vec:
2409    case INDEX_op_shri_vec:
2410    case INDEX_op_sari_vec:
2411        return C_O1_I1(w, w);
2412    case INDEX_op_dup2_vec:
2413    case INDEX_op_add_vec:
2414    case INDEX_op_mul_vec:
2415    case INDEX_op_smax_vec:
2416    case INDEX_op_smin_vec:
2417    case INDEX_op_ssadd_vec:
2418    case INDEX_op_sssub_vec:
2419    case INDEX_op_sub_vec:
2420    case INDEX_op_umax_vec:
2421    case INDEX_op_umin_vec:
2422    case INDEX_op_usadd_vec:
2423    case INDEX_op_ussub_vec:
2424    case INDEX_op_xor_vec:
2425    case INDEX_op_arm_sshl_vec:
2426    case INDEX_op_arm_ushl_vec:
2427        return C_O1_I2(w, w, w);
2428    case INDEX_op_arm_sli_vec:
2429        return C_O1_I2(w, 0, w);
2430    case INDEX_op_or_vec:
2431    case INDEX_op_andc_vec:
2432        return C_O1_I2(w, w, wO);
2433    case INDEX_op_and_vec:
2434    case INDEX_op_orc_vec:
2435        return C_O1_I2(w, w, wV);
2436    case INDEX_op_cmp_vec:
2437        return C_O1_I2(w, w, wZ);
2438    case INDEX_op_bitsel_vec:
2439        return C_O1_I3(w, w, w, w);
2440    default:
2441        g_assert_not_reached();
2442    }
2443}
2444
2445static void tcg_target_init(TCGContext *s)
2446{
2447    /* Only probe for the platform and capabilities if we havn't already
2448       determined maximum values at compile time.  */
2449#if !defined(use_idiv_instructions) || !defined(use_neon_instructions)
2450    {
2451        unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2452#ifndef use_idiv_instructions
2453        use_idiv_instructions = (hwcap & HWCAP_ARM_IDIVA) != 0;
2454#endif
2455#ifndef use_neon_instructions
2456        use_neon_instructions = (hwcap & HWCAP_ARM_NEON) != 0;
2457#endif
2458    }
2459#endif
2460
2461    if (__ARM_ARCH < 7) {
2462        const char *pl = (const char *)qemu_getauxval(AT_PLATFORM);
2463        if (pl != NULL && pl[0] == 'v' && pl[1] >= '4' && pl[1] <= '9') {
2464            arm_arch = pl[1] - '0';
2465        }
2466    }
2467
2468    tcg_target_available_regs[TCG_TYPE_I32] = ALL_GENERAL_REGS;
2469
2470    tcg_target_call_clobber_regs = 0;
2471    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
2472    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
2473    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
2474    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
2475    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
2476    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
2477
2478    if (use_neon_instructions) {
2479        tcg_target_available_regs[TCG_TYPE_V64]  = ALL_VECTOR_REGS;
2480        tcg_target_available_regs[TCG_TYPE_V128] = ALL_VECTOR_REGS;
2481
2482        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q0);
2483        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q1);
2484        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q2);
2485        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q3);
2486        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q8);
2487        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q9);
2488        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q10);
2489        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q11);
2490        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q12);
2491        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q13);
2492        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q14);
2493        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q15);
2494    }
2495
2496    s->reserved_regs = 0;
2497    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
2498    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP);
2499    tcg_regset_set_reg(s->reserved_regs, TCG_REG_PC);
2500    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP);
2501}
2502
2503static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
2504                       TCGReg arg1, intptr_t arg2)
2505{
2506    switch (type) {
2507    case TCG_TYPE_I32:
2508        tcg_out_ld32u(s, COND_AL, arg, arg1, arg2);
2509        return;
2510    case TCG_TYPE_V64:
2511        /* regs 1; size 8; align 8 */
2512        tcg_out_vldst(s, INSN_VLD1 | 0x7d0, arg, arg1, arg2);
2513        return;
2514    case TCG_TYPE_V128:
2515        /* regs 2; size 8; align 16 */
2516        tcg_out_vldst(s, INSN_VLD1 | 0xae0, arg, arg1, arg2);
2517        return;
2518    default:
2519        g_assert_not_reached();
2520    }
2521}
2522
2523static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
2524                       TCGReg arg1, intptr_t arg2)
2525{
2526    switch (type) {
2527    case TCG_TYPE_I32:
2528        tcg_out_st32(s, COND_AL, arg, arg1, arg2);
2529        return;
2530    case TCG_TYPE_V64:
2531        /* regs 1; size 8; align 8 */
2532        tcg_out_vldst(s, INSN_VST1 | 0x7d0, arg, arg1, arg2);
2533        return;
2534    case TCG_TYPE_V128:
2535        /* regs 2; size 8; align 16 */
2536        tcg_out_vldst(s, INSN_VST1 | 0xae0, arg, arg1, arg2);
2537        return;
2538    default:
2539        g_assert_not_reached();
2540    }
2541}
2542
2543static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
2544                               TCGReg base, intptr_t ofs)
2545{
2546    return false;
2547}
2548
2549static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
2550{
2551    if (ret == arg) {
2552        return true;
2553    }
2554    switch (type) {
2555    case TCG_TYPE_I32:
2556        if (ret < TCG_REG_Q0 && arg < TCG_REG_Q0) {
2557            tcg_out_mov_reg(s, COND_AL, ret, arg);
2558            return true;
2559        }
2560        return false;
2561
2562    case TCG_TYPE_V64:
2563    case TCG_TYPE_V128:
2564        /* "VMOV D,N" is an alias for "VORR D,N,N". */
2565        tcg_out_vreg3(s, INSN_VORR, type - TCG_TYPE_V64, 0, ret, arg, arg);
2566        return true;
2567
2568    default:
2569        g_assert_not_reached();
2570    }
2571}
2572
2573static void tcg_out_movi(TCGContext *s, TCGType type,
2574                         TCGReg ret, tcg_target_long arg)
2575{
2576    tcg_debug_assert(type == TCG_TYPE_I32);
2577    tcg_debug_assert(ret < TCG_REG_Q0);
2578    tcg_out_movi32(s, COND_AL, ret, arg);
2579}
2580
2581/* Type is always V128, with I64 elements.  */
2582static void tcg_out_dup2_vec(TCGContext *s, TCGReg rd, TCGReg rl, TCGReg rh)
2583{
2584    /* Move high element into place first. */
2585    /* VMOV Dd+1, Ds */
2586    tcg_out_vreg3(s, INSN_VORR | (1 << 12), 0, 0, rd, rh, rh);
2587    /* Move low element into place; tcg_out_mov will check for nop. */
2588    tcg_out_mov(s, TCG_TYPE_V64, rd, rl);
2589}
2590
2591static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
2592                            TCGReg rd, TCGReg rs)
2593{
2594    int q = type - TCG_TYPE_V64;
2595
2596    if (vece == MO_64) {
2597        if (type == TCG_TYPE_V128) {
2598            tcg_out_dup2_vec(s, rd, rs, rs);
2599        } else {
2600            tcg_out_mov(s, TCG_TYPE_V64, rd, rs);
2601        }
2602    } else if (rs < TCG_REG_Q0) {
2603        int b = (vece == MO_8);
2604        int e = (vece == MO_16);
2605        tcg_out32(s, INSN_VDUP_G | (b << 22) | (q << 21) | (e << 5) |
2606                  encode_vn(rd) | (rs << 12));
2607    } else {
2608        int imm4 = 1 << vece;
2609        tcg_out32(s, INSN_VDUP_S | (imm4 << 16) | (q << 6) |
2610                  encode_vd(rd) | encode_vm(rs));
2611    }
2612    return true;
2613}
2614
2615static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
2616                             TCGReg rd, TCGReg base, intptr_t offset)
2617{
2618    if (vece == MO_64) {
2619        tcg_out_ld(s, TCG_TYPE_V64, rd, base, offset);
2620        if (type == TCG_TYPE_V128) {
2621            tcg_out_dup2_vec(s, rd, rd, rd);
2622        }
2623    } else {
2624        int q = type - TCG_TYPE_V64;
2625        tcg_out_vldst(s, INSN_VLD1R | (vece << 6) | (q << 5),
2626                      rd, base, offset);
2627    }
2628    return true;
2629}
2630
2631static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
2632                             TCGReg rd, int64_t v64)
2633{
2634    int q = type - TCG_TYPE_V64;
2635    int cmode, imm8, i;
2636
2637    /* Test all bytes equal first.  */
2638    if (vece == MO_8) {
2639        tcg_out_vmovi(s, rd, q, 0, 0xe, v64);
2640        return;
2641    }
2642
2643    /*
2644     * Test all bytes 0x00 or 0xff second.  This can match cases that
2645     * might otherwise take 2 or 3 insns for MO_16 or MO_32 below.
2646     */
2647    for (i = imm8 = 0; i < 8; i++) {
2648        uint8_t byte = v64 >> (i * 8);
2649        if (byte == 0xff) {
2650            imm8 |= 1 << i;
2651        } else if (byte != 0) {
2652            goto fail_bytes;
2653        }
2654    }
2655    tcg_out_vmovi(s, rd, q, 1, 0xe, imm8);
2656    return;
2657 fail_bytes:
2658
2659    /*
2660     * Tests for various replications.  For each element width, if we
2661     * cannot find an expansion there's no point checking a larger
2662     * width because we already know by replication it cannot match.
2663     */
2664    if (vece == MO_16) {
2665        uint16_t v16 = v64;
2666
2667        if (is_shimm16(v16, &cmode, &imm8)) {
2668            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2669            return;
2670        }
2671        if (is_shimm16(~v16, &cmode, &imm8)) {
2672            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2673            return;
2674        }
2675
2676        /*
2677         * Otherwise, all remaining constants can be loaded in two insns:
2678         * rd = v16 & 0xff, rd |= v16 & 0xff00.
2679         */
2680        tcg_out_vmovi(s, rd, q, 0, 0x8, v16 & 0xff);
2681        tcg_out_vmovi(s, rd, q, 0, 0xb, v16 >> 8);   /* VORRI */
2682        return;
2683    }
2684
2685    if (vece == MO_32) {
2686        uint32_t v32 = v64;
2687
2688        if (is_shimm32(v32, &cmode, &imm8) ||
2689            is_soimm32(v32, &cmode, &imm8)) {
2690            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2691            return;
2692        }
2693        if (is_shimm32(~v32, &cmode, &imm8) ||
2694            is_soimm32(~v32, &cmode, &imm8)) {
2695            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2696            return;
2697        }
2698
2699        /*
2700         * Restrict the set of constants to those we can load with
2701         * two instructions.  Others we load from the pool.
2702         */
2703        i = is_shimm32_pair(v32, &cmode, &imm8);
2704        if (i) {
2705            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2706            tcg_out_vmovi(s, rd, q, 0, i | 1, extract32(v32, i * 4, 8));
2707            return;
2708        }
2709        i = is_shimm32_pair(~v32, &cmode, &imm8);
2710        if (i) {
2711            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2712            tcg_out_vmovi(s, rd, q, 1, i | 1, extract32(~v32, i * 4, 8));
2713            return;
2714        }
2715    }
2716
2717    /*
2718     * As a last resort, load from the constant pool.
2719     */
2720    if (!q || vece == MO_64) {
2721        new_pool_l2(s, R_ARM_PC11, s->code_ptr, 0, v64, v64 >> 32);
2722        /* VLDR Dd, [pc + offset] */
2723        tcg_out32(s, INSN_VLDR_D | encode_vd(rd) | (0xf << 16));
2724        if (q) {
2725            tcg_out_dup2_vec(s, rd, rd, rd);
2726        }
2727    } else {
2728        new_pool_label(s, (uint32_t)v64, R_ARM_PC8, s->code_ptr, 0);
2729        /* add tmp, pc, offset */
2730        tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_TMP, TCG_REG_PC, 0);
2731        tcg_out_dupm_vec(s, type, MO_32, rd, TCG_REG_TMP, 0);
2732    }
2733}
2734
2735static const ARMInsn vec_cmp_insn[16] = {
2736    [TCG_COND_EQ] = INSN_VCEQ,
2737    [TCG_COND_GT] = INSN_VCGT,
2738    [TCG_COND_GE] = INSN_VCGE,
2739    [TCG_COND_GTU] = INSN_VCGT_U,
2740    [TCG_COND_GEU] = INSN_VCGE_U,
2741};
2742
2743static const ARMInsn vec_cmp0_insn[16] = {
2744    [TCG_COND_EQ] = INSN_VCEQ0,
2745    [TCG_COND_GT] = INSN_VCGT0,
2746    [TCG_COND_GE] = INSN_VCGE0,
2747    [TCG_COND_LT] = INSN_VCLT0,
2748    [TCG_COND_LE] = INSN_VCLE0,
2749};
2750
2751static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2752                           unsigned vecl, unsigned vece,
2753                           const TCGArg *args, const int *const_args)
2754{
2755    TCGType type = vecl + TCG_TYPE_V64;
2756    unsigned q = vecl;
2757    TCGArg a0, a1, a2, a3;
2758    int cmode, imm8;
2759
2760    a0 = args[0];
2761    a1 = args[1];
2762    a2 = args[2];
2763
2764    switch (opc) {
2765    case INDEX_op_ld_vec:
2766        tcg_out_ld(s, type, a0, a1, a2);
2767        return;
2768    case INDEX_op_st_vec:
2769        tcg_out_st(s, type, a0, a1, a2);
2770        return;
2771    case INDEX_op_dupm_vec:
2772        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2773        return;
2774    case INDEX_op_dup2_vec:
2775        tcg_out_dup2_vec(s, a0, a1, a2);
2776        return;
2777    case INDEX_op_abs_vec:
2778        tcg_out_vreg2(s, INSN_VABS, q, vece, a0, a1);
2779        return;
2780    case INDEX_op_neg_vec:
2781        tcg_out_vreg2(s, INSN_VNEG, q, vece, a0, a1);
2782        return;
2783    case INDEX_op_not_vec:
2784        tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a1);
2785        return;
2786    case INDEX_op_add_vec:
2787        tcg_out_vreg3(s, INSN_VADD, q, vece, a0, a1, a2);
2788        return;
2789    case INDEX_op_mul_vec:
2790        tcg_out_vreg3(s, INSN_VMUL, q, vece, a0, a1, a2);
2791        return;
2792    case INDEX_op_smax_vec:
2793        tcg_out_vreg3(s, INSN_VMAX, q, vece, a0, a1, a2);
2794        return;
2795    case INDEX_op_smin_vec:
2796        tcg_out_vreg3(s, INSN_VMIN, q, vece, a0, a1, a2);
2797        return;
2798    case INDEX_op_sub_vec:
2799        tcg_out_vreg3(s, INSN_VSUB, q, vece, a0, a1, a2);
2800        return;
2801    case INDEX_op_ssadd_vec:
2802        tcg_out_vreg3(s, INSN_VQADD, q, vece, a0, a1, a2);
2803        return;
2804    case INDEX_op_sssub_vec:
2805        tcg_out_vreg3(s, INSN_VQSUB, q, vece, a0, a1, a2);
2806        return;
2807    case INDEX_op_umax_vec:
2808        tcg_out_vreg3(s, INSN_VMAX_U, q, vece, a0, a1, a2);
2809        return;
2810    case INDEX_op_umin_vec:
2811        tcg_out_vreg3(s, INSN_VMIN_U, q, vece, a0, a1, a2);
2812        return;
2813    case INDEX_op_usadd_vec:
2814        tcg_out_vreg3(s, INSN_VQADD_U, q, vece, a0, a1, a2);
2815        return;
2816    case INDEX_op_ussub_vec:
2817        tcg_out_vreg3(s, INSN_VQSUB_U, q, vece, a0, a1, a2);
2818        return;
2819    case INDEX_op_xor_vec:
2820        tcg_out_vreg3(s, INSN_VEOR, q, 0, a0, a1, a2);
2821        return;
2822    case INDEX_op_arm_sshl_vec:
2823        /*
2824         * Note that Vm is the data and Vn is the shift count,
2825         * therefore the arguments appear reversed.
2826         */
2827        tcg_out_vreg3(s, INSN_VSHL_S, q, vece, a0, a2, a1);
2828        return;
2829    case INDEX_op_arm_ushl_vec:
2830        /* See above. */
2831        tcg_out_vreg3(s, INSN_VSHL_U, q, vece, a0, a2, a1);
2832        return;
2833    case INDEX_op_shli_vec:
2834        tcg_out_vshifti(s, INSN_VSHLI, q, a0, a1, a2 + (8 << vece));
2835        return;
2836    case INDEX_op_shri_vec:
2837        tcg_out_vshifti(s, INSN_VSHRI, q, a0, a1, (16 << vece) - a2);
2838        return;
2839    case INDEX_op_sari_vec:
2840        tcg_out_vshifti(s, INSN_VSARI, q, a0, a1, (16 << vece) - a2);
2841        return;
2842    case INDEX_op_arm_sli_vec:
2843        tcg_out_vshifti(s, INSN_VSLI, q, a0, a2, args[3] + (8 << vece));
2844        return;
2845
2846    case INDEX_op_andc_vec:
2847        if (!const_args[2]) {
2848            tcg_out_vreg3(s, INSN_VBIC, q, 0, a0, a1, a2);
2849            return;
2850        }
2851        a2 = ~a2;
2852        /* fall through */
2853    case INDEX_op_and_vec:
2854        if (const_args[2]) {
2855            is_shimm1632(~a2, &cmode, &imm8);
2856            if (a0 == a1) {
2857                tcg_out_vmovi(s, a0, q, 1, cmode | 1, imm8); /* VBICI */
2858                return;
2859            }
2860            tcg_out_vmovi(s, a0, q, 1, cmode, imm8); /* VMVNI */
2861            a2 = a0;
2862        }
2863        tcg_out_vreg3(s, INSN_VAND, q, 0, a0, a1, a2);
2864        return;
2865
2866    case INDEX_op_orc_vec:
2867        if (!const_args[2]) {
2868            tcg_out_vreg3(s, INSN_VORN, q, 0, a0, a1, a2);
2869            return;
2870        }
2871        a2 = ~a2;
2872        /* fall through */
2873    case INDEX_op_or_vec:
2874        if (const_args[2]) {
2875            is_shimm1632(a2, &cmode, &imm8);
2876            if (a0 == a1) {
2877                tcg_out_vmovi(s, a0, q, 0, cmode | 1, imm8); /* VORRI */
2878                return;
2879            }
2880            tcg_out_vmovi(s, a0, q, 0, cmode, imm8); /* VMOVI */
2881            a2 = a0;
2882        }
2883        tcg_out_vreg3(s, INSN_VORR, q, 0, a0, a1, a2);
2884        return;
2885
2886    case INDEX_op_cmp_vec:
2887        {
2888            TCGCond cond = args[3];
2889
2890            if (cond == TCG_COND_NE) {
2891                if (const_args[2]) {
2892                    tcg_out_vreg3(s, INSN_VTST, q, vece, a0, a1, a1);
2893                } else {
2894                    tcg_out_vreg3(s, INSN_VCEQ, q, vece, a0, a1, a2);
2895                    tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a0);
2896                }
2897            } else {
2898                ARMInsn insn;
2899
2900                if (const_args[2]) {
2901                    insn = vec_cmp0_insn[cond];
2902                    if (insn) {
2903                        tcg_out_vreg2(s, insn, q, vece, a0, a1);
2904                        return;
2905                    }
2906                    tcg_out_dupi_vec(s, type, MO_8, TCG_VEC_TMP, 0);
2907                    a2 = TCG_VEC_TMP;
2908                }
2909                insn = vec_cmp_insn[cond];
2910                if (insn == 0) {
2911                    TCGArg t;
2912                    t = a1, a1 = a2, a2 = t;
2913                    cond = tcg_swap_cond(cond);
2914                    insn = vec_cmp_insn[cond];
2915                    tcg_debug_assert(insn != 0);
2916                }
2917                tcg_out_vreg3(s, insn, q, vece, a0, a1, a2);
2918            }
2919        }
2920        return;
2921
2922    case INDEX_op_bitsel_vec:
2923        a3 = args[3];
2924        if (a0 == a3) {
2925            tcg_out_vreg3(s, INSN_VBIT, q, 0, a0, a2, a1);
2926        } else if (a0 == a2) {
2927            tcg_out_vreg3(s, INSN_VBIF, q, 0, a0, a3, a1);
2928        } else {
2929            tcg_out_mov(s, type, a0, a1);
2930            tcg_out_vreg3(s, INSN_VBSL, q, 0, a0, a2, a3);
2931        }
2932        return;
2933
2934    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
2935    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
2936    default:
2937        g_assert_not_reached();
2938    }
2939}
2940
2941int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
2942{
2943    switch (opc) {
2944    case INDEX_op_add_vec:
2945    case INDEX_op_sub_vec:
2946    case INDEX_op_and_vec:
2947    case INDEX_op_andc_vec:
2948    case INDEX_op_or_vec:
2949    case INDEX_op_orc_vec:
2950    case INDEX_op_xor_vec:
2951    case INDEX_op_not_vec:
2952    case INDEX_op_shli_vec:
2953    case INDEX_op_shri_vec:
2954    case INDEX_op_sari_vec:
2955    case INDEX_op_ssadd_vec:
2956    case INDEX_op_sssub_vec:
2957    case INDEX_op_usadd_vec:
2958    case INDEX_op_ussub_vec:
2959    case INDEX_op_bitsel_vec:
2960        return 1;
2961    case INDEX_op_abs_vec:
2962    case INDEX_op_cmp_vec:
2963    case INDEX_op_mul_vec:
2964    case INDEX_op_neg_vec:
2965    case INDEX_op_smax_vec:
2966    case INDEX_op_smin_vec:
2967    case INDEX_op_umax_vec:
2968    case INDEX_op_umin_vec:
2969        return vece < MO_64;
2970    case INDEX_op_shlv_vec:
2971    case INDEX_op_shrv_vec:
2972    case INDEX_op_sarv_vec:
2973    case INDEX_op_rotli_vec:
2974    case INDEX_op_rotlv_vec:
2975    case INDEX_op_rotrv_vec:
2976        return -1;
2977    default:
2978        return 0;
2979    }
2980}
2981
2982void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
2983                       TCGArg a0, ...)
2984{
2985    va_list va;
2986    TCGv_vec v0, v1, v2, t1, t2, c1;
2987    TCGArg a2;
2988
2989    va_start(va, a0);
2990    v0 = temp_tcgv_vec(arg_temp(a0));
2991    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
2992    a2 = va_arg(va, TCGArg);
2993    va_end(va);
2994
2995    switch (opc) {
2996    case INDEX_op_shlv_vec:
2997        /*
2998         * Merely propagate shlv_vec to arm_ushl_vec.
2999         * In this way we don't set TCG_TARGET_HAS_shv_vec
3000         * because everything is done via expansion.
3001         */
3002        v2 = temp_tcgv_vec(arg_temp(a2));
3003        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(v0),
3004                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3005        break;
3006
3007    case INDEX_op_shrv_vec:
3008    case INDEX_op_sarv_vec:
3009        /* Right shifts are negative left shifts for NEON.  */
3010        v2 = temp_tcgv_vec(arg_temp(a2));
3011        t1 = tcg_temp_new_vec(type);
3012        tcg_gen_neg_vec(vece, t1, v2);
3013        if (opc == INDEX_op_shrv_vec) {
3014            opc = INDEX_op_arm_ushl_vec;
3015        } else {
3016            opc = INDEX_op_arm_sshl_vec;
3017        }
3018        vec_gen_3(opc, type, vece, tcgv_vec_arg(v0),
3019                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3020        tcg_temp_free_vec(t1);
3021        break;
3022
3023    case INDEX_op_rotli_vec:
3024        t1 = tcg_temp_new_vec(type);
3025        tcg_gen_shri_vec(vece, t1, v1, -a2 & ((8 << vece) - 1));
3026        vec_gen_4(INDEX_op_arm_sli_vec, type, vece,
3027                  tcgv_vec_arg(v0), tcgv_vec_arg(t1), tcgv_vec_arg(v1), a2);
3028        tcg_temp_free_vec(t1);
3029        break;
3030
3031    case INDEX_op_rotlv_vec:
3032        v2 = temp_tcgv_vec(arg_temp(a2));
3033        t1 = tcg_temp_new_vec(type);
3034        c1 = tcg_constant_vec(type, vece, 8 << vece);
3035        tcg_gen_sub_vec(vece, t1, v2, c1);
3036        /* Right shifts are negative left shifts for NEON.  */
3037        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t1),
3038                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3039        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(v0),
3040                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3041        tcg_gen_or_vec(vece, v0, v0, t1);
3042        tcg_temp_free_vec(t1);
3043        break;
3044
3045    case INDEX_op_rotrv_vec:
3046        v2 = temp_tcgv_vec(arg_temp(a2));
3047        t1 = tcg_temp_new_vec(type);
3048        t2 = tcg_temp_new_vec(type);
3049        c1 = tcg_constant_vec(type, vece, 8 << vece);
3050        tcg_gen_neg_vec(vece, t1, v2);
3051        tcg_gen_sub_vec(vece, t2, c1, v2);
3052        /* Right shifts are negative left shifts for NEON.  */
3053        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t1),
3054                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3055        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t2),
3056                  tcgv_vec_arg(v1), tcgv_vec_arg(t2));
3057        tcg_gen_or_vec(vece, v0, t1, t2);
3058        tcg_temp_free_vec(t1);
3059        tcg_temp_free_vec(t2);
3060        break;
3061
3062    default:
3063        g_assert_not_reached();
3064    }
3065}
3066
3067static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3068{
3069    int i;
3070    for (i = 0; i < count; ++i) {
3071        p[i] = INSN_NOP;
3072    }
3073}
3074
3075/* Compute frame size via macros, to share between tcg_target_qemu_prologue
3076   and tcg_register_jit.  */
3077
3078#define PUSH_SIZE  ((11 - 4 + 1 + 1) * sizeof(tcg_target_long))
3079
3080#define FRAME_SIZE \
3081    ((PUSH_SIZE \
3082      + TCG_STATIC_CALL_ARGS_SIZE \
3083      + CPU_TEMP_BUF_NLONGS * sizeof(long) \
3084      + TCG_TARGET_STACK_ALIGN - 1) \
3085     & -TCG_TARGET_STACK_ALIGN)
3086
3087#define STACK_ADDEND  (FRAME_SIZE - PUSH_SIZE)
3088
3089static void tcg_target_qemu_prologue(TCGContext *s)
3090{
3091    /* Calling convention requires us to save r4-r11 and lr.  */
3092    /* stmdb sp!, { r4 - r11, lr } */
3093    tcg_out32(s, (COND_AL << 28) | 0x092d4ff0);
3094
3095    /* Reserve callee argument and tcg temp space.  */
3096    tcg_out_dat_rI(s, COND_AL, ARITH_SUB, TCG_REG_CALL_STACK,
3097                   TCG_REG_CALL_STACK, STACK_ADDEND, 1);
3098    tcg_set_frame(s, TCG_REG_CALL_STACK, TCG_STATIC_CALL_ARGS_SIZE,
3099                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3100
3101    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3102
3103    tcg_out_bx(s, COND_AL, tcg_target_call_iarg_regs[1]);
3104
3105    /*
3106     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3107     * and fall through to the rest of the epilogue.
3108     */
3109    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3110    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, 0);
3111    tcg_out_epilogue(s);
3112}
3113
3114static void tcg_out_epilogue(TCGContext *s)
3115{
3116    /* Release local stack frame.  */
3117    tcg_out_dat_rI(s, COND_AL, ARITH_ADD, TCG_REG_CALL_STACK,
3118                   TCG_REG_CALL_STACK, STACK_ADDEND, 1);
3119
3120    /* ldmia sp!, { r4 - r11, pc } */
3121    tcg_out32(s, (COND_AL << 28) | 0x08bd8ff0);
3122}
3123
3124typedef struct {
3125    DebugFrameHeader h;
3126    uint8_t fde_def_cfa[4];
3127    uint8_t fde_reg_ofs[18];
3128} DebugFrame;
3129
3130#define ELF_HOST_MACHINE EM_ARM
3131
3132/* We're expecting a 2 byte uleb128 encoded value.  */
3133QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3134
3135static const DebugFrame debug_frame = {
3136    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3137    .h.cie.id = -1,
3138    .h.cie.version = 1,
3139    .h.cie.code_align = 1,
3140    .h.cie.data_align = 0x7c,             /* sleb128 -4 */
3141    .h.cie.return_column = 14,
3142
3143    /* Total FDE size does not include the "len" member.  */
3144    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3145
3146    .fde_def_cfa = {
3147        12, 13,                         /* DW_CFA_def_cfa sp, ... */
3148        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3149        (FRAME_SIZE >> 7)
3150    },
3151    .fde_reg_ofs = {
3152        /* The following must match the stmdb in the prologue.  */
3153        0x8e, 1,                        /* DW_CFA_offset, lr, -4 */
3154        0x8b, 2,                        /* DW_CFA_offset, r11, -8 */
3155        0x8a, 3,                        /* DW_CFA_offset, r10, -12 */
3156        0x89, 4,                        /* DW_CFA_offset, r9, -16 */
3157        0x88, 5,                        /* DW_CFA_offset, r8, -20 */
3158        0x87, 6,                        /* DW_CFA_offset, r7, -24 */
3159        0x86, 7,                        /* DW_CFA_offset, r6, -28 */
3160        0x85, 8,                        /* DW_CFA_offset, r5, -32 */
3161        0x84, 9,                        /* DW_CFA_offset, r4, -36 */
3162    }
3163};
3164
3165void tcg_register_jit(const void *buf, size_t buf_size)
3166{
3167    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3168}
3169